* [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel
@ 2024-06-21 10:13 Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 1/5] linux/dim: move useful macros to .h file Heng Qi
` (5 more replies)
0 siblings, 6 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki
The NetDIM library provides excellent acceleration for many modern
network cards. However, the default profiles of DIM limits its maximum
capabilities for different NICs, so providing a way which the NIC can
be custom configured is necessary.
Currently, the way is based on the commonly used "ethtool -C".
For example,
on the server side, the virtio-net NIC with rx dim enabled has 8
queues and runs nginx.
The client uses the following command to send traffic to the server:
./wrk http://server_ip:80 -c 64 -t 5 -d 30
Then adjust the default rx-profile for server dim to
{.usec = 1, .pkts = 256, .comps = n/a,},
{.usec = 8, .pkts = 256, .comps = n/a,},
{.usec = 30, .pkts = 256, .comps = n/a,},
{.usec = 64, .pkts = 256, .comps = n/a,},
{.usec = 128, .pkts = 256, .comps = n/a,}
The server PPS is improved by 20%+.
Please review, thank you very much!
Changelog
=====
v14->v15:
- Modify the mod bit and add some hints.
Jakub feedback: Use RESEND to refresh the review queue.
v13->v14:
- Make DIMLIB dependent on NET (patch 2/5).
v12->v13:
- Rebase net-next to fix the one-line conflict.
- Update tiny comments.
- Config ETHTOOL_NETLINK to select DIMLIB.
v11->v12:
- Remove the use of IS_ENABLED(DIMLIB).
- Update Simon's htmldoc hint.
v10->v11:
- Fix and clean up some issues from Kuba, thanks.
- Rebase net-next/main
v9->v10:
- Collect dim related flags/mode/work into one place.
- Use rx_profile + tx_profile instead of four profiles.
- Add several helps.
- Update commit logs.
v8->v9:
- Fix the compilation error of conflicting names of rx_profile in
dim.h and ice driver: in dim.h, rx_profile is replaced with
dim_rx_profile. So does tx_profile.
v7->v8:
- Use kmemdup() instead of kzalloc()/memcpy() in dev_dim_profile_init().
v6->v7:
- A new wrapper struct pointer is used in struct net_device.
- Add IS_ENABLED(CONFIG_DIMLIB) to avoid compiler warnings.
- Profile fields changed from u16 to u32.
v5->v6:
- Place the profile in netdevice to bypass the driver.
The interaction code of ethtool <-> kernel has not changed at all,
only the interaction part of kernel <-> driver has changed.
v4->v5:
- Update some snippets from Kuba.
v3->v4:
- Some tiny updates and patch 1 only add a new comment.
v2->v3:
- Break up the attributes to avoid the use of raw c structs.
- Use per-device profile instead of global profile in the driver.
v1->v2:
- Use ethtool tool instead of net-sysfs.
Heng Qi (5):
linux/dim: move useful macros to .h file
dim: make DIMLIB dependent on NET
ethtool: provide customized dim profile management
dim: add new interfaces for initialization and getting results
virtio-net: support dim profile fine-tuning
Documentation/netlink/specs/ethtool.yaml | 31 +++
Documentation/networking/ethtool-netlink.rst | 8 +
Documentation/networking/net_dim.rst | 42 +++
drivers/net/virtio_net.c | 54 +++-
drivers/soc/fsl/Kconfig | 2 +-
include/linux/dim.h | 113 ++++++++
include/linux/ethtool.h | 4 +-
include/linux/netdevice.h | 3 +
include/uapi/linux/ethtool_netlink.h | 22 ++
lib/Kconfig | 1 +
lib/dim/net_dim.c | 144 +++++++++-
net/Kconfig | 1 +
net/ethtool/coalesce.c | 273 ++++++++++++++++++-
13 files changed, 681 insertions(+), 17 deletions(-)
--
2.32.0.3.g01195cf9f
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH net-next v15 1/5] linux/dim: move useful macros to .h file
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
@ 2024-06-21 10:13 ` Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 2/5] dim: make DIMLIB dependent on NET Heng Qi
` (4 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki, Simon Horman
Useful macros will be used effectively elsewhere.
These will be utilized in subsequent patches.
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Simon Horman <horms@kernel.org>
---
include/linux/dim.h | 7 +++++++
lib/dim/net_dim.c | 6 ------
2 files changed, 7 insertions(+), 6 deletions(-)
diff --git a/include/linux/dim.h b/include/linux/dim.h
index f343bc9aa2ec..43398f5eade2 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -10,6 +10,13 @@
#include <linux/types.h>
#include <linux/workqueue.h>
+/* Number of DIM profiles and period mode. */
+#define NET_DIM_PARAMS_NUM_PROFILES 5
+#define NET_DIM_DEFAULT_RX_CQ_PKTS_FROM_EQE 256
+#define NET_DIM_DEFAULT_TX_CQ_PKTS_FROM_EQE 128
+#define NET_DIM_DEF_PROFILE_CQE 1
+#define NET_DIM_DEF_PROFILE_EQE 1
+
/*
* Number of events between DIM iterations.
* Causes a moderation of the algorithm run.
diff --git a/lib/dim/net_dim.c b/lib/dim/net_dim.c
index 4e32f7aaac86..67d5beb34dc3 100644
--- a/lib/dim/net_dim.c
+++ b/lib/dim/net_dim.c
@@ -11,12 +11,6 @@
* There are different set of profiles for RX/TX CQs.
* Each profile size must be of NET_DIM_PARAMS_NUM_PROFILES
*/
-#define NET_DIM_PARAMS_NUM_PROFILES 5
-#define NET_DIM_DEFAULT_RX_CQ_PKTS_FROM_EQE 256
-#define NET_DIM_DEFAULT_TX_CQ_PKTS_FROM_EQE 128
-#define NET_DIM_DEF_PROFILE_CQE 1
-#define NET_DIM_DEF_PROFILE_EQE 1
-
#define NET_DIM_RX_EQE_PROFILES { \
{.usec = 1, .pkts = NET_DIM_DEFAULT_RX_CQ_PKTS_FROM_EQE,}, \
{.usec = 8, .pkts = NET_DIM_DEFAULT_RX_CQ_PKTS_FROM_EQE,}, \
--
2.32.0.3.g01195cf9f
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH net-next v15 2/5] dim: make DIMLIB dependent on NET
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 1/5] linux/dim: move useful macros to .h file Heng Qi
@ 2024-06-21 10:13 ` Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 3/5] ethtool: provide customized dim profile management Heng Qi
` (3 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki, Simon Horman
DIMLIB's capabilities are supplied by the dim, net_dim, and
rdma_dim objects, and dim's interfaces solely act as a base for
net_dim and rdma_dim and are not explicitly used anywhere else.
rdma_dim is utilized by the infiniband driver, while net_dim
is for network devices, excluding the soc/fsl driver.
In this patch, net_dim relies on some NET's interfaces, thus
DIMLIB needs to explicitly depend on the NET Kconfig.
The soc/fsl driver uses the functions provided by net_dim, so
it also needs to depend on NET.
