*pos++ = '\0';
}
+static const struct hns3_dbg_item tx_spare_info_items[] = {
+ { "QUEUE_ID", 2 },
+ { "COPYBREAK", 2 },
+ { "LEN", 7 },
+ { "NTU", 4 },
+ { "NTC", 4 },
+ { "LTC", 4 },
+ { "DMA", 17 },
+};
+
+static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
+ int len, u32 ring_num, int *pos)
+{
+ char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN];
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+ char *result[ARRAY_SIZE(tx_spare_info_items)];
+ char content[HNS3_DBG_INFO_LEN];
+ u32 i, j;
+
+ if (!tx_spare) {
+ *pos += scnprintf(buf + *pos, len - *pos,
+ "tx spare buffer is not enabled\n");
+ return;
+ }
+
+ for (i = 0; i < ARRAY_SIZE(tx_spare_info_items); i++)
+ result[i] = &data_str[i][0];
+
+ *pos += scnprintf(buf + *pos, len - *pos, "tx spare buffer info\n");
+ hns3_dbg_fill_content(content, sizeof(content), tx_spare_info_items,
+ NULL, ARRAY_SIZE(tx_spare_info_items));
+ *pos += scnprintf(buf + *pos, len - *pos, "%s", content);
+
+ for (i = 0; i < ring_num; i++) {
+ j = 0;
+ sprintf(result[j++], "%8u", i);
+ sprintf(result[j++], "%9u", ring->tx_copybreak);
+ sprintf(result[j++], "%3u", tx_spare->len);
+ sprintf(result[j++], "%3u", tx_spare->next_to_use);
+ sprintf(result[j++], "%3u", tx_spare->next_to_clean);
+ sprintf(result[j++], "%3u", tx_spare->last_to_clean);
+ sprintf(result[j++], "%pad", &tx_spare->dma);
+ hns3_dbg_fill_content(content, sizeof(content),
+ tx_spare_info_items,
+ (const char **)result,
+ ARRAY_SIZE(tx_spare_info_items));
+ *pos += scnprintf(buf + *pos, len - *pos, "%s", content);
+ }
+}
+
static const struct hns3_dbg_item rx_queue_info_items[] = {
{ "QUEUE_ID", 2 },
{ "BD_NUM", 2 },
pos += scnprintf(buf + pos, len - pos, "%s", content);
}
+ hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos);
+
return 0;
}
module_param(debug, int, 0);
MODULE_PARM_DESC(debug, " Network interface message level setting");
+static unsigned int tx_spare_buf_size;
+module_param(tx_spare_buf_size, uint, 0400);
+MODULE_PARM_DESC(tx_spare_buf_size, "Size used to allocate tx spare buffer");
+
#define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
ops->request_update_promisc_mode(handle);
}
+static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
+{
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+ u32 ntc, ntu;
+
+ /* This smp_load_acquire() pairs with smp_store_release() in
+ * hns3_tx_spare_update() called in tx desc cleaning process.
+ */
+ ntc = smp_load_acquire(&tx_spare->last_to_clean);
+ ntu = tx_spare->next_to_use;
+
+ if (ntc > ntu)
+ return ntc - ntu - 1;
+
+ /* The free tx buffer is divided into two part, so pick the
+ * larger one.
+ */
+ return (ntc > (tx_spare->len - ntu) ? ntc :
+ (tx_spare->len - ntu)) - 1;
+}
+
+static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
+{
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+
+ if (!tx_spare ||
+ tx_spare->last_to_clean == tx_spare->next_to_clean)
+ return;
+
+ /* This smp_store_release() pairs with smp_load_acquire() in
+ * hns3_tx_spare_space() called in xmit process.
