@samitouri / QOSamiQemu / commits / bc034e8a95

vfio/pci: Add ats property

Add an "ats" OnOffAuto property to vfio-pci. When the device has an ATS extended capability in config space but we should not expose it (ats=off, or ats=auto and kernel reports IOMMU_HW_CAP_PCI_ATS_NOT_SUPPORTED), mask the capability so the guest does not see it. If ATS is explicitly requested but not supported by the kernel, fail device realize. This aligns with the kernel's per-device effective ATS reporting and allows vfio-pci to mask ATS when the host kernel reports ATS as unsupported. Emit a warning when ats=on is requested but the physical device does not advertise ATS, since ATS cannot be exposed to the guest in this case. Emit a warning when ats=auto, ats cap is present on the physical device, but kernel reports ATS as unsupported. Suggested-by: Shameer Kolothum <skolothumtho@nvidia.com> Signed-off-by: Nathan Chen <nathanc@nvidia.com> Reviewed-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Cédric Le Goater <clg@redhat.com> Link: https://lore.kernel.org/qemu-devel/20260623204943.989903-3-nathanc@nvidia.com Signed-off-by: Cédric Le Goater <clg@redhat.com>

Nathan Chen committed Jun 23, 2026 at 13:49 UTC bc034e8a959c07471d1214b3525bb68fccce1031

2 files changed +85 -4

hw/vfio/pci.c

+84 -4

@@ -2548,10 +2548,53 @@ static bool vfio_pci_synthesize_pasid_cap(VFIOPCIDevice *vdev, Error **errp)

2548 return true;

2549 }

2550

2551 -static void vfio_add_ext_cap(VFIOPCIDevice *vdev)

2551 +/*

2552 + * Determine whether ATS capability should be advertised for @vdev, based on

2553 + * whether it was enabled on the command line and whether it is supported

2554 + * according to the kernel.

2555 + *

2556 + * Store whether ATS capability should be advertised in @ats_needed.

2557 + *

2558 + * Returns false only when ats=on is explicitly requested but the kernel

2559 + * reports it is not supported. Returns true in all other cases.

2560 + */

2561 +static bool vfio_pci_ats_requested_and_supported(VFIOPCIDevice *vdev,

2562 + bool *ats_needed, Error **errp)

2563 +{

2564 + HostIOMMUDevice *hiod = vdev->vbasedev.hiod;

2565 + HostIOMMUDeviceClass *hiodc;

2566 + bool ats_supported;

2567 + *ats_needed = false;

2568 +

2569 + if (vdev->ats == ON_OFF_AUTO_OFF) {

2570 + return true;

2571 + }

2572 +

2573 + *ats_needed = true;

2574 + if (!hiod) {

2575 + return true;

2576 + }

2577 + hiodc = HOST_IOMMU_DEVICE_GET_CLASS(hiod);

2578 + if (!hiodc || !hiodc->support_ats) {

2579 + return true;

2580 + }

2581 +

2582 + ats_supported = hiodc->support_ats(hiod);

2583 + if (vdev->ats == ON_OFF_AUTO_ON && !ats_supported) {

2584 + error_setg(errp, "vfio-pci: ATS requested but not supported by kernel");

2585 + *ats_needed = false;

2586 + return false;

2587 + }

2588 +

2589 + *ats_needed = ats_supported;

2590 + return true;

2591 +}

2592 +

2593 +static void vfio_add_ext_cap(VFIOPCIDevice *vdev, bool ats_needed)

2594 {

2595 PCIDevice *pdev = PCI_DEVICE(vdev);

2596 bool pasid_cap_added = false;

2597 + bool ats_cap_present = false;

2598 Error *err = NULL;

2599 uint32_t header;

2600 uint16_t cap_id, next, size;

@@ -2637,7 +2680,19 @@ static void vfio_add_ext_cap(VFIOPCIDevice *vdev)

2680 */

2681 case PCI_EXT_CAP_ID_PASID:

2682 pasid_cap_added = true;

2640 - /* fallthrough */

2683 + pcie_add_capability(pdev, cap_id, cap_ver, next, size);

2684 + break;

2685 + case PCI_EXT_CAP_ID_ATS:

2686 + ats_cap_present = true;

2687 + /*

2688 + * If ATS is requested and supported according to the kernel, add

2689 + * the ATS capability. If not supported according to the kernel or

2690 + * disabled on the qemu command line, omit the ATS cap.

