This is a note to let you know that I've just added the patch titled
net: hns3: default enable tx bounce buffer when smmu enabled
to the 6.6-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
net-hns3-default-enable-tx-bounce-buffer-when-smmu-e.patch
and it can be found in the queue-6.6 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit ceb0ea1cbece9d1be93ef301b5c5dc69ea3f26cc
Author: Peiyang Wang <wangpeiyang1@xxxxxxxxxx>
Date: Fri Oct 25 17:29:30 2024 +0800
net: hns3: default enable tx bounce buffer when smmu enabled
[ Upstream commit e6ab19443b36a45ebfb392775cb17d6a78dd07ea ]
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: 295ba232a8c3 ("net: hns3: add device version to replace pci revision")
Signed-off-by: Peiyang Wang <wangpeiyang1@xxxxxxxxxx>
Signed-off-by: Jian Shen <shenjian15@xxxxxxxxxx>
Signed-off-by: Jijie Shao <shaojijie@xxxxxxxxxx>
Signed-off-by: Paolo Abeni <pabeni@xxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 14d086b535a2d..1f9bbf13214fb 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -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>
@@ -1032,6 +1033,8 @@ 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;
@@ -1073,6 +1076,7 @@ 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:
@@ -4872,6 +4876,30 @@ 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)
{
@@ -5106,6 +5134,7 @@ 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) {
@@ -5310,6 +5339,8 @@ 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);
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
index d36c4ed16d8dd..caf7a4df85852 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
@@ -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 {
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
index 78181eea93c1c..e03ba5aa3d182 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
@@ -1928,6 +1928,31 @@ static int hns3_set_tx_spare_buf_size(struct net_device *netdev,
return ret;
}
+static int hns3_check_tx_copybreak(struct net_device *netdev, u32 copybreak)
+{
+ struct hns3_nic_priv *priv = netdev_priv(netdev);
+
+ if (copybreak < priv->min_tx_copybreak) {
+ netdev_err(netdev, "tx copybreak %u should be no less than %u!\n",
+ copybreak, priv->min_tx_copybreak);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+static int hns3_check_tx_spare_buf_size(struct net_device *netdev, u32 buf_size)
+{
+ struct hns3_nic_priv *priv = netdev_priv(netdev);
+
+ if (buf_size < priv->min_tx_spare_buf_size) {
+ netdev_err(netdev,
+ "tx spare buf size %u should be no less than %u!\n",
+ buf_size, priv->min_tx_spare_buf_size);
+ return -EINVAL;
+ }
+ return 0;
+}
+
static int hns3_set_tunable(struct net_device *netdev,
const struct ethtool_tunable *tuna,
const void *data)
@@ -1944,6 +1969,10 @@ static int hns3_set_tunable(struct net_device *netdev,
switch (tuna->id) {
case ETHTOOL_TX_COPYBREAK:
+ ret = hns3_check_tx_copybreak(netdev, *(u32 *)data);
+ if (ret)
+ return ret;
+
priv->tx_copybreak = *(u32 *)data;
for (i = 0; i < h->kinfo.num_tqps; i++)
@@ -1958,6 +1987,10 @@ static int hns3_set_tunable(struct net_device *netdev,
break;
case ETHTOOL_TX_COPYBREAK_BUF_SIZE:
+ ret = hns3_check_tx_spare_buf_size(netdev, *(u32 *)data);
+ if (ret)
+ return ret;
+
old_tx_spare_buf_size = h->kinfo.tx_spare_buf_size;
new_tx_spare_buf_size = *(u32 *)data;
netdev_info(netdev, "request to set tx spare buf size from %u to %u\n",
