Hi Cristian,
Thank you for giving suggestion.
Will add in V7 patch.
Regards,
Li Zhang
> -Original Message-
> From: Dumitrescu, Cristian
> Sent: Wednesday, April 14, 2021 10:01 PM
> To: Asaf Penso ; NBU-Contact-Thomas Monjalon
> ; Li Zhang
> Cc: Ori Kam ; Slava Ovsiien
Thanks Cristian.
Will change it in V7 patch.
Regards,
Li Zhang
> -Original Message-
> From: Dumitrescu, Cristian
> Sent: Thursday, April 15, 2021 12:16 AM
> To: Li Zhang ; dek...@nvidia.com; Ori Kam
> ; Slava Ovsiienko ; Matan
> Azrad ; Shahaf Shuler ;
> li
Thanks Jasvinder.
Will change it in V7 patch.
Regards,
Li Zhang
> -Original Message-
> From: Singh, Jasvinder
> Sent: Thursday, April 15, 2021 6:22 AM
> To: Li Zhang ; dek...@nvidia.com; Ori Kam
> ; Slava Ovsiienko ; Matan
> Azrad ; Shahaf Shuler ;
> Dumit
lation doc files.
V3: Fix coding style issues.
V4: Fix comments about Depends-on
V5: Fix comments about rte_mtr_meter_policy_add.
V6: Delete policy example.
V7: Fix comments and place two common policy template as macros.
Haifei Luo (1):
app/testpmd: support policy actions per color
Li Zhang (1
From: Haifei Luo
Add the create/del policy CLIs to support actions per color.
The CLIs are:
Create: add port meter policy (port_id) (policy_id) g_actions (actions)
y_actions (actions) r_actions (actions)
Delete: del port meter policy (port_id) (policy_id)
Examples:
testpmd> add port meter poli
.
Signed-off-by: Li Zhang
Signed-off-by: Haifei Luo
Signed-off-by: Jiawei Wang
Acked-by: Matan Azrad
Acked-by: Ray Kinsella
---
app/test-flow-perf/main.c | 7 -
app/test-pmd/cmdline.c| 1 -
app/test-pmd/cmdline_mtr.c| 172
eries=16357
V2: Add MLX5_MTR_DEFAULT_POLICY_ID in MLX5 PMD
V3: Fix comments about Depends-on
V4: Fix comments about rte_mtr_meter_policy_add.
V5: Fix comments about destory meter on different ports.
Li Zhang (3):
net/mlx5: support meter policy operations
net/mlx5: support meter creation with policy
net/mlx5: pr
be split to
sub-policies per TIR in the flow creation time.
Create a function to prepare the policy and
its sub-policies for a configured flow with meter.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5_flow.h| 10 +++
drivers/net/mlx5/mlx5_flow_dv.c | 144
drop table.
4.Allocate meter resource per domain per color.
5.It can work with both ASO and legacy meter HW objects.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 13 +-
drivers/net/mlx5/mlx5.h| 62 ++---
drivers/net/mlx5/mlx5_flow.c
From: Shun Hao
Currently ASO meter must be followed by policy table, so this adds
the support that connecting meter and policy table.
There are several cases to be considered:
1. For non-termination policy, connect meter to the default policy
table.
2. For non-RSS termination policy case, simply
ess case, for example,
QUEUE\RSS\JUMP must be configured as last action for GREEN color.
All the above limitations will be validated.
create:
Validate the policy configuration.
Prepare the related tables and actions.
destroy:
Release the created policy resources.
Signed-off-by: Li Zhang
Acked-b
eries=16389
V2: Add MLX5_MTR_DEFAULT_POLICY_ID in MLX5 PMD
V3: Fix comments about Depends-on
V4: Fix comments about rte_mtr_meter_policy_add.
V5: Fix comments about destory meter on different ports.
V6: Fix comments about Depends-on
Li Zhang (3):
net/mlx5: support meter policy operations
net/mlx5: support meter
From: Shun Hao
Currently ASO meter must be followed by policy table, so this adds
the support that connecting meter and policy table.
