[dpdk-dev] [PATCH] doc: add information for Rx burst functions in mlx5

[dpdk-dev] [PATCH] doc: add information for Rx burst functions in mlx5

[Asaf Penso]

mlx5 pmd supports various rx_burst functions.
Each function is enabled differently and supports different features.

Add more comprehensive information about each function.

Signed-off-by: Asaf Penso <asafp@nvidia.com>
---
 doc/guides/nics/mlx5.rst | 55 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 55 insertions(+)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 205e5ba..8c1ce15 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1645,3 +1645,58 @@ all flows with assistance of external tools.
    .. code-block:: console
 
        mlx_steering_dump.py -f <output_file>
+
+RX burst functions
+------------------
+
+This section describes the different rx_burst functions and provides an
+elaborated information.
+
+#. Function name             - rx_burst
+   How to enable             - rx_vec_en=0
+   Scatter support           - Yes
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - The maximal packet size should be set to be the
+                               same as the MTU size. The mbuf size can be
+                               according to the application needs since the
+                               pmd can perform the scatter.
+
+#. Function name             - rx_burst_vec
+   How to enable             - rx_vec_en=1 (default value)
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - No. Scatter is not supported, so it means
+                               that the mbuf must be with the same size of
+                               the MTU. In this case, maximal packet size ==
+                               mbuf size == MTU size
+   Fallback                  - In scatter or LRO fallback to rx_burst
+   Notes                     - Improves CPU utilization for single-core and
+                               improves performance
+
+#. Function name             - rx_burst_mprq
+   How to enable             - mprq_en=1
+   Scatter support           - No
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Notes                     - Saving PCI bandwidth and getting higher
+                               performance, especially for small packets.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
+
+#. Function name             - rx_burst_mprq_vec
+   How to enable             - rx_vec_en=1 (default mode) and mprq_en=1
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Fallback                  - In scatter or LRO fallback to rx_burst_mprq
+   Notes                     - Improves CPU utilization of regular
+                               rx_burst_mprq for multi-core and improves
+                               performance.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
\ No newline at end of file
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH] doc: add information for Rx burst functions in mlx5

[Slava Ovsiienko]

Hi, Asaf

There is one more devargs to manage MPRQ enable: "rxqs_min_mprq".
This parameters defines the number of Rx queues limit to enable MPRQ.
If there is not enough Rx queues (below the limit) then MPRQ is disabled.
By default, "rxqs_min_mprq" is 12. I suppose we should mention this in
the list being added.

