* [net-next v4 5/5] ixgbe: Disable flow control for XFI
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem
Cc: Tony Nguyen, netdev, nhorman, sassmann, jogreene, Emil Tantilov,
Jeff Kirsher
In-Reply-To: <20170725234138.42283-1-jeffrey.t.kirsher@intel.com>
From: Tony Nguyen <anthony.l.nguyen@intel.com>
Flow control autonegotiation is not supported for XFI. Make sure that
ixgbe_device_supports_autoneg_fc() returns false and
hw->fc.disable_fc_autoneg is set to true to avoid running the fc_autoneg
function for that device.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/ethernet/intel/ixgbe/ixgbe_common.c | 5 ++-
drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c | 57 ++++++++++++++-----------
2 files changed, 35 insertions(+), 27 deletions(-)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
index 40ae7db468ea..2c19070d2a0b 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
@@ -97,7 +97,10 @@ bool ixgbe_device_supports_autoneg_fc(struct ixgbe_hw *hw)
break;
case ixgbe_media_type_backplane:
- supported = true;
+ if (hw->device_id == IXGBE_DEV_ID_X550EM_X_XFI)
+ supported = false;
+ else
+ supported = true;
break;
case ixgbe_media_type_copper:
/* only some copper devices support flow control autoneg */
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
index 95adbda36235..19fbb2f28ea4 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
@@ -2843,7 +2843,7 @@ static s32 ixgbe_setup_fc_x550em(struct ixgbe_hw *hw)
{
bool pause, asm_dir;
u32 reg_val;
- s32 rc;
+ s32 rc = 0;
/* Validate the requested mode */
if (hw->fc.strict_ieee && hw->fc.requested_mode == ixgbe_fc_rx_pause) {
@@ -2886,32 +2886,37 @@ static s32 ixgbe_setup_fc_x550em(struct ixgbe_hw *hw)
return IXGBE_ERR_CONFIG;
}
- if (hw->device_id != IXGBE_DEV_ID_X550EM_X_KR &&
- hw->device_id != IXGBE_DEV_ID_X550EM_A_KR &&
- hw->device_id != IXGBE_DEV_ID_X550EM_A_KR_L)
- return 0;
-
- rc = hw->mac.ops.read_iosf_sb_reg(hw,
- IXGBE_KRM_AN_CNTL_1(hw->bus.lan_id),
- IXGBE_SB_IOSF_TARGET_KR_PHY,
- ®_val);
- if (rc)
- return rc;
-
- reg_val &= ~(IXGBE_KRM_AN_CNTL_1_SYM_PAUSE |
- IXGBE_KRM_AN_CNTL_1_ASM_PAUSE);
- if (pause)
- reg_val |= IXGBE_KRM_AN_CNTL_1_SYM_PAUSE;
- if (asm_dir)
- reg_val |= IXGBE_KRM_AN_CNTL_1_ASM_PAUSE;
- rc = hw->mac.ops.write_iosf_sb_reg(hw,
- IXGBE_KRM_AN_CNTL_1(hw->bus.lan_id),
- IXGBE_SB_IOSF_TARGET_KR_PHY,
- reg_val);
-
- /* This device does not fully support AN. */
- hw->fc.disable_fc_autoneg = true;
+ switch (hw->device_id) {
+ case IXGBE_DEV_ID_X550EM_X_KR:
+ case IXGBE_DEV_ID_X550EM_A_KR:
+ case IXGBE_DEV_ID_X550EM_A_KR_L:
+ rc = hw->mac.ops.read_iosf_sb_reg(hw,
+ IXGBE_KRM_AN_CNTL_1(hw->bus.lan_id),
+ IXGBE_SB_IOSF_TARGET_KR_PHY,
+ ®_val);
+ if (rc)
+ return rc;
+ reg_val &= ~(IXGBE_KRM_AN_CNTL_1_SYM_PAUSE |
+ IXGBE_KRM_AN_CNTL_1_ASM_PAUSE);
+ if (pause)
+ reg_val |= IXGBE_KRM_AN_CNTL_1_SYM_PAUSE;
+ if (asm_dir)
+ reg_val |= IXGBE_KRM_AN_CNTL_1_ASM_PAUSE;
+ rc = hw->mac.ops.write_iosf_sb_reg(hw,
+ IXGBE_KRM_AN_CNTL_1(hw->bus.lan_id),
+ IXGBE_SB_IOSF_TARGET_KR_PHY,
+ reg_val);
+
+ /* This device does not fully support AN. */
+ hw->fc.disable_fc_autoneg = true;
+ break;
+ case IXGBE_DEV_ID_X550EM_X_XFI:
+ hw->fc.disable_fc_autoneg = true;
+ break;
+ default:
+ break;
+ }
return rc;
}
--
2.13.3
^ permalink raw reply related
* [net-next v4 3/5] ixgbe: Update NW_MNG_IF_SEL support for X553
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem
Cc: Tony Nguyen, netdev, nhorman, sassmann, jogreene, Paul Greenwalt,
Jeff Kirsher
In-Reply-To: <20170725234138.42283-1-jeffrey.t.kirsher@intel.com>
From: Tony Nguyen <anthony.l.nguyen@intel.com>
The MAC register NW_MNG_IF_SEL fields have been redefined for
X553. These changes impact the iXFI driver code flow. Since iXFI is
only supported in X552, add MAC checks for iXFI flows.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 2 +-
drivers/net/ethernet/intel/ixgbe/ixgbe_type.h | 4 ++--
drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c | 14 +++++++++++---
3 files changed, 14 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
index 0f867dcda65f..96606e3eb965 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
@@ -386,7 +386,7 @@ u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
if (ixgbe_removed(reg_addr))
return IXGBE_FAILED_READ_REG;
if (unlikely(hw->phy.nw_mng_if_sel &
- IXGBE_NW_MNG_IF_SEL_ENABLE_10_100M)) {
+ IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
struct ixgbe_adapter *adapter;
int i;
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h b/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
index 9c2460c5ef1b..ffa0ee5cd0f5 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
@@ -3778,8 +3778,8 @@ struct ixgbe_info {
#define IXGBE_NW_MNG_IF_SEL_PHY_SPEED_1G BIT(19)
#define IXGBE_NW_MNG_IF_SEL_PHY_SPEED_2_5G BIT(20)
#define IXGBE_NW_MNG_IF_SEL_PHY_SPEED_10G BIT(21)
-#define IXGBE_NW_MNG_IF_SEL_ENABLE_10_100M BIT(23)
-#define IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE BIT(24)
+#define IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE BIT(25)
+#define IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE BIT(24) /* X552 only */
#define IXGBE_NW_MNG_IF_SEL_MDIO_PHY_ADD_SHIFT 3
#define IXGBE_NW_MNG_IF_SEL_MDIO_PHY_ADD \
(0x1F << IXGBE_NW_MNG_IF_SEL_MDIO_PHY_ADD_SHIFT)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
index aa34e0b131bb..95adbda36235 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
@@ -1555,9 +1555,14 @@ static s32 ixgbe_restart_an_internal_phy_x550em(struct ixgbe_hw *hw)
**/
static s32 ixgbe_setup_ixfi_x550em(struct ixgbe_hw *hw, ixgbe_link_speed *speed)
{
+ struct ixgbe_mac_info *mac = &hw->mac;
s32 status;
u32 reg_val;
+ /* iXFI is only supported with X552 */
+ if (mac->type != ixgbe_mac_X550EM_x)
+ return IXGBE_ERR_LINK_SETUP;
+
/* Disable AN and force speed to 10G Serial. */
status = ixgbe_read_iosf_sb_reg_x550(hw,
IXGBE_KRM_LINK_CTRL_1(hw->bus.lan_id),
@@ -1874,8 +1879,10 @@ static s32 ixgbe_setup_mac_link_t_X550em(struct ixgbe_hw *hw,
else
force_speed = IXGBE_LINK_SPEED_1GB_FULL;
- /* If internal link mode is XFI, then setup XFI internal link. */
- if (!(hw->phy.nw_mng_if_sel & IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE)) {
+ /* If X552 and internal link mode is XFI, then setup XFI internal link.
+ */
+ if (hw->mac.type == ixgbe_mac_X550EM_x &&
+ !(hw->phy.nw_mng_if_sel & IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE)) {
status = ixgbe_setup_ixfi_x550em(hw, &force_speed);
if (status)
@@ -2628,7 +2635,8 @@ static s32 ixgbe_setup_internal_phy_t_x550em(struct ixgbe_hw *hw)
if (hw->mac.ops.get_media_type(hw) != ixgbe_media_type_copper)
return IXGBE_ERR_CONFIG;
- if (hw->phy.nw_mng_if_sel & IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE) {
+ if (!(hw->mac.type == ixgbe_mac_X550EM_x &&
+ !(hw->phy.nw_mng_if_sel & IXGBE_NW_MNG_IF_SEL_INT_PHY_MODE))) {
speed = IXGBE_LINK_SPEED_10GB_FULL |
IXGBE_LINK_SPEED_1GB_FULL;
return ixgbe_setup_kr_speed_x550em(hw, speed);
--
2.13.3
^ permalink raw reply related
* [net-next v4 4/5] ixgbe: Do not support flow control autonegotiation for X553
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem
Cc: Tony Nguyen, netdev, nhorman, sassmann, jogreene, Emil Tantilov,
Jeff Kirsher
In-Reply-To: <20170725234138.42283-1-jeffrey.t.kirsher@intel.com>
From: Tony Nguyen <anthony.l.nguyen@intel.com>
Flow control autonegotiation is not supported for fiber on X553. Add
device ID checks in ixgbe_device_supports_autoneg_fc() to return the
appropriate value.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/ethernet/intel/ixgbe/ixgbe_common.c | 25 +++++++++++++++++++------
1 file changed, 19 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
index 4e35e7017f3d..40ae7db468ea 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_common.c
@@ -79,13 +79,22 @@ bool ixgbe_device_supports_autoneg_fc(struct ixgbe_hw *hw)
switch (hw->phy.media_type) {
case ixgbe_media_type_fiber:
- hw->mac.ops.check_link(hw, &speed, &link_up, false);
- /* if link is down, assume supported */
- if (link_up)
- supported = speed == IXGBE_LINK_SPEED_1GB_FULL ?