There are several cases to be considered:
1. For non-termination policy, connect meter to the default policy
table.
2. For non-RSS termination policy case, simply
be split to
sub-policies per TIR in the flow creation time.
Create a function to prepare the policy and
its sub-policies for a configured flow with meter.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5_flow.h| 10 +++
drivers/net/mlx5/mlx5_flow_dv.c | 144
drop table.
4.Allocate meter resource per domain per color.
5.It can work with both ASO and legacy meter HW objects.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 13 +-
drivers/net/mlx5/mlx5.h| 62 ++---
drivers/net/mlx5/mlx5_flow.c
ess case, for example,
QUEUE\RSS\JUMP must be configured as last action for GREEN color.
All the above limitations will be validated.
create:
Validate the policy configuration.
Prepare the related tables and actions.
destroy:
Release the created policy resources.
Signed-off-by: Li Zhang
Acked-b
pport policy actions per color
Li Zhang (1):
ethdev: add pre-defined meter policy API
app/test-flow-perf/main.c | 7 -
app/test-pmd/cmdline.c| 14 +-
app/test-pmd/cmdline_flow.c | 118 -
app/test-pmd/cmdline_mtr.c
From: Haifei Luo
Add the create/del policy CLIs to support actions per color.
The CLIs are:
Create: add port meter policy (port_id) (policy_id) g_actions (actions)
y_actions (actions) r_actions (actions)
Delete: del port meter policy (port_id) (policy_id)
Examples:
testpmd> add port meter poli
.
Signed-off-by: Li Zhang
Signed-off-by: Haifei Luo
Signed-off-by: Jiawei Wang
Acked-by: Matan Azrad
Acked-by: Ray Kinsella
---
app/test-flow-perf/main.c | 7 -
app/test-pmd/cmdline.c| 1 -
app/test-pmd/cmdline_mtr.c| 172
Hi Raslan,
Thanks. I will fix in the next patch.
Regards,
Li Zhang
> -Original Message-
> From: Raslan Darawsheh
> Sent: Thursday, April 15, 2021 9:05 PM
> To: Li Zhang ; dek...@nvidia.com; Ori Kam
> ; Slava Ovsiienko ; Matan
> Azrad ; Shahaf Shuler
> Cc: dev@
e for ASO flow meter manage to HW.
V2: Fix compile error issue (mlx5_glue->destroy_flow_action)
V3: Fix comments.
V4: Fix comments and add Acked.
V5: Fix unused log_obj_size issue.
Li Zhang (10):
net/mlx5: optimize meter statistics
common/mlx5: add definitions for ASO flow meter
common/m
From: Suanming Mou
This commit adds table entry walk for the three level table.
Signed-off-by: Suanming Mou
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5_utils.h | 90 +++
1 file changed, 90 insertions(+)
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/
From: Shun Hao
8 bits are used for meter color in meter register. When the meter
register can be shared, the rest 24 bits can be used by others.
This adds the definination for the 24 bits that can be shared.
Signed-off-by: Shun Hao
Acked-by: Matan Azrad
---
drivers/common/mlx5/mlx5_prm.h | 6
From: Shun Hao
ASO meter feature may require to locate the flow
context tag action after the ASO action.
When color register is shared by meter_id/flow_id, it's like:
Bits[0-7] A meter color value set by the HW.
Bits[8-31] A flow id and meter id set by SW.
Currently the tag action for meter writ
This patch adds different PRM definitions, related to ASO flow meter
feature, in MLX5 PMD code.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/common/mlx5/mlx5_prm.h | 75 --
1 file changed, 71 insertions(+), 4 deletions(-)
diff --git a/drivers
From: Shun Hao
Currently, packets after meter will be steered to a global policer
table,
which includes green/red color rules for every meter, so as to have
counter statistics of each color in every meter.