+ */
+ smp_store_release(&tx_spare->last_to_clean,
+ tx_spare->next_to_clean);
+}
+
+static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
+ struct sk_buff *skb,
+ u32 space)
+{
+ u32 len = skb->len <= ring->tx_copybreak ? skb->len :
+ skb_headlen(skb);
+
+ if (len > ring->tx_copybreak)
+ return false;
+
+ if (ALIGN(len, dma_get_cache_alignment()) > space) {
+ u64_stats_update_begin(&ring->syncp);
+ ring->stats.tx_spare_full++;
+ u64_stats_update_end(&ring->syncp);
+ return false;
+ }
+
+ return true;
+}
+
+static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
+{
+ struct hns3_tx_spare *tx_spare;
+ struct page *page;
+ dma_addr_t dma;
+ int order;
+
+ if (!tx_spare_buf_size)
+ return;
+
+ order = get_order(tx_spare_buf_size);
+ tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
+ GFP_KERNEL);
+ if (!tx_spare) {
+ /* The driver still work without the tx spare buffer */
+ dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
+ return;
+ }
+
+ page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
+ GFP_KERNEL, order);
+ if (!page) {
+ dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
+ devm_kfree(ring_to_dev(ring), tx_spare);
+ return;
+ }
+
+ dma = dma_map_page(ring_to_dev(ring), page, 0,
+ PAGE_SIZE << order, DMA_TO_DEVICE);
+ if (dma_mapping_error(ring_to_dev(ring), dma)) {
+ dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
+ put_page(page);
+ devm_kfree(ring_to_dev(ring), tx_spare);
+ return;
+ }
+
+ tx_spare->dma = dma;
+ tx_spare->buf = page_address(page);
+ tx_spare->len = PAGE_SIZE << order;
+ ring->tx_spare = tx_spare;
+}
+
+/* Use hns3_tx_spare_space() to make sure there is enough buffer
+ * before calling below function to allocate tx buffer.
+ */
+static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
+ unsigned int size, dma_addr_t *dma,
+ u32 *cb_len)
+{
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+ u32 ntu = tx_spare->next_to_use;
+
+ size = ALIGN(size, dma_get_cache_alignment());
+ *cb_len = size;
+
+ /* Tx spare buffer wraps back here because the end of
+ * freed tx buffer is not enough.
+ */
+ if (ntu + size > tx_spare->len) {
+ *cb_len += (tx_spare->len - ntu);
+ ntu = 0;
+ }
+
+ tx_spare->next_to_use = ntu + size;
+ if (tx_spare->next_to_use == tx_spare->len)
+ tx_spare->next_to_use = 0;
+
+ *dma = tx_spare->dma + ntu;
+
+ return tx_spare->buf + ntu;
+}
+
+static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
+{
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+
+ if (len > tx_spare->next_to_use) {
+ len -= tx_spare->next_to_use;
+ tx_spare->next_to_use = tx_spare->len - len;
+ } else {
+ tx_spare->next_to_use -= len;
+ }
+}
+
+static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
+ struct hns3_desc_cb *cb)
+{
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+ u32 ntc = tx_spare->next_to_clean;
+ u32 len = cb->length;
+
+ tx_spare->next_to_clean += len;
+
+ if (tx_spare->next_to_clean >= tx_spare->len) {
+ tx_spare->next_to_clean -= tx_spare->len;
+
+ if (tx_spare->next_to_clean) {
+ ntc = 0;
+ len = tx_spare->next_to_clean;
+ }
+ }
+
+ /* This tx spare buffer is only really reclaimed after calling
+ * hns3_tx_spare_update(), so it is still safe to use the info in
+ * the tx buffer to do the dma sync after tx_spare->next_to_clean
+ * is moved forword.
+ */
+ if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
+ dma_addr_t dma = tx_spare->dma + ntc;
+
+ dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
+ DMA_TO_DEVICE);
+ }
+}
+
static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
{
return 0;
dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
+ } else if (type & DESC_TYPE_BOUNCE_HEAD) {
+ /* Head data has been filled in hns3_handle_tx_bounce(),
+ * just return 0 here.
