* [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces
@ 2012-08-03 8:40 Jack Morgenstein
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein,
dotanb-VPRAkNaXOzVWk0Htik3J/w, tziporet-VPRAkNaXOzVWk0Htik3J/w,
yevgenyp-VPRAkNaXOzVWk0Htik3J/w, liranl-VPRAkNaXOzVWk0Htik3J/w
This patch set adds SRIOV support for IB interfaces.
Patches 1-4 are "precondition" patches.
Patches 5-22 actually implement the feature.
NOTE: Patch 18 depends on patch IB/mlx4: fix possible deadlock with sm_lock spinlock
(a separately-submitted patch) being applied previously to the for-next git.
This patch set introduces Infiniband SRIOV support for ConnectX2 and ConnectX3
devices. Each function presents itself as an independent vHCA (virtual HCA) to
the host while a single HCA is observable by the network, which is unaware of
the vHCAs. No changes are required by the IB subsystem, ULPs, and apps to
support SRIOV, and vHCAs are interoperable with any existing (non-virtualized)
IB deployments.
We term this model for SRIOV implementation the shared-port model.
Sharing the same physical port(s) among multiple vHCAs is achieved as follows:
1. Each vHCA port presents its own virtual GID table.
Currently, the virtual GID table comprises a single entry (at index 0) that
maps to a unique index in the physical GID table. The vHCA of the PF maps to
physical GID index 0. To obtain GIDs for other vHCAs, alias GUIDs are requested
from the SM. These are GUIDs which the SM places, per port, in the port's guid
table after the 0'th slot (which is read-only and determined by the FW).
The host admin can assign GIDs to vHCAs using a sysfs interface (see below).
2. Each vHCA port presents its own virtual PKey table.
The virtual PKey table is a mapping of selected indexes of the physical pkey table.
The host admin can control which pkey indexes are mapped to which virtual indexes
using a sysfs interface (see below). Note that the physical PKey table may contain
both full and partial memberships of the same PKey to allow different membership
types in different virtual tables.
3. Each vHCA port has its own virtual port state.
A vHCA port is up if the following conditions apply:
- The physical port is up
- The virtual GID table contains the GIDs requested by the host admin
- The SM has acknowledged the requested GIDs since the last time that
the physical port came up
4. Other port attributes are shared, e.g., GID prefix, LID, SM LID, LMC mask.
5. Special QPs are para-virtualized.
vHCAs are not given direct access to QP0/1. Rather, these QPs are operated by a
special context hosted by the PF, which mediates access to/from vHCAs.
This is done by opening a “tunnel” per vHCA port per QP0/1. A tunnel comprises
a pair of UD QPs: a “Tunnel QP” in the PF-context and a “Proxy QP” in the vHCA.
All vHCA MAD traffic must pass through the corresponding tunnel.
vHCA QPs cannot be assigned to VL15 and are denied of the well-known QKey.
QP0 access is restricted to the PF vHCA. VF vHCAs also have (virtual) QP0’s,
but they never receive any SMPs and all SMPs sent are discarded.
QP1 traffic is allowed for all vHCAs, but special care is required to bridge
the gap between the host and network views.
Specifically:
- Transaction IDs are mapped to guarantee uniqueness among vHCAs
- CM para-virtualization
o Incoming requests are steered to the correct vHCA according to the embedded GID
o Local communication IDs are mapped to ensure uniqueness among vHCAs
- Multicast para-virtualization
o The PF context aggregates membership state from all vHCAs
o The SA is contacted only when the aggregate membership changes
o If the aggregate does not change, the PF context will provide the
requesting vHCA with the proper response
Incoming MADs are steered according to:
- the DGID If a GRH is present
- the mapped transaction ID for response MADs
- the embedded GID in CM requests
- the remote communication ID in other CM messages
To allow the host admin to control the virtual GID and PKey tables of vHCAs,
a new sysfs ‘iov’ sub-tree has been added under the PF infiniband device.
Details on this mechanism can be found in the change log of:
IB/mlx4: Add iov directory in sysfs under the ib device
Some Limitations
----------------
1. FMRs are not currently supported on slaves. This will be corrected in a
future submission.
2. RoCE is not currently supported on slaves. This will be corrected in a
future submission.
Changes for V2
--------------
1. Patches already accepted by Roland in the V1 set are removed from V2.
The following (precondition) patches submitted in V1 were accepted:
1: net/mlx4_core: Pass an invalid PCI id number to VFs
3: IB/mlx4: Add debug printouts
6: IB/sa: Add GuidInfoRecord query support
7: IB/core: move macros from cm_msgs.h to ib_cm.h
8: {NET,IB}/mlx4: Use port management change event instead of smp_snoop
9: net/mlx4_core: For SRIOV, initialize ib port-capabilities for all slaves
10: net/mlx4_core: Implement mechanism for reserved qkeys
11: net/mlx4_core: Allow guests to support IB ports
12: {NET,IB}/mlx4_core: place phys gid and pkey tbl sizes in mlx4_phys_caps
struct and paravirtualize them
2. patches (new numbering) "5: net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area"
and "8: IB/mlx4: SRIOV multiplex and demultiplex MADs"
were adjusted to account for changes introduced by commits to mlx4_core from other gits in the
kernel V3.5 merge
3. A fix to event processing in the V2 patch set was added to patch
"18: IB/mlx4: Miscellaneous adjustments to SRIOV IB support" :
Add spinlock to slave_event(), since it is called both in interrupt
context and in process context (due to patch 6, and also if smp_snoop
is used).
This fix was found and implemented by Saeed Mahameed <saeedm@mellanox.com>
4. Two new patches were added to V2:
21: {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations
22: IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes
5. The workaround for the KVM IRQ management problem is no longer needed. The KVM group
fixed the problem in kernel 3.5-rc7.
Changes for V1
--------------
1. librdmacm now supports multiple VF/PF on the same host (patch 29).
2. Several patches cleaned up (these were indicated in the V0 changelogs).
Major cleanups in patch 22 and patch 24.
3. Eliminated code duplication in Port Management Change event code (patch 8).
4. Now use pr_debug, instead of mlx4_ib_debug, and there is no module parameter
(Roland's recommendation).
5. mlx4_master_func_num() to get the master's "slave_id", to make code more readable.
6. Fixed illegal use of port num field for a force-loopback bit in ib_ah structure
(V0 patch 2 -- eliminated). The force-loopback bit is now set for Tunnel QPs in
mlx4_ib_post send (patch 17).
7. New patch 2 (not related to 6 above) to reserve bits in enum ib_qp_create_flags.
8. V0 patch 26 is now rolled into V1 patch 22. This allowed us to eliminate function
mlx4_ib_indexed_gid from patch 22 (replaced by using __mlx4_ib_query_gid() from V0 patch 26).
Amir Vadai (1):
IB/mlx4: Add CM paravirtualization
Erez Shitrit (1):
IB/sa: Add GuidInfoRecord query support.
Jack Morgenstein (26):
net/mlx4_core: Pass an invalid PCI id number to VFs
IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver
use
IB/mlx4: Add debug printouts
IB/core: change pkey table lookups to support full and partial
membership for the same pkey
IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey
match
IB/core: move macros from cm_msgs.h to ib_cm.h
{NET,IB}/mlx4: Use port management change event instead of smp_snoop
net/mlx4_core: For SRIOV, initialize ib port-capabilities for all
slaves
net/mlx4_core: Implement mechanism for reserved qkeys
net/mlx4_core: Allow guests to support IB ports
{NET,IB}/mlx4_core: place phys gid and pkey tbl sizes in
mlx4_phys_caps struct and paravirtualize them
IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area
IB/mlx4: Initialize SRIOV IB support for slaves in master context
{NET,IB}/mlx4: Implement QP paravirtualization and maintain
phys_pkey_cache for smp_snoop
IB/mlx4: SRIOV multiplex and demultiplex MADs
{NET,IB}/mlx4: MAD_IFC paravirtualization
net/mlx4_core: Add IB port-state machine, and port mgmt event
propagation infrastructure
{NET,IB}/mlx4: Add alias_guid mechanism
IB/mlx4: Propagate pkey and guid change port management events to
slaves
IB/mlx4: Add iov directory in sysfs under the ib device
net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB
net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode
IB/mlx4: Miscellaneous adjustments to SRIOV IB support
{NET,IB}/mlx4: Activate SRIOV mode for IB
{NET,IB}/mlx4: Paravirtualize Node Guids for slaves.
Oren Duer (1):
IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV
drivers/infiniband/core/cache.c | 42 +-
drivers/infiniband/core/cm_msgs.h | 12 -
drivers/infiniband/core/device.c | 17 +-
drivers/infiniband/core/sa_query.c | 133 ++
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 688 ++++++++
drivers/infiniband/hw/mlx4/cm.c | 437 +++++
drivers/infiniband/hw/mlx4/cq.c | 31 +-
drivers/infiniband/hw/mlx4/mad.c | 1684 +++++++++++++++++++-
drivers/infiniband/hw/mlx4/main.c | 285 +++-
drivers/infiniband/hw/mlx4/mcg.c | 1254 +++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 360 +++++-
drivers/infiniband/hw/mlx4/qp.c | 651 +++++++-
drivers/infiniband/hw/mlx4/sysfs.c | 794 +++++++++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 190 +++-
drivers/net/ethernet/mellanox/mlx4/en_main.c | 5 +-
drivers/net/ethernet/mellanox/mlx4/eq.c | 259 +++-
drivers/net/ethernet/mellanox/mlx4/fw.c | 227 +++-
drivers/net/ethernet/mellanox/mlx4/fw.h | 3 +
drivers/net/ethernet/mellanox/mlx4/intf.c | 5 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 124 ++-
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 116 +-
drivers/net/ethernet/mellanox/mlx4/port.c | 21 +-
drivers/net/ethernet/mellanox/mlx4/qp.c | 67 +-
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 220 +++-
include/linux/mlx4/device.h | 179 ++-
include/linux/mlx4/driver.h | 5 +-
include/linux/mlx4/qp.h | 3 +-
include/rdma/ib_cache.h | 16 +
include/rdma/ib_cm.h | 12 +
include/rdma/ib_sa.h | 33 +
include/rdma/ib_verbs.h | 3 +
32 files changed, 7520 insertions(+), 358 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c
create mode 100644 drivers/infiniband/hw/mlx4/cm.c
create mode 100644 drivers/infiniband/hw/mlx4/mcg.c
create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c
Cc: dotanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org
Cc: tziporet-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org
Cc: yevgenyp-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org
Cc: liranl-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org
Amir Vadai (1):
IB/mlx4: Add CM paravirtualization
Jack Morgenstein (20):
IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver
use
IB/core: change pkey table lookups to support full and partial
membership for the same pkey
IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey
match
IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area
IB/mlx4: Initialize SRIOV IB support for slaves in master context
{NET,IB}/mlx4: Implement QP paravirtualization and maintain
phys_pkey_cache for smp_snoop
IB/mlx4: SRIOV multiplex and demultiplex MADs
{NET,IB}/mlx4: MAD_IFC paravirtualization
net/mlx4_core: Add IB port-state machine, and port mgmt event
propagation infrastructure
{NET,IB}/mlx4: Add alias_guid mechanism
IB/mlx4: Propagate pkey and guid change port management events to
slaves
IB/mlx4: Add iov directory in sysfs under the ib device
net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB
net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode
IB/mlx4: Miscellaneous adjustments to SRIOV IB support
{NET,IB}/mlx4: Activate SRIOV mode for IB
{NET,IB}/mlx4: Paravirtualize Node Guids for slaves
{NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no
calculations
IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver
initializes
Oren Duer (1):
IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV
drivers/infiniband/core/cache.c | 42 +-
drivers/infiniband/core/device.c | 17 +-
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 688 +++++++++
drivers/infiniband/hw/mlx4/cm.c | 437 ++++++
drivers/infiniband/hw/mlx4/cq.c | 31 +-
drivers/infiniband/hw/mlx4/mad.c | 1573 +++++++++++++++++++-
drivers/infiniband/hw/mlx4/main.c | 273 +++-
drivers/infiniband/hw/mlx4/mcg.c | 1254 ++++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 341 +++++-
drivers/infiniband/hw/mlx4/qp.c | 656 +++++++-
drivers/infiniband/hw/mlx4/sysfs.c | 794 ++++++++++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 191 +++-
drivers/net/ethernet/mellanox/mlx4/eq.c | 245 +++-
drivers/net/ethernet/mellanox/mlx4/fw.c | 244 ++-
drivers/net/ethernet/mellanox/mlx4/fw.h | 11 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 91 +-
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 50 +
drivers/net/ethernet/mellanox/mlx4/port.c | 10 +
drivers/net/ethernet/mellanox/mlx4/qp.c | 100 ++-
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 220 +++-
include/linux/mlx4/device.h | 69 +-
include/linux/mlx4/driver.h | 2 +
include/linux/mlx4/qp.h | 3 +-
include/rdma/ib_cache.h | 16 +
include/rdma/ib_verbs.h | 3 +
26 files changed, 7080 insertions(+), 283 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c
create mode 100644 drivers/infiniband/hw/mlx4/cm.c
create mode 100644 drivers/infiniband/hw/mlx4/mcg.c
create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-2-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey Jack Morgenstein
` (22 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Reserve bits 26-31 for internal use by low-level drivers. Two
such bits are used in the mlx4 driver SRIOV IB implementation.
These enum additions guarantee that the core layer will never use
these bits, so that low level drivers may safely make use of them.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
include/rdma/ib_verbs.h | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 07996af..46bc045 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -614,6 +614,9 @@ enum ib_qp_type {
enum ib_qp_create_flags {
IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
+ /* reserve bits 26-31 for low level drivers' internal use */
+ IB_QP_CREATE_RESERVED_START = 1 << 26,
+ IB_QP_CREATE_RESERVED_END = 1 << 31,
};
struct ib_qp_init_attr {
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-3-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match Jack Morgenstein
` (21 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein, Liran Liss
Enhance the cached and non-cached pkey table lookups to enable limited and full
members of the same pkey to co-exist in the pkey table.
This is necessary for SRIOV to allow for a scheme where some guests would have the full
membership pkey in their virtual pkey table, where other guests on the same hypervisor
would have the limited one. In that sense, its an extension of the IBTA model for
non virtualized nodes.
To accomplish this, we need both the limited and full membership pkeys to be present
in the master's (hypervisor physical port) pkey table.
The algorithm for supporting pkey tables which contain both the limited and the full
membership versions of the same pkey works as follows:
When scanning the pkey table for a 15 bit pkey:
A. If there is a full member version of that pkey anywhere
in the table, return its index (even if a limited-member
version of the pkey exists earlier in the table).
B. If the full member version is not in the table,
but the limited-member version is in the table,
return the index of the limited pkey.
Signed-off-by: Liran Liss <liranl-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/core/cache.c | 14 +++++++++++---
drivers/infiniband/core/device.c | 17 +++++++++++++----
2 files changed, 24 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c
index 9353992..0f2f2b7 100644
--- a/drivers/infiniband/core/cache.c
+++ b/drivers/infiniband/core/cache.c
@@ -167,6 +167,7 @@ int ib_find_cached_pkey(struct ib_device *device,
unsigned long flags;
int i;
int ret = -ENOENT;
+ int partial_ix = -1;
if (port_num < start_port(device) || port_num > end_port(device))
return -EINVAL;
@@ -179,10 +180,17 @@ int ib_find_cached_pkey(struct ib_device *device,
for (i = 0; i < cache->table_len; ++i)
if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) {
- *index = i;
- ret = 0;
- break;
+ if (cache->table[i] & 0x8000) {
+ *index = i;
+ ret = 0;
+ break;
+ } else
+ partial_ix = i;
}
+ if (ret && partial_ix >= 0) {
+ *index = partial_ix;
+ ret = 0;
+ }
read_unlock_irqrestore(&device->cache.lock, flags);
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index e711de4..a645c68 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -707,18 +707,27 @@ int ib_find_pkey(struct ib_device *device,
{
int ret, i;
u16 tmp_pkey;
+ int partial_ix = -1;
for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
if (ret)
return ret;
-
if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
- *index = i;
- return 0;
+ /* if there is full-member pkey take it.*/
+ if (tmp_pkey & 0x8000) {
+ *index = i;
+ return 0;
+ }
+ if (partial_ix < 0)
+ partial_ix = i;
}
}
-
+ /*no full-member, if exists take the limited*/
+ if (partial_ix >= 0) {
+ *index = partial_ix;
+ return 0;
+ }
return -ENOENT;
}
EXPORT_SYMBOL(ib_find_pkey);
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-4-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support Jack Morgenstein
` (20 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
When port pkey table potentially contains both full and partial
membership copies for the same pkey, we need a function to find
the exact (16-bit) pkey index.
This is particularly necessary when the master forwards QP1 MADS
sent by guests. If the guest has sent the MAD with a limited
membership pkey, we wish to forward the MAD using the same limited
membership pkey. Since master may have both the limited and
the full member pkeys in its table, we must make sure to retrieve
the limited membership pkey in this case.
This requires the 16-bit pkey lookup function (which includes the
membership bit).
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/core/cache.c | 32 ++++++++++++++++++++++++++++++++
include/rdma/ib_cache.h | 16 ++++++++++++++++
2 files changed, 48 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c
index 0f2f2b7..d8a8c83 100644
--- a/drivers/infiniband/core/cache.c
+++ b/drivers/infiniband/core/cache.c
@@ -198,6 +198,38 @@ int ib_find_cached_pkey(struct ib_device *device,
}
EXPORT_SYMBOL(ib_find_cached_pkey);
+int ib_find_exact_cached_pkey(struct ib_device *device,
+ u8 port_num,
+ u16 pkey,
+ u16 *index)
+{
+ struct ib_pkey_cache *cache;
+ unsigned long flags;
+ int i;
+ int ret = -ENOENT;
+
+ if (port_num < start_port(device) || port_num > end_port(device))
+ return -EINVAL;
+
+ read_lock_irqsave(&device->cache.lock, flags);
+
+ cache = device->cache.pkey_cache[port_num - start_port(device)];
+
+ *index = -1;
+
+ for (i = 0; i < cache->table_len; ++i)
+ if (cache->table[i] == pkey) {
+ *index = i;
+ ret = 0;
+ break;
+ }
+
+ read_unlock_irqrestore(&device->cache.lock, flags);
+
+ return ret;
+}
+EXPORT_SYMBOL(ib_find_exact_cached_pkey);
+
int ib_get_cached_lmc(struct ib_device *device,
u8 port_num,
u8 *lmc)
diff --git a/include/rdma/ib_cache.h b/include/rdma/ib_cache.h
index 00a2b8e..ad9a3c2 100644
--- a/include/rdma/ib_cache.h
+++ b/include/rdma/ib_cache.h
@@ -101,6 +101,22 @@ int ib_find_cached_pkey(struct ib_device *device,
u16 *index);
/**
+ * ib_find_exact_cached_pkey - Returns the PKey table index where a specified
+ * PKey value occurs. Comparison uses the FULL 16 bits (incl membership bit)
+ * @device: The device to query.
+ * @port_num: The port number of the device to search for the PKey.
+ * @pkey: The PKey value to search for.
+ * @index: The index into the cached PKey table where the PKey was found.
+ *
+ * ib_find_exact_cached_pkey() searches the specified PKey table in
+ * the local software cache.
+ */
+int ib_find_exact_cached_pkey(struct ib_device *device,
+ u8 port_num,
+ u16 pkey,
+ u16 *index);
+
+/**
* ib_get_cached_lmc - Returns a cached lmc table entry
* @device: The device to query.
* @port_num: The port number of the device to query.
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (2 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-5-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 05/22] net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area Jack Morgenstein
` (19 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
1. Introduce the basic sriov parvirtualization context objects
for multiplexing and demultiplexing MADs.
2. Introduce support for the new proxy and tunnel QP types.
This patch introduces the objects required by the master
for managing QP paravirtualization for guests.
struct mlx4_ib_sriov{} is created by the master only.
It is a container for the following:
1. All the info required by the PPF to multiplex and de-multiplex MADs
(including those from the PF). (struct mlx4_ib_demux_ctx demux)
2. All the info required to manage alias GUIDs (i.e., the GUID at
index 0 that each guest perceives. In fact, this is not the
GUID which is actually at index 0, but is, in fact, the GUID
which is at index[<VF number>] in the physical table.
3. structures which are used to manage CM paravirtualization
4. structures for managing the real special QPs when running in
sriov mode.
The real SQPs are controlled by the PPF in this case. All SQPs
created and controlled by the ib core layer are proxy sqps
struct mlx4_ib_demux_ctx{} contains the information per port needed
to manage paravirtualization.
This includes:
1. All multicast paravirt info
2. All tunnel-qp paravirt info for the port.
3. GUID-table and GUID-prefix for the port
4. work queues.
struct mlx4_ib_demux_pv_ctx{} contains all the info for managing
the paravirtualized QPs for one slave/port.
struct mlx4_ib_demux_pv_qp{} contains the info need to run an
individual QP (either tunnel qp or real SQP).
Note: We made use of the 2 most-significant bits in enum
mlx4_ib_qp_flags (based upon enum ib_qp_create_flags (ib_verbs.h))
We need these bits in the low-level driver for internal purposes.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/cq.c | 31 ++-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 128 +++++++-
drivers/infiniband/hw/mlx4/qp.c | 616 ++++++++++++++++++++++++++++++----
include/linux/mlx4/device.h | 1 +
include/linux/mlx4/qp.h | 3 +-
5 files changed, 702 insertions(+), 77 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c
index 6d4ef71..342fabd 100644
--- a/drivers/infiniband/hw/mlx4/cq.c
+++ b/drivers/infiniband/hw/mlx4/cq.c
@@ -547,6 +547,26 @@ static int mlx4_ib_ipoib_csum_ok(__be16 status, __be16 checksum)
checksum == cpu_to_be16(0xffff);
}
+static int use_tunnel_data(struct mlx4_ib_qp *qp, struct mlx4_ib_cq *cq, struct ib_wc *wc,
+ unsigned tail, struct mlx4_cqe *cqe)
+{
+ struct mlx4_ib_proxy_sqp_hdr *hdr;
+
+ ib_dma_sync_single_for_cpu(qp->ibqp.device,
+ qp->sqp_proxy_rcv[tail].map,
+ sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ DMA_FROM_DEVICE);
+ hdr = (struct mlx4_ib_proxy_sqp_hdr *) (qp->sqp_proxy_rcv[tail].addr);
+ wc->pkey_index = be16_to_cpu(hdr->tun.pkey_index);
+ wc->slid = be16_to_cpu(hdr->tun.slid_mac_47_32);
+ wc->sl = (u8) (be16_to_cpu(hdr->tun.sl_vid) >> 12);
+ wc->src_qp = be32_to_cpu(hdr->tun.flags_src_qp) & 0xFFFFFF;
+ wc->wc_flags |= (hdr->tun.g_ml_path & 0x80) ? (IB_WC_GRH) : 0;
+ wc->dlid_path_bits = 0;
+
+ return 0;
+}
+
static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
struct mlx4_ib_qp **cur_qp,
struct ib_wc *wc)
@@ -559,6 +579,7 @@ static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
int is_error;
u32 g_mlpath_rqpn;
u16 wqe_ctr;
+ unsigned tail = 0;
repoll:
cqe = next_cqe_sw(cq);
@@ -634,7 +655,8 @@ repoll:
mlx4_ib_free_srq_wqe(srq, wqe_ctr);
} else {
wq = &(*cur_qp)->rq;
- wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+ tail = wq->tail & (wq->wqe_cnt - 1);
+ wc->wr_id = wq->wrid[tail];
++wq->tail;
}
@@ -717,6 +739,13 @@ repoll:
break;
}
+ if (mlx4_is_mfunc(to_mdev(cq->ibcq.device)->dev)) {
+ if ((*cur_qp)->mlx4_ib_qp_type &
+ (MLX4_IB_QPT_PROXY_SMI_OWNER |
+ MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI))
+ return use_tunnel_data(*cur_qp, cq, wc, tail, cqe);
+ }
+
wc->slid = be16_to_cpu(cqe->rlid);
g_mlpath_rqpn = be32_to_cpu(cqe->g_mlpath_rqpn);
wc->src_qp = g_mlpath_rqpn & 0xffffff;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index c136bb6..1248d57 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -133,8 +133,10 @@ struct mlx4_ib_wq {
};
enum mlx4_ib_qp_flags {
- MLX4_IB_QP_LSO = 1 << 0,
- MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
+ MLX4_IB_QP_LSO = IB_QP_CREATE_IPOIB_UD_LSO,
+ MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
+ MLX4_IB_SRIOV_TUNNEL_QP = 1 << 30,
+ MLX4_IB_SRIOV_SQP = 1 << 31,
};
struct mlx4_ib_gid_entry {
@@ -144,6 +146,68 @@ struct mlx4_ib_gid_entry {
u8 port;
};
+enum mlx4_ib_qp_type {
+ /*
+ * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
+ * here (and in that order) since the MAD layer uses them as
+ * indices into a 2-entry table.
+ */
+ MLX4_IB_QPT_SMI = IB_QPT_SMI,
+ MLX4_IB_QPT_GSI = IB_QPT_GSI,
+
+ MLX4_IB_QPT_RC = IB_QPT_RC,
+ MLX4_IB_QPT_UC = IB_QPT_UC,
+ MLX4_IB_QPT_UD = IB_QPT_UD,
+ MLX4_IB_QPT_RAW_IPV6 = IB_QPT_RAW_IPV6,
+ MLX4_IB_QPT_RAW_ETHERTYPE = IB_QPT_RAW_ETHERTYPE,
+ MLX4_IB_QPT_RAW_PACKET = IB_QPT_RAW_PACKET,
+ MLX4_IB_QPT_XRC_INI = IB_QPT_XRC_INI,
+ MLX4_IB_QPT_XRC_TGT = IB_QPT_XRC_TGT,
+
+ MLX4_IB_QPT_PROXY_SMI_OWNER = 1 << 16,
+ MLX4_IB_QPT_PROXY_SMI = 1 << 17,
+ MLX4_IB_QPT_PROXY_GSI = 1 << 18,
+ MLX4_IB_QPT_TUN_SMI_OWNER = 1 << 19,
+ MLX4_IB_QPT_TUN_SMI = 1 << 20,
+ MLX4_IB_QPT_TUN_GSI = 1 << 21,
+};
+
+#define MLX4_IB_QPT_ANY_SRIOV (MLX4_IB_QPT_PROXY_SMI_OWNER | \
+ MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \
+ MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI)
+
+struct mlx4_ib_tunnel_header {
+ struct mlx4_av av;
+ __be32 remote_qpn;
+ __be32 qkey;
+ __be16 vlan;
+ u8 mac[6];
+ __be16 pkey_index;
+ u8 reserved[6];
+};
+
+struct mlx4_ib_buf {
+ void *addr;
+ dma_addr_t map;
+};
+
+struct mlx4_rcv_tunnel_hdr {
+ __be32 flags_src_qp; /* flags[6:5] is defined for VLANs:
+ * 0x0 - no vlan was in the packet
+ * 0x01 - C-VLAN was in the packet */
+ u8 g_ml_path; /* gid bit stands for ipv6/4 header in RoCE */
+ u8 reserved;
+ __be16 pkey_index;
+ __be16 sl_vid;
+ __be16 slid_mac_47_32;
+ __be32 mac_31_0;
+};
+
+struct mlx4_ib_proxy_sqp_hdr {
+ struct ib_grh grh;
+ struct mlx4_rcv_tunnel_hdr tun;
+} __packed;
+
struct mlx4_ib_qp {
struct ib_qp ibqp;
struct mlx4_qp mqp;
@@ -159,6 +223,7 @@ struct mlx4_ib_qp {
int sq_spare_wqes;
struct mlx4_ib_wq sq;
+ enum mlx4_ib_qp_type mlx4_ib_qp_type;
struct ib_umem *umem;
struct mlx4_mtt mtt;
int buf_size;
@@ -174,6 +239,8 @@ struct mlx4_ib_qp {
int mlx_type;
struct list_head gid_list;
struct list_head steering_rules;
+ struct mlx4_ib_buf *sqp_proxy_rcv;
+
};
struct mlx4_ib_srq {
@@ -196,6 +263,55 @@ struct mlx4_ib_ah {
union mlx4_ext_av av;
};
+struct mlx4_ib_tun_tx_buf {
+ struct mlx4_ib_buf buf;
+ struct ib_ah *ah;
+};
+
+struct mlx4_ib_demux_pv_qp {
+ struct ib_qp *qp;
+ enum ib_qp_type proxy_qpt;
+ struct mlx4_ib_buf *ring;
+ struct mlx4_ib_tun_tx_buf *tx_ring;
+ spinlock_t tx_lock;
+ unsigned tx_ix_head;
+ unsigned tx_ix_tail;
+};
+
+struct mlx4_ib_demux_pv_ctx {
+ int port;
+ int slave;
+ int has_smi;
+ struct ib_device *ib_dev;
+ struct ib_cq *cq;
+ struct ib_pd *pd;
+ struct ib_mr *mr;
+ struct work_struct work;
+ struct workqueue_struct *wq;
+ struct mlx4_ib_demux_pv_qp qp[2];
+};
+
+struct mlx4_ib_demux_ctx {
+ struct ib_device *ib_dev;
+ int port;
+ struct workqueue_struct *wq;
+ struct workqueue_struct *ud_wq;
+ spinlock_t ud_lock;
+ __be64 subnet_prefix;
+ __be64 guid_cache[128];
+ struct mlx4_ib_dev *dev;
+ struct mlx4_ib_demux_pv_ctx **tun;
+};
+
+struct mlx4_ib_sriov {
+ struct mlx4_ib_demux_ctx demux[MLX4_MAX_PORTS];
+ struct mlx4_ib_demux_pv_ctx *sqps[MLX4_MAX_PORTS];
+ /* when using this spinlock you should use "irq" because
+ * it may be called from interrupt context.*/
+ spinlock_t going_down_lock;
+ int is_going_down;
+};
+
struct mlx4_ib_iboe {
spinlock_t lock;
struct net_device *netdevs[MLX4_MAX_PORTS];
@@ -216,6 +332,7 @@ struct mlx4_ib_dev {
struct ib_mad_agent *send_agent[MLX4_MAX_PORTS][2];
struct ib_ah *sm_ah[MLX4_MAX_PORTS];
spinlock_t sm_lock;
+ struct mlx4_ib_sriov sriov;
struct mutex cap_mask_mutex;
bool ib_active;
@@ -231,6 +348,13 @@ struct ib_event_work {
struct mlx4_eqe ib_eqe;
};
+struct mlx4_ib_qp_tunnel_init_attr {
+ struct ib_qp_init_attr init_attr;
+ int slave;
+ enum ib_qp_type proxy_qp_type;
+ u8 port;
+};
+
static inline struct mlx4_ib_dev *to_mdev(struct ib_device *ibdev)
{
return container_of(ibdev, struct mlx4_ib_dev, ib_dev);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index a6d8ea0..3a3a690 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -38,6 +38,7 @@
#include <rdma/ib_cache.h>
#include <rdma/ib_pack.h>
#include <rdma/ib_addr.h>
+#include <rdma/ib_mad.h>
#include <linux/mlx4/qp.h>
@@ -110,16 +111,38 @@ static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
return container_of(mqp, struct mlx4_ib_sqp, qp);
}
+static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
+{
+ if (!mlx4_is_master(dev->dev))
+ return 0;
+
+ return qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
+ qp->mqp.qpn < dev->dev->caps.base_sqpn +
+ 8 + 16 * MLX4_MFUNC_MAX;
+}
+
static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
- return qp->mqp.qpn >= dev->dev->caps.sqp_start &&
- qp->mqp.qpn <= dev->dev->caps.sqp_start + 3;
+ return ((mlx4_is_master(dev->dev) &&
+ qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
+ qp->mqp.qpn <= dev->dev->caps.base_sqpn + 3) ||
+ (qp->mqp.qpn >= dev->dev->caps.sqp_start &&
+ qp->mqp.qpn <= dev->dev->caps.sqp_start + 3));
}
+/* used for INIT/CLOSE port logic */
static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
- return qp->mqp.qpn >= dev->dev->caps.sqp_start &&
- qp->mqp.qpn <= dev->dev->caps.sqp_start + 1;
+ int qp0;
+
+ /* qp0 is either the proxy qp0, or the real qp0 */
+ qp0 = (qp->mqp.qpn >= dev->dev->caps.sqp_start &&
+ qp->mqp.qpn <= dev->dev->caps.sqp_start + 1) ||
+ (mlx4_is_master(dev->dev) &&
+ qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
+ qp->mqp.qpn <= dev->dev->caps.base_sqpn + 1);
+
+ return qp0;
}
static void *get_wqe(struct mlx4_ib_qp *qp, int offset)
@@ -270,7 +293,7 @@ static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type)
}
}
-static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
+static int send_wqe_overhead(enum mlx4_ib_qp_type type, u32 flags)
{
/*
* UD WQEs must have a datagram segment.
@@ -279,19 +302,29 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
* header and space for the ICRC).
*/
switch (type) {
- case IB_QPT_UD:
+ case MLX4_IB_QPT_UD:
return sizeof (struct mlx4_wqe_ctrl_seg) +
sizeof (struct mlx4_wqe_datagram_seg) +
((flags & MLX4_IB_QP_LSO) ? MLX4_IB_LSO_HEADER_SPARE : 0);
- case IB_QPT_UC:
+ case MLX4_IB_QPT_PROXY_SMI_OWNER:
+ case MLX4_IB_QPT_PROXY_SMI:
+ case MLX4_IB_QPT_PROXY_GSI:
+ return sizeof(struct mlx4_wqe_ctrl_seg) +
+ sizeof(struct mlx4_wqe_datagram_seg) + 64;
+ case MLX4_IB_QPT_TUN_SMI_OWNER:
+ case MLX4_IB_QPT_TUN_GSI:
+ return sizeof(struct mlx4_wqe_ctrl_seg) +
+ sizeof(struct mlx4_wqe_datagram_seg);
+
+ case MLX4_IB_QPT_UC:
return sizeof (struct mlx4_wqe_ctrl_seg) +
sizeof (struct mlx4_wqe_raddr_seg);
- case IB_QPT_RC:
+ case MLX4_IB_QPT_RC:
return sizeof (struct mlx4_wqe_ctrl_seg) +
sizeof (struct mlx4_wqe_atomic_seg) +
sizeof (struct mlx4_wqe_raddr_seg);
- case IB_QPT_SMI:
- case IB_QPT_GSI:
+ case MLX4_IB_QPT_SMI:
+ case MLX4_IB_QPT_GSI:
return sizeof (struct mlx4_wqe_ctrl_seg) +
ALIGN(MLX4_IB_UD_HEADER_SIZE +
DIV_ROUND_UP(MLX4_IB_UD_HEADER_SIZE,
@@ -345,7 +378,7 @@ static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
}
static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
- enum ib_qp_type type, struct mlx4_ib_qp *qp)
+ enum mlx4_ib_qp_type type, struct mlx4_ib_qp *qp)
{
int s;
@@ -360,7 +393,8 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
* For MLX transport we need 2 extra S/G entries:
* one for the header and one for the checksum at the end
*/
- if ((type == IB_QPT_SMI || type == IB_QPT_GSI) &&
+ if ((type == MLX4_IB_QPT_SMI || type == MLX4_IB_QPT_GSI ||
+ type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) &&
cap->max_send_sge + 2 > dev->dev->caps.max_sq_sg)
return -EINVAL;
@@ -404,7 +438,9 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
*/
if (dev->dev->caps.fw_ver >= MLX4_FW_VER_WQE_CTRL_NEC &&
qp->sq_signal_bits && BITS_PER_LONG == 64 &&
- type != IB_QPT_SMI && type != IB_QPT_GSI)
+ type != MLX4_IB_QPT_SMI && type != MLX4_IB_QPT_GSI &&
+ !(type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_PROXY_SMI |
+ MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER)))
qp->sq.wqe_shift = ilog2(64);
else
qp->sq.wqe_shift = ilog2(roundup_pow_of_two(s));
@@ -476,6 +512,54 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev,
return 0;
}
+static int alloc_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
+{
+ int i;
+
+ qp->sqp_proxy_rcv =
+ kmalloc(sizeof(struct mlx4_ib_buf) * qp->rq.wqe_cnt,
+ GFP_KERNEL);
+ if (!qp->sqp_proxy_rcv)
+ return -ENOMEM;
+ for (i = 0; i < qp->rq.wqe_cnt; i++) {
+ qp->sqp_proxy_rcv[i].addr =
+ kmalloc(sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ GFP_KERNEL);
+ if (!qp->sqp_proxy_rcv[i].addr)
+ goto err;
+ qp->sqp_proxy_rcv[i].map =
+ ib_dma_map_single(dev, qp->sqp_proxy_rcv[i].addr,
+ sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ DMA_FROM_DEVICE);
+ }
+ return 0;
+
+err:
+ while (i > 0) {
+ --i;
+ ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
+ sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ DMA_FROM_DEVICE);
+ kfree(qp->sqp_proxy_rcv[i].addr);
+ }
+ kfree(qp->sqp_proxy_rcv);
+ qp->sqp_proxy_rcv = NULL;
+ return -ENOMEM;
+}
+
+static void free_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
+{
+ int i;
+
+ for (i = 0; i < qp->rq.wqe_cnt; i++) {
+ ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
+ sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ DMA_FROM_DEVICE);
+ kfree(qp->sqp_proxy_rcv[i].addr);
+ }
+ kfree(qp->sqp_proxy_rcv);
+}
+
static int qp_has_rq(struct ib_qp_init_attr *attr)
{
if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
@@ -486,10 +570,71 @@ static int qp_has_rq(struct ib_qp_init_attr *attr)
static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
struct ib_qp_init_attr *init_attr,
- struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp)
+ struct ib_udata *udata, int sqpn, struct mlx4_ib_qp **caller_qp)
{
int qpn;
int err;
+ struct mlx4_ib_sqp *sqp;
+ struct mlx4_ib_qp *qp;
+ enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type;
+
+ /* When tunneling special qps, we use a plain UD qp */
+ if (sqpn) {
+ if (mlx4_is_mfunc(dev->dev) &&
+ (!mlx4_is_master(dev->dev) ||
+ !(init_attr->create_flags & MLX4_IB_SRIOV_SQP))) {
+ if (init_attr->qp_type == IB_QPT_GSI)
+ qp_type = MLX4_IB_QPT_PROXY_GSI;
+ else if (mlx4_is_master(dev->dev))
+ qp_type = MLX4_IB_QPT_PROXY_SMI_OWNER;
+ else
+ qp_type = MLX4_IB_QPT_PROXY_SMI;
+ }
+ qpn = sqpn;
+ /* add extra sg entry for tunneling */
+ init_attr->cap.max_recv_sge++;
+ } else if (init_attr->create_flags & MLX4_IB_SRIOV_TUNNEL_QP) {
+ struct mlx4_ib_qp_tunnel_init_attr *tnl_init =
+ container_of(init_attr,
+ struct mlx4_ib_qp_tunnel_init_attr, init_attr);
+ if ((tnl_init->proxy_qp_type != IB_QPT_SMI &&
+ tnl_init->proxy_qp_type != IB_QPT_GSI) ||
+ !mlx4_is_master(dev->dev))
+ return -EINVAL;
+ if (tnl_init->proxy_qp_type == IB_QPT_GSI)
+ qp_type = MLX4_IB_QPT_TUN_GSI;
+ else if (tnl_init->slave == mlx4_master_func_num(dev->dev))
+ qp_type = MLX4_IB_QPT_TUN_SMI_OWNER;
+ else
+ qp_type = MLX4_IB_QPT_TUN_SMI;
+ qpn = dev->dev->caps.base_tunnel_sqpn + 8 * tnl_init->slave +
+ tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
+ sqpn = qpn;
+ }
+
+ if (!*caller_qp) {
+ if (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI ||
+ (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER |
+ MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) {
+ sqp = kzalloc(sizeof(struct mlx4_ib_sqp), GFP_KERNEL);
+ if (!sqp)
+ return -ENOMEM;
+ qp = &sqp->qp;
+ } else {
+ qp = kzalloc(sizeof(struct mlx4_ib_qp), GFP_KERNEL);
+ if (!qp)
+ return -ENOMEM;
+ }
+ } else
+ qp = *caller_qp;
+
+ qp->mlx4_ib_qp_type = qp_type;
+
+ if (mlx4_is_mfunc(dev->dev) &&
+ (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI)) {
+ qpn -= 8;
+ sqpn -= 8;
+ }
mutex_init(&qp->mutex);
spin_lock_init(&qp->sq.lock);
@@ -550,7 +695,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)
qp->flags |= MLX4_IB_QP_LSO;
- err = set_kernel_sq_size(dev, &init_attr->cap, init_attr->qp_type, qp);
+ err = set_kernel_sq_size(dev, &init_attr->cap, qp_type, qp);
if (err)
goto err;
@@ -586,7 +731,13 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
}
if (sqpn) {
- qpn = sqpn;
+ if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+ MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) {
+ if (alloc_proxy_bufs(pd->device, qp)) {
+ err = -ENOMEM;
+ goto err_wrid;
+ }
+ }
} else {
/* Raw packet QPNs must be aligned to 8 bits. If not, the WQE
* BlueFlame setup flow wrongly causes VLAN insertion. */
@@ -595,7 +746,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
else
err = mlx4_qp_reserve_range(dev->dev, 1, 1, &qpn);
if (err)
- goto err_wrid;
+ goto err_proxy;
}
err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp);
@@ -613,13 +764,16 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);
qp->mqp.event = mlx4_ib_qp_event;
-
+ if (!*caller_qp)
+ *caller_qp = qp;
return 0;
err_qpn:
if (!sqpn)
mlx4_qp_release_range(dev->dev, qpn, 1);
-
+err_proxy:
+ if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
+ free_proxy_bufs(pd->device, qp);
err_wrid:
if (pd->uobject) {
if (qp_has_rq(init_attr))
@@ -643,6 +797,8 @@ err_db:
mlx4_db_free(dev->dev, &qp->db);
err:
+ if (!*caller_qp)
+ kfree(qp);
return err;
}
@@ -755,7 +911,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
mlx4_qp_free(dev->dev, &qp->mqp);
- if (!is_sqp(dev, qp))
+ if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp))
mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
mlx4_mtt_cleanup(dev->dev, &qp->mtt);
@@ -768,6 +924,9 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
} else {
kfree(qp->sq.wrid);
kfree(qp->rq.wrid);
+ if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+ MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI))
+ free_proxy_bufs(&dev->ib_dev, qp);
mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
if (qp->rq.wqe_cnt)
mlx4_db_free(dev->dev, &qp->db);
@@ -780,21 +939,25 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
{
- struct mlx4_ib_sqp *sqp;
- struct mlx4_ib_qp *qp;
+ struct mlx4_ib_qp *qp = NULL;
int err;
u16 xrcdn = 0;
/*
- * We only support LSO and multicast loopback blocking, and
- * only for kernel UD QPs.
+ * We only support LSO, vendor flag1, and multicast loopback blocking,
+ * and only for kernel UD QPs.
*/
- if (init_attr->create_flags & ~(IB_QP_CREATE_IPOIB_UD_LSO |
- IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK))
+ if (init_attr->create_flags & ~(MLX4_IB_QP_LSO |
+ MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK |
+ MLX4_IB_SRIOV_TUNNEL_QP | MLX4_IB_SRIOV_SQP))
return ERR_PTR(-EINVAL);
if (init_attr->create_flags &&
- (udata || init_attr->qp_type != IB_QPT_UD))
+ (udata ||
+ ((init_attr->create_flags & ~MLX4_IB_SRIOV_SQP) &&
+ init_attr->qp_type != IB_QPT_UD) ||
+ ((init_attr->create_flags & MLX4_IB_SRIOV_SQP) &&
+ init_attr->qp_type > IB_QPT_GSI)))
return ERR_PTR(-EINVAL);
switch (init_attr->qp_type) {
@@ -810,18 +973,17 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
/* fall through */
case IB_QPT_RC:
case IB_QPT_UC:
- case IB_QPT_UD:
case IB_QPT_RAW_PACKET:
- {
qp = kzalloc(sizeof *qp, GFP_KERNEL);
if (!qp)
return ERR_PTR(-ENOMEM);
-
- err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, 0, qp);
- if (err) {
- kfree(qp);
+ /* fall through */
+ case IB_QPT_UD:
+ {
+ err = create_qp_common(to_mdev(pd->device), pd, init_attr,
+ udata, 0, &qp);
+ if (err)
return ERR_PTR(err);
- }
qp->ibqp.qp_num = qp->mqp.qpn;
qp->xrcdn = xrcdn;
@@ -835,21 +997,13 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
if (udata)
return ERR_PTR(-EINVAL);
- sqp = kzalloc(sizeof *sqp, GFP_KERNEL);
- if (!sqp)
- return ERR_PTR(-ENOMEM);
-
- qp = &sqp->qp;
-
err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
to_mdev(pd->device)->dev->caps.sqp_start +
(init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
init_attr->port_num - 1,
- qp);
- if (err) {
- kfree(sqp);
+ &qp);
+ if (err)
return ERR_PTR(err);
- }
qp->port = init_attr->port_num;
qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
@@ -884,18 +1038,27 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
return 0;
}
-static int to_mlx4_st(enum ib_qp_type type)
+static int to_mlx4_st(struct mlx4_ib_dev *dev, enum mlx4_ib_qp_type type)
{
switch (type) {
- case IB_QPT_RC: return MLX4_QP_ST_RC;
- case IB_QPT_UC: return MLX4_QP_ST_UC;
- case IB_QPT_UD: return MLX4_QP_ST_UD;
- case IB_QPT_XRC_INI:
- case IB_QPT_XRC_TGT: return MLX4_QP_ST_XRC;
- case IB_QPT_SMI:
- case IB_QPT_GSI:
- case IB_QPT_RAW_PACKET: return MLX4_QP_ST_MLX;
- default: return -1;
+ case MLX4_IB_QPT_RC: return MLX4_QP_ST_RC;
+ case MLX4_IB_QPT_UC: return MLX4_QP_ST_UC;
+ case MLX4_IB_QPT_UD: return MLX4_QP_ST_UD;
+ case MLX4_IB_QPT_XRC_INI:
+ case MLX4_IB_QPT_XRC_TGT: return MLX4_QP_ST_XRC;
+ case MLX4_IB_QPT_SMI:
+ case MLX4_IB_QPT_GSI:
+ case MLX4_IB_QPT_RAW_PACKET: return MLX4_QP_ST_MLX;
+
+ case MLX4_IB_QPT_PROXY_SMI_OWNER:
+ case MLX4_IB_QPT_TUN_SMI_OWNER: return (mlx4_is_mfunc(dev->dev) ?
+ MLX4_QP_ST_MLX : -1);
+ case MLX4_IB_QPT_PROXY_SMI:
+ case MLX4_IB_QPT_TUN_SMI:
+ case MLX4_IB_QPT_PROXY_GSI:
+ case MLX4_IB_QPT_TUN_GSI: return (mlx4_is_mfunc(dev->dev) ?
+ MLX4_QP_ST_UD : -1);
+ default: return -1;
}
}
@@ -1043,7 +1206,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
return -ENOMEM;
context->flags = cpu_to_be32((to_mlx4_state(new_state) << 28) |
- (to_mlx4_st(ibqp->qp_type) << 16));
+ (to_mlx4_st(dev, qp->mlx4_ib_qp_type) << 16));
if (!(attr_mask & IB_QP_PATH_MIG_STATE))
context->flags |= cpu_to_be32(MLX4_QP_PM_MIGRATED << 11);
@@ -1121,13 +1284,16 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
}
if (attr_mask & IB_QP_PKEY_INDEX) {
+ if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
+ context->pri_path.disable_pkey_check = 0x40;
context->pri_path.pkey_index = attr->pkey_index;
optpar |= MLX4_QP_OPTPAR_PKEY_INDEX;
}
if (attr_mask & IB_QP_AV) {
if (mlx4_set_path(dev, &attr->ah_attr, &context->pri_path,
- attr_mask & IB_QP_PORT ? attr->port_num : qp->port))
+ attr_mask & IB_QP_PORT ?
+ attr->port_num : qp->port))
goto out;
optpar |= (MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH |
@@ -1210,8 +1376,24 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
if (attr_mask & IB_QP_RQ_PSN)
context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
+ /* proxy and tunnel qp qkeys will be changed in modify-qp wrappers */
if (attr_mask & IB_QP_QKEY) {
- context->qkey = cpu_to_be32(attr->qkey);
+ if (qp->mlx4_ib_qp_type &
+ (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))
+ context->qkey = cpu_to_be32(IB_QP_SET_QKEY);
+ else {
+ if (mlx4_is_mfunc(dev->dev) &&
+ !(qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) &&
+ (attr->qkey & MLX4_RESERVED_QKEY_MASK) ==
+ MLX4_RESERVED_QKEY_BASE) {
+ pr_err("Cannot use reserved QKEY"
+ " 0x%x (range 0xffff0000..0xffffffff"
+ " is reserved)\n", attr->qkey);
+ err = -EINVAL;
+ goto out;
+ }
+ context->qkey = cpu_to_be32(attr->qkey);
+ }
optpar |= MLX4_QP_OPTPAR_Q_KEY;
}
@@ -1227,10 +1409,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
ibqp->qp_type == IB_QPT_UD ||
ibqp->qp_type == IB_QPT_RAW_PACKET)) {
context->pri_path.sched_queue = (qp->port - 1) << 6;
- if (is_qp0(dev, qp))
+ if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI ||
+ qp->mlx4_ib_qp_type &
+ (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) {
context->pri_path.sched_queue |= MLX4_IB_DEFAULT_QP0_SCHED_QUEUE;
- else
+ if (qp->mlx4_ib_qp_type != MLX4_IB_QPT_SMI)
+ context->pri_path.fl = 0x80;
+ } else {
+ if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
+ context->pri_path.fl = 0x80;
context->pri_path.sched_queue |= MLX4_IB_DEFAULT_SCHED_QUEUE;
+ }
}
if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD &&
@@ -1346,7 +1535,7 @@ int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
}
if ((attr_mask & IB_QP_PORT) &&
- (attr->port_num == 0 || attr->port_num > dev->dev->caps.num_ports)) {
+ (attr->port_num == 0 || attr->port_num > dev->num_ports)) {
pr_debug("qpn 0x%x: invalid port number (%d) specified "
"for transition %d to %d. qp_type %d\n",
ibqp->qp_num, attr->port_num, cur_state,
@@ -1400,6 +1589,115 @@ out:
return err;
}
+static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
+ struct ib_send_wr *wr,
+ void *wqe, unsigned *mlx_seg_len)
+{
+ struct mlx4_ib_dev *mdev = to_mdev(sqp->qp.ibqp.device);
+ struct ib_device *ib_dev = &mdev->ib_dev;
+ struct mlx4_wqe_mlx_seg *mlx = wqe;
+ struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
+ struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+ u16 pkey;
+ u32 qkey;
+ int send_size;
+ int header_size;
+ int spc;
+ int i;
+
+ if (wr->opcode != IB_WR_SEND)
+ return -EINVAL;
+
+ send_size = 0;
+
+ for (i = 0; i < wr->num_sge; ++i)
+ send_size += wr->sg_list[i].length;
+
+ /* for proxy-qp0 sends, need to add in size of tunnel header */
+ /* for tunnel-qp0 sends, tunnel header is already in s/g list */
+ if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER)
+ send_size += sizeof(struct mlx4_ib_tunnel_header);
+
+ ib_ud_header_init(send_size, 1, 0, 0, 0, 0, &sqp->ud_header);
+
+ if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) {
+ sqp->ud_header.lrh.service_level =
+ be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
+ sqp->ud_header.lrh.destination_lid =
+ cpu_to_be16(ah->av.ib.g_slid & 0x7f);
+ sqp->ud_header.lrh.source_lid =
+ cpu_to_be16(ah->av.ib.g_slid & 0x7f);
+ }
+
+ mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
+
+ /* force loopback */
+ mlx->flags |= cpu_to_be32(MLX4_WQE_MLX_VL15 | 0x1 | MLX4_WQE_MLX_SLR);
+ mlx->rlid = sqp->ud_header.lrh.destination_lid;
+
+ sqp->ud_header.lrh.virtual_lane = 0;
+ sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
+ ib_get_cached_pkey(ib_dev, sqp->qp.port, 0, &pkey);
+ sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
+ if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
+ sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+ else
+ sqp->ud_header.bth.destination_qpn =
+ cpu_to_be32(mdev->dev->caps.base_tunnel_sqpn +
+ sqp->qp.port - 1);
+
+ sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
+ if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
+ return -EINVAL;
+ sqp->ud_header.deth.qkey = cpu_to_be32(qkey);
+ sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.mqp.qpn);
+
+ sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY;
+ sqp->ud_header.immediate_present = 0;
+
+ header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);
+
+ /*
+ * Inline data segments may not cross a 64 byte boundary. If
+ * our UD header is bigger than the space available up to the
+ * next 64 byte boundary in the WQE, use two inline data
+ * segments to hold the UD header.