There's a bug that all the rules in global policer table are matching
only color criteria,
Add DevX API to create ASO flow meter object.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/common/mlx5/mlx5_devx_cmds.c | 54
drivers/common/mlx5/mlx5_devx_cmds.h | 18 +-
drivers/common/mlx5/version.map | 1 +
3 files changed, 72
Read and store the device capability of FLOW_METER_ASO general object,
using the DevX API.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/common/mlx5/mlx5_devx_cmds.c | 14 ++
drivers/common/mlx5/mlx5_devx_cmds.h | 8
2 files changed, 22 insertions(+)
diff
Initialize the flow meter ASO SQ WQEs with
all the constant data that should not be updated
per enqueue operation.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 16
drivers/net/mlx5/meson.build | 2 +-
drivers/net/mlx5
ASO flow meter send WQE and CQE handle functions
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5.h| 4 +
drivers/net/mlx5/mlx5_flow_aso.c | 183 -
drivers/net/mlx5/mlx5_flow_meter.c | 143 +-
3 files
Synchronize ASO meter queue accesses from
different threads using a spinlock.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5.h | 1 +
drivers/net/mlx5/mlx5_flow_aso.c | 16 +---
2 files changed, 14 insertions(+), 3 deletions(-)
diff --git a
When ASO action is available, use it as the meter action
Signed-off-by: Shun Hao
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 4 +-
drivers/net/mlx5/mlx5.c| 12
drivers/net/mlx5/mlx5.h| 7 ++-
drivers/net/mlx5
flow table data entry, while
allocates the flow table, if the table level is same but the
different table id, the new table will be allocated with new
table object id. It supports 4M multiple flow tables on the
same level.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/m
Add ASO flow meter pool to manage meter object
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5.c| 2 +-
drivers/net/mlx5/mlx5.h| 207 -
drivers/net/mlx5/mlx5_flow.c | 70 -
drivers/net/mlx5/mlx5_flow.h | 196
color.
+*
+* See struct rte_flow_action_meter_color.
+*/
+ RTE_FLOW_ACTION_TYPE_METER_COLOR,
Regards,
Li Zhang
> -Original Message-
> From: Stephen Hemminger
> Sent: Wednesday, April 21, 2021 1:56 AM
> To: Li Zhang
> Cc: dek...@nvidia.com;
* Rebase.
V6:
* Fix comments about Depends-on.
V5:
* Fix comments about destory meter on different ports.
V4:
* Fix comments about rte_mtr_meter_policy_add.
V3:
* Fix comments about Depends-on.
V2:
* Add MLX5_MTR_DEFAULT_POLICY_ID in MLX5 PMD.
Li Zhang (3):
net/mlx5: support meter policy operations
be split to
sub-policies per TIR in the flow creation time.
Create a function to prepare the policy and
its sub-policies for a configured flow with meter.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5_flow.h| 10 +++
drivers/net/mlx5/mlx5_flow_dv.c | 144
ess case, for example,
QUEUE\RSS\JUMP must be configured as last action for GREEN color.
All the above limitations will be validated.
create:
Validate the policy configuration.
Prepare the related tables and actions.
destroy:
Release the created policy resources.
Signed-off-by: Li Zhang
Acked-b
drop table.
4.Allocate meter resource per domain per color.
5.It can work with both ASO and legacy meter HW objects.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 13 +-
drivers/net/mlx5/mlx5.h| 62 ++---
drivers/net/mlx5/mlx5_flow.c
From: Shun Hao
Currently ASO meter must be followed by policy table, so this adds
the support that connecting meter and policy table.
There are several cases to be considered:
1. For non-termination policy, connect meter to the default policy
table.
2. For non-RSS termination policy case, simply
Currently meter algorithms only supports bytes per second(BPS).
Such as Single Rate Three Color Marker (srTCM rfc2697)
Add packet_mode flag to support rate is packet per second.
So that it can meter traffic by packet per second (PPS).
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
doc
Add meter mode definitions in PRM file
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/common/mlx5/mlx5_prm.h | 8
1 file changed, 8 insertions(+)
diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h
index c6d8060bb9..efa5ae67bf 100644
--- a/drivers
oject/dpdk/list/?series=16312
Depends-on: series=16535 ("support meter policy operations")
https://patchwork.dpdk.org/project/dpdk/list/?series=16535
v4:
* Rebase.
