DPDK patches and discussions
 help / color / mirror / Atom feed
From: fengchengwen <fengchengwen@huawei.com>
To: Vamsi Krishna <vattunuru@marvell.com>, <thomas@monjalon.net>,
	<bruce.richardson@intel.com>, <mb@smartsharesystems.com>
Cc: <dev@dpdk.org>, <kevin.laatz@intel.com>, <jerinj@marvell.com>,
	<conor.walsh@intel.com>, <gmuthukrishn@marvell.com>,
	<vvelumuri@marvell.com>,  <g.singh@nxp.com>,
	<sachin.saxena@oss.nxp.com>, <hemant.agrawal@nxp.com>,
	Amit Prakash Shukla <amitprakashs@marvell.com>
Subject: Re: [PATCH v4 1/1] dmadev: support priority configuration
Date: Tue, 8 Oct 2024 10:36:34 +0800	[thread overview]
Message-ID: <5d66d7e0-7d94-459c-96ee-4274183f5735@huawei.com> (raw)
In-Reply-To: <20241003115347.1365370-1-vattunuru@marvell.com>

Hi,

Sorry to reply now, because I just got back from our National Day vacation.

From the commit it seem confused about the dmadev and HW channel, Current one dmadev is corresponding a HW channel if its a hardware device.
So I suggest try not to mention HW channel, but the dmadev or DMA device.

Suggest the tile of the commit: dmadev: support strict priority configuration

On 2024/10/3 19:53, Vamsi Krishna wrote:
> From: Vamsi Attunuru <vattunuru@marvell.com>
> 
> Some DMA controllers offer the ability to configure priority level
> for the hardware command queues, allowing for the prioritization of

Let align the queues to channels
for the hardware DMA channels,

> DMA command execution based on queue importance.

based on channel importance.

> 
> This patch introduces the necessary fields in the dmadev structures to
> retrieve information about the hardware-supported priority levels and to
> enable priority configuration from the application.

This patch supports such strict priority configuration. If the dmadev supports, it should
declared the capability flag RTE_DMA_CAPA_PRI_POLICY_SP, then application could enable strict priority configuration.

> 
> Signed-off-by: Vamsi Attunuru <vattunuru@marvell.com>
> Signed-off-by: Amit Prakash Shukla <amitprakashs@marvell.com>
> ---
> V4 changes:
> * Rebased onto the latest
> 
> V3 changes:
> * Corrected patch title
> 
> V2 changes:
> * Reverted removed text from release_24_11.rst
> 
> V1 changes:
> * Added trace support
> * Added new capability flag
> 
> Deprecation notice:
> https://patches.dpdk.org/project/dpdk/patch/20240730144612.2132848-1-amitprakashs@marvell.com/
> 
> * Assuming we do not anticipate any advanced scheduling schemes for dmadev queues,
>   this patch is intended to support a strict priority scheme.
> 
>  doc/guides/rel_notes/release_24_11.rst |  8 ++++++++
>  lib/dmadev/rte_dmadev.c                | 15 +++++++++++++++
>  lib/dmadev/rte_dmadev.h                | 21 +++++++++++++++++++++
>  lib/dmadev/rte_dmadev_trace.h          |  2 ++
>  4 files changed, 46 insertions(+)
> 
> diff --git a/doc/guides/rel_notes/release_24_11.rst b/doc/guides/rel_notes/release_24_11.rst
> index e0a9aa55a1..9672d8c679 100644
> --- a/doc/guides/rel_notes/release_24_11.rst
> +++ b/doc/guides/rel_notes/release_24_11.rst
> @@ -67,6 +67,11 @@ New Features
>  
>    The new statistics are useful for debugging and profiling.
>  
> +* **Added strict priority capability flag in dmadev.**
> +
> +  Added new capability flag ``RTE_DMA_CAPA_PRI_POLICY_SP`` to check if the
> +  DMA device supports assigning fixed priority to its channels, allowing
> +  for better control over resource allocation and scheduling.

Please refine:

Add strict priority capability for dmadev.

Added new capability flag ``RTE_DMA_CAPA_PRI_POLICY_SP`` to check if the DMA device supports assigned fixed priority, allowing
for better control over resource allocation and scheduling.