+ */
+ spc = MLX4_INLINE_ALIGN -
+ ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1));
+ if (header_size <= spc) {
+ inl->byte_count = cpu_to_be32(1 << 31 | header_size);
+ memcpy(inl + 1, sqp->header_buf, header_size);
+ i = 1;
+ } else {
+ inl->byte_count = cpu_to_be32(1 << 31 | spc);
+ memcpy(inl + 1, sqp->header_buf, spc);
+
+ inl = (void *) (inl + 1) + spc;
+ memcpy(inl + 1, sqp->header_buf + spc, header_size - spc);
+ /*
+ * Need a barrier here to make sure all the data is
+ * visible before the byte_count field is set.
+ * Otherwise the HCA prefetcher could grab the 64-byte
+ * chunk with this inline segment and get a valid (!=
+ * 0xffffffff) byte count but stale data, and end up
+ * generating a packet with bad headers.
+ *
+ * The first inline segment's byte_count field doesn't
+ * need a barrier, because it comes after a
+ * control/MLX segment and therefore is at an offset
+ * of 16 mod 64.
+ */
+ wmb();
+ inl->byte_count = cpu_to_be32(1 << 31 | (header_size - spc));
+ i = 2;
+ }
+
+ *mlx_seg_len =
+ ALIGN(i * sizeof(struct mlx4_wqe_inline_seg) + header_size, 16);
+ return 0;
+}
+
static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
void *wqe, unsigned *mlx_seg_len)
{
@@ -1417,6 +1715,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
int is_vlan = 0;
int is_grh;
u16 vlan;
+ int err = 0;
send_size = 0;
for (i = 0; i < wr->num_sge; ++i)
@@ -1425,8 +1724,24 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
is_eth = rdma_port_get_link_layer(sqp->qp.ibqp.device, sqp->qp.port) == IB_LINK_LAYER_ETHERNET;
is_grh = mlx4_ib_ah_grh_present(ah);
if (is_eth) {
- ib_get_cached_gid(ib_dev, be32_to_cpu(ah->av.ib.port_pd) >> 24,
- ah->av.ib.gid_index, &sgid);
+ if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
+ /* When multi-function is enabled, the ib_core gid
+ * indexes don't necessarily match the hw ones, so
+ * we must use our own cache */
+ sgid.global.subnet_prefix =
+ to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+ subnet_prefix;
+ sgid.global.interface_id =
+ to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+ guid_cache[ah->av.ib.gid_index];
+ } else {
+ err = ib_get_cached_gid(ib_dev,
+ be32_to_cpu(ah->av.ib.port_pd) >> 24,
+ ah->av.ib.gid_index, &sgid);
+ if (err)
+ return err;
+ }
+
vlan = rdma_get_vlan_id(&sgid);
is_vlan = vlan < 0x1000;
}
@@ -1445,8 +1760,21 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
sqp->ud_header.grh.flow_label =
ah->av.ib.sl_tclass_flowlabel & cpu_to_be32(0xfffff);
sqp->ud_header.grh.hop_limit = ah->av.ib.hop_limit;
- ib_get_cached_gid(ib_dev, be32_to_cpu(ah->av.ib.port_pd) >> 24,
- ah->av.ib.gid_index, &sqp->ud_header.grh.source_gid);
+ if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
+ /* When multi-function is enabled, the ib_core gid
+ * indexes don't necessarily match the hw ones, so
+ * we must use our own cache */
+ sqp->ud_header.grh.source_gid.global.subnet_prefix =
+ to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+ subnet_prefix;
+ sqp->ud_header.grh.source_gid.global.interface_id =
+ to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+ guid_cache[ah->av.ib.gid_index];
+ } else
+ ib_get_cached_gid(ib_dev,
+ be32_to_cpu(ah->av.ib.port_pd) >> 24,
+ ah->av.ib.gid_index,
+ &sqp->ud_header.grh.source_gid);
memcpy(sqp->ud_header.grh.destination_gid.raw,
ah->av.ib.dgid, 16);
}
@@ -1458,6 +1786,8 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
(sqp->ud_header.lrh.destination_lid ==
IB_LID_PERMISSIVE ? MLX4_WQE_MLX_SLR : 0) |
(sqp->ud_header.lrh.service_level << 8));
+ if (ah->av.ib.port_pd & cpu_to_be32(0x80000000))
+ mlx->flags |= cpu_to_be32(0x1); /* force loopback */
mlx->rlid = sqp->ud_header.lrh.destination_lid;
}
@@ -1663,6 +1993,63 @@ static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg,
memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6);
}
+static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
+ struct mlx4_wqe_datagram_seg *dseg,
+ struct ib_send_wr *wr, enum ib_qp_type qpt)
+{
+ union mlx4_ext_av *av = &to_mah(wr->wr.ud.ah)->av;
+ struct mlx4_av sqp_av = {0};
+ int port = *((u8 *) &av->ib.port_pd) & 0x3;
+
+ /* force loopback */
+ sqp_av.port_pd = av->ib.port_pd | cpu_to_be32(0x80000000);
+ sqp_av.g_slid = av->ib.g_slid & 0x7f; /* no GRH */
+ sqp_av.sl_tclass_flowlabel = av->ib.sl_tclass_flowlabel &
+ cpu_to_be32(0xf0000000);
+
+ memcpy(dseg->av, &sqp_av, sizeof(struct mlx4_av));
+ dseg->dqpn = cpu_to_be32(dev->dev->caps.base_tunnel_sqpn +
+ qpt * 2 + port - 1);
+ /* use well-known qkey from the QPC */
+ dseg->qkey = cpu_to_be32(0x80000000);
+}
+
+static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len)
+{
+ struct mlx4_wqe_inline_seg *inl = wqe;
+ struct mlx4_ib_tunnel_header hdr;
+ struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+ int spc;
+ int i;
+
+ memcpy(&hdr.av, &ah->av, sizeof hdr.av);
+ hdr.remote_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+ hdr.pkey_index = cpu_to_be16(wr->wr.ud.pkey_index);
+ hdr.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+
+ spc = MLX4_INLINE_ALIGN -
+ ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1));
+ if (sizeof(hdr) <= spc) {
+ memcpy(inl + 1, &hdr, sizeof(hdr));
+ wmb();
+ inl->byte_count = cpu_to_be32(1 << 31 | sizeof(hdr));
+ i = 1;
+ } else {
+ memcpy(inl + 1, &hdr, spc);
+ wmb();
+ inl->byte_count = cpu_to_be32(1 << 31 | spc);
+
+ inl = (void *) (inl + 1) + spc;
+ memcpy(inl + 1, (void *) &hdr + spc, sizeof(hdr) - spc);
+ wmb();
+ inl->byte_count = cpu_to_be32(1 << 31 | (sizeof(hdr) - spc));
+ i = 2;
+ }
+
+ *mlx_seg_len =
+ ALIGN(i * sizeof(struct mlx4_wqe_inline_seg) + sizeof(hdr), 16);
+}
+
static void set_mlx_icrc_seg(void *dseg)
{
u32 *t = dseg;
@@ -1744,6 +2131,13 @@ static __be32 send_ieth(struct ib_send_wr *wr)
}
}
+static void add_zero_len_inline(void *wqe)
+{
+ struct mlx4_wqe_inline_seg *inl = wqe;
+ memset(wqe, 0, 16);
+ inl->byte_count = cpu_to_be32(1 << 31);
+}
+
int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
struct ib_send_wr **bad_wr)
{
@@ -1802,9 +2196,9 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
wqe += sizeof *ctrl;
size = sizeof *ctrl / 16;
- switch (ibqp->qp_type) {
- case IB_QPT_RC:
- case IB_QPT_UC:
+ switch (qp->mlx4_ib_qp_type) {
+ case MLX4_IB_QPT_RC:
+ case MLX4_IB_QPT_UC:
switch (wr->opcode) {
case IB_WR_ATOMIC_CMP_AND_SWP:
case IB_WR_ATOMIC_FETCH_AND_ADD:
@@ -1865,7 +2259,25 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
}
break;
- case IB_QPT_UD:
+ case MLX4_IB_QPT_TUN_SMI_OWNER:
+ err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+ if (unlikely(err)) {
+ *bad_wr = wr;
+ goto out;
+ }
+ wqe += seglen;
+ size += seglen / 16;
+ break;
+ case MLX4_IB_QPT_TUN_SMI:
+ case MLX4_IB_QPT_TUN_GSI:
+ /* this is a UD qp used in MAD responses to slaves. */
+ set_datagram_seg(wqe, wr);
+ /* set the forced-loopback bit in the data seg av */
+ *(__be32 *) wqe |= cpu_to_be32(0x80000000);
+ wqe += sizeof(struct mlx4_wqe_datagram_seg);
+ size += sizeof(struct mlx4_wqe_datagram_seg) / 16;
+ break;
+ case MLX4_IB_QPT_UD:
set_datagram_seg(wqe, wr);
wqe += sizeof (struct mlx4_wqe_datagram_seg);
size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
@@ -1882,8 +2294,47 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
}
break;
- case IB_QPT_SMI:
- case IB_QPT_GSI:
+ case MLX4_IB_QPT_PROXY_SMI_OWNER:
+ if (unlikely(!mlx4_is_master(to_mdev(ibqp->device)->dev))) {
+ err = -ENOSYS;
+ *bad_wr = wr;
+ goto out;
+ }
+ err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+ if (unlikely(err)) {
+ *bad_wr = wr;
+ goto out;
+ }
+ wqe += seglen;
+ size += seglen / 16;
+ /* to start tunnel header on a cache-line boundary */
+ add_zero_len_inline(wqe);
+ wqe += 16;
+ size++;
+ build_tunnel_header(wr, wqe, &seglen);
+ wqe += seglen;
+ size += seglen / 16;
+ break;
+ case MLX4_IB_QPT_PROXY_SMI:
+ /* don't allow QP0 sends on guests */
+ err = -ENOSYS;
+ *bad_wr = wr;
+ goto out;
+ case MLX4_IB_QPT_PROXY_GSI:
+ /* If we are tunneling special qps, this is a UD qp.
+ * In this case we first add a UD segment targeting
+ * the tunnel qp, and then add a header with address
+ * information */
+ set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe, wr, ibqp->qp_type);
+ wqe += sizeof(struct mlx4_wqe_datagram_seg);
+ size += sizeof(struct mlx4_wqe_datagram_seg) / 16;
+ build_tunnel_header(wr, wqe, &seglen);
+ wqe += seglen;
+ size += seglen / 16;
+ break;
+
+ case MLX4_IB_QPT_SMI:
+ case MLX4_IB_QPT_GSI:
err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen);
if (unlikely(err)) {
*bad_wr = wr;
@@ -1909,8 +2360,10 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
size += wr->num_sge * (sizeof (struct mlx4_wqe_data_seg) / 16);
/* Add one more inline data segment for ICRC for MLX sends */
- if (unlikely(qp->ibqp.qp_type == IB_QPT_SMI ||
- qp->ibqp.qp_type == IB_QPT_GSI)) {
+ if (unlikely(qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI ||
+ qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI ||
+ qp->mlx4_ib_qp_type &
+ (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))) {
set_mlx_icrc_seg(dseg + 1);
size += sizeof (struct mlx4_wqe_data_seg) / 16;
}
@@ -2002,8 +2455,10 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
int err = 0;
int nreq;
int ind;
+ int max_gs;
int i;
+ max_gs = qp->rq.max_gs;
spin_lock_irqsave(&qp->rq.lock, flags);
ind = qp->rq.head & (qp->rq.wqe_cnt - 1);
@@ -2023,10 +2478,25 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
scat = get_recv_wqe(qp, ind);
+ if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+ MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) {
+ ib_dma_sync_single_for_device(ibqp->device,
+ qp->sqp_proxy_rcv[ind].map,
+ sizeof(struct mlx4_ib_proxy_sqp_hdr),
+ DMA_FROM_DEVICE);
+ scat->byte_count =
+ cpu_to_be32(sizeof(struct mlx4_ib_proxy_sqp_hdr));
+ /* use dma lkey from upper layer entry */
+ scat->lkey = cpu_to_be32(wr->sg_list->lkey);
+ scat->addr = cpu_to_be64(qp->sqp_proxy_rcv[ind].map);
+ scat++;
+ max_gs--;
+ }
+
for (i = 0; i < wr->num_sge; ++i)
__set_data_seg(scat + i, wr->sg_list + i);
- if (i < qp->rq.max_gs) {
+ if (i < max_gs) {
scat[i].byte_count = 0;
scat[i].lkey = cpu_to_be32(MLX4_INVALID_LKEY);
scat[i].addr = 0;
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index bd6c9fc..6faab99 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -389,6 +389,7 @@ struct mlx4_caps {
enum mlx4_port_type possible_type[MLX4_MAX_PORTS + 1];
u32 max_counters;
u8 port_ib_mtu[MLX4_MAX_PORTS + 1];
+ u16 sqp_demux;
};
struct mlx4_buf_list {
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 338388b..4b4ad6f 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -126,7 +126,8 @@ struct mlx4_rss_context {
struct mlx4_qp_path {
u8 fl;
- u8 reserved1[2];
+ u8 reserved1[1];
+ u8 disable_pkey_check;
u8 pkey_index;
u8 counter_index;
u8 grh_mylmc;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 05/22] net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (3 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 06/22] IB/mlx4: Initialize SRIOV IB support for slaves in master context Jack Morgenstein
` (18 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
In addition, pass the proxy and tunnel QP numbers to slaves so
the driver can perform sqp paravirtualization.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/net/ethernet/mellanox/mlx4/fw.c | 14 +++++++++
drivers/net/ethernet/mellanox/mlx4/fw.h | 3 ++
drivers/net/ethernet/mellanox/mlx4/main.c | 5 +++
drivers/net/ethernet/mellanox/mlx4/qp.c | 29 +++++++++++++++++--
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 3 +-
include/linux/mlx4/device.h | 13 ++++++++-
6 files changed, 62 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index c696484..0f557df 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -184,6 +184,8 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
#define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET 0x28
#define QUERY_FUNC_CAP_MAX_EQ_OFFSET 0x2c
#define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0X30
+#define QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET 0X44
+#define QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET 0X48
#define QUERY_FUNC_CAP_FMR_FLAG 0x80
#define QUERY_FUNC_CAP_FLAG_RDMA 0x40
@@ -247,6 +249,12 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
size = dev->caps.num_mgms + dev->caps.num_amgms;
MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
+ size = dev->caps.base_tunnel_sqpn + 8 * slave;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
+
+ size = dev->caps.sqp_start + 8 * slave;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
+
} else
err = -EINVAL;
@@ -312,6 +320,12 @@ int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap)
MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
func_cap->mcg_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
+ func_cap->base_tunnel_qpn = size & 0xFFFFFF;
+
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
+ func_cap->base_proxy_qpn = size & 0xFFFFFF;
+
for (i = 1; i <= func_cap->num_ports; ++i) {
err = mlx4_cmd_box(dev, 0, mailbox->dma, i, 1,
MLX4_CMD_QUERY_FUNC_CAP,
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h
index 83fcbbf..ced1de5 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.h
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.h
@@ -134,6 +134,9 @@ struct mlx4_func_cap {
int max_eq;
int reserved_eq;
int mcg_quota;
+ u32 base_qpn;
+ u32 base_tunnel_qpn;
+ u32 base_proxy_qpn;
u8 physical_port[MLX4_MAX_PORTS + 1];
u8 port_flags[MLX4_MAX_PORTS + 1];
};
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index e8f8ebb..892f2d6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -387,6 +387,7 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
+ dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
return 0;
}
/*The function checks if there are live vf, return the num of them*/
@@ -544,6 +545,10 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
return -ENODEV;
}
+ /* Calculate our sqp_start */
+ dev->caps.sqp_start = func_cap.base_proxy_qpn;
+ dev->caps.base_tunnel_sqpn = func_cap.base_tunnel_qpn;
+
return 0;
}
diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c
index fb2b367..b8da72b 100644
--- a/drivers/net/ethernet/mellanox/mlx4/qp.c
+++ b/drivers/net/ethernet/mellanox/mlx4/qp.c
@@ -406,7 +406,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
* We also reserve the MSB of the 24-bit QP number to indicate
* that a QP is an XRC QP.
*/
- dev->caps.sqp_start =
+ dev->caps.base_sqpn =
ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);
{
@@ -437,13 +437,36 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
}
+ /* Reserve 8 real SQPs in both native and SRIOV modes.
+ * In addition, in SRIOV mode, reserve 8 proxy SQPs per function
+ * (for all PFs and VFs), and 8 corresponding tunnel QPs.
+ * Each proxy SQP works opposite its own tunnel QP.
+ *
+ * The QPs are arranged as follows:
+ * a. 8 real SQPs
+ * b. All the proxy SQPs (8 per function)
+ * c. All the tunnel QPs (8 per function)
+ */
+
err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps,
- (1 << 23) - 1, dev->caps.sqp_start + 8,
+ (1 << 23) - 1, dev->caps.base_sqpn + 8 +
+ 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev),
reserved_from_top);
+
+ /* In mfunc, sqp_start is the base of the proxy SQPs, since the PF also
+ * uses paravirtualized SQPs.
+ * In native mode, sqp_start is the base of the real SQPs. */
+ if (mlx4_is_mfunc(dev)) {
+ dev->caps.sqp_start = dev->caps.base_sqpn +
+ 8 * (mlx4_master_func_num(dev) + 1);
+ dev->caps.base_tunnel_sqpn = dev->caps.sqp_start + 8 * MLX4_MFUNC_MAX;
+ } else
+ dev->caps.sqp_start = dev->caps.base_sqpn;
+
if (err)
return err;
- return mlx4_CONF_SPECIAL_QP(dev, dev->caps.sqp_start);
+ return mlx4_CONF_SPECIAL_QP(dev, dev->caps.base_sqpn);
}
void mlx4_cleanup_qp_table(struct mlx4_dev *dev)
diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
index 94ceddd..0f36ae9 100644
--- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
+++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
@@ -1104,7 +1104,8 @@ static void res_end_move(struct mlx4_dev *dev, int slave,
static int valid_reserved(struct mlx4_dev *dev, int slave, int qpn)
{
- return mlx4_is_qp_reserved(dev, qpn);
+ return mlx4_is_qp_reserved(dev, qpn) &&
+ (mlx4_is_master(dev) || mlx4_is_guest_proxy(dev, slave, qpn));
}
static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 6faab99..457503b 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -693,7 +693,18 @@ static inline int mlx4_is_master(struct mlx4_dev *dev)
static inline int mlx4_is_qp_reserved(struct mlx4_dev *dev, u32 qpn)
{
- return (qpn < dev->caps.sqp_start + 8);
+ return (qpn < dev->caps.base_sqpn + 8 +
+ 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev));
+}
+
+static inline int mlx4_is_guest_proxy(struct mlx4_dev *dev, int slave, u32 qpn)
+{
+ int base = dev->caps.sqp_start + slave * 8;
+
+ if (qpn >= base && qpn < base + 8)
+ return 1;
+
+ return 0;
}
static inline int mlx4_is_mfunc(struct mlx4_dev *dev)
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 06/22] IB/mlx4: Initialize SRIOV IB support for slaves in master context
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (4 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 05/22] net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 07/22] {NET,IB}/mlx4: Implement QP paravirtualization and maintain phys_pkey_cache for smp_snoop Jack Morgenstein
` (17 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Allocate sriov paravirtualization resources and mad demuxing contexts
on the master.
This has two parts. The first part is to initialize the structures
to contain the contexts. This is done at master startup time
(mlx4_ib_init_sriov).
The second part is to actually create the tunneling resources required
on the master to support a slave. This is performed the master detects
that a slave has started up (MLX4_DEV_EVENT_SLAVE_INIT event generated
when a slave initializes its comm channel).
For the master, there is no such startup event, so it creates its own
tunneling resources when it starts up. In addition, the master also
creates the real special QPs. (The ib_core layer on the master causes
creation of proxy special qp's, since the master is also
paravirtualized at the ib_core layer).
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 684 ++++++++++++++++++++++++++++++
drivers/infiniband/hw/mlx4/main.c | 80 ++++-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 34 ++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 3 +
include/linux/mlx4/device.h | 3 +-
include/linux/mlx4/driver.h | 2 +
6 files changed, 798 insertions(+), 8 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index c27141f..d955cc4 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -44,6 +44,35 @@ enum {
MLX4_IB_VENDOR_CLASS2 = 0xa
};
+#define MLX4_TUN_SEND_WRID_SHIFT 34
+#define MLX4_TUN_QPN_SHIFT 32
+#define MLX4_TUN_WRID_RECV (((u64) 1) << MLX4_TUN_SEND_WRID_SHIFT)
+#define MLX4_TUN_SET_WRID_QPN(a) (((u64) ((a) & 0x3)) << MLX4_TUN_QPN_SHIFT)
+
+#define MLX4_TUN_IS_RECV(a) (((a) >> MLX4_TUN_SEND_WRID_SHIFT) & 0x1)
+#define MLX4_TUN_WRID_QPN(a) (((a) >> MLX4_TUN_QPN_SHIFT) & 0x3)
+
+struct mlx4_mad_rcv_buf {
+ struct ib_grh grh;
+ u8 payload[256];
+} __packed;
+
+struct mlx4_mad_snd_buf {
+ u8 payload[256];
+} __packed;
+
+struct mlx4_tunnel_mad {
+ struct ib_grh grh;
+ struct mlx4_ib_tunnel_header hdr;
+ struct ib_mad mad;
+} __packed;
+
+struct mlx4_rcv_tunnel_mad {
+ struct mlx4_rcv_tunnel_hdr hdr;
+ struct ib_grh grh;
+ struct ib_mad mad;
+} __packed;
+
int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
void *in_mad, void *response_mad)
@@ -512,3 +541,658 @@ void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
ib_dispatch_event(&event);
}
+
+static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg)
+{
+ unsigned long flags;
+ struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context;
+ struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ if (!dev->sriov.is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE)
+ queue_work(ctx->wq, &ctx->work);
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
+ struct mlx4_ib_demux_pv_qp *tun_qp,
+ int index)
+{
+ struct ib_sge sg_list;
+ struct ib_recv_wr recv_wr, *bad_recv_wr;
+ int size;
+
+ size = (tun_qp->qp->qp_type == IB_QPT_UD) ?
+ sizeof(struct mlx4_tunnel_mad) : sizeof(struct mlx4_mad_rcv_buf);
+
+ sg_list.addr = tun_qp->ring[index].map;
+ sg_list.length = size;
+ sg_list.lkey = ctx->mr->lkey;
+
+ recv_wr.next = NULL;
+ recv_wr.sg_list = &sg_list;
+ recv_wr.num_sge = 1;
+ recv_wr.wr_id = (u64) index | MLX4_TUN_WRID_RECV |
+ MLX4_TUN_SET_WRID_QPN(tun_qp->proxy_qpt);
+ ib_dma_sync_single_for_device(ctx->ib_dev, tun_qp->ring[index].map,
+ size, DMA_FROM_DEVICE);
+ return ib_post_recv(tun_qp->qp, &recv_wr, &bad_recv_wr);
+}
+
+static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
+ enum ib_qp_type qp_type, int is_tun)
+{
+ int i;
+ struct mlx4_ib_demux_pv_qp *tun_qp;
+ int rx_buf_size, tx_buf_size;
+
+ if (qp_type > IB_QPT_GSI)
+ return -EINVAL;
+
+ tun_qp = &ctx->qp[qp_type];
+
+ tun_qp->ring = kzalloc(sizeof(struct mlx4_ib_buf) * MLX4_NUM_TUNNEL_BUFS,
+ GFP_KERNEL);
+ if (!tun_qp->ring)
+ return -ENOMEM;
+
+ tun_qp->tx_ring = kzalloc(sizeof(struct mlx4_ib_tun_tx_buf) *
+ MLX4_NUM_TUNNEL_BUFS,
+ GFP_KERNEL);
+ if (!tun_qp->tx_ring) {
+ kfree(tun_qp->ring);
+ tun_qp->ring = NULL;
+ return -ENOMEM;
+ }
+
+ if (is_tun) {
+ rx_buf_size = sizeof(struct mlx4_tunnel_mad);
+ tx_buf_size = sizeof(struct mlx4_rcv_tunnel_mad);
+ } else {
+ rx_buf_size = sizeof(struct mlx4_mad_rcv_buf);
+ tx_buf_size = sizeof(struct mlx4_mad_snd_buf);
+ }
+
+ for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+ tun_qp->ring[i].addr = kmalloc(rx_buf_size, GFP_KERNEL);
+ if (!tun_qp->ring[i].addr)
+ goto err;
+ tun_qp->ring[i].map = ib_dma_map_single(ctx->ib_dev,
+ tun_qp->ring[i].addr,
+ rx_buf_size,
+ DMA_FROM_DEVICE);
+ }
+
+ for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+ tun_qp->tx_ring[i].buf.addr =
+ kmalloc(tx_buf_size, GFP_KERNEL);
+ if (!tun_qp->tx_ring[i].buf.addr)
+ goto tx_err;
+ tun_qp->tx_ring[i].buf.map =
+ ib_dma_map_single(ctx->ib_dev,
+ tun_qp->tx_ring[i].buf.addr,
+ tx_buf_size,
+ DMA_TO_DEVICE);
+ tun_qp->tx_ring[i].ah = NULL;
+ }
+ spin_lock_init(&tun_qp->tx_lock);
+ tun_qp->tx_ix_head = 0;
+ tun_qp->tx_ix_tail = 0;
+ tun_qp->proxy_qpt = qp_type;
+
+ return 0;
+
+tx_err:
+ while (i > 0) {
+ --i;
+ ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map,
+ tx_buf_size, DMA_TO_DEVICE);
+ kfree(tun_qp->tx_ring[i].buf.addr);
+ }
+ kfree(tun_qp->tx_ring);
+ tun_qp->tx_ring = NULL;
+ i = MLX4_NUM_TUNNEL_BUFS;
+err:
+ while (i > 0) {
+ --i;
+ ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map,
+ rx_buf_size, DMA_FROM_DEVICE);
+ kfree(tun_qp->ring[i].addr);
+ }
+ kfree(tun_qp->ring);
+ tun_qp->ring = NULL;
+ return -ENOMEM;
+}
+
+static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
+ enum ib_qp_type qp_type, int is_tun)
+{
+ int i;
+ struct mlx4_ib_demux_pv_qp *tun_qp;
+ int rx_buf_size, tx_buf_size;
+
+ if (qp_type > IB_QPT_GSI)
+ return;
+
+ tun_qp = &ctx->qp[qp_type];
+ if (is_tun) {
+ rx_buf_size = sizeof(struct mlx4_tunnel_mad);
+ tx_buf_size = sizeof(struct mlx4_rcv_tunnel_mad);
+ } else {
+ rx_buf_size = sizeof(struct mlx4_mad_rcv_buf);
+ tx_buf_size = sizeof(struct mlx4_mad_snd_buf);
+ }
+
+
+ for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+ ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map,
+ rx_buf_size, DMA_FROM_DEVICE);
+ kfree(tun_qp->ring[i].addr);
+ }
+
+ for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+ ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map,
+ tx_buf_size, DMA_TO_DEVICE);
+ kfree(tun_qp->tx_ring[i].buf.addr);
+ if (tun_qp->tx_ring[i].ah)
+ ib_destroy_ah(tun_qp->tx_ring[i].ah);
+ }
+ kfree(tun_qp->tx_ring);
+ kfree(tun_qp->ring);
+}
+
+static void mlx4_ib_tunnel_comp_worker(struct work_struct *work)
+{
+ /* dummy until next patch in series */
+}
+
+static void pv_qp_event_handler(struct ib_event *event, void *qp_context)
+{
+ struct mlx4_ib_demux_pv_ctx *sqp = qp_context;
+
+ /* It's worse than that! He's dead, Jim! */
+ pr_err("Fatal error (%d) on a MAD QP on port %d\n",
+ event->event, sqp->port);
+}
+
+static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
+ enum ib_qp_type qp_type, int create_tun)
+{
+ int i, ret;
+ struct mlx4_ib_demux_pv_qp *tun_qp;
+ struct mlx4_ib_qp_tunnel_init_attr qp_init_attr;
+ struct ib_qp_attr attr;
+ int qp_attr_mask_INIT;
+
+ if (qp_type > IB_QPT_GSI)
+ return -EINVAL;
+
+ tun_qp = &ctx->qp[qp_type];
+
+ memset(&qp_init_attr, 0, sizeof qp_init_attr);
+ qp_init_attr.init_attr.send_cq = ctx->cq;
+ qp_init_attr.init_attr.recv_cq = ctx->cq;
+ qp_init_attr.init_attr.sq_sig_type = IB_SIGNAL_ALL_WR;
+ qp_init_attr.init_attr.cap.max_send_wr = MLX4_NUM_TUNNEL_BUFS;
+ qp_init_attr.init_attr.cap.max_recv_wr = MLX4_NUM_TUNNEL_BUFS;
+ qp_init_attr.init_attr.cap.max_send_sge = 1;
+ qp_init_attr.init_attr.cap.max_recv_sge = 1;
+ if (create_tun) {
+ qp_init_attr.init_attr.qp_type = IB_QPT_UD;
+ qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_TUNNEL_QP;
+ qp_init_attr.port = ctx->port;
+ qp_init_attr.slave = ctx->slave;
+ qp_init_attr.proxy_qp_type = qp_type;
+ qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX |
+ IB_QP_QKEY | IB_QP_PORT;
+ } else {
+ qp_init_attr.init_attr.qp_type = qp_type;
+ qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_SQP;
+ qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_QKEY;
+ }
+ qp_init_attr.init_attr.port_num = ctx->port;
+ qp_init_attr.init_attr.qp_context = ctx;
+ qp_init_attr.init_attr.event_handler = pv_qp_event_handler;
+ tun_qp->qp = ib_create_qp(ctx->pd, &qp_init_attr.init_attr);
+ if (IS_ERR(tun_qp->qp)) {
+ ret = PTR_ERR(tun_qp->qp);
+ tun_qp->qp = NULL;
+ pr_err("Couldn't create %s QP (%d)\n",
+ create_tun ? "tunnel" : "special", ret);
+ return ret;
+ }
+
+ memset(&attr, 0, sizeof attr);
+ attr.qp_state = IB_QPS_INIT;
+ attr.pkey_index =
+ to_mdev(ctx->ib_dev)->pkeys.virt2phys_pkey[ctx->slave][ctx->port - 1][0];
+ attr.qkey = IB_QP1_QKEY;
+ attr.port_num = ctx->port;
+ ret = ib_modify_qp(tun_qp->qp, &attr, qp_attr_mask_INIT);
+ if (ret) {
+ pr_err("Couldn't change %s qp state to INIT (%d)\n",
+ create_tun ? "tunnel" : "special", ret);
+ goto err_qp;
+ }
+ attr.qp_state = IB_QPS_RTR;
+ ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE);
+ if (ret) {
+ pr_err("Couldn't change %s qp state to RTR (%d)\n",
+ create_tun ? "tunnel" : "special", ret);
+ goto err_qp;
+ }
+ attr.qp_state = IB_QPS_RTS;
+ attr.sq_psn = 0;
+ ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE | IB_QP_SQ_PSN);
+ if (ret) {
+ pr_err("Couldn't change %s qp state to RTS (%d)\n",
+ create_tun ? "tunnel" : "special", ret);
+ goto err_qp;
+ }
+
+ for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+ ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp, i);
+ if (ret) {
+ pr_err(" mlx4_ib_post_pv_buf error"
+ " (err = %d, i = %d)\n", ret, i);
+ goto err_qp;
+ }
+ }
+ return 0;
+
+err_qp:
+ ib_destroy_qp(tun_qp->qp);
+ tun_qp->qp = NULL;
+ return ret;
+}
+
+/*
+ * IB MAD completion callback for real SQPs
+ */
+static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
+{
+ /* dummy until next patch in series */
+}
+
+static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port,
+ struct mlx4_ib_demux_pv_ctx **ret_ctx)
+{
+ struct mlx4_ib_demux_pv_ctx *ctx;
+
+ *ret_ctx = NULL;
+ ctx = kzalloc(sizeof(struct mlx4_ib_demux_pv_ctx), GFP_KERNEL);
+ if (!ctx) {
+ pr_err("failed allocating pv resource context "
+ "for port %d, slave %d\n", port, slave);
+ return -ENOMEM;
+ }
+
+ ctx->ib_dev = &dev->ib_dev;
+ ctx->port = port;
+ ctx->slave = slave;
+ *ret_ctx = ctx;
+ return 0;
+}
+
+static void free_pv_object(struct mlx4_ib_dev *dev, int slave, int port)
+{
+ if (dev->sriov.demux[port - 1].tun[slave]) {
+ kfree(dev->sriov.demux[port - 1].tun[slave]);
+ dev->sriov.demux[port - 1].tun[slave] = NULL;
+ }
+}
+
+static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
+ int create_tun, struct mlx4_ib_demux_pv_ctx *ctx)
+{
+ int ret, cq_size;
+
+ ctx->state = DEMUX_PV_STATE_STARTING;
+ /* have QP0 only on port owner, and only if link layer is IB */
+ if (ctx->slave == mlx4_master_func_num(to_mdev(ctx->ib_dev)->dev) &&
+ rdma_port_get_link_layer(ibdev, ctx->port) == IB_LINK_LAYER_INFINIBAND)
+ ctx->has_smi = 1;
+
+ if (ctx->has_smi) {
+ ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_SMI, create_tun);
+ if (ret) {
+ pr_err("Failed allocating qp0 tunnel bufs (%d)\n", ret);
+ goto err_out;
+ }
+ }
+
+ ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_GSI, create_tun);
+ if (ret) {
+ pr_err("Failed allocating qp1 tunnel bufs (%d)\n", ret);
+ goto err_out_qp0;
+ }
+
+ cq_size = 2 * MLX4_NUM_TUNNEL_BUFS;
+ if (ctx->has_smi)
+ cq_size *= 2;
+
+ ctx->cq = ib_create_cq(ctx->ib_dev, mlx4_ib_tunnel_comp_handler,
+ NULL, ctx, cq_size, 0);
+ if (IS_ERR(ctx->cq)) {
+ ret = PTR_ERR(ctx->cq);
+ pr_err("Couldn't create tunnel CQ (%d)\n", ret);
+ goto err_buf;
+ }
+
+ ctx->pd = ib_alloc_pd(ctx->ib_dev);
+ if (IS_ERR(ctx->pd)) {
+ ret = PTR_ERR(ctx->pd);
+ pr_err("Couldn't create tunnel PD (%d)\n", ret);
+ goto err_cq;
+ }
+
+ ctx->mr = ib_get_dma_mr(ctx->pd, IB_ACCESS_LOCAL_WRITE);
+ if (IS_ERR(ctx->mr)) {
+ ret = PTR_ERR(ctx->mr);
+ pr_err("Couldn't get tunnel DMA MR (%d)\n", ret);
+ goto err_pd;
+ }
+
+ if (ctx->has_smi) {
+ ret = create_pv_sqp(ctx, IB_QPT_SMI, create_tun);
+ if (ret) {
+ pr_err("Couldn't create %s QP0 (%d)\n",
+ create_tun ? "tunnel for" : "", ret);
+ goto err_mr;
+ }
+ }
+
+ ret = create_pv_sqp(ctx, IB_QPT_GSI, create_tun);
+ if (ret) {
+ pr_err("Couldn't create %s QP1 (%d)\n",
+ create_tun ? "tunnel for" : "", ret);
+ goto err_qp0;
+ }
+
+ if (create_tun)
+ INIT_WORK(&ctx->work, mlx4_ib_tunnel_comp_worker);
+ else
+ INIT_WORK(&ctx->work, mlx4_ib_sqp_comp_worker);
+
+ ctx->wq = to_mdev(ibdev)->sriov.demux[port - 1].wq;
+
+ ret = ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+ if (ret) {
+ pr_err("Couldn't arm tunnel cq (%d)\n", ret);
+ goto err_wq;
+ }
+ ctx->state = DEMUX_PV_STATE_ACTIVE;
+ return 0;
+
+err_wq:
+ ctx->wq = NULL;
+ ib_destroy_qp(ctx->qp[1].qp);
+ ctx->qp[1].qp = NULL;
+
+
+err_qp0:
+ if (ctx->has_smi)
+ ib_destroy_qp(ctx->qp[0].qp);
+ ctx->qp[0].qp = NULL;
+
+err_mr:
+ ib_dereg_mr(ctx->mr);
+ ctx->mr = NULL;
+
+err_pd:
+ ib_dealloc_pd(ctx->pd);
+ ctx->pd = NULL;
+
+err_cq:
+ ib_destroy_cq(ctx->cq);
+ ctx->cq = NULL;
+
+err_buf:
+ mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, create_tun);
+
+err_out_qp0:
+ if (ctx->has_smi)
+ mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, create_tun);
+err_out:
+ ctx->state = DEMUX_PV_STATE_DOWN;
+ return ret;
+}
+
+static void destroy_pv_resources(struct mlx4_ib_dev *dev, int slave, int port,
+ struct mlx4_ib_demux_pv_ctx *ctx, int flush)
+{
+ if (!ctx)
+ return;
+ if (ctx->state > DEMUX_PV_STATE_DOWN) {
+ ctx->state = DEMUX_PV_STATE_DOWNING;
+ if (flush)
+ flush_workqueue(ctx->wq);
+ if (ctx->has_smi) {
+ ib_destroy_qp(ctx->qp[0].qp);
+ ctx->qp[0].qp = NULL;
+ mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, 1);
+ }
+ ib_destroy_qp(ctx->qp[1].qp);
+ ctx->qp[1].qp = NULL;
+ mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, 1);
+ ib_dereg_mr(ctx->mr);
+ ctx->mr = NULL;
+ ib_dealloc_pd(ctx->pd);
+ ctx->pd = NULL;
+ ib_destroy_cq(ctx->cq);
+ ctx->cq = NULL;
+ ctx->state = DEMUX_PV_STATE_DOWN;
+ }
+}
+
+static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave,
+ int port, int do_init)
+{
+ int ret = 0;
+
+ if (!do_init) {
+ /* for master, destroy real sqp resources */
+ if (slave == mlx4_master_func_num(dev->dev))
+ destroy_pv_resources(dev, slave, port,
+ dev->sriov.sqps[port - 1], 1);
+ /* destroy the tunnel qp resources */
+ destroy_pv_resources(dev, slave, port,
+ dev->sriov.demux[port - 1].tun[slave], 1);
+ return 0;
+ }
+
+ /* create the tunnel qp resources */
+ ret = create_pv_resources(&dev->ib_dev, slave, port, 1,
+ dev->sriov.demux[port - 1].tun[slave]);
+
+ /* for master, create the real sqp resources */
+ if (!ret && slave == mlx4_master_func_num(dev->dev))
+ ret = create_pv_resources(&dev->ib_dev, slave, port, 0,
+ dev->sriov.sqps[port - 1]);
+ return ret;
+}
+
+void mlx4_ib_tunnels_update_work(struct work_struct *work)
+{
+ struct mlx4_ib_demux_work *dmxw;
+
+ dmxw = container_of(work, struct mlx4_ib_demux_work, work);
+ mlx4_ib_tunnels_update(dmxw->dev, dmxw->slave, (int) dmxw->port,
+ dmxw->do_init);
+ kfree(dmxw);
+ return;
+}
+
+static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
+ struct mlx4_ib_demux_ctx *ctx,
+ int port)
+{
+ char name[12];
+ int ret = 0;
+ int i;
+
+ ctx->tun = kzalloc(dev->dev->caps.sqp_demux *
+ sizeof(struct mlx4_ib_demux_pv_ctx *), GFP_KERNEL);
+ if (!ctx->tun)
+ return -ENOMEM;
+
+ ctx->dev = dev;
+ ctx->port = port;
+ ctx->ib_dev = &dev->ib_dev;
+
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+ ret = alloc_pv_object(dev, i, port, &ctx->tun[i]);
+ if (ret) {
+ ret = -ENOMEM;
+ goto err_wq;
+ }
+ }
+
+ snprintf(name, sizeof name, "mlx4_ibt%d", port);
+ ctx->wq = create_singlethread_workqueue(name);
+ if (!ctx->wq) {
+ pr_err("Failed to create tunnelling WQ for port %d\n", port);
+ ret = -ENOMEM;
+ goto err_wq;
+ }
+
+ snprintf(name, sizeof name, "mlx4_ibud%d", port);
+ ctx->ud_wq = create_singlethread_workqueue(name);
+ if (!ctx->ud_wq) {
+ pr_err("Failed to create up/down WQ for port %d\n", port);
+ ret = -ENOMEM;
+ goto err_udwq;
+ }
+
+ return 0;
+
+err_udwq:
+ destroy_workqueue(ctx->wq);
+ ctx->wq = NULL;
+
+err_wq:
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++)
+ free_pv_object(dev, i, port);
+ kfree(ctx->tun);
+ ctx->tun = NULL;
+ return ret;
+}
+
+static void mlx4_ib_free_sqp_ctx(struct mlx4_ib_demux_pv_ctx *sqp_ctx)
+{
+ if (sqp_ctx->state > DEMUX_PV_STATE_DOWN) {
+ sqp_ctx->state = DEMUX_PV_STATE_DOWNING;
+ flush_workqueue(sqp_ctx->wq);
+ if (sqp_ctx->has_smi) {
+ ib_destroy_qp(sqp_ctx->qp[0].qp);
+ sqp_ctx->qp[0].qp = NULL;
+ mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_SMI, 0);
+ }
+ ib_destroy_qp(sqp_ctx->qp[1].qp);
+ sqp_ctx->qp[1].qp = NULL;
+ mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_GSI, 0);
+ ib_dereg_mr(sqp_ctx->mr);
+ sqp_ctx->mr = NULL;
+ ib_dealloc_pd(sqp_ctx->pd);
+ sqp_ctx->pd = NULL;
+ ib_destroy_cq(sqp_ctx->cq);
+ sqp_ctx->cq = NULL;
+ sqp_ctx->state = DEMUX_PV_STATE_DOWN;
+ }
+}
+
+static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx)
+{
+ int i;
+ if (ctx) {
+ struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+ if (!ctx->tun[i])
+ continue;
+ if (ctx->tun[i]->state > DEMUX_PV_STATE_DOWN)
+ ctx->tun[i]->state = DEMUX_PV_STATE_DOWNING;
+ }
+ flush_workqueue(ctx->wq);
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+ destroy_pv_resources(dev, i, ctx->port, ctx->tun[i], 0);
+ free_pv_object(dev, i, ctx->port);
+ }
+ kfree(ctx->tun);
+ destroy_workqueue(ctx->ud_wq);
+ destroy_workqueue(ctx->wq);
+ }
+}
+
+static void mlx4_ib_master_tunnels(struct mlx4_ib_dev *dev, int do_init)
+{
+ int i;
+
+ if (!mlx4_is_master(dev->dev))
+ return;
+ /* initialize or tear down tunnel QPs for the master */
+ for (i = 0; i < dev->dev->caps.num_ports; i++)
+ mlx4_ib_tunnels_update(dev, mlx4_master_func_num(dev->dev), i + 1, do_init);
+ return;
+}
+
+int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
+{
+ int i = 0;
+ int err;
+
+ if (!mlx4_is_mfunc(dev->dev))
+ return 0;
+
+ dev->sriov.is_going_down = 0;
+ spin_lock_init(&dev->sriov.going_down_lock);
+
+ mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n");
+
+ if (mlx4_is_slave(dev->dev)) {
+ mlx4_ib_warn(&dev->ib_dev, "operating in qp1 tunnel mode\n");
+ return 0;
+ }
+
+ mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n",
+ dev->dev->caps.sqp_demux);
+ for (i = 0; i < dev->num_ports; i++) {
+ err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1,
+ &dev->sriov.sqps[i]);
+ if (err)
+ goto demux_err;
+ err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1);
+ if (err)
+ goto demux_err;
+ }
+ mlx4_ib_master_tunnels(dev, 1);
+ return 0;
+
+demux_err:
+ while (i > 0) {
+ free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
+ mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+ --i;
+ }
+
+ return err;
+}
+
+void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
+{
+ int i;
+ unsigned long flags;
+
+ if (!mlx4_is_mfunc(dev->dev))
+ return;
+
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ dev->sriov.is_going_down = 1;
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ if (mlx4_is_master(dev->dev))
+ for (i = 0; i < dev->num_ports; i++) {
+ flush_workqueue(dev->sriov.demux[i].ud_wq);
+ mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]);
+ kfree(dev->sriov.sqps[i]);
+ dev->sriov.sqps[i] = NULL;
+ mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+ }
+}
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fe2088c..075561f 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1356,11 +1356,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
if (mlx4_ib_mad_init(ibdev))
goto err_reg;
+ if (mlx4_ib_init_sriov(ibdev))
+ goto err_mad;
+
if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
iboe->nb.notifier_call = mlx4_ib_netdev_event;
err = register_netdevice_notifier(&iboe->nb);
if (err)
- goto err_reg;
+ goto err_sriov;
}
for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
@@ -1378,6 +1381,12 @@ err_notif:
pr_warn("failure unregistering notifier\n");
flush_workqueue(wq);
+err_sriov:
+ mlx4_ib_close_sriov(ibdev);
+
+err_mad:
+ mlx4_ib_mad_cleanup(ibdev);
+
err_reg:
ib_unregister_device(&ibdev->ib_dev);
@@ -1406,6 +1415,7 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
struct mlx4_ib_dev *ibdev = ibdev_ptr;
int p;
+ mlx4_ib_close_sriov(ibdev);
mlx4_ib_mad_cleanup(ibdev);
ib_unregister_device(&ibdev->ib_dev);
if (ibdev->iboe.nb.notifier_call) {
@@ -1427,6 +1437,51 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
ib_dealloc_device(&ibdev->ib_dev);
}
+static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
+{
+ struct mlx4_ib_demux_work **dm = NULL;
+ struct mlx4_dev *dev = ibdev->dev;
+ int i;
+ unsigned long flags;
+
+ if (!mlx4_is_master(dev))
+ return;
+
+ dm = kzalloc(sizeof(*dm) * dev->caps.num_ports, GFP_ATOMIC);
+ if (!dm) {
+ pr_err("failed to allocate memory for tunneling qp update\n");
+ goto out;
+ }
+
+ for (i = 0; i < dev->caps.num_ports; i++) {
+ dm[i] = kmalloc(sizeof(struct mlx4_ib_demux_work), GFP_ATOMIC);
+ if (!dm[i]) {
+ pr_err("failed to allocate memory for tunneling qp update work struct\n");
+ for (i = 0; i < dev->caps.num_ports; i++) {
+ if (dm[i])
+ kfree(dm[i]);
+ }
+ goto out;
+ }
+ }
+ /* initialize or tear down tunnel QPs for the slave */
+ for (i = 0; i < dev->caps.num_ports; i++) {
+ INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
+ dm[i]->port = i + 1;
+ dm[i]->slave = slave;
+ dm[i]->do_init = do_init;
+ dm[i]->dev = ibdev;
+ spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
+ if (!ibdev->sriov.is_going_down)
+ queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
+ spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
+ }
+out:
+ if (dm)
+ kfree(dm);
+ return;
+}
+
static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
enum mlx4_dev_event event, unsigned long param)
{
@@ -1434,22 +1489,23 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
struct mlx4_eqe *eqe = NULL;
struct ib_event_work *ew;
- int port = 0;
+ int p = 0;
if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
eqe = (struct mlx4_eqe *)param;
else
- port = (u8)param;
-
- if (port > ibdev->num_ports)
- return;
+ p = (int) param;
switch (event) {
case MLX4_DEV_EVENT_PORT_UP:
+ if (p > ibdev->num_ports)
+ return;
ibev.event = IB_EVENT_PORT_ACTIVE;
break;
case MLX4_DEV_EVENT_PORT_DOWN:
+ if (p > ibdev->num_ports)
+ return;
ibev.event = IB_EVENT_PORT_ERR;
break;
@@ -1471,12 +1527,22 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
handle_port_mgmt_change_event(&ew->work);
return;
+ case MLX4_DEV_EVENT_SLAVE_INIT:
+ /* here, p is the slave id */
+ do_slave_init(ibdev, p, 1);
+ return;
+
+ case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
+ /* here, p is the slave id */
+ do_slave_init(ibdev, p, 0);
+ return;
+
default:
return;
}
ibev.device = ibdev_ptr;
- ibev.element.port_num = port;
+ ibev.element.port_num = (u8) p;
ib_dispatch_event(&ibev);
}
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 1248d57..137941d 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -176,6 +176,10 @@ enum mlx4_ib_qp_type {
MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \
MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI)
+enum {
+ MLX4_NUM_TUNNEL_BUFS = 256,
+};
+
struct mlx4_ib_tunnel_header {
struct mlx4_av av;
__be32 remote_qpn;
@@ -263,6 +267,15 @@ struct mlx4_ib_ah {
union mlx4_ext_av av;
};
+struct mlx4_ib_demux_work {
+ struct work_struct work;
+ struct mlx4_ib_dev *dev;
+ int slave;
+ int do_init;
+ u8 port;
+
+};
+
struct mlx4_ib_tun_tx_buf {
struct mlx4_ib_buf buf;
struct ib_ah *ah;
@@ -278,9 +291,17 @@ struct mlx4_ib_demux_pv_qp {
unsigned tx_ix_tail;
};
+enum mlx4_ib_demux_pv_state {
+ DEMUX_PV_STATE_DOWN,
+ DEMUX_PV_STATE_STARTING,
+ DEMUX_PV_STATE_ACTIVE,
+ DEMUX_PV_STATE_DOWNING,
+};
+
struct mlx4_ib_demux_pv_ctx {
int port;
int slave;
+ enum mlx4_ib_demux_pv_state state;
int has_smi;
struct ib_device *ib_dev;
struct ib_cq *cq;
@@ -319,6 +340,13 @@ struct mlx4_ib_iboe {
union ib_gid gid_table[MLX4_MAX_PORTS][128];
};
+struct pkey_mgt {
+ u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+ u16 phys_pkey_cache[MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+ struct list_head pkey_port_list[MLX4_MFUNC_MAX];
+ struct kobject *device_parent[MLX4_MFUNC_MAX];
+};
+
struct mlx4_ib_dev {
struct ib_device ib_dev;
struct mlx4_dev *dev;
@@ -340,6 +368,7 @@ struct mlx4_ib_dev {
int counters[MLX4_MAX_PORTS];
int *eq_table;
int eq_added;
+ struct pkey_mgt pkeys;
};
struct ib_event_work {
@@ -424,6 +453,9 @@ static inline struct mlx4_ib_ah *to_mah(struct ib_ah *ibah)
return container_of(ibah, struct mlx4_ib_ah, ibah);
}
+int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev);
+void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev);
+
int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
struct mlx4_db *db);
void mlx4_ib_db_unmap_user(struct mlx4_ib_ucontext *context, struct mlx4_db *db);
@@ -515,4 +547,6 @@ int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
enum ib_event_type type);
+void mlx4_ib_tunnels_update_work(struct work_struct *work);
+
#endif /* MLX4_IB_H */
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 7e94987..1c2fc72 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -1295,6 +1295,8 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
if (MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd)
goto inform_slave_state;
+ mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN, slave);
+
/* write the version in the event field */
reply |= mlx4_comm_get_version();
@@ -1331,6 +1333,7 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
goto reset_slave;
slave_state[slave].vhcr_dma |= param;
slave_state[slave].active = true;
+ mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_INIT, slave);
break;
case MLX4_COMM_CMD_VHCR_POST:
if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) &&
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 457503b..d8004f2 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -54,7 +54,8 @@ enum {
};
enum {
- MLX4_MAX_PORTS = 2
+ MLX4_MAX_PORTS = 2,
+ MLX4_MAX_PORT_PKEYS = 128
};
/* base qkey for use in sriov tunnel-qp/proxy-qp communication.
diff --git a/include/linux/mlx4/driver.h b/include/linux/mlx4/driver.h
index d813704..c257e1b 100644
--- a/include/linux/mlx4/driver.h
+++ b/include/linux/mlx4/driver.h
@@ -45,6 +45,8 @@ enum mlx4_dev_event {
MLX4_DEV_EVENT_PORT_DOWN,
MLX4_DEV_EVENT_PORT_REINIT,
MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
+ MLX4_DEV_EVENT_SLAVE_INIT,
+ MLX4_DEV_EVENT_SLAVE_SHUTDOWN,
};
struct mlx4_interface {
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 07/22] {NET,IB}/mlx4: Implement QP paravirtualization and maintain phys_pkey_cache for smp_snoop
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (5 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 06/22] IB/mlx4: Initialize SRIOV IB support for slaves in master context Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 08/22] IB/mlx4: SRIOV multiplex and demultiplex MADs Jack Morgenstein
` (16 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
The requires:
1. Replacing the paravirtualized pkey index (inserted
by the guest) with the real pkey index
2. For UD qp's, placing the guest's true source gid index
in the address path structure mgid field, and setting
the ud_force_mgid bit so that the mgid is taken from
the qp context and not from the WQE when posting sends.
