The branch main has been updated by imp:
URL:
https://cgit.FreeBSD.org/src/commit/?id=d4959bfcd110ea471222c7dd87775ba1f4e3d1d9
commit d4959bfcd110ea471222c7dd87775ba1f4e3d1d9
Author: Warner Losh <i...@freebsd.org>
AuthorDate: 2023-08-25 16:10:08 +0000
Commit: Warner Losh <i...@freebsd.org>
CommitDate: 2023-08-25 16:11:15 +0000
nvme: Greatly improve error recovery
Next phase of error recovery: Eliminate the REOVERY_START phase, since
we don't need to wait to start recovery. Eliminate the RECOVERY_RESET
phase since it is transient, we now transition from RECOVERY_NORMAL into
RECOVERY_WAITING.
In normal mode, read the status of the controller. If it is in failed
state, or appears to be hot-plugged, jump directly to reset which will
sort out the proper things to do. This will cause all pending I/O to
complete with an abort status before the reset.
When in the NORMAL state, call the interrupt handler. This will complete
all pending transactions when interrupts are broken or temporarily
misbehaving. We then check all the pending completions for timeouts. If
we have abort enabled, then we'll send an abort. Otherwise we'll assume
the controller is wedged and needs a reset. By calling the interrupt
handler here, we'll avoid an issue with the current code where we
transitioned to RECOVERY_START which prevented any completions from
happening. Now completions happen. In addition and follow-on I/O that is
scheduled in the completion routines will be submitted, rather than
queued, because the recovery state is correct. This also fixes a problem
where I/O would timeout, but never complete, leading to hung I/O.
Resetting remains the same as before, just when we chose to reset has
changed.
A nice side effect of these changes is that we now do I/O when
interrupts to the card are totally broken. Followon commits will improve
the error reporting and logging when this happens. Performance will be
aweful, but will at least be minimally functional.
There is a small race when we're checking the completions if interrupts
are working, but this is handled in a future commit.
Sponsored by: Netflix
MFC After: 2 weeks
Differential Revision: https://reviews.freebsd.org/D36922
---
sys/dev/nvme/nvme_private.h | 2 -
sys/dev/nvme/nvme_qpair.c | 114 +++++++++++++++++++++++++-------------------
2 files changed, 66 insertions(+), 50 deletions(-)
diff --git a/sys/dev/nvme/nvme_private.h b/sys/dev/nvme/nvme_private.h
index dc69012cfd71..e4b319b9d8b7 100644
--- a/sys/dev/nvme/nvme_private.h
+++ b/sys/dev/nvme/nvme_private.h
@@ -149,8 +149,6 @@ struct nvme_tracker {
enum nvme_recovery {
RECOVERY_NONE = 0, /* Normal operations */
- RECOVERY_START, /* Deadline has passed, start
recovering */
- RECOVERY_RESET, /* This pass, initiate reset of
controller */
RECOVERY_WAITING, /* waiting for the reset to complete */
};
struct nvme_qpair {
diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c
index 0ad0b7cbe17f..6d34c7ddba2d 100644
--- a/sys/dev/nvme/nvme_qpair.c
+++ b/sys/dev/nvme/nvme_qpair.c
@@ -976,88 +976,106 @@ nvme_qpair_timeout(void *arg)
struct nvme_tracker *tr;
sbintime_t now;
bool idle;
- bool expired;
+ bool needs_reset;
uint32_t csts;
uint8_t cfs;
+
mtx_lock(&qpair->lock);
idle = TAILQ_EMPTY(&qpair->outstanding_tr);
-again:
switch (qpair->recovery_state) {
case RECOVERY_NONE:
+ /*
+ * Read csts to get value of cfs - controller fatal status. If
+ * we are in the hot-plug or controller failed status proceed
+ * directly to reset. We also bail early if the status reads all
+ * 1's or the control fatal status bit is now 1. The latter is
+ * always true when the former is true, but not vice versa. The
+ * intent of the code is that if the card is gone (all 1's) or
+ * we've failed, then try to do a reset (which someitmes
+ * unwedges a card reading all 1's that's not gone away, but
+ * usually doesn't).