>  
>  Removed Items
>  -------------
> @@ -112,6 +117,9 @@ ABI Changes
>     Also, make sure to start the actual text at the margin.
>     =======================================================
>  
> +* dmadev: Added ``nb_priorities`` field to ``rte_dma_info`` structure and
> +  ``priority`` field to ``rte_dma_conf`` structure to get device supported
> +  priority levels and configure required priority from the application.
>  
>  Known Issues
>  ------------
> diff --git a/lib/dmadev/rte_dmadev.c b/lib/dmadev/rte_dmadev.c
> index 845727210f..3d9063dee3 100644
> --- a/lib/dmadev/rte_dmadev.c
> +++ b/lib/dmadev/rte_dmadev.c
> @@ -497,6 +497,21 @@ rte_dma_configure(int16_t dev_id, const struct rte_dma_conf *dev_conf)
>  		return -EINVAL;
>  	}
>  
> +	if (dev_conf->priority && !(dev_info.dev_capa & RTE_DMA_CAPA_PRI_POLICY_SP)) {
> +		RTE_DMA_LOG(ERR, "Device %d don't support prioritization", dev_id);
> +		return -EINVAL;
> +	}
> +
> +	if (dev_info.nb_priorities == 1) {
> +		RTE_DMA_LOG(ERR, "Device %d must support more than 1 priority, or else 0", dev_id);
> +		return -EINVAL;
> +	}

Please consider other driver which don't set nb_priorities, then it will failed in the above branch.

Suggest add this verify in rte_dma_info_get(), make sure that this field should be > 1 if it supported PRI_POLICY_SP.

> +
> +	if (dev_info.nb_priorities && (dev_conf->priority >= dev_info.nb_priorities)) {
> +		RTE_DMA_LOG(ERR, "Device %d configure invalid priority", dev_id);
> +		return -EINVAL;
> +	}
> +
>  	if (*dev->dev_ops->dev_configure == NULL)
>  		return -ENOTSUP;
>  	ret = (*dev->dev_ops->dev_configure)(dev, dev_conf,
> diff --git a/lib/dmadev/rte_dmadev.h b/lib/dmadev/rte_dmadev.h
> index 5474a5281d..e5f730c327 100644
> --- a/lib/dmadev/rte_dmadev.h
> +++ b/lib/dmadev/rte_dmadev.h
> @@ -268,6 +268,16 @@ int16_t rte_dma_next_dev(int16_t start_dev_id);
>  #define RTE_DMA_CAPA_OPS_COPY_SG	RTE_BIT64(33)
>  /** Support fill operation. */
>  #define RTE_DMA_CAPA_OPS_FILL		RTE_BIT64(34)
> +/** Support strict prioritization at DMA HW channel level
> + *
> + * If device supports HW channel prioritization then application could
> + * assign fixed priority to the DMA HW channel using 'priority' field in
> + * struct rte_dma_conf. Number of supported priority levels will be known
> + * from 'nb_priorities' field in struct rte_dma_info.
> + *
> + * DMA devices which support prioritization can advertise this capability.

How about:
Support strict priority scheduling.

Application could assign fixed priority to the DMA device using 'priority' field in
struct rte_dma_conf. Number of supported priority levels will be known from 'nb_priorities'
field in struct rte_dma_info.

> + */
> +#define RTE_DMA_CAPA_PRI_POLICY_SP	RTE_BIT64(35)

This capa is a control plane flag, so please add after RTE_DMA_CAPA_M2D_AUTO_FREE.

>  /**@}*/
>  
>  /**
> @@ -297,6 +307,10 @@ struct rte_dma_info {
>  	int16_t numa_node;
>  	/** Number of virtual DMA channel configured. */
>  	uint16_t nb_vchans;
> +	/** Number of priority levels (must be > 1), if supported by DMA HW channel.
> +	 * 0 otherwise.

How about "Number of priority levels (must be > 1) if supported priority scheduling."

A DMA HW channel was just one dmadev, suggest don't introduce it.