+ /* flow control autoneg black list */
+ switch (hw->device_id) {
+ case IXGBE_DEV_ID_X550EM_A_SFP:
+ case IXGBE_DEV_ID_X550EM_A_SFP_N:
+ supported = false;
+ break;
+ default:
+ hw->mac.ops.check_link(hw, &speed, &link_up, false);
+ /* if link is down, assume supported */
+ if (link_up)
+ supported = speed == IXGBE_LINK_SPEED_1GB_FULL ?
true : false;
- else
- supported = true;
+ else
+ supported = true;
+ }
+
break;
case ixgbe_media_type_backplane:
supported = true;
@@ -111,6 +120,10 @@ bool ixgbe_device_supports_autoneg_fc(struct ixgbe_hw *hw)
break;
}
+ if (!supported)
+ hw_dbg(hw, "Device %x does not support flow control autoneg\n",
+ hw->device_id);
+
return supported;
}
--
2.13.3
^ permalink raw reply related
* [net-next v4 1/5] ixgbe: Ensure MAC filter was added before setting MACVLAN
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem; +Cc: Tony Nguyen, netdev, nhorman, sassmann, jogreene, Jeff Kirsher
In-Reply-To: <20170725234138.42283-1-jeffrey.t.kirsher@intel.com>
From: Tony Nguyen <anthony.l.nguyen@intel.com>
This patch adds a check to ensure that adding the MAC filter was
successful before setting the MACVLAN. If it was unsuccessful, propagate
the error.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c
index 0760bd7eeb01..112d24c6c9ce 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c
@@ -679,8 +679,9 @@ static void ixgbe_clear_vf_vlans(struct ixgbe_adapter *adapter, u32 vf)
static int ixgbe_set_vf_macvlan(struct ixgbe_adapter *adapter,
int vf, int index, unsigned char *mac_addr)
{
- struct list_head *pos;
struct vf_macvlans *entry;
+ struct list_head *pos;
+ int retval = 0;
if (index <= 1) {
list_for_each(pos, &adapter->vf_mvs.l) {
@@ -721,13 +722,15 @@ static int ixgbe_set_vf_macvlan(struct ixgbe_adapter *adapter,
if (!entry || !entry->free)
return -ENOSPC;
+ retval = ixgbe_add_mac_filter(adapter, mac_addr, vf);
+ if (retval < 0)
+ return retval;
+
entry->free = false;
entry->is_macvlan = true;
entry->vf = vf;
memcpy(entry->vf_macvlan, mac_addr, ETH_ALEN);
- ixgbe_add_mac_filter(adapter, mac_addr, vf);
-
return 0;
}
--
2.13.3
^ permalink raw reply related
* [net-next v4 2/5] ixgbe: Enable LASI interrupts for X552 devices
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem; +Cc: Tony Nguyen, netdev, nhorman, sassmann, jogreene, Jeff Kirsher
In-Reply-To: <20170725234138.42283-1-jeffrey.t.kirsher@intel.com>
From: Tony Nguyen <anthony.l.nguyen@intel.com>
Enable LASI interrupts on X552 devices in order to receive notifications of
link configurations of the external PHY and support the configuration of
the internal iXFI link since iXFI does not support auto-negotiation. This
is not required for X553 devices; add a check to avoid enabling LASI
interrupts for X553 devices.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c | 31 +++++++++++++++++++--------
1 file changed, 22 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
index 72d84a065e34..aa34e0b131bb 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
@@ -2404,17 +2404,30 @@ static s32 ixgbe_enable_lasi_ext_t_x550em(struct ixgbe_hw *hw)
status = ixgbe_get_lasi_ext_t_x550em(hw, &lsc);
/* Enable link status change alarm */
- status = hw->phy.ops.read_reg(hw, IXGBE_MDIO_PMA_TX_VEN_LASI_INT_MASK,
- MDIO_MMD_AN, ®);
- if (status)
- return status;
- reg |= IXGBE_MDIO_PMA_TX_VEN_LASI_INT_EN;
+ /* Enable the LASI interrupts on X552 devices to receive notifications
+ * of the link configurations of the external PHY and correspondingly
+ * support the configuration of the internal iXFI link, since iXFI does
+ * not support auto-negotiation. This is not required for X553 devices
+ * having KR support, which performs auto-negotiations and which is used
+ * as the internal link to the external PHY. Hence adding a check here
+ * to avoid enabling LASI interrupts for X553 devices.
+ */
+ if (hw->mac.type != ixgbe_mac_x550em_a) {
+ status = hw->phy.ops.read_reg(hw,
+ IXGBE_MDIO_PMA_TX_VEN_LASI_INT_MASK,
+ MDIO_MMD_AN, ®);
+ if (status)
+ return status;
+
+ reg |= IXGBE_MDIO_PMA_TX_VEN_LASI_INT_EN;
- status = hw->phy.ops.write_reg(hw, IXGBE_MDIO_PMA_TX_VEN_LASI_INT_MASK,
- MDIO_MMD_AN, reg);
- if (status)
- return status;
+ status = hw->phy.ops.write_reg(hw,
+ IXGBE_MDIO_PMA_TX_VEN_LASI_INT_MASK,
+ MDIO_MMD_AN, reg);
+ if (status)
+ return status;
+ }
/* Enable high temperature failure and global fault alarms */
status = hw->phy.ops.read_reg(hw, IXGBE_MDIO_GLOBAL_INT_MASK,
--
2.13.3
^ permalink raw reply related
* [net-next v4 0/5][pull request] 10GbE Intel Wired LAN Driver Updates 2017-07-25
From: Jeff Kirsher @ 2017-07-25 23:41 UTC (permalink / raw)
To: davem; +Cc: Jeff Kirsher, netdev, nhorman, sassmann, jogreene
This series contains updates to ixgbe only.
Tony provides all of the changes in the series, starting with adding a
check to ensure that adding a MAC filter was successful, before setting the
MACVLAN. In order to receive notifications of link configurations of the
external PHY and support the configuration of the internal iXFI link on
X552 devices, Tony enables LASI interrupts. Update the iXFI driver code
flow, since the MAC register NW_MNG_IF_SEL fields have been redefined for
X553 devices, so add MAC checks for iXFI flows. Added additional checks
for flow control autonegotiation, since it is not support for X553 fiber
and XFI devices.
v2: removed unnecessary parens noticed by David Miller in patch 6 of the
series.
v3: dropped patch 6 of the original series, while we work out a more
generic solution for malicious driver detection (MDD) support.
v4: updated patch 1 of the series with the comments from Joe Perches which
were:
- switched logic to return on error
- return 0 on success
- declare retval as an integer
The following are changes since commit 34c8c10a766c8daf42293664f5c9ca5516fa3f41:
Merge branch 'bnxt_en-Fix-kbuild-errors-and-rename-phys_port_name'
and are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue 10GbE
Tony Nguyen (5):
ixgbe: Ensure MAC filter was added before setting MACVLAN
ixgbe: Enable LASI interrupts for X552 devices
ixgbe: Update NW_MNG_IF_SEL support for X553
ixgbe: Do not support flow control autonegotiation for X553
ixgbe: Disable flow control for XFI
drivers/net/ethernet/intel/ixgbe/ixgbe_common.c | 30 +++++--
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 2 +-
drivers/net/ethernet/intel/ixgbe/ixgbe_sriov.c | 9 ++-
drivers/net/ethernet/intel/ixgbe/ixgbe_type.h | 4 +-
drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c | 102 +++++++++++++++---------
5 files changed, 96 insertions(+), 51 deletions(-)
--
2.13.3
^ permalink raw reply
* Re: [PATCH net] bpf: don't zero out the info struct in bpf_obj_get_info_by_fd()
From: Daniel Borkmann @ 2017-07-25 23:29 UTC (permalink / raw)
To: Jakub Kicinski; +Cc: netdev, oss-drivers, alexei.starovoitov, kafai
In-Reply-To: <20170725161547.64d1cba5@cakuba.netronome.com>
On 07/26/2017 01:15 AM, Jakub Kicinski wrote:
> On Wed, 26 Jul 2017 00:59:49 +0200, Daniel Borkmann wrote:
>>> @@ -418,6 +420,8 @@ static void test_bpf_obj_id(void)
>>> nr_id_found++;
>>>
>>> err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
>>> + prog_infos[i].jited_prog_insns = 0;
>>> + prog_infos[i].xlated_prog_insns = 0;
>>
>> Can you elaborate why this one above is needed?
>
> Ah, I removed the comment about it at the last minute. The check below
> compares the info we get here with info we got reading the programs in
> the earlier loop - using memcmp().
Yep, makes sense. I mistook it for 'length' given it is not NULL but 0,
but that is due to __aligned_u64. ;) Anyway, thanks for clarifying.
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
^ permalink raw reply
* Re: [PATCH net] bpf: don't zero out the info struct in bpf_obj_get_info_by_fd()
From: Jakub Kicinski @ 2017-07-25 23:20 UTC (permalink / raw)
To: Daniel Borkmann; +Cc: netdev, oss-drivers, alexei.starovoitov, kafai
In-Reply-To: <20170725161547.64d1cba5@cakuba.netronome.com>
On Tue, 25 Jul 2017 16:15:47 -0700, Jakub Kicinski wrote:
> On Wed, 26 Jul 2017 00:59:49 +0200, Daniel Borkmann wrote:
> > > @@ -418,6 +420,8 @@ static void test_bpf_obj_id(void)
> > > nr_id_found++;
> > >
> > > err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
> > > + prog_infos[i].jited_prog_insns = 0;
> > > + prog_infos[i].xlated_prog_insns = 0;
> >
> > Can you elaborate why this one above is needed?
>
> Ah, I removed the comment about it at the last minute. The check below
> compares the info we get here with info we got reading the programs in
> the earlier loop - using memcmp(). This call, however, doesn't fill in
> the pointers for jited and xlated images, so the memcmp() would fail.
>
> It used to work when bpf_obj_get_info_by_fd() was zeroing info, since
> the pointers would be cleared by it, and no dump ever returned, it
> didn't matter that the call sites differ.
FWIW the comment was this:
+ /* Clear the insns pointers, we're not requesting dumps here.
+ * Otherwise the byte-by-byte comparison below would fail.
+ */
> > > CHECK(err || info_len != sizeof(struct bpf_prog_info) ||
> > > memcmp(&prog_info, &prog_infos[i], info_len),
> > > "get-prog-info(next_id->fd)",
>
^ permalink raw reply
* Re: [PATCH net] bpf: don't zero out the info struct in bpf_obj_get_info_by_fd()
From: Jakub Kicinski @ 2017-07-25 23:15 UTC (permalink / raw)
To: Daniel Borkmann; +Cc: netdev, oss-drivers, alexei.starovoitov, kafai
In-Reply-To: <5977CD65.20504@iogearbox.net>
On Wed, 26 Jul 2017 00:59:49 +0200, Daniel Borkmann wrote:
> > @@ -418,6 +420,8 @@ static void test_bpf_obj_id(void)
> > nr_id_found++;
> >
> > err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
> > + prog_infos[i].jited_prog_insns = 0;
> > + prog_infos[i].xlated_prog_insns = 0;
>
> Can you elaborate why this one above is needed?
Ah, I removed the comment about it at the last minute. The check below
compares the info we get here with info we got reading the programs in
the earlier loop - using memcmp(). This call, however, doesn't fill in
the pointers for jited and xlated images, so the memcmp() would fail.
It used to work when bpf_obj_get_info_by_fd() was zeroing info, since
the pointers would be cleared by it, and no dump ever returned, it
didn't matter that the call sites differ.