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Simon Horman <horms@kernel.org>
---
drivers/soc/fsl/Kconfig | 2 +-
lib/Kconfig | 1 +
2 files changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/soc/fsl/Kconfig b/drivers/soc/fsl/Kconfig
index fcec6ed83d5e..a1e0bc8c1757 100644
--- a/drivers/soc/fsl/Kconfig
+++ b/drivers/soc/fsl/Kconfig
@@ -22,7 +22,7 @@ config FSL_GUTS
config FSL_MC_DPIO
tristate "QorIQ DPAA2 DPIO driver"
- depends on FSL_MC_BUS
+ depends on FSL_MC_BUS && NET
select SOC_BUS
select FSL_GUTS
select DIMLIB
diff --git a/lib/Kconfig b/lib/Kconfig
index b0a76dff5c18..b38849af6f13 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -623,6 +623,7 @@ config SIGNATURE
config DIMLIB
tristate
+ depends on NET
help
Dynamic Interrupt Moderation library.
Implements an algorithm for dynamically changing CQ moderation values
--
2.32.0.3.g01195cf9f
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH net-next v15 3/5] ethtool: provide customized dim profile management
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 1/5] linux/dim: move useful macros to .h file Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 2/5] dim: make DIMLIB dependent on NET Heng Qi
@ 2024-06-21 10:13 ` Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 4/5] dim: add new interfaces for initialization and getting results Heng Qi
` (2 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki, Simon Horman
The NetDIM library, currently leveraged by an array of NICs, delivers
excellent acceleration benefits. Nevertheless, NICs vary significantly
in their dim profile list prerequisites.
Specifically, virtio-net backends may present diverse sw or hw device
implementation, making a one-size-fits-all parameter list impractical.
On Alibaba Cloud, the virtio DPU's performance under the default DIM
profile falls short of expectations, partly due to a mismatch in
parameter configuration.
I also noticed that ice/idpf/ena and other NICs have customized
profilelist or placed some restrictions on dim capabilities.
Motivated by this, I tried adding new params for "ethtool -C" that provides
a per-device control to modify and access a device's interrupt parameters.
Usage
========
The target NIC is named ethx.
Assume that ethx only declares support for rx profile setting
(with DIM_PROFILE_RX flag set in profile_flags) and supports modification
of usec and pkt fields.
1. Query the currently customized list of the device
$ ethtool -c ethx
...
rx-profile:
{.usec = 1, .pkts = 256, .comps = n/a,},
{.usec = 8, .pkts = 256, .comps = n/a,},
{.usec = 64, .pkts = 256, .comps = n/a,},
{.usec = 128, .pkts = 256, .comps = n/a,},
{.usec = 256, .pkts = 256, .comps = n/a,}
tx-profile: n/a
2. Tune
$ ethtool -C ethx rx-profile 1,1,n_2,n,n_3,3,n_4,4,n_n,5,n
"n" means do not modify this field.
$ ethtool -c ethx
...
rx-profile:
{.usec = 1, .pkts = 1, .comps = n/a,},
{.usec = 2, .pkts = 256, .comps = n/a,},
{.usec = 3, .pkts = 3, .comps = n/a,},
{.usec = 4, .pkts = 4, .comps = n/a,},
{.usec = 256, .pkts = 5, .comps = n/a,}
tx-profile: n/a
3. Hint
If the device does not support some type of customized dim profiles,
the corresponding "n/a" will display.
If the "n/a" field is being modified, -EOPNOTSUPP will be reported.
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Simon Horman <horms@kernel.org>
---
Documentation/netlink/specs/ethtool.yaml | 31 +++
Documentation/networking/ethtool-netlink.rst | 8 +
Documentation/networking/net_dim.rst | 42 +++
include/linux/dim.h | 58 ++++
include/linux/ethtool.h | 4 +-
include/linux/netdevice.h | 3 +
include/uapi/linux/ethtool_netlink.h | 22 ++
lib/dim/net_dim.c | 70 +++++
net/Kconfig | 1 +
net/ethtool/coalesce.c | 273 ++++++++++++++++++-
10 files changed, 509 insertions(+), 3 deletions(-)
diff --git a/Documentation/netlink/specs/ethtool.yaml b/Documentation/netlink/specs/ethtool.yaml
index 00dc61358be8..6c2ab3d1c22f 100644
--- a/Documentation/netlink/specs/ethtool.yaml
+++ b/Documentation/netlink/specs/ethtool.yaml
@@ -414,6 +414,26 @@ attribute-sets:
name: combined-count
type: u32
+ -
+ name: irq-moderation
+ attributes:
+ -
+ name: usec
+ type: u32
+ -
+ name: pkts
+ type: u32
+ -
+ name: comps
+ type: u32
+ -
+ name: profile
+ attributes:
+ -
+ name: irq-moderation
+ type: nest
+ multi-attr: true
+ nested-attributes: irq-moderation
-
name: coalesce
attributes:
@@ -502,6 +522,15 @@ attribute-sets:
-
name: tx-aggr-time-usecs
type: u32
+ -
+ name: rx-profile
+ type: nest
+ nested-attributes: profile
+ -
+ name: tx-profile
+ type: nest
+ nested-attributes: profile
+
-
name: pause-stat
attributes:
@@ -1325,6 +1354,8 @@ operations:
- tx-aggr-max-bytes
- tx-aggr-max-frames
- tx-aggr-time-usecs
+ - rx-profile
+ - tx-profile
dump: *coalesce-get-op
-
name: coalesce-set
diff --git a/Documentation/networking/ethtool-netlink.rst b/Documentation/networking/ethtool-netlink.rst
index 160bfb0ae8ba..7ec08e903bab 100644
--- a/Documentation/networking/ethtool-netlink.rst
+++ b/Documentation/networking/ethtool-netlink.rst
@@ -1033,6 +1033,8 @@ Kernel response contents:
``ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES`` u32 max aggr size, Tx
``ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES`` u32 max aggr packets, Tx
``ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS`` u32 time (us), aggr, Tx
+ ``ETHTOOL_A_COALESCE_RX_PROFILE`` nested profile of DIM, Rx
+ ``ETHTOOL_A_COALESCE_TX_PROFILE`` nested profile of DIM, Tx
=========================================== ====== =======================
Attributes are only included in reply if their value is not zero or the
@@ -1062,6 +1064,10 @@ block should be sent.
This feature is mainly of interest for specific USB devices which does not cope
well with frequent small-sized URBs transmissions.
+``ETHTOOL_A_COALESCE_RX_PROFILE`` and ``ETHTOOL_A_COALESCE_TX_PROFILE`` refer
+to DIM parameters, see `Generic Network Dynamic Interrupt Moderation (Net DIM)
+<https://www.kernel.org/doc/Documentation/networking/net_dim.rst>`_.
