Driver handling for broken devices

These are just a couple simple patches for when the device stops working. I think handling the unload was mentioned on another thread somewhere on this mailing list. I'm not particularly thrilled with this since a truely broken device will take 2 minutes per queue for the unload or hot-unplug to complete, but I think it is a step in the right direction.

Keith Busch (2):
NVMe: Clear timed out IO
NVMe: Schedule timeout for sync commands

 drivers/block/nvme-core.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)

This cancels a timed out IO only once and clears the timeout so we don't log the same problem every second. This may also prove useful if we implement async events, which don't have timeouts.

---
 drivers/block/nvme-core.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

index 40908a0..e64beaa 100644
@@ -878,9 +878,11 @@ static void nvme_cancel_ios(struct nvme_queue *nvmeq, bool timeout)
 			.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1),
 		};
 
-		if (timeout && !time_after(now, info[cmdid].timeout))
+		if (timeout && (!time_after(now, info[cmdid].timeout) ||
+							!info[cmdid].timeout))
 			continue;
 		dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d\n", cmdid);
+		info[cmdid].timeout = 0;
 		ctx = cancel_cmdid(nvmeq, cmdid, &fn);
 		fn(nvmeq->dev, ctx, &cqe);
 	}

1.7.0.4

Schedule a timeout on sync commands in case the command times out and the device is not being polled for timeouts. This prevents device removal from hanging forever if the device has stopped responding.

---
 drivers/block/nvme-core.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

index e64beaa..2660b6c 100644
@@ -731,7 +731,7 @@ int nvme_submit_sync_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd,
 
 	set_current_state(TASK_KILLABLE);
 	nvme_submit_cmd(nvmeq, cmd);
-	schedule();
+	schedule_timeout(timeout);
 
 	if (cmdinfo.status == -EINTR) {
 		nvme_abort_command(nvmeq, cmdid);

1.7.0.4