> > CHECK(err || info_len != sizeof(struct bpf_prog_info) ||
> > memcmp(&prog_info, &prog_infos[i], info_len),
> > "get-prog-info(next_id->fd)",
^ permalink raw reply
* Re: [PATCH net] bpf: don't zero out the info struct in bpf_obj_get_info_by_fd()
From: Daniel Borkmann @ 2017-07-25 22:59 UTC (permalink / raw)
To: Jakub Kicinski, netdev; +Cc: oss-drivers, alexei.starovoitov, kafai
In-Reply-To: <20170725221612.6937-1-jakub.kicinski@netronome.com>
On 07/26/2017 12:16 AM, Jakub Kicinski wrote:
> The buffer passed to bpf_obj_get_info_by_fd() should be initialized
> to zeros. Kernel will enforce that to guarantee we can safely extend
> info structures in the future.
>
> Making the bpf_obj_get_info_by_fd() call in libbpf perform the zeroing
> is problematic, however, since some members of the info structures
> may need to be initialized by the callers (for instance pointers
> to buffers to which kernel is to dump translated and jited images).
>
> Remove the zeroing and fix up the in-tree callers before any kernel
> has been released with this code.
>
> As Daniel points out this seems to be the intended operation anyway,
> since commit 95b9afd3987f ("bpf: Test for bpf ID") is itself setting
> the buffer pointers before calling bpf_obj_get_info_by_fd().
>
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> ---
> I have a small patch to add checking if kernel actually populated
> the instruction dumps which I will post after this ends up in net-next.
>
> tools/lib/bpf/bpf.c | 1 -
> tools/testing/selftests/bpf/test_progs.c | 8 ++++++--
> 2 files changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
> index 412a7c82995a..256f571f2ab5 100644
> --- a/tools/lib/bpf/bpf.c
> +++ b/tools/lib/bpf/bpf.c
> @@ -314,7 +314,6 @@ int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len)
> int err;
>
> bzero(&attr, sizeof(attr));
> - bzero(info, *info_len);
> attr.info.bpf_fd = prog_fd;
> attr.info.info_len = *info_len;
> attr.info.info = ptr_to_u64(info);
> diff --git a/tools/testing/selftests/bpf/test_progs.c b/tools/testing/selftests/bpf/test_progs.c
> index 5855cd3d3d45..1f7dd35551b9 100644
> --- a/tools/testing/selftests/bpf/test_progs.c
> +++ b/tools/testing/selftests/bpf/test_progs.c
> @@ -340,6 +340,7 @@ static void test_bpf_obj_id(void)
>
> /* Check getting prog info */
> info_len = sizeof(struct bpf_prog_info) * 2;
> + bzero(&prog_infos[i], info_len);
> prog_infos[i].jited_prog_insns = ptr_to_u64(jited_insns);
> prog_infos[i].jited_prog_len = sizeof(jited_insns);
> prog_infos[i].xlated_prog_insns = ptr_to_u64(xlated_insns);
> @@ -369,6 +370,7 @@ static void test_bpf_obj_id(void)
>
> /* Check getting map info */
> info_len = sizeof(struct bpf_map_info) * 2;
> + bzero(&map_infos[i], info_len);
> err = bpf_obj_get_info_by_fd(map_fds[i], &map_infos[i],
> &info_len);
> if (CHECK(err ||
> @@ -394,7 +396,7 @@ static void test_bpf_obj_id(void)
> nr_id_found = 0;
> next_id = 0;
> while (!bpf_prog_get_next_id(next_id, &next_id)) {
> - struct bpf_prog_info prog_info;
> + struct bpf_prog_info prog_info = {};
> int prog_fd;
>
> info_len = sizeof(prog_info);
> @@ -418,6 +420,8 @@ static void test_bpf_obj_id(void)
> nr_id_found++;
>
> err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
> + prog_infos[i].jited_prog_insns = 0;
> + prog_infos[i].xlated_prog_insns = 0;
Can you elaborate why this one above is needed?
> CHECK(err || info_len != sizeof(struct bpf_prog_info) ||
> memcmp(&prog_info, &prog_infos[i], info_len),
> "get-prog-info(next_id->fd)",
> @@ -436,7 +440,7 @@ static void test_bpf_obj_id(void)
> nr_id_found = 0;
> next_id = 0;
> while (!bpf_map_get_next_id(next_id, &next_id)) {
> - struct bpf_map_info map_info;
> + struct bpf_map_info map_info = {};
> int map_fd;
>
> info_len = sizeof(map_info);
>
^ permalink raw reply
* Re: [PATCH net-next 10/10] bnxt_en: add support for port_attr_get and and get_phys_port_name
From: Jakub Kicinski @ 2017-07-25 22:27 UTC (permalink / raw)
To: Sathya Perla; +Cc: Michael Chan, David Miller, netdev
In-Reply-To: <CAKvpyk3=R-ER74wfwNE3yM19EgU+h2TUdp8epRUKEQ69=kco1g@mail.gmail.com>
On Tue, 25 Jul 2017 15:25:19 +0530, Sathya Perla wrote:
> >> Suggested naming convention is "swXpYsZ", where X is the switch name or ID, Y
> >> is the port name or ID, and Z is the sub-port name or ID. For example, sw1p1s0
> >> would be sub-port 0 on port 1 on switch 1.
> >
> > So for physical ports the convention is p<port_id>, and in case of
> > breakout p<port_id>s<subport_id>.
>
> I'm sending a follow-up patch that fixes the naming for both the
> external port and the VF-reps as proposed in your RFC ( switchdev:
> clarify ndo_get_phys_port_name() formats)
Thanks!
^ permalink raw reply
* Re: [RFC] switchdev: clarify ndo_get_phys_port_name() formats
From: Jakub Kicinski @ 2017-07-25 22:26 UTC (permalink / raw)
To: Andy Gospodarek
Cc: netdev, Jiri Pirko, Or Gerlitz, Michael Chan, Sathya Perla,
simon.horman, davem
In-Reply-To: <20170725152241.GA71091@C02RW35GFVH8.dhcp.broadcom.net>
On Tue, 25 Jul 2017 11:22:41 -0400, Andy Gospodarek wrote:
> On Mon, Jul 24, 2017 at 10:13:44PM -0700, Jakub Kicinski wrote:
> > We are still in position where we can suggest uniform naming
> > convention for ndo_get_phys_port_name(). switchdev.txt file
> > already contained a suggestion of how to name external ports.
> > Since the use of switchdev for SR-IOV NIC's eswitches is growing,
> > establish a format for ports of those devices as well.
> >
> > Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
>
> This is a nice addition and I suspect there could be even more done to
> update this file to cover the VF rep usage.
>
> > ---
> > Documentation/networking/switchdev.txt | 14 +++++++++++---
> > 1 file changed, 11 insertions(+), 3 deletions(-)
> >
> > diff --git a/Documentation/networking/switchdev.txt b/Documentation/networking/switchdev.txt
> > index 3e7b946dea27..7c4b6025fb4b 100644
> > --- a/Documentation/networking/switchdev.txt
> > +++ b/Documentation/networking/switchdev.txt
> > @@ -119,9 +119,17 @@ into 4 10G ports, resulting in 4 port netdevs, the device can give a unique
> > SUBSYSTEM=="net", ACTION=="add", ATTR{phys_switch_id}=="<phys_switch_id>", \
> > ATTR{phys_port_name}!="", NAME="swX$attr{phys_port_name}"
> >
> > -Suggested naming convention is "swXpYsZ", where X is the switch name or ID, Y
> > -is the port name or ID, and Z is the sub-port name or ID. For example, sw1p1s0
> > -would be sub-port 0 on port 1 on switch 1.
> > +Suggested formats of the port name returned by ndo_get_phys_port_name are:
> > + - pA for external ports;
> > + - pAsB for split external ports;
> > + - pfC for PF ports (so called PF representors);
> > + - pfCvfD for VF ports (so called VF representors).
>
> I hate to clutter this up, but might be also need to add:
>
> - pfCsB for split PF ports (so called PF representors);
> - pfCsBvfD for split VF ports (so called VF representors).
>
> or are we comfortable that these additions to the name for split ports
> are implied?
Hm.. What is a split PF port? Splits happen on the physical port - see
my rant on the thread this is a reply to ;) PFs are PCIe functions,
on the opposite side of the eswitch from the wires.
^ permalink raw reply
* [PATCH net] bpf: don't zero out the info struct in bpf_obj_get_info_by_fd()
From: Jakub Kicinski @ 2017-07-25 22:16 UTC (permalink / raw)
To: netdev, daniel; +Cc: oss-drivers, alexei.starovoitov, kafai, Jakub Kicinski
The buffer passed to bpf_obj_get_info_by_fd() should be initialized
to zeros. Kernel will enforce that to guarantee we can safely extend
info structures in the future.
Making the bpf_obj_get_info_by_fd() call in libbpf perform the zeroing
is problematic, however, since some members of the info structures
may need to be initialized by the callers (for instance pointers
to buffers to which kernel is to dump translated and jited images).
Remove the zeroing and fix up the in-tree callers before any kernel
has been released with this code.
As Daniel points out this seems to be the intended operation anyway,
since commit 95b9afd3987f ("bpf: Test for bpf ID") is itself setting
the buffer pointers before calling bpf_obj_get_info_by_fd().
