There are some bugfix for the HNS3 ethernet driver
---
changeLog:
v2 -> v3:
- Fixed a syntax error in git message, suggested by Ratheesh Kannoth
v2: https://lore.kernel.org/all/[email protected]/
v1 -> v2:
- Fixed some syntax errors in git messages and comments, suggested by Michal Kubiak
- Optimized the code by using a flag, suggested by Michal Kubiak
v1: https://lore.kernel.org/all/[email protected]/
---
Jian Shen (1):
net: hns3: mark unexcuted loopback test result as UNEXECUTED
Jie Wang (1):
net: hns3: fix index limit to support all queue stats
Yonglong Liu (1):
net: hns3: fix kernel crash when devlink reload during pf
initialization
.../hns3/hns3_common/hclge_comm_tqp_stats.c | 2 +-
.../ethernet/hisilicon/hns3/hns3_ethtool.c | 19 +++++++++++++++++--
.../hisilicon/hns3/hns3pf/hclge_devlink.c | 5 +++--
3 files changed, 21 insertions(+), 5 deletions(-)
--
2.30.0
From: Jian Shen <[email protected]>
Currently, loopback test may be skipped when resetting, but the test
result will still show as 'PASS', because the driver doesn't set
ETH_TEST_FL_FAILED flag. Fix it by setting the flag and
initializating the value to UNEXECUTED.
Fixes: 4c8dab1c709c ("net: hns3: reconstruct function hns3_self_test")
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Jijie Shao <[email protected]>
---
.../ethernet/hisilicon/hns3/hns3_ethtool.c | 19 +++++++++++++++++--
1 file changed, 17 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
index 999a0ee162a6..941cb529d671 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
@@ -78,6 +78,9 @@ static const struct hns3_stats hns3_rxq_stats[] = {
#define HNS3_NIC_LB_TEST_NO_MEM_ERR 1
#define HNS3_NIC_LB_TEST_TX_CNT_ERR 2
#define HNS3_NIC_LB_TEST_RX_CNT_ERR 3
+#define HNS3_NIC_LB_TEST_UNEXECUTED 4
+
+static int hns3_get_sset_count(struct net_device *netdev, int stringset);
static int hns3_lp_setup(struct net_device *ndev, enum hnae3_loop loop, bool en)
{
@@ -418,18 +421,26 @@ static void hns3_do_external_lb(struct net_device *ndev,
static void hns3_self_test(struct net_device *ndev,
struct ethtool_test *eth_test, u64 *data)
{
+ int cnt = hns3_get_sset_count(ndev, ETH_SS_TEST);
struct hns3_nic_priv *priv = netdev_priv(ndev);
struct hnae3_handle *h = priv->ae_handle;
int st_param[HNAE3_LOOP_NONE][2];
bool if_running = netif_running(ndev);
+ int i;
+
+ /* initialize the loopback test result, avoid marking an unexcuted
+ * loopback test as PASS.
+ */
+ for (i = 0; i < cnt; i++)
+ data[i] = HNS3_NIC_LB_TEST_UNEXECUTED;
if (hns3_nic_resetting(ndev)) {
netdev_err(ndev, "dev resetting!");
- return;
+ goto failure;
}
if (!(eth_test->flags & ETH_TEST_FL_OFFLINE))
- return;
+ goto failure;
if (netif_msg_ifdown(h))
netdev_info(ndev, "self test start\n");
@@ -451,6 +462,10 @@ static void hns3_self_test(struct net_device *ndev,
if (netif_msg_ifdown(h))
netdev_info(ndev, "self test end\n");
+ return;
+
+failure:
+ eth_test->flags |= ETH_TEST_FL_FAILED;
}
static void hns3_update_limit_promisc_mode(struct net_device *netdev,
--
2.30.0
From: Jie Wang <[email protected]>
Currently, hns hardware supports more than 512 queues and the index limit
in hclge_comm_tqps_update_stats is wrong. So this patch removes it.
Fixes: 287db5c40d15 ("net: hns3: create new set of common tqp stats APIs for PF and VF reuse")
Signed-off-by: Jie Wang <[email protected]>
Signed-off-by: Jijie Shao <[email protected]>
---
.../ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
index f3c9395d8351..618f66d9586b 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
@@ -85,7 +85,7 @@ int hclge_comm_tqps_update_stats(struct hnae3_handle *handle,
hclge_comm_cmd_setup_basic_desc(&desc, HCLGE_OPC_QUERY_TX_STATS,
true);
- desc.data[0] = cpu_to_le32(tqp->index & 0x1ff);
+ desc.data[0] = cpu_to_le32(tqp->index);
ret = hclge_comm_cmd_send(hw, &desc, 1);
if (ret) {
dev_err(&hw->cmq.csq.pdev->dev,
--
2.30.0
From: Yonglong Liu <[email protected]>
The devlink reload process will access the hardware resources,
but the register operation is done before the hardware is initialized.
so, if process the devlink reload during initialization, may lead to kernel
crash. This patch fixes this by checking whether the NIC is initialized.