+
COALESCE_SET
============
@@ -1098,6 +1104,8 @@ Request contents:
``ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES`` u32 max aggr size, Tx
``ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES`` u32 max aggr packets, Tx
``ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS`` u32 time (us), aggr, Tx
+ ``ETHTOOL_A_COALESCE_RX_PROFILE`` nested profile of DIM, Rx
+ ``ETHTOOL_A_COALESCE_TX_PROFILE`` nested profile of DIM, Tx
=========================================== ====== =======================
Request is rejected if it attributes declared as unsupported by driver (i.e.
diff --git a/Documentation/networking/net_dim.rst b/Documentation/networking/net_dim.rst
index 3bed9fd95336..8908fd7b0a8d 100644
--- a/Documentation/networking/net_dim.rst
+++ b/Documentation/networking/net_dim.rst
@@ -169,6 +169,48 @@ usage is not complete but it should make the outline of the usage clear.
...
}
+
+Tuning DIM
+==========
+
+Net DIM serves a range of network devices and delivers excellent acceleration
+benefits. Yet, it has been observed that some preset configurations of DIM may
+not align seamlessly with the varying specifications of network devices, and
+this discrepancy has been identified as a factor to the suboptimal performance
+outcomes of DIM-enabled network devices, related to a mismatch in profiles.
+
+To address this issue, Net DIM introduces a per-device control to modify and
+access a device's ``rx-profile`` and ``tx-profile`` parameters:
+Assume that the target network device is named ethx, and ethx only declares
+support for RX profile setting and supports modification of ``usec`` field
+and ``pkts`` field (See the data structure:
+:c:type:`struct dim_cq_moder <dim_cq_moder>`).
+
+You can use ethtool to modify the current RX DIM profile where all
+values are 64::
+
+ $ ethtool -C ethx rx-profile 1,1,n_2,2,n_3,n,n_n,4,n_n,n,n
+
+``n`` means do not modify this field, and ``_`` separates structure
+elements of the profile array.
+
+Querying the current profiles using::
+
+ $ ethtool -c ethx
+ ...
+ rx-profile:
+ {.usec = 1, .pkts = 1, .comps = n/a,},
+ {.usec = 2, .pkts = 2, .comps = n/a,},
+ {.usec = 3, .pkts = 64, .comps = n/a,},
+ {.usec = 64, .pkts = 4, .comps = n/a,},
+ {.usec = 64, .pkts = 64, .comps = n/a,}
+ tx-profile: n/a
+
+If the network device does not support specific fields of DIM profiles,
+the corresponding ``n/a`` will display. If the ``n/a`` field is being
+modified, error messages will be reported.
+
+
Dynamic Interrupt Moderation (DIM) library API
==============================================
diff --git a/include/linux/dim.h b/include/linux/dim.h
index 43398f5eade2..e0f39bd85432 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -10,6 +10,8 @@
#include <linux/types.h>
#include <linux/workqueue.h>
+struct net_device;
+
/* Number of DIM profiles and period mode. */
#define NET_DIM_PARAMS_NUM_PROFILES 5
#define NET_DIM_DEFAULT_RX_CQ_PKTS_FROM_EQE 256
@@ -45,12 +47,45 @@
* @pkts: CQ packet counter suggestion (by DIM)
* @comps: Completion counter
* @cq_period_mode: CQ period count mode (from CQE/EQE)
+ * @rcu: for asynchronous kfree_rcu
*/
struct dim_cq_moder {
u16 usec;
u16 pkts;
u16 comps;
u8 cq_period_mode;
+ struct rcu_head rcu;
+};
+
+#define DIM_PROFILE_RX BIT(0) /* support rx profile modification */
+#define DIM_PROFILE_TX BIT(1) /* support tx profile modification */
+
+#define DIM_COALESCE_USEC BIT(0) /* support usec field modification */
+#define DIM_COALESCE_PKTS BIT(1) /* support pkts field modification */
+#define DIM_COALESCE_COMPS BIT(2) /* support comps field modification */
+
+/**
+ * struct dim_irq_moder - Structure for irq moderation information.
+ * Used to collect irq moderation related information.
+ *
+ * @profile_flags: DIM_PROFILE_*
+ * @coal_flags: DIM_COALESCE_* for Rx and Tx
+ * @dim_rx_mode: Rx DIM period count mode: CQE or EQE
+ * @dim_tx_mode: Tx DIM period count mode: CQE or EQE
+ * @rx_profile: DIM profile list for Rx
+ * @tx_profile: DIM profile list for Tx
+ * @rx_dim_work: Rx DIM worker scheduled by net_dim()
+ * @tx_dim_work: Tx DIM worker scheduled by net_dim()
+ */
+struct dim_irq_moder {
+ u8 profile_flags;
+ u8 coal_flags;
+ u8 dim_rx_mode;
+ u8 dim_tx_mode;
+ struct dim_cq_moder __rcu *rx_profile;
+ struct dim_cq_moder __rcu *tx_profile;
+ void (*rx_dim_work)(struct work_struct *work);
+ void (*tx_dim_work)(struct work_struct *work);
};
/**
@@ -198,6 +233,29 @@ enum dim_step_result {
DIM_ON_EDGE,
};
+/**
+ * net_dim_init_irq_moder - collect information to initialize irq moderation
+ * @dev: target network device
+ * @profile_flags: Rx or Tx profile modification capability
+ * @coal_flags: irq moderation params flags
+ * @rx_mode: CQ period mode for Rx
+ * @tx_mode: CQ period mode for Tx
+ * @rx_dim_work: Rx worker called after dim decision
+ * @tx_dim_work: Tx worker called after dim decision
+ *
+ * Return: 0 on success or a negative error code.