+ */
+ csts = nvme_mmio_read_4(ctrlr, csts);
+ cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) &
NVME_CSTS_REG_CFS_MASK;
+ if (csts == NVME_GONE || cfs == 1)
+ goto do_reset;
+
+ /*
+ * Next, check to see if we have any completions. If we do,
+ * we've likely missed an interrupt, but the card is otherwise
+ * fine. This will also catch all the commands that are about
+ * to timeout (but there's still a tiny race). Since the timeout
+ * is long relative to the race between here and the check
below,
+ * this is still a win.
+ */
+ mtx_unlock(&qpair->lock);
+ nvme_qpair_process_completions(qpair);
+ mtx_lock(&qpair->lock);
+ if (qpair->recovery_state != RECOVERY_NONE) {
+ /*
+ * Somebody else adjusted recovery state while unlocked,
+ * we should bail. Unlock the qpair and return without
+ * doing anything else.
+ */
+ mtx_unlock(&qpair->lock);
+ return;
+ }
+
/*
* Check to see if we need to timeout any commands. If we do,
then
* we also enter a recovery phase.
*/
now = getsbinuptime();
- expired = false;
+ needs_reset = false;
TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq) {
if (tr->deadline == SBT_MAX)
continue;
- idle = false;
if (now > tr->deadline) {
- expired = true;
- nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
- nvme_abort_complete, tr);
+ if (tr->req->cb_fn != nvme_abort_complete &&
+ ctrlr->enable_aborts) {
+ /*
+ * This isn't an abort command, ask
+ * for a hardware abort.
+ */
+ nvme_ctrlr_cmd_abort(ctrlr, tr->cid,
+ qpair->id, nvme_abort_complete, tr);
+ } else {
+ /*
+ * Otherwise we have a live command in
+ * the card (either one we couldn't
+ * abort, or aborts weren't enabled).
+ * The only safe way to proceed is to do
+ * a reset.
+ */
+ needs_reset = true;
+ }
+ } else {
+ idle = false;
}
}
- if (!expired)
+ if (!needs_reset)
break;
/*
- * We're now passed our earliest deadline. We need to do
- * expensive things to cope, but next time. Flag that
- * and close the door to any further processing.
- */
- qpair->recovery_state = RECOVERY_START;
- nvme_printf(ctrlr, "RECOVERY_START %jd vs %jd\n",
- (uintmax_t)now, (uintmax_t)tr->deadline);
- /* FALLTHROUGH */
- case RECOVERY_START:
- /*
- * Read csts to get value of cfs - controller fatal status.
- * If no fatal status, try to call the completion routine, and
- * if completes transactions, report a missed interrupt and
- * return (this may need to be rate limited). Otherwise, if
- * aborts are enabled and the controller is not reporting
- * fatal status, abort the command. Otherwise, just reset the
- * controller and hope for the best.
- */
- csts = nvme_mmio_read_4(ctrlr, csts);
- cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) &
NVME_CSTS_REG_CFS_MASK;
- if (cfs) {
- nvme_printf(ctrlr, "Controller in fatal status,
resetting\n");
- qpair->recovery_state = RECOVERY_RESET;
- goto again;
- }
- mtx_unlock(&qpair->lock);
- if (nvme_qpair_process_completions(qpair)) {
- nvme_printf(ctrlr, "Completions present in output
without an interrupt\n");
- qpair->recovery_state = RECOVERY_NONE;
- } else {
- nvme_printf(ctrlr, "timeout with nothing complete,
resetting\n");
- qpair->recovery_state = RECOVERY_RESET;
- mtx_lock(&qpair->lock);
- goto again;
- }
- mtx_lock(&qpair->lock);
- break;
- case RECOVERY_RESET:
- /*
+ * We've had a command timeout that we weren't able to abort
+ *
* If we get here due to a possible surprise hot-unplug event,
* then we let nvme_ctrlr_reset confirm and fail the
* controller.
*/
+ do_reset:
nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n",
(csts == 0xffffffff) ? " and possible hot unplug" :
(cfs ? " and fatal error status" : ""));
nvme_printf(ctrlr, "RECOVERY_WAITING\n");
qpair->recovery_state = RECOVERY_WAITING;
nvme_ctrlr_reset(ctrlr);
+ idle = false; /* We want to keep polling */
break;
case RECOVERY_WAITING:
- nvme_printf(ctrlr, "waiting\n");
+ nvme_printf(ctrlr, "waiting for reset to complete\n");
break;
}