Fixes: b741269b2759 ("net: hns3: add support for registering devlink for PF")
Signed-off-by: Yonglong Liu <[email protected]>
Signed-off-by: Jijie Shao <[email protected]>
---
drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
index 9a939c0b217f..80db4f7b05f6 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
@@ -40,8 +40,9 @@ static int hclge_devlink_reload_down(struct devlink *devlink, bool netns_change,
struct pci_dev *pdev = hdev->pdev;
int ret;
- if (test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state)) {
- dev_err(&pdev->dev, "reset is handling\n");
+ if (test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state) ||
+ !test_bit(HCLGE_STATE_NIC_REGISTERED, &hdev->state)) {
+ dev_err(&pdev->dev, "reset is handling or driver removed\n");
return -EBUSY;
}
--
2.30.0
On Mon, Mar 18, 2024 at 09:29:48PM +0800, Jijie Shao wrote:
> From: Jian Shen <[email protected]>
>
> Currently, loopback test may be skipped when resetting, but the test
> result will still show as 'PASS', because the driver doesn't set
> ETH_TEST_FL_FAILED flag. Fix it by setting the flag and
> initializating the value to UNEXECUTED.
>
> Fixes: 4c8dab1c709c ("net: hns3: reconstruct function hns3_self_test")
> Signed-off-by: Jian Shen <[email protected]>
> Signed-off-by: Jijie Shao <[email protected]>
> ---
> .../ethernet/hisilicon/hns3/hns3_ethtool.c | 19 +++++++++++++++++--
> 1 file changed, 17 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
> index 999a0ee162a6..941cb529d671 100644
> --- a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
> +++ b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
> @@ -78,6 +78,9 @@ static const struct hns3_stats hns3_rxq_stats[] = {
> #define HNS3_NIC_LB_TEST_NO_MEM_ERR 1
> #define HNS3_NIC_LB_TEST_TX_CNT_ERR 2
> #define HNS3_NIC_LB_TEST_RX_CNT_ERR 3
> +#define HNS3_NIC_LB_TEST_UNEXECUTED 4
> +
> +static int hns3_get_sset_count(struct net_device *netdev, int stringset);
>
> static int hns3_lp_setup(struct net_device *ndev, enum hnae3_loop loop, bool en)
> {
> @@ -418,18 +421,26 @@ static void hns3_do_external_lb(struct net_device *ndev,
> static void hns3_self_test(struct net_device *ndev,
> struct ethtool_test *eth_test, u64 *data)
> {
> + int cnt = hns3_get_sset_count(ndev, ETH_SS_TEST);
> struct hns3_nic_priv *priv = netdev_priv(ndev);
> struct hnae3_handle *h = priv->ae_handle;
> int st_param[HNAE3_LOOP_NONE][2];
> bool if_running = netif_running(ndev);
> + int i;
> +
> + /* initialize the loopback test result, avoid marking an unexcuted
> + * loopback test as PASS.
> + */
> + for (i = 0; i < cnt; i++)
> + data[i] = HNS3_NIC_LB_TEST_UNEXECUTED;
>
> if (hns3_nic_resetting(ndev)) {
> netdev_err(ndev, "dev resetting!");
> - return;
> + goto failure;
> }
>
> if (!(eth_test->flags & ETH_TEST_FL_OFFLINE))
> - return;
> + goto failure;
>
> if (netif_msg_ifdown(h))
> netdev_info(ndev, "self test start\n");
> @@ -451,6 +462,10 @@ static void hns3_self_test(struct net_device *ndev,
>
> if (netif_msg_ifdown(h))
> netdev_info(ndev, "self test end\n");
> + return;
> +
> +failure:
> + eth_test->flags |= ETH_TEST_FL_FAILED;
> }
>
> static void hns3_update_limit_promisc_mode(struct net_device *netdev,
> --
> 2.30.0
>
Thanks!
Reviewed-by: Michal Kubiak <[email protected]>
On Mon, Mar 18, 2024 at 09:29:47PM +0800, Jijie Shao wrote:
> From: Yonglong Liu <[email protected]>
>
> The devlink reload process will access the hardware resources,
> but the register operation is done before the hardware is initialized.
> so, if process the devlink reload during initialization, may lead to kernel
This sentence still seems not so clear. How about:
"So, processing the devlink reload during initialization may lead to kernel
crash."
Thanks,
Michal
On Mon, Mar 18, 2024 at 09:29:46PM +0800, Jijie Shao wrote:
> From: Jie Wang <[email protected]>
>
> Currently, hns hardware supports more than 512 queues and the index limit
> in hclge_comm_tqps_update_stats is wrong. So this patch removes it.