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
I have a small patch to add checking if kernel actually populated
the instruction dumps which I will post after this ends up in net-next.
tools/lib/bpf/bpf.c | 1 -
tools/testing/selftests/bpf/test_progs.c | 8 ++++++--
2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
index 412a7c82995a..256f571f2ab5 100644
--- a/tools/lib/bpf/bpf.c
+++ b/tools/lib/bpf/bpf.c
@@ -314,7 +314,6 @@ int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len)
int err;
bzero(&attr, sizeof(attr));
- bzero(info, *info_len);
attr.info.bpf_fd = prog_fd;
attr.info.info_len = *info_len;
attr.info.info = ptr_to_u64(info);
diff --git a/tools/testing/selftests/bpf/test_progs.c b/tools/testing/selftests/bpf/test_progs.c
index 5855cd3d3d45..1f7dd35551b9 100644
--- a/tools/testing/selftests/bpf/test_progs.c
+++ b/tools/testing/selftests/bpf/test_progs.c
@@ -340,6 +340,7 @@ static void test_bpf_obj_id(void)
/* Check getting prog info */
info_len = sizeof(struct bpf_prog_info) * 2;
+ bzero(&prog_infos[i], info_len);
prog_infos[i].jited_prog_insns = ptr_to_u64(jited_insns);
prog_infos[i].jited_prog_len = sizeof(jited_insns);
prog_infos[i].xlated_prog_insns = ptr_to_u64(xlated_insns);
@@ -369,6 +370,7 @@ static void test_bpf_obj_id(void)
/* Check getting map info */
info_len = sizeof(struct bpf_map_info) * 2;
+ bzero(&map_infos[i], info_len);
err = bpf_obj_get_info_by_fd(map_fds[i], &map_infos[i],
&info_len);
if (CHECK(err ||
@@ -394,7 +396,7 @@ static void test_bpf_obj_id(void)
nr_id_found = 0;
next_id = 0;
while (!bpf_prog_get_next_id(next_id, &next_id)) {
- struct bpf_prog_info prog_info;
+ struct bpf_prog_info prog_info = {};
int prog_fd;
info_len = sizeof(prog_info);
@@ -418,6 +420,8 @@ static void test_bpf_obj_id(void)
nr_id_found++;
err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
+ prog_infos[i].jited_prog_insns = 0;
+ prog_infos[i].xlated_prog_insns = 0;
CHECK(err || info_len != sizeof(struct bpf_prog_info) ||
memcmp(&prog_info, &prog_infos[i], info_len),
"get-prog-info(next_id->fd)",
@@ -436,7 +440,7 @@ static void test_bpf_obj_id(void)
nr_id_found = 0;
next_id = 0;
while (!bpf_map_get_next_id(next_id, &next_id)) {
- struct bpf_map_info map_info;
+ struct bpf_map_info map_info = {};
int map_fd;
info_len = sizeof(map_info);
--
2.11.0
^ permalink raw reply related
* Re: [Patch net] bonding: commit link status change after propose
From: Mahesh Bandewar (महेश बंडेवार) @ 2017-07-25 22:14 UTC (permalink / raw)
To: Cong Wang; +Cc: linux-netdev, benjamin.gilbert
In-Reply-To: <20170725164425.9112-1-xiyou.wangcong@gmail.com>
On Tue, Jul 25, 2017 at 9:44 AM, Cong Wang <xiyou.wangcong@gmail.com> wrote:
> Commit de77ecd4ef02 ("bonding: improve link-status update in mii-monitoring")
> moves link status commitment into bond_mii_monitor(), but it still relies
> on the return value of bond_miimon_inspect() as the hint. We need to return
> non-zero as long as we propose a link status change.
>
> Fixes: de77ecd4ef02 ("bonding: improve link-status update in mii-monitoring")
> Reported-by: Benjamin Gilbert <benjamin.gilbert@coreos.com>
> Tested-by: Benjamin Gilbert <benjamin.gilbert@coreos.com>
> Cc: Mahesh Bandewar <maheshb@google.com>
> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Mahesh Bandewar <maheshb@google.com>
> ---
> drivers/net/bonding/bond_main.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> index 181839d6fbea..9bee6c1c70cc 100644
> --- a/drivers/net/bonding/bond_main.c
> +++ b/drivers/net/bonding/bond_main.c
> @@ -2050,6 +2050,7 @@ static int bond_miimon_inspect(struct bonding *bond)
> continue;
>
> bond_propose_link_state(slave, BOND_LINK_FAIL);
> + commit++;
> slave->delay = bond->params.downdelay;
> if (slave->delay) {
> netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
> @@ -2088,6 +2089,7 @@ static int bond_miimon_inspect(struct bonding *bond)
> continue;
>
> bond_propose_link_state(slave, BOND_LINK_BACK);
> + commit++;
> slave->delay = bond->params.updelay;
>
> if (slave->delay) {
> --
> 2.13.0
>
^ permalink raw reply
* Re: [PATCH] qtfnmac: Tidy up DMA mask setting
From: Sergey Matyukevich @ 2017-07-25 22:00 UTC (permalink / raw)
To: Robin Murphy
Cc: imitsyanko-P/7pdk10T0iB+jHODAdFcQ,
avinashp-P/7pdk10T0iB+jHODAdFcQ,
smatyukevich-P/7pdk10T0iB+jHODAdFcQ, kvalo-sgV2jX0FEOL9JmXXK+q4OQ,
linux-wireless-u79uwXL29TY76Z2rM5mHXA,
netdev-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA
In-Reply-To: <2c3f47792b2e38d5584d641d98e82f8318fefc72.1500917445.git.robin.murphy-5wv7dgnIgG8@public.gmane.org>
> As the only caller of dma_supported() outside of DMA API internals, the
> qtfnmac driver stands out and invites scrutiny. Thankfully, it's not
> being used for evil, but it is entirely redundant, since it open-codes a
> check that the DMA mask setting functions are going to perform anyway.
> In fact, the whole qtnf_pcie_init_dma_mask() function is nothing more
> than a rather long-winded implementation of dma_set_mask_and_coherent(),
> so let's just use that directly.
>
> Signed-off-by: Robin Murphy <robin.murphy-5wv7dgnIgG8@public.gmane.org>
Nice, thanks you !
Acked-by: Sergey Matyukevich <sergey.matyukevich.os-P/7pdk10T0iB+jHODAdFcQ@public.gmane.org>
^ permalink raw reply
* Re: [PATCH V2 net] net: fix tcp reset packet flowlabel for ipv6
From: Shaohua Li @ 2017-07-25 21:35 UTC (permalink / raw)
To: David Miller; +Cc: netdev, Kernel-team, shli, eric.dumazet, flo
In-Reply-To: <20170724.133455.562898067272108474.davem@davemloft.net>
On Mon, Jul 24, 2017 at 01:34:55PM -0700, David Miller wrote:
> From: Shaohua Li <shli@kernel.org>
> Date: Tue, 18 Jul 2017 12:03:37 -0700
>
> > + /* Since this is being sent on the wire obfuscate hash a bit
> > + * to minimize possbility that any useful information to an
> > + * attacker is leaked. Only lower 20 bits are relevant.
> > + */
> > + rol32(hash, 16);
>
> This doesn't help anything at all.
>
> I don't like things that try to give a sense of security (however
> little) but actually don't give any at all.
>
> I'd rather, therefore, that you remove this altogether.
ok, so you suggest removing the rol32() for ip6_make_flowlabel too, right? just
want to make sure.
Thanks,
Shaohua
^ permalink raw reply
* Re: [PATCH V2 net] net: fix tcp reset packet flowlabel for ipv6
From: Shaohua Li @ 2017-07-25 21:32 UTC (permalink / raw)
To: Cong Wang
Cc: David Miller, Linux Kernel Network Developers, kernel-team,
Shaohua Li, Eric Dumazet, flo
In-Reply-To: <CAM_iQpUGff36UigL39zoh7H=nuzh6EvB6bh4G+jSKsE96uw+aA@mail.gmail.com>
On Mon, Jul 24, 2017 at 04:12:53PM -0700, Cong Wang wrote:
> On Mon, Jul 24, 2017 at 1:34 PM, David Miller <davem@davemloft.net> wrote:
> > From: Shaohua Li <shli@kernel.org>
> > Date: Tue, 18 Jul 2017 12:03:37 -0700
> >
> >> + /* Since this is being sent on the wire obfuscate hash a bit
> >> + * to minimize possbility that any useful information to an
> >> + * attacker is leaked. Only lower 20 bits are relevant.
> >> + */
> >> + rol32(hash, 16);
> >
> > This doesn't help anything at all.
>
> I believe the above code is copy-n-pasted from ip6_make_flowlabel()
> (with few adjustments). Don't know why we can't refactor that function
> for reuse.
There are just several lines of code, I really don't want to add adhoc if-else
for fast path.
Thanks,
Shaohua
^ permalink raw reply
* Re: [PATCH net-next 3/5] netvsc: optimize receive completions
From: Stephen Hemminger @ 2017-07-25 21:07 UTC (permalink / raw)
To: kys, haiyangz, sthemmin; +Cc: devel, netdev
In-Reply-To: <20170725200422.13795-4-sthemmin@microsoft.com>
On Tue, 25 Jul 2017 13:04:20 -0700
Stephen Hemminger <stephen@networkplumber.org> wrote:
> Optimize how receive completion ring are managed.
> * Allocate only as many slots as needed for all buffers from host
> * Allocate before setting up sub channel for better error detection
> * Don't need to keep copy of initial receive section message
> * Only needt keep the transaction id, status doesn't matter
> * Precompute the watermark for when receive flushing is needed
> * Replace division with conditional test
> * Replace atomic per-device variable with per-channel check.
> * Handle corner case where receive completion send
> fails if ring buffer to host is full.
>
> Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
I probably pushed the shrinkage too far for now. Hiayang raised the
point that probably need to keep the status value. I dropped storing
it since it was always SUCCESS unless the host sent a bad packet,
or the system was out of skb's.
Will send V2 that keeps the status value for now.
^ permalink raw reply
* [PATCH v2 0/3] net: ethernet: ti: cpts: fix tx timestamping timeout
From: Grygorii Strashko @ 2017-07-25 20:24 UTC (permalink / raw)
To: David S. Miller, netdev, Richard Cochran
Cc: Sekhar Nori, linux-kernel, linux-omap, Wingman Kwok,
Ivan Khoronzhuk, Grygorii Strashko
Patch3 - With the low Ethernet connection speed cpdma notification about packet
processing can be received before CPTS TX timestamp event, which is set
when packet actually left CPSW while cpdma notification is sent when packet
pushed in CPSW fifo. As result, when connection is slow and CPU is fast
enough TX timestamping is not working properly.
Issue was discovered using timestamping tool on am57x boards with Ethernet link
speed forced to 100M and on am335x-evm with Ethernet link speed forced to 10M.
This series fixes it by introducing TX SKB queue to store PTP SKBs for which
Ethernet Transmit Event hasn't been received yet and then re-check this queue
with new Ethernet Transmit Events by scheduling CPTS overflow
work more often until TX SKB queue is not empty.
Patch 1,2 - As CPTS overflow work is time critical task it important to ensure
that its scheduling is not delayed. Unfortunately, There could be significant
delay in CPTS work schedule under high system load and on -RT which could cause
CPTS misbehavior due to internal counter overflow and there is no way to tune
CPTS overflow work execution policy and priority manually. The kthread_worker
can be used instead of workqueues, as it creates separate named kthread for
each worker and its its execution policy and priority can be configured
using chrt tool. Instead of modifying CPTS driver itself it was proposed to
it was proposed to add PTP auxiliary worker to the PHC subsystem [1], so
other drivers can benefit from this feature also.