+ */
+int net_dim_init_irq_moder(struct net_device *dev, u8 profile_flags,
+ u8 coal_flags, u8 rx_mode, u8 tx_mode,
+ void (*rx_dim_work)(struct work_struct *work),
+ void (*tx_dim_work)(struct work_struct *work));
+
+/**
+ * net_dim_free_irq_moder - free fields for irq moderation
+ * @dev: target network device
+ */
+void net_dim_free_irq_moder(struct net_device *dev);
+
/**
* dim_on_top - check if current state is a good place to stop (top location)
* @dim: DIM context
diff --git a/include/linux/ethtool.h b/include/linux/ethtool.h
index 6fd9107d3cc0..959196af7f5a 100644
--- a/include/linux/ethtool.h
+++ b/include/linux/ethtool.h
@@ -284,7 +284,9 @@ bool ethtool_convert_link_mode_to_legacy_u32(u32 *legacy_u32,
#define ETHTOOL_COALESCE_TX_AGGR_MAX_BYTES BIT(24)
#define ETHTOOL_COALESCE_TX_AGGR_MAX_FRAMES BIT(25)
#define ETHTOOL_COALESCE_TX_AGGR_TIME_USECS BIT(26)
-#define ETHTOOL_COALESCE_ALL_PARAMS GENMASK(26, 0)
+#define ETHTOOL_COALESCE_RX_PROFILE BIT(27)
+#define ETHTOOL_COALESCE_TX_PROFILE BIT(28)
+#define ETHTOOL_COALESCE_ALL_PARAMS GENMASK(28, 0)
#define ETHTOOL_COALESCE_USECS \
(ETHTOOL_COALESCE_RX_USECS | ETHTOOL_COALESCE_TX_USECS)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c83b390191d4..983ef1935c38 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2401,6 +2401,9 @@ struct net_device {
/** @page_pools: page pools created for this netdevice */
struct hlist_head page_pools;
#endif
+
+ /** @irq_moder: dim parameters used if IS_ENABLED(CONFIG_DIMLIB). */
+ struct dim_irq_moder *irq_moder;
};
#define to_net_dev(d) container_of(d, struct net_device, dev)
diff --git a/include/uapi/linux/ethtool_netlink.h b/include/uapi/linux/ethtool_netlink.h
index b49b804b9495..d15856c7e001 100644
--- a/include/uapi/linux/ethtool_netlink.h
+++ b/include/uapi/linux/ethtool_netlink.h
@@ -415,12 +415,34 @@ enum {
ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES, /* u32 */
ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES, /* u32 */
ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS, /* u32 */
+ /* nest - _A_PROFILE_IRQ_MODERATION */
+ ETHTOOL_A_COALESCE_RX_PROFILE,
+ /* nest - _A_PROFILE_IRQ_MODERATION */
+ ETHTOOL_A_COALESCE_TX_PROFILE,
/* add new constants above here */
__ETHTOOL_A_COALESCE_CNT,
ETHTOOL_A_COALESCE_MAX = (__ETHTOOL_A_COALESCE_CNT - 1)
};
+enum {
+ ETHTOOL_A_PROFILE_UNSPEC,
+ /* nest, _A_IRQ_MODERATION_* */
+ ETHTOOL_A_PROFILE_IRQ_MODERATION,
+ __ETHTOOL_A_PROFILE_CNT,
+ ETHTOOL_A_PROFILE_MAX = (__ETHTOOL_A_PROFILE_CNT - 1)
+};
+
+enum {
+ ETHTOOL_A_IRQ_MODERATION_UNSPEC,
+ ETHTOOL_A_IRQ_MODERATION_USEC, /* u32 */
+ ETHTOOL_A_IRQ_MODERATION_PKTS, /* u32 */
+ ETHTOOL_A_IRQ_MODERATION_COMPS, /* u32 */
+
+ __ETHTOOL_A_IRQ_MODERATION_CNT,
+ ETHTOOL_A_IRQ_MODERATION_MAX = (__ETHTOOL_A_IRQ_MODERATION_CNT - 1)
+};
+
/* PAUSE */
enum {
diff --git a/lib/dim/net_dim.c b/lib/dim/net_dim.c
index 67d5beb34dc3..0cd41277c7a3 100644
--- a/lib/dim/net_dim.c
+++ b/lib/dim/net_dim.c
@@ -4,6 +4,7 @@
*/
#include <linux/dim.h>
+#include <linux/rtnetlink.h>
/*
* Net DIM profiles:
@@ -95,6 +96,75 @@ net_dim_get_def_tx_moderation(u8 cq_period_mode)
}
EXPORT_SYMBOL(net_dim_get_def_tx_moderation);
+int net_dim_init_irq_moder(struct net_device *dev, u8 profile_flags,
+ u8 coal_flags, u8 rx_mode, u8 tx_mode,
+ void (*rx_dim_work)(struct work_struct *work),
+ void (*tx_dim_work)(struct work_struct *work))
+{
+ struct dim_cq_moder *rxp = NULL, *txp;
+ struct dim_irq_moder *moder;
+ int len;
+
+ dev->irq_moder = kzalloc(sizeof(*dev->irq_moder), GFP_KERNEL);
+ if (!dev->irq_moder)
+ return -ENOMEM;
+
+ moder = dev->irq_moder;
+ len = NET_DIM_PARAMS_NUM_PROFILES * sizeof(*moder->rx_profile);
+
+ moder->coal_flags = coal_flags;
+ moder->profile_flags = profile_flags;
+
+ if (profile_flags & DIM_PROFILE_RX) {
+ moder->rx_dim_work = rx_dim_work;
+ moder->dim_rx_mode = rx_mode;
+ rxp = kmemdup(rx_profile[rx_mode], len, GFP_KERNEL);
+ if (!rxp)
+ goto free_moder;
+
+ rcu_assign_pointer(moder->rx_profile, rxp);
+ }
+
+ if (profile_flags & DIM_PROFILE_TX) {
+ moder->tx_dim_work = tx_dim_work;
+ moder->dim_tx_mode = tx_mode;
+ txp = kmemdup(tx_profile[tx_mode], len, GFP_KERNEL);
+ if (!txp)
+ goto free_rxp;
+
+ rcu_assign_pointer(moder->tx_profile, txp);
+ }
+
+ return 0;
+
+free_rxp:
+ kfree(rxp);
+free_moder:
+ kfree(moder);
+ return -ENOMEM;
+}
+EXPORT_SYMBOL(net_dim_init_irq_moder);
+
+/* RTNL lock is held. */
+void net_dim_free_irq_moder(struct net_device *dev)
+{
+ struct dim_cq_moder *rxp, *txp;
+
+ if (!dev->irq_moder)
+ return;
+
+ rxp = rtnl_dereference(dev->irq_moder->rx_profile);
+ txp = rtnl_dereference(dev->irq_moder->tx_profile);
+
+ rcu_assign_pointer(dev->irq_moder->rx_profile, NULL);
+ rcu_assign_pointer(dev->irq_moder->tx_profile, NULL);
+
+ kfree_rcu(rxp, rcu);
+ kfree_rcu(txp, rcu);
+ kfree(dev->irq_moder);
+}
+EXPORT_SYMBOL(net_dim_free_irq_moder);
+
static int net_dim_step(struct dim *dim)
{
if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
diff --git a/net/Kconfig b/net/Kconfig
index 9fe65fa26e48..d27d0deac0bf 100644
--- a/net/Kconfig
+++ b/net/Kconfig
@@ -508,6 +508,7 @@ config FAILOVER
config ETHTOOL_NETLINK
bool "Netlink interface for ethtool"
+ select DIMLIB
default y
help
An alternative userspace interface for ethtool based on generic
diff --git a/net/ethtool/coalesce.c b/net/ethtool/coalesce.c
index 83112c1a71ae..759b16e3d134 100644
--- a/net/ethtool/coalesce.c
+++ b/net/ethtool/coalesce.c
@@ -1,5 +1,6 @@
// SPDX-License-Identifier: GPL-2.0-only
+#include <linux/dim.h>
#include "netlink.h"
#include "common.h"
@@ -82,6 +83,14 @@ static int coalesce_prepare_data(const struct ethnl_req_info *req_base,
static int coalesce_reply_size(const struct ethnl_req_info *req_base,
const struct ethnl_reply_data *reply_base)
{
+ int modersz = nla_total_size(0) + /* _PROFILE_IRQ_MODERATION, nest */
+ nla_total_size(sizeof(u32)) + /* _IRQ_MODERATION_USEC */
+ nla_total_size(sizeof(u32)) + /* _IRQ_MODERATION_PKTS */
+ nla_total_size(sizeof(u32)); /* _IRQ_MODERATION_COMPS */
+
+ int total_modersz = nla_total_size(0) + /* _{R,T}X_PROFILE, nest */
+ modersz * NET_DIM_PARAMS_NUM_PROFILES;
+
return nla_total_size(sizeof(u32)) + /* _RX_USECS */
nla_total_size(sizeof(u32)) + /* _RX_MAX_FRAMES */
nla_total_size(sizeof(u32)) + /* _RX_USECS_IRQ */
@@ -108,7 +117,8 @@ static int coalesce_reply_size(const struct ethnl_req_info *req_base,
nla_total_size(sizeof(u8)) + /* _USE_CQE_MODE_RX */
nla_total_size(sizeof(u32)) + /* _TX_AGGR_MAX_BYTES */
nla_total_size(sizeof(u32)) + /* _TX_AGGR_MAX_FRAMES */
- nla_total_size(sizeof(u32)); /* _TX_AGGR_TIME_USECS */
+ nla_total_size(sizeof(u32)) + /* _TX_AGGR_TIME_USECS */
+ total_modersz * 2; /* _{R,T}X_PROFILE */
}
static bool coalesce_put_u32(struct sk_buff *skb, u16 attr_type, u32 val,
@@ -127,14 +137,84 @@ static bool coalesce_put_bool(struct sk_buff *skb, u16 attr_type, u32 val,
return nla_put_u8(skb, attr_type, !!val);
}
+/**
+ * coalesce_put_profile - fill reply with a nla nest with four child nla nests.
+ * @skb: socket buffer the message is stored in
+ * @attr_type: nest attr type ETHTOOL_A_COALESCE_*X_PROFILE
+ * @profile: data passed to userspace
+ * @coal_flags: modifiable parameters supported by the driver
+ *
+ * Put a dim profile nest attribute. Refer to ETHTOOL_A_PROFILE_IRQ_MODERATION.
+ *
+ * Return: 0 on success or a negative error code.
+ */
+static int coalesce_put_profile(struct sk_buff *skb, u16 attr_type,
+ const struct dim_cq_moder *profile,
+ u8 coal_flags)
+{
+ struct nlattr *profile_attr, *moder_attr;
+ int i, ret;
+
+ if (!profile || !coal_flags)
+ return 0;
+
+ profile_attr = nla_nest_start(skb, attr_type);
+ if (!profile_attr)
+ return -EMSGSIZE;
+
+ for (i = 0; i < NET_DIM_PARAMS_NUM_PROFILES; i++) {
+ moder_attr = nla_nest_start(skb,
+ ETHTOOL_A_PROFILE_IRQ_MODERATION);
+ if (!moder_attr) {
+ ret = -EMSGSIZE;
+ goto cancel_profile;
+ }
+
+ if (coal_flags & DIM_COALESCE_USEC) {
+ ret = nla_put_u32(skb, ETHTOOL_A_IRQ_MODERATION_USEC,
+ profile[i].usec);
+ if (ret)
+ goto cancel_moder;
+ }
+
+ if (coal_flags & DIM_COALESCE_PKTS) {
+ ret = nla_put_u32(skb, ETHTOOL_A_IRQ_MODERATION_PKTS,
+ profile[i].pkts);
+ if (ret)
+ goto cancel_moder;
+ }
+
+ if (coal_flags & DIM_COALESCE_COMPS) {
+ ret = nla_put_u32(skb, ETHTOOL_A_IRQ_MODERATION_COMPS,
+ profile[i].comps);
+ if (ret)
+ goto cancel_moder;
+ }
+
+ nla_nest_end(skb, moder_attr);
+ }
+
+ nla_nest_end(skb, profile_attr);
+
+ return 0;
+
+cancel_moder:
+ nla_nest_cancel(skb, moder_attr);
+cancel_profile:
+ nla_nest_cancel(skb, profile_attr);
+ return ret;
+}
+
static int coalesce_fill_reply(struct sk_buff *skb,
const struct ethnl_req_info *req_base,
const struct ethnl_reply_data *reply_base)
{
const struct coalesce_reply_data *data = COALESCE_REPDATA(reply_base);
const struct kernel_ethtool_coalesce *kcoal = &data->kernel_coalesce;
+ struct dim_irq_moder *moder = req_base->dev->irq_moder;
const struct ethtool_coalesce *coal = &data->coalesce;
u32 supported = data->supported_params;
+ int ret = 0;
if (coalesce_put_u32(skb, ETHTOOL_A_COALESCE_RX_USECS,
coal->rx_coalesce_usecs, supported) ||
@@ -192,11 +272,41 @@ static int coalesce_fill_reply(struct sk_buff *skb,
kcoal->tx_aggr_time_usecs, supported))
return -EMSGSIZE;
- return 0;
+ if (!moder)
+ return 0;
+
+ rcu_read_lock();
+ if (moder->profile_flags & DIM_PROFILE_RX) {
+ ret = coalesce_put_profile(skb, ETHTOOL_A_COALESCE_RX_PROFILE,
+ rcu_dereference(moder->rx_profile),
+ moder->coal_flags);
+ if (ret)
+ goto out;
+ }
+
+ if (moder->profile_flags & DIM_PROFILE_TX)
+ ret = coalesce_put_profile(skb, ETHTOOL_A_COALESCE_TX_PROFILE,
+ rcu_dereference(moder->tx_profile),
+ moder->coal_flags);
+
+out:
+ rcu_read_unlock();
+ return ret;
}
/* COALESCE_SET */
+static const struct nla_policy coalesce_irq_moderation_policy[] = {
+ [ETHTOOL_A_IRQ_MODERATION_USEC] = { .type = NLA_U32 },
+ [ETHTOOL_A_IRQ_MODERATION_PKTS] = { .type = NLA_U32 },
+ [ETHTOOL_A_IRQ_MODERATION_COMPS] = { .type = NLA_U32 },
+};
+
+static const struct nla_policy coalesce_profile_policy[] = {
+ [ETHTOOL_A_PROFILE_IRQ_MODERATION] =
+ NLA_POLICY_NESTED(coalesce_irq_moderation_policy),
+};
+
const struct nla_policy ethnl_coalesce_set_policy[] = {
[ETHTOOL_A_COALESCE_HEADER] =
NLA_POLICY_NESTED(ethnl_header_policy),
@@ -227,6 +337,10 @@ const struct nla_policy ethnl_coalesce_set_policy[] = {
[ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES] = { .type = NLA_U32 },
[ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES] = { .type = NLA_U32 },
[ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS] = { .type = NLA_U32 },
+ [ETHTOOL_A_COALESCE_RX_PROFILE] =
+ NLA_POLICY_NESTED(coalesce_profile_policy),
+ [ETHTOOL_A_COALESCE_TX_PROFILE] =
+ NLA_POLICY_NESTED(coalesce_profile_policy),
};
static int
@@ -234,6 +348,7 @@ ethnl_set_coalesce_validate(struct ethnl_req_info *req_info,
struct genl_info *info)
{
const struct ethtool_ops *ops = req_info->dev->ethtool_ops;
+ struct dim_irq_moder *irq_moder = req_info->dev->irq_moder;
struct nlattr **tb = info->attrs;
u32 supported_params;
u16 a;
@@ -243,6 +358,12 @@ ethnl_set_coalesce_validate(struct ethnl_req_info *req_info,
/* make sure that only supported parameters are present */
supported_params = ops->supported_coalesce_params;
+ if (irq_moder && irq_moder->profile_flags & DIM_PROFILE_RX)
+ supported_params |= ETHTOOL_COALESCE_RX_PROFILE;
+
+ if (irq_moder && irq_moder->profile_flags & DIM_PROFILE_TX)
+ supported_params |= ETHTOOL_COALESCE_TX_PROFILE;
+
for (a = ETHTOOL_A_COALESCE_RX_USECS; a < __ETHTOOL_A_COALESCE_CNT; a++)
if (tb[a] && !(supported_params & attr_to_mask(a))) {
NL_SET_ERR_MSG_ATTR(info->extack, tb[a],
@@ -253,6 +374,138 @@ ethnl_set_coalesce_validate(struct ethnl_req_info *req_info,
return 1;
}
+/**
+ * ethnl_update_irq_moder - update a specific field in the given profile
+ * @irq_moder: place that collects dim related information
+ * @irq_field: field in profile to modify
+ * @attr_type: attr type ETHTOOL_A_IRQ_MODERATION_*
+ * @tb: netlink attribute with new values or null
+ * @coal_bit: DIM_COALESCE_* bit from coal_flags
+ * @mod: pointer to bool for modification tracking
+ * @extack: netlink extended ack
+ *
+ * Return: 0 on success or a negative error code.
