Messages in this thread Patch in this message | | | From | Chunguang Xu <> | Subject | [PATCH] nvme: donot try to resubmit a canceled IO | Date | Thu, 11 May 2023 18:20:03 +0800 |
| |
From: "chunguang.xu" <chunguang.xu@shopee.com>
Now if NVMe over rdma and NVMe over TCP module detected controller is INACTIVE on the IO timeout path, it will try to call nvmf_complete_timed_out_request() to terminated this IO. But nvme_complete_rq() will continue to retry this IO, as the request_queue is quiescing at this time, if the target cannot connected or the host actively executes disconnect, it will cause this IO hang in the hctx dispatch queue and cannot be processed, resulting in hung task, the calltrace as followed:
[ 1575.570245] INFO: task kworker/u129:6:758 blocked for more than 966 seconds. [ 1575.577829] Tainted: G OE 5.4.0-77-shopee-generic #86+5 [ 1575.585323] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 1575.593670] kworker/u129:6 D 0 758 2 0x80004000 [ 1575.593681] Workqueue: nvme-wq nvme_scan_work [nvme_core] [ 1575.593683] Call Trace: [ 1575.593689] __schedule+0x2ee/0x750 [ 1575.593691] schedule+0x42/0xb0 [ 1575.593693] io_schedule+0x16/0x40 [ 1575.593696] do_read_cache_page+0x438/0x840 [ 1575.593698] ? __switch_to_asm+0x40/0x70 [ 1575.593700] ? file_fdatawait_range+0x30/0x30 [ 1575.593702] read_cache_page+0x12/0x20 [ 1575.593704] read_dev_sector+0x27/0xd0 [ 1575.593705] read_lba+0xbd/0x220 [ 1575.593707] ? kmem_cache_alloc_trace+0x1b0/0x240 [ 1575.593708] efi_partition+0x1e0/0x700 [ 1575.593710] ? vsnprintf+0x39e/0x4e0 [ 1575.593712] ? snprintf+0x49/0x60 [ 1575.593714] check_partition+0x154/0x250 [ 1575.593715] rescan_partitions+0xae/0x280 [ 1575.593718] bdev_disk_changed+0x5f/0x70 [ 1575.593719] __blkdev_get+0x3e3/0x550 [ 1575.593720] blkdev_get+0x3d/0x150 [ 1575.593722] __device_add_disk+0x329/0x480 [ 1575.593723] device_add_disk+0x13/0x20 [ 1575.593727] nvme_mpath_set_live+0x125/0x130 [nvme_core] [ 1575.593731] nvme_mpath_add_disk+0x11e/0x130 [nvme_core] [ 1575.593734] nvme_validate_ns+0x6a8/0x9d0 [nvme_core] [ 1575.593736] ? __switch_to_asm+0x40/0x70 [ 1575.593739] nvme_scan_work+0x1e0/0x350 [nvme_core] [ 1575.593743] process_one_work+0x1eb/0x3b0 [ 1575.593745] worker_thread+0x4d/0x400 [ 1575.593747] kthread+0x104/0x140 [ 1575.593748] ? process_one_work+0x3b0/0x3b0 [ 1575.593750] ? kthread_park+0x90/0x90 [ 1575.593751] ret_from_fork+0x1f/0x40
This issue seems not fixed on lastes kernel, try to fix it here.
Signed-off-by: chunguang.xu <chunguang.xu@shopee.com> --- drivers/nvme/host/core.c | 3 +++ drivers/nvme/host/fabrics.h | 1 + 2 files changed, 4 insertions(+)
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c index ccb6eb1282f8..bf9273081595 100644 --- a/drivers/nvme/host/core.c +++ b/drivers/nvme/host/core.c @@ -349,6 +349,9 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req) if (likely(nvme_req(req)->status == 0)) return COMPLETE; + if (nvme_req(req)->flags & NVME_REQ_CANCELLED) + return COMPLETE; + if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED) return AUTHENTICATE; diff --git a/drivers/nvme/host/fabrics.h b/drivers/nvme/host/fabrics.h index dcac3df8a5f7..40a5d6659af0 100644 --- a/drivers/nvme/host/fabrics.h +++ b/drivers/nvme/host/fabrics.h @@ -199,6 +199,7 @@ static inline void nvmf_complete_timed_out_request(struct request *rq) { if (blk_mq_request_started(rq) && !blk_mq_request_completed(rq)) { nvme_req(rq)->status = NVME_SC_HOST_ABORTED_CMD; + nvme_req(rq)->flags |= NVME_REQ_CANCELLED; blk_mq_complete_request(rq); } } -- 2.25.1
| |