From mboxrd@z Thu Jan  1 00:00:00 1970
From: keith.busch@intel.com (Keith Busch)
Date: Tue, 29 Mar 2016 14:38:18 +0000
Subject: nvme device timeout
In-Reply-To: <1604554248.106551459228253752.JavaMail.weblogic@epmlwas03c>
References: <1604554248.106551459228253752.JavaMail.weblogic@epmlwas03c>
Message-ID: <20160329143818.GC27195@localhost.localdomain>


On Tue, Mar 29, 2016@05:10:53AM +0000, Amit Kumar Pandey wrote:
> I fear that test patch may not work for Tim's reported issue. Since nvme_ktherad will not process any CQ until reset_work is completed, it may not be able to complete identify cmd and will result into failure. 
> 
> static int nvme_kthread(void *data)
> /*
> * Skip controllers currently under reset.
> */
>  if (work_pending(&dev->reset_work) || work_busy(&dev->reset_work))
>   continue;

Thanks, good catch!

Have you confirmed that this controller is indeed relying on polling
completions during early init, or do you want to try a fix for kthread
polling? The test patch is appeneded below if you still need to confirm
it.

If the controller really doesn't work with legacy interrupts, would it
be possible to permanently set the PCI interrupt pin register to 0? You
shouldn't advertise the capability if it doesn't work, and the driver
would have used MSI-x if that was the case.

If we absolutely need to add a quirk, either polling or force using
MSI-x sound like possible solutions.

---
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 680f578..05eb152 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1371,12 +1371,6 @@ static int nvme_kthread(void *data)
 			int i;
 			u32 csts = readl(dev->bar + NVME_REG_CSTS);
 
-			/*
-			 * Skip controllers currently under reset.
-			 */
-			if (work_pending(&dev->reset_work) || work_busy(&dev->reset_work))
-				continue;
-
 			if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
 							csts & NVME_CSTS_CFS) {
 				if (queue_work(nvme_workq, &dev->reset_work)) {
@@ -1945,17 +1939,16 @@ static void nvme_reset_work(struct work_struct *work)
 	if (result)
 		goto out;
 
-	result = nvme_init_identify(&dev->ctrl);
+	dev->ctrl.event_limit = NVME_NR_AEN_COMMANDS;
+	result = nvme_dev_list_add(dev);
 	if (result)
 		goto out;
 
-	result = nvme_setup_io_queues(dev);
+	result = nvme_init_identify(&dev->ctrl);
 	if (result)
 		goto out;
 
-	dev->ctrl.event_limit = NVME_NR_AEN_COMMANDS;
-
-	result = nvme_dev_list_add(dev);
+	result = nvme_setup_io_queues(dev);
 	if (result)
 		goto out;
 
--