+ */
+static int ethnl_update_irq_moder(struct dim_irq_moder *irq_moder,
+ u16 *irq_field, u16 attr_type,
+ struct nlattr **tb,
+ u8 coal_bit, bool *mod,
+ struct netlink_ext_ack *extack)
+{
+ int ret = 0;
+ u32 val;
+
+ if (!tb[attr_type])
+ return 0;
+
+ if (irq_moder->coal_flags & coal_bit) {
+ val = nla_get_u32(tb[attr_type]);
+ if (*irq_field == val)
+ return 0;
+
+ *irq_field = val;
+ *mod = true;
+ } else {
+ NL_SET_BAD_ATTR(extack, tb[attr_type]);
+ ret = -EOPNOTSUPP;
+ }
+
+ return ret;
+}
+
+/**
+ * ethnl_update_profile - get a profile nest with child nests from userspace.
+ * @dev: netdevice to update the profile
+ * @dst: profile get from the driver and modified by ethnl_update_profile.
+ * @nests: nest attr ETHTOOL_A_COALESCE_*X_PROFILE to set profile.
+ * @mod: pointer to bool for modification tracking
+ * @extack: Netlink extended ack
+ *
+ * Layout of nests:
+ * Nested ETHTOOL_A_COALESCE_*X_PROFILE attr
+ * Nested ETHTOOL_A_PROFILE_IRQ_MODERATION attr
+ * ETHTOOL_A_IRQ_MODERATION_USEC attr
+ * ETHTOOL_A_IRQ_MODERATION_PKTS attr
+ * ETHTOOL_A_IRQ_MODERATION_COMPS attr
+ * ...
+ * Nested ETHTOOL_A_PROFILE_IRQ_MODERATION attr
+ * ETHTOOL_A_IRQ_MODERATION_USEC attr
+ * ETHTOOL_A_IRQ_MODERATION_PKTS attr
+ * ETHTOOL_A_IRQ_MODERATION_COMPS attr
+ *
+ * Return: 0 on success or a negative error code.
+ */
+static int ethnl_update_profile(struct net_device *dev,
+ struct dim_cq_moder __rcu **dst,
+ const struct nlattr *nests,
+ bool *mod,
+ struct netlink_ext_ack *extack)
+{
+ int len_irq_moder = ARRAY_SIZE(coalesce_irq_moderation_policy);
+ struct nlattr *tb[ARRAY_SIZE(coalesce_irq_moderation_policy)];
+ struct dim_irq_moder *irq_moder = dev->irq_moder;
+ struct dim_cq_moder *new_profile, *old_profile;
+ int ret, rem, i = 0, len;
+ struct nlattr *nest;
+
+ if (!nests)
+ return 0;
+
+ if (!*dst)
+ return -EOPNOTSUPP;
+
+ old_profile = rtnl_dereference(*dst);
+ len = NET_DIM_PARAMS_NUM_PROFILES * sizeof(*old_profile);
+ new_profile = kmemdup(old_profile, len, GFP_KERNEL);
+ if (!new_profile)
+ return -ENOMEM;
+
+ nla_for_each_nested_type(nest, ETHTOOL_A_PROFILE_IRQ_MODERATION,
+ nests, rem) {
+ ret = nla_parse_nested(tb, len_irq_moder - 1, nest,
+ coalesce_irq_moderation_policy,
+ extack);
+ if (ret)
+ goto err_out;
+
+ ret = ethnl_update_irq_moder(irq_moder, &new_profile[i].usec,
+ ETHTOOL_A_IRQ_MODERATION_USEC,
+ tb, DIM_COALESCE_USEC,
+ mod, extack);
+ if (ret)
+ goto err_out;
+
+ ret = ethnl_update_irq_moder(irq_moder, &new_profile[i].pkts,
+ ETHTOOL_A_IRQ_MODERATION_PKTS,
+ tb, DIM_COALESCE_PKTS,
+ mod, extack);
+ if (ret)
+ goto err_out;
+
+ ret = ethnl_update_irq_moder(irq_moder, &new_profile[i].comps,
+ ETHTOOL_A_IRQ_MODERATION_COMPS,
+ tb, DIM_COALESCE_COMPS,
+ mod, extack);
+ if (ret)
+ goto err_out;
+
+ i++;
+ }
+
+ /* After the profile is modified, dim itself is a dynamic
+ * mechanism and will quickly fit to the appropriate
+ * coalescing parameters according to the new profile.
+ */
+ rcu_assign_pointer(*dst, new_profile);
+ kfree_rcu(old_profile, rcu);
+
+ return 0;
+
+err_out:
+ kfree(new_profile);
+ return ret;
+}
+
static int
__ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info,
bool *dual_change)
@@ -317,6 +570,22 @@ __ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info,
ethnl_update_u32(&kernel_coalesce.tx_aggr_time_usecs,
tb[ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS], &mod);
+ if (dev->irq_moder && dev->irq_moder->profile_flags & DIM_PROFILE_RX) {
+ ret = ethnl_update_profile(dev, &dev->irq_moder->rx_profile,
+ tb[ETHTOOL_A_COALESCE_RX_PROFILE],
+ &mod, info->extack);
+ if (ret < 0)
+ return ret;
+ }
+
+ if (dev->irq_moder && dev->irq_moder->profile_flags & DIM_PROFILE_TX) {
+ ret = ethnl_update_profile(dev, &dev->irq_moder->tx_profile,
+ tb[ETHTOOL_A_COALESCE_TX_PROFILE],
+ &mod, info->extack);
+ if (ret < 0)
+ return ret;
+ }
+
/* Update operation modes */
ethnl_update_bool32(&coalesce.use_adaptive_rx_coalesce,
tb[ETHTOOL_A_COALESCE_USE_ADAPTIVE_RX], &mod_mode);
--
2.32.0.3.g01195cf9f
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH net-next v15 4/5] dim: add new interfaces for initialization and getting results
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
` (2 preceding siblings ...)
