DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ferruh Yigit <ferruh.yigit@intel.com>
To: Dmitry Kozlyuk <dkozlyuk@oss.nvidia.com>, <dev@dpdk.org>
Cc: Raslan Darawsheh <rasland@oss.nvidia.com>,
	Matan Azrad <matan@oss.nvidia.com>,
	Viacheslav Ovsiienko <viacheslavo@oss.nvidia.com>
Subject: Re: [dpdk-dev] [PATCH] net/mlx5: fix split buffer Rx
Date: Mon, 8 Nov 2021 13:15:56 +0000	[thread overview]
Message-ID: <53a46630-2028-b298-12c2-553119327b75@intel.com> (raw)
In-Reply-To: <dc27c6db-04eb-93c0-53fd-58bf014123ed@intel.com>

On 11/8/2021 1:10 PM, Ferruh Yigit wrote:
> On 11/1/2021 9:56 AM, Dmitry Kozlyuk wrote:
>> Routine to lookup LKey on Rx was assuming that the mbuf address
>> always belongs to a single mempool: the one associated with an RxQ
>> or the MPRQ mempool. This assumption is false for split buffers case.
>> A wrong LKey was looked up, resulting in completion errors.
>> Modify lookup routines to lookup LKey in the mbuf->pool
>> for non-MPRQ cases both on Rx datapath and on queue initialization.
>>
>> Fixes: fec28ca0e3a9 ("net/mlx5: support mempool registration")
>>
>> Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
>> Reviewed-by: Matan Azrad <matan@nvidia.com>
>> Reviewed-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
>> ---
>>   drivers/net/mlx5/mlx5_rx.c |  5 ++++-
>>   drivers/net/mlx5/mlx5_rx.h | 37 +++++++++++++++++++++++++++++++++++--
>>   2 files changed, 39 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/net/mlx5/mlx5_rx.c b/drivers/net/mlx5/mlx5_rx.c
>> index 258a645314..7296e81534 100644
>> --- a/drivers/net/mlx5/mlx5_rx.c
>> +++ b/drivers/net/mlx5/mlx5_rx.c
>> @@ -356,6 +356,7 @@ mlx5_rxq_initialize(struct mlx5_rxq_data *rxq)
>>           volatile struct mlx5_wqe_data_seg *scat;
>>           uintptr_t addr;
>>           uint32_t byte_count;
>> +        uint32_t lkey;
>>           if (mlx5_rxq_mprq_enabled(rxq)) {
>>               struct mlx5_mprq_buf *buf = (*rxq->mprq_bufs)[i];
>> @@ -366,6 +367,7 @@ mlx5_rxq_initialize(struct mlx5_rxq_data *rxq)
>>                                1 << rxq->strd_num_n);
>>               byte_count = (1 << rxq->strd_sz_n) *
>>                       (1 << rxq->strd_num_n);
>> +            lkey = mlx5_rx_addr2mr(rxq, addr);
>>           } else {
>>               struct rte_mbuf *buf = (*rxq->elts)[i];
>> @@ -373,13 +375,14 @@ mlx5_rxq_initialize(struct mlx5_rxq_data *rxq)
>>                       rxq->wqes)[i];
>>               addr = rte_pktmbuf_mtod(buf, uintptr_t);
>>               byte_count = DATA_LEN(buf);
>> +            lkey = mlx5_rx_mb2mr(rxq, buf);
>>           }
>>           /* scat->addr must be able to store a pointer. */
>>           MLX5_ASSERT(sizeof(scat->addr) >= sizeof(uintptr_t));
>>           *scat = (struct mlx5_wqe_data_seg){
>>               .addr = rte_cpu_to_be_64(addr),
>>               .byte_count = rte_cpu_to_be_32(byte_count),
>> -            .lkey = mlx5_rx_addr2mr(rxq, addr),
>> +            .lkey = lkey,
>>           };
>>       }
>>       rxq->consumed_strd = 0;
>> diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
>> index 4952fe1455..5903776e7a 100644
>> --- a/drivers/net/mlx5/mlx5_rx.h
>> +++ b/drivers/net/mlx5/mlx5_rx.h