[1] https://www.spinics.net/lists/netdev/msg445392.html
changes in v2:
- added PTP auxiliary worker to the PHC subsystem
Link on v1:
- https://www.spinics.net/lists/netdev/msg445387.html
Grygorii Strashko (3):
ptp: introduce ptp auxiliary worker
net: ethernet: ti: cpts: convert to use ptp auxiliary worker
net: ethernet: ti: cpts: fix tx timestamping timeout
drivers/net/ethernet/ti/cpts.c | 111 +++++++++++++++++++++++++++++++++------
drivers/net/ethernet/ti/cpts.h | 2 +-
drivers/ptp/ptp_clock.c | 40 ++++++++++++++
drivers/ptp/ptp_private.h | 3 ++
include/linux/ptp_clock_kernel.h | 19 +++++++
5 files changed, 158 insertions(+), 17 deletions(-)
--
2.10.1
^ permalink raw reply
* [PATCH v2 3/3] net: ethernet: ti: cpts: fix tx timestamping timeout
From: Grygorii Strashko @ 2017-07-25 20:24 UTC (permalink / raw)
To: David S. Miller, netdev, Richard Cochran
Cc: Sekhar Nori, linux-kernel, linux-omap, Wingman Kwok,
Ivan Khoronzhuk, Grygorii Strashko
In-Reply-To: <20170725202420.28229-1-grygorii.strashko@ti.com>
With the low speed Ethernet connection CPDMA notification about packet
processing can be received before CPTS TX timestamp event, which is set
when packet actually left CPSW while cpdma notification is sent when packet
pushed in CPSW fifo. As result, when connection is slow and CPU is fast
enough TX timestamping is not working properly.
Fix it, by introducing TX SKB queue to store PTP SKBs for which Ethernet
Transmit Event hasn't been received yet and then re-check this queue
with new Ethernet Transmit Events by scheduling CPTS overflow
work more often (every 1 jiffies) until TX SKB queue is not empty.
Side effect of this change is:
- User space tools require to take into account possible delay in TX
timestamp processing (for example ptp4l works with tx_timestamp_timeout=400
under net traffic and tx_timestamp_timeout=25 in idle).
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
drivers/net/ethernet/ti/cpts.c | 86 ++++++++++++++++++++++++++++++++++++++++--
drivers/net/ethernet/ti/cpts.h | 1 +
2 files changed, 84 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index 3ed438e..c2121d2 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -31,9 +31,18 @@
#include "cpts.h"
+#define CPTS_SKB_TX_WORK_TIMEOUT 1 /* jiffies */
+
+struct cpts_skb_cb_data {
+ unsigned long tmo;
+};
+
#define cpts_read32(c, r) readl_relaxed(&c->reg->r)
#define cpts_write32(c, v, r) writel_relaxed(v, &c->reg->r)
+static int cpts_match(struct sk_buff *skb, unsigned int ptp_class,
+ u16 ts_seqid, u8 ts_msgtype);
+
static int event_expired(struct cpts_event *event)
{
return time_after(jiffies, event->tmo);
@@ -77,6 +86,47 @@ static int cpts_purge_events(struct cpts *cpts)
return removed ? 0 : -1;
}
+static bool cpts_match_tx_ts(struct cpts *cpts, struct cpts_event *event)
+{
+ struct sk_buff *skb, *tmp;
+ u16 seqid;
+ u8 mtype;
+ bool found = false;
+
+ mtype = (event->high >> MESSAGE_TYPE_SHIFT) & MESSAGE_TYPE_MASK;
+ seqid = (event->high >> SEQUENCE_ID_SHIFT) & SEQUENCE_ID_MASK;
+
+ /* no need to grab txq.lock as access is always done under cpts->lock */
+ skb_queue_walk_safe(&cpts->txq, skb, tmp) {
+ struct skb_shared_hwtstamps ssh;
+ unsigned int class = ptp_classify_raw(skb);
+ struct cpts_skb_cb_data *skb_cb =
+ (struct cpts_skb_cb_data *)skb->cb;
+
+ if (cpts_match(skb, class, seqid, mtype)) {
+ u64 ns = timecounter_cyc2time(&cpts->tc, event->low);
+
+ memset(&ssh, 0, sizeof(ssh));
+ ssh.hwtstamp = ns_to_ktime(ns);
+ skb_tstamp_tx(skb, &ssh);
+ found = true;
+ __skb_unlink(skb, &cpts->txq);
+ dev_consume_skb_any(skb);
+ dev_dbg(cpts->dev, "match tx timestamp mtype %u seqid %04x\n",
+ mtype, seqid);
+ } else if (time_after(jiffies, skb_cb->tmo)) {
+ /* timeout any expired skbs over 1s */
+ dev_dbg(cpts->dev,
+ "expiring tx timestamp mtype %u seqid %04x\n",
+ mtype, seqid);
+ __skb_unlink(skb, &cpts->txq);
+ dev_consume_skb_any(skb);
+ }
+ }
+
+ return found;
+}
+
/*
* Returns zero if matching event type was found.
*/
@@ -101,9 +151,15 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
event->low = lo;
type = event_type(event);
switch (type) {
+ case CPTS_EV_TX:
+ if (cpts_match_tx_ts(cpts, event)) {
+ /* if the new event matches an existing skb,
+ * then don't queue it
+ */
+ break;
+ }
case CPTS_EV_PUSH:
case CPTS_EV_RX:
- case CPTS_EV_TX:
list_del_init(&event->list);
list_add_tail(&event->list, &cpts->events);
break;
@@ -229,8 +285,15 @@ static long cpts_overflow_check(struct ptp_clock_info *ptp)
struct cpts *cpts = container_of(ptp, struct cpts, info);
unsigned long delay = cpts->ov_check_period;
struct timespec64 ts;
+ unsigned long flags;
+
+ spin_lock_irqsave(&cpts->lock, flags);
+ ts = ns_to_timespec64(timecounter_read(&cpts->tc));
+
+ if (!skb_queue_empty(&cpts->txq))
+ delay = CPTS_SKB_TX_WORK_TIMEOUT;
+ spin_unlock_irqrestore(&cpts->lock, flags);
- cpts_ptp_gettime(&cpts->info, &ts);
pr_debug("cpts overflow check at %lld.%09lu\n", ts.tv_sec, ts.tv_nsec);
return (long)delay;
}
@@ -301,7 +364,7 @@ static u64 cpts_find_ts(struct cpts *cpts, struct sk_buff *skb, int ev_type)
return 0;
spin_lock_irqsave(&cpts->lock, flags);
- cpts_fifo_read(cpts, CPTS_EV_PUSH);
+ cpts_fifo_read(cpts, -1);
list_for_each_safe(this, next, &cpts->events) {
event = list_entry(this, struct cpts_event, list);
if (event_expired(event)) {
@@ -319,6 +382,19 @@ static u64 cpts_find_ts(struct cpts *cpts, struct sk_buff *skb, int ev_type)
break;
}
}
+
+ if (ev_type == CPTS_EV_TX && !ns) {
+ struct cpts_skb_cb_data *skb_cb =
+ (struct cpts_skb_cb_data *)skb->cb;
+ /* Not found, add frame to queue for processing later.
+ * The periodic FIFO check will handle this.
+ */
+ skb_get(skb);
+ /* get the timestamp for timeouts */
+ skb_cb->tmo = jiffies + msecs_to_jiffies(100);
+ __skb_queue_tail(&cpts->txq, skb);
+ ptp_schedule_worker(cpts->clock, 0);
+ }
spin_unlock_irqrestore(&cpts->lock, flags);
return ns;
@@ -360,6 +436,7 @@ int cpts_register(struct cpts *cpts)
{
int err, i;
+ skb_queue_head_init(&cpts->txq);
INIT_LIST_HEAD(&cpts->events);
INIT_LIST_HEAD(&cpts->pool);
for (i = 0; i < CPTS_MAX_EVENTS; i++)
@@ -400,6 +477,9 @@ void cpts_unregister(struct cpts *cpts)
cpts_write32(cpts, 0, int_enable);
cpts_write32(cpts, 0, control);
+ /* Drop all packet */
+ skb_queue_purge(&cpts->txq);
+
clk_disable(cpts->refclk);
}
EXPORT_SYMBOL_GPL(cpts_unregister);
diff --git a/drivers/net/ethernet/ti/cpts.h b/drivers/net/ethernet/ti/cpts.h
index 586edd9..73d73fa 100644
--- a/drivers/net/ethernet/ti/cpts.h
+++ b/drivers/net/ethernet/ti/cpts.h
@@ -125,6 +125,7 @@ struct cpts {
struct list_head pool;
struct cpts_event pool_data[CPTS_MAX_EVENTS];
unsigned long ov_check_period;
+ struct sk_buff_head txq;
};
void cpts_rx_timestamp(struct cpts *cpts, struct sk_buff *skb);
--
2.10.1
^ permalink raw reply related
* [PATCH v2 2/3] net: ethernet: ti: cpts: convert to use ptp auxiliary worker
From: Grygorii Strashko @ 2017-07-25 20:24 UTC (permalink / raw)
To: David S. Miller, netdev, Richard Cochran
Cc: Sekhar Nori, linux-kernel, linux-omap, Wingman Kwok,
Ivan Khoronzhuk, Grygorii Strashko
In-Reply-To: <20170725202420.28229-1-grygorii.strashko@ti.com>
There could be significant delay in CPTS work schedule under high system
load and on -RT which could cause CPTS misbehavior due to internal counter
overflow. Usage of own kthread_worker allows to avoid such kind of issues
and makes it possible to tune priority of CPTS kthread_worker thread on -RT.