2024-06-21 10:13 ` [PATCH net-next v15 3/5] ethtool: provide customized dim profile management Heng Qi
@ 2024-06-21 10:13 ` Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 5/5] virtio-net: support dim profile fine-tuning Heng Qi
2024-06-26 0:50 ` [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel patchwork-bot+netdevbpf
5 siblings, 0 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki, Simon Horman
DIM-related mode and work have been collected in one same place,
so new interfaces are added to provide convenience.
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Simon Horman <horms@kernel.org>
---
include/linux/dim.h | 48 ++++++++++++++++++++++++++++++++
lib/dim/net_dim.c | 68 +++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 116 insertions(+)
diff --git a/include/linux/dim.h b/include/linux/dim.h
index e0f39bd85432..1b581ff25a15 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -256,6 +256,54 @@ int net_dim_init_irq_moder(struct net_device *dev, u8 profile_flags,
*/
void net_dim_free_irq_moder(struct net_device *dev);
+/**
+ * net_dim_setting - initialize DIM's cq mode and schedule worker
+ * @dev: target network device
+ * @dim: DIM context
+ * @is_tx: true indicates the tx direction, false indicates the rx direction
+ */
+void net_dim_setting(struct net_device *dev, struct dim *dim, bool is_tx);
+
+/**
+ * net_dim_work_cancel - synchronously cancel dim's worker
+ * @dim: DIM context
+ */
+void net_dim_work_cancel(struct dim *dim);
+
+/**
+ * net_dim_get_rx_irq_moder - get DIM rx results based on profile_ix
+ * @dev: target network device
+ * @dim: DIM context
+ *
+ * Return: DIM irq moderation
+ */
+struct dim_cq_moder
+net_dim_get_rx_irq_moder(struct net_device *dev, struct dim *dim);
+
+/**
+ * net_dim_get_tx_irq_moder - get DIM tx results based on profile_ix
+ * @dev: target network device
+ * @dim: DIM context
+ *
+ * Return: DIM irq moderation
+ */
+struct dim_cq_moder
+net_dim_get_tx_irq_moder(struct net_device *dev, struct dim *dim);
+
+/**
+ * net_dim_set_rx_mode - set DIM rx cq mode
+ * @dev: target network device
+ * @rx_mode: target rx cq mode
+ */
+void net_dim_set_rx_mode(struct net_device *dev, u8 rx_mode);
+
+/**
+ * net_dim_set_tx_mode - set DIM tx cq mode
+ * @dev: target network device
+ * @tx_mode: target tx cq mode
+ */
+void net_dim_set_tx_mode(struct net_device *dev, u8 tx_mode);
+
/**
* dim_on_top - check if current state is a good place to stop (top location)
* @dim: DIM context
diff --git a/lib/dim/net_dim.c b/lib/dim/net_dim.c
index 0cd41277c7a3..d7e7028e9b19 100644
--- a/lib/dim/net_dim.c
+++ b/lib/dim/net_dim.c
@@ -165,6 +165,74 @@ void net_dim_free_irq_moder(struct net_device *dev)
}
EXPORT_SYMBOL(net_dim_free_irq_moder);
+void net_dim_setting(struct net_device *dev, struct dim *dim, bool is_tx)
+{
+ struct dim_irq_moder *irq_moder = dev->irq_moder;
+
+ if (!irq_moder)
+ return;
+
+ if (is_tx) {
+ INIT_WORK(&dim->work, irq_moder->tx_dim_work);
+ dim->mode = READ_ONCE(irq_moder->dim_tx_mode);
+ return;
+ }
+
+ INIT_WORK(&dim->work, irq_moder->rx_dim_work);
+ dim->mode = READ_ONCE(irq_moder->dim_rx_mode);
+}
+EXPORT_SYMBOL(net_dim_setting);
+
+void net_dim_work_cancel(struct dim *dim)
+{
+ cancel_work_sync(&dim->work);
+}
+EXPORT_SYMBOL(net_dim_work_cancel);
+
+struct dim_cq_moder net_dim_get_rx_irq_moder(struct net_device *dev,
+ struct dim *dim)
+{
+ struct dim_cq_moder res, *profile;
+
+ rcu_read_lock();
+ profile = rcu_dereference(dev->irq_moder->rx_profile);
+ res = profile[dim->profile_ix];
+ rcu_read_unlock();
+
+ res.cq_period_mode = dim->mode;
+
+ return res;
+}
+EXPORT_SYMBOL(net_dim_get_rx_irq_moder);
+
+struct dim_cq_moder net_dim_get_tx_irq_moder(struct net_device *dev,
+ struct dim *dim)
+{
+ struct dim_cq_moder res, *profile;
+
+ rcu_read_lock();
+ profile = rcu_dereference(dev->irq_moder->tx_profile);
+ res = profile[dim->profile_ix];
+ rcu_read_unlock();
+
+ res.cq_period_mode = dim->mode;
+
+ return res;
+}
+EXPORT_SYMBOL(net_dim_get_tx_irq_moder);
+
+void net_dim_set_rx_mode(struct net_device *dev, u8 rx_mode)
+{
+ WRITE_ONCE(dev->irq_moder->dim_rx_mode, rx_mode);
+}
+EXPORT_SYMBOL(net_dim_set_rx_mode);
+
+void net_dim_set_tx_mode(struct net_device *dev, u8 tx_mode)
+{
+ WRITE_ONCE(dev->irq_moder->dim_tx_mode, tx_mode);
+}
+EXPORT_SYMBOL(net_dim_set_tx_mode);
+
static int net_dim_step(struct dim *dim)
{
if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
--
2.32.0.3.g01195cf9f
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH net-next v15 5/5] virtio-net: support dim profile fine-tuning
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
` (3 preceding siblings ...)
2024-06-21 10:13 ` [PATCH net-next v15 4/5] dim: add new interfaces for initialization and getting results Heng Qi
@ 2024-06-21 10:13 ` Heng Qi
2024-06-26 0:50 ` [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel patchwork-bot+netdevbpf
5 siblings, 0 replies; 7+ messages in thread
From: Heng Qi @ 2024-06-21 10:13 UTC (permalink / raw)
To: netdev, virtualization
Cc: Jakub Kicinski, David S . Miller, Paolo Abeni, Eric Dumazet,
Jason Wang, Michael S . Tsirkin, Brett Creeley, Ratheesh Kannoth,
Alexander Lobakin, Xuan Zhuo, Tal Gilboa, Jonathan Corbet,
linux-doc, Maxime Chevallier, Jiri Pirko, Paul Greenwalt,
Ahmed Zaki, Vladimir Oltean, Kory Maincent, Andrew Lunn,
justinstitt, donald.hunter, Eugenio Pérez, Andrew Morton,
Dragos Tatulea, Rahul Rameshbabu, Heiner Kallweit,
Przemek Kitszel, awel Dembicki, Simon Horman
Virtio-net has different types of back-end device implementations.
In order to effectively optimize the dim library's gains for different
device implementations, let's use the new interface params to
initialize and query dim results from a customized profile list.
