From: Stefan Roese <sr@xxxxxxx>
commit 83dbf898a2d45289be875deb580e93050ba67529 upstream.
Masking all unused MSI-X entries is done to ensure that a crash kernel
starts from a clean slate, which correponds to the reset state of the
device as defined in the PCI-E specificion 3.0 and later:
Vector Control for MSI-X Table Entries
--------------------------------------
"00: Mask bit: When this bit is set, the function is prohibited from
sending a message using this MSI-X Table entry.
...
This bit’s state after reset is 1 (entry is masked)."
A Marvell NVME device fails to deliver MSI interrupts after trying to
enable MSI-X interrupts due to that masking. It seems to take the MSI-X
mask bits into account even when MSI-X is disabled.
While not specification compliant, this can be cured by moving the masking
into the success path, so that the MSI-X table entries stay in device reset
state when the MSI-X setup fails.
[ tglx: Move it into the success path, add comment and amend changelog ]
Fixes: aa8092c1d1f1 ("PCI/MSI: Mask all unused MSI-X entries")
Signed-off-by: Stefan Roese <sr@xxxxxxx>
Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: linux-pci@xxxxxxxxxxxxxxx
Cc: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
Cc: Michal Simek <michal.simek@xxxxxxxxxx>
Cc: Marek Vasut <marex@xxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Link: https://lore.kernel.org/r/20211210161025.3287927-1-sr@xxxxxxx
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
drivers/pci/msi.c | 13 ++++++++++---
1 file changed, 10 insertions(+), 3 deletions(-)
--- a/drivers/pci/msi.c
+++ b/drivers/pci/msi.c
@@ -796,9 +796,6 @@ static int msix_capability_init(struct p
goto out_disable;
}
- /* Ensure that all table entries are masked. */
- msix_mask_all(base, tsize);
-
ret = msix_setup_entries(dev, base, entries, nvec, affd);
if (ret)
goto out_disable;
@@ -821,6 +818,16 @@ static int msix_capability_init(struct p
/* Set MSI-X enabled bits and unmask the function */
pci_intx_for_msi(dev, 0);
dev->msix_enabled = 1;
+
+ /*
+ * Ensure that all table entries are masked to prevent
+ * stale entries from firing in a crash kernel.
+ *
+ * Done late to deal with a broken Marvell NVME device
+ * which takes the MSI-X mask bits into account even
+ * when MSI-X is disabled, which prevents MSI delivery.
+ */
+ msix_mask_all(base, tsize);
pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
pcibios_free_irq(dev);