>
> Fixes: 287db5c40d15 ("net: hns3: create new set of common tqp stats APIs for PF and VF reuse")
> Signed-off-by: Jie Wang <[email protected]>
> Signed-off-by: Jijie Shao <[email protected]>
> ---
> .../ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> index f3c9395d8351..618f66d9586b 100644
> --- a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> +++ b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> @@ -85,7 +85,7 @@ int hclge_comm_tqps_update_stats(struct hnae3_handle *handle,
> hclge_comm_cmd_setup_basic_desc(&desc, HCLGE_OPC_QUERY_TX_STATS,
> true);
>
> - desc.data[0] = cpu_to_le32(tqp->index & 0x1ff);
> + desc.data[0] = cpu_to_le32(tqp->index);
> ret = hclge_comm_cmd_send(hw, &desc, 1);
> if (ret) {
> dev_err(&hw->cmq.csq.pdev->dev,
> --
> 2.30.0
>
Thanks,
Reviewed-by: Michal Kubiak <[email protected]>
On Mon, Mar 18, 2024 at 7:05 PM Jijie Shao <[email protected]> wrote:
>
> From: Jie Wang <[email protected]>
>
> Currently, hns hardware supports more than 512 queues and the index limit
> in hclge_comm_tqps_update_stats is wrong. So this patch removes it.
>
> Fixes: 287db5c40d15 ("net: hns3: create new set of common tqp stats APIs for PF and VF reuse")
> Signed-off-by: Jie Wang <[email protected]>
> Signed-off-by: Jijie Shao <[email protected]>
Only question I have is whether this is a fix or an enhancement.
Nevertheless changes look good to me.
Reviewed-by: Kalesh AP <[email protected]>
> ---
> .../ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> index f3c9395d8351..618f66d9586b 100644
> --- a/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> +++ b/drivers/net/ethernet/hisilicon/hns3/hns3_common/hclge_comm_tqp_stats.c
> @@ -85,7 +85,7 @@ int hclge_comm_tqps_update_stats(struct hnae3_handle *handle,
> hclge_comm_cmd_setup_basic_desc(&desc, HCLGE_OPC_QUERY_TX_STATS,
> true);
>
> - desc.data[0] = cpu_to_le32(tqp->index & 0x1ff);
> + desc.data[0] = cpu_to_le32(tqp->index);
> ret = hclge_comm_cmd_send(hw, &desc, 1);
> if (ret) {
> dev_err(&hw->cmq.csq.pdev->dev,
> --
> 2.30.0
>
>
--
Regards,
Kalesh A P
Mon, Mar 18, 2024 at 02:29:47PM CET, [email protected] wrote:
>From: Yonglong Liu <[email protected]>
>
>The devlink reload process will access the hardware resources,
>but the register operation is done before the hardware is initialized.
>so, if process the devlink reload during initialization, may lead to kernel
>crash. This patch fixes this by checking whether the NIC is initialized.
Fix your locking, you should take devl_lock during your init. That would
disallow reload to race with it.
pw-bot: cr
>
>Fixes: b741269b2759 ("net: hns3: add support for registering devlink for PF")
>Signed-off-by: Yonglong Liu <[email protected]>
>Signed-off-by: Jijie Shao <[email protected]>
>---
> drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
>diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
>index 9a939c0b217f..80db4f7b05f6 100644
>--- a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
>+++ b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_devlink.c
>@@ -40,8 +40,9 @@ static int hclge_devlink_reload_down(struct devlink *devlink, bool netns_change,
> struct pci_dev *pdev = hdev->pdev;
> int ret;
>
>- if (test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state)) {
>- dev_err(&pdev->dev, "reset is handling\n");
>+ if (test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state) ||
>+ !test_bit(HCLGE_STATE_NIC_REGISTERED, &hdev->state)) {
>+ dev_err(&pdev->dev, "reset is handling or driver removed\n");
> return -EBUSY;
> }
>
>--
>2.30.0
>
>
on 2024/3/18 22:36, Kalesh Anakkur Purayil wrote:
> On Mon, Mar 18, 2024 at 7:05 PM Jijie Shao <[email protected]> wrote:
>> From: Jie Wang <[email protected]>
>>
>> Currently, hns hardware supports more than 512 queues and the index limit
>> in hclge_comm_tqps_update_stats is wrong. So this patch removes it.
>>
>> Fixes: 287db5c40d15 ("net: hns3: create new set of common tqp stats APIs for PF and VF reuse")
>> Signed-off-by: Jie Wang <[email protected]>
>> Signed-off-by: Jijie Shao <[email protected]>
> Only question I have is whether this is a fix or an enhancement.
> Nevertheless changes look good to me.
Actually, it's a fix. The capability of supporting more than 512 queues
was not implemented by modifying this statistics. this bug prevents us
from querying the statistics of more than 512 queues.
on 2024/3/19 19:17, Jiri Pirko wrote:
> Mon, Mar 18, 2024 at 02:29:47PM CET, [email protected] wrote:
>> From: Yonglong Liu <[email protected]>
>>
>> The devlink reload process will access the hardware resources,
>> but the register operation is done before the hardware is initialized.
>> so, if process the devlink reload during initialization, may lead to kernel
>> crash. This patch fixes this by checking whether the NIC is initialized.
> Fix your locking, you should take devl_lock during your init. That would
> disallow reload to race with it.
>
> pw-bot: cr
Thanks，
We have fixed this in v4.
Jijie Shao