V3:
* Rebase.
* Fix comments about Depends-on.
V2:
* Fix comments about commit-log
Li Zhang (2):
common/mlx5: add mete
Hi Thomas,
Thanks for your comments.
Since this patch already accepted, I will fix it in RC2.
Regards,
Li Zhang
> -Original Message-
> From: Thomas Monjalon
> Sent: Thursday, April 22, 2021 3:44 AM
> To: Jiawei(Jonny) Wang
> Cc: Matan Azrad ; Ori Kam ; Slava
>
nts about commit-log
Li Zhang (1):
net/mlx5: support meter PPS profile
doc/guides/nics/mlx5.rst | 1 +
doc/guides/rel_notes/release_21_05.rst | 1 +
drivers/common/mlx5/mlx5_prm.h | 8 +++
drivers/net/mlx5/mlx5_flow_aso.c | 17 +++---
drivers/net/mlx5/mlx5_f
Currently meter algorithms only supports bytes units for meter profiles.
Using ASO feature, the driver can support metering in per packet units.
Add support for packet units in meter profiles.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
doc/guides/nics/mlx5.rst | 1 +
doc
* Fix comments about release-note
v7:
* Rebase.
V6:
* Fix comments about Depends-on.
V5:
* Fix comments about destory meter on different ports.
V4:
* Fix comments about rte_mtr_meter_policy_add.
V3:
* Fix comments about Depends-on.
V2:
* Add MLX5_MTR_DEFAULT_POLICY_ID in MLX5 PMD.
Li Zhang (3):
net
ess case, for example,
QUEUE\RSS\JUMP must be configured as last action for GREEN color.
All the above limitations will be validated.
create:
Validate the policy configuration.
Prepare the related tables and actions.
destroy:
Release the created policy resources.
Signed-off-by: Li Zhang
Acked-b
drop table.
4.Allocate meter resource per domain per color.
5.It can work with both ASO and legacy meter HW objects.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/linux/mlx5_os.c | 13 +-
drivers/net/mlx5/mlx5.h| 62 ++---
drivers/net/mlx5/mlx5_flow.c
be split to
sub-policies per TIR in the flow creation time.
Create a function to prepare the policy and
its sub-policies for a configured flow with meter.
Signed-off-by: Li Zhang
Acked-by: Matan Azrad
---
drivers/net/mlx5/mlx5_flow.h| 10 +++
drivers/net/mlx5/mlx5_flow_dv.c | 145
From: Shun Hao
Currently ASO meter must be followed by policy table, so this adds
the support that connecting meter and policy table.
There are several cases to be considered:
1. For non-termination policy, connect meter to the default policy
table.
2. For non-RSS termination policy case, simply
When creating the meter sub-policy RSS rule,
the RSS descriptor was used before its update.
It also need update tunnel bit in RSS descriptor
after flow translate.
Use it only when it is updated.
Fixes: ec962bad14e ("net/mlx5: fix metering cleanup on stop")
Cc: sta...@dpdk.org
Signed-
1868
RFC ("Add vDPA multi-threads optiomization")
https://patchwork.dpdk.org/project/dpdk/cover/20220408075606.33056-1-l...@nvidia.com/
Li Zhang (12):
vdpa/mlx5: fix usage of capability for max number of virtqs
common/mlx5: extend virtq modifiable fields
vdpa/mlx5: pre-create virt
The driver wrongly takes the capability value for
the number of virtq pairs instead of just the number of virtqs.
Adjust all the usages of it to be the number of virtqs.
Fixes: c2eb33a ("vdpa/mlx5: manage virtqs by array")
Cc: sta...@dpdk.org
Signed-off-by: Li Zhang
---
drivers
From: Yajun Wu
Add device removal in function rte_eal_cleanup. This is the last chance
device remove get called for sanity. Loop vdev bus first and then all bus
for all device, calling rte_dev_remove.