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Simon Horman <horms@kernel.org>
---
drivers/net/virtio_net.c | 54 ++++++++++++++++++++++++++++++++++------
1 file changed, 47 insertions(+), 7 deletions(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b1f8b720733e..77c23979f687 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -2468,6 +2468,13 @@ static int virtnet_enable_queue_pair(struct virtnet_info *vi, int qp_index)
return err;
}
+static void virtnet_cancel_dim(struct virtnet_info *vi, struct dim *dim)
+{
+ if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
+ return;
+ net_dim_work_cancel(dim);
+}
+
static int virtnet_open(struct net_device *dev)
{
struct virtnet_info *vi = netdev_priv(dev);
@@ -2494,7 +2501,7 @@ static int virtnet_open(struct net_device *dev)
for (i--; i >= 0; i--) {
virtnet_disable_queue_pair(vi, i);
- cancel_work_sync(&vi->rq[i].dim.work);
+ virtnet_cancel_dim(vi, &vi->rq[i].dim);
}
return err;
@@ -2670,7 +2677,7 @@ static int virtnet_rx_resize(struct virtnet_info *vi,
if (running) {
napi_disable(&rq->napi);
- cancel_work_sync(&rq->dim.work);
+ virtnet_cancel_dim(vi, &rq->dim);
}
err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_unmap_free_buf);
@@ -2931,7 +2938,7 @@ static int virtnet_close(struct net_device *dev)
for (i = 0; i < vi->max_queue_pairs; i++) {
virtnet_disable_queue_pair(vi, i);
- cancel_work_sync(&vi->rq[i].dim.work);
+ virtnet_cancel_dim(vi, &vi->rq[i].dim);
}
return 0;
@@ -4457,7 +4464,7 @@ static void virtnet_rx_dim_work(struct work_struct *work)
if (!rq->dim_enabled)
goto out;
- update_moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
+ update_moder = net_dim_get_rx_irq_moder(dev, dim);
if (update_moder.usec != rq->intr_coal.max_usecs ||
update_moder.pkts != rq->intr_coal.max_packets) {
err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, qnum,
@@ -5157,6 +5164,36 @@ static void virtnet_tx_timeout(struct net_device *dev, unsigned int txqueue)
jiffies_to_usecs(jiffies - READ_ONCE(txq->trans_start)));
}
+static int virtnet_init_irq_moder(struct virtnet_info *vi)
+{
+ u8 profile_flags = 0, coal_flags = 0;
+ int ret, i;
+
+ profile_flags |= DIM_PROFILE_RX;
+ coal_flags |= DIM_COALESCE_USEC | DIM_COALESCE_PKTS;
+ ret = net_dim_init_irq_moder(vi->dev, profile_flags, coal_flags,
+ DIM_CQ_PERIOD_MODE_START_FROM_EQE,
+ 0, virtnet_rx_dim_work, NULL);
+
+ if (ret)
+ return ret;
+
+ for (i = 0; i < vi->max_queue_pairs; i++)
+ net_dim_setting(vi->dev, &vi->rq[i].dim, false);
+
+ return 0;
+}
+
+static void virtnet_free_irq_moder(struct virtnet_info *vi)
+{
+ if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
+ return;
+
+ rtnl_lock();
+ net_dim_free_irq_moder(vi->dev);
+ rtnl_unlock();
+}
+
static const struct net_device_ops virtnet_netdev = {
.ndo_open = virtnet_open,
.ndo_stop = virtnet_close,
@@ -5436,9 +5473,6 @@ static int virtnet_alloc_queues(struct virtnet_info *vi)
virtnet_poll_tx,
napi_tx ? napi_weight : 0);
- INIT_WORK(&vi->rq[i].dim.work, virtnet_rx_dim_work);
- vi->rq[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
-
sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
@@ -5867,6 +5901,10 @@ static int virtnet_probe(struct virtio_device *vdev)
for (i = 0; i < vi->max_queue_pairs; i++)
if (vi->sq[i].napi.weight)
vi->sq[i].intr_coal.max_packets = 1;
+
+ err = virtnet_init_irq_moder(vi);
+ if (err)
+ goto free;
}
#ifdef CONFIG_SYSFS
@@ -6018,6 +6056,8 @@ static void virtnet_remove(struct virtio_device *vdev)
disable_rx_mode_work(vi);
flush_work(&vi->rx_mode_work);
+ virtnet_free_irq_moder(vi);
+
unregister_netdev(vi->dev);
net_failover_destroy(vi->failover);
--
2.32.0.3.g01195cf9f
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
` (4 preceding siblings ...)
2024-06-21 10:13 ` [PATCH net-next v15 5/5] virtio-net: support dim profile fine-tuning Heng Qi
@ 2024-06-26 0:50 ` patchwork-bot+netdevbpf
5 siblings, 0 replies; 7+ messages in thread
From: patchwork-bot+netdevbpf @ 2024-06-26 0:50 UTC (permalink / raw)
To: Heng Qi
Cc: netdev, virtualization, kuba, davem, pabeni, edumazet, jasowang,
mst, bcreeley, rkannoth, aleksander.lobakin, xuanzhuo, talgi,
corbet, linux-doc, maxime.chevallier, jiri, paul.greenwalt,
ahmed.zaki, vladimir.oltean, kory.maincent, andrew, justinstitt,
donald.hunter, eperezma, akpm, dtatulea, rrameshbabu, hkallweit1,
przemyslaw.kitszel, paweldembicki
Hello:
This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Fri, 21 Jun 2024 18:13:48 +0800 you wrote:
> The NetDIM library provides excellent acceleration for many modern
> network cards. However, the default profiles of DIM limits its maximum
> capabilities for different NICs, so providing a way which the NIC can
> be custom configured is necessary.
>
> Currently, the way is based on the commonly used "ethtool -C".
>
> [...]
Here is the summary with links:
- [net-next,v15,1/5] linux/dim: move useful macros to .h file
https://git.kernel.org/netdev/net-next/c/0e942053e4dc
- [net-next,v15,2/5] dim: make DIMLIB dependent on NET
https://git.kernel.org/netdev/net-next/c/b65e697a7c9e
- [net-next,v15,3/5] ethtool: provide customized dim profile management
https://git.kernel.org/netdev/net-next/c/f750dfe825b9
- [net-next,v15,4/5] dim: add new interfaces for initialization and getting results
https://git.kernel.org/netdev/net-next/c/13ba28c5cd04
- [net-next,v15,5/5] virtio-net: support dim profile fine-tuning
https://git.kernel.org/netdev/net-next/c/dcb67f6a9ead
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2024-06-26 0:50 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-06-21 10:13 [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 1/5] linux/dim: move useful macros to .h file Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 2/5] dim: make DIMLIB dependent on NET Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 3/5] ethtool: provide customized dim profile management Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 4/5] dim: add new interfaces for initialization and getting results Heng Qi
2024-06-21 10:13 ` [PATCH net-next v15 5/5] virtio-net: support dim profile fine-tuning Heng Qi
2024-06-26 0:50 ` [PATCH net-next v15 0/5] ethtool: provide the dim profile fine-tuning channel patchwork-bot+netdevbpf
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).