3. For UC and RC QPs, placing the guest's true source gid index
in the address path structure mgid field.
4. For tunnel and proxy QPs, setting the QKEY value reserved for
that proxy/tunnel pair.
Since not all the above adjustments occur in all the QP transitions,
the QP transitions require separate wrapper functions.
Secondly, initialize the pkey virtualization table to its default
values: Master virtualized table is 1-1 with the real pkey table,
guest virtualized table has pkey index 0 mapped to the real
pkey index 0, and all the other pkey indices mapped to the
reserved (invalid) pkey at index 127.
Finally, Add logic in smp_snoop for maintaining the phys_pkey_cache.
and generating events on the master only if a pkey actually changed.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 33 +++-
drivers/infiniband/hw/mlx4/main.c | 35 +++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 12 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 11 +
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 47 +++++
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 217 +++++++++++++++++++-
include/linux/mlx4/device.h | 3 +
7 files changed, 344 insertions(+), 14 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index d955cc4..2413a08 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -184,6 +184,10 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
{
struct ib_port_info *pinfo;
u16 lid;
+ __be16 *base;
+ u32 bn, pkey_change_bitmap;
+ int i;
+
struct mlx4_ib_dev *dev = to_mdev(ibdev);
if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
@@ -208,8 +212,33 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
break;
case IB_SMP_ATTR_PKEY_TABLE:
- mlx4_ib_dispatch_event(dev, port_num,
- IB_EVENT_PKEY_CHANGE);
+ if (!mlx4_is_mfunc(dev->dev)) {
+ mlx4_ib_dispatch_event(dev, port_num,
+ IB_EVENT_PKEY_CHANGE);
+ break;
+ }
+
+ bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod) & 0xFFFF;
+ base = (__be16 *) &(((struct ib_smp *)mad)->data[0]);
+ pkey_change_bitmap = 0;
+ for (i = 0; i < 32; i++) {
+ pr_debug("PKEY[%d] = x%x\n",
+ i + bn*32, be16_to_cpu(base[i]));
+ if (be16_to_cpu(base[i]) !=
+ dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32]) {
+ pkey_change_bitmap |= (1 << i);
+ dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32] =
+ be16_to_cpu(base[i]);
+ }
+ }
+ pr_debug("PKEY Change event: port=%d, "
+ "block=0x%x, change_bitmap=0x%x\n",
+ port_num, bn, pkey_change_bitmap);
+
+ if (pkey_change_bitmap)
+ mlx4_ib_dispatch_event(dev, port_num,
+ IB_EVENT_PKEY_CHANGE);
+
break;
case IB_SMP_ATTR_GUID_INFO:
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 075561f..bdfd3ae 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1120,6 +1120,38 @@ static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event
return NOTIFY_DONE;
}
+static void init_pkeys(struct mlx4_ib_dev *ibdev)
+{
+ int port;
+ int slave;
+ int i;
+
+ if (mlx4_is_master(ibdev->dev)) {
+ for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
+ for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
+ for (i = 0;
+ i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
+ ++i) {
+ ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
+ /* master has the identity virt2phys pkey mapping */
+ (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
+ ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
+ mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
+ ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
+ }
+ }
+ }
+ /* initialize pkey cache */
+ for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
+ for (i = 0;
+ i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
+ ++i)
+ ibdev->pkeys.phys_pkey_cache[port-1][i] =
+ (i) ? 0 : 0xFFFF;
+ }
+ }
+}
+
static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
{
char name[32];
@@ -1374,6 +1406,9 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
ibdev->ib_active = true;
+ if (mlx4_is_mfunc(ibdev->dev))
+ init_pkeys(ibdev);
+
return ibdev;
err_notif:
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 1c2fc72..89feaed 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -905,7 +905,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_INIT2INIT_QP_wrapper
},
{
.opcode = MLX4_CMD_INIT2RTR_QP,
@@ -923,7 +923,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_RTR2RTS_QP_wrapper
},
{
.opcode = MLX4_CMD_RTS2RTS_QP,
@@ -932,7 +932,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_RTS2RTS_QP_wrapper
},
{
.opcode = MLX4_CMD_SQERR2RTS_QP,
@@ -941,7 +941,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_SQERR2RTS_QP_wrapper
},
{
.opcode = MLX4_CMD_2ERR_QP,
@@ -968,7 +968,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_SQD2SQD_QP_wrapper
},
{
.opcode = MLX4_CMD_SQD2RTS_QP,
@@ -977,7 +977,7 @@ static struct mlx4_cmd_info cmd_info[] = {
.out_is_imm = false,
.encode_slave_id = false,
.verify = NULL,
- .wrapper = mlx4_GEN_QP_wrapper
+ .wrapper = mlx4_SQD2RTS_QP_wrapper
},
{
.opcode = MLX4_CMD_2RST_QP,
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index 892f2d6..cd82607 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -427,6 +427,17 @@ int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
}
EXPORT_SYMBOL(mlx4_get_parav_qkey);
+void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
+{
+ struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
+
+ if (!mlx4_is_master(dev))
+ return;
+
+ priv->virt2phys_pkey[slave][port - 1][i] = val;
+}
+EXPORT_SYMBOL(mlx4_sync_pkey_table);
+
int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
{
struct mlx4_priv *priv = mlx4_priv(dev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 59ebc03..cd77a74 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -731,6 +731,8 @@ struct mlx4_priv {
struct io_mapping *bf_mapping;
int reserved_mtts;
int fs_hash_mode;
+ u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+
};
static inline struct mlx4_priv *mlx4_priv(struct mlx4_dev *dev)
@@ -935,16 +937,61 @@ int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_cmd_mailbox *inbox,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd);
+int mlx4_INIT2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
int mlx4_INIT2RTR_QP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd);
+int mlx4_RTR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_RTS2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_SQERR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_2ERR_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_RTS2SQD_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_SQD2SQD_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
+int mlx4_SQD2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
int mlx4_2RST_QP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd);
+int mlx4_QUERY_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
int mlx4_GEN_EQE(struct mlx4_dev *dev, int slave, struct mlx4_eqe *eqe);
diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
index 0f36ae9..27fb4e5 100644
--- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
+++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
@@ -241,6 +241,15 @@ static int res_tracker_insert(struct rb_root *root, struct res_common *res)
return 0;
}
+enum qp_transition {
+ QP_TRANS_INIT2RTR,
+ QP_TRANS_RTR2RTS,
+ QP_TRANS_RTS2RTS,
+ QP_TRANS_SQERR2RTS,
+ QP_TRANS_SQD2SQD,
+ QP_TRANS_SQD2RTS
+};
+
/* For Debug uses */
static const char *ResourceType(enum mlx4_resource rt)
{
@@ -307,14 +316,41 @@ void mlx4_free_resource_tracker(struct mlx4_dev *dev,
}
}
-static void update_ud_gid(struct mlx4_dev *dev,
- struct mlx4_qp_context *qp_ctx, u8 slave)
+static void update_pkey_index(struct mlx4_dev *dev, int slave,
+ struct mlx4_cmd_mailbox *inbox)
{
- u32 ts = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff;
+ u8 sched = *(u8 *)(inbox->buf + 64);
+ u8 orig_index = *(u8 *)(inbox->buf + 35);
+ u8 new_index;
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ int port;
+
+ port = (sched >> 6 & 1) + 1;
+
+ new_index = priv->virt2phys_pkey[slave][port - 1][orig_index];
+ *(u8 *)(inbox->buf + 35) = new_index;
+
+ mlx4_dbg(dev, "port = %d, orig pkey index = %d, "
+ "new pkey index = %d\n", port, orig_index, new_index);
+}
+
+static void update_gid(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *inbox,
+ u8 slave)
+{
+ struct mlx4_qp_context *qp_ctx = inbox->buf + 8;
+ enum mlx4_qp_optpar optpar = be32_to_cpu(*(__be32 *) inbox->buf);
+ u32 ts = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff;
if (MLX4_QP_ST_UD == ts)
qp_ctx->pri_path.mgid_index = 0x80 | slave;
+ if (MLX4_QP_ST_RC == ts || MLX4_QP_ST_UC == ts) {
+ if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH)
+ qp_ctx->pri_path.mgid_index = slave & 0x7F;
+ if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH)
+ qp_ctx->alt_path.mgid_index = slave & 0x7F;
+ }
+
mlx4_dbg(dev, "slave %d, new gid index: 0x%x ",
slave, qp_ctx->pri_path.mgid_index);
}
@@ -1108,6 +1144,11 @@ static int valid_reserved(struct mlx4_dev *dev, int slave, int qpn)
(mlx4_is_master(dev) || mlx4_is_guest_proxy(dev, slave, qpn));
}
+static int fw_reserved(struct mlx4_dev *dev, int qpn)
+{
+ return qpn < dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
+}
+
static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
u64 in_param, u64 *out_param)
{
@@ -1145,7 +1186,7 @@ static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
if (err)
return err;
- if (!valid_reserved(dev, slave, qpn)) {
+ if (!fw_reserved(dev, qpn)) {
err = __mlx4_qp_alloc_icm(dev, qpn);
if (err) {
res_abort_move(dev, slave, RES_QP, qpn);
@@ -1498,7 +1539,7 @@ static int qp_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
if (err)
return err;
- if (!valid_reserved(dev, slave, qpn))
+ if (!fw_reserved(dev, qpn))
__mlx4_qp_free_icm(dev, qpn);
res_end_move(dev, slave, RES_QP, qpn);
@@ -1938,6 +1979,19 @@ static u32 qp_get_srqn(struct mlx4_qp_context *qpc)
return be32_to_cpu(qpc->srqn) & 0x1ffffff;
}
+static void adjust_proxy_tun_qkey(struct mlx4_dev *dev, struct mlx4_vhcr *vhcr,
+ struct mlx4_qp_context *context)
+{
+ u32 qpn = vhcr->in_modifier & 0xffffff;
+ u32 qkey = 0;
+
+ if (mlx4_get_parav_qkey(dev, qpn, &qkey))
+ return;
+
+ /* adjust qkey in qp context */
+ context->qkey = cpu_to_be32(qkey);
+}
+
int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
@@ -1990,6 +2044,8 @@ int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
goto ex_put_scq;
}
+ adjust_proxy_tun_qkey(dev, vhcr, qpc);
+ update_pkey_index(dev, slave, inbox);
err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
if (err)
goto ex_put_srq;
@@ -2135,6 +2191,48 @@ static int get_containing_mtt(struct mlx4_dev *dev, int slave, int start,
return err;
}
+static int verify_qp_parameters(struct mlx4_dev *dev,
+ struct mlx4_cmd_mailbox *inbox,
+ enum qp_transition transition, u8 slave)
+{
+ u32 qp_type;
+ struct mlx4_qp_context *qp_ctx;
+ enum mlx4_qp_optpar optpar;
+
+ qp_ctx = inbox->buf + 8;
+ qp_type = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff;
+ optpar = be32_to_cpu(*(__be32 *) inbox->buf);
+
+ switch (qp_type) {
+ case MLX4_QP_ST_RC:
+ case MLX4_QP_ST_UC:
+ switch (transition) {
+ case QP_TRANS_INIT2RTR:
+ case QP_TRANS_RTR2RTS:
+ case QP_TRANS_RTS2RTS:
+ case QP_TRANS_SQD2SQD:
+ case QP_TRANS_SQD2RTS:
+ if (slave != mlx4_master_func_num(dev))
+ /* slaves have only gid index 0 */
+ if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH)
+ if (qp_ctx->pri_path.mgid_index)
+ return -EINVAL;
+ if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH)
+ if (qp_ctx->alt_path.mgid_index)
+ return -EINVAL;
+ break;
+ default:
+ break;
+ }
+
+ break;
+ default:
+ break;
+ }
+
+ return 0;
+}
+
int mlx4_WRITE_MTT_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
@@ -2622,16 +2720,123 @@ out:
return err;
}
+int mlx4_INIT2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ struct mlx4_qp_context *context = inbox->buf + 8;
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ update_pkey_index(dev, slave, inbox);
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
int mlx4_INIT2RTR_QP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd)
{
+ int err;
struct mlx4_qp_context *qpc = inbox->buf + 8;
- update_ud_gid(dev, qpc, (u8)slave);
+ err = verify_qp_parameters(dev, inbox, QP_TRANS_INIT2RTR, slave);
+ if (err)
+ return err;
+
+ update_pkey_index(dev, slave, inbox);
+ update_gid(dev, inbox, (u8)slave);
+ adjust_proxy_tun_qkey(dev, vhcr, qpc);
+
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
+int mlx4_RTR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ int err;
+ struct mlx4_qp_context *context = inbox->buf + 8;
+
+ err = verify_qp_parameters(dev, inbox, QP_TRANS_RTR2RTS, slave);
+ if (err)
+ return err;
+
+ update_pkey_index(dev, slave, inbox);
+ update_gid(dev, inbox, (u8)slave);
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
+int mlx4_RTS2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ int err;
+ struct mlx4_qp_context *context = inbox->buf + 8;
+
+ err = verify_qp_parameters(dev, inbox, QP_TRANS_RTS2RTS, slave);
+ if (err)
+ return err;
+
+ update_pkey_index(dev, slave, inbox);
+ update_gid(dev, inbox, (u8)slave);
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
+
+int mlx4_SQERR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ struct mlx4_qp_context *context = inbox->buf + 8;
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
+int mlx4_SQD2SQD_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ int err;
+ struct mlx4_qp_context *context = inbox->buf + 8;
+
+ err = verify_qp_parameters(dev, inbox, QP_TRANS_SQD2SQD, slave);
+ if (err)
+ return err;
+
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ update_gid(dev, inbox, (u8)slave);
+ update_pkey_index(dev, slave, inbox);
+ return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
+}
+
+int mlx4_SQD2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ int err;
+ struct mlx4_qp_context *context = inbox->buf + 8;
+
+ err = verify_qp_parameters(dev, inbox, QP_TRANS_SQD2RTS, slave);
+ if (err)
+ return err;
+ adjust_proxy_tun_qkey(dev, vhcr, context);
+ update_gid(dev, inbox, (u8)slave);
+ update_pkey_index(dev, slave, inbox);
return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
}
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index d8004f2..8e2f03f 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -927,6 +927,9 @@ int mlx4_flow_attach(struct mlx4_dev *dev,
struct mlx4_net_trans_rule *rule, u64 *reg_id);
int mlx4_flow_detach(struct mlx4_dev *dev, u64 reg_id);
+void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port,
+ int i, int val);
+
int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey);
#endif /* MLX4_DEVICE_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 08/22] IB/mlx4: SRIOV multiplex and demultiplex MADs
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (6 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 07/22] {NET,IB}/mlx4: Implement QP paravirtualization and maintain phys_pkey_cache for smp_snoop Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 09/22] {NET,IB}/mlx4: MAD_IFC paravirtualization Jack Morgenstein
` (15 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Special QPs are para-virtualized.
vHCAs are not given direct access to QP0/1. Rather, these QPs are operated by a
special context hosted by the PF, which mediates access to/from vHCAs.
This is done by opening a “tunnel” per vHCA port per QP0/1. A tunnel comprises
a pair of UD QPs: a “Tunnel QP” in the PF-context and a “Proxy QP” in the vHCA.
All vHCA MAD traffic must pass through the corresponding tunnel.
vHCA QPs cannot be assigned to VL15 and are denied of the well-known QKey.
Outgoing messages are "de-multiplexed" (i.e., directed to the wire via
the real special QP).
Incoming messages are "multiplexed" (i.e. steered by the PPF to the correct
VF or to the PF)
QP0 access is restricted to the PF vHCA. VF vHCAs also have (virtual) QP0’s,
but they never receive any SMPs and all SMPs sent are discarded.
QP1 traffic is allowed for all vHCAs, but special care is required to bridge
the gap between the host and network views.
Specifically:
- Transaction IDs are mapped to guarantee uniqueness among vHCAs
- CM para-virtualization
o Incoming requests are steered to the correct vHCA according to the embedded GID
o Local communication IDs are mapped to ensure uniqueness among vHCAs
(see the patch that adds CM paravirtualization.)
- Multicast para-virtualization
o The PF context aggregates membership state from all vHCAs
o The SA is contacted only when the aggregate membership changes
o If the aggregate does not change, the PF context will provide the
requesting vHCA with the proper response
(see the patch that adds multicast group paravirtualization.)
Incoming MADs are steered according to:
- the DGID If a GRH is present
- the mapped transaction ID for response MADs
- the embedded GID in CM requests
- the remote communication ID in other CM messages
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 567 +++++++++++++++++++++++++++++++++++++-
1 files changed, 565 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 2413a08..a392a5c 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -32,6 +32,8 @@
#include <rdma/ib_mad.h>
#include <rdma/ib_smi.h>
+#include <rdma/ib_sa.h>
+#include <rdma/ib_cache.h>
#include <linux/mlx4/cmd.h>
#include <linux/gfp.h>
@@ -296,6 +298,254 @@ static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *ma
}
}
+static int mlx4_ib_demux_sa_handler(struct ib_device *ibdev, int port, int slave,
+ struct ib_sa_mad *sa_mad)
+{
+ return 0;
+}
+
+int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ int i;
+
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+ if (dev->sriov.demux[port - 1].guid_cache[i] == guid)
+ return i;
+ }
+ return -1;
+}
+
+
+static int get_pkey_phys_indices(struct mlx4_ib_dev *ibdev, u8 port, u8 ph_pkey_ix,
+ u8 *full_pk_ix, u8 *partial_pk_ix,
+ int *is_full_member)
+{
+ u16 search_pkey;
+ int fm;
+ int err = 0;
+ u16 pk;
+
+ err = ib_get_cached_pkey(&ibdev->ib_dev, port, ph_pkey_ix, &search_pkey);
+ if (err)
+ return err;
+
+ fm = (search_pkey & 0x8000) ? 1 : 0;
+ if (fm) {
+ *full_pk_ix = ph_pkey_ix;
+ search_pkey &= 0x7FFF;
+ } else {
+ *partial_pk_ix = ph_pkey_ix;
+ search_pkey |= 0x8000;
+ }
+
+ if (ib_find_exact_cached_pkey(&ibdev->ib_dev, port, search_pkey, &pk))
+ pk = 0xFFFF;
+
+ if (fm)
+ *partial_pk_ix = (pk & 0xFF);
+ else
+ *full_pk_ix = (pk & 0xFF);
+
+ *is_full_member = fm;
+ return err;
+}
+
+int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
+ enum ib_qp_type dest_qpt, struct ib_wc *wc,
+ struct ib_grh *grh, struct ib_mad *mad)
+{
+ struct ib_sge list;
+ struct ib_send_wr wr, *bad_wr;
+ struct mlx4_ib_demux_pv_ctx *tun_ctx;
+ struct mlx4_ib_demux_pv_qp *tun_qp;
+ struct mlx4_rcv_tunnel_mad *tun_mad;
+ struct ib_ah_attr attr;
+ struct ib_ah *ah;
+ struct ib_qp *src_qp = NULL;
+ unsigned tun_tx_ix = 0;
+ int dqpn;
+ int ret = 0;
+ int i;
+ int is_full_member = 0;
+ u16 tun_pkey_ix;
+ u8 ph_pkey_ix, full_pk_ix = 0, partial_pk_ix = 0;
+
+ if (dest_qpt > IB_QPT_GSI)
+ return -EINVAL;
+
+ tun_ctx = dev->sriov.demux[port-1].tun[slave];
+
+ /* check if proxy qp created */
+ if (!tun_ctx || tun_ctx->state != DEMUX_PV_STATE_ACTIVE)
+ return -EAGAIN;
+
+ /* QP0 forwarding only for Dom0 */
+ if (!dest_qpt && (mlx4_master_func_num(dev->dev) != slave))
+ return -EINVAL;
+
+ if (!dest_qpt)
+ tun_qp = &tun_ctx->qp[0];
+ else
+ tun_qp = &tun_ctx->qp[1];
+
+ /* compute pkey index for slave */
+ /* get physical pkey -- virtualized Dom0 pkey to phys*/
+ if (dest_qpt) {
+ ph_pkey_ix =
+ dev->pkeys.virt2phys_pkey[mlx4_master_func_num(dev->dev)][port - 1][wc->pkey_index];
+
+ /* now, translate this to the slave pkey index */
+ ret = get_pkey_phys_indices(dev, port, ph_pkey_ix, &full_pk_ix,
+ &partial_pk_ix, &is_full_member);
+ if (ret)
+ return -EINVAL;
+
+ for (i = 0; i < dev->dev->caps.pkey_table_len[port]; i++) {
+ if ((dev->pkeys.virt2phys_pkey[slave][port - 1][i] == full_pk_ix) ||
+ (is_full_member &&
+ (dev->pkeys.virt2phys_pkey[slave][port - 1][i] == partial_pk_ix)))
+ break;
+ }
+ if (i == dev->dev->caps.pkey_table_len[port])
+ return -EINVAL;
+ tun_pkey_ix = i;
+ } else
+ tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+
+ dqpn = dev->dev->caps.sqp_start + 8 * slave + port + (dest_qpt * 2) - 1;
+
+ /* get tunnel tx data buf for slave */
+ src_qp = tun_qp->qp;
+
+ /* create ah. Just need an empty one with the port num for the post send.
+ * The driver will set the force loopback bit in post_send */
+ memset(&attr, 0, sizeof attr);
+ attr.port_num = port;
+ ah = ib_create_ah(tun_ctx->pd, &attr);
+ if (IS_ERR(ah))
+ return -ENOMEM;
+
+ /* allocate tunnel tx buf after pass failure returns */
+ spin_lock(&tun_qp->tx_lock);
+ if (tun_qp->tx_ix_head - tun_qp->tx_ix_tail >=
+ (MLX4_NUM_TUNNEL_BUFS - 1))
+ ret = -EAGAIN;
+ else
+ tun_tx_ix = (++tun_qp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
+ spin_unlock(&tun_qp->tx_lock);
+ if (ret)
+ goto out;
+
+ tun_mad = (struct mlx4_rcv_tunnel_mad *) (tun_qp->tx_ring[tun_tx_ix].buf.addr);
+ if (tun_qp->tx_ring[tun_tx_ix].ah)
+ ib_destroy_ah(tun_qp->tx_ring[tun_tx_ix].ah);
+ tun_qp->tx_ring[tun_tx_ix].ah = ah;
+ ib_dma_sync_single_for_cpu(&dev->ib_dev,
+ tun_qp->tx_ring[tun_tx_ix].buf.map,
+ sizeof(struct mlx4_rcv_tunnel_mad),
+ DMA_TO_DEVICE);
+
+ /* copy over to tunnel buffer */
+ if (grh)
+ memcpy(&tun_mad->grh, grh, sizeof(*grh));
+ memcpy(&tun_mad->mad, mad, sizeof(*mad));
+
+ /* adjust tunnel data */
+ tun_mad->hdr.pkey_index = cpu_to_be16(tun_pkey_ix);
+ tun_mad->hdr.sl_vid = cpu_to_be16(((u16)(wc->sl)) << 12);
+ tun_mad->hdr.slid_mac_47_32 = cpu_to_be16(wc->slid);
+ tun_mad->hdr.flags_src_qp = cpu_to_be32(wc->src_qp & 0xFFFFFF);
+ tun_mad->hdr.g_ml_path = (grh && (wc->wc_flags & IB_WC_GRH)) ? 0x80 : 0;
+
+ ib_dma_sync_single_for_device(&dev->ib_dev,
+ tun_qp->tx_ring[tun_tx_ix].buf.map,
+ sizeof(struct mlx4_rcv_tunnel_mad),
+ DMA_TO_DEVICE);
+
+ list.addr = tun_qp->tx_ring[tun_tx_ix].buf.map;
+ list.length = sizeof(struct mlx4_rcv_tunnel_mad);
+ list.lkey = tun_ctx->mr->lkey;
+
+ wr.wr.ud.ah = ah;
+ wr.wr.ud.port_num = port;
+ wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
+ wr.wr.ud.remote_qpn = dqpn;
+ wr.next = NULL;
+ wr.wr_id = ((u64) tun_tx_ix) | MLX4_TUN_SET_WRID_QPN(dest_qpt);
+ wr.sg_list = &list;
+ wr.num_sge = 1;
+ wr.opcode = IB_WR_SEND;
+ wr.send_flags = IB_SEND_SIGNALED;
+
+ ret = ib_post_send(src_qp, &wr, &bad_wr);
+out:
+ if (ret)
+ ib_destroy_ah(ah);
+ return ret;
+}
+
+static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
+ struct ib_wc *wc, struct ib_grh *grh,
+ struct ib_mad *mad)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ int err;
+ int slave;
+ u8 *slave_id;
+
+ /* Initially assume that this mad is for us */
+ slave = mlx4_master_func_num(dev->dev);
+
+ /* See if the slave id is encoded in a response mad */
+ if (mad->mad_hdr.method & 0x80) {
+ slave_id = (u8 *) &mad->mad_hdr.tid;
+ slave = *slave_id;
+ if (slave != 255) /*255 indicates the dom0*/
+ *slave_id = 0; /* remap tid */
+ }
+
+ /* If a grh is present, we demux according to it */
+ if (wc->wc_flags & IB_WC_GRH) {
+ slave = mlx4_ib_find_real_gid(ibdev, port, grh->dgid.global.interface_id);
+ if (slave < 0) {
+ mlx4_ib_warn(ibdev, "failed matching grh\n");
+ return -ENOENT;
+ }
+ }
+ /* Class-specific handling */
+ switch (mad->mad_hdr.mgmt_class) {
+ case IB_MGMT_CLASS_SUBN_ADM:
+ if (mlx4_ib_demux_sa_handler(ibdev, port, slave,
+ (struct ib_sa_mad *) mad))
+ return 0;
+ break;
+ case IB_MGMT_CLASS_DEVICE_MGMT:
+ if (mad->mad_hdr.method != IB_MGMT_METHOD_GET_RESP)
+ return 0;
+ break;
+ default:
+ /* Drop unsupported classes for slaves in tunnel mode */
+ if (slave != mlx4_master_func_num(dev->dev)) {
+ pr_debug("dropping unsupported ingress mad from class:%d "
+ "for slave:%d\n", mad->mad_hdr.mgmt_class, slave);
+ return 0;
+ }
+ }
+ /*make sure that no slave==255 was not handled yet.*/
+ if (slave >= dev->dev->caps.sqp_demux) {
+ mlx4_ib_warn(ibdev, "slave id: %d is bigger than allowed:%d\n",
+ slave, dev->dev->caps.sqp_demux);
+ return -ENOENT;
+ }
+
+ err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad);
+ if (err)
+ pr_debug("failed sending to slave %d via tunnel qp (%d)\n",
+ slave, err);
+ return 0;
+}
+
static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
struct ib_wc *in_wc, struct ib_grh *in_grh,
struct ib_mad *in_mad, struct ib_mad *out_mad)
@@ -607,6 +857,216 @@ static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
return ib_post_recv(tun_qp->qp, &recv_wr, &bad_recv_wr);
}
+static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
+ int slave, struct ib_sa_mad *sa_mad)
+{
+ return 0;
+}
+
+static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
+{
+ int slave_start = dev->dev->caps.sqp_start + 8 * slave;
+
+ return (qpn >= slave_start && qpn <= slave_start + 1);
+}
+
+
+int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
+ enum ib_qp_type dest_qpt, u16 pkey_index, u32 remote_qpn,
+ u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad)
+{
+ struct ib_sge list;
+ struct ib_send_wr wr, *bad_wr;
+ struct mlx4_ib_demux_pv_ctx *sqp_ctx;
+ struct mlx4_ib_demux_pv_qp *sqp;
+ struct mlx4_mad_snd_buf *sqp_mad;
+ struct ib_ah *ah;
+ struct ib_qp *send_qp = NULL;
+ unsigned wire_tx_ix = 0;
+ int ret = 0;
+ u16 wire_pkey_ix;
+ int src_qpnum;
+ u8 sgid_index;
+
+
+ sqp_ctx = dev->sriov.sqps[port-1];
+
+ /* check if proxy qp created */
+ if (!sqp_ctx || sqp_ctx->state != DEMUX_PV_STATE_ACTIVE)
+ return -EAGAIN;
+
+ /* QP0 forwarding only for Dom0 */
+ if (dest_qpt == IB_QPT_SMI && (mlx4_master_func_num(dev->dev) != slave))
+ return -EINVAL;
+
+ if (dest_qpt == IB_QPT_SMI) {
+ src_qpnum = 0;
+ sqp = &sqp_ctx->qp[0];
+ wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+ } else {
+ src_qpnum = 1;
+ sqp = &sqp_ctx->qp[1];
+ wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][pkey_index];
+ }
+
+ send_qp = sqp->qp;
+
+ /* create ah */
+ sgid_index = attr->grh.sgid_index;
+ attr->grh.sgid_index = 0;
+ ah = ib_create_ah(sqp_ctx->pd, attr);
+ if (IS_ERR(ah))
+ return -ENOMEM;
+ attr->grh.sgid_index = sgid_index;
+ to_mah(ah)->av.ib.gid_index = sgid_index;
+ /* get rid of force-loopback bit */
+ to_mah(ah)->av.ib.port_pd &= cpu_to_be32(0x7FFFFFFF);
+ spin_lock(&sqp->tx_lock);
+ if (sqp->tx_ix_head - sqp->tx_ix_tail >=
+ (MLX4_NUM_TUNNEL_BUFS - 1))
+ ret = -EAGAIN;
+ else
+ wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
+ spin_unlock(&sqp->tx_lock);
+ if (ret)
+ goto out;
+
+ sqp_mad = (struct mlx4_mad_snd_buf *) (sqp->tx_ring[wire_tx_ix].buf.addr);
+ if (sqp->tx_ring[wire_tx_ix].ah)
+ ib_destroy_ah(sqp->tx_ring[wire_tx_ix].ah);
+ sqp->tx_ring[wire_tx_ix].ah = ah;
+ ib_dma_sync_single_for_cpu(&dev->ib_dev,
+ sqp->tx_ring[wire_tx_ix].buf.map,
+ sizeof(struct mlx4_mad_snd_buf),
+ DMA_TO_DEVICE);
+
+ memcpy(&sqp_mad->payload, mad, sizeof(*mad));
+
+ ib_dma_sync_single_for_device(&dev->ib_dev,
+ sqp->tx_ring[wire_tx_ix].buf.map,
+ sizeof(struct mlx4_mad_snd_buf),
+ DMA_TO_DEVICE);
+
+ list.addr = sqp->tx_ring[wire_tx_ix].buf.map;
+ list.length = sizeof(struct mlx4_mad_snd_buf);
+ list.lkey = sqp_ctx->mr->lkey;
+
+ wr.wr.ud.ah = ah;
+ wr.wr.ud.port_num = port;
+ wr.wr.ud.pkey_index = wire_pkey_ix;
+ wr.wr.ud.remote_qkey = qkey;
+ wr.wr.ud.remote_qpn = remote_qpn;
+ wr.next = NULL;
+ wr.wr_id = ((u64) wire_tx_ix) | MLX4_TUN_SET_WRID_QPN(src_qpnum);
+ wr.sg_list = &list;
+ wr.num_sge = 1;
+ wr.opcode = IB_WR_SEND;
+ wr.send_flags = IB_SEND_SIGNALED;
+
+ ret = ib_post_send(send_qp, &wr, &bad_wr);
+out:
+ if (ret)
+ ib_destroy_ah(ah);
+ return ret;
+}
+
+static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc *wc)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+ struct mlx4_ib_demux_pv_qp *tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc->wr_id)];
+ int wr_ix = wc->wr_id & (MLX4_NUM_TUNNEL_BUFS - 1);
+ struct mlx4_tunnel_mad *tunnel = tun_qp->ring[wr_ix].addr;
+ struct mlx4_ib_ah ah;
+ struct ib_ah_attr ah_attr;
+ u8 *slave_id;
+ int slave;
+
+ /* Get slave that sent this packet */
+ if (wc->src_qp < dev->dev->caps.sqp_start ||
+ wc->src_qp >= dev->dev->caps.base_tunnel_sqpn ||
+ (wc->src_qp & 0x1) != ctx->port - 1 ||
+ wc->src_qp & 0x4) {
+ mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp);
+ return;
+ }
+ slave = ((wc->src_qp & ~0x7) - dev->dev->caps.sqp_start) / 8;
+ if (slave != ctx->slave) {
+ mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: "
+ "belongs to another slave\n", wc->src_qp);
+ return;
+ }
+ if (slave != mlx4_master_func_num(dev->dev) && !(wc->src_qp & 0x2)) {
+ mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: "
+ "non-master trying to send QP0 packets\n", wc->src_qp);
+ return;
+ }
+
+ /* Map transaction ID */
+ ib_dma_sync_single_for_cpu(ctx->ib_dev, tun_qp->ring[wr_ix].map,
+ sizeof(struct mlx4_tunnel_mad),
+ DMA_FROM_DEVICE);
+ switch (tunnel->mad.mad_hdr.method) {
+ case IB_MGMT_METHOD_SET:
+ case IB_MGMT_METHOD_GET:
+ case IB_MGMT_METHOD_REPORT:
+ case IB_SA_METHOD_GET_TABLE:
+ case IB_SA_METHOD_DELETE:
+ case IB_SA_METHOD_GET_MULTI:
+ case IB_SA_METHOD_GET_TRACE_TBL:
+ slave_id = (u8 *) &tunnel->mad.mad_hdr.tid;
+ if (*slave_id) {
+ mlx4_ib_warn(ctx->ib_dev, "egress mad has non-null tid msb:%d "
+ "class:%d slave:%d\n", *slave_id,
+ tunnel->mad.mad_hdr.mgmt_class, slave);
+ return;
+ } else
+ *slave_id = slave;
+ default:
+ /* nothing */;
+ }
+
+ /* Class-specific handling */
+ switch (tunnel->mad.mad_hdr.mgmt_class) {
+ case IB_MGMT_CLASS_SUBN_ADM:
+ if (mlx4_ib_multiplex_sa_handler(ctx->ib_dev, ctx->port, slave,
+ (struct ib_sa_mad *) &tunnel->mad))
+ return;
+ break;
+ case IB_MGMT_CLASS_DEVICE_MGMT:
+ if (tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_GET &&
+ tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_SET)
+ return;
+ break;
+ default:
+ /* Drop unsupported classes for slaves in tunnel mode */
+ if (slave != mlx4_master_func_num(dev->dev)) {
+ mlx4_ib_warn(ctx->ib_dev, "dropping unsupported egress mad from class:%d "
+ "for slave:%d\n", tunnel->mad.mad_hdr.mgmt_class, slave);
+ return;
+ }
+ }
+
+ /* We are using standard ib_core services to send the mad, so generate a
+ * stadard address handle by decoding the tunnelled mlx4_ah fields */
+ memcpy(&ah.av, &tunnel->hdr.av, sizeof(struct mlx4_av));
+ ah.ibah.device = ctx->ib_dev;
+ mlx4_ib_query_ah(&ah.ibah, &ah_attr);
+ if ((ah_attr.ah_flags & IB_AH_GRH) &&
+ (ah_attr.grh.sgid_index != slave)) {
+ mlx4_ib_warn(ctx->ib_dev, "slave:%d accessed invalid sgid_index:%d\n",
+ slave, ah_attr.grh.sgid_index);
+ return;
+ }
+
+ mlx4_ib_send_to_wire(dev, slave, ctx->port,
+ is_proxy_qp0(dev, wc->src_qp, slave) ?
+ IB_QPT_SMI : IB_QPT_GSI,
+ be16_to_cpu(tunnel->hdr.pkey_index),
+ be32_to_cpu(tunnel->hdr.remote_qpn),
+ be32_to_cpu(tunnel->hdr.qkey),
+ &ah_attr, &tunnel->mad);
+}
+
static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
enum ib_qp_type qp_type, int is_tun)
{
@@ -731,7 +1191,57 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
static void mlx4_ib_tunnel_comp_worker(struct work_struct *work)
{
- /* dummy until next patch in series */
+ struct mlx4_ib_demux_pv_ctx *ctx;
+ struct mlx4_ib_demux_pv_qp *tun_qp;
+ struct ib_wc wc;
+ int ret;
+ ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work);
+ ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+
+ while (ib_poll_cq(ctx->cq, 1, &wc) == 1) {
+ tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)];
+ if (wc.status == IB_WC_SUCCESS) {
+ switch (wc.opcode) {
+ case IB_WC_RECV:
+ mlx4_ib_multiplex_mad(ctx, &wc);
+ ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp,
+ wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1));
+ if (ret)
+ pr_err("Failed reposting tunnel "
+ "buf:%lld\n", wc.wr_id);
+ break;
+ case IB_WC_SEND:
+ pr_debug("received tunnel send completion:"
+ "wrid=0x%llx, status=0x%x\n",
+ wc.wr_id, wc.status);
+ ib_destroy_ah(tun_qp->tx_ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+ tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+ = NULL;
+ spin_lock(&tun_qp->tx_lock);
+ tun_qp->tx_ix_tail++;
+ spin_unlock(&tun_qp->tx_lock);
+
+ break;
+ default:
+ break;
+ }
+ } else {
+ pr_debug("mlx4_ib: completion error in tunnel: %d."
+ " status = %d, wrid = 0x%llx\n",
+ ctx->slave, wc.status, wc.wr_id);
+ if (!MLX4_TUN_IS_RECV(wc.wr_id)) {
+ ib_destroy_ah(tun_qp->tx_ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+ tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+ = NULL;
+ spin_lock(&tun_qp->tx_lock);
+ tun_qp->tx_ix_tail++;
+ spin_unlock(&tun_qp->tx_lock);
+ }
+ }
+ }
}
static void pv_qp_event_handler(struct ib_event *event, void *qp_context)
@@ -839,7 +1349,60 @@ err_qp:
*/
static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
{
- /* dummy until next patch in series */
+ struct mlx4_ib_demux_pv_ctx *ctx;
+ struct mlx4_ib_demux_pv_qp *sqp;
+ struct ib_wc wc;
+ struct ib_grh *grh;
+ struct ib_mad *mad;
+
+ ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work);
+ ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+
+ while (mlx4_ib_poll_cq(ctx->cq, 1, &wc) == 1) {
+ sqp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)];
+ if (wc.status == IB_WC_SUCCESS) {
+ switch (wc.opcode) {
+ case IB_WC_SEND:
+ ib_destroy_ah(sqp->tx_ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+ sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+ = NULL;
+ spin_lock(&sqp->tx_lock);
+ sqp->tx_ix_tail++;
+ spin_unlock(&sqp->tx_lock);
+ break;
+ case IB_WC_RECV:
+ mad = (struct ib_mad *) &(((struct mlx4_mad_rcv_buf *)
+ (sqp->ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].addr))->payload);
+ grh = &(((struct mlx4_mad_rcv_buf *)
+ (sqp->ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].addr))->grh);
+ mlx4_ib_demux_mad(ctx->ib_dev, ctx->port, &wc, grh, mad);
+ if (mlx4_ib_post_pv_qp_buf(ctx, sqp, wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)))
+ pr_err("Failed reposting SQP "
+ "buf:%lld\n", wc.wr_id);
+ break;
+ default:
+ BUG_ON(1);
+ break;
+ }
+ } else {
+ pr_debug("mlx4_ib: completion error in tunnel: %d."
+ " status = %d, wrid = 0x%llx\n",
+ ctx->slave, wc.status, wc.wr_id);
+ if (!MLX4_TUN_IS_RECV(wc.wr_id)) {
+ ib_destroy_ah(sqp->tx_ring[wc.wr_id &
+ (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+ sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+ = NULL;
+ spin_lock(&sqp->tx_lock);
+ sqp->tx_ix_tail++;
+ spin_unlock(&sqp->tx_lock);
+ }
+ }
+ }
}
static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port,
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 09/22] {NET,IB}/mlx4: MAD_IFC paravirtualization
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (7 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 08/22] IB/mlx4: SRIOV multiplex and demultiplex MADs Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 10/22] IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV Jack Morgenstein
` (14 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
The MAD_IFC firmware command fulfills two functions.
In the first case, it is used in the QP0/QP1 MAD-handling flow to
obtain information from the FW (for answering queries), and for
setting variables in the HCA (MAD SET packets).
In this function, MAD_IFC should provide the FW (physical) view
of the data. This is the view that OpenSM needs for its functioning.
We call this the "network view".
In the second case, MAD_IFC is used by various verbs to obtain data
regarding the local HCA (e.g., ib_query_device() ).
We call this the "host view".
This data needs to be paravirtualized.
MAD_IFC thus needs a wrapper function, and also needs another flag
indicating whether it should provide the network view (when it is
called by ib_process_mad in special-qp packet handling), or the
host view (when it is called while implementing a verb).
There are currently 2 flag parameters in mlx4_MAD_IFC already:
ignore_bkey and ignore_mkey. These two parameters are replaced
by a single "mad_ifc_flags" parameter, with different bits set
for each flag. A third flag is added: "network-view/host-view".
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 20 ++--
drivers/infiniband/hw/mlx4/main.c | 64 +++++++++---
drivers/infiniband/hw/mlx4/mlx4_ib.h | 14 +++-
drivers/net/ethernet/mellanox/mlx4/cmd.c | 162 ++++++++++++++++++++++++++++++
4 files changed, 234 insertions(+), 26 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index a392a5c..4c8650f 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -75,7 +75,7 @@ struct mlx4_rcv_tunnel_mad {
struct ib_mad mad;
} __packed;
-int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags,
int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
void *in_mad, void *response_mad)
{
@@ -102,10 +102,13 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
* Key check traps can't be generated unless we have in_wc to
* tell us where to send the trap.
*/
- if (ignore_mkey || !in_wc)
+ if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_MKEY) || !in_wc)
op_modifier |= 0x1;
- if (ignore_bkey || !in_wc)
+ if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_BKEY) || !in_wc)
op_modifier |= 0x2;
+ if (mlx4_is_mfunc(dev->dev) &&
+ (mad_ifc_flags & MLX4_MAD_IFC_NET_VIEW || in_wc))
+ op_modifier |= 0x8;
if (in_wc) {
struct {
@@ -138,10 +141,10 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
in_modifier |= in_wc->slid << 16;
}
- err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma,
- in_modifier, op_modifier,
+ err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma, in_modifier,
+ mlx4_is_master(dev->dev) ? (op_modifier & ~0x8) : op_modifier,
MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
- MLX4_CMD_NATIVE);
+ (op_modifier & 0x8) ? MLX4_CMD_NATIVE : MLX4_CMD_WRAPPED);
if (!err)
memcpy(response_mad, outmailbox->buf, 256);
@@ -610,8 +613,9 @@ static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
prev_lid = pattr.lid;
err = mlx4_MAD_IFC(to_mdev(ibdev),
- mad_flags & IB_MAD_IGNORE_MKEY,
- mad_flags & IB_MAD_IGNORE_BKEY,
+ (mad_flags & IB_MAD_IGNORE_MKEY ? MLX4_MAD_IFC_IGNORE_MKEY : 0) |
+ (mad_flags & IB_MAD_IGNORE_BKEY ? MLX4_MAD_IFC_IGNORE_BKEY : 0) |
+ MLX4_MAD_IFC_NET_VIEW,
port_num, in_wc, in_grh, in_mad, out_mad);
if (err)
return IB_MAD_RESULT_FAILURE;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index bdfd3ae..48cda35 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -98,7 +98,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
init_query_mad(in_mad);
in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
+ err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
+ 1, NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -182,11 +183,12 @@ mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
}
static int ib_link_query_port(struct ib_device *ibdev, u8 port,
- struct ib_port_attr *props)
+ struct ib_port_attr *props, int netw_view)
{
struct ib_smp *in_mad = NULL;
struct ib_smp *out_mad = NULL;
int ext_active_speed;
+ int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
int err = -ENOMEM;
in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -198,7 +200,10 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
in_mad->attr_mod = cpu_to_be32(port);
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL,
+ if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
+ mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+ err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
in_mad, out_mad);
if (err)
goto out;
@@ -211,7 +216,10 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
props->state = out_mad->data[32] & 0xf;
props->phys_state = out_mad->data[33] >> 4;
props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
- props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
+ if (netw_view)
+ props->gid_tbl_len = out_mad->data[50];
+ else
+ props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
@@ -244,7 +252,7 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
in_mad->attr_mod = cpu_to_be32(port);
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
+ err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -270,7 +278,7 @@ static u8 state_to_phys_state(enum ib_port_state state)
}
static int eth_link_query_port(struct ib_device *ibdev, u8 port,
- struct ib_port_attr *props)
+ struct ib_port_attr *props, int netw_view)
{
struct mlx4_ib_dev *mdev = to_mdev(ibdev);
@@ -320,20 +328,27 @@ out:
return err;
}
-static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
- struct ib_port_attr *props)
+int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+ struct ib_port_attr *props, int netw_view)
{
int err;
memset(props, 0, sizeof *props);
err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
- ib_link_query_port(ibdev, port, props) :
- eth_link_query_port(ibdev, port, props);
+ ib_link_query_port(ibdev, port, props, netw_view) :
+ eth_link_query_port(ibdev, port, props, netw_view);
return err;
}
+static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+ struct ib_port_attr *props)
+{
+ /* returns host view */
+ return __mlx4_ib_query_port(ibdev, port, props, 0);
+}
+
static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
union ib_gid *gid)
{
@@ -350,7 +365,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
in_mad->attr_mod = cpu_to_be32(port);
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+ err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port,
+ NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -360,7 +376,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
in_mad->attr_mod = cpu_to_be32(index / 8);
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+ err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port,
+ NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -391,11 +408,12 @@ static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
return iboe_query_gid(ibdev, port, index, gid);
}
-static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
- u16 *pkey)
+int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+ u16 *pkey, int netw_view)
{
struct ib_smp *in_mad = NULL;
struct ib_smp *out_mad = NULL;
+ int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
int err = -ENOMEM;
in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -407,7 +425,11 @@ static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
in_mad->attr_mod = cpu_to_be32(index / 32);
- err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+ if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
+ mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+ err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
+ in_mad, out_mad);
if (err)
goto out;
@@ -419,6 +441,11 @@ out:
return err;
}
+static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
+{
+ return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
+}
+
static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
struct ib_device_modify *props)
{
@@ -848,6 +875,7 @@ static int init_node_data(struct mlx4_ib_dev *dev)
{
struct ib_smp *in_mad = NULL;
struct ib_smp *out_mad = NULL;
+ int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
int err = -ENOMEM;
in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -857,8 +885,10 @@ static int init_node_data(struct mlx4_ib_dev *dev)
init_query_mad(in_mad);
in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
+ if (mlx4_is_master(dev->dev))
+ mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
- err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+ err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -866,7 +896,7 @@ static int init_node_data(struct mlx4_ib_dev *dev)
in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
- err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+ err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
if (err)
goto out;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 137941d..ac71d56 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -176,6 +176,14 @@ enum mlx4_ib_qp_type {
MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \
MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI)
+enum mlx4_ib_mad_ifc_flags {
+ MLX4_MAD_IFC_IGNORE_MKEY = 1,
+ MLX4_MAD_IFC_IGNORE_BKEY = 2,
+ MLX4_MAD_IFC_IGNORE_KEYS = (MLX4_MAD_IFC_IGNORE_MKEY |
+ MLX4_MAD_IFC_IGNORE_BKEY),
+ MLX4_MAD_IFC_NET_VIEW = 4,
+};
+
enum {
MLX4_NUM_TUNNEL_BUFS = 256,
};
@@ -512,7 +520,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
struct ib_recv_wr **bad_wr);
-int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags,
int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
void *in_mad, void *response_mad);
int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
@@ -527,6 +535,10 @@ int mlx4_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, int npages,
u64 iova);
int mlx4_ib_unmap_fmr(struct list_head *fmr_list);
int mlx4_ib_fmr_dealloc(struct ib_fmr *fmr);
+int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+ struct ib_port_attr *props, int netw_view);
+int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+ u16 *pkey, int netw_view);
int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr,
u8 *mac, int *is_mcast, u8 port);
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 89feaed..28fca65 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -40,6 +40,7 @@
#include <linux/mlx4/cmd.h>
#include <linux/semaphore.h>
+#include <rdma/ib_smi.h>
#include <asm/io.h>
@@ -582,6 +583,149 @@ static int mlx4_ACCESS_MEM(struct mlx4_dev *dev, u64 master_addr,
MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
}
+static int query_pkey_block(struct mlx4_dev *dev, u8 port, u16 index, u16 *pkey,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox)
+{
+ struct ib_smp *in_mad = (struct ib_smp *)(inbox->buf);
+ struct ib_smp *out_mad = (struct ib_smp *)(outbox->buf);
+ int err;
+ int i;
+
+ if (index & 0x1f)
+ return -EINVAL;
+
+ in_mad->attr_mod = cpu_to_be32(index / 32);
+
+ err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, port, 3,
+ MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
+ MLX4_CMD_NATIVE);
+ if (err)
+ return err;
+
+ for (i = 0; i < 32; ++i)
+ pkey[i] = be16_to_cpu(((__be16 *) out_mad->data)[i]);
+
+ return err;
+}
+
+static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox)
+{
+ int i;
+ int err;
+
+ for (i = 0; i < dev->caps.pkey_table_len[port]; i += 32) {
+ err = query_pkey_block(dev, port, i, table + i, inbox, outbox);
+ if (err)
+ return err;
+ }
+
+ return 0;
+}
+#define PORT_CAPABILITY_LOCATION_IN_SMP 20
+#define PORT_STATE_OFFSET 32
+
+static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf)
+{
+ /* will be modified when add alias_guid feature */
+ return IB_PORT_DOWN;
+}
+
+static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ struct ib_smp *smp = inbox->buf;
+ u32 index;
+ u8 port;
+ u16 *table;
+ int err;
+ int vidx, pidx;
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ struct ib_smp *outsmp = outbox->buf;
+ __be16 *outtab = (__be16 *)(outsmp->data);
+ __be32 slave_cap_mask;
+ port = vhcr->in_modifier;
+
+ if (smp->base_version == 1 &&
+ smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
+ smp->class_version == 1) {
+ if (smp->method == IB_MGMT_METHOD_GET) {
+ if (smp->attr_id == IB_SMP_ATTR_PKEY_TABLE) {
+ index = be32_to_cpu(smp->attr_mod);
+ if (port < 1 || port > dev->caps.num_ports)
+ return -EINVAL;
+ table = kzalloc(dev->caps.pkey_table_len[port] * sizeof *table, GFP_KERNEL);
+ if (!table)
+ return -ENOMEM;
+ /* need to get the full pkey table because the paravirtualized
+ * pkeys may be scattered among several pkey blocks.