Cc: sta...@dpdk.org
Signed-off-by: Yajun Wu
---
lib/eal/freebsd/eal.c | 33 +
From: Yajun Wu
Add calling rte_dev_remove in vDPA example application exit. Otherwise
rte_dev_remove never get called.
Fixes: edbed86d1cc ("examples/vdpa: introduce a new sample for vDPA")
Cc: sta...@dpdk.org
Signed-off-by: Yajun Wu
---
examples/vdpa/main.c | 4
1 file changed, 4 inserti
From: Yajun Wu
Move rte_eal_cleanup to function vdpa_sample_quit which
handling all example app quit.
Otherwise rte_eal_cleanup won't be called on receiving signal
like SIGINT(control + c).
Fixes: 10aa3757 ("examples: add eal cleanup to examples")
Cc: sta...@dpdk.org
Signed-off-by: Yajun Wu
--
From: Yajun Wu
Support set QP to RESET state.
Signed-off-by: Yajun Wu
---
drivers/common/mlx5/mlx5_devx_cmds.c | 7 +++
drivers/common/mlx5/mlx5_prm.h | 17 +
2 files changed, 24 insertions(+)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/
From: Yajun Wu
The motivation of this change is to reduce vDPA device queue creation
time by create some queue resource in vDPA device probe stage.
In VM live migration scenario, this can reduce 0.8ms for each queue
creation, thus reduce LM network downtime.
To create queue resource(umem/counte
From: Yajun Wu
The motivation of this change is to reduce vDPA device queue creation
time by create some queue resource in vDPA device probe stage.
In VM live migration scenario, this can reduce 0.8ms for each queue
creation, thus reduce LM network downtime.
To create queue resource(umem/counte
From: Yajun Wu
Support set QP to RESET state.
Signed-off-by: Yajun Wu
---
drivers/common/mlx5/mlx5_devx_cmds.c | 7 +++
drivers/common/mlx5/mlx5_prm.h | 17 +
2 files changed, 24 insertions(+)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/
From: Yajun Wu
To speed up queue create time, event qp and cq will create only once.
Each virtq creation will reuse same event qp and cq.
Because FW will set event qp to error state during virtq destroy,
need modify event qp to RESET state, then modify qp to RTS state as
usual. This can save abo
mode: event_mode/event_qpn_or_msix
Signed-off-by: Li Zhang
---
drivers/common/mlx5/mlx5_devx_cmds.c | 70 +++-
drivers/common/mlx5/mlx5_devx_cmds.h | 6 ++-
drivers/common/mlx5/mlx5_prm.h | 13 +-
3 files changed, 76 insertions(+), 13 deletions(-)
diff --git a
From: Yajun Wu
To speed up queue create time, event qp and cq will create only once.
Each virtq creation will reuse same event qp and cq.
Because FW will set event qp to error state during virtq destroy,
need modify event qp to RESET state, then modify qp to RTS state as
usual. This can save abo
accelerates the LM process and
reduces its time by 70%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 4 +
drivers/vdpa/mlx5/mlx5_vdpa_lm.c| 13 +-
drivers/vdpa/mlx5/mlx5_vdpa_virtq.c | 257 +---
3 files changed, 170 insertions(+), 104 deletions
mode: event_mode/event_qpn_or_msix
Signed-off-by: Li Zhang
---
drivers/common/mlx5/mlx5_devx_cmds.c | 70 +++-
drivers/common/mlx5/mlx5_devx_cmds.h | 6 ++-
drivers/common/mlx5/mlx5_prm.h | 13 +-
3 files changed, 76 insertions(+), 13 deletions(-)
diff --git a
accelerates the LM process and
reduces its time by 70%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 4 +
drivers/vdpa/mlx5/mlx5_vdpa_lm.c| 13 +-
drivers/vdpa/mlx5/mlx5_vdpa_virtq.c | 257 +---
3 files changed, 170 insertions(+), 104 deletions
.
Signed-off-by: Li Zhang
---
doc/guides/vdpadevs/mlx5.rst | 11 +++
drivers/vdpa/mlx5/meson.build | 1 +
drivers/vdpa/mlx5/mlx5_vdpa.c | 41
drivers/vdpa/mlx5/mlx5_vdpa.h | 36 +++
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 129
polling and
parallel configurations on the same virtq.