+ */
+ return 0;
} else {
skb_frag_t *frag = (skb_frag_t *)priv;
if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
DMA_TO_DEVICE);
+ else if (desc_cb->type &
+ (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
+ hns3_tx_spare_rollback(ring, desc_cb->length);
else if (desc_cb->length)
dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
DMA_TO_DEVICE);
desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
}
+static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
+ struct sk_buff *skb)
+{
+ struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
+ unsigned int type = DESC_TYPE_BOUNCE_HEAD;
+ unsigned int size = skb_headlen(skb);
+ dma_addr_t dma;
+ int bd_num = 0;
+ u32 cb_len;
+ void *buf;
+ int ret;
+
+ if (skb->len <= ring->tx_copybreak) {
+ size = skb->len;
+ type = DESC_TYPE_BOUNCE_ALL;
+ }
+
+ /* hns3_can_use_tx_bounce() is called to ensure the below
+ * function can always return the tx buffer.
+ */
+ buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
+
+ ret = skb_copy_bits(skb, 0, buf, size);
+ if (unlikely(ret < 0)) {
+ hns3_tx_spare_rollback(ring, cb_len);
+ u64_stats_update_begin(&ring->syncp);
+ ring->stats.copy_bits_err++;
+ u64_stats_update_end(&ring->syncp);
+ return ret;
+ }
+
+ desc_cb->priv = skb;
+ desc_cb->length = cb_len;
+ desc_cb->dma = dma;
+ desc_cb->type = type;
+
+ bd_num += hns3_fill_desc(ring, dma, size);
+
+ if (type == DESC_TYPE_BOUNCE_HEAD) {
+ ret = hns3_fill_skb_to_desc(ring, skb,
+ DESC_TYPE_BOUNCE_HEAD);
+ if (unlikely(ret < 0))
+ return ret;
+
+ bd_num += ret;
+ }
+
+ dma_sync_single_for_device(ring_to_dev(ring), dma, size,
+ DMA_TO_DEVICE);
+
+ u64_stats_update_begin(&ring->syncp);
+ ring->stats.tx_bounce++;
+ u64_stats_update_end(&ring->syncp);
+ return bd_num;
+}
+
+static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
+ struct sk_buff *skb)
+{
+ u32 space;
+
+ if (!ring->tx_spare)
+ goto out;
+
+ space = hns3_tx_spare_space(ring);
+
+ if (hns3_can_use_tx_bounce(ring, skb, space))
+ return hns3_handle_tx_bounce(ring, skb);
+
+out:
+ return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
+}
+
netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
{
struct hns3_nic_priv *priv = netdev_priv(netdev);
* zero, which is unlikely, and 'ret > 0' means how many tx desc
* need to be notified to the hw.
*/
- ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
+ ret = hns3_handle_desc_filling(ring, skb);
if (unlikely(ret <= 0))
goto fill_err;
tx_drop += ring->stats.tx_tso_err;
tx_drop += ring->stats.over_max_recursion;
tx_drop += ring->stats.hw_limitation;
+ tx_drop += ring->stats.copy_bits_err;
tx_errors += ring->stats.sw_err_cnt;
tx_errors += ring->stats.tx_vlan_err;
tx_errors += ring->stats.tx_l4_proto_err;
tx_errors += ring->stats.tx_tso_err;
tx_errors += ring->stats.over_max_recursion;
tx_errors += ring->stats.hw_limitation;
+ tx_errors += ring->stats.copy_bits_err;
} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
/* fetch the rx stats */
static void hns3_free_buffer(struct hns3_enet_ring *ring,
struct hns3_desc_cb *cb, int budget)
{
- if (cb->type & DESC_TYPE_SKB)
+ if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
+ DESC_TYPE_BOUNCE_ALL))
napi_consume_skb(cb->priv, budget);
else if (!HNAE3_IS_TX_RING(ring) && cb->pagecnt_bias)
__page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
ring_to_dma_dir(ring));
- else if (cb->length)
+ else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
ring_to_dma_dir(ring));
+ else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD))
+ hns3_tx_spare_reclaim_cb(ring, cb);
}
static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
desc_cb = &ring->desc_cb[ntc];
- if (desc_cb->type & DESC_TYPE_SKB) {
+ if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
+ DESC_TYPE_BOUNCE_HEAD)) {
(*pkts)++;
(*bytes) += desc_cb->send_bytes;
}
* ring_space called by hns3_nic_net_xmit.