+ */
+ err = get_full_pkey_table(dev, port, table, inbox, outbox);
+ if (!err) {
+ for (vidx = index * 32; vidx < (index + 1) * 32; ++vidx) {
+ pidx = priv->virt2phys_pkey[slave][port - 1][vidx];
+ outtab[vidx % 32] = cpu_to_be16(table[pidx]);
+ }
+ }
+ kfree(table);
+ return err;
+ }
+ if (smp->attr_id == IB_SMP_ATTR_PORT_INFO) {
+ /*get the slave specific caps:*/
+ /*do the command */
+ err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
+ vhcr->in_modifier, vhcr->op_modifier,
+ vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
+ /* modify the response for slaves */
+ if (!err && slave != mlx4_master_func_num(dev)) {
+ u8 *state = outsmp->data + PORT_STATE_OFFSET;
+
+ *state = (*state & 0xf0) | vf_port_state(dev, port, slave);
+ slave_cap_mask = priv->mfunc.master.slave_state[slave].ib_cap_mask[port];
+ memcpy(outsmp->data + PORT_CAPABILITY_LOCATION_IN_SMP, &slave_cap_mask, 4);
+ }
+ return err;
+ }
+ if (smp->attr_id == IB_SMP_ATTR_GUID_INFO) {
+ /* compute slave's gid block */
+ smp->attr_mod = cpu_to_be32(slave / 8);
+ /* execute cmd */
+ err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
+ vhcr->in_modifier, vhcr->op_modifier,
+ vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
+ if (!err) {
+ /* if needed, move slave gid to index 0 */
+ if (slave % 8)
+ memcpy(outsmp->data,
+ outsmp->data + (slave % 8) * 8, 8);
+ /* delete all other gids */
+ memset(outsmp->data + 8, 0, 56);
+ }
+ return err;
+ }
+ }
+ }
+ if (slave != mlx4_master_func_num(dev) &&
+ ((smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) ||
+ (smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
+ smp->method == IB_MGMT_METHOD_SET))) {
+ mlx4_err(dev, "slave %d is trying to execute a Subnet MGMT MAD, "
+ "class 0x%x, method 0x%x for attr 0x%x. Rejecting\n",
+ slave, smp->method, smp->mgmt_class,
+ be16_to_cpu(smp->attr_id));
+ return -EPERM;
+ }
+ /*default:*/
+ return mlx4_cmd_box(dev, inbox->dma, outbox->dma,
+ vhcr->in_modifier, vhcr->op_modifier,
+ vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
+}
+
int mlx4_DMA_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
@@ -1016,6 +1160,24 @@ static struct mlx4_cmd_info cmd_info[] = {
.wrapper = mlx4_GEN_QP_wrapper
},
{
+ .opcode = MLX4_CMD_CONF_SPECIAL_QP,
+ .has_inbox = false,
+ .has_outbox = false,
+ .out_is_imm = false,
+ .encode_slave_id = false,
+ .verify = NULL, /* XXX verify: only demux can do this */
+ .wrapper = NULL
+ },
+ {
+ .opcode = MLX4_CMD_MAD_IFC,
+ .has_inbox = true,
+ .has_outbox = true,
+ .out_is_imm = false,
+ .encode_slave_id = false,
+ .verify = NULL,
+ .wrapper = mlx4_MAD_IFC_wrapper
+ },
+ {
.opcode = MLX4_CMD_QUERY_IF_STAT,
.has_inbox = false,
.has_outbox = true,
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 10/22] IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (8 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 09/22] {NET,IB}/mlx4: MAD_IFC paravirtualization Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 11/22] IB/mlx4: Add CM paravirtualization Jack Morgenstein
` (13 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Oren Duer, Eli Cohen,
Jack Morgenstein
From: Oren Duer <oren-VPRAkNaXOzVS1MOuV/RT9w@public.gmane.org>
MCG para-virtualization support includes:
- Creating multicast groups by VFs, and keeping accounting of them
- Leaving multicast groups by VFs
- SM will only be updated with real changes in the overall picture of MCGs status
- Creation of MGID=0 groups (let SM choose MGID)
Note that the MCG module maintains its own internal MCG object reference
counts. The reason for this is that the IB core is used to track only
the multicast groups joins generated by the PF it runs over.
The PF ib-core layer is unaware of slaves, so it cannot be used to keep
track of MCG joins they generate.
Signed-off-by: Oren Duer <oren-VPRAkNaXOzVS1MOuV/RT9w@public.gmane.org>
Signed-off-by: Eli Cohen <eli-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/mad.c | 60 ++-
drivers/infiniband/hw/mlx4/main.c | 18 +-
drivers/infiniband/hw/mlx4/mcg.c | 1187 ++++++++++++++++++++++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 30 +
5 files changed, 1285 insertions(+), 12 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/mcg.c
diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile
index 70f09c7..20d627d 100644
--- a/drivers/infiniband/hw/mlx4/Makefile
+++ b/drivers/infiniband/hw/mlx4/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o
-mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o
+mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 4c8650f..195b211 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -75,6 +75,14 @@ struct mlx4_rcv_tunnel_mad {
struct ib_mad mad;
} __packed;
+static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num);
+
+__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx)
+{
+ return cpu_to_be64(atomic_inc_return(&ctx->tid)) |
+ cpu_to_be64(0xff00000000000000LL);
+}
+
int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags,
int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
void *in_mad, void *response_mad)
@@ -208,8 +216,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
pinfo->neighbormtu_mastersmsl & 0xf);
if (pinfo->clientrereg_resv_subnetto & 0x80)
- mlx4_ib_dispatch_event(dev, port_num,
- IB_EVENT_CLIENT_REREGISTER);
+ handle_client_rereg_event(dev, port_num);
if (prev_lid != lid)
mlx4_ib_dispatch_event(dev, port_num,
@@ -304,7 +311,17 @@ static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *ma
static int mlx4_ib_demux_sa_handler(struct ib_device *ibdev, int port, int slave,
struct ib_sa_mad *sa_mad)
{
- return 0;
+ int ret = 0;
+
+ /* dispatch to different sa handlers */
+ switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) {
+ case IB_SA_ATTR_MC_MEMBER_REC:
+ ret = mlx4_ib_mcg_demux_handler(ibdev, port, slave, sa_mad);
+ break;
+ default:
+ break;
+ }
+ return ret;
}
int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid)
@@ -764,6 +781,16 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
}
}
+static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num)
+{
+ /* re-configure the mcg's */
+ if (mlx4_is_master(dev->dev)) {
+ if (!dev->sriov.is_going_down)
+ mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0);
+ }
+ mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_CLIENT_REREGISTER);
+}
+
void handle_port_mgmt_change_event(struct work_struct *work)
{
struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
@@ -793,8 +820,7 @@ void handle_port_mgmt_change_event(struct work_struct *work)
mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK)
- mlx4_ib_dispatch_event(dev, port,
- IB_EVENT_CLIENT_REREGISTER);
+ handle_client_rereg_event(dev, port);
break;
case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE:
@@ -864,7 +890,17 @@ static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
int slave, struct ib_sa_mad *sa_mad)
{
- return 0;
+ int ret = 0;
+
+ /* dispatch to different sa handlers */
+ switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) {
+ case IB_SA_ATTR_MC_MEMBER_REC:
+ ret = mlx4_ib_mcg_multiplex_handler(ibdev, port, slave, sa_mad);
+ break;
+ default:
+ break;
+ }
+ return ret;
}
static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
@@ -1586,6 +1622,7 @@ static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave,
int ret = 0;
if (!do_init) {
+ clean_vf_mcast(&dev->sriov.demux[port - 1], slave);
/* for master, destroy real sqp resources */
if (slave == mlx4_master_func_num(dev->dev))
destroy_pv_resources(dev, slave, port,
@@ -1639,10 +1676,16 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
ret = alloc_pv_object(dev, i, port, &ctx->tun[i]);
if (ret) {
ret = -ENOMEM;
- goto err_wq;
+ goto err_mcg;
}
}
+ ret = mlx4_ib_mcg_port_init(ctx);
+ if (ret) {
+ pr_err("Failed initializing mcg para-virt (%d)\n", ret);
+ goto err_mcg;
+ }
+
snprintf(name, sizeof name, "mlx4_ibt%d", port);
ctx->wq = create_singlethread_workqueue(name);
if (!ctx->wq) {
@@ -1666,6 +1709,8 @@ err_udwq:
ctx->wq = NULL;
err_wq:
+ mlx4_ib_mcg_port_cleanup(ctx, 1);
+err_mcg:
for (i = 0; i < dev->dev->caps.sqp_demux; i++)
free_pv_object(dev, i, port);
kfree(ctx->tun);
@@ -1701,6 +1746,7 @@ static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx)
int i;
if (ctx) {
struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+ mlx4_ib_mcg_port_cleanup(ctx, 1);
for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
if (!ctx->tun[i])
continue;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 48cda35..fce64ec 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1627,18 +1627,28 @@ static int __init mlx4_ib_init(void)
if (!wq)
return -ENOMEM;
+ err = mlx4_ib_mcg_init();
+ if (err)
+ goto clean_wq;
+
err = mlx4_register_interface(&mlx4_ib_interface);
- if (err) {
- destroy_workqueue(wq);
- return err;
- }
+ if (err)
+ goto clean_mcg;
return 0;
+
+clean_mcg:
+ mlx4_ib_mcg_destroy();
+
+clean_wq:
+ destroy_workqueue(wq);
+ return err;
}
static void __exit mlx4_ib_cleanup(void)
{
mlx4_unregister_interface(&mlx4_ib_interface);
+ mlx4_ib_mcg_destroy();
destroy_workqueue(wq);
}
diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c
new file mode 100644
index 0000000..1ee2e3a
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/mcg.c
@@ -0,0 +1,1187 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_cache.h>
+#include <rdma/ib_sa.h>
+
+#include <linux/mlx4/cmd.h>
+#include <linux/rbtree.h>
+#include <linux/delay.h>
+
+#include "mlx4_ib.h"
+
+#define MAX_VFS 80
+#define MAX_PEND_REQS_PER_FUNC 4
+#define MAD_TIMEOUT_MS 2000
+
+#define mcg_warn(fmt, arg...) pr_warn("MCG WARNING: " fmt, ##arg)
+#define mcg_error(fmt, arg...) pr_err(fmt, ##arg)
+#define mcg_warn_group(group, format, arg...) \
+ pr_warn("%s-%d: %16s (port %d): WARNING: " format, __func__, __LINE__,\
+ (group)->name, group->demux->port, ## arg)
+
+#define mcg_error_group(group, format, arg...) \
+ pr_err(" %16s: " format, (group)->name, ## arg)
+
+
+static union ib_gid mgid0;
+
+static struct workqueue_struct *clean_wq;
+
+enum mcast_state {
+ MCAST_NOT_MEMBER = 0,
+ MCAST_MEMBER,
+};
+
+enum mcast_group_state {
+ MCAST_IDLE,
+ MCAST_JOIN_SENT,
+ MCAST_LEAVE_SENT,
+ MCAST_RESP_READY
+};
+
+struct mcast_member {
+ enum mcast_state state;
+ uint8_t join_state;
+ int num_pend_reqs;
+ struct list_head pending;
+};
+
+struct ib_sa_mcmember_data {
+ union ib_gid mgid;
+ union ib_gid port_gid;
+ __be32 qkey;
+ __be16 mlid;
+ u8 mtusel_mtu;
+ u8 tclass;
+ __be16 pkey;
+ u8 ratesel_rate;
+ u8 lifetmsel_lifetm;
+ __be32 sl_flowlabel_hoplimit;
+ u8 scope_join_state;
+ u8 proxy_join;
+ u8 reserved[2];
+};
+
+struct mcast_group {
+ struct ib_sa_mcmember_data rec;
+ struct rb_node node;
+ struct list_head mgid0_list;
+ struct mlx4_ib_demux_ctx *demux;
+ struct mcast_member func[MAX_VFS];
+ struct mutex lock;
+ struct work_struct work;
+ struct list_head pending_list;
+ int members[3];
+ enum mcast_group_state state;
+ enum mcast_group_state prev_state;
+ struct ib_sa_mad response_sa_mad;
+ __be64 last_req_tid;
+
+ char name[33]; /* MGID string */
+
+ /* refcount is the reference count for the following:
+ 1. Each queued request
+ 2. Each invocation of the worker thread
+ 3. Membership of the port at the SA
+ */
+ atomic_t refcount;
+
+ /* delayed work to clean pending SM request */
+ struct delayed_work timeout_work;
+ struct list_head cleanup_list;
+};
+
+struct mcast_req {
+ int func;
+ struct ib_sa_mad sa_mad;
+ struct list_head group_list;
+ struct list_head func_list;
+ struct mcast_group *group;
+ int clean;
+};
+
+
+#define safe_atomic_dec(ref) \
+ do {\
+ if (atomic_dec_and_test(ref)) \
+ mcg_warn_group(group, "did not expect to reach zero\n"); \
+ } while (0)
+
+static const char *get_state_string(enum mcast_group_state state)
+{
+ switch (state) {
+ case MCAST_IDLE:
+ return "MCAST_IDLE";
+ case MCAST_JOIN_SENT:
+ return "MCAST_JOIN_SENT";
+ case MCAST_LEAVE_SENT:
+ return "MCAST_LEAVE_SENT";
+ case MCAST_RESP_READY:
+ return "MCAST_RESP_READY";
+ }
+ return "Invalid State";
+}
+
+static struct mcast_group *mcast_find(struct mlx4_ib_demux_ctx *ctx,
+ union ib_gid *mgid)
+{
+ struct rb_node *node = ctx->mcg_table.rb_node;
+ struct mcast_group *group;
+ int ret;
+
+ while (node) {
+ group = rb_entry(node, struct mcast_group, node);
+ ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid);
+ if (!ret)
+ return group;
+
+ if (ret < 0)
+ node = node->rb_left;
+ else
+ node = node->rb_right;
+ }
+ return NULL;
+}
+
+static struct mcast_group *mcast_insert(struct mlx4_ib_demux_ctx *ctx,
+ struct mcast_group *group)
+{
+ struct rb_node **link = &ctx->mcg_table.rb_node;
+ struct rb_node *parent = NULL;
+ struct mcast_group *cur_group;
+ int ret;
+
+ while (*link) {
+ parent = *link;
+ cur_group = rb_entry(parent, struct mcast_group, node);
+
+ ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw,
+ sizeof group->rec.mgid);
+ if (ret < 0)
+ link = &(*link)->rb_left;
+ else if (ret > 0)
+ link = &(*link)->rb_right;
+ else
+ return cur_group;
+ }
+ rb_link_node(&group->node, parent, link);
+ rb_insert_color(&group->node, &ctx->mcg_table);
+ return NULL;
+}
+
+static int send_mad_to_wire(struct mlx4_ib_demux_ctx *ctx, struct ib_mad *mad)
+{
+ struct mlx4_ib_dev *dev = ctx->dev;
+ struct ib_ah_attr ah_attr;
+
+ spin_lock(&dev->sm_lock);
+ if (!dev->sm_ah[ctx->port - 1]) {
+ /* port is not yet Active, sm_ah not ready */
+ spin_unlock(&dev->sm_lock);
+ return -EAGAIN;
+ }
+ mlx4_ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);
+ spin_unlock(&dev->sm_lock);
+ return mlx4_ib_send_to_wire(dev, mlx4_master_func_num(dev->dev), ctx->port,
+ IB_QPT_GSI, 0, 1, IB_QP1_QKEY, &ah_attr, mad);
+}
+
+static int send_mad_to_slave(int slave, struct mlx4_ib_demux_ctx *ctx,
+ struct ib_mad *mad)
+{
+ struct mlx4_ib_dev *dev = ctx->dev;
+ struct ib_mad_agent *agent = dev->send_agent[ctx->port - 1][1];
+ struct ib_wc wc;
+ struct ib_ah_attr ah_attr;
+
+ /* Our agent might not yet be registered when mads start to arrive */
+ if (!agent)
+ return -EAGAIN;
+
+ ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);
+
+ wc.pkey_index = 0;
+ wc.sl = 0;
+ wc.dlid_path_bits = 0;
+ wc.port_num = ctx->port;
+ wc.slid = ah_attr.dlid; /* opensm lid */
+ wc.src_qp = 1;
+ return mlx4_ib_send_to_slave(dev, slave, ctx->port, IB_QPT_GSI, &wc, NULL, mad);
+}
+
+static int send_join_to_wire(struct mcast_group *group, struct ib_sa_mad *sa_mad)
+{
+ struct ib_sa_mad mad;
+ struct ib_sa_mcmember_data *sa_mad_data = (struct ib_sa_mcmember_data *)&mad.data;
+ int ret;
+
+ /* we rely on a mad request as arrived from a VF */
+ memcpy(&mad, sa_mad, sizeof mad);
+
+ /* fix port GID to be the real one (slave 0) */
+ sa_mad_data->port_gid.global.interface_id = group->demux->guid_cache[0];
+
+ /* assign our own TID */
+ mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
+ group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */
+
+ ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
+ /* set timeout handler */
+ if (!ret) {
+ /* calls mlx4_ib_mcg_timeout_handler */
+ queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
+ msecs_to_jiffies(MAD_TIMEOUT_MS));
+ }
+
+ return ret;
+}
+
+static int send_leave_to_wire(struct mcast_group *group, u8 join_state)
+{
+ struct ib_sa_mad mad;
+ struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
+ int ret;
+
+ memset(&mad, 0, sizeof mad);
+ mad.mad_hdr.base_version = 1;
+ mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
+ mad.mad_hdr.class_version = 2;
+ mad.mad_hdr.method = IB_SA_METHOD_DELETE;
+ mad.mad_hdr.status = cpu_to_be16(0);
+ mad.mad_hdr.class_specific = cpu_to_be16(0);
+ mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
+ group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */
+ mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
+ mad.mad_hdr.attr_mod = cpu_to_be32(0);
+ mad.sa_hdr.sm_key = 0x0;
+ mad.sa_hdr.attr_offset = cpu_to_be16(7);
+ mad.sa_hdr.comp_mask = IB_SA_MCMEMBER_REC_MGID |
+ IB_SA_MCMEMBER_REC_PORT_GID | IB_SA_MCMEMBER_REC_JOIN_STATE;
+
+ *sa_data = group->rec;
+ sa_data->scope_join_state = join_state;
+
+ ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
+ if (ret)
+ group->state = MCAST_IDLE;
+
+ /* set timeout handler */
+ if (!ret) {
+ /* calls mlx4_ib_mcg_timeout_handler */
+ queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
+ msecs_to_jiffies(MAD_TIMEOUT_MS));
+ }
+
+ return ret;
+}
+
+static int send_reply_to_slave(int slave, struct mcast_group *group,
+ struct ib_sa_mad *req_sa_mad, u16 status)
+{
+ struct ib_sa_mad mad;
+ struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
+ struct ib_sa_mcmember_data *req_sa_data = (struct ib_sa_mcmember_data *)&req_sa_mad->data;
+ int ret;
+
+ memset(&mad, 0, sizeof mad);
+ mad.mad_hdr.base_version = 1;
+ mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
+ mad.mad_hdr.class_version = 2;
+ mad.mad_hdr.method = IB_MGMT_METHOD_GET_RESP;
+ mad.mad_hdr.status = cpu_to_be16(status);
+ mad.mad_hdr.class_specific = cpu_to_be16(0);
+ mad.mad_hdr.tid = req_sa_mad->mad_hdr.tid;
+ *(u8 *)&mad.mad_hdr.tid = 0; /* resetting tid to 0 */
+ mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
+ mad.mad_hdr.attr_mod = cpu_to_be32(0);
+ mad.sa_hdr.sm_key = req_sa_mad->sa_hdr.sm_key;
+ mad.sa_hdr.attr_offset = cpu_to_be16(7);
+ mad.sa_hdr.comp_mask = 0; /* ignored on responses, see IBTA spec */
+
+ *sa_data = group->rec;
+
+ /* reconstruct VF's requested join_state and port_gid */
+ sa_data->scope_join_state &= 0xf0;
+ sa_data->scope_join_state |= (group->func[slave].join_state & 0x0f);
+ memcpy(&sa_data->port_gid, &req_sa_data->port_gid, sizeof req_sa_data->port_gid);
+
+ ret = send_mad_to_slave(slave, group->demux, (struct ib_mad *)&mad);
+ return ret;
+}
+
+static int check_selector(ib_sa_comp_mask comp_mask,
+ ib_sa_comp_mask selector_mask,
+ ib_sa_comp_mask value_mask,
+ u8 src_value, u8 dst_value)
+{
+ int err;
+ u8 selector = dst_value >> 6;
+ dst_value &= 0x3f;
+ src_value &= 0x3f;
+
+ if (!(comp_mask & selector_mask) || !(comp_mask & value_mask))
+ return 0;
+
+ switch (selector) {
+ case IB_SA_GT:
+ err = (src_value <= dst_value);
+ break;
+ case IB_SA_LT:
+ err = (src_value >= dst_value);
+ break;
+ case IB_SA_EQ:
+ err = (src_value != dst_value);
+ break;
+ default:
+ err = 0;
+ break;
+ }
+
+ return err;
+}
+
+static u16 cmp_rec(struct ib_sa_mcmember_data *src,
+ struct ib_sa_mcmember_data *dst, ib_sa_comp_mask comp_mask)
+{
+ /* src is group record, dst is request record */
+ /* MGID must already match */
+ /* Port_GID we always replace to our Port_GID, so it is a match */
+
+#define MAD_STATUS_REQ_INVALID 0x0200
+ if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey)
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid)
+ return MAD_STATUS_REQ_INVALID;
+ if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR,
+ IB_SA_MCMEMBER_REC_MTU,
+ src->mtusel_mtu, dst->mtusel_mtu))
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS &&
+ src->tclass != dst->tclass)
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey)
+ return MAD_STATUS_REQ_INVALID;
+ if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR,
+ IB_SA_MCMEMBER_REC_RATE,
+ src->ratesel_rate, dst->ratesel_rate))
+ return MAD_STATUS_REQ_INVALID;
+ if (check_selector(comp_mask,
+ IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR,
+ IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME,
+ src->lifetmsel_lifetm, dst->lifetmsel_lifetm))
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_SL &&
+ (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0xf0000000) !=
+ (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0xf0000000))
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL &&
+ (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x0fffff00) !=
+ (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x0fffff00))
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT &&
+ (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x000000ff) !=
+ (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x000000ff))
+ return MAD_STATUS_REQ_INVALID;
+ if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE &&
+ (src->scope_join_state & 0xf0) !=
+ (dst->scope_join_state & 0xf0))
+ return MAD_STATUS_REQ_INVALID;
+
+ /* join_state checked separately, proxy_join ignored */
+
+ return 0;
+}
+
+/* release group, return 1 if this was last release and group is destroyed
+ * timout work is canceled sync */
+static int release_group(struct mcast_group *group, int from_timeout_handler)
+{
+ struct mlx4_ib_demux_ctx *ctx = group->demux;
+ int nzgroup;
+
+ mutex_lock(&ctx->mcg_table_lock);
+ mutex_lock(&group->lock);
+ if (atomic_dec_and_test(&group->refcount)) {
+ if (!from_timeout_handler) {
+ if (group->state != MCAST_IDLE &&
+ !cancel_delayed_work(&group->timeout_work)) {
+ atomic_inc(&group->refcount);
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ return 0;
+ }
+ }
+
+ nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0);
+ if (!list_empty(&group->pending_list))
+ mcg_warn_group(group, "releasing a group with non empty pending list\n");
+ if (nzgroup)
+ rb_erase(&group->node, &ctx->mcg_table);
+ list_del_init(&group->mgid0_list);
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ kfree(group);
+ return 1;
+ } else {
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ }
+ return 0;
+}
+
+static void adjust_membership(struct mcast_group *group, u8 join_state, int inc)
+{
+ int i;
+
+ for (i = 0; i < 3; i++, join_state >>= 1)
+ if (join_state & 0x1)
+ group->members[i] += inc;
+}
+
+static u8 get_leave_state(struct mcast_group *group)
+{
+ u8 leave_state = 0;
+ int i;
+
+ for (i = 0; i < 3; i++)
+ if (!group->members[i])
+ leave_state |= (1 << i);
+
+ return leave_state & (group->rec.scope_join_state & 7);
+}
+
+static int join_group(struct mcast_group *group, int slave, u8 join_mask)
+{
+ int ret = 0;
+ u8 join_state;
+
+ /* remove bits that slave is already member of, and adjust */
+ join_state = join_mask & (~group->func[slave].join_state);
+ adjust_membership(group, join_state, 1);
+ group->func[slave].join_state |= join_state;
+ if (group->func[slave].state != MCAST_MEMBER && join_state) {
+ group->func[slave].state = MCAST_MEMBER;
+ ret = 1;
+ }
+ return ret;
+}
+
+static int leave_group(struct mcast_group *group, int slave, u8 leave_state)
+{
+ int ret = 0;
+
+ adjust_membership(group, leave_state, -1);
+ group->func[slave].join_state &= ~leave_state;
+ if (!group->func[slave].join_state) {
+ group->func[slave].state = MCAST_NOT_MEMBER;
+ ret = 1;
+ }
+ return ret;
+}
+
+static int check_leave(struct mcast_group *group, int slave, u8 leave_mask)
+{
+ if (group->func[slave].state != MCAST_MEMBER)
+ return MAD_STATUS_REQ_INVALID;
+
+ /* make sure we're not deleting unset bits */
+ if (~group->func[slave].join_state & leave_mask)
+ return MAD_STATUS_REQ_INVALID;
+
+ if (!leave_mask)
+ return MAD_STATUS_REQ_INVALID;
+
+ return 0;
+}
+
+static void mlx4_ib_mcg_timeout_handler(struct work_struct *work)
+{
+ struct delayed_work *delay = to_delayed_work(work);
+ struct mcast_group *group;
+ struct mcast_req *req = NULL;
+
+ group = container_of(delay, typeof(*group), timeout_work);
+
+ mutex_lock(&group->lock);
+ if (group->state == MCAST_JOIN_SENT) {
+ if (!list_empty(&group->pending_list)) {
+ req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ --group->func[req->func].num_pend_reqs;
+ mutex_unlock(&group->lock);
+ kfree(req);
+ if (memcmp(&group->rec.mgid, &mgid0, sizeof mgid0)) {
+ if (release_group(group, 1))
+ return;
+ } else {
+ kfree(group);
+ return;
+ }
+ mutex_lock(&group->lock);
+ } else
+ mcg_warn_group(group, "DRIVER BUG\n");
+ } else if (group->state == MCAST_LEAVE_SENT) {
+ if (group->rec.scope_join_state & 7)
+ group->rec.scope_join_state &= 0xf8;
+ group->state = MCAST_IDLE;
+ mutex_unlock(&group->lock);
+ if (release_group(group, 1))
+ return;
+ mutex_lock(&group->lock);
+ } else
+ mcg_warn_group(group, "invalid state %s\n", get_state_string(group->state));
+ group->state = MCAST_IDLE;
+ atomic_inc(&group->refcount);
+ if (!queue_work(group->demux->mcg_wq, &group->work))
+ safe_atomic_dec(&group->refcount);
+
+ mutex_unlock(&group->lock);
+}
+
+static int handle_leave_req(struct mcast_group *group, u8 leave_mask,
+ struct mcast_req *req)
+{
+ u16 status;
+
+ if (req->clean)
+ leave_mask = group->func[req->func].join_state;
+
+ status = check_leave(group, req->func, leave_mask);
+ if (!status)
+ leave_group(group, req->func, leave_mask);
+
+ if (!req->clean)
+ send_reply_to_slave(req->func, group, &req->sa_mad, status);
+ --group->func[req->func].num_pend_reqs;
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ return 1;
+}
+
+static int handle_join_req(struct mcast_group *group, u8 join_mask,
+ struct mcast_req *req)
+{
+ u8 group_join_state = group->rec.scope_join_state & 7;
+ int ref = 0;
+ u16 status;
+ struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;
+
+ if (join_mask == (group_join_state & join_mask)) {
+ /* port's membership need not change */
+ status = cmp_rec(&group->rec, sa_data, req->sa_mad.sa_hdr.comp_mask);
+ if (!status)
+ join_group(group, req->func, join_mask);
+
+ --group->func[req->func].num_pend_reqs;
+ send_reply_to_slave(req->func, group, &req->sa_mad, status);
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ ++ref;
+ } else {
+ /* port's membership needs to be updated */
+ group->prev_state = group->state;
+ if (send_join_to_wire(group, &req->sa_mad)) {
+ --group->func[req->func].num_pend_reqs;
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ ref = 1;
+ group->state = group->prev_state;
+ } else
+ group->state = MCAST_JOIN_SENT;
+ }
+
+ return ref;
+}
+
+static void mlx4_ib_mcg_work_handler(struct work_struct *work)
+{
+ struct mcast_group *group;
+ struct mcast_req *req = NULL;
+ struct ib_sa_mcmember_data *sa_data;
+ u8 req_join_state;
+ int rc = 1; /* release_count - this is for the scheduled work */
+ u16 status;
+ u8 method;
+
+ group = container_of(work, typeof(*group), work);
+
+ mutex_lock(&group->lock);
+
+ /* First, let's see if a response from SM is waiting regarding this group.
+ * If so, we need to update the group's REC. If this is a bad response, we
+ * may need to send a bad response to a VF waiting for it. If VF is waiting
+ * and this is a good response, the VF will be answered later in this func. */
+ if (group->state == MCAST_RESP_READY) {
+ /* cancels mlx4_ib_mcg_timeout_handler */
+ cancel_delayed_work(&group->timeout_work);
+ status = be16_to_cpu(group->response_sa_mad.mad_hdr.status);
+ method = group->response_sa_mad.mad_hdr.method;
+ if (group->last_req_tid != group->response_sa_mad.mad_hdr.tid) {
+ mcg_warn_group(group, "Got MAD response to existing MGID but wrong TID, dropping. Resp TID=%llx, group TID=%llx\n",
+ be64_to_cpu(group->response_sa_mad.mad_hdr.tid),
+ be64_to_cpu(group->last_req_tid));
+ group->state = group->prev_state;
+ goto process_requests;
+ }
+ if (status) {
+ if (!list_empty(&group->pending_list))
+ req = list_first_entry(&group->pending_list,
+ struct mcast_req, group_list);
+ if ((method == IB_MGMT_METHOD_GET_RESP)) {
+ if (req) {
+ send_reply_to_slave(req->func, group, &req->sa_mad, status);
+ --group->func[req->func].num_pend_reqs;
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ ++rc;
+ } else
+ mcg_warn_group(group, "no request for failed join\n");
+ } else if (method == IB_SA_METHOD_DELETE_RESP && group->demux->flushing)
+ ++rc;
+ } else {
+ u8 resp_join_state;
+ u8 cur_join_state;
+
+ resp_join_state = ((struct ib_sa_mcmember_data *)
+ group->response_sa_mad.data)->scope_join_state & 7;
+ cur_join_state = group->rec.scope_join_state & 7;
+
+ if (method == IB_MGMT_METHOD_GET_RESP) {
+ /* successfull join */
+ if (!cur_join_state && resp_join_state)
+ --rc;
+ } else if (!resp_join_state)
+ ++rc;
+ memcpy(&group->rec, group->response_sa_mad.data, sizeof group->rec);
+ }
+ group->state = MCAST_IDLE;
+ }
+
+process_requests:
+ /* We should now go over pending join/leave requests, as long as we are idle. */
+ while (!list_empty(&group->pending_list) && group->state == MCAST_IDLE) {
+ req = list_first_entry(&group->pending_list, struct mcast_req,
+ group_list);
+ sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;
+ req_join_state = sa_data->scope_join_state & 0x7;
+
+ /* For a leave request, we will immediately answer the VF, and
+ * update our internal counters. The actual leave will be sent
+ * to SM later, if at all needed. We dequeue the request now. */
+ if (req->sa_mad.mad_hdr.method == IB_SA_METHOD_DELETE)
+ rc += handle_leave_req(group, req_join_state, req);
+ else
+ rc += handle_join_req(group, req_join_state, req);
+ }
+
+ /* Handle leaves */
+ if (group->state == MCAST_IDLE) {
+ req_join_state = get_leave_state(group);
+ if (req_join_state) {
+ group->rec.scope_join_state &= ~req_join_state;
+ group->prev_state = group->state;
+ if (send_leave_to_wire(group, req_join_state)) {
+ group->state = group->prev_state;
+ ++rc;
+ } else
+ group->state = MCAST_LEAVE_SENT;
+ }
+ }
+
+ if (!list_empty(&group->pending_list) && group->state == MCAST_IDLE)
+ goto process_requests;
+ mutex_unlock(&group->lock);
+
+ while (rc--)
+ release_group(group, 0);
+}
+
+static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx *ctx,
+ __be64 tid,
+ union ib_gid *new_mgid)
+{
+ struct mcast_group *group = NULL, *cur_group;
+ struct mcast_req *req;
+ struct list_head *pos;
+ struct list_head *n;
+
+ mutex_lock(&ctx->mcg_table_lock);
+ list_for_each_safe(pos, n, &ctx->mcg_mgid0_list) {
+ group = list_entry(pos, struct mcast_group, mgid0_list);
+ mutex_lock(&group->lock);
+ if (group->last_req_tid == tid) {
+ if (memcmp(new_mgid, &mgid0, sizeof mgid0)) {
+ group->rec.mgid = *new_mgid;
+ sprintf(group->name, "%016llx%016llx",
+ be64_to_cpu(group->rec.mgid.global.subnet_prefix),
+ be64_to_cpu(group->rec.mgid.global.interface_id));
+ list_del_init(&group->mgid0_list);
+ cur_group = mcast_insert(ctx, group);
+ if (cur_group) {
+ /* A race between our code and SM. Silently cleaning the new one */
+ req = list_first_entry(&group->pending_list,
+ struct mcast_req, group_list);
+ --group->func[req->func].num_pend_reqs;
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ release_group(group, 0);
+ return NULL;
+ }
+
+ atomic_inc(&group->refcount);
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ return group;
+ } else {
+ struct mcast_req *tmp1, *tmp2;
+
+ list_del(&group->mgid0_list);
+ if (!list_empty(&group->pending_list) && group->state != MCAST_IDLE)
+ cancel_delayed_work_sync(&group->timeout_work);
+
+ list_for_each_entry_safe(tmp1, tmp2, &group->pending_list, group_list) {
+ list_del(&tmp1->group_list);
+ kfree(tmp1);
+ }
+ mutex_unlock(&group->lock);
+ mutex_unlock(&ctx->mcg_table_lock);
+ kfree(group);
+ return NULL;
+ }
+ }
+ mutex_unlock(&group->lock);
+ }
+ mutex_unlock(&ctx->mcg_table_lock);
+
+ return NULL;
+}
+
+static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
+ union ib_gid *mgid, int create,
+ gfp_t gfp_mask)
+{
+ struct mcast_group *group, *cur_group;
+ int is_mgid0;
+ int i;
+
+ is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
+ if (!is_mgid0) {
+ group = mcast_find(ctx, mgid);
+ if (group)
+ goto found;
+ }
+
+ if (!create)
+ return ERR_PTR(-ENOENT);
+
+ group = kzalloc(sizeof *group, gfp_mask);
+ if (!group)
+ return ERR_PTR(-ENOMEM);
+
+ group->demux = ctx;
+ group->rec.mgid = *mgid;
+ INIT_LIST_HEAD(&group->pending_list);
+ INIT_LIST_HEAD(&group->mgid0_list);
+ for (i = 0; i < MAX_VFS; ++i)
+ INIT_LIST_HEAD(&group->func[i].pending);
+ INIT_WORK(&group->work, mlx4_ib_mcg_work_handler);
+ INIT_DELAYED_WORK(&group->timeout_work, mlx4_ib_mcg_timeout_handler);
+ mutex_init(&group->lock);
+ sprintf(group->name, "%016llx%016llx",
+ be64_to_cpu(group->rec.mgid.global.subnet_prefix),
+ be64_to_cpu(group->rec.mgid.global.interface_id));
+ group->state = MCAST_IDLE;
+
+ if (is_mgid0) {
+ list_add(&group->mgid0_list, &ctx->mcg_mgid0_list);
+ goto found;
+ }
+
+ cur_group = mcast_insert(ctx, group);
+ if (cur_group) {
+ mcg_warn("group just showed up %s - confused\n", cur_group->name);
+ kfree(group);
+ return ERR_PTR(-EINVAL);
+ }
+
+found:
+ atomic_inc(&group->refcount);
+ return group;
+}
+
+static void queue_req(struct mcast_req *req)
+{
+ struct mcast_group *group = req->group;
+
+ atomic_inc(&group->refcount); /* for the request */
+ atomic_inc(&group->refcount); /* for scheduling the work */
+ list_add_tail(&req->group_list, &group->pending_list);
+ list_add_tail(&req->func_list, &group->func[req->func].pending);
+ /* calls mlx4_ib_mcg_work_handler */
+ if (!queue_work(group->demux->mcg_wq, &group->work))
+ safe_atomic_dec(&group->refcount);
+}
+
+int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
+ struct ib_sa_mad *mad)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)mad->data;
+ struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+ struct mcast_group *group;
+
+ switch (mad->mad_hdr.method) {
+ case IB_MGMT_METHOD_GET_RESP:
+ case IB_SA_METHOD_DELETE_RESP:
+ mutex_lock(&ctx->mcg_table_lock);
+ group = acquire_group(ctx, &rec->mgid, 0, GFP_KERNEL);
+ mutex_unlock(&ctx->mcg_table_lock);
+ if (IS_ERR(group)) {
+ if (mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP) {
+ __be64 tid = mad->mad_hdr.tid;
+ *(u8 *)(&tid) = (u8)slave; /* in group we kept the modified TID */
+ group = search_relocate_mgid0_group(ctx, tid, &rec->mgid);
+ } else
+ group = NULL;
+ }
+
+ if (!group)
+ return 1;
+
+ mutex_lock(&group->lock);
+ group->response_sa_mad = *mad;
+ group->prev_state = group->state;
+ group->state = MCAST_RESP_READY;
+ /* calls mlx4_ib_mcg_work_handler */
+ atomic_inc(&group->refcount);
+ if (!queue_work(ctx->mcg_wq, &group->work))
+ safe_atomic_dec(&group->refcount);
+ mutex_unlock(&group->lock);
+ release_group(group, 0);
+ return 1; /* consumed */
+ case IB_MGMT_METHOD_SET:
+ case IB_SA_METHOD_GET_TABLE:
+ case IB_SA_METHOD_GET_TABLE_RESP:
+ case IB_SA_METHOD_DELETE:
+ return 0; /* not consumed, pass-through to guest over tunnel */
+ default:
+ mcg_warn("In demux, port %d: unexpected MCMember method: 0x%x, dropping\n",
+ port, mad->mad_hdr.method);
+ return 1; /* consumed */
+ }
+}
+
+int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port,
+ int slave, struct ib_sa_mad *sa_mad)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)sa_mad->data;
+ struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+ struct mcast_group *group;
+ struct mcast_req *req;
+ int may_create = 0;
+
+ if (ctx->flushing)
+ return -EAGAIN;
+
+ switch (sa_mad->mad_hdr.method) {
+ case IB_MGMT_METHOD_SET:
+ may_create = 1;
+ case IB_SA_METHOD_DELETE:
+ req = kzalloc(sizeof *req, GFP_KERNEL);
+ if (!req)
+ return -ENOMEM;
+
+ req->func = slave;
+ req->sa_mad = *sa_mad;
+
+ mutex_lock(&ctx->mcg_table_lock);
+ group = acquire_group(ctx, &rec->mgid, may_create, GFP_KERNEL);
+ mutex_unlock(&ctx->mcg_table_lock);
+ if (IS_ERR(group)) {
+ kfree(req);
+ return PTR_ERR(group);
+ }
+ mutex_lock(&group->lock);
+ if (group->func[slave].num_pend_reqs > MAX_PEND_REQS_PER_FUNC) {
+ mutex_unlock(&group->lock);
+ mcg_warn_group(group, "Port %d, Func %d has too many pending requests (%d), dropping\n",
+ port, slave, MAX_PEND_REQS_PER_FUNC);
+ release_group(group, 0);
+ kfree(req);
+ return -ENOMEM;
+ }
+ ++group->func[slave].num_pend_reqs;
+ req->group = group;
+ queue_req(req);
+ mutex_unlock(&group->lock);
+ release_group(group, 0);
+ return 1; /* consumed */
+ case IB_SA_METHOD_GET_TABLE:
+ case IB_MGMT_METHOD_GET_RESP:
+ case IB_SA_METHOD_GET_TABLE_RESP:
+ case IB_SA_METHOD_DELETE_RESP:
+ return 0; /* not consumed, pass-through */
+ default:
+ mcg_warn("In multiplex, port %d, func %d: unexpected MCMember method: 0x%x, dropping\n",
+ port, slave, sa_mad->mad_hdr.method);
+ return 1; /* consumed */
+ }
+}
+
+int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx)
+{
+ char name[20];
+
+ atomic_set(&ctx->tid, 0);
+ sprintf(name, "mlx4_ib_mcg%d", ctx->port);
+ ctx->mcg_wq = create_singlethread_workqueue(name);
+ if (!ctx->mcg_wq)
+ return -ENOMEM;
+
+ mutex_init(&ctx->mcg_table_lock);
+ ctx->mcg_table = RB_ROOT;
+ INIT_LIST_HEAD(&ctx->mcg_mgid0_list);
+ ctx->flushing = 0;
+
+ return 0;
+}
+
+static void force_clean_group(struct mcast_group *group)
+{
+ struct mcast_req *req, *tmp
+ ;
+ list_for_each_entry_safe(req, tmp, &group->pending_list, group_list) {
+ list_del(&req->group_list);
+ kfree(req);
+ }
+ rb_erase(&group->node, &group->demux->mcg_table);
+ kfree(group);
+}
+
+static void _mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
+{
+ int i;
+ struct rb_node *p;
+ struct mcast_group *group;
+ unsigned long end;
+ int count;
+
+ if (ctx->flushing)
+ return;
+
+ ctx->flushing = 1;
+ for (i = 0; i < MAX_VFS; ++i)
+ clean_vf_mcast(ctx, i);
+
+ end = jiffies + msecs_to_jiffies(MAD_TIMEOUT_MS + 3000);
+ do {
+ count = 0;
+ mutex_lock(&ctx->mcg_table_lock);
+ for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p))
+ ++count;
+ mutex_unlock(&ctx->mcg_table_lock);
+ if (!count)
+ break;
+
+ msleep(1);
+ } while (time_after(end, jiffies));
+
+ flush_workqueue(ctx->mcg_wq);
+ if (destroy_wq)
+ destroy_workqueue(ctx->mcg_wq);
+
+ mutex_lock(&ctx->mcg_table_lock);
+ while ((p = rb_first(&ctx->mcg_table)) != NULL) {
+ group = rb_entry(p, struct mcast_group, node);
+ if (atomic_read(&group->refcount))
+ mcg_warn_group(group, "group refcount %d!!! (pointer %p)\n", atomic_read(&group->refcount), group);
+
+ force_clean_group(group);
+ }
+ mutex_unlock(&ctx->mcg_table_lock);
+
+ if (!destroy_wq)
+ ctx->flushing = 0;
+}
+
+struct clean_work {
+ struct work_struct work;
+ struct mlx4_ib_demux_ctx *ctx;
+ int destroy_wq;
+};
+
+static void mcg_clean_task(struct work_struct *work)
+{
+ struct clean_work *cw = container_of(work, struct clean_work, work);
+
+ _mlx4_ib_mcg_port_cleanup(cw->ctx, cw->destroy_wq);
+ kfree(cw);
+}
+
+void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
+{
+ struct clean_work *work;
+
+ if (destroy_wq) {
+ _mlx4_ib_mcg_port_cleanup(ctx, destroy_wq);
+ return;
+ }
+
+ work = kmalloc(sizeof *work, GFP_KERNEL);
+ if (!work) {
+ mcg_warn("failed allocating work for cleanup\n");
+ return;
+ }
+
+ work->ctx = ctx;
+ work->destroy_wq = destroy_wq;
+ INIT_WORK(&work->work, mcg_clean_task);
+ queue_work(clean_wq, &work->work);
+}
+
+static void build_leave_mad(struct mcast_req *req)
+{
+ struct ib_sa_mad *mad = &req->sa_mad;
+
+ mad->mad_hdr.method = IB_SA_METHOD_DELETE;
+}
+
+
+static void clear_pending_reqs(struct mcast_group *group, int vf)
+{
+ struct mcast_req *req, *tmp, *group_first = NULL;
+ int clear;
+ int pend = 0;
+
+ if (!list_empty(&group->pending_list))
+ group_first = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+
+ list_for_each_entry_safe(req, tmp, &group->func[vf].pending, func_list) {
+ clear = 1;
+ if (group_first == req &&
+ (group->state == MCAST_JOIN_SENT ||
+ group->state == MCAST_LEAVE_SENT)) {
+ clear = cancel_delayed_work(&group->timeout_work);
+ pend = !clear;
+ group->state = MCAST_IDLE;
+ }
+ if (clear) {
+ --group->func[vf].num_pend_reqs;
+ list_del(&req->group_list);
+ list_del(&req->func_list);
+ kfree(req);
+ atomic_dec(&group->refcount);
+ }
+ }
+
+ if (!pend && (!list_empty(&group->func[vf].pending) || group->func[vf].num_pend_reqs)) {
+ mcg_warn_group(group, "DRIVER BUG: list_empty %d, num_pend_reqs %d\n",
+ list_empty(&group->func[vf].pending), group->func[vf].num_pend_reqs);
+ }
+}
+
+static int push_deleteing_req(struct mcast_group *group, int slave)
+{
+ struct mcast_req *req;
+ struct mcast_req *pend_req;
+
+ if (!group->func[slave].join_state)
+ return 0;
+
+ req = kzalloc(sizeof *req, GFP_KERNEL);
+ if (!req) {
+ mcg_warn_group(group, "failed allocation - may leave stall groups\n");
+ return -ENOMEM;
+ }
+
+ if (!list_empty(&group->func[slave].pending)) {
+ pend_req = list_entry(group->func[slave].pending.prev, struct mcast_req, group_list);
+ if (pend_req->clean) {
+ kfree(req);
+ return 0;
+ }
+ }
+
+ req->clean = 1;
+ req->func = slave;
+ req->group = group;
+ ++group->func[slave].num_pend_reqs;
+ build_leave_mad(req);
+ queue_req(req);
+ return 0;
+}
+
+void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave)
+{
+ struct mcast_group *group;
+ struct rb_node *p;
+
+ mutex_lock(&ctx->mcg_table_lock);
+ for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p)) {
+ group = rb_entry(p, struct mcast_group, node);
+ mutex_lock(&group->lock);
+ if (atomic_read(&group->refcount)) {
+ /* clear pending requests of this VF */
+ clear_pending_reqs(group, slave);
+ push_deleteing_req(group, slave);
+ }
+ mutex_unlock(&group->lock);
+ }
+ mutex_unlock(&ctx->mcg_table_lock);
+}
+
+
+int mlx4_ib_mcg_init(void)
+{
+ clean_wq = create_singlethread_workqueue("mlx4_ib_mcg");
+ if (!clean_wq)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void mlx4_ib_mcg_destroy(void)
+{
+ destroy_workqueue(clean_wq);
+}
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index ac71d56..01ba9f1 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -37,9 +37,11 @@
#include <linux/compiler.h>
#include <linux/list.h>
#include <linux/mutex.h>
+#include <linux/idr.h>
#include <rdma/ib_verbs.h>
#include <rdma/ib_umem.h>
+#include <rdma/ib_mad.h>
#include <linux/mlx4/device.h>
#include <linux/mlx4/doorbell.h>
@@ -329,7 +331,14 @@ struct mlx4_ib_demux_ctx {
__be64 subnet_prefix;
__be64 guid_cache[128];
struct mlx4_ib_dev *dev;
+ /* the following lock protects both mcg_table and mcg_mgid0_list */
+ struct mutex mcg_table_lock;
+ struct rb_root mcg_table;
+ struct list_head mcg_mgid0_list;
+ struct workqueue_struct *mcg_wq;
struct mlx4_ib_demux_pv_ctx **tun;
+ atomic_t tid;
+ int flushing; /* flushing the work queue */
};
struct mlx4_ib_sriov {
@@ -553,6 +562,19 @@ static inline int mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah)
return !!(ah->av.ib.g_slid & 0x80);
}
+int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx);
+void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq);
+void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave);
+int mlx4_ib_mcg_init(void);
+void mlx4_ib_mcg_destroy(void);
+
+int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid);
+
+int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port, int slave,
+ struct ib_sa_mad *sa_mad);
+int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
+ struct ib_sa_mad *mad);
+
int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
union ib_gid *gid);
@@ -561,4 +583,12 @@ void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
void mlx4_ib_tunnels_update_work(struct work_struct *work);
+int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
+ enum ib_qp_type qpt, struct ib_wc *wc,
+ struct ib_grh *grh, struct ib_mad *mad);
+int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
+ enum ib_qp_type dest_qpt, u16 pkey_index, u32 remote_qpn,
+ u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad);
+__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx);
+
#endif /* MLX4_IB_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 11/22] IB/mlx4: Add CM paravirtualization
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (9 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 10/22] IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-12-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 12/22] net/mlx4_core: Add IB port-state machine, and port mgmt event propagation infrastructure Jack Morgenstein
` (12 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Amir Vadai, Jack Morgenstein
From: Amir Vadai <amirv-VPRAkNaXOzVS1MOuV/RT9w@public.gmane.org>
In CM para-virtualization:
1. Incoming requests are steered to the correct vHCA according to the embedded GID
2. Communication IDs on outgoing requests are replaced by a globally unique ID, generated
by the PPF, since there is no synchronization of ID generation between guests (and so these
IDs are not guaranteed to be globally unique).
The guest's comm ID is stored, and is returned to the response MAD when it arrives.