With best regards, Slava

> -----Original Message-----
> From: Asaf Penso <asafp@nvidia.com>
> Sent: Sunday, November 22, 2020 14:36
> To: dev@dpdk.org
> Cc: NBU-Contact-Thomas Monjalon <thomas@monjalon.net>; Slava Ovsiienko
> <viacheslavo@nvidia.com>; Matan Azrad <matan@nvidia.com>; Raslan
> Darawsheh <rasland@nvidia.com>
> Subject: [PATCH] doc: add information for Rx burst functions in mlx5
> 
> mlx5 pmd supports various rx_burst functions.
> Each function is enabled differently and supports different features.
> 
> Add more comprehensive information about each function.
> 
> Signed-off-by: Asaf Penso <asafp@nvidia.com>
> ---
>  doc/guides/nics/mlx5.rst | 55
> ++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 55 insertions(+)
> 
> diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst index
> 205e5ba..8c1ce15 100644
> --- a/doc/guides/nics/mlx5.rst
> +++ b/doc/guides/nics/mlx5.rst
> @@ -1645,3 +1645,58 @@ all flows with assistance of external tools.
>     .. code-block:: console
> 
>         mlx_steering_dump.py -f <output_file>
> +
> +RX burst functions
> +------------------
> +
> +This section describes the different rx_burst functions and provides an
> +elaborated information.
> +
> +#. Function name             - rx_burst
> +   How to enable             - rx_vec_en=0
> +   Scatter support           - Yes
> +   RX error recovery support - Yes
> +   CQE compression support   - Yes
> +   Large MTU support         - The maximal packet size should be set to be the
> +                               same as the MTU size. The mbuf size can be
> +                               according to the application needs since the
> +                               pmd can perform the scatter.
> +
> +#. Function name             - rx_burst_vec
> +   How to enable             - rx_vec_en=1 (default value)
> +   Scatter support           - No
> +   RX error recovery support - Supported only in case CQE compression is off
> +   CQE compression support   - Yes
> +   Large MTU support         - No. Scatter is not supported, so it means
> +                               that the mbuf must be with the same size of
> +                               the MTU. In this case, maximal packet size ==
> +                               mbuf size == MTU size
> +   Fallback                  - In scatter or LRO fallback to rx_burst
> +   Notes                     - Improves CPU utilization for single-core and
> +                               improves performance
> +
> +#. Function name             - rx_burst_mprq
> +   How to enable             - mprq_en=1
> +   Scatter support           - No
> +   RX error recovery support - Yes
> +   CQE compression support   - Yes
> +   Large MTU support         - Yes. Needs to be turned on by specifying the
> +                               stride size
> +   Notes                     - Saving PCI bandwidth and getting higher
> +                               performance, especially for small packets.
> +                               Used mainly for multi-core use cases.
> +                               Uses externally attached buffers only.
> +
> +#. Function name             - rx_burst_mprq_vec
> +   How to enable             - rx_vec_en=1 (default mode) and mprq_en=1
> +   Scatter support           - No
> +   RX error recovery support - Supported only in case CQE compression is off
> +   CQE compression support   - Yes
> +   Large MTU support         - Yes. Needs to be turned on by specifying the
> +                               stride size
> +   Fallback                  - In scatter or LRO fallback to rx_burst_mprq
> +   Notes                     - Improves CPU utilization of regular
> +                               rx_burst_mprq for multi-core and improves
> +                               performance.
> +                               Used mainly for multi-core use cases.
> +                               Uses externally attached buffers only.
> \ No newline at end of file
> --
> 1.8.3.1

[dpdk-dev] [PATCH v2] doc: add information for Rx burst functions in mlx5

[Asaf Penso]

mlx5 pmd supports various rx_burst functions.
Each function is enabled differently and supports different features.

Add more comprehensive information about each function.

Signed-off-by: Asaf Penso <asafp@nvidia.com>
---
v2: for mprq added note for rxqs_min_mprq
---
 doc/guides/nics/mlx5.rst | 57 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 57 insertions(+)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 205e5ba..46cfc3f 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1645,3 +1645,60 @@ all flows with assistance of external tools.
    .. code-block:: console
 
        mlx_steering_dump.py -f <output_file>
+
+RX burst functions
+------------------
+
+This section describes the different rx_burst functions and provides an
+elaborated information.
+
+#. Function name             - rx_burst
+   How to enable             - rx_vec_en=0
+   Scatter support           - Yes
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - The maximal packet size should be set to be the
+                               same as the MTU size. The mbuf size can be
+                               according to the application needs since the
+                               pmd can perform the scatter.
+
+#. Function name             - rx_burst_vec
+   How to enable             - rx_vec_en=1 (default value)
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - No. Scatter is not supported, so it means
+                               that the mbuf must be with the same size of
+                               the MTU. In this case, maximal packet size ==
+                               mbuf size == MTU size
+   Fallback                  - In scatter or LRO fallback to rx_burst
+   Notes                     - Improves CPU utilization for single-core and
+                               improves performance
+
+#. Function name             - rx_burst_mprq
+   How to enable             - mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq   
+   Scatter support           - No
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Notes                     - Saving PCI bandwidth and getting higher
+                               performance, especially for small packets.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
+
+#. Function name             - rx_burst_mprq_vec
+   How to enable             - rx_vec_en=1 (default mode) and mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq   
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Fallback                  - In scatter or LRO fallback to rx_burst_mprq
+   Notes                     - Improves CPU utilization of regular
+                               rx_burst_mprq for multi-core and improves
+                               performance.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
-- 
1.8.3.1

[dpdk-dev] [PATCH v3] doc: add information for Rx burst functions in mlx5

[Asaf Penso]

mlx5 pmd supports various rx_burst functions.
Each function is enabled differently and supports different features.