2.A doorbell lock synchronizes doorbell update,
which is shared for all the virtqs in the device.
3.A steering lock for the shared steering objects updates.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 24 ---
drivers
polling and
parallel configurations on the same virtq.
2.A doorbell lock synchronizes doorbell update,
which is shared for all the virtqs in the device.
3.A steering lock for the shared steering objects updates.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 24 ---
drivers
polls its ring and dequeue tasks.
That’s why the ring should be in multi-producer
and single consumer mode.
Anatomic counter manages the tasks completion notification.
The threads report errors to the caller by
a dedicated error counter per task.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5
.
Signed-off-by: Li Zhang
---
doc/guides/vdpadevs/mlx5.rst | 11 +++
drivers/vdpa/mlx5/meson.build | 1 +
drivers/vdpa/mlx5/mlx5_vdpa.c | 41
drivers/vdpa/mlx5/mlx5_vdpa.h | 36 +++
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 129
direct MRs in parallel using the MT mechanism.
After completion, the primary thread creates the indirect MR
needed for the following virtqs configurations.
This optimization accelerrate the LM process and
reduce its time by 5%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c
polls its ring and dequeue tasks.
That’s why the ring should be in multi-producer
and single consumer mode.
Anatomic counter manages the tasks completion notification.
The threads report errors to the caller by
a dedicated error counter per task.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5
The virtq object and all its sub-resources use a lot of
FW commands and can be accelerated by the MT management.
Split the virtqs creation between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h
Split the virtqs LM log between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 3 +
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 34 +++
drivers/vdpa/mlx5/mlx5_vdpa_lm.c | 85
direct MRs in parallel using the MT mechanism.
After completion, the primary thread creates the indirect MR
needed for the following virtqs configurations.
This optimization accelerrate the LM process and
reduce its time by 5%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c
Split the virtqs device close tasks after
stopping virt-queue between the configuration threads.
This accelerates the LM process and
reduces its time by 50%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 56 +--
drivers/vdpa/mlx5/mlx5_vdpa.h
The virtq object and all its sub-resources use a lot of
FW commands and can be accelerated by the MT management.
Split the virtqs creation between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h
Split the virtqs LM log between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 3 +
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 34 +++
drivers/vdpa/mlx5/mlx5_vdpa_lm.c | 85
Split the virtqs device close tasks after
stopping virt-queue between the configuration threads.
This accelerates the LM process and
reduces its time by 50%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 56 +--
drivers/vdpa/mlx5/mlx5_vdpa.h
pre-created virt-queue sub-resource in device probe stage
and then modify virtqueue in device config stage.
Steer table also need to support dummy virt-queue.
This accelerates the LM process and reduces its time by 40%.
Signed-off-by: Li Zhang
Signed-off-by: Yajun Wu
---
drivers/vdpa/mlx5
Split the virtqs virt-queue resource between
the configuration threads.
Also need pre-created virt-queue resource
after virtq destruction.
This accelerates the LM process and reduces its time by 30%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 115
pre-created virt-queue sub-resource in device probe stage
and then modify virtqueue in device config stage.
Steer table also need to support dummy virt-queue.
This accelerates the LM process and reduces its time by 40%.
Signed-off-by: Li Zhang
Signed-off-by: Yajun Wu
---
drivers/vdpa/mlx5
Split the virtqs virt-queue resource between
the configuration threads.
Also need pre-created virt-queue resource
after virtq destruction.
This accelerates the LM process and reduces its time by 30%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 115
1868
RFC ("Add vDPA multi-threads optiomization")
https://patchwork.dpdk.org/project/dpdk/cover/20220408075606.33056-1-l...@nvidia.com/
Li Zhang (12):
vdpa/mlx5: fix usage of capability for max number of virtqs
common/mlx5: extend virtq modifiable fields
vdpa/mlx5: pre-create virt
The driver wrongly takes the capability value for
the number of virtq pairs instead of just the number of virtqs.
