[PATCH] PCI/MSI: Fix restoring of MSI-X vector control's mask bit

[Jian-Hong Pan]

MSI-X vector control's bit 0 is the mask bit, which masks the
corresponding interrupt request, or not. Other reserved bits might be
used for other purpose by device vendors. For example, the values of
Kingston NVMe SSD's MSI-X vector control are neither 0, nor 1, but other
values [1].

The original restoring logic in system resuming uses the whole MSI-X
vector control value as the flag to set/clear the mask bit. However,
this logic conflicts the idea mentioned above. It may mislead system to
disable the MSI-X vector entries. That makes system get no interrupt
from Kingston NVMe SSD after resume and usually get NVMe I/O timeout
error.

[  174.715534] nvme nvme0: I/O 978 QID 3 timeout, completion polled

This patch takes only the mask bit of original MSI-X vector control
value as the flag to fix this issue.

[1] https://bugzilla.kernel.org/show_bug.cgi?id=204887#c8

Buglink: https://bugzilla.kernel.org/show_bug.cgi?id=204887
Fixed: f2440d9acbe8 ("PCI MSI: Refactor interrupt masking code")
Signed-off-by: Jian-Hong Pan <jian-hong@xxxxxxxxxxxx>
---
 drivers/pci/msi.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
index 0884bedcfc7a..deae3d5acaf6 100644
--- a/drivers/pci/msi.c
+++ b/drivers/pci/msi.c
@@ -433,6 +433,7 @@ static void __pci_restore_msi_state(struct pci_dev *dev)
 static void __pci_restore_msix_state(struct pci_dev *dev)
 {
 	struct msi_desc *entry;
+	u32 flag;
 
 	if (!dev->msix_enabled)
 		return;
@@ -444,8 +445,10 @@ static void __pci_restore_msix_state(struct pci_dev *dev)
 				PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
 
 	arch_restore_msi_irqs(dev);
-	for_each_pci_msi_entry(entry, dev)
-		msix_mask_irq(entry, entry->masked);
+	for_each_pci_msi_entry(entry, dev) {
+		flag = entry->masked & PCI_MSIX_ENTRY_CTRL_MASKBIT;
+		msix_mask_irq(entry, flag);
+	}
 
 	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
 }
-- 
2.23.0