Signed-off-by: Amir Vadai <amirv-VPRAkNaXOzVS1MOuV/RT9w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/cm.c | 437 ++++++++++++++++++++++++++++++++++
drivers/infiniband/hw/mlx4/mad.c | 16 ++-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 15 ++
4 files changed, 468 insertions(+), 2 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/cm.c
diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile
index 20d627d..bf0aa90 100644
--- a/drivers/infiniband/hw/mlx4/Makefile
+++ b/drivers/infiniband/hw/mlx4/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o
-mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o
+mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
new file mode 100644
index 0000000..0a52146
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -0,0 +1,437 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+
+#include <linux/mlx4/cmd.h>
+#include <linux/rbtree.h>
+#include <linux/idr.h>
+#include <rdma/ib_cm.h>
+
+#include "mlx4_ib.h"
+
+#define CM_CLEANUP_CACHE_TIMEOUT (5 * HZ)
+
+struct id_map_entry {
+ struct rb_node node;
+
+ u32 sl_cm_id;
+ u32 pv_cm_id;
+ int slave_id;
+ int scheduled_delete;
+ struct mlx4_ib_dev *dev;
+
+ struct list_head list;
+ struct delayed_work timeout;
+};
+
+struct cm_generic_msg {
+ struct ib_mad_hdr hdr;
+
+ __be32 local_comm_id;
+ __be32 remote_comm_id;
+};
+
+struct cm_req_msg {
+ unsigned char unused[0x60];
+ union ib_gid primary_path_sgid;
+};
+
+
+static void set_local_comm_id(struct ib_mad *mad, u32 cm_id)
+{
+ struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+ msg->local_comm_id = cpu_to_be32(cm_id);
+}
+
+static u32 get_local_comm_id(struct ib_mad *mad)
+{
+ struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+
+ return be32_to_cpu(msg->local_comm_id);
+}
+
+static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id)
+{
+ struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+ msg->remote_comm_id = cpu_to_be32(cm_id);
+}
+
+static u32 get_remote_comm_id(struct ib_mad *mad)
+{
+ struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+
+ return be32_to_cpu(msg->remote_comm_id);
+}
+
+static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad)
+{
+ struct cm_req_msg *msg = (struct cm_req_msg *)mad;
+
+ return msg->primary_path_sgid;
+}
+
+/* Lock should be taken before called */
+static struct id_map_entry *
+id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id)
+{
+ struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+ struct rb_node *node = sl_id_map->rb_node;
+
+ while (node) {
+ struct id_map_entry *id_map_entry =
+ rb_entry(node, struct id_map_entry, node);
+
+ if (id_map_entry->sl_cm_id > sl_cm_id)
+ node = node->rb_left;
+ else if (id_map_entry->sl_cm_id < sl_cm_id)
+ node = node->rb_right;
+ else if (id_map_entry->slave_id > slave_id)
+ node = node->rb_left;
+ else if (id_map_entry->slave_id < slave_id)
+ node = node->rb_right;
+ else
+ return id_map_entry;
+ }
+ return NULL;
+}
+
+static void id_map_ent_timeout(struct work_struct *work)
+{
+ struct delayed_work *delay = to_delayed_work(work);
+ struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout);
+ struct id_map_entry *db_ent, *found_ent;
+ struct mlx4_ib_dev *dev = ent->dev;
+ struct mlx4_ib_sriov *sriov = &dev->sriov;
+ struct rb_root *sl_id_map = &sriov->sl_id_map;
+ int pv_id = (int) ent->pv_cm_id;
+
+ spin_lock(&sriov->id_map_lock);
+ db_ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_id);
+ if (!db_ent)
+ goto out;
+ found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id);
+ if (found_ent && found_ent == ent)
+ rb_erase(&found_ent->node, sl_id_map);
+ idr_remove(&sriov->pv_id_table, pv_id);
+
+out:
+ list_del(&ent->list);
+ spin_unlock(&sriov->id_map_lock);
+ kfree(ent);
+}
+
+static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
+{
+ struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ struct rb_root *sl_id_map = &sriov->sl_id_map;
+ struct id_map_entry *ent, *found_ent;
+
+ spin_lock(&sriov->id_map_lock);
+ ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_cm_id);
+ if (!ent)
+ goto out;
+ found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id);
+ if (found_ent && found_ent == ent)
+ rb_erase(&found_ent->node, sl_id_map);
+ idr_remove(&sriov->pv_id_table, pv_cm_id);
+out:
+ spin_unlock(&sriov->id_map_lock);
+}
+
+static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
+{
+ struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+ struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
+ struct id_map_entry *ent;
+ int slave_id = new->slave_id;
+ int sl_cm_id = new->sl_cm_id;
+
+ ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
+ if (ent) {
+ pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n",
+ sl_cm_id);
+
+ rb_replace_node(&ent->node, &new->node, sl_id_map);
+ return;
+ }
+
+ /* Go to the bottom of the tree */
+ while (*link) {
+ parent = *link;
+ ent = rb_entry(parent, struct id_map_entry, node);
+
+ if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id))
+ link = &(*link)->rb_left;
+ else
+ link = &(*link)->rb_right;
+ }
+
+ rb_link_node(&new->node, parent, link);
+ rb_insert_color(&new->node, sl_id_map);
+}
+
+static struct id_map_entry *
+id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
+{
+ int ret, id;
+ static int next_id;
+ struct id_map_entry *ent;
+ struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+
+ ent = kmalloc(sizeof(struct id_map_entry), GFP_KERNEL);
+ if (!ent) {
+ mlx4_ib_warn(ibdev, "Couldn't allocate id cache entry - out of memory\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ ent->sl_cm_id = sl_cm_id;
+ ent->slave_id = slave_id;
+ ent->scheduled_delete = 0;
+ ent->dev = to_mdev(ibdev);
+ INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
+
+ do {
+ spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
+ ret = idr_get_new_above(&sriov->pv_id_table, ent,
+ next_id, &id);
+ if (!ret) {
+ next_id = ((unsigned) id + 1) & MAX_ID_MASK;
+ ent->pv_cm_id = (u32)id;
+ sl_id_map_add(ibdev, ent);
+ }
+
+ spin_unlock(&sriov->id_map_lock);
+ } while (ret == -EAGAIN && idr_pre_get(&sriov->pv_id_table, GFP_KERNEL));
+ /*the function idr_get_new_above can return -ENOSPC, so don't insert in that case.*/
+ if (!ret) {
+ spin_lock(&sriov->id_map_lock);
+ list_add_tail(&ent->list, &sriov->cm_list);
+ spin_unlock(&sriov->id_map_lock);
+ return ent;
+ }
+ /*error flow*/
+ kfree(ent);
+ mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret);
+ return ERR_PTR(-ENOMEM);
+}
+
+static struct id_map_entry *
+id_map_get(struct ib_device *ibdev, int *pv_cm_id, int sl_cm_id, int slave_id)
+{
+ struct id_map_entry *ent;
+ struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+
+ spin_lock(&sriov->id_map_lock);
+ if (*pv_cm_id == -1) {
+ ent = id_map_find_by_sl_id(ibdev, sl_cm_id, slave_id);
+ if (ent)
+ *pv_cm_id = (int) ent->pv_cm_id;
+ } else
+ ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, *pv_cm_id);
+ spin_unlock(&sriov->id_map_lock);
+
+ return ent;
+}
+
+static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
+{
+ struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ unsigned long flags;
+
+ spin_lock_irqsave(&sriov->going_down_lock, flags);
+ spin_lock(&sriov->id_map_lock);
+ /*make sure that there is no schedule inside the scheduled work.*/
+ if (!sriov->is_going_down) {
+ id->scheduled_delete = 1;
+ schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
+ }
+ spin_unlock(&sriov->id_map_lock);
+ spin_unlock_irqrestore(&sriov->going_down_lock, flags);
+}
+
+int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
+ struct ib_mad *mad)
+{
+ struct id_map_entry *id;
+ u32 sl_cm_id;
+ int pv_cm_id = -1;
+
+ sl_cm_id = get_local_comm_id(mad);
+
+ if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
+ mad->mad_hdr.attr_id == CM_REP_ATTR_ID) {
+ id = id_map_alloc(ibdev, slave_id, sl_cm_id);
+ if (IS_ERR(id)) {
+ mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n",
+ __func__, slave_id, sl_cm_id);
+ return PTR_ERR(id);
+ }
+ } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID) {
+ return 0;
+ } else {
+ id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
+ }
+
+ if (!id) {
+ pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
+ slave_id, sl_cm_id);
+ return -EINVAL;
+ }
+
+ set_local_comm_id(mad, id->pv_cm_id);
+
+ if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
+ schedule_delayed(ibdev, id);
+ else if (mad->mad_hdr.attr_id == CM_DREP_ATTR_ID)
+ id_map_find_del(ibdev, pv_cm_id);
+
+ return 0;
+}
+
+int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
+ struct ib_mad *mad)
+{
+ u32 pv_cm_id;
+ struct id_map_entry *id;
+
+ if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID) {
+ union ib_gid gid;
+
+ gid = gid_from_req_msg(ibdev, mad);
+ *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
+ if (*slave < 0) {
+ mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n",
+ gid.global.interface_id);
+ return -ENOENT;
+ }
+ return 0;
+ }
+
+ pv_cm_id = get_remote_comm_id(mad);
+ id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
+
+ if (!id) {
+ pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
+ return -ENOENT;
+ }
+
+ *slave = id->slave_id;
+ set_remote_comm_id(mad, id->sl_cm_id);
+
+ if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
+ schedule_delayed(ibdev, id);
+ else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
+ mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) {
+ id_map_find_del(ibdev, (int) pv_cm_id);
+ }
+
+ return 0;
+}
+
+void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
+{
+ spin_lock_init(&dev->sriov.id_map_lock);
+ INIT_LIST_HEAD(&dev->sriov.cm_list);
+ dev->sriov.sl_id_map = RB_ROOT;
+ idr_init(&dev->sriov.pv_id_table);
+ idr_pre_get(&dev->sriov.pv_id_table, GFP_KERNEL);
+}
+
+/* slave = -1 ==> all slaves */
+/* TBD -- call paravirt clean for single slave. Need for slave RESET event */
+void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
+{
+ struct mlx4_ib_sriov *sriov = &dev->sriov;
+ struct rb_root *sl_id_map = &sriov->sl_id_map;
+ struct list_head lh;
+ struct rb_node *nd;
+ int need_flush = 1;
+ struct id_map_entry *map, *tmp_map;
+ /* cancel all delayed work queue entries */
+ INIT_LIST_HEAD(&lh);
+ spin_lock(&sriov->id_map_lock);
+ list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+ if (slave < 0 || slave == map->slave_id) {
+ if (map->scheduled_delete)
+ need_flush &= !!cancel_delayed_work(&map->timeout);
+ }
+ }
+
+ spin_unlock(&sriov->id_map_lock);
+
+ if (!need_flush)
+ flush_scheduled_work(); /* make sure all timers were flushed */
+
+ /* now, remove all leftover entries from databases*/
+ spin_lock(&sriov->id_map_lock);
+ if (slave < 0) {
+ while (rb_first(sl_id_map)) {
+ struct id_map_entry *ent =
+ rb_entry(rb_first(sl_id_map),
+ struct id_map_entry, node);
+
+ rb_erase(&ent->node, sl_id_map);
+ idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id);
+ }
+ list_splice_init(&dev->sriov.cm_list, &lh);
+ } else {
+ /* first, move nodes belonging to slave to db remove list */
+ nd = rb_first(sl_id_map);
+ while (nd) {
+ struct id_map_entry *ent =
+ rb_entry(nd, struct id_map_entry, node);
+ nd = rb_next(nd);
+ if (ent->slave_id == slave)
+ list_move_tail(&ent->list, &lh);
+ }
+ /* remove those nodes from databases */
+ list_for_each_entry_safe(map, tmp_map, &lh, list) {
+ rb_erase(&map->node, sl_id_map);
+ idr_remove(&sriov->pv_id_table, (int) map->pv_cm_id);
+ }
+
+ /* add remaining nodes from cm_list */
+ list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+ if (slave == map->slave_id)
+ list_move_tail(&map->list, &lh);
+ }
+ }
+
+ spin_unlock(&sriov->id_map_lock);
+
+ /* free any map entries left behind due to cancel_delayed_work above */
+ list_for_each_entry_safe(map, tmp_map, &lh, list) {
+ list_del(&map->list);
+ kfree(map);
+ }
+}
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 195b211..ef40254 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -540,6 +540,10 @@ static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
(struct ib_sa_mad *) mad))
return 0;
break;
+ case IB_MGMT_CLASS_CM:
+ if (mlx4_ib_demux_cm_handler(ibdev, port, &slave, mad))
+ return 0;
+ break;
case IB_MGMT_CLASS_DEVICE_MGMT:
if (mad->mad_hdr.method != IB_MGMT_METHOD_GET_RESP)
return 0;
@@ -1072,6 +1076,11 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc
(struct ib_sa_mad *) &tunnel->mad))
return;
break;
+ case IB_MGMT_CLASS_CM:
+ if (mlx4_ib_multiplex_cm_handler(ctx->ib_dev, ctx->port, slave,
+ (struct ib_mad *) &tunnel->mad))
+ return;
+ break;
case IB_MGMT_CLASS_DEVICE_MGMT:
if (tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_GET &&
tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_SET)
@@ -1786,6 +1795,7 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
dev->sriov.is_going_down = 0;
spin_lock_init(&dev->sriov.going_down_lock);
+ mlx4_ib_cm_paravirt_init(dev);
mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n");
@@ -1814,6 +1824,7 @@ demux_err:
mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
--i;
}
+ mlx4_ib_cm_paravirt_clean(dev, -1);
return err;
}
@@ -1829,7 +1840,7 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
dev->sriov.is_going_down = 1;
spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
- if (mlx4_is_master(dev->dev))
+ if (mlx4_is_master(dev->dev)) {
for (i = 0; i < dev->num_ports; i++) {
flush_workqueue(dev->sriov.demux[i].ud_wq);
mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]);
@@ -1837,4 +1848,7 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
dev->sriov.sqps[i] = NULL;
mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
}
+
+ mlx4_ib_cm_paravirt_clean(dev, -1);
+ }
}
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 01ba9f1..7476e24 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -348,6 +348,12 @@ struct mlx4_ib_sriov {
* it may be called from interrupt context.*/
spinlock_t going_down_lock;
int is_going_down;
+
+ /* CM paravirtualization fields */
+ struct list_head cm_list;
+ spinlock_t id_map_lock;
+ struct rb_root sl_id_map;
+ struct idr pv_id_table;
};
struct mlx4_ib_iboe {
@@ -591,4 +597,13 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad);
__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx);
+int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
+ struct ib_mad *mad);
+
+int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
+ struct ib_mad *mad);
+
+void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev);
+void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave_id);
+
#endif /* MLX4_IB_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 12/22] net/mlx4_core: Add IB port-state machine, and port mgmt event propagation infrastructure
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (10 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 11/22] IB/mlx4: Add CM paravirtualization Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 13/22] {NET,IB}/mlx4: Add alias_guid mechanism Jack Morgenstein
` (11 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
For an IB port, a slave should not show port active until that
slave has a valid alias-guid (provided by the subnet manager).
Therefore the port-up event should be passed to a slave only after
both the port is up, and the slave's alias-guid has been set.
Also, provide the infrastructure for propagating port-management
events (client-reregister, etc) to slaves.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/net/ethernet/mellanox/mlx4/eq.c | 237 +++++++++++++++++++++++++++--
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 +
include/linux/mlx4/device.h | 28 ++++
3 files changed, 250 insertions(+), 16 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx4/eq.c b/drivers/net/ethernet/mellanox/mlx4/eq.c
index 99a0464..c425826 100644
--- a/drivers/net/ethernet/mellanox/mlx4/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx4/eq.c
@@ -200,6 +200,196 @@ static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
slave_event(dev, slave, eqe);
}
+int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
+{
+ struct mlx4_eqe eqe;
+
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
+
+ if (!s_slave->active)
+ return 0;
+
+ memset(&eqe, 0, sizeof eqe);
+
+ eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
+ eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
+ eqe.event.port_mgmt_change.port = port;
+
+ return mlx4_GEN_EQE(dev, slave, &eqe);
+}
+EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
+
+int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
+{
+ struct mlx4_eqe eqe;
+
+ /*don't send if we don't have the that slave */
+ if (dev->num_vfs < slave)
+ return 0;
+ memset(&eqe, 0, sizeof eqe);
+
+ eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
+ eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
+ eqe.event.port_mgmt_change.port = port;
+
+ return mlx4_GEN_EQE(dev, slave, &eqe);
+}
+EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
+
+int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
+ u8 port_subtype_change)
+{
+ struct mlx4_eqe eqe;
+
+ /*don't send if we don't have the that slave */
+ if (dev->num_vfs < slave)
+ return 0;
+ memset(&eqe, 0, sizeof eqe);
+
+ eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
+ eqe.subtype = port_subtype_change;
+ eqe.event.port_change.port = cpu_to_be32(port << 28);
+
+ mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
+ port_subtype_change, slave, port);
+ return mlx4_GEN_EQE(dev, slave, &eqe);
+}
+EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
+
+enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
+ if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS) {
+ pr_err("%s: Error: asking for slave:%d, port:%d\n",
+ __func__, slave, port);
+ return SLAVE_PORT_DOWN;
+ }
+ return s_state[slave].port_state[port];
+}
+EXPORT_SYMBOL(mlx4_get_slave_port_state);
+
+static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
+ enum slave_port_state state)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
+
+ if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
+ pr_err("%s: Error: asking for slave:%d, port:%d\n",
+ __func__, slave, port);
+ return -1;
+ }
+ s_state[slave].port_state[port] = state;
+
+ return 0;
+}
+
+static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
+{
+ int i;
+ enum slave_port_gen_event gen_event;
+
+ for (i = 0; i < dev->num_slaves; i++)
+ set_and_calc_slave_port_state(dev, i, port, event, &gen_event);
+}
+/**************************************************************************
+ The function get as input the new event to that port,
+ and according to the prev state change the slave's port state.
+ The events are:
+ MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
+ MLX4_PORT_STATE_DEV_EVENT_PORT_UP
+ MLX4_PORT_STATE_IB_EVENT_GID_VALID
+ MLX4_PORT_STATE_IB_EVENT_GID_INVALID
+***************************************************************************/
+int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
+ u8 port, int event,
+ enum slave_port_gen_event *gen_event)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ struct mlx4_slave_state *ctx = NULL;
+ unsigned long flags;
+ int ret = -1;
+ enum slave_port_state cur_state =
+ mlx4_get_slave_port_state(dev, slave, port);
+
+ *gen_event = SLAVE_PORT_GEN_EVENT_NONE;
+
+ if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
+ pr_err("%s: Error: asking for slave:%d, port:%d\n",
+ __func__, slave, port);
+ return ret;
+ }
+
+ ctx = &priv->mfunc.master.slave_state[slave];
+ spin_lock_irqsave(&ctx->lock, flags);
+
+ mlx4_dbg(dev, "%s: slave: %d, current state: %d new event :%d\n",
+ __func__, slave, cur_state, event);
+
+ switch (cur_state) {
+ case SLAVE_PORT_DOWN:
+ if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
+ mlx4_set_slave_port_state(dev, slave, port,
+ SLAVE_PENDING_UP);
+ break;
+ case SLAVE_PENDING_UP:
+ if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
+ mlx4_set_slave_port_state(dev, slave, port,
+ SLAVE_PORT_DOWN);
+ else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
+ mlx4_set_slave_port_state(dev, slave, port,
+ SLAVE_PORT_UP);
+ *gen_event = SLAVE_PORT_GEN_EVENT_UP;
+ }
+ break;
+ case SLAVE_PORT_UP:
+ if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
+ mlx4_set_slave_port_state(dev, slave, port,
+ SLAVE_PORT_DOWN);
+ *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
+ } else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
+ event) {
+ mlx4_set_slave_port_state(dev, slave, port,
+ SLAVE_PENDING_UP);
+ *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
+ }
+ break;
+ default:
+ pr_err("%s: BUG!!! UNKNOWN state: "
+ "slave:%d, port:%d\n", __func__, slave, port);
+ goto out;
+ }
+ ret = mlx4_get_slave_port_state(dev, slave, port);
+ mlx4_dbg(dev, "%s: slave: %d, current state: %d new event"
+ " :%d gen_event: %d\n",
+ __func__, slave, cur_state, event, *gen_event);
+
+out:
+ spin_unlock_irqrestore(&ctx->lock, flags);
+ return ret;
+}
+
+EXPORT_SYMBOL(set_and_calc_slave_port_state);
+
+int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
+{
+ struct mlx4_eqe eqe;
+
+ memset(&eqe, 0, sizeof eqe);
+
+ eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
+ eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
+ eqe.event.port_mgmt_change.port = port;
+ eqe.event.port_mgmt_change.params.port_info.changed_attr =
+ cpu_to_be32((u32) attr);
+
+ slave_event(dev, ALL_SLAVES, &eqe);
+ return 0;
+}
+EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
+
void mlx4_master_handle_slave_flr(struct work_struct *work)
{
struct mlx4_mfunc_master_ctx *master =
@@ -251,6 +441,7 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
u32 flr_slave;
u8 update_slave_state;
int i;
+ enum slave_port_gen_event gen_event;
while ((eqe = next_eqe_sw(eq))) {
/*
@@ -347,35 +538,49 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
case MLX4_EVENT_TYPE_PORT_CHANGE:
port = be32_to_cpu(eqe->event.port_change.port) >> 28;
if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
- mlx4_dispatch_event(dev,
- MLX4_DEV_EVENT_PORT_DOWN,
+ mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
port);
mlx4_priv(dev)->sense.do_sense_port[port] = 1;
- if (mlx4_is_master(dev))
- /*change the state of all slave's port
- * to down:*/
- for (i = 0; i < dev->num_slaves; i++) {
- mlx4_dbg(dev, "%s: Sending "
- "MLX4_PORT_CHANGE_SUBTYPE_DOWN"
+ if (!mlx4_is_master(dev))
+ break;
+ for (i = 0; i < dev->num_slaves; i++) {
+ if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
+ if (i == mlx4_master_func_num(dev))
+ continue;
+ mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN"
" to slave: %d, port:%d\n",
__func__, i, port);
- if (i == dev->caps.function)
- continue;
mlx4_slave_event(dev, i, eqe);
+ } else { /* IB port */
+ set_and_calc_slave_port_state(dev, i, port,
+ MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
+ &gen_event);
+ /*we can be in pending state, then do not send port_down event*/
+ if (SLAVE_PORT_GEN_EVENT_DOWN == gen_event) {
+ if (i == mlx4_master_func_num(dev))
+ continue;
+ mlx4_slave_event(dev, i, eqe);
+ }
}
+ }
} else {
- mlx4_dispatch_event(dev,
- MLX4_DEV_EVENT_PORT_UP,
- port);
+ mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
+
mlx4_priv(dev)->sense.do_sense_port[port] = 0;
- if (mlx4_is_master(dev)) {
+ if (!mlx4_is_master(dev))
+ break;
+ if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
for (i = 0; i < dev->num_slaves; i++) {
- if (i == dev->caps.function)
+ if (i == mlx4_master_func_num(dev))
continue;
mlx4_slave_event(dev, i, eqe);
}
- }
+ else /* IB port */
+ /* port-up event will be sent to a slave when the
+ * slave's alias-guid is set. This is done in alias_GUID.c
+ */
+ set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
}
break;
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index cd77a74..91e5ead 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -452,6 +452,7 @@ struct mlx4_slave_state {
/*initialized via the kzalloc*/
u8 is_slave_going_down;
u32 cookie;
+ enum slave_port_state port_state[MLX4_MAX_PORTS + 1];
};
struct slave_list {
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 8e2f03f..8fed1fc 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -192,6 +192,25 @@ enum {
MLX4_FATAL_WARNING_SUBTYPE_WARMING = 0,
};
+enum slave_port_state {
+ SLAVE_PORT_DOWN = 0,
+ SLAVE_PENDING_UP,
+ SLAVE_PORT_UP,
+};
+
+enum slave_port_gen_event {
+ SLAVE_PORT_GEN_EVENT_DOWN = 0,
+ SLAVE_PORT_GEN_EVENT_UP,
+ SLAVE_PORT_GEN_EVENT_NONE,
+};
+
+enum slave_port_state_event {
+ MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
+ MLX4_PORT_STATE_DEV_EVENT_PORT_UP,
+ MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID,
+ MLX4_PORT_STATE_IB_EVENT_GID_INVALID,
+};
+
enum {
MLX4_PERM_LOCAL_READ = 1 << 10,
MLX4_PERM_LOCAL_WRITE = 1 << 11,
@@ -932,4 +951,13 @@ void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port,
int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey);
+int mlx4_is_slave_active(struct mlx4_dev *dev, int slave);
+int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port);
+int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port);
+int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr);
+int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port, u8 port_subtype_change);
+enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port);
+int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, int event, enum slave_port_gen_event *gen_event);
+
+
#endif /* MLX4_DEVICE_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 13/22] {NET,IB}/mlx4: Add alias_guid mechanism
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (11 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 12/22] net/mlx4_core: Add IB port-state machine, and port mgmt event propagation infrastructure Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 14/22] IB/mlx4: Propagate pkey and guid change port management events to slaves Jack Morgenstein
` (10 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
For IB ports, we paravirtualize the GUID at index 0 on slaves.
The GUID at index 0 seen by a slave is the actual GUID occupying
the GUID table at the slave-id index.
The driver, by default, request at startup time that subnet manager
populate its entire guid table with GUIDs. These guids are then
mapped (paravirtualized) to the slaves, and appear for each slave
as its GUID at index 0.
Until each slave has such a guid, its port status is DOWN.
The guid table is cached to support SQP paravirtualization, and event
propagation to slaves on guid change (we test to see if the guid really
changed before propagating an event to the slave).
To support this caching, added capability to __mlx4_ib_query_gid()
to obtain the network view (i.e., physical view) gid at index X,
not just the host (paravirtualized) view.
Based on a patch from Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 688 ++++++++++++++++++++++++++++++
drivers/infiniband/hw/mlx4/mad.c | 19 +-
drivers/infiniband/hw/mlx4/main.c | 37 ++-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 74 ++++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 6 +-
6 files changed, 816 insertions(+), 10 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c
diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile
index bf0aa90..31d4c8a 100644
--- a/drivers/infiniband/hw/mlx4/Makefile
+++ b/drivers/infiniband/hw/mlx4/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o
-mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o
+mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o
diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c
new file mode 100644
index 0000000..9db1581
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/alias_GUID.c
@@ -0,0 +1,688 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+ /***********************************************************/
+/*This file support the handling of the Alias GUID feature. */
+/***********************************************************/
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_cache.h>
+#include <rdma/ib_sa.h>
+#include <rdma/ib_pack.h>
+#include <linux/mlx4/cmd.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <rdma/ib_user_verbs.h>
+#include <linux/delay.h>
+#include "mlx4_ib.h"
+
+/*
+The driver keeps the current state of all guids, as they are in the HW.
+Whenever we receive an smp mad GUIDInfo record, the data will be cached.
+*/
+
+struct mlx4_alias_guid_work_context {
+ u8 port;
+ struct mlx4_ib_dev *dev ;
+ struct ib_sa_query *sa_query;
+ struct completion done;
+ int query_id;
+ struct list_head list;
+ int block_num;
+};
+
+struct mlx4_next_alias_guid_work {
+ u8 port;
+ u8 block_num;
+ struct mlx4_sriov_alias_guid_info_rec_det rec_det;
+};
+
+
+void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num,
+ u8 port_num, u8 *p_data)
+{
+ int i;
+ u64 guid_indexes;
+ int slave_id;
+ int port_index = port_num - 1;
+
+ if (!mlx4_is_master(dev->dev))
+ return;
+
+ guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+ ports_guid[port_num - 1].
+ all_rec_per_port[block_num].guid_indexes);
+ pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+
+ for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+ /* The location of the specific index starts from bit number 4
+ * until bit num 11 */
+ if (test_bit(i + 4, (unsigned long *)&guid_indexes)) {
+ slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+ if (slave_id >= dev->dev->num_slaves) {
+ pr_debug("The last slave: %d\n", slave_id);
+ return;
+ }
+
+ /* cache the guid: */
+ memcpy(&dev->sriov.demux[port_index].guid_cache[slave_id],
+ &p_data[i * GUID_REC_SIZE],
+ GUID_REC_SIZE);
+ } else
+ pr_debug("Guid number: %d in block: %d"
+ " was not updated\n", i, block_num);
+ }
+}
+
+static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index)
+{
+ if (index >= NUM_ALIAS_GUID_PER_PORT) {
+ pr_err("%s: ERROR: asked for index:%d\n", __func__, index);
+ return (__force __be64) ((u64) 0xFFFFFFFFFFFFFFFFUL);
+ }
+ return *(__be64 *)&dev->sriov.demux[port - 1].guid_cache[index];
+}
+
+
+static ib_sa_comp_mask get_aguid_comp_mask_from_ix(int index)
+{
+ return IB_SA_COMP_MASK(4 + index);
+}
+
+/*
+ * Whenever new GUID is set/unset (guid table change) create event and
+ * notify the relevant slave (master also should be notified).
+ * If the GUID value is not as we have in the cache the slave will not be
+ * updated; in this case it waits for the smp_snoop or the port management
+ * event to call the function and to update the slave.
+ * block_number - the index of the block (16 blocks available)
+ * port_number - 1 or 2
+ */
+void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
+ int block_num, u8 port_num,
+ u8 *p_data)
+{
+ int i;
+ u64 guid_indexes;
+ int slave_id;
+ enum slave_port_state new_state;
+ enum slave_port_state prev_state;
+ __be64 tmp_cur_ag, form_cache_ag;
+ enum slave_port_gen_event gen_event;
+
+ if (!mlx4_is_master(dev->dev))
+ return;
+
+ guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+ ports_guid[port_num - 1].
+ all_rec_per_port[block_num].guid_indexes);
+ pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+
+ /*calculate the slaves and notify them*/
+ for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+ /* the location of the specific index runs from bits 4..11 */
+ if (!(test_bit(i + 4, (unsigned long *)&guid_indexes)))
+ continue;
+
+ slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+ if (slave_id >= dev->dev->num_slaves)
+ return;
+ tmp_cur_ag = *(__be64 *)&p_data[i * GUID_REC_SIZE];
+ form_cache_ag = get_cached_alias_guid(dev, port_num,
+ (NUM_ALIAS_GUID_IN_REC * block_num) + i);
+ /*
+ * Check if guid is not the same as in the cache,
+ * If it is different, wait for the snoop_smp or the port mgmt
+ * change event to update the slave on its port state change
+ */
+ if (tmp_cur_ag != form_cache_ag)
+ continue;
+ mlx4_gen_guid_change_eqe(dev->dev, slave_id, port_num);
+
+ /*2 cases: Valid GUID, and Invalid Guid*/
+
+ if (tmp_cur_ag != MLX4_NOT_SET_GUID) { /*valid GUID*/
+ prev_state = mlx4_get_slave_port_state(dev->dev, slave_id, port_num);
+ new_state = set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+ MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID,
+ &gen_event);
+ pr_debug("slave: %d, port: %d prev_port_state: %d,"
+ " new_port_state: %d, gen_event: %d\n",
+ slave_id, port_num, prev_state, new_state, gen_event);
+ if (gen_event == SLAVE_PORT_GEN_EVENT_UP) {
+ pr_debug("sending PORT_UP event to slave: %d, port: %d\n",
+ slave_id, port_num);
+ mlx4_gen_port_state_change_eqe(dev->dev, slave_id,
+ port_num, MLX4_PORT_CHANGE_SUBTYPE_ACTIVE);
+ }
+ } else { /* request to invalidate GUID */
+ set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+ MLX4_PORT_STATE_IB_EVENT_GID_INVALID,
+ &gen_event);
+ pr_debug("sending PORT DOWN event to slave: %d, port: %d\n",
+ slave_id, port_num);
+ mlx4_gen_port_state_change_eqe(dev->dev, slave_id, port_num,
+ MLX4_PORT_CHANGE_SUBTYPE_DOWN);
+ }
+ }
+}
+
+static void aliasguid_query_handler(int status,
+ struct ib_sa_guidinfo_rec *guid_rec,
+ void *context)
+{
+ struct mlx4_ib_dev *dev;
+ struct mlx4_alias_guid_work_context *cb_ctx = context;
+ u8 port_index ;
+ int i;
+ struct mlx4_sriov_alias_guid_info_rec_det *rec;
+ unsigned long flags, flags1;
+
+ if (!context)
+ return;
+
+ dev = cb_ctx->dev;
+ port_index = cb_ctx->port - 1;
+ rec = &dev->sriov.alias_guid.ports_guid[port_index].
+ all_rec_per_port[cb_ctx->block_num];
+
+ if (status) {
+ rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+ pr_debug("(port: %d) failed: status = %d\n",
+ cb_ctx->port, status);
+ goto out;
+ }
+
+ if (guid_rec->block_num != cb_ctx->block_num) {
+ pr_err("block num mismatch: %d != %d\n",
+ cb_ctx->block_num, guid_rec->block_num);
+ goto out;
+ }
+
+ pr_debug("lid/port: %d/%d, block_num: %d\n",
+ be16_to_cpu(guid_rec->lid), cb_ctx->port,
+ guid_rec->block_num);
+
+ rec = &dev->sriov.alias_guid.ports_guid[port_index].
+ all_rec_per_port[guid_rec->block_num];
+
+ rec->status = MLX4_GUID_INFO_STATUS_SET;
+ rec->method = MLX4_GUID_INFO_RECORD_SET;
+
+ for (i = 0 ; i < NUM_ALIAS_GUID_IN_REC; i++) {
+ __be64 tmp_cur_ag;
+ tmp_cur_ag = *(__be64 *)&guid_rec->guid_info_list[i * GUID_REC_SIZE];
+ /* check if the SM didn't assign one of the records.
+ * if it didn't, if it was not sysadmin request:
+ * ask the SM to give a new GUID, (instead of the driver request).
+ */
+ if (tmp_cur_ag == MLX4_NOT_SET_GUID) {
+ mlx4_ib_warn(&dev->ib_dev, "%s:Record num %d in "
+ "block_num: %d was declined by SM, "
+ "ownership by %d (0 = driver, 1=sysAdmin,"
+ " 2=None)\n", __func__, i,
+ guid_rec->block_num, rec->ownership);
+ if (rec->ownership == MLX4_GUID_DRIVER_ASSIGN) {
+ /* if it is driver assign, asks for new GUID from SM*/
+ *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE] =
+ MLX4_NOT_SET_GUID;
+
+ /* Mark the record as not assigned, and let it
+ * be sent again in the next work sched.*/
+ rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+ rec->guid_indexes |= get_aguid_comp_mask_from_ix(i);
+ }
+ } else {
+ /* properly assigned record. */
+ /* We save the GUID we just got from the SM in the
+ * admin_guid in order to be persistent, and in the
+ * request from the sm the process will ask for the same GUID */
+ if (rec->ownership == MLX4_GUID_SYSADMIN_ASSIGN &&
+ tmp_cur_ag != *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE]) {
+ /* the sysadmin assignment failed.*/
+ mlx4_ib_warn(&dev->ib_dev, "%s: Failed to set"
+ " admin guid after SysAdmin "
+ "configuration. "
+ "Record num %d in block_num:%d "
+ "was declined by SM, "
+ "new val(0x%llx) was kept\n",
+ __func__, i,
+ guid_rec->block_num,
+ be64_to_cpu(*(__be64 *) &
+ rec->all_recs[i * GUID_REC_SIZE]));
+ } else {
+ memcpy(&rec->all_recs[i * GUID_REC_SIZE],
+ &guid_rec->guid_info_list[i * GUID_REC_SIZE],
+ GUID_REC_SIZE);
+ }
+ }
+ }
+ /*
+ The func is call here to close the cases when the
+ sm doesn't send smp, so in the sa response the driver
+ notifies the slave.
+ */
+ mlx4_ib_notify_slaves_on_guid_change(dev, guid_rec->block_num,
+ cb_ctx->port,
+ guid_rec->guid_info_list);
+out:
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ if (!dev->sriov.is_going_down)
+ queue_delayed_work(dev->sriov.alias_guid.ports_guid[port_index].wq,
+ &dev->sriov.alias_guid.ports_guid[port_index].
+ alias_guid_work, 0);
+ if (cb_ctx->sa_query) {
+ list_del(&cb_ctx->list);
+ kfree(cb_ctx);
+ } else
+ complete(&cb_ctx->done);
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
+{
+ int i;
+ u64 cur_admin_val;
+ ib_sa_comp_mask comp_mask = 0;
+
+ dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].status
+ = MLX4_GUID_INFO_STATUS_IDLE;
+ dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].method
+ = MLX4_GUID_INFO_RECORD_SET;
+
+ /* calculate the comp_mask for that record.*/
+ for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+ cur_admin_val =
+ *(u64 *)&dev->sriov.alias_guid.ports_guid[port - 1].
+ all_rec_per_port[index].all_recs[GUID_REC_SIZE * i];
+ /*
+ check the admin value: if it's for delete (~00LL) or
+ it is the first guid of the first record (hw guid) or
+ the records is not in ownership of the sysadmin and the sm doesn't
+ need to assign GUIDs, then don't put it up for assignment.
+ */
+ if (MLX4_GUID_FOR_DELETE_VAL == cur_admin_val ||
+ (!index && !i) ||
+ MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid.
+ ports_guid[port - 1].all_rec_per_port[index].ownership)
+ continue;
+ comp_mask |= get_aguid_comp_mask_from_ix(i);
+ }
+ dev->sriov.alias_guid.ports_guid[port - 1].
+ all_rec_per_port[index].guid_indexes = comp_mask;
+}
+
+static int set_guid_rec(struct ib_device *ibdev,
+ u8 port, int index,
+ struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+{
+ int err;
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ struct ib_sa_guidinfo_rec guid_info_rec;
+ ib_sa_comp_mask comp_mask;
+ struct ib_port_attr attr;
+ struct mlx4_alias_guid_work_context *callback_context;
+ unsigned long resched_delay, flags, flags1;
+ struct list_head *head =
+ &dev->sriov.alias_guid.ports_guid[port - 1].cb_list;
+
+ err = __mlx4_ib_query_port(ibdev, port, &attr, 1);
+ if (err) {
+ pr_debug("mlx4_ib_query_port failed (err: %d), port: %d\n",
+ err, port);
+ return err;
+ }
+ /*check the port was configured by the sm, otherwise no need to send */
+ if (attr.state != IB_PORT_ACTIVE) {
+ pr_debug("port %d not active...rescheduling\n", port);
+ resched_delay = 5 * HZ;
+ err = -EAGAIN;
+ goto new_schedule;
+ }
+
+ callback_context = kmalloc(sizeof *callback_context, GFP_KERNEL);
+ if (!callback_context) {
+ err = -ENOMEM;
+ resched_delay = HZ * 5;
+ goto new_schedule;
+ }
+ callback_context->port = port;
+ callback_context->dev = dev;
+ callback_context->block_num = index;
+
+ memset(&guid_info_rec, 0, sizeof(struct ib_sa_guidinfo_rec));
+
+ guid_info_rec.lid = cpu_to_be16(attr.lid);
+ guid_info_rec.block_num = index;
+
+ memcpy(guid_info_rec.guid_info_list, rec_det->all_recs,
+ GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC);
+ comp_mask = IB_SA_GUIDINFO_REC_LID | IB_SA_GUIDINFO_REC_BLOCK_NUM |
+ rec_det->guid_indexes;
+
+ init_completion(&callback_context->done);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ list_add_tail(&callback_context->list, head);
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+
+ callback_context->query_id =
+ ib_sa_guid_info_rec_query(dev->sriov.alias_guid.sa_client,
+ ibdev, port, &guid_info_rec,
+ comp_mask, rec_det->method, 1000,
+ GFP_KERNEL, aliasguid_query_handler,
+ callback_context,
+ &callback_context->sa_query);
+ if (callback_context->query_id < 0) {
+ pr_debug("ib_sa_guid_info_rec_query failed, query_id: "
+ "%d. will reschedule to the next 1 sec.\n",
+ callback_context->query_id);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ list_del(&callback_context->list);
+ kfree(callback_context);
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ resched_delay = 1 * HZ;
+ err = -EAGAIN;
+ goto new_schedule;
+ }
+ err = 0;
+ goto out;
+
+new_schedule:
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ invalidate_guid_record(dev, port, index);
+ if (!dev->sriov.is_going_down) {
+ queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+ &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+ resched_delay);
+ }
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+
+out:
+ return err;
+}
+
+void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port)
+{
+ int i;
+ unsigned long flags, flags1;
+
+ pr_debug("port %d\n", port);
+
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ for (i = 0; i < NUM_ALIAS_GUID_REC_IN_PORT; i++)
+ invalidate_guid_record(dev, port, i);
+
+ if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) {
+ /*
+ make sure no work waits in the queue, if the work is already
+ queued(not on the timer) the cancel will fail. That is not a problem
+ because we just want the work started.
+ */
+ __cancel_delayed_work(&dev->sriov.alias_guid.
+ ports_guid[port - 1].alias_guid_work);
+ queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+ &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+ 0);
+ }
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+/* The function returns the next record that was
+ * not configured (or failed to be configured) */
+static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port,
+ struct mlx4_next_alias_guid_work *rec)
+{
+ int j;
+ unsigned long flags;
+
+ for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+ if (dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status ==
+ MLX4_GUID_INFO_STATUS_IDLE) {
+ memcpy(&rec->rec_det,
+ &dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j],
+ sizeof(struct mlx4_sriov_alias_guid_info_rec_det));
+ rec->port = port;
+ rec->block_num = j;
+ dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status =
+ MLX4_GUID_INFO_STATUS_PENDING;
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+ return 0;
+ }
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+ }
+ return -ENOENT;
+}
+
+static void set_administratively_guid_record(struct mlx4_ib_dev *dev, int port,
+ int rec_index,
+ struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+{
+ dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].guid_indexes =
+ rec_det->guid_indexes;
+ memcpy(dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].all_recs,
+ rec_det->all_recs, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+ dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].status =
+ rec_det->status;
+}
+
+static void set_all_slaves_guids(struct mlx4_ib_dev *dev, int port)
+{
+ int j;
+ struct mlx4_sriov_alias_guid_info_rec_det rec_det ;
+
+ for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT ; j++) {
+ memset(rec_det.all_recs, 0, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+ rec_det.guid_indexes = (!j ? 0 : IB_SA_GUIDINFO_REC_GID0) |
+ IB_SA_GUIDINFO_REC_GID1 | IB_SA_GUIDINFO_REC_GID2 |
+ IB_SA_GUIDINFO_REC_GID3 | IB_SA_GUIDINFO_REC_GID4 |
+ IB_SA_GUIDINFO_REC_GID5 | IB_SA_GUIDINFO_REC_GID6 |
+ IB_SA_GUIDINFO_REC_GID7;
+ rec_det.status = MLX4_GUID_INFO_STATUS_IDLE;
+ set_administratively_guid_record(dev, port, j, &rec_det);
+ }
+}
+
+static void alias_guid_work(struct work_struct *work)
+{
+ struct delayed_work *delay = to_delayed_work(work);
+ int ret = 0;
+ struct mlx4_next_alias_guid_work *rec;
+ struct mlx4_sriov_alias_guid_port_rec_det *sriov_alias_port =
+ container_of(delay, struct mlx4_sriov_alias_guid_port_rec_det,
+ alias_guid_work);
+ struct mlx4_sriov_alias_guid *sriov_alias_guid = sriov_alias_port->parent;
+ struct mlx4_ib_sriov *ib_sriov = container_of(sriov_alias_guid,
+ struct mlx4_ib_sriov,
+ alias_guid);
+ struct mlx4_ib_dev *dev = container_of(ib_sriov, struct mlx4_ib_dev, sriov);
+
+ rec = kzalloc(sizeof *rec, GFP_KERNEL);
+ if (!rec) {
+ pr_err("alias_guid_work: No Memory\n");
+ return;
+ }
+
+ pr_debug("starting [port: %d]...\n", sriov_alias_port->port + 1);
+ ret = get_next_record_to_update(dev, sriov_alias_port->port, rec);
+ if (ret) {
+ pr_debug("No more records to update.\n");
+ goto out;
+ }
+
+ set_guid_rec(&dev->ib_dev, rec->port + 1, rec->block_num,
+ &rec->rec_det);
+
+out:
+ kfree(rec);
+}
+
+
+void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port)
+{
+ unsigned long flags, flags1;
+
+ if (!mlx4_is_master(dev->dev))
+ return;
+ spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ if (!dev->sriov.is_going_down) {
+ queue_delayed_work(dev->sriov.alias_guid.ports_guid[port].wq,
+ &dev->sriov.alias_guid.ports_guid[port].alias_guid_work, 0);
+ }
+ spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev)
+{
+ int i;
+ struct mlx4_ib_sriov *sriov = &dev->sriov;
+ struct mlx4_alias_guid_work_context *cb_ctx;
+ struct mlx4_sriov_alias_guid_port_rec_det *det;
+ struct ib_sa_query *sa_query;
+ unsigned long flags;
+
+ for (i = 0 ; i < dev->num_ports; i++) {
+ cancel_delayed_work(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work);
+ det = &sriov->alias_guid.ports_guid[i];
+ spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+ while (!list_empty(&det->cb_list)) {
+ cb_ctx = list_entry(det->cb_list.next,
+ struct mlx4_alias_guid_work_context,
+ list);
+ sa_query = cb_ctx->sa_query;
+ cb_ctx->sa_query = NULL;
+ list_del(&cb_ctx->list);
+ spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+ ib_sa_cancel_query(cb_ctx->query_id, sa_query);
+ wait_for_completion(&cb_ctx->done);
+ kfree(cb_ctx);
+ spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+ }
+ spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+ }
+ for (i = 0 ; i < dev->num_ports; i++) {
+ flush_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+ destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+ }
+ ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+ kfree(dev->sriov.alias_guid.sa_client);
+}
+
+int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
+{
+ char alias_wq_name[15];
+ int ret = 0;
+ int i, j, k;
+ union ib_gid gid;
+
+ if (!mlx4_is_master(dev->dev))
+ return 0;
+ dev->sriov.alias_guid.sa_client =
+ kzalloc(sizeof *dev->sriov.alias_guid.sa_client, GFP_KERNEL);
+ if (!dev->sriov.alias_guid.sa_client)
+ return -ENOMEM;
+
+ ib_sa_register_client(dev->sriov.alias_guid.sa_client);
+
+ spin_lock_init(&dev->sriov.alias_guid.ag_work_lock);
+
+ for (i = 1; i <= dev->num_ports; ++i) {
+ if (dev->ib_dev.query_gid(&dev->ib_dev , i, 0, &gid)) {
+ ret = -EFAULT;
+ goto err_unregister;
+ }
+ }
+
+ for (i = 0 ; i < dev->num_ports; i++) {
+ memset(&dev->sriov.alias_guid.ports_guid[i], 0,
+ sizeof(struct mlx4_sriov_alias_guid_port_rec_det));
+ /*Check if the SM doesn't need to assign the GUIDs*/
+ for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+ if (mlx4_ib_sm_guid_assign) {
+ dev->sriov.alias_guid.ports_guid[i].
+ all_rec_per_port[j].
+ ownership = MLX4_GUID_DRIVER_ASSIGN;
+ continue;
+ }
+ dev->sriov.alias_guid.ports_guid[i].all_rec_per_port[j].
+ ownership = MLX4_GUID_NONE_ASSIGN;
+ /*mark each val as it was deleted,
+ till the sysAdmin will give it valid val*/
+ for (k = 0; k < NUM_ALIAS_GUID_IN_REC; k++) {
+ *(__be64 *)&dev->sriov.alias_guid.ports_guid[i].
+ all_rec_per_port[j].all_recs[GUID_REC_SIZE * k] =
+ cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL);
+ }
+ }
+ INIT_LIST_HEAD(&dev->sriov.alias_guid.ports_guid[i].cb_list);
+ /*prepare the records, set them to be allocated by sm*/
+ for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT; j++)
+ invalidate_guid_record(dev, i + 1, j);
+
+ dev->sriov.alias_guid.ports_guid[i].parent = &dev->sriov.alias_guid;
+ dev->sriov.alias_guid.ports_guid[i].port = i;
+ if (mlx4_ib_sm_guid_assign)
+ set_all_slaves_guids(dev, i);
+
+ snprintf(alias_wq_name, sizeof alias_wq_name, "alias_guid%d", i);
+ dev->sriov.alias_guid.ports_guid[i].wq =
+ create_singlethread_workqueue(alias_wq_name);
+ if (!dev->sriov.alias_guid.ports_guid[i].wq) {
+ ret = -ENOMEM;
+ goto err_thread;
+ }
+ INIT_DELAYED_WORK(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work,
+ alias_guid_work);
+ }
+ return 0;
+
+err_thread:
+ for (--i; i >= 0; i--) {
+ destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+ dev->sriov.alias_guid.ports_guid[i].wq = NULL;
+ }
+
+err_unregister:
+ ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+ kfree(dev->sriov.alias_guid.sa_client);
+ dev->sriov.alias_guid.sa_client = NULL;
+ pr_err("init_alias_guid_service: Failed. (ret:%d)\n", ret);
+ return ret;
+}
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index ef40254..0db3e9c 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -787,8 +787,10 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num)
{
- /* re-configure the mcg's */
+ /* re-configure the alias-guid and mcg's */
if (mlx4_is_master(dev->dev)) {
+ mlx4_ib_invalidate_all_guid_record(dev, port_num);
+
if (!dev->sriov.is_going_down)
mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0);
}
@@ -1804,9 +1806,20 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
return 0;
}
+ err = mlx4_ib_init_alias_guid_service(dev);
+ if (err) {
+ mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n");
+ goto paravirt_err;
+ }
+
mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n",
dev->dev->caps.sqp_demux);
for (i = 0; i < dev->num_ports; i++) {
+ union ib_gid gid;
+ err = __mlx4_ib_query_gid(&dev->ib_dev, i + 1, 0, &gid, 1);
+ if (err)
+ goto demux_err;
+ dev->sriov.demux[i].guid_cache[0] = gid.global.interface_id;
err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1,
&dev->sriov.sqps[i]);
if (err)
@@ -1824,6 +1837,9 @@ demux_err:
mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
--i;
}
+ mlx4_ib_destroy_alias_guid_service(dev);
+
+paravirt_err:
mlx4_ib_cm_paravirt_clean(dev, -1);
return err;
@@ -1850,5 +1866,6 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
}
mlx4_ib_cm_paravirt_clean(dev, -1);
+ mlx4_ib_destroy_alias_guid_service(dev);
}
}
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fce64ec..778fdb8 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -59,6 +59,10 @@ MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
MODULE_LICENSE("Dual BSD/GPL");
MODULE_VERSION(DRV_VERSION);
+int mlx4_ib_sm_guid_assign = 1;
+module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
+MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
+
static const char mlx4_ib_version[] =
DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
DRV_VERSION " (" DRV_RELDATE ")\n";
@@ -349,12 +353,15 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
return __mlx4_ib_query_port(ibdev, port, props, 0);
}
-static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
- union ib_gid *gid)
+int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+ union ib_gid *gid, int netw_view)
{
struct ib_smp *in_mad = NULL;
struct ib_smp *out_mad = NULL;
int err = -ENOMEM;
+ struct mlx4_ib_dev *dev = to_mdev(ibdev);
+ int clear = 0;
+ int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -365,18 +372,29 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
in_mad->attr_mod = cpu_to_be32(port);
- err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port,
- NULL, NULL, in_mad, out_mad);
+ if (mlx4_is_mfunc(dev->dev) && netw_view)
+ mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+ err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
if (err)
goto out;
memcpy(gid->raw, out_mad->data + 8, 8);
+ if (mlx4_is_mfunc(dev->dev) && !netw_view) {
+ if (index) {
+ /* For any index > 0, return the null guid */
+ err = 0;
+ clear = 1;
+ goto out;
+ }
+ }
+
init_query_mad(in_mad);
in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
in_mad->attr_mod = cpu_to_be32(index / 8);
- err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port,
+ err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
NULL, NULL, in_mad, out_mad);
if (err)
goto out;
@@ -384,6 +402,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
out:
+ if (clear)
+ memset(gid->raw + 8, 0, 8);
kfree(in_mad);
kfree(out_mad);
return err;
@@ -403,7 +423,7 @@ static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
union ib_gid *gid)
{
if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
- return __mlx4_ib_query_gid(ibdev, port, index, gid);
+ return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
else
return iboe_query_gid(ibdev, port, index, gid);
}
@@ -1565,6 +1585,11 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
case MLX4_DEV_EVENT_PORT_UP:
if (p > ibdev->num_ports)
return;
+ if (mlx4_is_master(dev) &&
+ rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
+ IB_LINK_LAYER_INFINIBAND) {
+ mlx4_ib_invalidate_all_guid_record(ibdev, p);
+ }
ibev.event = IB_EVENT_PORT_ACTIVE;
break;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 7476e24..f3f75f8 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -42,6 +42,7 @@
#include <rdma/ib_verbs.h>
#include <rdma/ib_umem.h>
#include <rdma/ib_mad.h>
+#include <rdma/ib_sa.h>
#include <linux/mlx4/device.h>
#include <linux/mlx4/doorbell.h>
@@ -64,6 +65,9 @@ enum {
#define MLX4_IB_SQ_HEADROOM(shift) ((MLX4_IB_MAX_HEADROOM >> (shift)) + 1)
#define MLX4_IB_SQ_MAX_SPARE (MLX4_IB_SQ_HEADROOM(MLX4_IB_SQ_MIN_WQE_SHIFT))
+/*module param to indicate if SM assigns the alias_GUID*/
+extern int mlx4_ib_sm_guid_assign;
+
struct mlx4_ib_ucontext {
struct ib_ucontext ibucontext;
struct mlx4_uar uar;
@@ -277,6 +281,57 @@ struct mlx4_ib_ah {
union mlx4_ext_av av;
};
+/****************************************/
+/* alias guid support */
+/****************************************/
+#define NUM_PORT_ALIAS_GUID 2
+#define NUM_ALIAS_GUID_IN_REC 8
+#define NUM_ALIAS_GUID_REC_IN_PORT 16
+#define GUID_REC_SIZE 8
+#define NUM_ALIAS_GUID_PER_PORT 128
+#define MLX4_NOT_SET_GUID (0x00LL)
+#define MLX4_GUID_FOR_DELETE_VAL (~(0x00LL))
+
+enum mlx4_guid_alias_rec_status {
+ MLX4_GUID_INFO_STATUS_IDLE,
+ MLX4_GUID_INFO_STATUS_SET,
+ MLX4_GUID_INFO_STATUS_PENDING,
+};
+
+enum mlx4_guid_alias_rec_ownership {
+ MLX4_GUID_DRIVER_ASSIGN,
+ MLX4_GUID_SYSADMIN_ASSIGN,
+ MLX4_GUID_NONE_ASSIGN, /*init state of each record*/
+};
+
+enum mlx4_guid_alias_rec_method {
+ MLX4_GUID_INFO_RECORD_SET = IB_MGMT_METHOD_SET,
+ MLX4_GUID_INFO_RECORD_DELETE = IB_SA_METHOD_DELETE,
+};
+
+struct mlx4_sriov_alias_guid_info_rec_det {
+ u8 all_recs[GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC];
+ ib_sa_comp_mask guid_indexes; /*indicates what from the 8 records are valid*/
+ enum mlx4_guid_alias_rec_status status; /*indicates the administraively status of the record.*/
+ u8 method; /*set or delete*/
+ enum mlx4_guid_alias_rec_ownership ownership; /*indicates who assign that alias_guid record*/
+};
+
+struct mlx4_sriov_alias_guid_port_rec_det {
+ struct mlx4_sriov_alias_guid_info_rec_det all_rec_per_port[NUM_ALIAS_GUID_REC_IN_PORT];
+ struct workqueue_struct *wq;
+ struct delayed_work alias_guid_work;
+ u8 port;
+ struct mlx4_sriov_alias_guid *parent;
+ struct list_head cb_list;
+};
+
+struct mlx4_sriov_alias_guid {
+ struct mlx4_sriov_alias_guid_port_rec_det ports_guid[MLX4_MAX_PORTS];
+ spinlock_t ag_work_lock;
+ struct ib_sa_client *sa_client;
+};
+
struct mlx4_ib_demux_work {
struct work_struct work;
struct mlx4_ib_dev *dev;
@@ -349,6 +404,8 @@ struct mlx4_ib_sriov {
spinlock_t going_down_lock;
int is_going_down;
+ struct mlx4_sriov_alias_guid alias_guid;
+
/* CM paravirtualization fields */
struct list_head cm_list;
spinlock_t id_map_lock;
@@ -555,6 +612,9 @@ int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
u16 *pkey, int netw_view);
+int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+ union ib_gid *gid, int netw_view);
+
int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr,
u8 *mac, int *is_mcast, u8 port);
@@ -606,4 +666,18 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id
void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev);
void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave_id);
+/* alias guid support */
+void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port);
+int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev);
+void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev);
+void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port);
+
+void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
+ int block_num,
+ u8 port_num, u8 *p_data);
+
+void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev,
+ int block_num, u8 port_num,
+ u8 *p_data);
+
#endif /* MLX4_IB_H */
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 28fca65..c13e988 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -629,8 +629,10 @@ static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table,
static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf)
{
- /* will be modified when add alias_guid feature */
- return IB_PORT_DOWN;
+ if (mlx4_get_slave_port_state(dev, vf, port) == SLAVE_PORT_UP)
+ return IB_PORT_ACTIVE;
+ else
+ return IB_PORT_DOWN;
}
static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 14/22] IB/mlx4: Propagate pkey and guid change port management events to slaves
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (12 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 13/22] {NET,IB}/mlx4: Add alias_guid mechanism Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 15/22] IB/mlx4: Add iov directory in sysfs under the ib device Jack Morgenstein
` (9 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
pkey change and guid change events are not of interest to all slaves,
but only to those slaves which "see" the table slots whose contents
have change.
