hw/xen/xen_pt.c | 60 ++++++++++++++++++++++++++++++++++++++++++++ include/hw/pci/pci.h | 4 +++ 2 files changed, 64 insertions(+)
In PVH dom0, when passthrough a device to domU, QEMU code
xen_pt_realize->xc_physdev_map_pirq wants to use gsi, but in current codes
the gsi number is got from file /sys/bus/pci/devices/<sbdf>/irq, that is
wrong, because irq is not equal with gsi, they are in different spaces, so
pirq mapping fails.
To solve above problem, use new interface of Xen, xc_pcidev_get_gsi to get
gsi and use xc_physdev_map_pirq_gsi to map pirq when dom0 is PVH.
Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com>
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com>
---
Hi All,
This is v10 to support passthrough on Xen when dom0 is PVH.
v9->v10 changes:
* Check the return value of function fgets.
* Added comment to explain why set the length of array 'type' to be 10.
Best regards,
Jiqian Chen
v8->v9 changes:
* Moved the definition of PCI_SBDF from /hw/xen/xen_pt.c to /include/hw/pci/pci.h.
* Renamed xen_run_qemu_on_hvm to xen_pt_need_gsi.
* Renamed xen_map_pirq_for_gsi to xen_pt_map_pirq_for_gsi.
* Through reading /sys/hypervisor/guest_type to get dom type instead of using xc_domain_getinfo_single.
v7->v8 changes:
* Since xc_physdev_gsi_from_dev was renamed to xc_pcidev_get_gsi, changed it.
* Added xen_run_qemu_on_hvm to check if Qemu run on PV dom0, if not use xc_physdev_map_pirq_gsi to map pirq.
* Used CONFIG_XEN_CTRL_INTERFACE_VERSION to wrap the new part for compatibility.
* Added "#define DOMID_RUN_QEMU 0" to represent the id of domain that Qemu run on.
v6->v7 changes:
* Because the function of obtaining gsi was changed on the kernel and Xen side. Changed to use
xc_physdev_gsi_from_dev, that requires passing in sbdf instead of irq.
v5->v6 changes:
* Because the function of obtaining gsi was changed on the kernel and Xen side. Changed to use
xc_physdev_gsi_from_irq, instead of gsi sysfs.
* Since function changed, removed the Review-by of Stefano.
v4->v5 changes:
* Added Review-by Stefano.
v3->v4 changes:
* Added gsi into struct XenHostPCIDevice and used gsi number that read from gsi sysfs
if it exists, if there is no gsi sysfs, still use irq.
v2->v3 changes:
* Due to changes in the implementation of the second patch on kernel side(that adds
a new sysfs for gsi instead of a new syscall), so read gsi number from the sysfs of gsi.
v1 and v2:
We can record the relation between gsi and irq, then when userspace(qemu) want
to use gsi, we can do a translation. The third patch of kernel(xen/privcmd: Add new syscall
to get gsi from irq) records all the relations in acpi_register_gsi_xen_pvh() when dom0
initialize pci devices, and provide a syscall for userspace to get the gsi from irq. The
third patch of xen(tools: Add new function to get gsi from irq) add a new function
xc_physdev_gsi_from_irq() to call the new syscall added on kernel side.
And then userspace can use that function to get gsi. Then xc_physdev_map_pirq() will success.
Issues we encountered:
1. failed to map pirq for gsi
Problem: qemu will call xc_physdev_map_pirq() to map a passthrough device's gsi to pirq in
function xen_pt_realize(). But failed.
Reason: According to the implement of xc_physdev_map_pirq(), it needs gsi instead of irq,
but qemu pass irq to it and treat irq as gsi, it is got from file
/sys/bus/pci/devices/xxxx:xx:xx.x/irq in function xen_host_pci_device_get(). But actually
the gsi number is not equal with irq. They are in different space.
---
hw/xen/xen_pt.c | 60 ++++++++++++++++++++++++++++++++++++++++++++
include/hw/pci/pci.h | 4 +++
2 files changed, 64 insertions(+)
diff --git a/hw/xen/xen_pt.c b/hw/xen/xen_pt.c
index 3635d1b39f79..5e6230c4a60f 100644
--- a/hw/xen/xen_pt.c
+++ b/hw/xen/xen_pt.c
@@ -766,6 +766,57 @@ static void xen_pt_destroy(PCIDevice *d) {
}
/* init */
+#if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 42000
+static bool xen_pt_need_gsi(void)
+{
+ FILE *fp;
+ int len;
+ /*
+ * The max length of guest_type is "PVH"+'\n'+'\0', it is 5,
+ * so here set the length of type to be twice.
