RE: [EXT] Re: [v1 04/10] bus/dpaa: optimize bman acquire/release

[Jun Yang <jun.yang@nxp.com>]

Hi,
> +     rte_memcpy(bm_bufs, mcr->acquire.bufs,
> +             sizeof(struct bm_buffer) * rst);
"mcr->acquire.bufs" is located in register space, it is supposed to improve performance to copy it to DDR(bm_bufs) to parse and read.

BR
Jun

-----Original Message-----
From: Stephen Hemminger <stephen@networkplumber.org> 
Sent: Wednesday, May 28, 2025 10:31 PM
To: Vanshika Shukla <vanshika.shukla@nxp.com>
Cc: dev@dpdk.org; Hemant Agrawal <hemant.agrawal@nxp.com>; Sachin Saxena <sachin.saxena@nxp.com>; Jun Yang <jun.yang@nxp.com>
Subject: [EXT] Re: [v1 04/10] bus/dpaa: optimize bman acquire/release

Caution: This is an external email. Please take care when clicking links or opening attachments. When in doubt, report the message using the 'Report this email' button


On Wed, 28 May 2025 16:09:28 +0530
vanshika.shukla@nxp.com wrote:

> +RTE_EXPORT_INTERNAL_SYMBOL(bman_acquire_fast)
> +int
> +bman_acquire_fast(struct bman_pool *pool, uint64_t *bufs, uint8_t 
> +num) {
> +     struct bman_portal *p = get_affine_portal();
> +     struct bm_mc_command *mcc;
> +     struct bm_mc_result *mcr;
> +     uint8_t i, rst;
> +     struct bm_hw_buf_desc bm_bufs[FSL_BM_BURST_MAX];
> +
> +#ifdef RTE_LIBRTE_DPAA_HWDEBUG
> +     if (!num || (num > FSL_BM_BURST_MAX))
> +             return -EINVAL;
> +     if (pool->params.flags & BMAN_POOL_FLAG_ONLY_RELEASE)
> +             return -EINVAL;
> +#endif
> +
> +     mcc = bm_mc_start(&p->p);
> +     mcc->acquire.bpid = pool->params.bpid;
> +     bm_mc_commit(&p->p, BM_MCC_VERB_CMD_ACQUIRE |
> +                     (num & BM_MCC_VERB_ACQUIRE_BUFCOUNT));
> +     while (!(mcr = bm_mc_result(&p->p)))
> +             ;
> +     rst = mcr->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT;
> +     if (unlikely(!rst))
> +             return 0;
> +
> +     rte_memcpy(bm_bufs, mcr->acquire.bufs,
> +             sizeof(struct bm_buffer) * rst);
> +
> +     for (i = 0; i < rst; i++)
> +             bufs[i] = bman_hw_extract_addr(&bm_bufs[i]);
> +

Why do you need multiple the local copy?
Does hw_extract_addr modify the buffers?