On Wed, 16 Oct 2019 23:03:08 +0200, Lorenzo Bianconi wrote:
> Refactor mvneta_rx_swbm code introducing mvneta_swbm_rx_frame and
> mvneta_swbm_add_rx_fragment routines. Rely on build_skb in oreder to
> allocate skb since the previous patch introduced buffer recycling using
> the page_pool API.
> This patch fixes even an issue in the original driver where dma buffers
> are accessed before dma sync.
> mvneta driver can run on not cache coherent devices so it is
> necessary to sync DMA buffers before sending them to the device
> in order to avoid memory corruptions. Running perf analysis we can
> see a performance cost associated with this DMA-sync (anyway it is
> already there in the original driver code). In follow up patches we
> will add more logic to reduce DMA-sync as much as possible.
>
> Signed-off-by: Ilias Apalodimas <ilias.apalodi...@linaro.org>
> Signed-off-by: Jesper Dangaard Brouer <bro...@redhat.com>
> Signed-off-by: Lorenzo Bianconi <lore...@kernel.org>
> ---
> drivers/net/ethernet/marvell/mvneta.c | 201 ++++++++++++++------------
> 1 file changed, 107 insertions(+), 94 deletions(-)
>
> diff --git a/drivers/net/ethernet/marvell/mvneta.c
> b/drivers/net/ethernet/marvell/mvneta.c
> index 9e4d828787fd..2f8123224f6e 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -323,6 +323,11 @@
> ETH_HLEN + ETH_FCS_LEN, \
> cache_line_size())
>
> +#define MVNETA_SKB_PAD (SKB_DATA_ALIGN(sizeof(struct skb_shared_info)
> + \
> + NET_SKB_PAD))
> +#define MVNETA_SKB_SIZE(len) (SKB_DATA_ALIGN(len) + MVNETA_SKB_PAD)
> +#define MVNETA_MAX_RX_BUF_SIZE (PAGE_SIZE - MVNETA_SKB_PAD)
> +
> #define IS_TSO_HEADER(txq, addr) \
> ((addr >= txq->tso_hdrs_phys) && \
> (addr < txq->tso_hdrs_phys + txq->size * TSO_HEADER_SIZE))
> @@ -646,7 +651,6 @@ static int txq_number = 8;
> static int rxq_def;
>
> static int rx_copybreak __read_mostly = 256;
> -static int rx_header_size __read_mostly = 128;
>
> /* HW BM need that each port be identify by a unique ID */
> static int global_port_id;
> @@ -1817,6 +1821,7 @@ static int mvneta_rx_refill(struct mvneta_port *pp,
> struct mvneta_rx_queue *rxq,
> gfp_t gfp_mask)
> {
> + enum dma_data_direction dma_dir;
> dma_addr_t phys_addr;
> struct page *page;
>
> @@ -1826,6 +1831,9 @@ static int mvneta_rx_refill(struct mvneta_port *pp,
> return -ENOMEM;
>
> phys_addr = page_pool_get_dma_addr(page) + pp->rx_offset_correction;
> + dma_dir = page_pool_get_dma_dir(rxq->page_pool);
> + dma_sync_single_for_device(pp->dev->dev.parent, phys_addr,
> + MVNETA_MAX_RX_BUF_SIZE, dma_dir);
This belongs in patch 2.
> mvneta_rx_desc_fill(rx_desc, phys_addr, page, rxq);
>
> return 0;
> @@ -1942,30 +1950,101 @@ int mvneta_rx_refill_queue(struct mvneta_port *pp,
> struct mvneta_rx_queue *rxq)
> return i;
> }
> /* Main rx processing when using software buffer management */
> static int mvneta_rx_swbm(struct napi_struct *napi,
> struct mvneta_port *pp, int budget,
> struct mvneta_rx_queue *rxq)
> {
> - struct net_device *dev = pp->dev;
> - int rx_todo, rx_proc;
> - int refill = 0;
> - u32 rcvd_pkts = 0;
> - u32 rcvd_bytes = 0;
> + int rcvd_pkts = 0, rcvd_bytes = 0;
> + int rx_pending, refill, done = 0;
>
> /* Get number of received packets */
> - rx_todo = mvneta_rxq_busy_desc_num_get(pp, rxq);
> - rx_proc = 0;
> + rx_pending = mvneta_rxq_busy_desc_num_get(pp, rxq);
>
> /* Fairness NAPI loop */
> - while ((rcvd_pkts < budget) && (rx_proc < rx_todo)) {
> + while (done < budget && done < rx_pending) {
> struct mvneta_rx_desc *rx_desc = mvneta_rxq_next_desc_get(rxq);
> unsigned char *data;
> struct page *page;
> - dma_addr_t phys_addr;
> - u32 rx_status, index;
> - int rx_bytes, skb_size, copy_size;
> - int frag_num, frag_size, frag_offset;
> + int index;
>
> index = rx_desc - rxq->descs;
> page = (struct page *)rxq->buf_virt_addr[index];
> @@ -1973,98 +2052,33 @@ static int mvneta_rx_swbm(struct napi_struct *napi,
> /* Prefetch header */
> prefetch(data);
>
> - phys_addr = rx_desc->buf_phys_addr;
> - rx_status = rx_desc->status;
> - rx_proc++;
> rxq->refill_num++;
> + done++;
> +
> + if (rx_desc->status & MVNETA_RXD_FIRST_DESC) {
> + int err;
>
> - if (rx_status & MVNETA_RXD_FIRST_DESC) {
If I may nit pick - not removing the rx_status and dev local variables
could make this chunk a little bit more readable - could you experiment
in moving that to a separate patch?
