On 6/23/26 19:01, Shameer Kolothum Thodi wrote:
-----Original Message-----
From: Nathan Chen <[email protected]>
Sent: 23 June 2026 03:36
To: [email protected]
Cc: Yi Liu <[email protected]>; Eric Auger <[email protected]>;
Zhenzhong Duan <[email protected]>; Alex Williamson
<[email protected]>; Cédric Le Goater <[email protected]>; Matt Ochs
<[email protected]>; Nicolin Chen <[email protected]>; Shameer
Kolothum Thodi <[email protected]>; Nathan Chen
<[email protected]>
Subject: [PATCH v3 2/2] vfio/pci: Add ats property
From: Nathan Chen <[email protected]>
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.
Suggested-by: Shameer Kolothum <[email protected]>
Signed-off-by: Nathan Chen <[email protected]>
---
hw/vfio/pci.h | 1 +
hw/vfio/pci.c | 88
++++++++++++++++++++++++++++++++++++++++++++++++---
2 files changed, 85 insertions(+), 4 deletions(-)
diff --git a/hw/vfio/pci.h b/hw/vfio/pci.h
index c3a1f53d35..f2934f2d84 100644
--- a/hw/vfio/pci.h
+++ b/hw/vfio/pci.h
@@ -188,6 +188,7 @@ struct VFIOPCIDevice {
bool clear_parent_atomics_on_exit;
bool skip_vsc_check;
uint16_t vpasid_cap_offset;
+ OnOffAuto ats;
VFIODisplay *dpy;
Notifier irqchip_change_notifier;
VFIOPCICPR cpr;
diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c
index 9c06b25e63..c0436b4c04 100644
--- a/hw/vfio/pci.c
+++ b/hw/vfio/pci.c
@@ -2546,10 +2546,58 @@ static bool
vfio_pci_synthesize_pasid_cap(VFIOPCIDevice *vdev, Error **errp)
return true;
}
-static void vfio_add_ext_cap(VFIOPCIDevice *vdev)
+/*
+ * Determine whether ATS capability should be advertised for @vdev, based
on
+ * whether it was enabled on the command line and whether it is supported
+ * according to the kernel.
+ *
+ * Store whether ATS capability should be advertised in @ats_needed.
+ *
+ * Returns false only when ats=on is explicitly requested but the kernel
+ * reports it is not supported. Returns true in all other cases.
+ */
+static bool vfio_pci_ats_requested_and_supported(VFIOPCIDevice *vdev,
+ bool *ats_needed, Error
**errp)
+{
+ HostIOMMUDevice *hiod = vdev->vbasedev.hiod;
+ HostIOMMUDeviceClass *hiodc;
+ bool ats_supported;
+ *ats_needed = false;
+
+ if (vdev->ats == ON_OFF_AUTO_OFF) {
+ return true;
+ }
+
+ *ats_needed = true;
+ if (!hiod) {
+ return true;
+ }
+ hiodc = HOST_IOMMU_DEVICE_GET_CLASS(hiod);
+ if (!hiodc || !hiodc->support_ats) {
+ return true;
+ }
+
+ ats_supported = hiodc->support_ats(hiod);
+ if (vdev->ats == ON_OFF_AUTO_ON && !ats_supported) {
+ error_setg(errp, "vfio-pci: ATS requested but not supported by
kernel");
+ *ats_needed = false;
+ return false;
+ }
+
+ if (vdev->ats == ON_OFF_AUTO_AUTO && !ats_supported) {
+ warn_report("vfio-pci: host kernel reports ATS unsupported; "
+ "ATS capability will be masked");
+ }
This will be slightly misleading if dev doesn't have the ATS CAP at all.
Isn't it? I think, the best place to do this warning is below where we know
ats_cap_present is set or not.
Thanks,
Shameer
+
+ *ats_needed = ats_supported;
+ return true;
+}
+
+static void vfio_add_ext_cap(VFIOPCIDevice *vdev, bool ats_needed)
{
PCIDevice *pdev = PCI_DEVICE(vdev);
bool pasid_cap_added = false;
+ bool ats_cap_present = false;
Error *err = NULL;
uint32_t header;
uint16_t cap_id, next, size;
@@ -2635,7 +2683,19 @@ static void vfio_add_ext_cap(VFIOPCIDevice
*vdev)
*/
case PCI_EXT_CAP_ID_PASID:
pasid_cap_added = true;
- /* fallthrough */
+ pcie_add_capability(pdev, cap_id, cap_ver, next, size);
+ break;
+ case PCI_EXT_CAP_ID_ATS:
+ ats_cap_present = true;
+ /*
+ * If ATS is requested and supported according to the kernel, add
+ * the ATS capability. If not supported according to the kernel or
+ * disabled on the qemu command line, omit the ATS cap.