Add more comprehensive information about each function.

Signed-off-by: Asaf Penso <asafp@nvidia.com>
---
v3: checkpatch cleanup
v2: for mprq added note for rxqs_min_mprq
---
 doc/guides/nics/mlx5.rst | 57 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 57 insertions(+)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 205e5ba..758351a 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1645,3 +1645,60 @@ all flows with assistance of external tools.
    .. code-block:: console
 
        mlx_steering_dump.py -f <output_file>
+
+RX burst functions
+------------------
+
+This section describes the different rx_burst functions and provides an
+elaborated information.
+
+#. Function name             - rx_burst
+   How to enable             - rx_vec_en=0
+   Scatter support           - Yes
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - The maximal packet size should be set to be the
+                               same as the MTU size. The mbuf size can be
+                               according to the application needs since the
+                               pmd can perform the scatter.
+
+#. Function name             - rx_burst_vec
+   How to enable             - rx_vec_en=1 (default value)
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - No. Scatter is not supported, so it means
+                               that the mbuf must be with the same size of
+                               the MTU. In this case, maximal packet size ==
+                               mbuf size == MTU size
+   Fallback                  - In scatter or LRO fallback to rx_burst
+   Notes                     - Improves CPU utilization for single-core and
+                               improves performance
+
+#. Function name             - rx_burst_mprq
+   How to enable             - mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq
+   Scatter support           - No
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Notes                     - Saving PCI bandwidth and getting higher
+                               performance, especially for small packets.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
+
+#. Function name             - rx_burst_mprq_vec
+   How to enable             - rx_vec_en=1 (default mode) and mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Fallback                  - In scatter or LRO fallback to rx_burst_mprq
+   Notes                     - Improves CPU utilization of regular
+                               rx_burst_mprq for multi-core and improves
+                               performance.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v3] doc: add information for Rx burst functions in mlx5

[Slava Ovsiienko]

> -----Original Message-----
> From: Asaf Penso <asafp@nvidia.com>
> Sent: Monday, November 23, 2020 10:05
> To: dev@dpdk.org
> Cc: NBU-Contact-Thomas Monjalon <thomas@monjalon.net>; Slava Ovsiienko
> <viacheslavo@nvidia.com>; Matan Azrad <matan@nvidia.com>; Raslan
> Darawsheh <rasland@nvidia.com>
> Subject: [PATCH v3] doc: add information for Rx burst functions in mlx5
> 
> mlx5 pmd supports various rx_burst functions.
> Each function is enabled differently and supports different features.
> 
> Add more comprehensive information about each function.
> 
> Signed-off-by: Asaf Penso <asafp@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>

[dpdk-dev] [PATCH v4] doc: add information for Rx burst functions in mlx5

[Asaf Penso]

mlx5 pmd supports various rx_burst functions.
Each function is enabled differently and supports different features.

Add more comprehensive information about each function.

Signed-off-by: Asaf Penso <asafp@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>

---
v4: update the fallback for mprq burst functions
v3: checkpatch cleanup
v2: for mprq added note for rxqs_min_mprq
---
 doc/guides/nics/mlx5.rst | 59 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 59 insertions(+)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 205e5ba..c5a0fee 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1645,3 +1645,62 @@ all flows with assistance of external tools.
    .. code-block:: console
 