For example, if the guid at port 1, index 5 has changed in the
PPF, we wish to propagate the gid-change event only to the function
which has that guid index mapped to its port/guid table (in this case
it is slave #5). Other functions should not get the event,
since the event does not affect them.
Similarly with pkeys -- pkey change events are forwarded
only to slaves which have that pkey index mapped to their
virtual pkey table.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 162 +++++++++++++++++++++++++++++--
drivers/net/ethernet/mellanox/mlx4/fw.c | 6 +
2 files changed, 161 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 0db3e9c..ef9842d 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -54,6 +54,15 @@ enum {
#define MLX4_TUN_IS_RECV(a) (((a) >> MLX4_TUN_SEND_WRID_SHIFT) & 0x1)
#define MLX4_TUN_WRID_QPN(a) (((a) >> MLX4_TUN_QPN_SHIFT) & 0x3)
+ /* Port mgmt change event handling */
+
+#define GET_BLK_PTR_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.block_ptr)
+#define GET_MASK_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.tbl_entries_mask)
+#define NUM_IDX_IN_PKEY_TBL_BLK 32
+#define GUID_TBL_ENTRY_SIZE 8 /* size in bytes */
+#define GUID_TBL_BLK_NUM_ENTRIES 8
+#define GUID_TBL_BLK_SIZE (GUID_TBL_ENTRY_SIZE * GUID_TBL_BLK_NUM_ENTRIES)
+
struct mlx4_mad_rcv_buf {
struct ib_grh grh;
u8 payload[256];
@@ -76,6 +85,9 @@ struct mlx4_rcv_tunnel_mad {
} __packed;
static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num);
+static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num);
+static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+ int block, u32 change_bitmap);
__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx)
{
@@ -219,8 +231,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
handle_client_rereg_event(dev, port_num);
if (prev_lid != lid)
- mlx4_ib_dispatch_event(dev, port_num,
- IB_EVENT_LID_CHANGE);
+ handle_lid_change_event(dev, port_num);
break;
case IB_SMP_ATTR_PKEY_TABLE:
@@ -230,6 +241,9 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
break;
}
+ /* at this point, we are running in the master.
+ * Slaves do not receive SMPs.
+ */
bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod) & 0xFFFF;
base = (__be16 *) &(((struct ib_smp *)mad)->data[0]);
pkey_change_bitmap = 0;
@@ -247,10 +261,13 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
"block=0x%x, change_bitmap=0x%x\n",
port_num, bn, pkey_change_bitmap);
- if (pkey_change_bitmap)
+ if (pkey_change_bitmap) {
mlx4_ib_dispatch_event(dev, port_num,
IB_EVENT_PKEY_CHANGE);
-
+ if (!dev->sriov.is_going_down)
+ __propagate_pkey_ev(dev, port_num, bn,
+ pkey_change_bitmap);
+ }
break;
case IB_SMP_ATTR_GUID_INFO:
@@ -258,12 +275,56 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
if (!mlx4_is_master(dev->dev))
mlx4_ib_dispatch_event(dev, port_num,
IB_EVENT_GID_CHANGE);
+ /*if master, notify relevant slaves*/
+ if (mlx4_is_master(dev->dev) &&
+ !dev->sriov.is_going_down) {
+ bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod);
+ mlx4_ib_update_cache_on_guid_change(dev, bn, port_num,
+ (u8 *)(&((struct ib_smp *)mad)->data));
+ mlx4_ib_notify_slaves_on_guid_change(dev, bn, port_num,
+ (u8 *)(&((struct ib_smp *)mad)->data));
+ }
break;
+
default:
break;
}
}
+static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+ int block, u32 change_bitmap)
+{
+ int i, ix, slave, err;
+ int have_event = 0;
+
+ for (slave = 0; slave < dev->dev->caps.sqp_demux; slave++) {
+ if (slave == mlx4_master_func_num(dev->dev))
+ continue;
+ if (!mlx4_is_slave_active(dev->dev, slave))
+ continue;
+
+ have_event = 0;
+ for (i = 0; i < 32; i++) {
+ if (!(change_bitmap & (1 << i)))
+ continue;
+ for (ix = 0;
+ ix < dev->dev->caps.pkey_table_len[port_num]; ix++) {
+ if (dev->pkeys.virt2phys_pkey[slave][port_num - 1]
+ [ix] == i + 32 * block) {
+ err = mlx4_gen_pkey_eqe(dev->dev, slave, port_num);
+ pr_debug("propagate_pkey_ev: slave %d,"
+ " port %d, ix %d (%d)\n",
+ slave, port_num, ix, err);
+ have_event = 1;
+ break;
+ }
+ }
+ if (have_event)
+ break;
+ }
+ }
+}
+
static void node_desc_override(struct ib_device *dev,
struct ib_mad *mad)
{
@@ -785,18 +846,90 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
}
}
+static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num)
+{
+ mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_LID_CHANGE);
+
+ if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+ mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
+ MLX4_EQ_PORT_INFO_LID_CHANGE_MASK);
+}
+
static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num)
{
/* re-configure the alias-guid and mcg's */
if (mlx4_is_master(dev->dev)) {
mlx4_ib_invalidate_all_guid_record(dev, port_num);
- if (!dev->sriov.is_going_down)
+ if (!dev->sriov.is_going_down) {
mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0);
+ mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
+ MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK);
+ }
}
mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_CLIENT_REREGISTER);
}
+static void propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+ struct mlx4_eqe *eqe)
+{
+ __propagate_pkey_ev(dev, port_num, GET_BLK_PTR_FROM_EQE(eqe),
+ GET_MASK_FROM_EQE(eqe));
+}
+
+static void handle_slaves_guid_change(struct mlx4_ib_dev *dev, u8 port_num,
+ u32 guid_tbl_blk_num, u32 change_bitmap)
+{
+ struct ib_smp *in_mad = NULL;
+ struct ib_smp *out_mad = NULL;
+ u16 i;
+
+ if (!mlx4_is_mfunc(dev->dev) || !mlx4_is_master(dev->dev))
+ return;
+
+ in_mad = kmalloc(sizeof *in_mad, GFP_KERNEL);
+ out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+ if (!in_mad || !out_mad) {
+ mlx4_ib_warn(&dev->ib_dev, "failed to allocate memory for guid info mads\n");
+ goto out;
+ }
+
+ guid_tbl_blk_num *= 4;
+
+ for (i = 0; i < 4; i++) {
+ if (change_bitmap && (!((change_bitmap >> (8 * i)) & 0xff)))
+ continue;
+ memset(in_mad, 0, sizeof *in_mad);
+ memset(out_mad, 0, sizeof *out_mad);
+
+ in_mad->base_version = 1;
+ in_mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+ in_mad->class_version = 1;
+ in_mad->method = IB_MGMT_METHOD_GET;
+ in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
+ in_mad->attr_mod = cpu_to_be32(guid_tbl_blk_num + i);
+
+ if (mlx4_MAD_IFC(dev,
+ MLX4_MAD_IFC_IGNORE_KEYS | MLX4_MAD_IFC_NET_VIEW,
+ port_num, NULL, NULL, in_mad, out_mad)) {
+ mlx4_ib_warn(&dev->ib_dev, "Failed in get GUID INFO MAD_IFC\n");
+ goto out;
+ }
+
+ mlx4_ib_update_cache_on_guid_change(dev, guid_tbl_blk_num + i,
+ port_num,
+ (u8 *)(&((struct ib_smp *)out_mad)->data));
+ mlx4_ib_notify_slaves_on_guid_change(dev, guid_tbl_blk_num + i,
+ port_num,
+ (u8 *)(&((struct ib_smp *)out_mad)->data));
+ }
+
+out:
+ kfree(in_mad);
+ kfree(out_mad);
+ return;
+}
+
void handle_port_mgmt_change_event(struct work_struct *work)
{
struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
@@ -804,6 +937,8 @@ void handle_port_mgmt_change_event(struct work_struct *work)
struct mlx4_eqe *eqe = &(ew->ib_eqe);
u8 port = eqe->event.port_mgmt_change.port;
u32 changed_attr;
+ u32 tbl_block;
+ u32 change_bitmap;
switch (eqe->subtype) {
case MLX4_DEV_PMC_SUBTYPE_PORT_INFO:
@@ -819,11 +954,16 @@ void handle_port_mgmt_change_event(struct work_struct *work)
/* Check if it is a lid change event */
if (changed_attr & MLX4_EQ_PORT_INFO_LID_CHANGE_MASK)
- mlx4_ib_dispatch_event(dev, port, IB_EVENT_LID_CHANGE);
+ handle_lid_change_event(dev, port);
/* Generate GUID changed event */
- if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK)
+ if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK) {
mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
+ /*if master, notify all slaves*/
+ if (mlx4_is_master(dev->dev))
+ mlx4_gen_slaves_port_mgt_ev(dev->dev, port,
+ MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK);
+ }
if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK)
handle_client_rereg_event(dev, port);
@@ -831,11 +971,19 @@ void handle_port_mgmt_change_event(struct work_struct *work)
case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE:
mlx4_ib_dispatch_event(dev, port, IB_EVENT_PKEY_CHANGE);
+ if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+ propagate_pkey_ev(dev, port, eqe);
break;
case MLX4_DEV_PMC_SUBTYPE_GUID_INFO:
/* paravirtualized master's guid is guid 0 -- does not change */
if (!mlx4_is_master(dev->dev))
mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
+ /*if master, notify relevant slaves*/
+ else if (!dev->sriov.is_going_down) {
+ tbl_block = GET_BLK_PTR_FROM_EQE(eqe);
+ change_bitmap = GET_MASK_FROM_EQE(eqe);
+ handle_slaves_guid_change(dev, port, tbl_block, change_bitmap);
+ }
break;
default:
pr_warn("Unsupported subtype 0x%x for "
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 0f557df..88d61ff 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -729,6 +729,7 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd)
{
+ u64 flags;
int err = 0;
u8 field;
@@ -737,6 +738,11 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
if (err)
return err;
+ /* add port mng change event capability unconditionally to slaves */
+ MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
+ flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV;
+ MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
+
/* For guests, report Blueflame disabled */
MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_BF_OFFSET);
field &= 0x7f;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 15/22] IB/mlx4: Add iov directory in sysfs under the ib device
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (13 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 14/22] IB/mlx4: Propagate pkey and guid change port management events to slaves Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 16/22] net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB Jack Morgenstein
` (8 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
This directory is added only for the master -- slaves do
not have it.
The sysfs iov directory is used to manage and examine the port
pkey and guid paravirtualization.
Under iov/ports, the administrator may examine the gid and pkey
tables as they are present in the device (and as are seen in the
"network view" presented to the SM).
Under the iov/<pci slot number> directories, the admin may map
the index numbers in the physical tables (as under iov/ports)
to the paravirtualized index numbers that guests see.
Thus, for example, if the administrator, for port 1 on guest 2, say,
maps physical pkey index 10 to virtual index 1, that guest, whenever
it uses its pkey index 1 will actually be using the real pkey index 10.
Based on patch from Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 6 +-
drivers/infiniband/hw/mlx4/mad.c | 9 +
drivers/infiniband/hw/mlx4/mcg.c | 67 +++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 43 ++
drivers/infiniband/hw/mlx4/sysfs.c | 794 +++++++++++++++++++++++++++++++
6 files changed, 917 insertions(+), 4 deletions(-)
create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c
diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile
index 31d4c8a..f4213b3 100644
--- a/drivers/infiniband/hw/mlx4/Makefile
+++ b/drivers/infiniband/hw/mlx4/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o
-mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o
+mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o sysfs.o
diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c
index 9db1581..c4252e7 100644
--- a/drivers/infiniband/hw/mlx4/alias_GUID.c
+++ b/drivers/infiniband/hw/mlx4/alias_GUID.c
@@ -113,7 +113,7 @@ static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index
}
-static ib_sa_comp_mask get_aguid_comp_mask_from_ix(int index)
+ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index)
{
return IB_SA_COMP_MASK(4 + index);
}
@@ -259,7 +259,7 @@ static void aliasguid_query_handler(int status,
/* Mark the record as not assigned, and let it
* be sent again in the next work sched.*/
rec->status = MLX4_GUID_INFO_STATUS_IDLE;
- rec->guid_indexes |= get_aguid_comp_mask_from_ix(i);
+ rec->guid_indexes |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
}
} else {
/* properly assigned record. */
@@ -337,7 +337,7 @@ static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid.
ports_guid[port - 1].all_rec_per_port[index].ownership)
continue;
- comp_mask |= get_aguid_comp_mask_from_ix(i);
+ comp_mask |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
}
dev->sriov.alias_guid.ports_guid[port - 1].
all_rec_per_port[index].guid_indexes = comp_mask;
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index ef9842d..ec64077 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1959,6 +1959,11 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n");
goto paravirt_err;
}
+ err = mlx4_ib_device_register_sysfs(dev);
+ if (err) {
+ mlx4_ib_warn(&dev->ib_dev, "Failed to register sysfs\n");
+ goto sysfs_err;
+ }
mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n",
dev->dev->caps.sqp_demux);
@@ -1985,6 +1990,9 @@ demux_err:
mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
--i;
}
+ mlx4_ib_device_unregister_sysfs(dev);
+
+sysfs_err:
mlx4_ib_destroy_alias_guid_service(dev);
paravirt_err:
@@ -2015,5 +2023,6 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
mlx4_ib_cm_paravirt_clean(dev, -1);
mlx4_ib_destroy_alias_guid_service(dev);
+ mlx4_ib_device_unregister_sysfs(dev);
}
}
diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c
index 1ee2e3a..3c3b54c 100644
--- a/drivers/infiniband/hw/mlx4/mcg.c
+++ b/drivers/infiniband/hw/mlx4/mcg.c
@@ -110,6 +110,7 @@ struct mcast_group {
__be64 last_req_tid;
char name[33]; /* MGID string */
+ struct device_attribute dentry;
/* refcount is the reference count for the following:
1. Each queued request
@@ -445,6 +446,8 @@ static int release_group(struct mcast_group *group, int from_timeout_handler)
}
nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0);
+ if (nzgroup)
+ del_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
if (!list_empty(&group->pending_list))
mcg_warn_group(group, "releasing a group with non empty pending list\n");
if (nzgroup)
@@ -769,6 +772,7 @@ static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx
}
atomic_inc(&group->refcount);
+ add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
mutex_unlock(&group->lock);
mutex_unlock(&ctx->mcg_table_lock);
return group;
@@ -796,6 +800,9 @@ static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx
return NULL;
}
+static ssize_t sysfs_show_group(struct device *dev,
+ struct device_attribute *attr, char *buf);
+
static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
union ib_gid *mgid, int create,
gfp_t gfp_mask)
@@ -830,6 +837,11 @@ static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
sprintf(group->name, "%016llx%016llx",
be64_to_cpu(group->rec.mgid.global.subnet_prefix),
be64_to_cpu(group->rec.mgid.global.interface_id));
+ sysfs_attr_init(&group->dentry.attr);
+ group->dentry.show = sysfs_show_group;
+ group->dentry.store = NULL;
+ group->dentry.attr.name = group->name;
+ group->dentry.attr.mode = 0400;
group->state = MCAST_IDLE;
if (is_mgid0) {
@@ -844,6 +856,8 @@ static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
return ERR_PTR(-EINVAL);
}
+ add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
+
found:
atomic_inc(&group->refcount);
return group;
@@ -969,6 +983,58 @@ int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port,
}
}
+static ssize_t sysfs_show_group(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct mcast_group *group =
+ container_of(attr, struct mcast_group, dentry);
+ struct mcast_req *req = NULL;
+ char pending_str[40];
+ char state_str[40];
+ ssize_t len = 0;
+ int f;
+
+ if (group->state == MCAST_IDLE)
+ sprintf(state_str, "%s", get_state_string(group->state));
+ else
+ sprintf(state_str, "%s(TID=0x%llx)",
+ get_state_string(group->state),
+ be64_to_cpu(group->last_req_tid));
+ if (list_empty(&group->pending_list)) {
+ sprintf(pending_str, "No");
+ } else {
+ req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+ sprintf(pending_str, "Yes(TID=0x%llx)",
+ be64_to_cpu(req->sa_mad.mad_hdr.tid));
+ }
+ len += sprintf(buf + len, "%1d [%02d,%02d,%02d] %4d %4s %5s ",
+ group->rec.scope_join_state & 0xf,
+ group->members[2], group->members[1], group->members[0],
+ atomic_read(&group->refcount),
+ pending_str,
+ state_str);
+ for (f = 0; f < MAX_VFS; ++f)
+ if (group->func[f].state == MCAST_MEMBER)
+ len += sprintf(buf + len, "%d[%1x] ",
+ f, group->func[f].join_state);
+
+ len += sprintf(buf + len, "\t\t(%4hx %4x %2x %2x %2x %2x %2x "
+ "%4x %4x %2x %2x)\n",
+ be16_to_cpu(group->rec.pkey),
+ be32_to_cpu(group->rec.qkey),
+ (group->rec.mtusel_mtu & 0xc0) >> 6,
+ group->rec.mtusel_mtu & 0x3f,
+ group->rec.tclass,
+ (group->rec.ratesel_rate & 0xc0) >> 6,
+ group->rec.ratesel_rate & 0x3f,
+ (be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0xf0000000) >> 28,
+ (be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x0fffff00) >> 8,
+ be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x000000ff,
+ group->rec.proxy_join);
+
+ return len;
+}
+
int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx)
{
char name[20];
@@ -995,6 +1061,7 @@ static void force_clean_group(struct mcast_group *group)
list_del(&req->group_list);
kfree(req);
}
+ del_sysfs_port_mcg_attr(group->demux->dev, group->demux->port, &group->dentry.attr);
rb_erase(&group->node, &group->demux->mcg_table);
kfree(group);
}
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index f3f75f8..e57a220 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -427,6 +427,35 @@ struct pkey_mgt {
struct kobject *device_parent[MLX4_MFUNC_MAX];
};
+struct mlx4_ib_iov_sysfs_attr {
+ void *ctx;
+ struct kobject *kobj;
+ unsigned long data;
+ u32 entry_num;
+ char name[15];
+ struct device_attribute dentry;
+ struct device *dev;
+};
+
+struct mlx4_ib_iov_sysfs_attr_ar {
+ struct mlx4_ib_iov_sysfs_attr dentries[3 * NUM_ALIAS_GUID_PER_PORT + 1];
+};
+
+struct mlx4_ib_iov_port {
+ char name[100];
+ u8 num;
+ struct mlx4_ib_dev *dev;
+ struct list_head list;
+ struct mlx4_ib_iov_sysfs_attr_ar *dentr_ar;
+ struct ib_port_attr attr;
+ struct kobject *cur_port;
+ struct kobject *admin_alias_parent;
+ struct kobject *gids_parent;
+ struct kobject *pkeys_parent;
+ struct kobject *mcgs_parent;
+ struct mlx4_ib_iov_sysfs_attr mcg_dentry;
+};
+
struct mlx4_ib_dev {
struct ib_device ib_dev;
struct mlx4_dev *dev;
@@ -448,6 +477,10 @@ struct mlx4_ib_dev {
int counters[MLX4_MAX_PORTS];
int *eq_table;
int eq_added;
+ struct kobject *iov_parent;
+ struct kobject *ports_parent;
+ struct kobject *dev_ports_parent[MLX4_MFUNC_MAX];
+ struct mlx4_ib_iov_port iov_ports[MLX4_MAX_PORTS];
struct pkey_mgt pkeys;
};
@@ -680,4 +713,14 @@ void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev,
int block_num, u8 port_num,
u8 *p_data);
+int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+ struct attribute *attr);
+void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+ struct attribute *attr);
+ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index);
+
+int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *device) ;
+
+void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device);
+
#endif /* MLX4_IB_H */
diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
new file mode 100644
index 0000000..a2be11a
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/sysfs.c
@@ -0,0 +1,794 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*#include "core_priv.h"*/
+#include "mlx4_ib.h"
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/stat.h>
+
+#include <rdma/ib_mad.h>
+/*show_admin_alias_guid returns the administratively assigned value of that GUID.
+ * Values returned in buf parameter string:
+ * 0 - requests opensm to assign a value.
+ * ffffffffffffffff - delete this entry.
+ * other - value assigned by administrator.
+ */
+static ssize_t show_admin_alias_guid(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ int record_num;/*0-15*/
+ int guid_index_in_rec; /*0 - 7*/
+ struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+ container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+ struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+ struct mlx4_ib_dev *mdev = port->dev;
+
+ record_num = mlx4_ib_iov_dentry->entry_num / 8 ;
+ guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8 ;
+
+ return sprintf(buf, "%llx\n",
+ be64_to_cpu(*(__be64 *)&mdev->sriov.alias_guid.
+ ports_guid[port->num - 1].
+ all_rec_per_port[record_num].
+ all_recs[8 * guid_index_in_rec]));
+}
+
+/* store_admin_alias_guid stores the (new) administratively assigned value of that GUID.
+ * Values in buf parameter string:
+ * 0 - requests opensm to assign a value.
+ * 0xffffffffffffffff - delete this entry.
+ * other - guid value assigned by the administrator.
+ */
+static ssize_t store_admin_alias_guid(struct device *dev,
+ struct device_attribute *attr,
+ const char *buf, size_t count)
+{
+ int record_num;/*0-15*/
+ int guid_index_in_rec; /*0 - 7*/
+ struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+ container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+ struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+ struct mlx4_ib_dev *mdev = port->dev;
+ u64 sysadmin_ag_val;
+
+ record_num = mlx4_ib_iov_dentry->entry_num / 8;
+ guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8;
+ if (0 == record_num && 0 == guid_index_in_rec) {
+ pr_err("GUID 0 block 0 is RO\n");
+ return count;
+ }
+ sscanf(buf, "%llx", &sysadmin_ag_val);
+ *(__be64 *)&mdev->sriov.alias_guid.ports_guid[port->num - 1].
+ all_rec_per_port[record_num].
+ all_recs[GUID_REC_SIZE * guid_index_in_rec] =
+ cpu_to_be64(sysadmin_ag_val);
+
+ /* Change the state to be pending for update */
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].status
+ = MLX4_GUID_INFO_STATUS_IDLE ;
+
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method
+ = MLX4_GUID_INFO_RECORD_SET;
+
+ switch (sysadmin_ag_val) {
+ case MLX4_GUID_FOR_DELETE_VAL:
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method
+ = MLX4_GUID_INFO_RECORD_DELETE;
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+ = MLX4_GUID_SYSADMIN_ASSIGN;
+ break;
+ /* The sysadmin requests the SM to re-assign */
+ case MLX4_NOT_SET_GUID:
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+ = MLX4_GUID_DRIVER_ASSIGN;
+ break;
+ /* The sysadmin requests a specific value.*/
+ default:
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+ = MLX4_GUID_SYSADMIN_ASSIGN;
+ break;
+ }
+
+ /* set the record index */
+ mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].guid_indexes
+ = mlx4_ib_get_aguid_comp_mask_from_ix(guid_index_in_rec);
+
+ mlx4_ib_init_alias_guid_work(mdev, port->num - 1);
+
+ return count;
+}
+
+static ssize_t show_port_gid(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+ container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+ struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+ struct mlx4_ib_dev *mdev = port->dev;
+ union ib_gid gid;
+ ssize_t ret;
+
+ ret = __mlx4_ib_query_gid(&mdev->ib_dev, port->num,
+ mlx4_ib_iov_dentry->entry_num, &gid, 1);
+ if (ret)
+ return ret;
+ ret = sprintf(buf, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n",
+ be16_to_cpu(((__be16 *) gid.raw)[0]),
+ be16_to_cpu(((__be16 *) gid.raw)[1]),
+ be16_to_cpu(((__be16 *) gid.raw)[2]),
+ be16_to_cpu(((__be16 *) gid.raw)[3]),
+ be16_to_cpu(((__be16 *) gid.raw)[4]),
+ be16_to_cpu(((__be16 *) gid.raw)[5]),
+ be16_to_cpu(((__be16 *) gid.raw)[6]),
+ be16_to_cpu(((__be16 *) gid.raw)[7]));
+ return ret;
+}
+
+static ssize_t show_phys_port_pkey(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+ container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+ struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+ struct mlx4_ib_dev *mdev = port->dev;
+ u16 pkey;
+ ssize_t ret;
+
+ ret = __mlx4_ib_query_pkey(&mdev->ib_dev, port->num,
+ mlx4_ib_iov_dentry->entry_num, &pkey, 1);
+ if (ret)
+ return ret;
+
+ return sprintf(buf, "0x%04x\n", pkey);
+}
+
+#define DENTRY_REMOVE(_dentry) \
+do { \
+ sysfs_remove_file((_dentry)->kobj, &(_dentry)->dentry.attr); \
+} while (0);
+
+static int create_sysfs_entry(void *_ctx, struct mlx4_ib_iov_sysfs_attr *_dentry,
+ char *_name, struct kobject *_kobj,
+ ssize_t (*show)(struct device *dev,
+ struct device_attribute *attr,
+ char *buf),
+ ssize_t (*store)(struct device *dev,
+ struct device_attribute *attr,
+ const char *buf, size_t count)
+ )
+{
+ int ret = 0;
+ struct mlx4_ib_iov_sysfs_attr *vdentry = _dentry;
+
+ vdentry->ctx = _ctx;
+ vdentry->dentry.show = show;
+ vdentry->dentry.store = store;
+ sysfs_attr_init(&vdentry->dentry.attr);
+ vdentry->dentry.attr.name = vdentry->name;
+ vdentry->dentry.attr.mode = 0;
+ vdentry->kobj = _kobj;
+ snprintf(vdentry->name, 15, "%s", _name);
+
+ if (vdentry->dentry.store)
+ vdentry->dentry.attr.mode |= S_IWUSR;
+
+ if (vdentry->dentry.show)
+ vdentry->dentry.attr.mode |= S_IRUGO;
+
+ ret = sysfs_create_file(vdentry->kobj, &vdentry->dentry.attr);
+ if (ret) {
+ pr_err("failed to create %s\n", vdentry->dentry.attr.name);
+ vdentry->ctx = NULL;
+ return ret;
+ }
+
+ return ret;
+}
+
+int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+ struct attribute *attr)
+{
+ struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1];
+ int ret;
+
+ ret = sysfs_create_file(port->mcgs_parent, attr);
+ if (ret)
+ pr_err("failed to create %s\n", attr->name);
+
+ return ret;
+}
+
+void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+ struct attribute *attr)
+{
+ struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1];
+
+ sysfs_remove_file(port->mcgs_parent, attr);
+}
+
+static int add_port_entries(struct mlx4_ib_dev *device, int port_num)
+{
+ int i;
+ char buff[10];
+ struct mlx4_ib_iov_port *port = NULL;
+ int ret = 0 ;
+ struct ib_port_attr attr;
+
+ /* get the physical gid and pkey table sizes.*/
+ ret = __mlx4_ib_query_port(&device->ib_dev, port_num, &attr, 1);
+ if (ret)
+ goto err;
+
+ port = &device->iov_ports[port_num - 1];
+ port->dev = device;
+ port->num = port_num;
+ /* Directory structure:
+ * iov -
+ * port num -
+ * admin_guids
+ * gids (operational)
+ * mcg_table
+ */
+ port->dentr_ar = kzalloc(sizeof(struct mlx4_ib_iov_sysfs_attr_ar),
+ GFP_KERNEL);
+ if (!port->dentr_ar) {
+ ret = -ENOMEM;
+ goto err;
+ }
+ sprintf(buff, "%d", port_num);
+ port->cur_port = kobject_create_and_add(buff,
+ kobject_get(device->ports_parent));
+ if (!port->cur_port) {
+ ret = -ENOMEM;
+ goto kobj_create_err;
+ }
+ /* admin GUIDs */
+ port->admin_alias_parent = kobject_create_and_add("admin_guids",
+ kobject_get(port->cur_port));
+ if (!port->admin_alias_parent) {
+ ret = -ENOMEM;
+ goto err_admin_guids;
+ }
+ for (i = 0 ; i < attr.gid_tbl_len; i++) {
+ sprintf(buff, "%d", i);
+ port->dentr_ar->dentries[i].entry_num = i;
+ ret = create_sysfs_entry(port, &port->dentr_ar->dentries[i],
+ buff, port->admin_alias_parent,
+ show_admin_alias_guid, store_admin_alias_guid);
+ if (ret)
+ goto err_admin_alias_parent;
+ }
+
+ /* gids subdirectory (operational gids) */
+ port->gids_parent = kobject_create_and_add("gids",
+ kobject_get(port->cur_port));
+ if (!port->gids_parent) {
+ ret = -ENOMEM;
+ goto err_gids;
+ }
+
+ for (i = 0 ; i < attr.gid_tbl_len; i++) {
+ sprintf(buff, "%d", i);
+ port->dentr_ar->dentries[attr.gid_tbl_len + i].entry_num = i;
+ ret = create_sysfs_entry(port,
+ &port->dentr_ar->dentries[attr.gid_tbl_len + i],
+ buff,
+ port->gids_parent, show_port_gid, NULL);
+ if (ret)
+ goto err_gids_parent;
+ }
+
+ /* physical port pkey table */
+ port->pkeys_parent =
+ kobject_create_and_add("pkeys", kobject_get(port->cur_port));
+ if (!port->pkeys_parent) {
+ ret = -ENOMEM;
+ goto err_pkeys;
+ }
+
+ for (i = 0 ; i < attr.pkey_tbl_len; i++) {
+ sprintf(buff, "%d", i);
+ port->dentr_ar->dentries[2 * attr.gid_tbl_len + i].entry_num = i;
+ ret = create_sysfs_entry(port,
+ &port->dentr_ar->dentries[2 * attr.gid_tbl_len + i],
+ buff, port->pkeys_parent,
+ show_phys_port_pkey, NULL);
+ if (ret)
+ goto err_pkeys_parent;
+ }
+
+ /* MCGs table */
+ port->mcgs_parent =
+ kobject_create_and_add("mcgs", kobject_get(port->cur_port));
+ if (!port->mcgs_parent) {
+ ret = -ENOMEM;
+ goto err_mcgs;
+ }
+ return 0;
+
+err_mcgs:
+ kobject_put(port->cur_port);
+
+err_pkeys_parent:
+ kobject_put(port->pkeys_parent);
+
+err_pkeys:
+ kobject_put(port->cur_port);
+
+err_gids_parent:
+ kobject_put(port->gids_parent);
+
+err_gids:
+ kobject_put(port->cur_port);
+
+err_admin_alias_parent:
+ kobject_put(port->admin_alias_parent);
+
+err_admin_guids:
+ kobject_put(port->cur_port);
+ kobject_put(port->cur_port); /* once more for create_and_add buff */
+
+kobj_create_err:
+ kobject_put(device->ports_parent);
+ kfree(port->dentr_ar);
+
+err:
+ pr_err("add_port_entries FAILED: for port:%d, error: %d\n",
+ port_num, ret);
+ return ret;
+}
+
+static void get_name(struct mlx4_ib_dev *dev, char *name, int i, int max)
+{
+ char base_name[9];
+
+ /* pci_name format is: bus:dev:func -> xxxx:yy:zz.n */
+ strlcpy(name, pci_name(dev->dev->pdev), max);
+ strncpy(base_name, name, 8); /*till xxxx:yy:*/
+ base_name[8] = '\0';
+ /* with no ARI only 3 last bits are used so when the fn is higher than 8
+ * need to add it to the dev num, so count in the last number will be
+ * modulo 8 */
+ sprintf(name, "%s%.2d.%d", base_name, (i/8), (i%8));
+}
+
+struct mlx4_port {
+ struct kobject kobj;
+ struct mlx4_ib_dev *dev;
+ struct attribute_group pkey_group;
+ struct attribute_group gid_group;
+ u8 port_num;
+ int slave;
+};
+
+
+static void mlx4_port_release(struct kobject *kobj)
+{
+ struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+ struct attribute *a;
+ int i;
+
+ for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
+ kfree(a);
+ kfree(p->pkey_group.attrs);
+ for (i = 0; (a = p->gid_group.attrs[i]); ++i)
+ kfree(a);
+ kfree(p->gid_group.attrs);
+ kfree(p);
+}
+
+struct port_attribute {
+ struct attribute attr;
+ ssize_t (*show)(struct mlx4_port *, struct port_attribute *, char *buf);
+ ssize_t (*store)(struct mlx4_port *, struct port_attribute *,
+ const char *buf, size_t count);
+};
+
+static ssize_t port_attr_show(struct kobject *kobj,
+ struct attribute *attr, char *buf)
+{
+ struct port_attribute *port_attr =
+ container_of(attr, struct port_attribute, attr);
+ struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+
+ if (!port_attr->show)
+ return -EIO;
+ return port_attr->show(p, port_attr, buf);
+}
+
+static ssize_t port_attr_store(struct kobject *kobj,
+ struct attribute *attr,
+ const char *buf, size_t size)
+{
+ struct port_attribute *port_attr =
+ container_of(attr, struct port_attribute, attr);
+ struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+
+ if (!port_attr->store)
+ return -EIO;
+ return port_attr->store(p, port_attr, buf, size);
+}
+
+static const struct sysfs_ops port_sysfs_ops = {
+ .show = port_attr_show,
+ .store = port_attr_store,
+};
+
+static struct kobj_type port_type = {
+ .release = mlx4_port_release,
+ .sysfs_ops = &port_sysfs_ops,
+};
+
+struct port_table_attribute {
+ struct port_attribute attr;
+ char name[8];
+ int index;
+};
+
+static ssize_t show_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
+ char *buf)
+{
+ struct port_table_attribute *tab_attr =
+ container_of(attr, struct port_table_attribute, attr);
+ ssize_t ret = -ENODEV;
+
+ if (p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1][tab_attr->index] >=
+ (p->dev->dev->caps.pkey_table_len[p->port_num]))
+ ret = sprintf(buf, "none\n");
+ else
+ ret = sprintf(buf, "%d\n",
+ p->dev->pkeys.virt2phys_pkey[p->slave]
+ [p->port_num - 1][tab_attr->index]);
+ return ret;
+}
+
+static ssize_t store_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct port_table_attribute *tab_attr =
+ container_of(attr, struct port_table_attribute, attr);
+ int idx;
+ int err;
+
+ /* do not allow remapping Dom0 virtual pkey table */
+ if (p->slave == mlx4_master_func_num(p->dev->dev))
+ return -EINVAL;
+
+ if (!strncasecmp(buf, "no", 2))
+ idx = p->dev->dev->phys_caps.pkey_phys_table_len[p->port_num] - 1;
+ else if (sscanf(buf, "%i", &idx) != 1 ||
+ idx >= p->dev->dev->caps.pkey_table_len[p->port_num] ||
+ idx < 0)
+ return -EINVAL;
+
+ p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1]
+ [tab_attr->index] = idx;
+ mlx4_sync_pkey_table(p->dev->dev, p->slave, p->port_num,
+ tab_attr->index, idx);
+ err = mlx4_gen_pkey_eqe(p->dev->dev, p->slave, p->port_num);
+ if (err) {
+ pr_err("mlx4_gen_pkey_eqe failed for slave %d,"
+ " port %d, index %d\n", p->slave, p->port_num, idx);
+ return err;
+ }
+ return count;
+}
+
+static ssize_t show_port_gid_idx(struct mlx4_port *p,
+ struct port_attribute *attr, char *buf)
+{
+ return sprintf(buf, "%d\n", p->slave);
+}
+
+static struct attribute **
+alloc_group_attrs(ssize_t (*show)(struct mlx4_port *,
+ struct port_attribute *, char *buf),
+ ssize_t (*store)(struct mlx4_port *, struct port_attribute *,
+ const char *buf, size_t count),
+ int len)
+{
+ struct attribute **tab_attr;
+ struct port_table_attribute *element;
+ int i;
+
+ tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
+ if (!tab_attr)
+ return NULL;
+
+ for (i = 0; i < len; i++) {
+ element = kzalloc(sizeof(struct port_table_attribute),
+ GFP_KERNEL);
+ if (!element)
+ goto err;
+ if (snprintf(element->name, sizeof(element->name),
+ "%d", i) >= sizeof(element->name)) {
+ kfree(element);
+ goto err;
+ }
+ sysfs_attr_init(&element->attr.attr);
+ element->attr.attr.name = element->name;
+ if (store) {
+ element->attr.attr.mode = S_IWUSR | S_IRUGO;
+ element->attr.store = store;
+ } else
+ element->attr.attr.mode = S_IRUGO;
+
+ element->attr.show = show;
+ element->index = i;
+ tab_attr[i] = &element->attr.attr;
+ }
+ return tab_attr;
+
+err:
+ while (--i >= 0)
+ kfree(tab_attr[i]);
+ kfree(tab_attr);
+ return NULL;
+}
+
+static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
+{
+ struct mlx4_port *p;
+ int i;
+ int ret;
+
+ p = kzalloc(sizeof *p, GFP_KERNEL);
+ if (!p)
+ return -ENOMEM;
+
+ p->dev = dev;
+ p->port_num = port_num;
+ p->slave = slave;
+
+ ret = kobject_init_and_add(&p->kobj, &port_type,
+ kobject_get(dev->dev_ports_parent[slave]),
+ "%d", port_num);
+ if (ret)
+ goto err_alloc;
+
+ p->pkey_group.name = "pkey_idx";
+ p->pkey_group.attrs =
+ alloc_group_attrs(show_port_pkey, store_port_pkey,
+ dev->dev->caps.pkey_table_len[port_num]);
+ if (!p->pkey_group.attrs)
+ goto err_alloc;
+
+ ret = sysfs_create_group(&p->kobj, &p->pkey_group);
+ if (ret)
+ goto err_free_pkey;
+
+ p->gid_group.name = "gid_idx";
+ p->gid_group.attrs = alloc_group_attrs(show_port_gid_idx, NULL, 1);
+ if (!p->gid_group.attrs)
+ goto err_free_pkey;
+
+ ret = sysfs_create_group(&p->kobj, &p->gid_group);
+ if (ret)
+ goto err_free_gid;
+
+ list_add_tail(&p->kobj.entry, &dev->pkeys.pkey_port_list[slave]);
+ return 0;
+
+err_free_gid:
+ kfree(p->gid_group.attrs[0]);
+ kfree(p->gid_group.attrs);
+
+err_free_pkey:
+ for (i = 0; i < dev->dev->caps.pkey_table_len[port_num]; ++i)
+ kfree(p->pkey_group.attrs[i]);
+ kfree(p->pkey_group.attrs);
+
+err_alloc:
+ kobject_put(dev->dev_ports_parent[slave]);
+ kfree(p);
+ return ret;
+}
+
+static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave)
+{
+ char name[32];
+ int err;
+ int port;
+ struct kobject *p, *t;
+ struct mlx4_port *mport;
+
+ get_name(dev, name, slave, sizeof name);
+
+ dev->pkeys.device_parent[slave] =
+ kobject_create_and_add(name, kobject_get(dev->iov_parent));
+
+ if (!dev->pkeys.device_parent[slave]) {
+ err = -ENOMEM;
+ goto fail_dev;
+ }
+
+ INIT_LIST_HEAD(&dev->pkeys.pkey_port_list[slave]);
+
+ dev->dev_ports_parent[slave] =
+ kobject_create_and_add("ports",
+ kobject_get(dev->pkeys.device_parent[slave]));
+
+ if (!dev->dev_ports_parent[slave]) {
+ err = -ENOMEM;
+ goto err_ports;
+ }
+
+ for (port = 1; port <= dev->dev->caps.num_ports; ++port) {
+ err = add_port(dev, port, slave);
+ if (err)
+ goto err_add;
+ }
+ return 0;
+
+err_add:
+ list_for_each_entry_safe(p, t,
+ &dev->pkeys.pkey_port_list[slave],
+ entry) {
+ list_del(&p->entry);
+ mport = container_of(p, struct mlx4_port, kobj);
+ sysfs_remove_group(p, &mport->pkey_group);
+ sysfs_remove_group(p, &mport->gid_group);
+ kobject_put(p);
+ }
+ kobject_put(dev->dev_ports_parent[slave]);
+
+err_ports:
+ kobject_put(dev->pkeys.device_parent[slave]);
+ /* extra put for the device_parent create_and_add */
+ kobject_put(dev->pkeys.device_parent[slave]);
+
+fail_dev:
+ kobject_put(dev->iov_parent);
+ return err;
+}
+
+static int register_pkey_tree(struct mlx4_ib_dev *device)
+{
+ int i;
+
+ if (!mlx4_is_master(device->dev))
+ return 0;
+
+ for (i = 0; i <= device->dev->num_vfs; ++i)
+ register_one_pkey_tree(device, i);
+
+ return 0;
+}
+
+static void unregister_pkey_tree(struct mlx4_ib_dev *device)
+{
+ int slave;
+ struct kobject *p, *t;
+ struct mlx4_port *port;
+
+ if (!mlx4_is_master(device->dev))
+ return;
+
+ for (slave = device->dev->num_vfs; slave >= 0; --slave) {
+ list_for_each_entry_safe(p, t,
+ &device->pkeys.pkey_port_list[slave],
+ entry) {
+ list_del(&p->entry);
+ port = container_of(p, struct mlx4_port, kobj);
+ sysfs_remove_group(p, &port->pkey_group);
+ sysfs_remove_group(p, &port->gid_group);
+ kobject_put(p);
+ kobject_put(device->dev_ports_parent[slave]);
+ }
+ kobject_put(device->dev_ports_parent[slave]);
+ kobject_put(device->pkeys.device_parent[slave]);
+ kobject_put(device->pkeys.device_parent[slave]);
+ kobject_put(device->iov_parent);
+ }
+}
+
+int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *dev)
+{
+ int i;
+ int ret = 0;
+
+ if (!mlx4_is_master(dev->dev))
+ return 0;
+
+ dev->iov_parent =
+ kobject_create_and_add("iov",
+ kobject_get(dev->ib_dev.ports_parent->parent));
+ if (!dev->iov_parent) {
+ ret = -ENOMEM;
+ goto err;
+ }
+ dev->ports_parent =
+ kobject_create_and_add("ports",
+ kobject_get(dev->iov_parent));
+ if (!dev->iov_parent) {
+ ret = -ENOMEM;
+ goto err_ports;
+ }
+
+ for (i = 1; i <= dev->ib_dev.phys_port_cnt; ++i) {
+ ret = add_port_entries(dev, i);
+ if (ret)
+ goto err_add_entries;
+ }
+
+ ret = register_pkey_tree(dev);
+ if (ret)
+ goto err_add_entries;
+ return 0;
+
+err_add_entries:
+ kobject_put(dev->ports_parent);
+
+err_ports:
+ kobject_put(dev->iov_parent);
+err:
+ kobject_put(dev->ib_dev.ports_parent->parent);
+ pr_err("mlx4_ib_device_register_sysfs error (%d)\n", ret);
+ return ret;
+}
+
+static void unregister_alias_guid_tree(struct mlx4_ib_dev *device)
+{
+ struct mlx4_ib_iov_port *p;
+ int i;
+
+ if (!mlx4_is_master(device->dev))
+ return;
+
+ for (i = 0; i < device->dev->caps.num_ports; i++) {
+ p = &device->iov_ports[i];
+ kobject_put(p->admin_alias_parent);
+ kobject_put(p->gids_parent);
+ kobject_put(p->pkeys_parent);
+ kobject_put(p->mcgs_parent);
+ kobject_put(p->cur_port);
+ kobject_put(p->cur_port);
+ kobject_put(p->cur_port);
+ kobject_put(p->cur_port);
+ kobject_put(p->cur_port);
+ kobject_put(p->dev->ports_parent);
+ kfree(p->dentr_ar);
+ }
+}
+
+void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device)
+{
+ unregister_alias_guid_tree(device);
+ unregister_pkey_tree(device);
+ kobject_put(device->ports_parent);
+ kobject_put(device->iov_parent);
+ kobject_put(device->iov_parent);
+ kobject_put(device->ib_dev.ports_parent->parent);
+}
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 16/22] net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (14 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 15/22] IB/mlx4: Add iov directory in sysfs under the ib device Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 17/22] net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode Jack Morgenstein
` (7 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
1. Slave may not set the IS_SM capability for the port.
2. No DEV_MGR in multifunc mode.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/net/ethernet/mellanox/mlx4/port.c | 10 ++++++++++
include/linux/mlx4/device.h | 5 +++++
2 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx4/port.c b/drivers/net/ethernet/mellanox/mlx4/port.c
index e36dd0f..8ead556 100644
--- a/drivers/net/ethernet/mellanox/mlx4/port.c
+++ b/drivers/net/ethernet/mellanox/mlx4/port.c
@@ -732,6 +732,16 @@ static int mlx4_common_set_port(struct mlx4_dev *dev, int slave, u32 in_mod,
new_cap_mask = ((__be32 *) inbox->buf)[1];
}
+ /* slave may not set the IS_SM capability for the port */
+ if (slave != mlx4_master_func_num(dev) &&
+ (be32_to_cpu(new_cap_mask) & MLX4_PORT_CAP_IS_SM))
+ return -EINVAL;
+
+ /* NO DEV_MGR in multifunc mode */
+ if (mlx4_is_mfunc(dev) &&
+ (be32_to_cpu(new_cap_mask) & MLX4_PORT_CAP_DEV_MGMT_SUP))
+ return -EINVAL;
+
agg_cap_mask = 0;
slave_cap_mask =
priv->mfunc.master.slave_state[slave].ib_cap_mask[port];
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 8fed1fc..863fcea 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -54,6 +54,11 @@ enum {
};
enum {
+ MLX4_PORT_CAP_IS_SM = 1 << 1,
+ MLX4_PORT_CAP_DEV_MGMT_SUP = 1 << 19,
+};
+
+enum {
MLX4_MAX_PORTS = 2,
MLX4_MAX_PORT_PKEYS = 128
};
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 17/22] net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (15 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 16/22] net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 18/22] IB/mlx4: Miscellaneous adjustments to SRIOV IB support Jack Morgenstein
` (6 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Normally, INIT_PORT and CLOSE_PORT are invoked when special QP0
transitions to RTR, or transitions to ERR/RESET respectively.
In SRIOV mode, however, the master is also paravirtualized.
This in turn requires that we not do INIT_PORT until the
entire QP0 path (real QP0 and proxy QP0) is ready to receive.