+ */
+ if (ats_needed) {
+ pcie_add_capability(pdev, cap_id, cap_ver, next, size);
+ }
+ break;
default:
pcie_add_capability(pdev, cap_id, cap_ver, next, size);
}
@@ -2646,6 +2706,11 @@ static void vfio_add_ext_cap(VFIOPCIDevice
*vdev)
error_report_err(err);
}
+ if (ats_needed && !ats_cap_present) {
+ warn_report("vfio-pci: ats=on requested, but host device has no "
+ "ATS extended capability");
+ }
"ats=on requested" is wrong here. OnOffAuto_str() could be used instead.
Also, ats_needed is true for "ats=auto" too, and since it's the default
setting, and most devices lack ATS and any modern host kernel support it,
this warning will be emitted for nearly all devices.
How about :
if (vdev->ats == ON_OFF_AUTO_ON && !ats_cap_present) {
warn_report("vfio-pci: ats=on requested, but host device has no "
"ATS extended capability");
}
?
Thanks,
C.
+
/* Cleanup chain head ID if necessary */
if (pci_get_word(pdev->config + PCI_CONFIG_SPACE_SIZE) == 0xFFFF) {
pci_set_word(pdev->config + PCI_CONFIG_SPACE_SIZE, 0);
@@ -2657,6 +2722,7 @@ static void vfio_add_ext_cap(VFIOPCIDevice
*vdev)
bool vfio_pci_add_capabilities(VFIOPCIDevice *vdev, Error **errp)
{
PCIDevice *pdev = PCI_DEVICE(vdev);
+ bool ats_needed = false;
if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST) ||
!pdev->config[PCI_CAPABILITY_LIST]) {
@@ -2667,7 +2733,11 @@ bool vfio_pci_add_capabilities(VFIOPCIDevice
*vdev, Error **errp)
return false;
}
- vfio_add_ext_cap(vdev);
+ if (!vfio_pci_ats_requested_and_supported(vdev, &ats_needed, errp)) {
+ return false;
+ }
+
+ vfio_add_ext_cap(vdev, ats_needed);
return true;
}
@@ -3815,6 +3885,7 @@ static const Property vfio_pci_properties[] = {
DEFINE_PROP_BOOL("skip-vsc-check", VFIOPCIDevice, skip_vsc_check,
true),
DEFINE_PROP_UINT16("x-vpasid-cap-offset", VFIOPCIDevice,
vpasid_cap_offset, 0),
+ DEFINE_PROP_ON_OFF_AUTO("ats", VFIOPCIDevice, ats,
ON_OFF_AUTO_AUTO),
};
static void vfio_pci_set_fd(Object *obj, const char *str, Error **errp)
@@ -3966,13 +4037,22 @@ static void vfio_pci_class_init(ObjectClass *klass,
const void *data)
"destination when doing live "
"migration of device state via "
"multifd channels");
- object_class_property_set_description(klass, /* 11.0 */
+ object_class_property_set_description(klass, /* 11.0 */
"x-vpasid-cap-offset",
"PCIe extended configuration space
offset at which to
place a "
"synthetic PASID extended
capability when PASID is
enabled via "
"a vIOMMU. A value of 0 (default)
places the capability
at the "
"end of the extended configuration
space. The offset
must be "
"4-byte aligned and within the PCIe
extended
configuration space");
+ object_class_property_set_description(klass, /* 11.1 */
+ "ats",
+ "Control guest visibility of the ATS
PCIe extended
capability. "
+ "Valid values are on, off, and auto
(default). "
+ "'off' always masks ATS. "
+ "'on' requires ATS support for the
device and fails realize
if the "
+ "host kernel reports ATS as unavailable
for this device. "
+ "'auto' masks ATS only when the host
kernel reports "
+ "ATS as unavailable");
}
static const TypeInfo vfio_pci_info = {
--
2.43.0