Hence, the CPTS driver is converted to use PTP auxiliary worker as PHC
subsystem implements such functionality in a generic way now.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
drivers/net/ethernet/ti/cpts.c | 27 +++++++++++++--------------
drivers/net/ethernet/ti/cpts.h | 1 -
2 files changed, 13 insertions(+), 15 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index 32279d2..3ed438e 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -224,6 +224,17 @@ static int cpts_ptp_enable(struct ptp_clock_info *ptp,
return -EOPNOTSUPP;
}
+static long cpts_overflow_check(struct ptp_clock_info *ptp)
+{
+ struct cpts *cpts = container_of(ptp, struct cpts, info);
+ unsigned long delay = cpts->ov_check_period;
+ struct timespec64 ts;
+
+ cpts_ptp_gettime(&cpts->info, &ts);
+ pr_debug("cpts overflow check at %lld.%09lu\n", ts.tv_sec, ts.tv_nsec);
+ return (long)delay;
+}
+
static struct ptp_clock_info cpts_info = {
.owner = THIS_MODULE,
.name = "CTPS timer",
@@ -236,18 +247,9 @@ static struct ptp_clock_info cpts_info = {
.gettime64 = cpts_ptp_gettime,
.settime64 = cpts_ptp_settime,
.enable = cpts_ptp_enable,
+ .do_aux_work = cpts_overflow_check,
};
-static void cpts_overflow_check(struct work_struct *work)
-{
- struct timespec64 ts;
- struct cpts *cpts = container_of(work, struct cpts, overflow_work.work);
-
- cpts_ptp_gettime(&cpts->info, &ts);
- pr_debug("cpts overflow check at %lld.%09lu\n", ts.tv_sec, ts.tv_nsec);
- schedule_delayed_work(&cpts->overflow_work, cpts->ov_check_period);
-}
-
static int cpts_match(struct sk_buff *skb, unsigned int ptp_class,
u16 ts_seqid, u8 ts_msgtype)
{
@@ -378,7 +380,7 @@ int cpts_register(struct cpts *cpts)
}
cpts->phc_index = ptp_clock_index(cpts->clock);
- schedule_delayed_work(&cpts->overflow_work, cpts->ov_check_period);
+ ptp_schedule_worker(cpts->clock, cpts->ov_check_period);
return 0;
err_ptp:
@@ -392,8 +394,6 @@ void cpts_unregister(struct cpts *cpts)
if (WARN_ON(!cpts->clock))
return;
- cancel_delayed_work_sync(&cpts->overflow_work);
-
ptp_clock_unregister(cpts->clock);
cpts->clock = NULL;
@@ -476,7 +476,6 @@ struct cpts *cpts_create(struct device *dev, void __iomem *regs,
cpts->dev = dev;
cpts->reg = (struct cpsw_cpts __iomem *)regs;
spin_lock_init(&cpts->lock);
- INIT_DELAYED_WORK(&cpts->overflow_work, cpts_overflow_check);
ret = cpts_of_parse(cpts, node);
if (ret)
diff --git a/drivers/net/ethernet/ti/cpts.h b/drivers/net/ethernet/ti/cpts.h
index 01ea82b..586edd9 100644
--- a/drivers/net/ethernet/ti/cpts.h
+++ b/drivers/net/ethernet/ti/cpts.h
@@ -119,7 +119,6 @@ struct cpts {
u32 cc_mult; /* for the nominal frequency */
struct cyclecounter cc;
struct timecounter tc;
- struct delayed_work overflow_work;
int phc_index;
struct clk *refclk;
struct list_head events;
--
2.10.1
^ permalink raw reply related
* [PATCH v2 1/3] ptp: introduce ptp auxiliary worker
From: Grygorii Strashko @ 2017-07-25 20:24 UTC (permalink / raw)
To: David S. Miller, netdev, Richard Cochran
Cc: Sekhar Nori, linux-kernel, linux-omap, Wingman Kwok,
Ivan Khoronzhuk, Grygorii Strashko
In-Reply-To: <20170725202420.28229-1-grygorii.strashko@ti.com>
Many PTP drivers required to perform some asynchronous or periodic work,
like periodically handling PHC counter overflow or handle delayed timestamp
for RX/TX network packets. In most of the cases, such work is implemented
using workqueues. Unfortunately, Kernel workqueues might introduce
significant delay in work scheduling under high system load and on -RT,
which could cause misbehavior of PTP drivers due to internal counter
overflow, for example, and there is no way to tune its execution policy and
priority manually.
Hence, The kthread_worker can be used instead of workqueues, as it creates
separate named kthread for each worker and its its execution policy and
priority can be configured using chrt tool.
This problem was reported for two drivers TI CPSW CPTS and dp83640, so
instead of modifying each of these driver it was proposed to add PTP
auxiliary worker to the PHC subsystem [1].
The patch adds PTP auxiliary worker in PHC subsystem using kthread_worker
and kthread_delayed_work and introduces two new PHC subsystem APIs:
- long (*do_aux_work)(struct ptp_clock_info *ptp) callback in
ptp_clock_info structure, which driver should assign if it require to
perform asynchronous or periodic work. Driver should return the delay of
the PTP next auxiliary work scheduling time (>=0) or negative value in case
further scheduling is not required.
- int ptp_schedule_worker(struct ptp_clock *ptp, unsigned long delay) which
allows schedule PTP auxiliary work.
The name of kthread_worker thread corresponds PTP PHC device name "ptp%d".
[1] https://www.spinics.net/lists/netdev/msg445392.html
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
drivers/ptp/ptp_clock.c | 40 ++++++++++++++++++++++++++++++++++++++++
drivers/ptp/ptp_private.h | 3 +++
include/linux/ptp_clock_kernel.h | 19 +++++++++++++++++++
3 files changed, 62 insertions(+)
diff --git a/drivers/ptp/ptp_clock.c b/drivers/ptp/ptp_clock.c
index b774357..7c6f3e1 100644
--- a/drivers/ptp/ptp_clock.c
+++ b/drivers/ptp/ptp_clock.c
@@ -28,6 +28,7 @@
#include <linux/slab.h>
#include <linux/syscalls.h>
#include <linux/uaccess.h>
+#include <uapi/linux/sched/types.h>
#include "ptp_private.h"
@@ -184,6 +185,19 @@ static void delete_ptp_clock(struct posix_clock *pc)
kfree(ptp);
}
+static void ptp_aux_kworker(struct kthread_work *work)
+{
+ struct ptp_clock *ptp = container_of(work, struct ptp_clock,
+ aux_work.work);
+ struct ptp_clock_info *info = ptp->info;
+ long delay;
+
+ delay = info->do_aux_work(info);
+
+ if (delay >= 0)
+ kthread_queue_delayed_work(ptp->kworker, &ptp->aux_work, delay);
+}
+
/* public interface */
struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
@@ -217,6 +231,19 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
mutex_init(&ptp->pincfg_mux);
init_waitqueue_head(&ptp->tsev_wq);
+ if (ptp->info->do_aux_work) {
+ char *worker_name = kasprintf(GFP_KERNEL, "ptp%d", ptp->index);
+
+ kthread_init_delayed_work(&ptp->aux_work, ptp_aux_kworker);
+ ptp->kworker = kthread_create_worker(0, worker_name ?
+ worker_name : info->name);
+ if (IS_ERR(ptp->kworker)) {
+ pr_err("failed to create ptp aux_worker task %ld\n",
+ PTR_ERR(ptp->kworker));
+ return ERR_CAST(ptp->kworker);
+ }
+ }
+
err = ptp_populate_pin_groups(ptp);
if (err)
goto no_pin_groups;
@@ -274,6 +301,11 @@ int ptp_clock_unregister(struct ptp_clock *ptp)
ptp->defunct = 1;
wake_up_interruptible(&ptp->tsev_wq);
+ if (ptp->kworker) {
+ kthread_cancel_delayed_work_sync(&ptp->aux_work);
+ kthread_destroy_worker(ptp->kworker);
+ }
+
/* Release the clock's resources. */
if (ptp->pps_source)
pps_unregister_source(ptp->pps_source);
@@ -339,6 +371,14 @@ int ptp_find_pin(struct ptp_clock *ptp,
}
EXPORT_SYMBOL(ptp_find_pin);
+int ptp_schedule_worker(struct ptp_clock *ptp, unsigned long delay)
+{
+ if (!ptp->kworker)
+ return -EOPNOTSUPP;
+ return kthread_mod_delayed_work(ptp->kworker, &ptp->aux_work, delay);
+}
+EXPORT_SYMBOL(ptp_schedule_worker);
+
/* module operations */
static void __exit ptp_exit(void)
diff --git a/drivers/ptp/ptp_private.h b/drivers/ptp/ptp_private.h
index d958889..b86f1bf 100644
--- a/drivers/ptp/ptp_private.h
+++ b/drivers/ptp/ptp_private.h
@@ -22,6 +22,7 @@
#include <linux/cdev.h>
#include <linux/device.h>
+#include <linux/kthread.h>
#include <linux/mutex.h>
#include <linux/posix-clock.h>
#include <linux/ptp_clock.h>
@@ -56,6 +57,8 @@ struct ptp_clock {
struct attribute_group pin_attr_group;
/* 1st entry is a pointer to the real group, 2nd is NULL terminator */
const struct attribute_group *pin_attr_groups[2];
+ struct kthread_worker *kworker;
+ struct kthread_delayed_work aux_work;
};
/*
diff --git a/include/linux/ptp_clock_kernel.h b/include/linux/ptp_clock_kernel.h
index a026bfd..ec86fd2 100644
--- a/include/linux/ptp_clock_kernel.h
+++ b/include/linux/ptp_clock_kernel.h
@@ -98,6 +98,10 @@ struct system_device_crosststamp;
* parameter pin: index of the pin in question.
* parameter func: the desired function to use.
* parameter chan: the function channel index to use.
+ * @do_work: Request driver to perform auxiliary (periodic) operations
+ * Driver should return delay of the next auxiliary work scheduling
+ * time (>=0) or negative value in case further scheduling
+ * is not required.
*
* Drivers should embed their ptp_clock_info within a private
* structure, obtaining a reference to it using container_of().
@@ -126,6 +130,7 @@ struct ptp_clock_info {
struct ptp_clock_request *request, int on);
int (*verify)(struct ptp_clock_info *ptp, unsigned int pin,
enum ptp_pin_function func, unsigned int chan);
+ long (*do_aux_work)(struct ptp_clock_info *ptp);
};
struct ptp_clock;
@@ -211,6 +216,16 @@ extern int ptp_clock_index(struct ptp_clock *ptp);
int ptp_find_pin(struct ptp_clock *ptp,
enum ptp_pin_function func, unsigned int chan);
+/**
+ * ptp_schedule_worker() - schedule ptp auxiliary work
+ *
+ * @ptp: The clock obtained from ptp_clock_register().
+ * @delay: number of jiffies to wait before queuing
+ * See kthread_queue_delayed_work() for more info.
+ */
+
+int ptp_schedule_worker(struct ptp_clock *ptp, unsigned long delay);
+
#else
static inline struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
struct device *parent)
@@ -225,6 +240,10 @@ static inline int ptp_clock_index(struct ptp_clock *ptp)
static inline int ptp_find_pin(struct ptp_clock *ptp,
enum ptp_pin_function func, unsigned int chan)
{ return -1; }
+static inline int ptp_schedule_worker(struct ptp_clock *ptp,
+ unsigned long delay)
+{ return -EOPNOTSUPP; }
+
#endif
#endif
--
2.10.1
^ permalink raw reply related
* [PATCH net-next 5/5] netvsc: allow smaller send/recv buffer size
From: Stephen Hemminger @ 2017-07-25 20:04 UTC (permalink / raw)
To: kys, haiyangz, sthemmin; +Cc: devel, netdev
In-Reply-To: <20170725200422.13795-1-sthemmin@microsoft.com>
The default value of send and receive buffer area for host DMA
is much larger than it needs to be. Experimentation shows that
4M receive and 1M send is sufficient.