> +	 */
> +	uint16_t nb_priorities;
>  };
>  
>  /**
> @@ -332,6 +346,13 @@ struct rte_dma_conf {
>  	 * @see RTE_DMA_CAPA_SILENT
>  	 */
>  	bool enable_silent;
> +	/* The priority of the DMA HW channel.
> +	 * This value cannot be greater than or equal to the field 'nb_priorities'
> +	 * of struct rte_dma_info which get from rte_dma_info_get().
> +	 * Among the values between '0' and 'nb_priorities - 1', lowest value
> +	 * indicates higher priority and vice-versa.

How about (you could retouch it. I'm not a native English speaker.):
The priority of the DMA device.
If the the DMA device don't support priority scheduling, this value should be zero.
Otherwise, the value should lower than the field 'nb_priorities' of struct rte_dma_info
which get from rte_dma_info_get(). And also lowest value indicates higher priority and vice-versa.

> +	 */
> +	uint16_t priority;
>  };
>  
>  /**
> diff --git a/lib/dmadev/rte_dmadev_trace.h b/lib/dmadev/rte_dmadev_trace.h
> index e55c4c6091..be089c065c 100644
> --- a/lib/dmadev/rte_dmadev_trace.h
> +++ b/lib/dmadev/rte_dmadev_trace.h
> @@ -35,6 +35,7 @@ RTE_TRACE_POINT(
>  	rte_trace_point_emit_u16(dev_info->max_sges);
>  	rte_trace_point_emit_i16(dev_info->numa_node);
>  	rte_trace_point_emit_u16(dev_info->nb_vchans);
> +	rte_trace_point_emit_u16(dev_info->nb_priorities);
>  )
>  
>  RTE_TRACE_POINT(
> @@ -48,6 +49,7 @@ RTE_TRACE_POINT(
>  	int enable_silent = (int)dev_conf->enable_silent;
>  	rte_trace_point_emit_i16(dev_id);
>  	rte_trace_point_emit_u16(dev_conf->nb_vchans);
> +	rte_trace_point_emit_u16(dev_conf->priority);
>  	rte_trace_point_emit_int(enable_silent);
>  	rte_trace_point_emit_int(ret);
>  )

Please also modify dma_capability_name and dmadev_handle_dev_info,


  parent reply	other threads:[~2024-10-08  2:36 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-09-13 12:10 [RFC v0 1/1] dmadev: provide priority configuration support Vamsi Krishna
2024-09-14  9:41 ` fengchengwen
2024-09-16 16:34   ` [EXTERNAL] " Vamsi Krishna Attunuru
2024-09-18  9:40 ` [RFC v1 1/1] dmadev: support priority configuration Vamsi Krishna
2024-09-18 10:13   ` [RFC v2 " Vamsi Krishna
2024-10-03 11:41     ` [PATCH v3 " Vamsi Krishna
2024-10-03 11:53       ` [PATCH v4 " Vamsi Krishna
2024-10-03 13:07         ` [EXTERNAL] " Anoob Joseph
2024-10-03 13:15           ` Vamsi Krishna Attunuru
2024-10-03 13:31             ` Vamsi Krishna Attunuru
2024-10-03 14:28             ` Anoob Joseph
2024-10-03 15:48               ` Vamsi Krishna Attunuru
2024-10-08  2:36         ` fengchengwen [this message]
2024-10-11  3:37           ` [EXTERNAL] " Vamsi Krishna Attunuru
2024-10-11  9:13       ` [PATCH v5 1/1] dmadev: support strict " Vamsi Krishna
2024-10-11  9:35         ` fengchengwen
2024-10-11 10:12         ` [PATCH v6 " Vamsi Krishna
2024-10-17 17:39           ` Thomas Monjalon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=5d66d7e0-7d94-459c-96ee-4274183f5735@huawei.com \
    --to=fengchengwen@huawei.com \
    --cc=amitprakashs@marvell.com \
    --cc=bruce.richardson@intel.com \
    --cc=conor.walsh@intel.com \
    --cc=dev@dpdk.org \
    --cc=g.singh@nxp.com \
    --cc=gmuthukrishn@marvell.com \
    --cc=hemant.agrawal@nxp.com \
    --cc=jerinj@marvell.com \
    --cc=kevin.laatz@intel.com \
    --cc=mb@smartsharesystems.com \
    --cc=sachin.saxena@oss.nxp.com \
    --cc=thomas@monjalon.net \
    --cc=vattunuru@marvell.com \
    --cc=vvelumuri@marvell.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).