PCI: vmd: Do not disable MSI-X remapping if interrupt remapping is enabled by IOMMU

When enabling VMD in BIOS setup (Ice Lake Processor: Whitley platform),
the host OS cannot boot successfully with the following error message:

  nvme nvme0: I/O 12 QID 0 timeout, completion polled
  nvme nvme0: Shutdown timeout set to 6 seconds
  DMAR: DRHD: handling fault status reg 2
  DMAR: [INTR-REMAP] Request device [0x00:0x00.5] fault index 0xa00 [fault reason 0x25] Blocked a compatibility format interrupt request

The request device is the VMD controller:
  # lspci -s 0000:00.5 -nn
  0000:00:00.5 RAID bus controller [0104]: Intel Corporation Volume
  Management Device NVMe RAID Controller [8086:28c0] (rev 04)

`git bisect` points to this offending commit ee81ee8 ("PCI:
vmd: Disable MSI-X remapping when possible"), which disables VMD MSI
remapping. The IOMMU hardware blocks the compatibility format
interrupt request because Interrupt Remapping Enable Status (IRES) and
Extended Interrupt Mode Enable (EIME) are enabled. Please refer to
section "5.1.4 Interrupt-Remapping Hardware Operation" in Intel VT-d
spec.

To fix the issue, VMD driver still enables the interrupt remapping
irrespective of VMD_FEAT_CAN_BYPASS_MSI_REMAP if the IOMMU subsystem
enables the interrupt remapping.

Test configuration is shown as follows:
  * Two VMD controllers
    1. 8086:28c0 (Whitley's VMD)
    2. 8086:201d (Purley's VMD: The issue does not appear in this
       controller. Just make sure if any side effect occurs.)
  * w/wo intremap=off

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214219
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Adrian Huang <[email protected]>
Signed-off-by: Lorenzo Pieralisi <[email protected]>
Reviewed-by: Jon Derrick <[email protected]>
Cc: Jon Derrick <[email protected]>
Cc: Nirmal Patel <[email protected]>
Cc: Joerg Roedel <[email protected]>

Author: Adrian Huang

drivers/pci/controller/vmd.c

@@ -6,6 +6,7 @@
 
 #include <linux/device.h>
 #include <linux/interrupt.h>
+#include <linux/iommu.h>
 #include <linux/irq.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
@@ -765,7 +766,8 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 	 * acceptable because the guest is usually CPU-limited and MSI
 	 * remapping doesn't become a performance bottleneck.
 	 */
-	if (!(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
+	if (iommu_capable(vmd->dev->dev.bus, IOMMU_CAP_INTR_REMAP) ||
+	    !(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
 	    offset[0] || offset[1]) {
 		ret = vmd_alloc_irqs(vmd);
 		if (ret)