+ */
+ char type[10];
+ const char *guest_type = "/sys/hypervisor/guest_type";
+
+ fp = fopen(guest_type, "r");
+ if (!fp) {
+ error_report("Cannot open %s: %s", guest_type, strerror(errno));
+ return false;
+ }
+
+ if (fgets(type, sizeof(type), fp)) {
+ len = strlen(type);
+ if (len) {
+ type[len - 1] = '\0';
+ if (!strcmp(type, "PVH")) {
+ fclose(fp);
+ return true;
+ }
+ }
+ }
+
+ fclose(fp);
+ return false;
+}
+
+static int xen_pt_map_pirq_for_gsi(PCIDevice *d, int *pirq)
+{
+ int gsi;
+ XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
+
+ gsi = xc_pcidev_get_gsi(xen_xc,
+ PCI_SBDF(s->real_device.domain,
+ s->real_device.bus,
+ s->real_device.dev,
+ s->real_device.func));
+ if (gsi >= 0) {
+ return xc_physdev_map_pirq_gsi(xen_xc, xen_domid, gsi, pirq);
+ }
+
+ return gsi;
+}
+#endif
+
static void xen_pt_realize(PCIDevice *d, Error **errp)
{
ERRP_GUARD();
@@ -847,7 +898,16 @@ static void xen_pt_realize(PCIDevice *d, Error **errp)
goto out;
}
+#if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 42000
+ if (xen_pt_need_gsi()) {
+ rc = xen_pt_map_pirq_for_gsi(d, &pirq);
+ } else {
+ rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
+ }
+#else
rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
+#endif
+
if (rc < 0) {
XEN_PT_ERR(d, "Mapping machine irq %u to pirq %i failed, (err: %d)\n",
machine_irq, pirq, errno);
diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h
index eb26cac81098..07805aa8a5f3 100644
--- a/include/hw/pci/pci.h
+++ b/include/hw/pci/pci.h
@@ -23,6 +23,10 @@ extern bool pci_available;
#define PCI_SLOT_MAX 32
#define PCI_FUNC_MAX 8
+#define PCI_SBDF(seg, bus, dev, func) \
+ ((((uint32_t)(seg)) << 16) | \
+ (PCI_BUILD_BDF(bus, PCI_DEVFN(dev, func))))
+
/* Class, Vendor and Device IDs from Linux's pci_ids.h */
#include "hw/pci/pci_ids.h"
--
2.34.1On Wed, Nov 06, 2024 at 02:14:18PM +0800, Jiqian Chen wrote: > In PVH dom0, when passthrough a device to domU, QEMU code > xen_pt_realize->xc_physdev_map_pirq wants to use gsi, but in current codes > the gsi number is got from file /sys/bus/pci/devices/<sbdf>/irq, that is > wrong, because irq is not equal with gsi, they are in different spaces, so > pirq mapping fails. > > To solve above problem, use new interface of Xen, xc_pcidev_get_gsi to get > gsi and use xc_physdev_map_pirq_gsi to map pirq when dom0 is PVH. > > Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> > Signed-off-by: Huang Rui <ray.huang@amd.com> > Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> Acked-by: Anthony PERARD <anthony@xenproject.org> But, this following change probably needs an ack from PCI maintaners, CCed. > diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h > index eb26cac81098..07805aa8a5f3 100644 > --- a/include/hw/pci/pci.h > +++ b/include/hw/pci/pci.h > @@ -23,6 +23,10 @@ extern bool pci_available; > #define PCI_SLOT_MAX 32 > #define PCI_FUNC_MAX 8 > > +#define PCI_SBDF(seg, bus, dev, func) \ > + ((((uint32_t)(seg)) << 16) | \ > + (PCI_BUILD_BDF(bus, PCI_DEVFN(dev, func)))) > + > /* Class, Vendor and Device IDs from Linux's pci_ids.h */ > #include "hw/pci/pci_ids.h" Thanks, -- Anthony PERARD
Hi MST and Marcel, On 2024/11/19 00:05, Anthony PERARD wrote: > On Wed, Nov 06, 2024 at 02:14:18PM +0800, Jiqian Chen wrote: >> In PVH dom0, when passthrough a device to domU, QEMU code >> xen_pt_realize->xc_physdev_map_pirq wants to use gsi, but in current codes >> the gsi number is got from file /sys/bus/pci/devices/<sbdf>/irq, that is >> wrong, because irq is not equal with gsi, they are in different spaces, so >> pirq mapping fails. >> >> To solve above problem, use new interface of Xen, xc_pcidev_get_gsi to get >> gsi and use xc_physdev_map_pirq_gsi to map pirq when dom0 is PVH. >> >> Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> >> Signed-off-by: Huang Rui <ray.huang@amd.com> >> Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> > > Acked-by: Anthony PERARD <anthony@xenproject.org> > > But, this following change probably needs an ack from PCI maintaners, > CCed. > >> diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h >> index eb26cac81098..07805aa8a5f3 100644 >> --- a/include/hw/pci/pci.h >> +++ b/include/hw/pci/pci.h >> @@ -23,6 +23,10 @@ extern bool pci_available; >> #define PCI_SLOT_MAX 32 >> #define PCI_FUNC_MAX 8 >> >> +#define PCI_SBDF(seg, bus, dev, func) \ >> + ((((uint32_t)(seg)) << 16) | \ >> + (PCI_BUILD_BDF(bus, PCI_DEVFN(dev, func)))) >> + >> /* Class, Vendor and Device IDs from Linux's pci_ids.h */ >> #include "hw/pci/pci_ids.h" May I get your ACK about this change? > > Thanks, > -- Best regards, Jiqian Chen.
On 11/6/24 01:14, Jiqian Chen wrote: > In PVH dom0, when passthrough a device to domU, QEMU code > xen_pt_realize->xc_physdev_map_pirq wants to use gsi, but in current codes > the gsi number is got from file /sys/bus/pci/devices/<sbdf>/irq, that is > wrong, because irq is not equal with gsi, they are in different spaces, so > pirq mapping fails. > > To solve above problem, use new interface of Xen, xc_pcidev_get_gsi to get > gsi and use xc_physdev_map_pirq_gsi to map pirq when dom0 is PVH. > > Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> > Signed-off-by: Huang Rui <ray.huang@amd.com> > Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com> Reviewed-by: Stewart Hildebrand <stewart.hildebrand@amd.com>
© 2016 - 2024 Red Hat, Inc.