VMD MSI remapping is disabled by default for all the CPUs with 28c0 VMD
deviceID. We used to disable remapping because drives supported more
vectors than the VMD so the performance was better without remapping.
Now with CPUs that support more than 64 (128 VMD MSIx vectors for gen4
VMD) we no longer need to disable this feature.
Note, pci_msix_vec_count() failure is translated to ENODEV per typical
expectations that drivers may return ENODEV when some driver-known
fundamental detail of the device is missing.
Signed-off-by: Nirmal Patel <nirmal.patel@xxxxxxxxxxxxxxx>
Acked-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
v1->v2: Updating commit message.
v2->v3: Use VMD MSI count instead of cpu count.
v3->v4: Updating commit message.
---
drivers/pci/controller/vmd.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
index 769eedeb8802..ba63af70bb63 100644
--- a/drivers/pci/controller/vmd.c
+++ b/drivers/pci/controller/vmd.c
@@ -34,6 +34,8 @@
#define MB2_SHADOW_OFFSET 0x2000
#define MB2_SHADOW_SIZE 16
+#define VMD_MIN_MSI_VECTOR_COUNT 64
+
enum vmd_features {
/*
* Device may contain registers which hint the physical location of the
@@ -807,13 +809,20 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
sd->node = pcibus_to_node(vmd->dev->bus);
+ vmd->msix_count = pci_msix_vec_count(vmd->dev);
+ if (vmd->msix_count < 0)
+ return -ENODEV;
+
/*
* Currently MSI remapping must be enabled in guest passthrough mode
* due to some missing interrupt remapping plumbing. This is probably
* acceptable because the guest is usually CPU-limited and MSI
* remapping doesn't become a performance bottleneck.
+ * Disable MSI remapping only if supported by VMD hardware and when
+ * VMD MSI count is less than or equal to minimum MSI count.
*/
if (!(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
+ vmd->msix_count > VMD_MIN_MSI_VECTOR_COUNT ||
offset[0] || offset[1]) {
ret = vmd_alloc_irqs(vmd);
if (ret)
--
2.31.1
