From 12cda920212a49fa22d9e8b9492ac4ea013310a4 Mon Sep 17 00:00:00 2001 From: Yonglong Liu Date: Wed, 5 Jun 2024 15:20:57 +0800 Subject: [PATCH 1/2] net: hns3: fix kernel crash problem in concurrent scenario When link status change, the nic driver need to notify the roce driver to handle this event, but at this time, the roce driver may uninit, then cause kernel crash. To fix the problem, when link status change, need to check whether the roce registered, and when uninit, need to wait link update finish. Fixes: 45e92b7e4e27 ("net: hns3: add calling roce callback function when link status change") Signed-off-by: Yonglong Liu Signed-off-by: Jijie Shao Reviewed-by: Simon Horman Signed-off-by: David S. Miller --- .../hisilicon/hns3/hns3pf/hclge_main.c | 21 ++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c index 43cc6ee4d87d..82574ce0194f 100644 --- a/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c +++ b/drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c @@ -3086,9 +3086,7 @@ static void hclge_push_link_status(struct hclge_dev *hdev) static void hclge_update_link_status(struct hclge_dev *hdev) { - struct hnae3_handle *rhandle = &hdev->vport[0].roce; struct hnae3_handle *handle = &hdev->vport[0].nic; - struct hnae3_client *rclient = hdev->roce_client; struct hnae3_client *client = hdev->nic_client; int state; int ret; @@ -3112,8 +3110,15 @@ static void hclge_update_link_status(struct hclge_dev *hdev) client->ops->link_status_change(handle, state); hclge_config_mac_tnl_int(hdev, state); - if (rclient && rclient->ops->link_status_change) - rclient->ops->link_status_change(rhandle, state); + + if (test_bit(HCLGE_STATE_ROCE_REGISTERED, &hdev->state)) { + struct hnae3_handle *rhandle = &hdev->vport[0].roce; + struct hnae3_client *rclient = hdev->roce_client; + + if (rclient && rclient->ops->link_status_change) + rclient->ops->link_status_change(rhandle, + state); + } hclge_push_link_status(hdev); } @@ -11319,6 +11324,12 @@ clear_roce: return ret; } +static bool hclge_uninit_need_wait(struct hclge_dev *hdev) +{ + return test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state) || + test_bit(HCLGE_STATE_LINK_UPDATING, &hdev->state); +} + static void hclge_uninit_client_instance(struct hnae3_client *client, struct hnae3_ae_dev *ae_dev) { @@ -11327,7 +11338,7 @@ static void hclge_uninit_client_instance(struct hnae3_client *client, if (hdev->roce_client) { clear_bit(HCLGE_STATE_ROCE_REGISTERED, &hdev->state); - while (test_bit(HCLGE_STATE_RST_HANDLING, &hdev->state)) + while (hclge_uninit_need_wait(hdev)) msleep(HCLGE_WAIT_RESET_DONE); hdev->roce_client->ops->uninit_instance(&vport->roce, 0); From 968fde83841a8c23558dfbd0a0c69d636db52b55 Mon Sep 17 00:00:00 2001 From: Jie Wang Date: Wed, 5 Jun 2024 15:20:58 +0800 Subject: [PATCH 2/2] net: hns3: add cond_resched() to hns3 ring buffer init process Currently hns3 ring buffer init process would hold cpu too long with big Tx/Rx ring depth. This could cause soft lockup. So this patch adds cond_resched() to the process. Then cpu can break to run other tasks instead of busy looping. Fixes: a723fb8efe29 ("net: hns3: refine for set ring parameters") Signed-off-by: Jie Wang Signed-off-by: Jijie Shao Reviewed-by: Simon Horman Signed-off-by: David S. Miller --- drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 4 ++++ drivers/net/ethernet/hisilicon/hns3/hns3_enet.h | 2 ++ 2 files changed, 6 insertions(+) diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c index ff71fb1eced9..a5fc0209d628 100644 --- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c +++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c @@ -3535,6 +3535,9 @@ static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring) ret = hns3_alloc_and_attach_buffer(ring, i); if (ret) goto out_buffer_fail; + + if (!(i % HNS3_RESCHED_BD_NUM)) + cond_resched(); } return 0; @@ -5107,6 +5110,7 @@ int hns3_init_all_ring(struct hns3_nic_priv *priv) } u64_stats_init(&priv->ring[i].syncp); + cond_resched(); } return 0; diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h index acd756b0c7c9..d36c4ed16d8d 100644 --- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h +++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h @@ -214,6 +214,8 @@ enum hns3_nic_state { #define HNS3_CQ_MODE_EQE 1U #define HNS3_CQ_MODE_CQE 0U +#define HNS3_RESCHED_BD_NUM 1024 + enum hns3_pkt_l2t_type { HNS3_L2_TYPE_UNICAST, HNS3_L2_TYPE_MULTICAST,