net: hns3: default enable tx bounce buffer when smmu enabled

The SMMU engine on HIP09 chip has a hardware issue.
SMMU pagetable prefetch features may prefetch and use a invalid PTE
even the PTE is valid at that time. This will cause the device trigger
fake pagefaults. The solution is to avoid prefetching by adding a
SYNC command when smmu mapping a iova. But the performance of nic has a
sharp drop. Then we do this workaround, always enable tx bounce buffer,
avoid mapping/unmapping on TX path.

This issue only affects HNS3, so we always enable
tx bounce buffer when smmu enabled to improve performance.

Fixes: 295ba232a8 ("net: hns3: add device version to replace pci revision")
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250722125423.1270673-5-shaojijie@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This commit is contained in:
Jijie Shao
2025-07-22 20:54:23 +08:00
committed by Paolo Abeni
parent b3e75c0bcc
commit 49ade8630f
2 changed files with 33 additions and 0 deletions

View File

@@ -11,6 +11,7 @@
#include <linux/irq.h>
#include <linux/ip.h>
#include <linux/ipv6.h>
#include <linux/iommu.h>
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/skbuff.h>
@@ -1039,6 +1040,8 @@ static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
{
u32 alloc_size = ring->tqp->handle->kinfo.tx_spare_buf_size;
struct net_device *netdev = ring_to_netdev(ring);
struct hns3_nic_priv *priv = netdev_priv(netdev);
struct hns3_tx_spare *tx_spare;
struct page *page;
dma_addr_t dma;
@@ -1080,6 +1083,7 @@ static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
tx_spare->buf = page_address(page);
tx_spare->len = PAGE_SIZE << order;
ring->tx_spare = tx_spare;
ring->tx_copybreak = priv->tx_copybreak;
return;
dma_mapping_error:
@@ -4874,6 +4878,30 @@ static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
devm_kfree(&pdev->dev, priv->tqp_vector);
}
static void hns3_update_tx_spare_buf_config(struct hns3_nic_priv *priv)
{
#define HNS3_MIN_SPARE_BUF_SIZE (2 * 1024 * 1024)
#define HNS3_MAX_PACKET_SIZE (64 * 1024)
struct iommu_domain *domain = iommu_get_domain_for_dev(priv->dev);
struct hnae3_ae_dev *ae_dev = hns3_get_ae_dev(priv->ae_handle);
struct hnae3_handle *handle = priv->ae_handle;
if (ae_dev->dev_version < HNAE3_DEVICE_VERSION_V3)
return;
if (!(domain && iommu_is_dma_domain(domain)))
return;
priv->min_tx_copybreak = HNS3_MAX_PACKET_SIZE;
priv->min_tx_spare_buf_size = HNS3_MIN_SPARE_BUF_SIZE;
if (priv->tx_copybreak < priv->min_tx_copybreak)
priv->tx_copybreak = priv->min_tx_copybreak;
if (handle->kinfo.tx_spare_buf_size < priv->min_tx_spare_buf_size)
handle->kinfo.tx_spare_buf_size = priv->min_tx_spare_buf_size;
}
static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
unsigned int ring_type)
{
@@ -5107,6 +5135,7 @@ int hns3_init_all_ring(struct hns3_nic_priv *priv)
int i, j;
int ret;
hns3_update_tx_spare_buf_config(priv);
for (i = 0; i < ring_num; i++) {
ret = hns3_alloc_ring_memory(&priv->ring[i]);
if (ret) {
@@ -5311,6 +5340,8 @@ static int hns3_client_init(struct hnae3_handle *handle)
priv->ae_handle = handle;
priv->tx_timeout_count = 0;
priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
priv->min_tx_copybreak = 0;
priv->min_tx_spare_buf_size = 0;
set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);

View File

@@ -596,6 +596,8 @@ struct hns3_nic_priv {
struct hns3_enet_coalesce rx_coal;
u32 tx_copybreak;
u32 rx_copybreak;
u32 min_tx_copybreak;
u32 min_tx_spare_buf_size;
};
union l3_hdr_info {