Patchwork vfio-pci: Add missing kvm_enabled() checks

login
register
mail settings
Submitter Stefan Hajnoczi
Date Dec. 20, 2012, 9:14 p.m.
Message ID <1356038055-29507-1-git-send-email-stefanha@redhat.com>
Download mbox | patch
Permalink /patch/207711/
State New
Headers show

Comments

Stefan Hajnoczi - Dec. 20, 2012, 9:14 p.m.
It's necessary to check kvm_enabled() since a QEMU binary can be
compiled with KVM support but TCG can be used at run-time.
kvm_check_extension() segfaults if kvm_state is NULL.

Reported-by: Joe Terranova <joeterranova@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 hw/vfio_pci.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
Alex Williamson - Dec. 20, 2012, 10:35 p.m.
On Thu, 2012-12-20 at 22:14 +0100, Stefan Hajnoczi wrote:
> It's necessary to check kvm_enabled() since a QEMU binary can be
> compiled with KVM support but TCG can be used at run-time.
> kvm_check_extension() segfaults if kvm_state is NULL.

Ugh, this is exactly why we added kvm_irqfds_enabled() which doesn't
call kvm_check_extension().  How is it possible to fall through to a
kvm_check_extension() with that in place?  Thanks,

Alex

> Reported-by: Joe Terranova <joeterranova@gmail.com>
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
>  hw/vfio_pci.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/vfio_pci.c b/hw/vfio_pci.c
> index fbfe670..677dc48 100644
> --- a/hw/vfio_pci.c
> +++ b/hw/vfio_pci.c
> @@ -275,7 +275,8 @@ static void vfio_enable_intx_kvm(VFIODevice *vdev)
>      int ret, argsz;
>      int32_t *pfd;
>  
> -    if (!kvm_irqfds_enabled() ||
> +    if (!kvm_enabled() ||
> +        !kvm_irqfds_enabled() ||
>          vdev->intx.route.mode != PCI_INTX_ENABLED ||
>          !kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) {
>          return;
> @@ -438,7 +439,8 @@ static int vfio_enable_intx(VFIODevice *vdev)
>       * Only conditional to avoid generating error messages on platforms
>       * where we won't actually use the result anyway.
>       */
> -    if (kvm_irqfds_enabled() &&
> +    if (kvm_enabled() &&
> +        kvm_irqfds_enabled() &&
>          kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) {
>          vdev->intx.route = pci_device_route_intx_to_irq(&vdev->pdev,
>                                                          vdev->intx.pin);
Stefan Hajnoczi - Dec. 21, 2012, 7:11 a.m.
On Thu, Dec 20, 2012 at 11:35 PM, Alex Williamson
<alex.williamson@redhat.com> wrote:
> On Thu, 2012-12-20 at 22:14 +0100, Stefan Hajnoczi wrote:
>> It's necessary to check kvm_enabled() since a QEMU binary can be
>> compiled with KVM support but TCG can be used at run-time.
>> kvm_check_extension() segfaults if kvm_state is NULL.
>
> Ugh, this is exactly why we added kvm_irqfds_enabled() which doesn't
> call kvm_check_extension().  How is it possible to fall through to a
> kvm_check_extension() with that in place?  Thanks,

Ah, this explains it: Joe was using QEMU 1.3.0 which does not contain
the kvm_irqfds_enabled() fix.

My patch is unnecessary and can be dropped.

Stefan

Patch

diff --git a/hw/vfio_pci.c b/hw/vfio_pci.c
index fbfe670..677dc48 100644
--- a/hw/vfio_pci.c
+++ b/hw/vfio_pci.c
@@ -275,7 +275,8 @@  static void vfio_enable_intx_kvm(VFIODevice *vdev)
     int ret, argsz;
     int32_t *pfd;
 
-    if (!kvm_irqfds_enabled() ||
+    if (!kvm_enabled() ||
+        !kvm_irqfds_enabled() ||
         vdev->intx.route.mode != PCI_INTX_ENABLED ||
         !kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) {
         return;
@@ -438,7 +439,8 @@  static int vfio_enable_intx(VFIODevice *vdev)
      * Only conditional to avoid generating error messages on platforms
      * where we won't actually use the result anyway.
      */
-    if (kvm_irqfds_enabled() &&
+    if (kvm_enabled() &&
+        kvm_irqfds_enabled() &&
         kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) {
         vdev->intx.route = pci_device_route_intx_to_irq(&vdev->pdev,
                                                         vdev->intx.pin);