Commit 6d9d3f87 authored Dec 09, 2020 by Jim Harris Committed by Tomasz Zawadzki Dec 11, 2020

bdev_nvme: don't try to read csts on fabrics cmd timeout



It is recommended to read CSTS when there is a timeout.
If CSTS.CFS (Controller Fatal Status) is set, we should
reset the controller.

But if an admin command on a fabrics controller times
out, reading CSTS submits another fabrics command that
could also timeout.  Even worse, we are recursively
polling the admin queue for completions in this case.

Fixes issue #1716.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I23d31f6302375c52eba6f4370748d622fbd25ca7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5513


Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>

parent 037f6bda

module/bdev/nvme/bdev_nvme.c

+12 −5

Original line number	Diff line number	Diff line
		@@ -1231,12 +1231,19 @@ timeout_cb(void cb_arg, struct spdk_nvme_ctrlr ctrlr,

		SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid);

		/* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
		* queue. (Note: qpair == NULL when there's an admin cmd timeout.) Otherwise we
		* would submit another fabrics cmd on the admin queue to read CSTS and check for its
		* completion recursively.
		*/
		if (nvme_bdev_ctrlr->connected_trid->trtype == SPDK_NVME_TRANSPORT_PCIE \|\| qpair != NULL) {
		csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
		if (csts.bits.cfs) {
		SPDK_ERRLOG("Controller Fatal Status, reset required\n");
		_bdev_nvme_reset(nvme_bdev_ctrlr, NULL);
		return;
		}
		}

		switch (g_opts.action_on_timeout) {
		case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT: