From 974c70e63f8827712e4173c55cda0aa05842cf4d Mon Sep 17 00:00:00 2001 From: Shuhei Matsumoto Date: Tue, 22 Dec 2020 15:49:25 +0900 Subject: [PATCH] lib/nvme: nvme_ctrlr_fail() return immediately if ctrlr is already failed As log shows 00:06:32.300 [2020-12-18 21:13:35.511014] nvme_ctrlr.c:1414:spdk_nvme_ctrlr_reset: *ERROR*: Controller reinitialization failed. 00:06:32.300 [2020-12-18 21:13:35.511104] nvme_ctrlr.c: 925:nvme_ctrlr_fail: *ERROR*: ctrlr 192.168.100.8 in failed state. 00:06:32.300 [2020-12-18 21:13:35.511132] bdev_nvme.c: 392:_bdev_nvme_reset_complete: *ERROR*: Resetting controller failed. 00:06:32.300 [2020-12-18 21:13:35.511240] nvme_ctrlr.c: 925:nvme_ctrlr_fail: *ERROR*: ctrlr 192.168.100.8 in failed state. 00:06:32.300 [2020-12-18 21:13:35.511511] bdev_nvme.c: 556:bdev_nvme_failover: *NOTICE*: Unable to perform reset, already in progress. if spdk_nvme_ctrlr_reset() failed, nvme_ctrlr_fail() is called, and then if spdk_nvme_ctrlr_process_admin_completions() failed, nvme_ctrlr_fail() is called. We don't know which one comes first but nvme_ctrlr_fail() should do nothing if the ctrlr is already failed. Hence we should avoid setting ctrlr->is_failed and calling nvme_transport_ctrlr_disconnect_qpair() twice. However we should set ctrlr->is_removed if the parameter hot_remove is true. We do these changes in this patch. Signed-off-by: Shuhei Matsumoto Change-Id: Iac37c892e054fb59d78e69346ca7f0575d596235 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5694 Community-CI: Broadcom CI Tested-by: SPDK CI Jenkins Reviewed-by: Aleksey Marchuk Reviewed-by: Changpeng Liu --- lib/nvme/nvme_ctrlr.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/lib/nvme/nvme_ctrlr.c b/lib/nvme/nvme_ctrlr.c index 69cf545c9..25ab7fef0 100644 --- a/lib/nvme/nvme_ctrlr.c +++ b/lib/nvme/nvme_ctrlr.c @@ -869,6 +869,12 @@ nvme_ctrlr_fail(struct spdk_nvme_ctrlr *ctrlr, bool hot_remove) if (hot_remove) { ctrlr->is_removed = true; } + + if (ctrlr->is_failed) { + SPDK_NOTICELOG("ctrlr %s is already in failed state\n", ctrlr->trid.traddr); + return; + } + ctrlr->is_failed = true; nvme_transport_ctrlr_disconnect_qpair(ctrlr, ctrlr->adminq); SPDK_ERRLOG("ctrlr %s in failed state.\n", ctrlr->trid.traddr);