On Sat, Aug 09, 2025 at 08:28:39PM +0700, Ammar Faizi wrote: > On Sat, Aug 09, 2025 at 06:34:29AM +0200, Nam Cao wrote: > > On Sat, Aug 09, 2025 at 07:52:30AM +0900, Ammar Faizi wrote: > > > I can do that. Send me a git diff. I'll test it and back with the dmesg > > > output. > > > > That would be very helpful, thanks! > > > > Please bear with me, this may take a few iterations. > > > > Let's first try the below. > > I just got home from a family outing. A bit slow response. > > Here's the result: > > https://gist.github.com/ammarfaizi2/ef5f98123ed3868f8d64ed41662edd63#file-dmesg_pci_debug_001-txt-L853 Thanks! Here's the problem: [ 1.037223] pcieport 10000:e0:1d.0: __pci_enable_msix_range:840 err=hwsize The PCIe port driver enables interrupt, trying MSI-X first. However, the device does not support MSI-X, so it tries MSI instead, which triggers the WARN_ON() in VMD driver. What's strange is that, the VMD doc says: "Intel VMD only supports MSIx Interrupts from child devices and therefore the BIOS must enable PCIe Hot Plug and MSIx interrups" Is it lying to us? Can you please try: Revert d5c647b08ee0 ("PCI: vmd: Fix wrong kfree() in vmd_msi_free()") Revert d7d8ab87e3e7 ("PCI: vmd: Switch to msi_create_parent_irq_domain()") So that the driver is back to the original state before I touched it. And apply the diff below. This will tell us if my commit breaks the driver somehow, or VMD has been allowing MSI all this time. diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c index d9b893bf4e45..e99d8cefb78d 100644 --- a/drivers/pci/controller/vmd.c +++ b/drivers/pci/controller/vmd.c @@ -255,9 +255,15 @@ static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info, msi_alloc_info_t *arg) { struct msi_desc *desc = arg->desc; + struct pci_dev *pci_dev = msi_desc_to_pci_dev(desc); struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus); struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL); + if (!pci_msix_vec_count(pci_dev)) + pr_err("But VMD only supports MSIx Interrupts from child devices!\n"); + else + pr_err("MSI-X, looking good...\n"); + dump_stack(); if (!vmdirq) return -ENOMEM;