2691 + */

2692 + if (ats_needed) {

2693 + pcie_add_capability(pdev, cap_id, cap_ver, next, size);

2694 + }

2695 + break;

2696 default:

2697 pcie_add_capability(pdev, cap_id, cap_ver, next, size);

2698 }

@@ -2648,6 +2703,16 @@ static void vfio_add_ext_cap(VFIOPCIDevice *vdev)

2703 error_report_err(err);

2704 }

2705

2706 + if (vdev->ats == ON_OFF_AUTO_ON && !ats_cap_present) {

2707 + warn_report("vfio-pci: ats=on requested, but host device has no "

2708 + "ATS extended capability");

2709 + }

2710 +

2711 + if (vdev->ats == ON_OFF_AUTO_AUTO && ats_cap_present && !ats_needed) {

2712 + warn_report("vfio-pci: host kernel reports ATS unsupported; "

2713 + "ATS capability will be masked");

2714 + }

2715 +

2716 /* Cleanup chain head ID if necessary */

2717 if (pci_get_word(pdev->config + PCI_CONFIG_SPACE_SIZE) == 0xFFFF) {

2718 pci_set_word(pdev->config + PCI_CONFIG_SPACE_SIZE, 0);

@@ -2659,6 +2724,7 @@ static void vfio_add_ext_cap(VFIOPCIDevice *vdev)

2724 bool vfio_pci_add_capabilities(VFIOPCIDevice *vdev, Error **errp)

2725 {

2726 PCIDevice *pdev = PCI_DEVICE(vdev);

2727 + bool ats_needed = false;

2728

2729 if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST) ||

2730 !pdev->config[PCI_CAPABILITY_LIST]) {

@@ -2669,7 +2735,11 @@ bool vfio_pci_add_capabilities(VFIOPCIDevice *vdev, Error **errp)

2735 return false;

2736 }

2737

2672 - vfio_add_ext_cap(vdev);

2738 + if (!vfio_pci_ats_requested_and_supported(vdev, &ats_needed, errp)) {

2739 + return false;

2740 + }

2741 +

2742 + vfio_add_ext_cap(vdev, ats_needed);

2743 return true;

2744 }

2745

@@ -3819,6 +3889,7 @@ static const Property vfio_pci_properties[] = {

3889 DEFINE_PROP_BOOL("skip-vsc-check", VFIOPCIDevice, skip_vsc_check, true),

3890 DEFINE_PROP_UINT16("x-vpasid-cap-offset", VFIOPCIDevice,

3891 vpasid_cap_offset, 0),

3892 + DEFINE_PROP_ON_OFF_AUTO("ats", VFIOPCIDevice, ats, ON_OFF_AUTO_AUTO),

3893 };

3894

3895 static void vfio_pci_set_fd(Object *obj, const char *str, Error **errp)

@@ -3970,13 +4041,22 @@ static void vfio_pci_class_init(ObjectClass *klass, const void *data)

4041 "destination when doing live "

4042 "migration of device state via "

4043 "multifd channels");

3973 - object_class_property_set_description(klass, /* 11.0 */

4044 + object_class_property_set_description(klass, /* 11.0 */

4045 "x-vpasid-cap-offset",

4046 "PCIe extended configuration space offset at which to place a "

4047 "synthetic PASID extended capability when PASID is enabled via "

4048 "a vIOMMU. A value of 0 (default) places the capability at the "

4049 "end of the extended configuration space. The offset must be "

4050 "4-byte aligned and within the PCIe extended configuration space");

4051 + object_class_property_set_description(klass, /* 11.1 */

4052 + "ats",

4053 + "Control guest visibility of the ATS PCIe extended capability. "

4054 + "Valid values are on, off, and auto (default). "

4055 + "'off' always masks ATS. "

4056 + "'on' requires ATS support for the device and fails realize if the "

4057 + "host kernel reports ATS as unavailable for this device. "

4058 + "'auto' masks ATS only when the host kernel reports "

4059 + "ATS as unavailable");

4060 }

4061

4062 static const TypeInfo vfio_pci_info = {

hw/vfio/pci.h

@@ -188,6 +188,7 @@ struct VFIOPCIDevice {

188 bool clear_parent_atomics_on_exit;

189 bool skip_vsc_check;

190 uint16_t vpasid_cap_offset;

191 + OnOffAuto ats;

192 VFIODisplay *dpy;

193 Notifier irqchip_change_notifier;