>> @@ -266,7 +266,7 @@ uint16_t mlx5_rx_burst_mprq_vec(void *dpdk_rxq, struct rte_mbuf **pkts,
>>   static int mlx5_rxq_mprq_enabled(struct mlx5_rxq_data *rxq);
>>   /**
>> - * Query LKey from a packet buffer for Rx. No need to flush local caches
>> + * Query LKey for an address on Rx. No need to flush local caches
>>    * as the Rx mempool database entries are valid for the lifetime of the queue.
>>    *
>>    * @param rxq
>> @@ -301,7 +301,40 @@ mlx5_rx_addr2mr(struct mlx5_rxq_data *rxq, uintptr_t addr)
>>                        mr_ctrl, mp, addr);
>>   }
>> -#define mlx5_rx_mb2mr(rxq, mb) mlx5_rx_addr2mr(rxq, (uintptr_t)((mb)->buf_addr))
>> +/**
>> + * Query LKey from a packet buffer for Rx. No need to flush local caches
>> + * as the Rx mempool database entries are valid for the lifetime of the queue.
>> + *
>> + * @param rxq
>> + *   Pointer to Rx queue structure.
>> + * @param mb
>> + *   Buffer to search the address of.
>> + *
>> + * @return
>> + *   Searched LKey on success, UINT32_MAX on no match.
>> + *   This function always succeeds on valid input.
>> + */
>> +static __rte_always_inline uint32_t
>> +mlx5_rx_mb2mr(struct mlx5_rxq_data *rxq, struct rte_mbuf *mb)
>> +{
>> +    struct mlx5_mr_ctrl *mr_ctrl = &rxq->mr_ctrl;
>> +    uintptr_t addr = (uintptr_t)mb->buf_addr;
>> +    struct mlx5_rxq_ctrl *rxq_ctrl;
>> +    uint32_t lkey;
>> +
>> +    /* Linear search on MR cache array. */
>> +    lkey = mlx5_mr_lookup_lkey(mr_ctrl->cache, &mr_ctrl->mru,
>> +                   MLX5_MR_CACHE_N, addr);
>> +    if (likely(lkey != UINT32_MAX))
>> +        return lkey;
>> +    /*
>> +     * Slower search in the mempool database on miss.
>> +     * During queue creation rxq->sh is not yet set, so we use rxq_ctrl.
>> +     */
>> +    rxq_ctrl = container_of(rxq, struct mlx5_rxq_ctrl, rxq);
>> +    return mlx5_mr_mempool2mr_bh(&rxq_ctrl->priv->sh->cdev->mr_scache,
>> +                     mr_ctrl, mb->pool, addr);
> 
> Hi Dmitry,
> 
> I am getting build errors with this patch, although I can see CI build is passed,
> can you please check if I am missing something, build error is:
> 
> ../drivers/net/mlx5/mlx5_rx.h: In function ‘mlx5_rx_mb2mr’:
> ../drivers/net/mlx5/mlx5_rx.h:354:47: error: ‘struct mlx5_rxq_ctrl’ has no member named ‘priv’
>    354 |         return mlx5_mr_mempool2mr_bh(&rxq_ctrl->priv->sh->cdev->mr_scache,
>        |                                               ^~
> 
> 

My bad, Raslan pointed out that there is a v2 already, will continue with that one.



      parent reply	other threads:[~2021-11-08 13:16 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-01  9:56 Dmitry Kozlyuk
2021-11-08 11:17 ` [dpdk-dev] [PATCH v2] " Dmitry Kozlyuk
2021-11-08 12:57   ` Raslan Darawsheh
2021-11-08 13:10 ` [dpdk-dev] [PATCH] " Ferruh Yigit
2021-11-08 13:14   ` Dmitry Kozlyuk
2021-11-08 13:15   ` Ferruh Yigit [this message]

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=53a46630-2028-b298-12c2-553119327b75@intel.com \
    --to=ferruh.yigit@intel.com \
    --cc=dev@dpdk.org \
    --cc=dkozlyuk@oss.nvidia.com \
    --cc=matan@oss.nvidia.com \
    --cc=rasland@oss.nvidia.com \
    --cc=viacheslavo@oss.nvidia.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).