*/
smp_store_release(&ring->next_to_clean, ntc);
+
+ hns3_tx_spare_update(ring);
+
return true;
}
ring = &priv->ring[q->tqp_index];
desc_num = priv->ae_handle->kinfo.num_tx_desc;
ring->queue_index = q->tqp_index;
+ ring->tx_copybreak = priv->tx_copybreak;
+ ring->last_to_use = 0;
} else {
ring = &priv->ring[q->tqp_index + queue_num];
desc_num = priv->ae_handle->kinfo.num_rx_desc;
ring->desc_num = desc_num;
ring->next_to_use = 0;
ring->next_to_clean = 0;
- ring->last_to_use = 0;
}
static void hns3_queue_to_ring(struct hnae3_queue *tqp,
ret = hns3_alloc_ring_buffers(ring);
if (ret)
goto out_with_desc;
+ } else {
+ hns3_init_tx_spare_buffer(ring);
}
return 0;
ring->next_to_use = 0;
ring->last_to_use = 0;
ring->pending_buf = 0;
- if (ring->skb) {
+ if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
dev_kfree_skb_any(ring->skb);
ring->skb = NULL;
+ } else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
+ struct hns3_tx_spare *tx_spare = ring->tx_spare;
+
+ dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
+ DMA_TO_DEVICE);
+ free_pages((unsigned long)tx_spare->buf,
+ get_order(tx_spare->len));
+ devm_kfree(ring_to_dev(ring), tx_spare);
+ ring->tx_spare = NULL;
}
}
HNS3_TQP_STAT("tso_err", tx_tso_err),
HNS3_TQP_STAT("over_max_recursion", over_max_recursion),
HNS3_TQP_STAT("hw_limitation", hw_limitation),
+ HNS3_TQP_STAT("bounce", tx_bounce),
+ HNS3_TQP_STAT("spare_full", tx_spare_full),
+ HNS3_TQP_STAT("copy_bits_err", copy_bits_err),
};
#define HNS3_TXQ_STATS_COUNT ARRAY_SIZE(hns3_txq_stats)
return 0;
}
+static int hns3_get_tunable(struct net_device *netdev,
+ const struct ethtool_tunable *tuna,
+ void *data)
+{
+ struct hns3_nic_priv *priv = netdev_priv(netdev);
+ int ret = 0;
+
+ switch (tuna->id) {
+ case ETHTOOL_TX_COPYBREAK:
+ /* all the tx rings have the same tx_copybreak */
+ *(u32 *)data = priv->tx_copybreak;
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+
+ return ret;
+}
+
+static int hns3_set_tunable(struct net_device *netdev,
+ const struct ethtool_tunable *tuna,
+ const void *data)
+{
+ struct hns3_nic_priv *priv = netdev_priv(netdev);
+ struct hnae3_handle *h = priv->ae_handle;
+ int i, ret = 0;
+
+ switch (tuna->id) {
+ case ETHTOOL_TX_COPYBREAK:
+ priv->tx_copybreak = *(u32 *)data;
+
+ for (i = 0; i < h->kinfo.num_tqps; i++)
+ priv->ring[i].tx_copybreak = priv->tx_copybreak;
+
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+
+ return ret;
+}
+
#define HNS3_ETHTOOL_COALESCE (ETHTOOL_COALESCE_USECS | \
ETHTOOL_COALESCE_USE_ADAPTIVE | \
ETHTOOL_COALESCE_RX_USECS_HIGH | \
.set_msglevel = hns3_set_msglevel,
.get_priv_flags = hns3_get_priv_flags,
.set_priv_flags = hns3_set_priv_flags,
+ .get_tunable = hns3_get_tunable,
+ .set_tunable = hns3_set_tunable,
};
static const struct ethtool_ops hns3_ethtool_ops = {
.get_priv_flags = hns3_get_priv_flags,
.set_priv_flags = hns3_set_priv_flags,
.get_ts_info = hns3_get_ts_info,
+ .get_tunable = hns3_get_tunable,
+ .set_tunable = hns3_set_tunable,
};
void hns3_ethtool_set_ops(struct net_device *netdev)
projects / linux-stable / commitdiff