When the real QP0 goes down, we should indicate that
the port is not active.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/net/ethernet/mellanox/mlx4/fw.c | 71 +++++++++++++++++++++++++------
drivers/net/ethernet/mellanox/mlx4/qp.c | 38 +++++++++++++++--
2 files changed, 92 insertions(+), 17 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 88d61ff..1d11e81 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -1365,6 +1365,19 @@ out:
return err;
}
+/* for IB-type ports only in SRIOV mode. Checks that both proxy QP0
+ * and real QP0 are active, so that the paravirtualized QP0 is ready
+ * to operate */
+static int check_qp0_state(struct mlx4_dev *dev, int function, int port)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ /* irrelevant if not infiniband */
+ if (priv->mfunc.master.qp0_state[port].proxy_qp0_active &&
+ priv->mfunc.master.qp0_state[port].qp0_active)
+ return 1;
+ return 0;
+}
+
int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
@@ -1381,14 +1394,29 @@ int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave,
if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB)
return -ENODEV;
- /* Enable port only if it was previously disabled */
- if (!priv->mfunc.master.init_port_ref[port]) {
- err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
- MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
- if (err)
- return err;
+ if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
+ /* Enable port only if it was previously disabled */
+ if (!priv->mfunc.master.init_port_ref[port]) {
+ err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
+ MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
+ if (err)
+ return err;
+ }
+ priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
+ } else {
+ if (slave == mlx4_master_func_num(dev)) {
+ if (check_qp0_state(dev, slave, port) &&
+ !priv->mfunc.master.qp0_state[port].port_active) {
+ err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
+ MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
+ if (err)
+ return err;
+ priv->mfunc.master.qp0_state[port].port_active = 1;
+ priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
+ }
+ } else
+ priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
}
- priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
++priv->mfunc.master.init_port_ref[port];
return 0;
}
@@ -1463,13 +1491,30 @@ int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave,
if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB)
return -ENODEV;
- if (priv->mfunc.master.init_port_ref[port] == 1) {
- err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT, 1000,
- MLX4_CMD_NATIVE);
- if (err)
- return err;
+
+ if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
+ if (priv->mfunc.master.init_port_ref[port] == 1) {
+ err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
+ 1000, MLX4_CMD_NATIVE);
+ if (err)
+ return err;
+ }
+ priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
+ } else {
+ /* infiniband port */
+ if (slave == mlx4_master_func_num(dev)) {
+ if (!priv->mfunc.master.qp0_state[port].qp0_active &&
+ priv->mfunc.master.qp0_state[port].port_active) {
+ err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
+ 1000, MLX4_CMD_NATIVE);
+ if (err)
+ return err;
+ priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
+ priv->mfunc.master.qp0_state[port].port_active = 0;
+ }
+ } else
+ priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
}
- priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
--priv->mfunc.master.init_port_ref[port];
return 0;
}
diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c
index b8da72b..436ef6c 100644
--- a/drivers/net/ethernet/mellanox/mlx4/qp.c
+++ b/drivers/net/ethernet/mellanox/mlx4/qp.c
@@ -67,10 +67,18 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type)
complete(&qp->free);
}
-static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp)
+/* used for INIT/CLOSE port logic */
+static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0)
{
- return qp->qpn >= dev->caps.sqp_start &&
+ /* qp0 is either the proxy qp0, or the real qp0 */
+ *proxy_qp0 = qp->qpn >= dev->caps.sqp_start &&
qp->qpn <= dev->caps.sqp_start + 1;
+
+ *real_qp0 = mlx4_is_master(dev) &&
+ qp->qpn >= dev->caps.base_sqpn &&
+ qp->qpn <= dev->caps.base_sqpn + 1;
+
+ return *real_qp0 || *proxy_qp0;
}
static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
@@ -122,6 +130,8 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
struct mlx4_priv *priv = mlx4_priv(dev);
struct mlx4_cmd_mailbox *mailbox;
int ret = 0;
+ int real_qp0 = 0;
+ int proxy_qp0 = 0;
u8 port;
if (cur_state >= MLX4_QP_NUM_STATE || new_state >= MLX4_QP_NUM_STATE ||
@@ -133,9 +143,12 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native);
if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR &&
cur_state != MLX4_QP_STATE_RST &&
- is_qp0(dev, qp)) {
+ is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
port = (qp->qpn & 1) + 1;
- priv->mfunc.master.qp0_state[port].qp0_active = 0;
+ if (proxy_qp0)
+ priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0;
+ else
+ priv->mfunc.master.qp0_state[port].qp0_active = 0;
}
return ret;
}
@@ -162,6 +175,23 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
new_state == MLX4_QP_STATE_RST ? 2 : 0,
op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native);
+ if (mlx4_is_master(dev) && is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
+ port = (qp->qpn & 1) + 1;
+ if (cur_state != MLX4_QP_STATE_ERR &&
+ cur_state != MLX4_QP_STATE_RST &&
+ new_state == MLX4_QP_STATE_ERR) {
+ if (proxy_qp0)
+ priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0;
+ else
+ priv->mfunc.master.qp0_state[port].qp0_active = 0;
+ } else if (new_state == MLX4_QP_STATE_RTR) {
+ if (proxy_qp0)
+ priv->mfunc.master.qp0_state[port].proxy_qp0_active = 1;
+ else
+ priv->mfunc.master.qp0_state[port].qp0_active = 1;
+ }
+ }
+
mlx4_free_cmd_mailbox(dev, mailbox);
return ret;
}
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 18/22] IB/mlx4: Miscellaneous adjustments to SRIOV IB support
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (16 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 17/22] net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 19/22] {NET,IB}/mlx4: Activate SRIOV mode for IB Jack Morgenstein
` (5 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
1. allow only master to change node description
2. prevent ah leakage in send mads
3. take device part number from PCI structure, so that
guests see the VF part number (and not the PF part number)
4. place the device revision ID into caps structure at startup
5. SET_PORT in update_gids_task needs to go through wrapper on master.
6. in mlx4_ib_event, PORT_MGMT_EVENT needs be handled in a work queue
on the master, since it propagates events to slaves using GEN_EQE
7. Do not support FMR on slaves.
8. Add spinlock to slave_event(), since it is called both in interrupt
context and in process context (due to 6 above, and also if smp_snoop
is used).
This fix was found and implemented by Saeed Mahameed <saeedm-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 6 +++++-
drivers/infiniband/hw/mlx4/main.c | 26 ++++++++++++++++++--------
drivers/net/ethernet/mellanox/mlx4/cmd.c | 1 +
drivers/net/ethernet/mellanox/mlx4/eq.c | 8 ++++++--
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 +
5 files changed, 31 insertions(+), 11 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index ec64077..cee7bf6 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -705,7 +705,9 @@ static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
if (!out_mad->mad_hdr.status) {
if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV))
smp_snoop(ibdev, port_num, in_mad, prev_lid);
- node_desc_override(ibdev, out_mad);
+ /* slaves get node desc from FW */
+ if (!mlx4_is_slave(to_mdev(ibdev)->dev))
+ node_desc_override(ibdev, out_mad);
}
/* set return bit in status of directed route responses */
@@ -788,6 +790,8 @@ int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
static void send_handler(struct ib_mad_agent *agent,
struct ib_mad_send_wc *mad_send_wc)
{
+ if (mad_send_wc->send_buf->context[0])
+ ib_destroy_ah(mad_send_wc->send_buf->context[0]);
ib_free_send_mad(mad_send_wc->send_buf);
}
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 778fdb8..1166379 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -138,7 +138,7 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
0xffffff;
- props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
+ props->vendor_part_id = dev->dev->pdev->device;
props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
@@ -477,6 +477,9 @@ static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
return 0;
+ if (mlx4_is_slave(to_mdev(ibdev)->dev))
+ return -EOPNOTSUPP;
+
spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
memcpy(ibdev->node_desc, props->node_desc, 64);
spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
@@ -492,7 +495,7 @@ static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
memset(mailbox->buf, 0, 256);
memcpy(mailbox->buf, props->node_desc, 64);
mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
- MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
+ MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
@@ -920,6 +923,7 @@ static int init_node_data(struct mlx4_ib_dev *dev)
if (err)
goto out;
+ dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
out:
@@ -1008,7 +1012,7 @@ static void update_gids_task(struct work_struct *work)
err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
- MLX4_CMD_NATIVE);
+ MLX4_CMD_WRAPPED);
if (err)
pr_warn("set port command failed\n");
else {
@@ -1399,10 +1403,12 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
- ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
- ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
- ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
- ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
+ if (!mlx4_is_slave(ibdev->dev)) {
+ ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
+ ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
+ ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
+ ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
+ }
if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
@@ -1614,7 +1620,11 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
INIT_WORK(&ew->work, handle_port_mgmt_change_event);
memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
ew->ib_dev = ibdev;
- handle_port_mgmt_change_event(&ew->work);
+ /* need to queue only for port owner, which uses GEN_EQE */
+ if (mlx4_is_master(dev))
+ queue_work(wq, &ew->work);
+ else
+ handle_port_mgmt_change_event(&ew->work);
return;
case MLX4_DEV_EVENT_SLAVE_INIT:
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index c13e988..9717002 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -1712,6 +1712,7 @@ int mlx4_multi_func_init(struct mlx4_dev *dev)
INIT_WORK(&priv->mfunc.master.slave_flr_event_work,
mlx4_master_handle_slave_flr);
spin_lock_init(&priv->mfunc.master.slave_state_lock);
+ spin_lock_init(&priv->mfunc.master.slave_eq.event_lock);
priv->mfunc.master.comm_wq =
create_singlethread_workqueue("mlx4_comm");
if (!priv->mfunc.master.comm_wq)
diff --git a/drivers/net/ethernet/mellanox/mlx4/eq.c b/drivers/net/ethernet/mellanox/mlx4/eq.c
index c425826..51c7649 100644
--- a/drivers/net/ethernet/mellanox/mlx4/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx4/eq.c
@@ -164,13 +164,16 @@ static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
{
struct mlx4_priv *priv = mlx4_priv(dev);
struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
- struct mlx4_eqe *s_eqe =
- &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
+ struct mlx4_eqe *s_eqe;
+ unsigned long flags;
+ spin_lock_irqsave(&slave_eq->event_lock, flags);
+ s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
if ((!!(s_eqe->owner & 0x80)) ^
(!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. "
"No free EQE on slave events queue\n", slave);
+ spin_unlock_irqrestore(&slave_eq->event_lock, flags);
return;
}
@@ -183,6 +186,7 @@ static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
queue_work(priv->mfunc.master.comm_wq,
&priv->mfunc.master.slave_event_work);
+ spin_unlock_irqrestore(&slave_eq->event_lock, flags);
}
static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 91e5ead..5c30488 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -473,6 +473,7 @@ struct mlx4_slave_event_eq {
u32 eqn;
u32 cons;
u32 prod;
+ spinlock_t event_lock;
struct mlx4_eqe event_eqe[SLAVE_EVENT_EQ_SIZE];
};
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 19/22] {NET,IB}/mlx4: Activate SRIOV mode for IB
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (17 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 18/22] IB/mlx4: Miscellaneous adjustments to SRIOV IB support Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 20/22] {NET,IB}/mlx4: Paravirtualize Node Guids for slaves Jack Morgenstein
` (4 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Remove the error returns for IB ports from mlx4_ib_add,
mlx4_INIT_PORT_wrapper, and mlx4_CLOSE_PORT_wrapper.
Currently, SRIOV is supported only for devices for which the
link-layer is IB on all ports; RoCE support will be implemented at later time.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/main.c | 8 ++++++--
drivers/net/ethernet/mellanox/mlx4/fw.c | 6 ------
include/linux/mlx4/device.h | 4 ++++
3 files changed, 10 insertions(+), 8 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 1166379..00384b3 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1292,11 +1292,15 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
pr_info_once("%s", mlx4_ib_version);
- if (mlx4_is_mfunc(dev)) {
- pr_warn("IB not yet supported in SRIOV\n");
+ mlx4_foreach_non_ib_transport_port(i, dev)
+ num_ports++;
+
+ if (mlx4_is_mfunc(dev) && num_ports) {
+ dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
return NULL;
}
+ num_ports = 0;
mlx4_foreach_ib_transport_port(i, dev)
num_ports++;
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 1d11e81..e36cbdb 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -1391,9 +1391,6 @@ int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave,
if (priv->mfunc.master.slave_state[slave].init_port_mask & (1 << port))
return 0;
- if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB)
- return -ENODEV;
-
if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
/* Enable port only if it was previously disabled */
if (!priv->mfunc.master.init_port_ref[port]) {
@@ -1489,9 +1486,6 @@ int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave,
(1 << port)))
return 0;
- if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB)
- return -ENODEV;
-
if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
if (priv->mfunc.master.init_port_ref[port] == 1) {
err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 863fcea..f8c9316 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -697,6 +697,10 @@ struct mlx4_init_port_param {
for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \
if ((type) == (dev)->caps.port_mask[(port)])
+#define mlx4_foreach_non_ib_transport_port(port, dev) \
+ for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \
+ if (((dev)->caps.port_mask[port] != MLX4_PORT_TYPE_IB))
+
#define mlx4_foreach_ib_transport_port(port, dev) \
for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \
if (((dev)->caps.port_mask[port] == MLX4_PORT_TYPE_IB) || \
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 20/22] {NET,IB}/mlx4: Paravirtualize Node Guids for slaves
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (18 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 19/22] {NET,IB}/mlx4: Activate SRIOV mode for IB Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 21/22] {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations Jack Morgenstein
` (3 subsequent siblings)
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
This is necessary in order to support > 1 VF/PF in a VM for software that
uses the node guid as a discriminator, such as librdmacm.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 14 ++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 3 +++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 11 +++++++++++
drivers/net/ethernet/mellanox/mlx4/main.c | 22 ++++++++++++++++++++++
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 +
include/linux/mlx4/device.h | 2 ++
6 files changed, 53 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index cee7bf6..f7b9a41 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -35,6 +35,7 @@
#include <rdma/ib_sa.h>
#include <rdma/ib_cache.h>
+#include <linux/random.h>
#include <linux/mlx4/cmd.h>
#include <linux/gfp.h>
#include <rdma/ib_pma.h>
@@ -89,6 +90,12 @@ static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num);
static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
int block, u32 change_bitmap);
+__be64 mlx4_ib_gen_node_guid(void)
+{
+#define NODE_GUID_HI ((u64) (((u64)IB_OPENIB_OUI) << 40))
+ return (cpu_to_be64( NODE_GUID_HI | random32()));
+}
+
__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx)
{
return cpu_to_be64(atomic_inc_return(&ctx->tid)) |
@@ -1958,6 +1965,13 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
return 0;
}
+ for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+ if (i == mlx4_master_func_num(dev->dev))
+ mlx4_put_slave_node_guid(dev->dev, i, dev->ib_dev.node_guid);
+ else
+ mlx4_put_slave_node_guid(dev->dev, i, mlx4_ib_gen_node_guid());
+ }
+
err = mlx4_ib_init_alias_guid_service(dev);
if (err) {
mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n");
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index e57a220..e04cbc9 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -723,4 +723,7 @@ int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *device) ;
void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device);
+__be64 mlx4_ib_gen_node_guid(void);
+
+
#endif /* MLX4_IB_H */
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 9717002..33df1db 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -651,6 +651,7 @@ static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
struct ib_smp *outsmp = outbox->buf;
__be16 *outtab = (__be16 *)(outsmp->data);
__be32 slave_cap_mask;
+ __be64 slave_node_guid;
port = vhcr->in_modifier;
if (smp->base_version == 1 &&
@@ -710,6 +711,16 @@ static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
}
return err;
}
+ if (smp->attr_id == IB_SMP_ATTR_NODE_INFO) {
+ err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
+ vhcr->in_modifier, vhcr->op_modifier,
+ vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
+ if (!err) {
+ slave_node_guid = mlx4_get_slave_node_guid(dev, slave);
+ memcpy(outsmp->data + 12, &slave_node_guid, 8);
+ }
+ return err;
+ }
}
}
if (slave != mlx4_master_func_num(dev) &&
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index cd82607..d5e4238 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -438,6 +438,28 @@ void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int
}
EXPORT_SYMBOL(mlx4_sync_pkey_table);
+void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
+{
+ struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
+
+ if (!mlx4_is_master(dev))
+ return;
+
+ priv->slave_node_guids[slave] = guid;
+}
+EXPORT_SYMBOL(mlx4_put_slave_node_guid);
+
+__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
+{
+ struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
+
+ if (!mlx4_is_master(dev))
+ return 0;
+
+ return priv->slave_node_guids[slave];
+}
+EXPORT_SYMBOL(mlx4_get_slave_node_guid);
+
int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
{
struct mlx4_priv *priv = mlx4_priv(dev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 5c30488..d2ccec6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -734,6 +734,7 @@ struct mlx4_priv {
int reserved_mtts;
int fs_hash_mode;
u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+ __be64 slave_node_guids[MLX4_MFUNC_MAX];
};
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index f8c9316..a8cb149 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -968,5 +968,7 @@ int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port, u8
enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port);
int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, int event, enum slave_port_gen_event *gen_event);
+void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid);
+__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave);
#endif /* MLX4_DEVICE_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 21/22] {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (19 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 20/22] {NET,IB}/mlx4: Paravirtualize Node Guids for slaves Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
[not found] ` <1343983258-6268-22-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 22/22] IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes Jack Morgenstein
` (2 subsequent siblings)
23 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
Previously, the structure of a guest's proxy QPs followed the structure of the PPF
special qps (qp0 port 1, qp0 port 2, qp1 port 1, qp1 port 2, ...).
The guest then did offset calculations on the sqp_base qp number that the PPF passed
to it in QUERY_FUNC_CAP().
This is now changed so that the guest does no offset calculations regarding proxy or
tunnel QPs to use. This change frees the PPF from needing to adhere to a specific
order in allocating proxy and tunnel QPs.
Now, QUERY_FUNC_CAP provides each port individually with its proxy qp0, proxy qp1,
tunnel qp0, and tunnel qp1 QP numbers, and these are used directly where required
(with no offset calculations).
To accomplish this change, several fields were added to the phys_caps structure
for use by the PPF and by non-SRIOV mode:
base_sqpn -- in non-sriov mode, this was formerly sqp_start.
base_proxy_sqpn -- the first physical proxy qp number -- used by PPF
base_tunnel_sqpn -- the first physical tunnel qp number -- used by PPF.
The current code in the PPF still adheres to the previous layout of sqp's, proxy-sqp's,
and tunnel-sqp's. However, the PPF can change this layout without affecting
VF or (paravirtualized) PF code.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 12 +-
drivers/infiniband/hw/mlx4/qp.c | 104 +++++++++++------
drivers/net/ethernet/mellanox/mlx4/fw.c | 175 ++++++++++++++++------------
drivers/net/ethernet/mellanox/mlx4/fw.h | 14 ++-
drivers/net/ethernet/mellanox/mlx4/main.c | 61 ++++++++--
drivers/net/ethernet/mellanox/mlx4/qp.c | 71 +++++++++----
include/linux/mlx4/device.h | 16 ++-
7 files changed, 292 insertions(+), 161 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index f7b9a41..31b0559 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -501,7 +501,7 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
} else
tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
- dqpn = dev->dev->caps.sqp_start + 8 * slave + port + (dest_qpt * 2) - 1;
+ dqpn = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave + port + (dest_qpt * 2) - 1;
/* get tunnel tx data buf for slave */
src_qp = tun_qp->qp;
@@ -1070,9 +1070,9 @@ static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
{
- int slave_start = dev->dev->caps.sqp_start + 8 * slave;
+ int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave;
- return (qpn >= slave_start && qpn <= slave_start + 1);
+ return (qpn >= proxy_start && qpn <= proxy_start + 1);
}
@@ -1187,14 +1187,14 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc
int slave;
/* Get slave that sent this packet */
- if (wc->src_qp < dev->dev->caps.sqp_start ||
- wc->src_qp >= dev->dev->caps.base_tunnel_sqpn ||
+ if (wc->src_qp < dev->dev->phys_caps.base_proxy_sqpn ||
+ wc->src_qp >= dev->dev->phys_caps.base_proxy_sqpn + 8 * MLX4_MFUNC_MAX ||
(wc->src_qp & 0x1) != ctx->port - 1 ||
wc->src_qp & 0x4) {
mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp);
return;
}
- slave = ((wc->src_qp & ~0x7) - dev->dev->caps.sqp_start) / 8;
+ slave = ((wc->src_qp & ~0x7) - dev->dev->phys_caps.base_proxy_sqpn) / 8;
if (slave != ctx->slave) {
mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: "
"belongs to another slave\n", wc->src_qp);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 3a3a690..24dcff8 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -116,33 +116,57 @@ static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
if (!mlx4_is_master(dev->dev))
return 0;
- return qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
- qp->mqp.qpn < dev->dev->caps.base_sqpn +
- 8 + 16 * MLX4_MFUNC_MAX;
+ return qp->mqp.qpn >= dev->dev->phys_caps.base_tunnel_sqpn &&
+ qp->mqp.qpn < dev->dev->phys_caps.base_tunnel_sqpn +
+ 8 * MLX4_MFUNC_MAX;
}
static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
- return ((mlx4_is_master(dev->dev) &&
- qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
- qp->mqp.qpn <= dev->dev->caps.base_sqpn + 3) ||
- (qp->mqp.qpn >= dev->dev->caps.sqp_start &&
- qp->mqp.qpn <= dev->dev->caps.sqp_start + 3));
+ int proxy_sqp = 0;
+ int real_sqp = 0;
+ int i;
+ /* PPF or Native -- real SQP */
+ real_sqp = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+ qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+ qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 3);
+ if (real_sqp)
+ return 1;
+ /* VF or PF -- proxy SQP */
+ if (mlx4_is_mfunc(dev->dev)) {
+ for (i = 0; i < dev->dev->caps.num_ports; i++) {
+ if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i] ||
+ qp->mqp.qpn == dev->dev->caps.qp1_proxy[i]) {
+ proxy_sqp = 1;
+ break;
+ }
+ }
+ }
+ return proxy_sqp;
}
/* used for INIT/CLOSE port logic */
static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
- int qp0;
-
- /* qp0 is either the proxy qp0, or the real qp0 */
- qp0 = (qp->mqp.qpn >= dev->dev->caps.sqp_start &&
- qp->mqp.qpn <= dev->dev->caps.sqp_start + 1) ||
- (mlx4_is_master(dev->dev) &&
- qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
- qp->mqp.qpn <= dev->dev->caps.base_sqpn + 1);
-
- return qp0;
+ int proxy_qp0 = 0;
+ int real_qp0 = 0;
+ int i;
+ /* PPF or Native -- real QP0 */
+ real_qp0 = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+ qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+ qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 1);
+ if (real_qp0)
+ return 1;
+ /* VF or PF -- proxy QP0 */
+ if (mlx4_is_mfunc(dev->dev)) {
+ for (i = 0; i < dev->dev->caps.num_ports; i++) {
+ if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i]) {
+ proxy_qp0 = 1;
+ break;
+ }
+ }
+ }
+ return proxy_qp0;
}
static void *get_wqe(struct mlx4_ib_qp *qp, int offset)
@@ -607,8 +631,10 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
qp_type = MLX4_IB_QPT_TUN_SMI_OWNER;
else
qp_type = MLX4_IB_QPT_TUN_SMI;
- qpn = dev->dev->caps.base_tunnel_sqpn + 8 * tnl_init->slave +
- tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
+ /* we are definitely in the PPF here, since we are creating
+ * tunnel QPs. base_tunnel_sqpn is therefore valid. */
+ qpn = dev->dev->phys_caps.base_tunnel_sqpn + 8 * tnl_init->slave
+ + tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
sqpn = qpn;
}
@@ -630,12 +656,6 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
qp->mlx4_ib_qp_type = qp_type;
- if (mlx4_is_mfunc(dev->dev) &&
- (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI)) {
- qpn -= 8;
- sqpn -= 8;
- }
-
mutex_init(&qp->mutex);
spin_lock_init(&qp->sq.lock);
spin_lock_init(&qp->rq.lock);
@@ -935,6 +955,23 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
del_gid_entries(qp);
}
+static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr)
+{
+ /* Native or PPF */
+ if (!mlx4_is_mfunc(dev->dev) ||
+ (mlx4_is_master(dev->dev) &&
+ attr->create_flags & MLX4_IB_SRIOV_SQP)) {
+ return dev->dev->phys_caps.base_sqpn +
+ (attr->qp_type == IB_QPT_SMI ? 0 : 2) +
+ attr->port_num - 1;
+ }
+ /* PF or VF -- creating proxies */
+ if (attr->qp_type == IB_QPT_SMI)
+ return dev->dev->caps.qp0_proxy[attr->port_num - 1];
+ else
+ return dev->dev->caps.qp1_proxy[attr->port_num - 1];
+}
+
struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
@@ -998,9 +1035,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
return ERR_PTR(-EINVAL);
err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
- to_mdev(pd->device)->dev->caps.sqp_start +
- (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
- init_attr->port_num - 1,
+ get_sqp_num(to_mdev(pd->device), init_attr),
&qp);
if (err)
return ERR_PTR(err);
@@ -1643,8 +1678,7 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
else
sqp->ud_header.bth.destination_qpn =
- cpu_to_be32(mdev->dev->caps.base_tunnel_sqpn +
- sqp->qp.port - 1);
+ cpu_to_be32(mdev->dev->caps.qp0_tunnel[sqp->qp.port - 1]);
sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
@@ -2008,10 +2042,10 @@ static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
cpu_to_be32(0xf0000000);
memcpy(dseg->av, &sqp_av, sizeof(struct mlx4_av));
- dseg->dqpn = cpu_to_be32(dev->dev->caps.base_tunnel_sqpn +
- qpt * 2 + port - 1);
- /* use well-known qkey from the QPC */
- dseg->qkey = cpu_to_be32(0x80000000);
+ /* This function used only for sending on QP1 proxies */
+ dseg->dqpn = cpu_to_be32(dev->dev->caps.qp1_tunnel[port - 1]);
+ /* Use QKEY from the QP context, which is set by master */
+ dseg->qkey = cpu_to_be32(IB_QP_SET_QKEY);
}
static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len)
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index e36cbdb..27531e8 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -184,8 +184,6 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
#define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET 0x28
#define QUERY_FUNC_CAP_MAX_EQ_OFFSET 0x2c
#define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0X30
-#define QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET 0X44
-#define QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET 0X48
#define QUERY_FUNC_CAP_FMR_FLAG 0x80
#define QUERY_FUNC_CAP_FLAG_RDMA 0x40
@@ -196,21 +194,41 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
#define QUERY_FUNC_CAP_RDMA_PROPS_OFFSET 0x8
#define QUERY_FUNC_CAP_ETH_PROPS_OFFSET 0xc
+#define QUERY_FUNC_CAP_QP0_TUNNEL 0x10
+#define QUERY_FUNC_CAP_QP0_PROXY 0x14
+#define QUERY_FUNC_CAP_QP1_TUNNEL 0x18
+#define QUERY_FUNC_CAP_QP1_PROXY 0x1c
+
#define QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC 0x40
#define QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN 0x80
+
+
#define QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID 0x80
if (vhcr->op_modifier == 1) {
- field = vhcr->in_modifier;
- MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
-
field = 0;
/* ensure force vlan and force mac bits are not set */
MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
/* ensure that phy_wqe_gid bit is not set */
MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
+ field = vhcr->in_modifier; /* phys-port = logical-port */
+ MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
+
+ /* size is now the QP number */
+ size = dev->phys_caps.base_tunnel_sqpn + 8 * slave + field - 1;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_TUNNEL);
+
+ size += 2;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_TUNNEL);
+
+ size = dev->phys_caps.base_proxy_sqpn + 8 * slave + field - 1;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_PROXY);
+
+ size += 2;
+ MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_PROXY);
+
} else if (vhcr->op_modifier == 0) {
/* enable rdma and ethernet interfaces */
field = (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA);
@@ -249,117 +267,124 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
size = dev->caps.num_mgms + dev->caps.num_amgms;
MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
- size = dev->caps.base_tunnel_sqpn + 8 * slave;
- MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
-
- size = dev->caps.sqp_start + 8 * slave;
- MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
-
} else
err = -EINVAL;
return err;
}
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap)
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+ struct mlx4_func_cap *func_cap)
{
struct mlx4_cmd_mailbox *mailbox;
u32 *outbox;
- u8 field;
+ u8 field, op_modifier;
u32 size;
- int i;
int err = 0;
+ op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */
mailbox = mlx4_alloc_cmd_mailbox(dev);
if (IS_ERR(mailbox))
return PTR_ERR(mailbox);
- err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_FUNC_CAP,
+ err = mlx4_cmd_box(dev, 0, mailbox->dma, gen_or_port, op_modifier,
+ MLX4_CMD_QUERY_FUNC_CAP,
MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
if (err)
goto out;
outbox = mailbox->buf;
- MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
- if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
- mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
- err = -EPROTONOSUPPORT;
- goto out;
- }
- func_cap->flags = field;
-
- MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
- func_cap->num_ports = field;
+ if (!op_modifier) {
+ MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
+ if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
+ mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
+ err = -EPROTONOSUPPORT;
+ goto out;
+ }
+ func_cap->flags = field;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
- func_cap->pf_context_behaviour = size;
+ MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
+ func_cap->num_ports = field;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
- func_cap->qp_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
+ func_cap->pf_context_behaviour = size;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
- func_cap->srq_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
+ func_cap->qp_quota = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
- func_cap->cq_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
+ func_cap->srq_quota = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
- func_cap->max_eq = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
+ func_cap->cq_quota = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
- func_cap->reserved_eq = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
+ func_cap->max_eq = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
- func_cap->mpt_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
+ func_cap->reserved_eq = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
- func_cap->mtt_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
+ func_cap->mpt_quota = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
- func_cap->mcg_quota = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
+ func_cap->mtt_quota = size & 0xFFFFFF;
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
- func_cap->base_tunnel_qpn = size & 0xFFFFFF;
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
+ func_cap->mcg_quota = size & 0xFFFFFF;
+ goto out;
+ }
- MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
- func_cap->base_proxy_qpn = size & 0xFFFFFF;
+ /* logical port query */
+ if (gen_or_port > dev->caps.num_ports) {
+ err = -EINVAL;
+ goto out;
+ }
- for (i = 1; i <= func_cap->num_ports; ++i) {
- err = mlx4_cmd_box(dev, 0, mailbox->dma, i, 1,
- MLX4_CMD_QUERY_FUNC_CAP,
- MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
- if (err)
+ if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_ETH) {
+ MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
+ if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) {
+ mlx4_err(dev, "VLAN is enforced on this port\n");
+ err = -EPROTONOSUPPORT;
goto out;
+ }
- if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH) {
- MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
- if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) {
- mlx4_err(dev, "VLAN is enforced on this port\n");
- err = -EPROTONOSUPPORT;
- goto out;
- }
-
- if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) {
- mlx4_err(dev, "Force mac is enabled on this port\n");
- err = -EPROTONOSUPPORT;
- goto out;
- }
- } else if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB) {
- MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
- if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) {
- mlx4_err(dev, "phy_wqe_gid is "
- "enforced on this ib port\n");
- err = -EPROTONOSUPPORT;
- goto out;
- }
+ if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) {
+ mlx4_err(dev, "Force mac is enabled on this port\n");
+ err = -EPROTONOSUPPORT;
+ goto out;
}
+ } else if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_IB) {
+ MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
+ if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) {
+ mlx4_err(dev, "phy_wqe_gid is "
+ "enforced on this ib port\n");
+ err = -EPROTONOSUPPORT;
+ goto out;
+ }
+ }
- MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
- func_cap->physical_port[i] = field;
+ MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
+ func_cap->physical_port = field;
+ if (func_cap->physical_port != gen_or_port) {
+ err = -ENOSYS;
+ goto out;
}
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_TUNNEL);
+ func_cap->qp0_tunnel_qpn = size & 0xFFFFFF;
+
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_PROXY);
+ func_cap->qp0_proxy_qpn = size & 0xFFFFFF;
+
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_TUNNEL);
+ func_cap->qp1_tunnel_qpn = size & 0xFFFFFF;
+
+ MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_PROXY);
+ func_cap->qp1_proxy_qpn = size & 0xFFFFFF;
+
/* All other resources are allocated by the master, but we still report
* 'num' and 'reserved' capabilities as follows:
* - num remains the maximum resource index
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h
index ced1de5..85abe9c 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.h
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.h
@@ -134,11 +134,12 @@ struct mlx4_func_cap {
int max_eq;
int reserved_eq;
int mcg_quota;
- u32 base_qpn;
- u32 base_tunnel_qpn;
- u32 base_proxy_qpn;
- u8 physical_port[MLX4_MAX_PORTS + 1];
- u8 port_flags[MLX4_MAX_PORTS + 1];
+ u32 qp0_tunnel_qpn;
+ u32 qp0_proxy_qpn;
+ u32 qp1_tunnel_qpn;
+ u32 qp1_proxy_qpn;
+ u8 physical_port;
+ u8 port_flags;
};
struct mlx4_adapter {
@@ -195,7 +196,8 @@ struct mlx4_set_ib_param {
};
int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap);
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap);
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+ struct mlx4_func_cap *func_cap);
int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_vhcr *vhcr,
struct mlx4_cmd_mailbox *inbox,
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index d5e4238..aacc577 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -413,15 +413,16 @@ static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
{
u32 qk = MLX4_RESERVED_QKEY_BASE;
- if (qpn >= dev->caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
- qpn < dev->caps.sqp_start)
+
+ if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
+ qpn < dev->phys_caps.base_proxy_sqpn)
return -EINVAL;
- if (qpn >= dev->caps.base_tunnel_sqpn)
+ if (qpn >= dev->phys_caps.base_tunnel_sqpn)
/* tunnel qp */
- qk += qpn - dev->caps.base_tunnel_sqpn;
+ qk += qpn - dev->phys_caps.base_tunnel_sqpn;
else
- qk += qpn - dev->caps.sqp_start;
+ qk += qpn - dev->phys_caps.base_proxy_sqpn;
*qkey = qk;
return 0;
}
@@ -530,9 +531,10 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
}
memset(&func_cap, 0, sizeof(func_cap));
- err = mlx4_QUERY_FUNC_CAP(dev, &func_cap);
+ err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
if (err) {
- mlx4_err(dev, "QUERY_FUNC_CAP command failed, aborting.\n");
+ mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
+ err);
return err;
}
@@ -560,12 +562,33 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
return -ENODEV;
}
+ dev->caps.qp0_tunnel = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp0_proxy = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp1_tunnel = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp1_proxy = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+
+ if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
+ !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
+ err = -ENOMEM;
+ goto err_mem;
+ }
+
for (i = 1; i <= dev->caps.num_ports; ++i) {
+ err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
+ if (err) {
+ mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
+ " port %d, aborting (%d).\n", i, err);
+ goto err_mem;
+ }
+ dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
+ dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
+ dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
+ dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
dev->caps.port_mask[i] = dev->caps.port_type[i];
if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
&dev->caps.gid_table_len[i],
&dev->caps.pkey_table_len[i]))
- return -ENODEV;
+ goto err_mem;
}
if (dev->caps.uar_page_size * (dev->caps.num_uars -
@@ -575,14 +598,20 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
"PCI resource 2 size of 0x%llx, aborting.\n",
dev->caps.uar_page_size * dev->caps.num_uars,
(unsigned long long) pci_resource_len(dev->pdev, 2));
- return -ENODEV;
+ goto err_mem;
}
- /* Calculate our sqp_start */
- dev->caps.sqp_start = func_cap.base_proxy_qpn;
- dev->caps.base_tunnel_sqpn = func_cap.base_tunnel_qpn;
-
return 0;
+
+err_mem:
+ kfree(dev->caps.qp0_tunnel);
+ kfree(dev->caps.qp0_proxy);
+ kfree(dev->caps.qp1_tunnel);
+ kfree(dev->caps.qp1_proxy);
+ dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
+ dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
+
+ return err;
}
/*
@@ -2256,6 +2285,12 @@ static void mlx4_remove_one(struct pci_dev *pdev)
if (!mlx4_is_slave(dev))
mlx4_free_ownership(dev);
+
+ kfree(dev->caps.qp0_tunnel);
+ kfree(dev->caps.qp0_proxy);
+ kfree(dev->caps.qp1_tunnel);
+ kfree(dev->caps.qp1_proxy);
+
kfree(priv);
pci_release_regions(pdev);
pci_disable_device(pdev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c
index 436ef6c..52d41b4 100644
--- a/drivers/net/ethernet/mellanox/mlx4/qp.c
+++ b/drivers/net/ethernet/mellanox/mlx4/qp.c
@@ -68,15 +68,15 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type)
}
/* used for INIT/CLOSE port logic */
-static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0)
+static int is_master_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0)
{
+ /* this procedure is called after we already know we are on the master */
/* qp0 is either the proxy qp0, or the real qp0 */
- *proxy_qp0 = qp->qpn >= dev->caps.sqp_start &&
- qp->qpn <= dev->caps.sqp_start + 1;
+ u32 pf_proxy_offset = dev->phys_caps.base_proxy_sqpn + 8 * mlx4_master_func_num(dev);
+ *proxy_qp0 = qp->qpn >= pf_proxy_offset && qp->qpn <= pf_proxy_offset + 1;
- *real_qp0 = mlx4_is_master(dev) &&
- qp->qpn >= dev->caps.base_sqpn &&
- qp->qpn <= dev->caps.base_sqpn + 1;
+ *real_qp0 = qp->qpn >= dev->phys_caps.base_sqpn &&
+ qp->qpn <= dev->phys_caps.base_sqpn + 1;
return *real_qp0 || *proxy_qp0;
}
@@ -143,7 +143,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native);
if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR &&
cur_state != MLX4_QP_STATE_RST &&
- is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
+ is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
port = (qp->qpn & 1) + 1;
if (proxy_qp0)
priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0;
@@ -175,7 +175,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
new_state == MLX4_QP_STATE_RST ? 2 : 0,
op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native);
- if (mlx4_is_master(dev) && is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
+ if (mlx4_is_master(dev) && is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
port = (qp->qpn & 1) + 1;
if (cur_state != MLX4_QP_STATE_ERR &&
cur_state != MLX4_QP_STATE_RST &&
@@ -422,6 +422,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table;
int err;
int reserved_from_top = 0;
+ int k;
spin_lock_init(&qp_table->lock);
INIT_RADIX_TREE(&dev->qp_table_tree, GFP_ATOMIC);
@@ -436,7 +437,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
* We also reserve the MSB of the 24-bit QP number to indicate
* that a QP is an XRC QP.
*/
- dev->caps.base_sqpn =
+ dev->phys_caps.base_sqpn =
ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);
{
@@ -479,24 +480,54 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
*/
err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps,
- (1 << 23) - 1, dev->caps.base_sqpn + 8 +
+ (1 << 23) - 1, dev->phys_caps.base_sqpn + 8 +
16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev),
reserved_from_top);
+ if (err)
+ return err;
- /* In mfunc, sqp_start is the base of the proxy SQPs, since the PF also
- * uses paravirtualized SQPs.
- * In native mode, sqp_start is the base of the real SQPs. */
if (mlx4_is_mfunc(dev)) {
- dev->caps.sqp_start = dev->caps.base_sqpn +
- 8 * (mlx4_master_func_num(dev) + 1);
- dev->caps.base_tunnel_sqpn = dev->caps.sqp_start + 8 * MLX4_MFUNC_MAX;
- } else
- dev->caps.sqp_start = dev->caps.base_sqpn;
+ /* for PPF use */
+ dev->phys_caps.base_proxy_sqpn = dev->phys_caps.base_sqpn + 8;
+ dev->phys_caps.base_tunnel_sqpn = dev->phys_caps.base_sqpn + 8 + 8 * MLX4_MFUNC_MAX;
+
+ /* In mfunc, calculate proxy and tunnel qp offsets for the PF here,
+ * since the PF does not call mlx4_slave_caps */
+ dev->caps.qp0_tunnel = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp0_proxy = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp1_tunnel = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+ dev->caps.qp1_proxy = kzalloc((sizeof(u32)) * dev->caps.num_ports, GFP_KERNEL);
+
+ if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
+ !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
+ err = -ENOMEM;
+ goto err_mem;
+ }
+ for (k = 0; k < dev->caps.num_ports; k++) {
+ dev->caps.qp0_proxy[k] = dev->phys_caps.base_proxy_sqpn +
+ 8 * mlx4_master_func_num(dev) + k;
+ dev->caps.qp0_tunnel[k] = dev->caps.qp0_proxy[k] + 8 * MLX4_MFUNC_MAX;
+ dev->caps.qp1_proxy[k] = dev->phys_caps.base_proxy_sqpn +
+ 8 * mlx4_master_func_num(dev) + MLX4_MAX_PORTS + k;
+ dev->caps.qp1_tunnel[k] = dev->caps.qp1_proxy[k] + 8 * MLX4_MFUNC_MAX;
+ }
+ }
+
+
+ err = mlx4_CONF_SPECIAL_QP(dev, dev->phys_caps.base_sqpn);
if (err)
- return err;
+ goto err_mem;
+ return 0;
- return mlx4_CONF_SPECIAL_QP(dev, dev->caps.base_sqpn);
+err_mem:
+ kfree(dev->caps.qp0_tunnel);
+ kfree(dev->caps.qp0_proxy);
+ kfree(dev->caps.qp1_tunnel);
+ kfree(dev->caps.qp1_proxy);
+ dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
+ dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
+ return err;
}
void mlx4_cleanup_qp_table(struct mlx4_dev *dev)
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index a8cb149..12d1036 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -328,6 +328,9 @@ struct mlx4_phys_caps {
u32 gid_phys_table_len[MLX4_MAX_PORTS + 1];
u32 pkey_phys_table_len[MLX4_MAX_PORTS + 1];
u32 num_phys_eqs;
+ u32 base_sqpn;
+ u32 base_proxy_sqpn;
+ u32 base_tunnel_sqpn;
};
struct mlx4_caps {
@@ -358,9 +361,10 @@ struct mlx4_caps {
int max_rq_desc_sz;
int max_qp_init_rdma;
int max_qp_dest_rdma;
- int sqp_start;
- u32 base_sqpn;
- u32 base_tunnel_sqpn;
+ u32 *qp0_proxy;
+ u32 *qp1_proxy;
+ u32 *qp0_tunnel;
+ u32 *qp1_tunnel;
int num_srqs;
int max_srq_wqes;
int max_srq_sge;
@@ -722,15 +726,15 @@ static inline int mlx4_is_master(struct mlx4_dev *dev)
static inline int mlx4_is_qp_reserved(struct mlx4_dev *dev, u32 qpn)
{
- return (qpn < dev->caps.base_sqpn + 8 +
+ return (qpn < dev->phys_caps.base_sqpn + 8 +
16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev));
}
static inline int mlx4_is_guest_proxy(struct mlx4_dev *dev, int slave, u32 qpn)
{
- int base = dev->caps.sqp_start + slave * 8;
+ int guest_proxy_base = dev->phys_caps.base_proxy_sqpn + slave * 8;
- if (qpn >= base && qpn < base + 8)
+ if (qpn >= guest_proxy_base && qpn < guest_proxy_base + 8)
return 1;
return 0;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* [PATCH for-next V2 22/22] IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (20 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 21/22] {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations Jack Morgenstein
@ 2012-08-03 8:40 ` Jack Morgenstein
2012-08-03 9:00 ` [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces Tziporet Koren
2012-08-12 9:34 ` Or Gerlitz
23 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-08-03 8:40 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Jack Morgenstein
When have VFs and PFs on same host, the VFs are activated within the
mlx4_core module before the mlx4_ib kernel module is loaded.
When the mlx4_ib module initializes the PF (master), it now creates
mad paravirtualization contexts for any VFs already active.
Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
---
drivers/infiniband/hw/mlx4/mad.c | 3 +++
drivers/infiniband/hw/mlx4/main.c | 11 +++++++++++
2 files changed, 14 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 31b0559..090a2ad 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1648,6 +1648,9 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
{
int ret, cq_size;
+ if (ctx->state != DEMUX_PV_STATE_DOWN)
+ return -EEXIST;
+
ctx->state = DEMUX_PV_STATE_STARTING;
/* have QP0 only on port owner, and only if link layer is IB */
if (ctx->slave == mlx4_master_func_num(to_mdev(ctx->ib_dev)->dev) &&
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 00384b3..82b7231 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -74,6 +74,8 @@ struct update_gid_work {
int port;
};
+static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
+
static struct workqueue_struct *wq;
static void init_query_mad(struct ib_smp *mad)
@@ -1469,6 +1471,15 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
if (mlx4_is_mfunc(ibdev->dev))
init_pkeys(ibdev);
+ /* create paravirt contexts for any VFs which are active */
+ if (mlx4_is_master(ibdev->dev)) {
+ for (j = 0; j < MLX4_MFUNC_MAX; j++) {
+ if (j == mlx4_master_func_num(ibdev->dev))
+ continue;
+ if (mlx4_is_slave_active(ibdev->dev, j))
+ do_slave_init(ibdev, j, 1);
+ }
+ }
return ibdev;
err_notif:
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 48+ messages in thread
* RE: [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (21 preceding siblings ...)
2012-08-03 8:40 ` [PATCH for-next V2 22/22] IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes Jack Morgenstein
@ 2012-08-03 9:00 ` Tziporet Koren
2012-08-12 9:34 ` Or Gerlitz
23 siblings, 0 replies; 48+ messages in thread
From: Tziporet Koren @ 2012-08-03 9:00 UTC (permalink / raw)
To: Jack Morgenstein, roland-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Or Gerlitz,
Dotan Barak, Yevgeny Petrilin, Liran Liss
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 15787 bytes --]
many thanks Jack
Tziporet
-----Original Message-----
From: linux-rdma-owner@vger.kernel.org [mailto:linux-rdma-owner@vger.kernel.org] On Behalf Of Jack Morgenstein
Sent: Friday, August 03, 2012 11:41 AM
To: roland@kernel.org
Cc: linux-rdma@vger.kernel.org; Or Gerlitz; Jack Morgenstein; Dotan Barak; Tziporet Koren; Yevgeny Petrilin; Liran Liss
Subject: [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces
This patch set adds SRIOV support for IB interfaces.
Patches 1-4 are "precondition" patches.
Patches 5-22 actually implement the feature.
NOTE: Patch 18 depends on patch IB/mlx4: fix possible deadlock with sm_lock spinlock
(a separately-submitted patch) being applied previously to the for-next git.
This patch set introduces Infiniband SRIOV support for ConnectX2 and ConnectX3 devices. Each function presents itself as an independent vHCA (virtual HCA) to the host while a single HCA is observable by the network, which is unaware of the vHCAs. No changes are required by the IB subsystem, ULPs, and apps to support SRIOV, and vHCAs are interoperable with any existing (non-virtualized) IB deployments.
We term this model for SRIOV implementation the shared-port model.
Sharing the same physical port(s) among multiple vHCAs is achieved as follows:
1. Each vHCA port presents its own virtual GID table.
Currently, the virtual GID table comprises a single entry (at index 0) that maps to a unique index in the physical GID table. The vHCA of the PF maps to physical GID index 0. To obtain GIDs for other vHCAs, alias GUIDs are requested from the SM. These are GUIDs which the SM places, per port, in the port's guid table after the 0'th slot (which is read-only and determined by the FW).
The host admin can assign GIDs to vHCAs using a sysfs interface (see below).
2. Each vHCA port presents its own virtual PKey table.
The virtual PKey table is a mapping of selected indexes of the physical pkey table.
The host admin can control which pkey indexes are mapped to which virtual indexes using a sysfs interface (see below). Note that the physical PKey table may contain both full and partial memberships of the same PKey to allow different membership types in different virtual tables.
3. Each vHCA port has its own virtual port state.
A vHCA port is up if the following conditions apply:
- The physical port is up
- The virtual GID table contains the GIDs requested by the host admin
- The SM has acknowledged the requested GIDs since the last time that
the physical port came up
4. Other port attributes are shared, e.g., GID prefix, LID, SM LID, LMC mask.
5. Special QPs are para-virtualized.
vHCAs are not given direct access to QP0/1. Rather, these QPs are operated by a special context hosted by the PF, which mediates access to/from vHCAs.
This is done by opening a “tunnel†per vHCA port per QP0/1. A tunnel comprises a pair of UD QPs: a “Tunnel QP†in the PF-context and a “Proxy QP†in the vHCA.
All vHCA MAD traffic must pass through the corresponding tunnel.
vHCA QPs cannot be assigned to VL15 and are denied of the well-known QKey.
QP0 access is restricted to the PF vHCA. VF vHCAs also have (virtual) QP0’s, but they never receive any SMPs and all SMPs sent are discarded.
QP1 traffic is allowed for all vHCAs, but special care is required to bridge the gap between the host and network views.
Specifically:
- Transaction IDs are mapped to guarantee uniqueness among vHCAs
- CM para-virtualization
o Incoming requests are steered to the correct vHCA according to the embedded GID
o Local communication IDs are mapped to ensure uniqueness among vHCAs
- Multicast para-virtualization
o The PF context aggregates membership state from all vHCAs
o The SA is contacted only when the aggregate membership changes
o If the aggregate does not change, the PF context will provide the
requesting vHCA with the proper response
Incoming MADs are steered according to:
- the DGID If a GRH is present
- the mapped transaction ID for response MADs
- the embedded GID in CM requests
- the remote communication ID in other CM messages
To allow the host admin to control the virtual GID and PKey tables of vHCAs, a new sysfs ‘iov’ sub-tree has been added under the PF infiniband device.
Details on this mechanism can be found in the change log of:
IB/mlx4: Add iov directory in sysfs under the ib device
Some Limitations
----------------
1. FMRs are not currently supported on slaves. This will be corrected in a
future submission.
2. RoCE is not currently supported on slaves. This will be corrected in a
future submission.
Changes for V2
--------------
1. Patches already accepted by Roland in the V1 set are removed from V2.
The following (precondition) patches submitted in V1 were accepted:
1: net/mlx4_core: Pass an invalid PCI id number to VFs
3: IB/mlx4: Add debug printouts
6: IB/sa: Add GuidInfoRecord query support
7: IB/core: move macros from cm_msgs.h to ib_cm.h
8: {NET,IB}/mlx4: Use port management change event instead of smp_snoop
9: net/mlx4_core: For SRIOV, initialize ib port-capabilities for all slaves
10: net/mlx4_core: Implement mechanism for reserved qkeys
11: net/mlx4_core: Allow guests to support IB ports
12: {NET,IB}/mlx4_core: place phys gid and pkey tbl sizes in mlx4_phys_caps
struct and paravirtualize them
2. patches (new numbering) "5: net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area"
and "8: IB/mlx4: SRIOV multiplex and demultiplex MADs"
were adjusted to account for changes introduced by commits to mlx4_core from other gits in the
kernel V3.5 merge
3. A fix to event processing in the V2 patch set was added to patch
"18: IB/mlx4: Miscellaneous adjustments to SRIOV IB support" :
Add spinlock to slave_event(), since it is called both in interrupt
context and in process context (due to patch 6, and also if smp_snoop
is used).
This fix was found and implemented by Saeed Mahameed <saeedm@mellanox.com>
4. Two new patches were added to V2:
21: {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations
22: IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes
5. The workaround for the KVM IRQ management problem is no longer needed. The KVM group
fixed the problem in kernel 3.5-rc7.
Changes for V1
--------------
1. librdmacm now supports multiple VF/PF on the same host (patch 29).
2. Several patches cleaned up (these were indicated in the V0 changelogs).
Major cleanups in patch 22 and patch 24.
3. Eliminated code duplication in Port Management Change event code (patch 8).
4. Now use pr_debug, instead of mlx4_ib_debug, and there is no module parameter
(Roland's recommendation).
5. mlx4_master_func_num() to get the master's "slave_id", to make code more readable.
6. Fixed illegal use of port num field for a force-loopback bit in ib_ah structure
(V0 patch 2 -- eliminated). The force-loopback bit is now set for Tunnel QPs in
mlx4_ib_post send (patch 17).
7. New patch 2 (not related to 6 above) to reserve bits in enum ib_qp_create_flags.
8. V0 patch 26 is now rolled into V1 patch 22. This allowed us to eliminate function
mlx4_ib_indexed_gid from patch 22 (replaced by using __mlx4_ib_query_gid() from V0 patch 26).
Amir Vadai (1):
IB/mlx4: Add CM paravirtualization
Erez Shitrit (1):
IB/sa: Add GuidInfoRecord query support.
Jack Morgenstein (26):
net/mlx4_core: Pass an invalid PCI id number to VFs
IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver
use
IB/mlx4: Add debug printouts
IB/core: change pkey table lookups to support full and partial
membership for the same pkey
IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey
match
IB/core: move macros from cm_msgs.h to ib_cm.h
{NET,IB}/mlx4: Use port management change event instead of smp_snoop
net/mlx4_core: For SRIOV, initialize ib port-capabilities for all
slaves
net/mlx4_core: Implement mechanism for reserved qkeys
net/mlx4_core: Allow guests to support IB ports
{NET,IB}/mlx4_core: place phys gid and pkey tbl sizes in
mlx4_phys_caps struct and paravirtualize them
IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area
IB/mlx4: Initialize SRIOV IB support for slaves in master context
{NET,IB}/mlx4: Implement QP paravirtualization and maintain
phys_pkey_cache for smp_snoop
IB/mlx4: SRIOV multiplex and demultiplex MADs
{NET,IB}/mlx4: MAD_IFC paravirtualization
net/mlx4_core: Add IB port-state machine, and port mgmt event
propagation infrastructure
{NET,IB}/mlx4: Add alias_guid mechanism
IB/mlx4: Propagate pkey and guid change port management events to
slaves
IB/mlx4: Add iov directory in sysfs under the ib device
net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB
net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode
IB/mlx4: Miscellaneous adjustments to SRIOV IB support
{NET,IB}/mlx4: Activate SRIOV mode for IB
{NET,IB}/mlx4: Paravirtualize Node Guids for slaves.