Make the size a module parameter so that it can be adjusted
as needed for testing or special needs.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
drivers/net/hyperv/hyperv_net.h | 2 ++
drivers/net/hyperv/netvsc.c | 18 ++++++++++++------
drivers/net/hyperv/netvsc_drv.c | 31 ++++++++++++++++++++++++++++++-
3 files changed, 44 insertions(+), 7 deletions(-)
diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index b0259b12c5ee..40677de411b7 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -148,6 +148,8 @@ struct netvsc_device_info {
unsigned char mac_adr[ETH_ALEN];
int ring_size;
u32 num_chn;
+ u32 recv_buf_size;
+ u32 send_buf_size;
};
enum rndis_device_state {
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index f5d7cae6ba70..0b03ace42ceb 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -482,14 +482,16 @@ static int negotiate_nvsp_ver(struct hv_device *device,
}
static int netvsc_connect_vsp(struct hv_device *device,
- struct netvsc_device *net_device)
+ struct netvsc_device *net_device,
+ const struct netvsc_device_info *device_info)
{
const u32 ver_list[] = {
NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
};
struct nvsp_message *init_packet;
- int ndis_version, i, ret;
+ u32 max_recv_buf_size, ndis_version;
+ int i, ret;
init_packet = &net_device->channel_init_pkt;
@@ -534,10 +536,14 @@ static int netvsc_connect_vsp(struct hv_device *device,
/* Post the big receive buffer to NetVSP */
if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
- net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
+ max_recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
else
- net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
- net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
+ max_recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
+
+ net_device->recv_buf_size = min(max_recv_buf_size,
+ device_info->recv_buf_size);
+ net_device->send_buf_size = min_t(u32, NETVSC_SEND_BUFFER_SIZE,
+ device_info->send_buf_size);
ret = netvsc_init_buf(device, net_device);
@@ -1296,7 +1302,7 @@ struct netvsc_device *netvsc_device_add(struct hv_device *device,
rcu_assign_pointer(net_device_ctx->nvdev, net_device);
/* Connect with the NetVsp */
- ret = netvsc_connect_vsp(device, net_device);
+ ret = netvsc_connect_vsp(device, net_device, device_info);
if (ret != 0) {
netdev_err(ndev,
"unable to connect to NetVSP - %d\n", ret);
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index a04f2efbbc25..4a9562c7db01 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -44,13 +44,23 @@
#include "hyperv_net.h"
-#define RING_SIZE_MIN 64
+#define RING_SIZE_MIN 64
+#define RECV_BUFFER_MIN 16
+#define SEND_BUFFER_MIN 4
#define LINKCHANGE_INT (2 * HZ)
static int ring_size = 128;
module_param(ring_size, int, S_IRUGO);
MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
+static unsigned int recv_buffer_size = (4 * 1024 * 1024) / PAGE_SIZE;
+module_param(recv_buffer_size, uint, S_IRUGO);
+MODULE_PARM_DESC(recv_buffer_size, "Receive buffer size (# of pages)");
+
+static unsigned int send_buffer_size = (1024 * 1024) / PAGE_SIZE;
+module_param(send_buffer_size, uint, S_IRUGO);
+MODULE_PARM_DESC(send_buffer_size, "Send buffer size (# of pages)");
+
static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
NETIF_MSG_LINK | NETIF_MSG_IFUP |
NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
@@ -753,6 +763,8 @@ static int netvsc_set_channels(struct net_device *net,
memset(&device_info, 0, sizeof(device_info));
device_info.num_chn = count;
device_info.ring_size = ring_size;
+ device_info.send_buf_size = nvdev->send_buf_size;
+ device_info.recv_buf_size = nvdev->recv_buf_size;
nvdev = rndis_filter_device_add(dev, &device_info);
if (!IS_ERR(nvdev)) {
@@ -850,6 +862,8 @@ static int netvsc_change_mtu(struct net_device *ndev, int mtu)
memset(&device_info, 0, sizeof(device_info));
device_info.ring_size = ring_size;
device_info.num_chn = nvdev->num_chn;
+ device_info.send_buf_size = nvdev->send_buf_size;
+ device_info.recv_buf_size = nvdev->recv_buf_size;
rndis_filter_device_remove(hdev, nvdev);
@@ -1520,6 +1534,8 @@ static int netvsc_probe(struct hv_device *dev,
memset(&device_info, 0, sizeof(device_info));
device_info.ring_size = ring_size;
device_info.num_chn = VRSS_CHANNEL_DEFAULT;
+ device_info.send_buf_size = send_buffer_size * PAGE_SIZE;
+ device_info.recv_buf_size = recv_buffer_size * PAGE_SIZE;
nvdev = rndis_filter_device_add(dev, &device_info);
if (IS_ERR(nvdev)) {
@@ -1671,6 +1687,19 @@ static int __init netvsc_drv_init(void)
pr_info("Increased ring_size to %d (min allowed)\n",
ring_size);
}
+
+ if (recv_buffer_size < RECV_BUFFER_MIN) {
+ recv_buffer_size = RECV_BUFFER_MIN;
+ pr_notice("Increased receive buffer size to %u (min allowed)\n",
+ recv_buffer_size);
+ }
+
+ if (send_buffer_size < SEND_BUFFER_MIN) {
+ send_buffer_size = SEND_BUFFER_MIN;
+ pr_notice("Increased receive buffer size to %u (min allowed)\n",
+ send_buffer_size);
+ }
+
ret = vmbus_driver_register(&netvsc_drv);
if (ret)
--
2.11.0
^ permalink raw reply related
* [PATCH net-next 4/5] netvsc: signal host if receive ring is emptied
From: Stephen Hemminger @ 2017-07-25 20:04 UTC (permalink / raw)
To: kys, haiyangz, sthemmin; +Cc: devel, netdev
In-Reply-To: <20170725200422.13795-1-sthemmin@microsoft.com>
Latency improvement related to NAPI conversion.
If all packets are processed from receive ring then need
to signal host.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
drivers/net/hyperv/netvsc.c | 11 ++++++++---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 03c1fec762c7..f5d7cae6ba70 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -1189,10 +1189,15 @@ int netvsc_poll(struct napi_struct *napi, int budget)
nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
}
- /* If send of pending receive completions suceeded
- * and did not exhaust NAPI budget
+ /* if ring is empty, signal host */
+ if (!nvchan->desc)
+ hv_pkt_iter_close(channel);
+
+ /* If send of pending receive completions suceeded
+ * and did not exhaust NAPI budget this time
* and not doing busy poll
- * then reschedule if more data has arrived from host
+ * then re-enable host interrupts
+ * and reschedule if ring is not empty.
*/
if (send_recv_completions(net_device, channel, q_idx) == 0 &&
work_done < budget &&
--
2.11.0
^ permalink raw reply related
* [PATCH net-next 3/5] netvsc: optimize receive completions
From: Stephen Hemminger @ 2017-07-25 20:04 UTC (permalink / raw)
To: kys, haiyangz, sthemmin; +Cc: devel, netdev
In-Reply-To: <20170725200422.13795-1-sthemmin@microsoft.com>
Optimize how receive completion ring are managed.
* Allocate only as many slots as needed for all buffers from host
* Allocate before setting up sub channel for better error detection
* Don't need to keep copy of initial receive section message
* Only needt keep the transaction id, status doesn't matter
* Precompute the watermark for when receive flushing is needed
* Replace division with conditional test
* Replace atomic per-device variable with per-channel check.
* Handle corner case where receive completion send
fails if ring buffer to host is full.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
drivers/net/hyperv/hyperv_net.h | 19 +--
drivers/net/hyperv/netvsc.c | 270 +++++++++++++++-----------------------
drivers/net/hyperv/rndis_filter.c | 20 +--
3 files changed, 125 insertions(+), 184 deletions(-)
diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index fb62ea632914..b0259b12c5ee 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -186,6 +186,7 @@ struct net_device_context;
struct netvsc_device *netvsc_device_add(struct hv_device *device,
const struct netvsc_device_info *info);
+int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx);
void netvsc_device_remove(struct hv_device *device);
int netvsc_send(struct net_device_context *ndc,
struct hv_netvsc_packet *packet,
@@ -652,18 +653,10 @@ struct multi_send_data {
u32 count; /* counter of batched packets */
};
-struct recv_comp_data {
- u64 tid; /* transaction id */
- u32 status;
-};
-
-/* Netvsc Receive Slots Max */
-#define NETVSC_RECVSLOT_MAX (NETVSC_RECEIVE_BUFFER_SIZE / ETH_DATA_LEN + 1)
-
struct multi_recv_comp {
- void *buf; /* queued receive completions */
- u32 first; /* first data entry */
- u32 next; /* next entry for writing */
+ u64 *slots; /* pending completion transactions */
+ u32 first; /* first data entry */
+ u32 next; /* next entry for writing */
};
struct netvsc_stats {
@@ -750,7 +743,7 @@ struct netvsc_device {
u32 recv_buf_size;
u32 recv_buf_gpadl_handle;
u32 recv_section_cnt;
- struct nvsp_1_receive_buffer_section *recv_section;
+ u32 recv_completion_cnt;
/* Send buffer allocated by us */
void *send_buf;
@@ -778,8 +771,6 @@ struct netvsc_device {
u32 max_pkt; /* max number of pkt in one send, e.g. 8 */
u32 pkt_align; /* alignment bytes, e.g. 8 */
- atomic_t num_outstanding_recvs;
-
atomic_t open_cnt;
struct netvsc_channel chan_table[VRSS_CHANNEL_MAX];
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 94c00acac58a..03c1fec762c7 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -72,9 +72,6 @@ static struct netvsc_device *alloc_net_device(void)
if (!net_device)
return NULL;
- net_device->chan_table[0].mrc.buf
- = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
-
init_waitqueue_head(&net_device->wait_drain);
net_device->destroy = false;
atomic_set(&net_device->open_cnt, 0);
@@ -92,7 +89,7 @@ static void free_netvsc_device(struct rcu_head *head)
int i;
for (i = 0; i < VRSS_CHANNEL_MAX; i++)
- vfree(nvdev->chan_table[i].mrc.buf);
+ vfree(nvdev->chan_table[i].mrc.slots);
kfree(nvdev);
}
@@ -171,12 +168,6 @@ static void netvsc_destroy_buf(struct hv_device *device)
net_device->recv_buf = NULL;
}
- if (net_device->recv_section) {
- net_device->recv_section_cnt = 0;
- kfree(net_device->recv_section);
- net_device->recv_section = NULL;
- }
-
/* Deal with the send buffer we may have setup.
* If we got a send section size, it means we received a
* NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
@@ -239,11 +230,26 @@ static void netvsc_destroy_buf(struct hv_device *device)
kfree(net_device->send_section_map);
}
+int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
+{
+ struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
+ int node = cpu_to_node(nvchan->channel->target_cpu);
+ size_t size;
+
+ size = net_device->recv_completion_cnt * sizeof(u64);
+ nvchan->mrc.slots = vzalloc_node(size, node);
+ if (!nvchan->mrc.slots)
+ nvchan->mrc.slots = vzalloc(size);
+
+ return nvchan->mrc.slots ? 0 : -ENOMEM;
+}
+
static int netvsc_init_buf(struct hv_device *device,
struct netvsc_device *net_device)
{
int ret = 0;
struct nvsp_message *init_packet;
+ struct nvsp_1_message_send_receive_buffer_complete *resp;
struct net_device *ndev;
size_t map_words;
int node;
@@ -300,43 +306,41 @@ static int netvsc_init_buf(struct hv_device *device,
wait_for_completion(&net_device->channel_init_wait);
/* Check the response */
- if (init_packet->msg.v1_msg.