        mlx_steering_dump.py -f <output_file>
+
+RX burst functions
+------------------
+
+This section describes the different rx_burst functions and provides an
+elaborated information.
+
+#. Function name             - rx_burst
+   How to enable             - rx_vec_en=0
+   Scatter support           - Yes
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - The maximal packet size should be set to be the
+                               same as the MTU size. The mbuf size can be
+                               according to the application needs since the
+                               pmd can perform the scatter.
+
+#. Function name             - rx_burst_vec
+   How to enable             - rx_vec_en=1 (default value)
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - No. Scatter is not supported, so it means
+                               that the mbuf must be with the same size of
+                               the MTU. In this case, maximal packet size ==
+                               mbuf size == MTU size
+   Fallback                  - In scatter or LRO fallback to rx_burst
+   Notes                     - Improves CPU utilization for single-core and
+                               improves performance
+
+#. Function name             - rx_burst_mprq
+   How to enable             - mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq
+   Scatter support           - No
+   RX error recovery support - Yes
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Fallback                  - In scatter fallback to rx_burst
+   Notes                     - Saving PCI bandwidth and getting higher
+                               performance, especially for small packets.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
+
+#. Function name             - rx_burst_mprq_vec
+   How to enable             - rx_vec_en=1 (default mode) and mprq_en=1
+                             - RX queues number should be >= rxqs_min_mprq
+   Scatter support           - No
+   RX error recovery support - Supported only in case CQE compression is off
+   CQE compression support   - Yes
+   Large MTU support         - Yes. Needs to be turned on by specifying the
+                               stride size
+   Fallback                  - In scatter fallback to rx_burst.
+                               In LRO fallback to rx_burst_mprq.
+   Notes                     - Improves CPU utilization of regular
+                               rx_burst_mprq for multi-core and improves
+                               performance.
+                               Used mainly for multi-core use cases.
+                               Uses externally attached buffers only.
-- 
1.8.3.1

[dpdk-dev] [PATCH v5] doc: add Rx functions limitations in mlx5 guide

[Thomas Monjalon]

From: Asaf Penso <asafp@nvidia.com>

The mlx5 PMD supports various Rx burst functions.
Each function is enabled differently and supports different features.

Signed-off-by: Asaf Penso <asafp@nvidia.com>
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
---
v5: fix formatting by using a table
---
 doc/guides/nics/mlx5.rst | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index e96aca21eb..6858e81f72 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1345,6 +1345,29 @@ Performance tuning
    - Configure per-lcore cache when creating Mempools for packet buffer.
    - Refrain from dynamically allocating/freeing memory in run-time.
 
+Rx burst functions
+------------------
+
+There are multiple Rx burst functions with different advantages and limitations.
+
+.. table:: Rx burst functions
+
+   +-------------------+------------------------+---------+-----------------+------+-------+
+   || Function Name    || Enabler               || Scatter|| Error Recovery || CQE || Large|
+   |                   |                        |         |                 || comp|| MTU  |
+   +===================+========================+=========+=================+======+=======+
+   | rx_burst          | rx_vec_en=0            |   Yes   | Yes             |  Yes |  Yes  |
+   +-------------------+------------------------+---------+-----------------+------+-------+
+   | rx_burst_vec      | rx_vec_en=1 (default)  |   No    | if CQE comp off |  Yes |  No   |
+   +-------------------+------------------------+---------+-----------------+------+-------+
+   | rx_burst_mprq     || mprq_en=1             |   No    | Yes             |  Yes |  Yes  |
+   |                   || RxQs >= rxqs_min_mprq |         |                 |      |       |
+   +-------------------+------------------------+---------+-----------------+------+-------+
+   | rx_burst_mprq_vec || rx_vec_en=1 (default) |   No    | if CQE comp off |  Yes |  Yes  |
+   |                   || mprq_en=1             |         |                 |      |       |
+   |                   || RxQs >= rxqs_min_mprq |         |                 |      |       |
+   +-------------------+------------------------+---------+-----------------+------+-------+
+
 .. _mlx5_offloads_support:
 
 Supported hardware offloads
-- 
2.28.0

Re: [dpdk-dev] [PATCH v5] doc: add Rx functions limitations in mlx5 guide

[Thomas Monjalon]

27/11/2020 00:55, Thomas Monjalon:
> From: Asaf Penso <asafp@nvidia.com>
> 
> The mlx5 PMD supports various Rx burst functions.
> Each function is enabled differently and supports different features.
> 
> Signed-off-by: Asaf Penso <asafp@nvidia.com>
> Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
> ---
> v5: fix formatting by using a table

Applied