Oren Duer (1):
IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV
drivers/infiniband/core/cache.c | 42 +-
drivers/infiniband/core/cm_msgs.h | 12 -
drivers/infiniband/core/device.c | 17 +-
drivers/infiniband/core/sa_query.c | 133 ++
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 688 ++++++++
drivers/infiniband/hw/mlx4/cm.c | 437 +++++
drivers/infiniband/hw/mlx4/cq.c | 31 +-
drivers/infiniband/hw/mlx4/mad.c | 1684 +++++++++++++++++++-
drivers/infiniband/hw/mlx4/main.c | 285 +++-
drivers/infiniband/hw/mlx4/mcg.c | 1254 +++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 360 +++++-
drivers/infiniband/hw/mlx4/qp.c | 651 +++++++-
drivers/infiniband/hw/mlx4/sysfs.c | 794 +++++++++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 190 +++-
drivers/net/ethernet/mellanox/mlx4/en_main.c | 5 +-
drivers/net/ethernet/mellanox/mlx4/eq.c | 259 +++-
drivers/net/ethernet/mellanox/mlx4/fw.c | 227 +++-
drivers/net/ethernet/mellanox/mlx4/fw.h | 3 +
drivers/net/ethernet/mellanox/mlx4/intf.c | 5 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 124 ++-
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 116 +-
drivers/net/ethernet/mellanox/mlx4/port.c | 21 +-
drivers/net/ethernet/mellanox/mlx4/qp.c | 67 +-
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 220 +++-
include/linux/mlx4/device.h | 179 ++-
include/linux/mlx4/driver.h | 5 +-
include/linux/mlx4/qp.h | 3 +-
include/rdma/ib_cache.h | 16 +
include/rdma/ib_cm.h | 12 +
include/rdma/ib_sa.h | 33 +
include/rdma/ib_verbs.h | 3 +
32 files changed, 7520 insertions(+), 358 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c
create mode 100644 drivers/infiniband/hw/mlx4/cm.c create mode 100644 drivers/infiniband/hw/mlx4/mcg.c create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c
Cc: dotanb@mellanox.com
Cc: tziporet@mellanox.com
Cc: yevgenyp@mellanox.com
Cc: liranl@mellanox.com
Amir Vadai (1):
IB/mlx4: Add CM paravirtualization
Jack Morgenstein (20):
IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver
use
IB/core: change pkey table lookups to support full and partial
membership for the same pkey
IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey
match
IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area
IB/mlx4: Initialize SRIOV IB support for slaves in master context
{NET,IB}/mlx4: Implement QP paravirtualization and maintain
phys_pkey_cache for smp_snoop
IB/mlx4: SRIOV multiplex and demultiplex MADs
{NET,IB}/mlx4: MAD_IFC paravirtualization
net/mlx4_core: Add IB port-state machine, and port mgmt event
propagation infrastructure
{NET,IB}/mlx4: Add alias_guid mechanism
IB/mlx4: Propagate pkey and guid change port management events to
slaves
IB/mlx4: Add iov directory in sysfs under the ib device
net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB
net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode
IB/mlx4: Miscellaneous adjustments to SRIOV IB support
{NET,IB}/mlx4: Activate SRIOV mode for IB
{NET,IB}/mlx4: Paravirtualize Node Guids for slaves
{NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no
calculations
IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver
initializes
Oren Duer (1):
IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV
drivers/infiniband/core/cache.c | 42 +-
drivers/infiniband/core/device.c | 17 +-
drivers/infiniband/hw/mlx4/Makefile | 2 +-
drivers/infiniband/hw/mlx4/alias_GUID.c | 688 +++++++++
drivers/infiniband/hw/mlx4/cm.c | 437 ++++++
drivers/infiniband/hw/mlx4/cq.c | 31 +-
drivers/infiniband/hw/mlx4/mad.c | 1573 +++++++++++++++++++-
drivers/infiniband/hw/mlx4/main.c | 273 +++-
drivers/infiniband/hw/mlx4/mcg.c | 1254 ++++++++++++++++
drivers/infiniband/hw/mlx4/mlx4_ib.h | 341 +++++-
drivers/infiniband/hw/mlx4/qp.c | 656 +++++++-
drivers/infiniband/hw/mlx4/sysfs.c | 794 ++++++++++
drivers/net/ethernet/mellanox/mlx4/cmd.c | 191 +++-
drivers/net/ethernet/mellanox/mlx4/eq.c | 245 +++-
drivers/net/ethernet/mellanox/mlx4/fw.c | 244 ++-
drivers/net/ethernet/mellanox/mlx4/fw.h | 11 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 91 +-
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 50 +
drivers/net/ethernet/mellanox/mlx4/port.c | 10 +
drivers/net/ethernet/mellanox/mlx4/qp.c | 100 ++-
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 220 +++-
include/linux/mlx4/device.h | 69 +-
include/linux/mlx4/driver.h | 2 +
include/linux/mlx4/qp.h | 3 +-
include/rdma/ib_cache.h | 16 +
include/rdma/ib_verbs.h | 3 +
26 files changed, 7080 insertions(+), 283 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c
create mode 100644 drivers/infiniband/hw/mlx4/cm.c create mode 100644 drivers/infiniband/hw/mlx4/mcg.c create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
N‹§²æìr¸›yúèšØb²X¬¶Ç§vØ^–)Þº{.nÇ+‰·¥Š{±ÙšŠ{ayº\x1dʇڙë,j\a¢f£¢·hš‹»öì\x17/oSc¾™Ú³9˜uÀ¦æå‰È&jw¨®\x03(階ŠÝ¢j"ú\x1a¶^[m§ÿïêäz¹Þ–Šàþf£¢·hšˆ§~ˆmš
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
` (22 preceding siblings ...)
2012-08-03 9:00 ` [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces Tziporet Koren
@ 2012-08-12 9:34 ` Or Gerlitz
23 siblings, 0 replies; 48+ messages in thread
From: Or Gerlitz @ 2012-08-12 9:34 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: Jack Morgenstein, linux-rdma-u79uwXL29TY76Z2rM5mHXA
On 03/08/2012 11:40, Jack Morgenstein wrote:
> This patch set adds SRIOV support for IB interfaces.
>
> Patches 1-4 are "precondition" patches.
> Patches 5-22 actually implement the feature.
>
> NOTE: Patch 18 depends on patch IB/mlx4: fix possible deadlock with sm_lock spinlock
> (a separately-submitted patch) being applied previously to the for-next git.
Hi Roland,
Any feedback on the patches? we want to see things are converging
towards the 3.7 merge
window, such that people who want IB on their SRIOV systems have it. So
if taking into account the volume of changes there's not too much
time... can the review be continued?
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use
[not found] ` <1343983258-6268-2-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-05 14:55 ` Doug Ledford
[not found] ` <504767EB.6090004-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-24 19:34 ` Roland Dreier
1 sibling, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-05 14:55 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
[-- Attachment #1: Type: text/plain, Size: 1435 bytes --]
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> Reserve bits 26-31 for internal use by low-level drivers. Two
> such bits are used in the mlx4 driver SRIOV IB implementation.
>
> These enum additions guarantee that the core layer will never use
> these bits, so that low level drivers may safely make use of them.
>
> Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
> ---
> include/rdma/ib_verbs.h | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
> index 07996af..46bc045 100644
> --- a/include/rdma/ib_verbs.h
> +++ b/include/rdma/ib_verbs.h
> @@ -614,6 +614,9 @@ enum ib_qp_type {
> enum ib_qp_create_flags {
> IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
> IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
> + /* reserve bits 26-31 for low level drivers' internal use */
> + IB_QP_CREATE_RESERVED_START = 1 << 26,
> + IB_QP_CREATE_RESERVED_END = 1 << 31,
> };
>
> struct ib_qp_init_attr {
>
Reserving 6 bits for driver use out of 32 seems reasonable.
Acked-by: Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use
[not found] ` <504767EB.6090004-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-06 10:06 ` Jack Morgenstein
0 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-09-06 10:06 UTC (permalink / raw)
To: Doug Ledford
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
Thanks, Doug!
-Jack
On Wednesday 05 September 2012 17:55, Doug Ledford wrote:
> On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> > Reserve bits 26-31 for internal use by low-level drivers. Two
> > such bits are used in the mlx4 driver SRIOV IB implementation.
> >
> > These enum additions guarantee that the core layer will never use
> > these bits, so that low level drivers may safely make use of them.
> >
> > Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
> > ---
> > include/rdma/ib_verbs.h | 3 +++
> > 1 files changed, 3 insertions(+), 0 deletions(-)
> >
> > diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
> > index 07996af..46bc045 100644
> > --- a/include/rdma/ib_verbs.h
> > +++ b/include/rdma/ib_verbs.h
> > @@ -614,6 +614,9 @@ enum ib_qp_type {
> > enum ib_qp_create_flags {
> > IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
> > IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
> > + /* reserve bits 26-31 for low level drivers' internal use */
> > + IB_QP_CREATE_RESERVED_START = 1 << 26,
> > + IB_QP_CREATE_RESERVED_END = 1 << 31,
> > };
> >
> > struct ib_qp_init_attr {
> >
>
> Reserving 6 bits for driver use out of 32 seems reasonable.
>
> Acked-by: Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <1343983258-6268-3-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-11 16:52 ` Doug Ledford
[not found] ` <504F6C4F.6050207-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 16:52 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Liran Liss
[-- Attachment #1: Type: text/plain, Size: 4298 bytes --]
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> Enhance the cached and non-cached pkey table lookups to enable limited and full
> members of the same pkey to co-exist in the pkey table.
>
> This is necessary for SRIOV to allow for a scheme where some guests would have the full
> membership pkey in their virtual pkey table, where other guests on the same hypervisor
> would have the limited one. In that sense, its an extension of the IBTA model for
> non virtualized nodes.
OK, maybe I'm not getting something, but I'm curious why we always pick
the full pkey in preference to the partial pkey. Shouldn't we pick the
pkey that's appropriate for the vHCA sending the message?
Also, given the rule of least surprise, don't you think it would be best
to rename this function ib_find_cached_full_or_parital_pkey and in your
next patch instead of naming it ib_find_exact_pkey just call that one
ib_find_cached_pkey?
> To accomplish this, we need both the limited and full membership pkeys to be present
> in the master's (hypervisor physical port) pkey table.
>
> The algorithm for supporting pkey tables which contain both the limited and the full
> membership versions of the same pkey works as follows:
>
> When scanning the pkey table for a 15 bit pkey:
>
> A. If there is a full member version of that pkey anywhere
> in the table, return its index (even if a limited-member
> version of the pkey exists earlier in the table).
>
> B. If the full member version is not in the table,
> but the limited-member version is in the table,
> return the index of the limited pkey.
>
> Signed-off-by: Liran Liss <liranl-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
> Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> ---
> drivers/infiniband/core/cache.c | 14 +++++++++++---
> drivers/infiniband/core/device.c | 17 +++++++++++++----
> 2 files changed, 24 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c
> index 9353992..0f2f2b7 100644
> --- a/drivers/infiniband/core/cache.c
> +++ b/drivers/infiniband/core/cache.c
> @@ -167,6 +167,7 @@ int ib_find_cached_pkey(struct ib_device *device,
> unsigned long flags;
> int i;
> int ret = -ENOENT;
> + int partial_ix = -1;
>
> if (port_num < start_port(device) || port_num > end_port(device))
> return -EINVAL;
> @@ -179,10 +180,17 @@ int ib_find_cached_pkey(struct ib_device *device,
>
> for (i = 0; i < cache->table_len; ++i)
> if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) {
> - *index = i;
> - ret = 0;
> - break;
> + if (cache->table[i] & 0x8000) {
> + *index = i;
> + ret = 0;
> + break;
> + } else
> + partial_ix = i;
> }
> + if (ret && partial_ix >= 0) {
> + *index = partial_ix;
> + ret = 0;
> + }
>
> read_unlock_irqrestore(&device->cache.lock, flags);
>
> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
> index e711de4..a645c68 100644
> --- a/drivers/infiniband/core/device.c
> +++ b/drivers/infiniband/core/device.c
> @@ -707,18 +707,27 @@ int ib_find_pkey(struct ib_device *device,
> {
> int ret, i;
> u16 tmp_pkey;
> + int partial_ix = -1;
>
> for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
> ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
> if (ret)
> return ret;
> -
> if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
> - *index = i;
> - return 0;
> + /* if there is full-member pkey take it.*/
> + if (tmp_pkey & 0x8000) {
> + *index = i;
> + return 0;
> + }
> + if (partial_ix < 0)
> + partial_ix = i;
> }
> }
> -
> + /*no full-member, if exists take the limited*/
> + if (partial_ix >= 0) {
> + *index = partial_ix;
> + return 0;
> + }
> return -ENOENT;
> }
> EXPORT_SYMBOL(ib_find_pkey);
>
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <1343983258-6268-4-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-11 16:53 ` Doug Ledford
2012-09-11 17:12 ` Doug Ledford
1 sibling, 0 replies; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 16:53 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
[-- Attachment #1: Type: text/plain, Size: 3840 bytes --]
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> When port pkey table potentially contains both full and partial
> membership copies for the same pkey, we need a function to find
> the exact (16-bit) pkey index.
The code on this patch is fine, just see my previous email about the
function naming...
> This is particularly necessary when the master forwards QP1 MADS
> sent by guests. If the guest has sent the MAD with a limited
> membership pkey, we wish to forward the MAD using the same limited
> membership pkey. Since master may have both the limited and
> the full member pkeys in its table, we must make sure to retrieve
> the limited membership pkey in this case.
>
> This requires the 16-bit pkey lookup function (which includes the
> membership bit).
>
> Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
> Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> ---
> drivers/infiniband/core/cache.c | 32 ++++++++++++++++++++++++++++++++
> include/rdma/ib_cache.h | 16 ++++++++++++++++
> 2 files changed, 48 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c
> index 0f2f2b7..d8a8c83 100644
> --- a/drivers/infiniband/core/cache.c
> +++ b/drivers/infiniband/core/cache.c
> @@ -198,6 +198,38 @@ int ib_find_cached_pkey(struct ib_device *device,
> }
> EXPORT_SYMBOL(ib_find_cached_pkey);
>
> +int ib_find_exact_cached_pkey(struct ib_device *device,
> + u8 port_num,
> + u16 pkey,
> + u16 *index)
> +{
> + struct ib_pkey_cache *cache;
> + unsigned long flags;
> + int i;
> + int ret = -ENOENT;
> +
> + if (port_num < start_port(device) || port_num > end_port(device))
> + return -EINVAL;
> +
> + read_lock_irqsave(&device->cache.lock, flags);
> +
> + cache = device->cache.pkey_cache[port_num - start_port(device)];
> +
> + *index = -1;
> +
> + for (i = 0; i < cache->table_len; ++i)
> + if (cache->table[i] == pkey) {
> + *index = i;
> + ret = 0;
> + break;
> + }
> +
> + read_unlock_irqrestore(&device->cache.lock, flags);
> +
> + return ret;
> +}
> +EXPORT_SYMBOL(ib_find_exact_cached_pkey);
> +
> int ib_get_cached_lmc(struct ib_device *device,
> u8 port_num,
> u8 *lmc)
> diff --git a/include/rdma/ib_cache.h b/include/rdma/ib_cache.h
> index 00a2b8e..ad9a3c2 100644
> --- a/include/rdma/ib_cache.h
> +++ b/include/rdma/ib_cache.h
> @@ -101,6 +101,22 @@ int ib_find_cached_pkey(struct ib_device *device,
> u16 *index);
>
> /**
> + * ib_find_exact_cached_pkey - Returns the PKey table index where a specified
> + * PKey value occurs. Comparison uses the FULL 16 bits (incl membership bit)
> + * @device: The device to query.
> + * @port_num: The port number of the device to search for the PKey.
> + * @pkey: The PKey value to search for.
> + * @index: The index into the cached PKey table where the PKey was found.
> + *
> + * ib_find_exact_cached_pkey() searches the specified PKey table in
> + * the local software cache.
> + */
> +int ib_find_exact_cached_pkey(struct ib_device *device,
> + u8 port_num,
> + u16 pkey,
> + u16 *index);
> +
> +/**
> * ib_get_cached_lmc - Returns a cached lmc table entry
> * @device: The device to query.
> * @port_num: The port number of the device to query.
>
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
[not found] ` <1343983258-6268-5-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-11 17:10 ` Doug Ledford
[not found] ` <504F7068.6020606-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 17:10 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
[-- Attachment #1: Type: text/plain, Size: 1926 bytes --]
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> 1. Introduce the basic sriov parvirtualization context objects
> for multiplexing and demultiplexing MADs.
> 2. Introduce support for the new proxy and tunnel QP types.
>
> This patch introduces the objects required by the master
> for managing QP paravirtualization for guests.
>
> struct mlx4_ib_sriov{} is created by the master only.
> It is a container for the following:
> 1. All the info required by the PPF to multiplex and de-multiplex MADs
> (including those from the PF). (struct mlx4_ib_demux_ctx demux)
OK, so can we have at least a single reference to the various
abbreviations before using them exclusively? I know PF and PPF may be
common, but it might be nice that they were used once in full form
before abbreviated in commit messages.
> 2. All the info required to manage alias GUIDs (i.e., the GUID at
> index 0 that each guest perceives. In fact, this is not the
> GUID which is actually at index 0, but is, in fact, the GUID
> which is at index[<VF number>] in the physical table.
OK, this has been one of the things that has made reviewing this
difficult. I freely admit that I've steadfastly ignored SRIOV for as
long as I can, so maybe this is just me. But, in the context of this
driver, how am I supposed to know which code paths will be on the host
and which on the guest?
Also, I note that you do math every time you want to know if you are on
a parent device or a virtual device. Do you really want to do math all
the time, or would it be better to save off your status on device init
and just refer to that when you would do math in this patch?
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <1343983258-6268-4-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-11 16:53 ` Doug Ledford
@ 2012-09-11 17:12 ` Doug Ledford
[not found] ` <504F70FB.6030806-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
1 sibling, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 17:12 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Marciniszyn, Mike,
Mascarenhas, Edward
[-- Attachment #1: Type: text/plain, Size: 1234 bytes --]
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> When port pkey table potentially contains both full and partial
> membership copies for the same pkey, we need a function to find
> the exact (16-bit) pkey index.
>
> This is particularly necessary when the master forwards QP1 MADS
> sent by guests. If the guest has sent the MAD with a limited
> membership pkey, we wish to forward the MAD using the same limited
> membership pkey. Since master may have both the limited and
> the full member pkeys in its table, we must make sure to retrieve
> the limited membership pkey in this case.
>
> This requires the 16-bit pkey lookup function (which includes the
> membership bit).
As a second note, I would like to know why Intel (previously QLogic)
does not use these functions in their driver and what it would take to
get all drivers to use the functions. Do we need to add more to them?
In my opinion these should be generally useful and used by all drivers.
Mike?
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <504F70FB.6030806-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-11 19:07 ` Roland Dreier
[not found] ` <CAL1RGDW6D25s+R8HuxK-DMrsS_wvKDTL+LSp2X-TK8gB8Vm2Fg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Roland Dreier @ 2012-09-11 19:07 UTC (permalink / raw)
To: Doug Ledford
Cc: Jack Morgenstein, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Marciniszyn, Mike,
Mascarenhas, Edward
On Tue, Sep 11, 2012 at 10:12 AM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> As a second note, I would like to know why Intel (previously QLogic)
> does not use these functions in their driver and what it would take to
> get all drivers to use the functions. Do we need to add more to them?
> In my opinion these should be generally useful and used by all drivers.
Use which functions? The P_Key lookup functions?
What would a low-level driver use them for? I thought these are for
use by upper-level protocols.
- R.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <CAL1RGDW6D25s+R8HuxK-DMrsS_wvKDTL+LSp2X-TK8gB8Vm2Fg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2012-09-11 20:34 ` Doug Ledford
[not found] ` <504FA064.6040002-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 20:34 UTC (permalink / raw)
To: Roland Dreier
Cc: Jack Morgenstein, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Marciniszyn, Mike,
Mascarenhas, Edward
[-- Attachment #1: Type: text/plain, Size: 1086 bytes --]
On 9/11/2012 3:07 PM, Roland Dreier wrote:
> On Tue, Sep 11, 2012 at 10:12 AM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
>> As a second note, I would like to know why Intel (previously QLogic)
>> does not use these functions in their driver and what it would take to
>> get all drivers to use the functions. Do we need to add more to them?
>> In my opinion these should be generally useful and used by all drivers.
>
> Use which functions? The P_Key lookup functions?
>
> What would a low-level driver use them for? I thought these are for
> use by upper-level protocols.
Well, at this point, the mlx4 driver uses them, the rdmacm kernel driver
uses them, and both QLogic/Intel drivers have their own internal pkey
table implementation. So, it isn't so much upper layer as it is drivers.
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <504FA064.6040002-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-11 20:43 ` Roland Dreier
[not found] ` <CAL1RGDXpeK3KjrmzyivQs8FOs2dg5MqSmuVsdRE+bVD5OXe6BA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Roland Dreier @ 2012-09-11 20:43 UTC (permalink / raw)
To: Doug Ledford
Cc: Jack Morgenstein, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Marciniszyn, Mike,
Mascarenhas, Edward
On Tue, Sep 11, 2012 at 1:34 PM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> Well, at this point, the mlx4 driver uses them, the rdmacm kernel driver
> uses them, and both QLogic/Intel drivers have their own internal pkey
> table implementation. So, it isn't so much upper layer as it is drivers.
rdmacm is an upper-level protocol (it's above the midlayer / hardware
abstraction).
mlx4 and mthca look up P_Keys because of internal details of how they
send MADs, and really they should move to maintaining their own P_Key
table too.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match
[not found] ` <CAL1RGDXpeK3KjrmzyivQs8FOs2dg5MqSmuVsdRE+bVD5OXe6BA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2012-09-11 22:51 ` Doug Ledford
0 siblings, 0 replies; 48+ messages in thread
From: Doug Ledford @ 2012-09-11 22:51 UTC (permalink / raw)
To: Roland Dreier
Cc: Jack Morgenstein, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Marciniszyn, Mike,
Mascarenhas, Edward
[-- Attachment #1: Type: text/plain, Size: 1111 bytes --]
On 9/11/2012 4:43 PM, Roland Dreier wrote:
> On Tue, Sep 11, 2012 at 1:34 PM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
>> Well, at this point, the mlx4 driver uses them, the rdmacm kernel driver
>> uses them, and both QLogic/Intel drivers have their own internal pkey
>> table implementation. So, it isn't so much upper layer as it is drivers.
>
> rdmacm is an upper-level protocol (it's above the midlayer / hardware
> abstraction).
Yeah, I know. My point wasn't that it was a low level item, just that
it's the only upper layer consumer that I saw.
> mlx4 and mthca look up P_Keys because of internal details of how they
> send MADs, and really they should move to maintaining their own P_Key
> table too.
Why not make the routines useful for all users instead of multiple
implementations of the same thing?
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
Infiniband specific RPMs available at
http://people.redhat.com/dledford/Infiniband
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 895 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <504F6C4F.6050207-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-12 7:56 ` Jack Morgenstein
[not found] ` <201209121056.00309.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-13 15:53 ` Or Gerlitz
1 sibling, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-09-12 7:56 UTC (permalink / raw)
To: Doug Ledford
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Liran Liss
On Tuesday 11 September 2012 19:52, Doug Ledford wrote:
> On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> > Enhance the cached and non-cached pkey table lookups to enable limited and full
> > members of the same pkey to co-exist in the pkey table.
> >
> > This is necessary for SRIOV to allow for a scheme where some guests would have the full
> > membership pkey in their virtual pkey table, where other guests on the same hypervisor
> > would have the limited one. In that sense, its an extension of the IBTA model for
> > non virtualized nodes.
>
> OK, maybe I'm not getting something, but I'm curious why we always pick
> the full pkey in preference to the partial pkey.
>
The pkey model that we are working with is the following:
- guests do NOT have both limited and full membership forms of a given pkey;
they have either the limited member or the full member
- The guests will see a virtualized pkey table, where pkeys appearing in that table
are "cherry-picked" by the administrator from the underlying physical (device) pkey
table contents. The slot (or index) occupied by a mapped pkey in the virtual table
is also determined by the administrator.
The Hypervisor (primary phys function or PPF), which sees the physical pkey table used
for the mappings, will therefor have both forms of a given pkey -- with the limited form
mapped to some slot in a guest's pkey table, and the full form mapped to some slot in
a different guest's pkey table.
- The only reason that a pkey table would contain both full and limited membership versions
of a given pkey is to support SRIOV guest pkey paravirtualization. In "native" mode, there
is no reason to have both forms in a given pkey table. (The driver clearly reflects this
in its original implementation of "ib_find_cached_pkey" -- it returns the index of the first
15-bit pkey it finds in the table, and assumes that these 15 bits are unique in the table).
Thus, on guests ib_find_cached_pkey will operate as before, since the 15-bit pkey in the guest
table will be unique.
On the Hypervisor, however, we assume that if both versions of the pkey are in its pkey table,
then for its own infiniband operation (as opposed to performing its pkey virtualizing function),
it should operate with the highest membership type in its table for a given 15-bit pkey.
> Shouldn't we pick the
> pkey that's appropriate for the vHCA sending the message?
We do. When QPs on the guests are created, the modify-qp commands are not executed on the guest,
but rather are passed to the PPF for processing. The PPF replaces the guest-provided virtual pkey-index
value with the appropriate physical pkey-index value. See procedure "update_pkey_index" in file
resource_tracker.c, and all the places it is called (i.e., in the wrapper functions for the various
modify-qp firmware commands).
> Also, given the rule of least surprise, don't you think it would be best
> to rename this function ib_find_cached_full_or_parital_pkey and in your
> next patch instead of naming it ib_find_exact_pkey just call that one
> ib_find_cached_pkey?
The naming of "ib_find_cached_pkey" refers to the 15-bit pkey value. I also
did not want to change all the places that "ib_find_cached_pkey" is used, to
call "ib_find_cached_full_or_partial_pkey" instead.
I also think it is more confusing to use the old name "ib_find_cached_pkey" in
a completely new (and incompatible) way.
Maybe I can change "ib_find_exact_pkey" to "ib_find_16_bit_pkey" ?
In this case, "ib_find_cached_pkey" will be backwards compatible and will find the
appropriate 15-bit pkey of either form (with the proviso that if both forms of the
pkey are in the table, it will return the full membership pkey index).
The ib_find_16_bit_pkey() looks for the exact match, and returns error if it does not find
it (-- specifically, if it is looking for one membership form, and only the other form is in
the table, it will return error).
> > To accomplish this, we need both the limited and full membership pkeys to be present
> > in the master's (hypervisor physical port) pkey table.
> >
> > The algorithm for supporting pkey tables which contain both the limited and the full
> > membership versions of the same pkey works as follows:
> >
> > When scanning the pkey table for a 15 bit pkey:
> >
> > A. If there is a full member version of that pkey anywhere
> > in the table, return its index (even if a limited-member
> > version of the pkey exists earlier in the table).
> >
> > B. If the full member version is not in the table,
> > but the limited-member version is in the table,
> > return the index of the limited pkey.
> >
> > Signed-off-by: Liran Liss <liranl-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> > Signed-off-by: Jack Morgenstein <jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
> > Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <201209121056.00309.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-12 16:48 ` Doug Ledford
[not found] ` <5050BCDD.50106-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Doug Ledford @ 2012-09-12 16:48 UTC (permalink / raw)
To: Jack Morgenstein
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Liran Liss
[-- Attachment #1: Type: text/plain, Size: 1840 bytes --]
On 09/12/2012 03:56 AM, Jack Morgenstein wrote:
> On the Hypervisor, however, we assume that if both versions of the pkey are in its pkey table,
> then for its own infiniband operation (as opposed to performing its pkey virtualizing function),
> it should operate with the highest membership type in its table for a given 15-bit pkey.
That's what I was looking for. So, how can you know this assumption is
correct? It seems to me that if someone wanted to restrict membership
of the hypervisor as part of a security lockdown, then give full
membership to a guest because that guest is some high security, single
task guest, then this assumption would break things (the user would be
able to assign the full membership key to the guest OK, but regardless
of how they wanted the hypervisor to be subscribed to that particular
pkey, it would always get the full membership from the guest).
>> Shouldn't we pick the
>> pkey that's appropriate for the vHCA sending the message?
>
> We do. When QPs on the guests are created, the modify-qp commands are not executed on the guest,
> but rather are passed to the PPF for processing. The PPF replaces the guest-provided virtual pkey-index
> value with the appropriate physical pkey-index value. See procedure "update_pkey_index" in file
> resource_tracker.c, and all the places it is called (i.e., in the wrapper functions for the various
> modify-qp firmware commands).
That's fine for the guest, but I don't see how this solves the
assumption issue. My concern is that I could see a valid scenario that
a user might want to implement for security reasons that I think makes
your assumption above incorrect.
--
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
GPG KeyID: 0E572FDD
http://people.redhat.com/dledford
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 897 bytes --]
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <5050BCDD.50106-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-13 7:35 ` Jack Morgenstein
[not found] ` <201209131035.15318.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Jack Morgenstein @ 2012-09-13 7:35 UTC (permalink / raw)
To: Doug Ledford
Cc: roland-DgEjT+Ai2ygdnm+yROfE0A, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Liran Liss, Dotan Barak
On Wednesday 12 September 2012 19:48, Doug Ledford wrote:
> > On the Hypervisor, however, we assume that if both versions of the pkey are in its pkey table,
> > then for its own infiniband operation (as opposed to performing its pkey virtualizing function),
> > it should operate with the highest membership type in its table for a given 15-bit pkey.
>
> That's what I was looking for. So, how can you know this assumption is
> correct? It seems to me that if someone wanted to restrict membership
> of the hypervisor as part of a security lockdown, then give full
> membership to a guest because that guest is some high security, single
> task guest, then this assumption would break things (the user would be
> able to assign the full membership key to the guest OK, but regardless
> of how they wanted the hypervisor to be subscribed to that particular
> pkey, it would always get the full membership from the guest).
>
This issue, unfortunately, opens up a real "can of worms".
ib_find_cached_pkey() is used by the CM in determining the (15-bit) p_key
to be used for the connection (although 16-bit pkeys are placed in the
CM_REQ message (see IB Spec 1.2.1, table 99, page 667). The REQ handler on
the remote side finds the index in its pkey table which contains the
(15-bit) pkey enclosed in the REQ message. This index is then used
when creating the local RC qp as its pkey_index.
AFAIK, no check is performed regarding compatible membership forms (i.e.,
that at least one of the two sides has the full membership form of the
15-bit pkey) -- it is the network administrator's responsibility to see
that the pkey configurations are correct.
Up to now, the ib_core driver has assumed that only 1 membership form per
15-bit pkey is contained in the local pkey table.
Now that both forms may exist in the Hypervisor's table, we do have a problem.
1. We cannot depend on ib_find_cached_pkey simply finding the first 15-bit pkey
which matches its search, since we cannot depend on the order of full vs limited
membership forms appearing in the pkey table (the order is not controllable -- see
the example at the end of this post).
2. We therefore need a consistent policy for preferentially retrieving either the full
or the partial member for a given 15-bit pkey.
3. This policy must be configurable by the administrator per host, per HCA.
The problems that I see:
1. What if the Hypervisor needs to have limited membership for some connections, but
full membership for others? Such a split policy is complex to implement -- it would
require specifying PER PKEY whether the preferential return should be full, or should
be limited.
2. What if the Hypervisor has several HCAs, and wishes a different policy for each HCA?
There may be more problems as well. Liran?
In any event, the issue of multiple pkey forms in a given pkey table is actually separate
from SRIOV (it's just that SRIOV needs multiple forms, so the issue came up).
Solving it will require changes in the ib core driver.
I don't yet have a concrete proposal to fix this at present. Any ideas would be appreciated.
-Jack
P.S. We have the same issue in procedure ib_find_pkey(), also in this patch, in core/device.c .
This procedure is used in lots of places: ipoib, core/multicast, sa_query.
I seem to recall that there were problems with IPoIB when partial membership pkeys are used.
Liran, do you recall something?
======================================
It is also impossible to demand that the pkey table be ordered so that if both limited
and full pkeys are present, then the full membership value always appears first.
A simple example:
1. Fill the entire pkey table with full-membership pkeys. Say pkey A and pkey B are both
in the table, and that pkey A appears first.
2. Delete full-membership pkey A, and add the limited-membership value of Pkey B. This
new limited-pkey-B will be entered into the pkey table by OpenSM in the position just
vacated by the deleted pkey A -- it is the only available slot!.
3. The Limited-membership pkey-B value will therefore unavoidably appear in the table
before the full-membership pkey-B value.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <201209131035.15318.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-13 8:18 ` Or Gerlitz
2012-09-13 8:20 ` Or Gerlitz
1 sibling, 0 replies; 48+ messages in thread
From: Or Gerlitz @ 2012-09-13 8:18 UTC (permalink / raw)
To: Jack Morgenstein
Cc: Doug Ledford, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Liran Liss, Dotan Barak
On 13/09/2012 10:35, Jack Morgenstein wrote:
> I seem to recall that there were problems with IPoIB when partial membership pkeys are used.
There are some issues in the overall solution, since ARPs sent over the
broadcast group reach
also nodes with partial membership their HCA generated pkey violation
errors which is a bit annoying,
the IBTA has been dealing with this for a while, not sure where it stands.
But, all in all, the approach taken by IPoIB is to always use the full
form of a pkey e.g 0x8001
and not 0x0001 when dealing with the core helper functions and within
ipoib internally.
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <201209131035.15318.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-13 8:18 ` Or Gerlitz
@ 2012-09-13 8:20 ` Or Gerlitz
1 sibling, 0 replies; 48+ messages in thread
From: Or Gerlitz @ 2012-09-13 8:20 UTC (permalink / raw)
To: Jack Morgenstein
Cc: Doug Ledford, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Liran Liss, Dotan Barak
On 13/09/2012 10:35, Jack Morgenstein wrote:
> I seem to recall that there were problems with IPoIB when partial membership pkeys are used.
There are some issues in the overall solution, since ARPs sent over the
broadcast group reach
also nodes with partial membership their HCA generated pkey violation
errors which is a bit annoying,
the IBTA has been dealing with this for a while, not sure where it
stands. I'm not aware to other issues.
All in all, the approach taken by IPoIB is to always use the full form
of a pkey e.g 0x8001
and not 0x0001 when dealing with the core helper functions and within
ipoib internally and
externally -- e.g in the sysfs (and also in the rtnl link ops I'm
adding now) interfaces - of add/delete child and
show the pkey.
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <504F6C4F.6050207-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-12 7:56 ` Jack Morgenstein
@ 2012-09-13 15:53 ` Or Gerlitz
[not found] ` <50520193.2010304-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
1 sibling, 1 reply; 48+ messages in thread
From: Or Gerlitz @ 2012-09-13 15:53 UTC (permalink / raw)
To: Doug Ledford, Liran Liss
Cc: Jack Morgenstein, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA
On 11/09/2012 19:52, Doug Ledford wrote:
> On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
>> >Enhance the cached and non-cached pkey table lookups to enable limited and full
>> >members of the same pkey to co-exist in the pkey table.
>> >
>> >This is necessary for SRIOV to allow for a scheme where some guests would have the full
>> >membership pkey in their virtual pkey table, where other guests on the same hypervisor
>> >would have the limited one. In that sense, its an extension of the IBTA model for
>> >non virtualized nodes.
> OK, maybe I'm not getting something, but I'm curious why we always pick
> the full pkey in preference to the partial pkey. Shouldn't we pick the
> pkey that's appropriate for the vHCA sending the message?
>
> Also, given the rule of least surprise, don't you think it would be best
> to rename this function ib_find_cached_full_or_parital_pkey and in your
> next patch instead of naming it ib_find_exact_pkey just call that one
> ib_find_cached_pkey?
>
reply from Liran, our architect:
The physical PKey table can contain both full and partial memberships of
the same Pkey.
This is needed to serve 2 VFs that are granted access to the same PKey,
albeit with different membership types.
Indeed, there is no reason that the virtual PKey table of any *single*
VF will contain both membership types.However, the PF PKey table
certainly can because it mirrors the physical Pkey table.This patch is
required for ULPs running on top of the PF.
--Liran
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey
[not found] ` <50520193.2010304-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
@ 2012-09-13 16:00 ` Or Gerlitz
0 siblings, 0 replies; 48+ messages in thread
From: Or Gerlitz @ 2012-09-13 16:00 UTC (permalink / raw)
To: Doug Ledford, Liran Liss
Cc: Jack Morgenstein, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA
On 13/09/2012 18:53, Or Gerlitz wrote:
> The physical PKey table can contain both full and partial memberships
> of the same Pkey.
> This is needed to serve 2 VFs that are granted access to the same
> PKey, albeit with different membership types.
Example use case -- RDMA or IPoIB network storage (file/block) server
running on a VM
and clients running on other VMs. Some of the client VMs may be located
on the
same hypervisor node as the server, and we want the clients to use limited
partition such that they can't talk to each other on the storage partition.
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
[not found] ` <504F7068.6020606-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
@ 2012-09-20 23:15 ` Or Gerlitz
[not found] ` <CAJZOPZ+-1EUpGozrG+XsyPmS0RL791zP0=7OT8JMsmhwgL7QPQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
0 siblings, 1 reply; 48+ messages in thread
From: Or Gerlitz @ 2012-09-20 23:15 UTC (permalink / raw)
To: Doug Ledford
Cc: Jack Morgenstein, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
On Tue, Sep 11, 2012 at 8:10 PM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> > struct mlx4_ib_sriov{} is created by the master only.
> > It is a container for the following:
> > 1. All the info required by the PPF to multiplex and de-multiplex MADs
> > (including those from the PF). (struct mlx4_ib_demux_ctx demux)
>
> OK, so can we have at least a single reference to the various
> abbreviations before using them exclusively? I know PF and PPF may be
> common, but it might be nice that they were used once in full form
> before abbreviated in commit messages.
PF is physical function and PPF primary physical function
>
> > 2. All the info required to manage alias GUIDs (i.e., the GUID at
> > index 0 that each guest perceives. In fact, this is not the
> > GUID which is actually at index 0, but is, in fact, the GUID
> > which is at index[<VF number>] in the physical table.
>
> OK, this has been one of the things that has made reviewing this
> difficult. I freely admit that I've steadfastly ignored SRIOV for as
> long as I can, so maybe this is just me. But, in the context of this
> driver, how am I supposed to know which code paths will be on the host
> and which on the guest?
For the mlx4 driver the approach taken was to para-virtualize mlx4_core
such that both the PPF and VFs run the same driver but within that
driver two flows are operative. In mlx4_core it should be pretty clear
to see where are you now, in mlx4_ib sometimes less easy, I'll leave
that to Jack
to address with more details.
> Also, I note that you do math every time you want to know if you are on
> a parent device or a virtual device. Do you really want to do math all
> the time, or would it be better to save off your status on device init
> and just refer to that when you would do math in this patch?
this is addressed in down stream patch #21 of this series
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support
[not found] ` <CAJZOPZ+-1EUpGozrG+XsyPmS0RL791zP0=7OT8JMsmhwgL7QPQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2012-09-21 7:42 ` Jack Morgenstein
0 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-09-21 7:42 UTC (permalink / raw)
To: Or Gerlitz
Cc: Doug Ledford, roland-DgEjT+Ai2ygdnm+yROfE0A,
linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
On Friday 21 September 2012 02:15, Or Gerlitz wrote:
> On Tue, Sep 11, 2012 at 8:10 PM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> > On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> > > struct mlx4_ib_sriov{} is created by the master only.
> > > It is a container for the following:
> > > 1. All the info required by the PPF to multiplex and de-multiplex MADs
> > > (including those from the PF). (struct mlx4_ib_demux_ctx demux)
> >
> > OK, so can we have at least a single reference to the various
> > abbreviations before using them exclusively? I know PF and PPF may be
> > common, but it might be nice that they were used once in full form
> > before abbreviated in commit messages.
>
> PF is physical function and PPF primary physical function
I'll do that during the review/resubmission process
>
> >
> > > 2. All the info required to manage alias GUIDs (i.e., the GUID at
> > > index 0 that each guest perceives. In fact, this is not the
> > > GUID which is actually at index 0, but is, in fact, the GUID
> > > which is at index[<VF number>] in the physical table.
> >
> > OK, this has been one of the things that has made reviewing this
> > difficult. I freely admit that I've steadfastly ignored SRIOV for as
> > long as I can, so maybe this is just me. But, in the context of this
> > driver, how am I supposed to know which code paths will be on the host
> > and which on the guest?
>
> For the mlx4 driver the approach taken was to para-virtualize mlx4_core
> such that both the PPF and VFs run the same driver but within that
> driver two flows are operative. In mlx4_core it should be pretty clear
> to see where are you now, in mlx4_ib sometimes less easy, I'll leave
> that to Jack
> to address with more details.
>
The flows are indeed complex. The issue we have is that there is a single driver
for both "Native" (i.e., non-sriov) mode, for the SRIOV master (PF/PPF) and for SRIOV slaves (VFs).
We have macros to assist in routing these flows:
mlx4_is_mfunc() -- test for multifunc active (i.e, we are in an sriov master or an sriov slave).
mlx4_is_master() -- we are an sriov master. False if we are an SRIOV slave, or if we are in Native mode.
mlx4_is_slave() -- we are an SRIOV slave. False if we are an SRIOV master, or if we are in Native mode.
You will see
if (!mlx4_is_master)
return;
in various procedures, to indicate that what follows is code that only an SRIOV master should execute.
For example, look at procedure
void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
{
struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
if (!mlx4_is_master(dev))
return;
priv->virt2phys_pkey[slave][port - 1][i] = val;
}
You will also see the tests in combination (e.g., if (!mlx4_is_master && !<some other condition>) do something)
Finally, you will see lots of places where the test is positive, around a block of code:
if (mlx4_is_master()) {
do something;
}
-Jack
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 21/22] {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations
[not found] ` <1343983258-6268-22-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-22 10:25 ` Roland Dreier
0 siblings, 0 replies; 48+ messages in thread
From: Roland Dreier @ 2012-09-22 10:25 UTC (permalink / raw)
To: Jack Morgenstein
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
On Fri, Aug 3, 2012 at 1:40 AM, Jack Morgenstein
<jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org> wrote:
> Previously, the structure of a guest's proxy QPs followed the structure of the PPF
> special qps (qp0 port 1, qp0 port 2, qp1 port 1, qp1 port 2, ...).
> The guest then did offset calculations on the sqp_base qp number that the PPF passed
> to it in QUERY_FUNC_CAP().
>
> This is now changed so that the guest does no offset calculations regarding proxy or
> tunnel QPs to use. This change frees the PPF from needing to adhere to a specific
> order in allocating proxy and tunnel QPs.
It seems a little strange that you rewrite code that is only added
earlier in the same
patch series. Is there any reason why we can't just add the right
code in the first place?
- R.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 11/22] IB/mlx4: Add CM paravirtualization
[not found] ` <1343983258-6268-12-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
@ 2012-09-24 19:33 ` Roland Dreier
0 siblings, 0 replies; 48+ messages in thread
From: Roland Dreier @ 2012-09-24 19:33 UTC (permalink / raw)
To: Jack Morgenstein
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w, Amir Vadai
On Fri, Aug 3, 2012 at 1:40 AM, Jack Morgenstein
<jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org> wrote:
> +static struct id_map_entry *
> +id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
> +{
> + int ret, id;
> + static int next_id;
> + struct id_map_entry *ent;
> + struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
> +
> + ent = kmalloc(sizeof(struct id_map_entry), GFP_KERNEL);
> + if (!ent) {
> + mlx4_ib_warn(ibdev, "Couldn't allocate id cache entry - out of memory\n");
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + ent->sl_cm_id = sl_cm_id;
> + ent->slave_id = slave_id;
> + ent->scheduled_delete = 0;
> + ent->dev = to_mdev(ibdev);
> + INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
> +
> + do {
> + spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
> + ret = idr_get_new_above(&sriov->pv_id_table, ent,
> + next_id, &id);
> + if (!ret) {
> + next_id = ((unsigned) id + 1) & MAX_ID_MASK;
> + ent->pv_cm_id = (u32)id;
> + sl_id_map_add(ibdev, ent);
> + }
> +
> + spin_unlock(&sriov->id_map_lock);
> + } while (ret == -EAGAIN && idr_pre_get(&sriov->pv_id_table, GFP_KERNEL));
> + /*the function idr_get_new_above can return -ENOSPC, so don't insert in that case.*/
> + if (!ret) {
> + spin_lock(&sriov->id_map_lock);
> + list_add_tail(&ent->list, &sriov->cm_list);
> + spin_unlock(&sriov->id_map_lock);
> + return ent;
> + }
> + /*error flow*/
> + kfree(ent);
> + mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret);
> + return ERR_PTR(-ENOMEM);
> +}
This looks strange. We have one spin lock and IDR per device, but next_id
is just a static variable in the function. So it seems we'll get unexpected
results if we have multiple physical HCAs running this code in parallel
(since they can both do read-modify-write of next_id and step on each other).
Should next_id be moved into the per-device structure? Is there any need
for idr_get_new_above() at all, or could this be replaced by idr_get_new()?
(I guess we want to avoid CM ID reuse for as long as possible)
- R.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use
[not found] ` <1343983258-6268-2-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-05 14:55 ` Doug Ledford
@ 2012-09-24 19:34 ` Roland Dreier
[not found] ` <CAL1RGDXrJ+c2tgxYsLMZQVz+os7E3FZROMJ9D7oqPiQbdk1g7w-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
1 sibling, 1 reply; 48+ messages in thread
From: Roland Dreier @ 2012-09-24 19:34 UTC (permalink / raw)
To: Jack Morgenstein
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
So I applied this whole series, with the plan to merge this for 3.7.
Please send any changes as patches on top of what's already merged.
Thanks,
Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
* Re: [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use
[not found] ` <CAL1RGDXrJ+c2tgxYsLMZQVz+os7E3FZROMJ9D7oqPiQbdk1g7w-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2012-09-25 8:28 ` Jack Morgenstein
0 siblings, 0 replies; 48+ messages in thread
From: Jack Morgenstein @ 2012-09-25 8:28 UTC (permalink / raw)
To: Roland Dreier
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
ogerlitz-VPRAkNaXOzVWk0Htik3J/w
Thanks, Roland!
-Jack
On Monday 24 September 2012 21:34, Roland Dreier wrote:
> So I applied this whole series, with the plan to merge this for 3.7.
>
> Please send any changes as patches on top of what's already merged.
>
> Thanks,
> Roland
>
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 48+ messages in thread
end of thread, other threads:[~2012-09-25 8:28 UTC | newest]
Thread overview: 48+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-08-03 8:40 [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces Jack Morgenstein
[not found] ` <1343983258-6268-1-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-08-03 8:40 ` [PATCH for-next V2 01/22] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use Jack Morgenstein
[not found] ` <1343983258-6268-2-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-05 14:55 ` Doug Ledford
[not found] ` <504767EB.6090004-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-06 10:06 ` Jack Morgenstein
2012-09-24 19:34 ` Roland Dreier
[not found] ` <CAL1RGDXrJ+c2tgxYsLMZQVz+os7E3FZROMJ9D7oqPiQbdk1g7w-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2012-09-25 8:28 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 02/22] IB/core: change pkey table lookups to support full and partial membership for the same pkey Jack Morgenstein
[not found] ` <1343983258-6268-3-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-11 16:52 ` Doug Ledford
[not found] ` <504F6C4F.6050207-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-12 7:56 ` Jack Morgenstein
[not found] ` <201209121056.00309.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-12 16:48 ` Doug Ledford
[not found] ` <5050BCDD.50106-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-13 7:35 ` Jack Morgenstein
[not found] ` <201209131035.15318.jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-13 8:18 ` Or Gerlitz
2012-09-13 8:20 ` Or Gerlitz
2012-09-13 15:53 ` Or Gerlitz
[not found] ` <50520193.2010304-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
2012-09-13 16:00 ` Or Gerlitz
2012-08-03 8:40 ` [PATCH for-next V2 03/22] IB/core: Add ib_find_exact_cached_pkey() to search for 16-bit pkey match Jack Morgenstein
[not found] ` <1343983258-6268-4-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-11 16:53 ` Doug Ledford
2012-09-11 17:12 ` Doug Ledford
[not found] ` <504F70FB.6030806-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-11 19:07 ` Roland Dreier
[not found] ` <CAL1RGDW6D25s+R8HuxK-DMrsS_wvKDTL+LSp2X-TK8gB8Vm2Fg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2012-09-11 20:34 ` Doug Ledford
[not found] ` <504FA064.6040002-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-11 20:43 ` Roland Dreier
[not found] ` <CAL1RGDXpeK3KjrmzyivQs8FOs2dg5MqSmuVsdRE+bVD5OXe6BA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2012-09-11 22:51 ` Doug Ledford
2012-08-03 8:40 ` [PATCH for-next V2 04/22] IB/mlx4: SRIOV IB context objects and proxy/tunnel sqp support Jack Morgenstein
[not found] ` <1343983258-6268-5-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-11 17:10 ` Doug Ledford
[not found] ` <504F7068.6020606-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
2012-09-20 23:15 ` Or Gerlitz
[not found] ` <CAJZOPZ+-1EUpGozrG+XsyPmS0RL791zP0=7OT8JMsmhwgL7QPQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2012-09-21 7:42 ` Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 05/22] net/mlx4_core: Add proxy and tunnel QPs to the reserved QP area Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 06/22] IB/mlx4: Initialize SRIOV IB support for slaves in master context Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 07/22] {NET,IB}/mlx4: Implement QP paravirtualization and maintain phys_pkey_cache for smp_snoop Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 08/22] IB/mlx4: SRIOV multiplex and demultiplex MADs Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 09/22] {NET,IB}/mlx4: MAD_IFC paravirtualization Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 10/22] IB/mlx4: Added Multicast Groups (MCG) para-virtualization for SRIOV Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 11/22] IB/mlx4: Add CM paravirtualization Jack Morgenstein
[not found] ` <1343983258-6268-12-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-24 19:33 ` Roland Dreier
2012-08-03 8:40 ` [PATCH for-next V2 12/22] net/mlx4_core: Add IB port-state machine, and port mgmt event propagation infrastructure Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 13/22] {NET,IB}/mlx4: Add alias_guid mechanism Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 14/22] IB/mlx4: Propagate pkey and guid change port management events to slaves Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 15/22] IB/mlx4: Add iov directory in sysfs under the ib device Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 16/22] net/mlx4_core: Adjustments to SET_PORT for SRIOV-IB Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 17/22] net/mlx4_core: INIT/CLOSE port logic for IB ports in SRIOV mode Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 18/22] IB/mlx4: Miscellaneous adjustments to SRIOV IB support Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 19/22] {NET,IB}/mlx4: Activate SRIOV mode for IB Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 20/22] {NET,IB}/mlx4: Paravirtualize Node Guids for slaves Jack Morgenstein
2012-08-03 8:40 ` [PATCH for-next V2 21/22] {NET,IB}/mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations Jack Morgenstein
[not found] ` <1343983258-6268-22-git-send-email-jackm-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
2012-09-22 10:25 ` Roland Dreier
2012-08-03 8:40 ` [PATCH for-next V2 22/22] IB/mlx4: Create pv contexts for active VFs when PF (master) ib driver initializes Jack Morgenstein
2012-08-03 9:00 ` [PATCH for-next V2 00/22] Add SRIOV support for IB interfaces Tziporet Koren
2012-08-12 9:34 ` Or Gerlitz
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).