- send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
- netdev_err(ndev, "Unable to complete receive buffer "
- "initialization with NetVsp - status %d\n",
- init_packet->msg.v1_msg.
- send_recv_buf_complete.status);
+ resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
+ if (resp->status != NVSP_STAT_SUCCESS) {
+ netdev_err(ndev,
+ "Unable to complete receive buffer initialization with NetVsp - status %d\n",
+ resp->status);
ret = -EINVAL;
goto cleanup;
}
/* Parse the response */
+ netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
+ resp->num_sections, resp->sections[0].sub_alloc_size,
+ resp->sections[0].num_sub_allocs);
- net_device->recv_section_cnt = init_packet->msg.
- v1_msg.send_recv_buf_complete.num_sections;
-
- net_device->recv_section = kmemdup(
- init_packet->msg.v1_msg.send_recv_buf_complete.sections,
- net_device->recv_section_cnt *
- sizeof(struct nvsp_1_receive_buffer_section),
- GFP_KERNEL);
- if (net_device->recv_section == NULL) {
- ret = -EINVAL;
- goto cleanup;
- }
+ net_device->recv_section_cnt = resp->num_sections;
/*
* For 1st release, there should only be 1 section that represents the
* entire receive buffer
*/
if (net_device->recv_section_cnt != 1 ||
- net_device->recv_section->offset != 0) {
+ resp->sections[0].offset != 0) {
ret = -EINVAL;
goto cleanup;
}
- /* Now setup the send buffer.
- */
+ /* Setup receive completion ring */
+ net_device->recv_completion_cnt
+ = round_up(resp->sections[0].num_sub_allocs + 1,
+ PAGE_SIZE / sizeof(u64));
+ ret = netvsc_alloc_recv_comp_ring(net_device, 0);
+ if (ret)
+ goto cleanup;
+
+ /* Now setup the send buffer. */
net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
if (!net_device->send_buf)
net_device->send_buf = vzalloc(net_device->send_buf_size);
@@ -950,139 +954,96 @@ int netvsc_send(struct net_device_context *ndev_ctx,
return ret;
}
-static int netvsc_send_recv_completion(struct vmbus_channel *channel,
- u64 transaction_id, u32 status)
-{
- struct nvsp_message recvcompMessage;
- int ret;
-
- recvcompMessage.hdr.msg_type =
- NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
-
- recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
-
- /* Send the completion */
- ret = vmbus_sendpacket(channel, &recvcompMessage,
- sizeof(struct nvsp_message_header) + sizeof(u32),
- transaction_id, VM_PKT_COMP, 0);
-
- return ret;
-}
-
-static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
- u32 *filled, u32 *avail)
+/* Send pending recv completions */
+static int send_recv_completions(struct netvsc_device *nvdev,
+ struct vmbus_channel *channel, u16 q_idx)
{
struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
- u32 first = mrc->first;
- u32 next = mrc->next;
+ struct recv_comp_msg {
+ struct nvsp_message_header hdr;
+ u32 status;
+ } __packed;
+ struct recv_comp_msg msg = {
+ .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
+ .status = NVSP_STAT_SUCCESS,
+ };
+ int ret;
- *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
- next - first;
+ while (mrc->first != mrc->next) {
+ u64 tid = mrc->slots[mrc->first];
- *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
-}
+ ret = vmbus_sendpacket(channel, &msg, sizeof(msg),
+ tid, VM_PKT_COMP, 0);
+ if (unlikely(ret))
+ return ret;
-/* Read the first filled slot, no change to index */
-static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
- *nvdev, u16 q_idx)
-{
- struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
- u32 filled, avail;
-
- if (unlikely(!mrc->buf))
- return NULL;
+ if (++mrc->first == nvdev->recv_completion_cnt)
+ mrc->first = 0;
+ }
- count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
- if (!filled)
- return NULL;
+ /* receive completion ring has been emptied */
+ if (unlikely(nvdev->destroy))
+ wake_up(&nvdev->wait_drain);
- return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
+ return 0;
}
-/* Put the first filled slot back to available pool */
-static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
+/* Count how many receive completions are outstanding */
+static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
+ const struct multi_recv_comp *mrc,
+ u32 *filled, u32 *avail)
{
- struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
- int num_recv;
+ u32 count = nvdev->recv_completion_cnt;
- mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
-
- num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
+ if (mrc->next >= mrc->first)
+ *filled = mrc->next - mrc->first;
+ else
+ *filled = (count - mrc->first) + mrc->next;
- if (nvdev->destroy && num_recv == 0)
- wake_up(&nvdev->wait_drain);
+ *avail = count - *filled - 1;
}
-/* Check and send pending recv completions */
-static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
- struct vmbus_channel *channel, u16 q_idx)
+/* Add receive complete to ring to send to host. */
+static void enq_receive_complete(struct net_device *ndev,
+ struct netvsc_device *nvdev, u16 q_idx,
+ u64 tid)
{
- struct recv_comp_data *rcd;
- int ret;
-
- while (true) {
- rcd = read_recv_comp_slot(nvdev, q_idx);
- if (!rcd)
- break;
+ struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
+ struct multi_recv_comp *mrc = &nvchan->mrc;
+ u32 filled, avail;
- ret = netvsc_send_recv_completion(channel, rcd->tid,
- rcd->status);
- if (ret)
- break;
+ recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
- put_recv_comp_slot(nvdev, q_idx);
+ if (unlikely(filled > NAPI_POLL_WEIGHT)) {
+ send_recv_completions(nvdev, nvchan->channel, q_idx);
+ recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
}
-}
-
-#define NETVSC_RCD_WATERMARK 80
-
-/* Get next available slot */
-static inline struct recv_comp_data *get_recv_comp_slot(
- struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
-{
- struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
- u32 filled, avail, next;
- struct recv_comp_data *rcd;
-
- if (unlikely(!nvdev->recv_section))
- return NULL;
-
- if (unlikely(!mrc->buf))
- return NULL;
-
- if (atomic_read(&nvdev->num_outstanding_recvs) >
- nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
- netvsc_chk_recv_comp(nvdev, channel, q_idx);
-
- count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
- if (!avail)
- return NULL;
-
- next = mrc->next;
- rcd = mrc->buf + next * sizeof(struct recv_comp_data);
- mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
- atomic_inc(&nvdev->num_outstanding_recvs);
+ if (unlikely(!avail)) {
+ netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
+ q_idx, tid);
+ return;
+ }
- return rcd;
+ mrc->slots[mrc->next] = tid;
+ if (++mrc->next == nvdev->recv_completion_cnt)
+ mrc->next = 0;
}
static int netvsc_receive(struct net_device *ndev,
- struct netvsc_device *net_device,
- struct net_device_context *net_device_ctx,
- struct hv_device *device,
- struct vmbus_channel *channel,
- const struct vmpacket_descriptor *desc,
- struct nvsp_message *nvsp)
+ struct netvsc_device *net_device,
+ struct net_device_context *net_device_ctx,
+ struct hv_device *device,
+ struct vmbus_channel *channel,
+ const struct vmpacket_descriptor *desc,
+ struct nvsp_message *nvsp)
{
const struct vmtransfer_page_packet_header *vmxferpage_packet
= container_of(desc, const struct vmtransfer_page_packet_header, d);
u16 q_idx = channel->offermsg.offer.sub_channel_index;
char *recv_buf = net_device->recv_buf;
- u32 status = NVSP_STAT_SUCCESS;
int i;
int count = 0;
- int ret;
/* Make sure this is a valid nvsp packet */
if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
@@ -1109,29 +1070,13 @@ static int netvsc_receive(struct net_device *ndev,
u32 buflen = vmxferpage_packet->ranges[i].byte_count;
/* Pass it to the upper layer */
- status = rndis_filter_receive(ndev, net_device, device,
- channel, data, buflen);
+ rndis_filter_receive(ndev, net_device, device,
+ channel, data, buflen);
}
- if (net_device->chan_table[q_idx].mrc.buf) {
- struct recv_comp_data *rcd;
+ enq_receive_complete(ndev, net_device, q_idx,
+ vmxferpage_packet->d.trans_id);
- rcd = get_recv_comp_slot(net_device, channel, q_idx);
- if (rcd) {
- rcd->tid = vmxferpage_packet->d.trans_id;
- rcd->status = status;
- } else {
- netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
- q_idx, vmxferpage_packet->d.trans_id);
- }
- } else {
- ret = netvsc_send_recv_completion(channel,
- vmxferpage_packet->d.trans_id,
- status);
- if (ret)
- netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
- q_idx, vmxferpage_packet->d.trans_id, ret);
- }
return count;
}
@@ -1244,17 +1189,18 @@ int netvsc_poll(struct napi_struct *napi, int budget)
nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
}
- /* If receive ring was exhausted
- * and not doing busy poll
- * then re-enable host interrupts
- * and reschedule if ring is not empty.
+ /* If send of pending receive completions suceeded
+ * and did not exhaust NAPI budget
+ * and not doing busy poll
+ * then reschedule if more data has arrived from host
*/
- if (work_done < budget &&
+ if (send_recv_completions(net_device, channel, q_idx) == 0 &&
+ work_done < budget &&
napi_complete_done(napi, work_done) &&
- hv_end_read(&channel->inbound) != 0)
+ hv_end_read(&channel->inbound)) {
+ hv_begin_read(&channel->inbound);
napi_reschedule(napi);
-
- netvsc_chk_recv_comp(net_device, channel, q_idx);
+ }
/* Driver may overshoot since multiple packets per descriptor */
return min(work_done, budget);
diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index bf21ea92c743..db9bac31dd09 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -928,12 +928,12 @@ static bool netvsc_device_idle(const struct netvsc_device *nvdev)
{
int i;
- if (atomic_read(&nvdev->num_outstanding_recvs) > 0)
- return false;
-
for (i = 0; i < nvdev->num_chn; i++) {
const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
+ if (nvchan->mrc.first != nvchan->mrc.next)
+ return false;
+
if (atomic_read(&nvchan->queue_sends) > 0)
return false;
}
@@ -1031,11 +1031,6 @@ static void netvsc_sc_open(struct vmbus_channel *new_sc)
return;
nvchan = nvscdev->chan_table + chn_index;
- nvchan->mrc.buf
- = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
-
- if (!nvchan->mrc.buf)
- return;
/* Because the device uses NAPI, all the interrupt batching and
* control is done via Net softirq, not the channel handling
@@ -1225,6 +1220,15 @@ struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
if (num_rss_qs == 0)
return net_device;
+ for (i = 1; i < net_device->num_chn; i++) {
+ ret = netvsc_alloc_recv_comp_ring(net_device, i);
+ if (ret) {
+ while (--i != 0)
+ vfree(net_device->chan_table[i].mrc.slots);
+ goto out;
+ }
+ }
+
refcount_set(&net_device->sc_offered, num_rss_qs);
vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
--
2.11.0
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
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).