Adjust all the usages of it to be the number of virtqs.
Fixes: c2eb33a ("vdpa/mlx5: manage virtqs by array")
Cc: sta...@dpdk.org
Signed-off-by: Li Zhang
---
drivers
From: Yajun Wu
Add device removal in function rte_eal_cleanup. This is the last chance
device remove get called for sanity. Loop vdev bus first and then all bus
for all device, calling rte_dev_remove.
Cc: sta...@dpdk.org
Signed-off-by: Yajun Wu
---
lib/eal/freebsd/eal.c | 33 +
From: Yajun Wu
Move rte_eal_cleanup to function vdpa_sample_quit which
handling all example app quit.
Otherwise rte_eal_cleanup won't be called on receiving signal
like SIGINT(control + c).
Fixes: 10aa3757 ("examples: add eal cleanup to examples")
Cc: sta...@dpdk.org
Signed-off-by: Yajun Wu
--
From: Yajun Wu
The motivation of this change is to reduce vDPA device queue creation
time by create some queue resource in vDPA device probe stage.
In VM live migration scenario, this can reduce 0.8ms for each queue
creation, thus reduce LM network downtime.
To create queue resource(umem/counte
From: Yajun Wu
Support set QP to RESET state.
Signed-off-by: Yajun Wu
---
drivers/common/mlx5/mlx5_devx_cmds.c | 7 +++
drivers/common/mlx5/mlx5_prm.h | 17 +
2 files changed, 24 insertions(+)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/
From: Yajun Wu
To speed up queue create time, event qp and cq will create only once.
Each virtq creation will reuse same event qp and cq.
Because FW will set event qp to error state during virtq destroy,
need modify event qp to RESET state, then modify qp to RTS state as
usual. This can save abo
mode: event_mode/event_qpn_or_msix
Signed-off-by: Li Zhang
---
drivers/common/mlx5/mlx5_devx_cmds.c | 70 +++-
drivers/common/mlx5/mlx5_devx_cmds.h | 6 ++-
drivers/common/mlx5/mlx5_prm.h | 13 +-
3 files changed, 76 insertions(+), 13 deletions(-)
diff --git a
accelerates the LM process and
reduces its time by 70%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 4 +
drivers/vdpa/mlx5/mlx5_vdpa_lm.c| 13 +-
drivers/vdpa/mlx5/mlx5_vdpa_virtq.c | 257 +---
3 files changed, 170 insertions(+), 104 deletions
polling and
parallel configurations on the same virtq.
2.A doorbell lock synchronizes doorbell update,
which is shared for all the virtqs in the device.
3.A steering lock for the shared steering objects updates.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c | 24 ---
drivers
.
Signed-off-by: Li Zhang
---
doc/guides/vdpadevs/mlx5.rst | 11 +++
drivers/vdpa/mlx5/meson.build | 1 +
drivers/vdpa/mlx5/mlx5_vdpa.c | 41
drivers/vdpa/mlx5/mlx5_vdpa.h | 36 +++
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 129
polls its ring and dequeue tasks.
That’s why the ring should be in multi-producer
and single consumer mode.
Anatomic counter manages the tasks completion notification.
The threads report errors to the caller by
a dedicated error counter per task.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5
direct MRs in parallel using the MT mechanism.
After completion, the primary thread creates the indirect MR
needed for the following virtqs configurations.
This optimization accelerrate the LM process and
reduce its time by 5%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.c
The virtq object and all its sub-resources use a lot of
FW commands and can be accelerated by the MT management.
Split the virtqs creation between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h
Split the virtqs LM log between the configuration threads.
This accelerates the LM process and reduces its time by 20%.
Signed-off-by: Li Zhang
---
drivers/vdpa/mlx5/mlx5_vdpa.h | 3 +
drivers/vdpa/mlx5/mlx5_vdpa_cthread.c | 34 +++
drivers/vdpa/mlx5/mlx5_vdpa_lm.c | 85
1 - 100 of 393 matches
Mail list logo