> /* Check errors only for FIRST descriptor */
> - if (rx_status & MVNETA_RXD_ERR_SUMMARY) {
> + if (rx_desc->status & MVNETA_RXD_ERR_SUMMARY) {
> mvneta_rx_error(pp, rx_desc);
> - dev->stats.rx_errors++;
> + pp->dev->stats.rx_errors++;
> /* leave the descriptor untouched */
> continue;
> }
> - rx_bytes = rx_desc->data_size -
> - (ETH_FCS_LEN + MVNETA_MH_SIZE);
>
> - /* Allocate small skb for each new packet */
> - skb_size = max(rx_copybreak, rx_header_size);
> - rxq->skb = netdev_alloc_skb_ip_align(dev, skb_size);
> - if (unlikely(!rxq->skb)) {
> - netdev_err(dev,
> - "Can't allocate skb on queue %d\n",
> - rxq->id);
> - dev->stats.rx_dropped++;
> - rxq->skb_alloc_err++;
> + err = mvneta_swbm_rx_frame(pp, rx_desc, rxq, page);
> + if (err)
> continue;
> - }
> - copy_size = min(skb_size, rx_bytes);
> -
> - /* Copy data from buffer to SKB, skip Marvell header */
> - memcpy(rxq->skb->data, data + MVNETA_MH_SIZE,
> - copy_size);
> - skb_put(rxq->skb, copy_size);
> - rxq->left_size = rx_bytes - copy_size;
> -
> - mvneta_rx_csum(pp, rx_status, rxq->skb);
> - if (rxq->left_size == 0) {
> - int size = copy_size + MVNETA_MH_SIZE;
> -
> - dma_sync_single_range_for_cpu(dev->dev.parent,
> - phys_addr, 0,
> - size,
> - DMA_FROM_DEVICE);
> -
> - /* leave the descriptor and buffer untouched */
> - } else {
> - /* refill descriptor with new buffer later */
> - rx_desc->buf_phys_addr = 0;
> -
> - frag_num = 0;
> - frag_offset = copy_size + MVNETA_MH_SIZE;
> - frag_size = min(rxq->left_size,
> - (int)(PAGE_SIZE - frag_offset));
> - skb_add_rx_frag(rxq->skb, frag_num, page,
> - frag_offset, frag_size,
> - PAGE_SIZE);
> - page_pool_release_page(rxq->page_pool, page);
> - rxq->left_size -= frag_size;
> - }
> } else {
> - /* Middle or Last descriptor */
> if (unlikely(!rxq->skb)) {
> pr_debug("no skb for rx_status 0x%x\n",
> - rx_status);
> + rx_desc->status);
> continue;
> }
> - if (!rxq->left_size) {
> - /* last descriptor has only FCS */
> - /* and can be discarded */
> - dma_sync_single_range_for_cpu(dev->dev.parent,
> - phys_addr, 0,
> - ETH_FCS_LEN,
> - DMA_FROM_DEVICE);
> - /* leave the descriptor and buffer untouched */
> - } else {
> - /* refill descriptor with new buffer later */
> - rx_desc->buf_phys_addr = 0;
> -
> - frag_num = skb_shinfo(rxq->skb)->nr_frags;
> - frag_offset = 0;
> - frag_size = min(rxq->left_size,
> - (int)(PAGE_SIZE - frag_offset));
> - skb_add_rx_frag(rxq->skb, frag_num, page,
> - frag_offset, frag_size,
> - PAGE_SIZE);
> -
> - page_pool_release_page(rxq->page_pool, page);
> - rxq->left_size -= frag_size;
> - }
> + mvneta_swbm_add_rx_fragment(pp, rx_desc, rxq, page);
> } /* Middle or Last descriptor */
>
> - if (!(rx_status & MVNETA_RXD_LAST_DESC))
> + if (!(rx_desc->status & MVNETA_RXD_LAST_DESC))
> /* no last descriptor this time */
> continue;
>
> @@ -2080,13 +2094,12 @@ static int mvneta_rx_swbm(struct napi_struct *napi,
> rcvd_bytes += rxq->skb->len;
>
> /* Linux processing */
> - rxq->skb->protocol = eth_type_trans(rxq->skb, dev);
> + rxq->skb->protocol = eth_type_trans(rxq->skb, pp->dev);
>
> napi_gro_receive(napi, rxq->skb);
>
> /* clean uncomplete skb pointer in queue */
> rxq->skb = NULL;
> - rxq->left_size = 0;
> }
>
> mvneta_update_stats(pp, rcvd_pkts, rcvd_bytes, false);
