pcie_sriov: make pcie_sriov_pf_exit() safe on non-SR-IOV devices

[PATCH] pcie_sriov: make pcie_sriov_pf_exit() safe on non-SR-IOV devices

Posted by Stefan Hajnoczi 3 days, 23 hours ago

Commit 3f9cfaa92c96 ("virtio-pci: Implement SR-IOV PF") added an
unconditional call from virtio_pci_exit() to pcie_sriov_pf_exit().

pcie_sriov_pf_exit() reads from the SR-IOV Capability in Configuration
Space:

  uint8_t *cfg = dev->config + dev->exp.sriov_cap;
  ...
  unparent_vfs(dev, pci_get_word(cfg + PCI_SRIOV_TOTAL_VF));
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This results in undefined behavior when dev->exp.sriov_cap is 0 because
this is not an SR-IOV device. For example, unparent_vfs() segfaults when
total_vfs happens to be non-zero.

Fix this by returning early from pcie_sriov_pf_exit() when
dev->exp.sriov_cap is 0 because this is not an SR-IOV device.

Cc: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
Cc: Michael S. Tsirkin <mst@redhat.com>
Reported-by: Qing Wang <qinwang@redhat.com>
Buglink: https://issues.redhat.com/browse/RHEL-116443
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 hw/pci/pcie_sriov.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/hw/pci/pcie_sriov.c b/hw/pci/pcie_sriov.c
index 8a4bf0d6f7..cf1b5b5c05 100644
--- a/hw/pci/pcie_sriov.c
+++ b/hw/pci/pcie_sriov.c
@@ -195,7 +195,9 @@ bool pcie_sriov_pf_init(PCIDevice *dev, uint16_t offset,
 
 void pcie_sriov_pf_exit(PCIDevice *dev)
 {
-    uint8_t *cfg = dev->config + dev->exp.sriov_cap;
+    if (dev->exp.sriov_cap == 0) {
+        return;
+    }
 
     if (dev->exp.sriov_pf.vf_user_created) {
         uint16_t ven_id = pci_get_word(dev->config + PCI_VENDOR_ID);
@@ -211,6 +213,8 @@ void pcie_sriov_pf_exit(PCIDevice *dev)
             pci_config_set_device_id(dev->exp.sriov_pf.vf[i]->config, vf_dev_id);
         }
     } else {
+        uint8_t *cfg = dev->config + dev->exp.sriov_cap;
+
         unparent_vfs(dev, pci_get_word(cfg + PCI_SRIOV_TOTAL_VF));
     }
 }
-- 
2.51.0

Re: [PATCH] pcie_sriov: make pcie_sriov_pf_exit() safe on non-SR-IOV devices

Posted by Stefan Hajnoczi 3 days, 21 hours ago

On Wed, Sep 24, 2025 at 11:51:53AM -0400, Stefan Hajnoczi wrote:
> Commit 3f9cfaa92c96 ("virtio-pci: Implement SR-IOV PF") added an
> unconditional call from virtio_pci_exit() to pcie_sriov_pf_exit().
> 
> pcie_sriov_pf_exit() reads from the SR-IOV Capability in Configuration
> Space:
> 
>   uint8_t *cfg = dev->config + dev->exp.sriov_cap;
>   ...
>   unparent_vfs(dev, pci_get_word(cfg + PCI_SRIOV_TOTAL_VF));
>                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> This results in undefined behavior when dev->exp.sriov_cap is 0 because
> this is not an SR-IOV device. For example, unparent_vfs() segfaults when
> total_vfs happens to be non-zero.
> 
> Fix this by returning early from pcie_sriov_pf_exit() when
> dev->exp.sriov_cap is 0 because this is not an SR-IOV device.
> 
> Cc: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
> Cc: Michael S. Tsirkin <mst@redhat.com>
> Reported-by: Qing Wang <qinwang@redhat.com>
> Buglink: https://issues.redhat.com/browse/RHEL-116443
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
>  hw/pci/pcie_sriov.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)

CCing qemu-stable

> 
> diff --git a/hw/pci/pcie_sriov.c b/hw/pci/pcie_sriov.c
> index 8a4bf0d6f7..cf1b5b5c05 100644
> --- a/hw/pci/pcie_sriov.c
> +++ b/hw/pci/pcie_sriov.c
> @@ -195,7 +195,9 @@ bool pcie_sriov_pf_init(PCIDevice *dev, uint16_t offset,
>  
>  void pcie_sriov_pf_exit(PCIDevice *dev)
>  {
> -    uint8_t *cfg = dev->config + dev->exp.sriov_cap;
> +    if (dev->exp.sriov_cap == 0) {
> +        return;
> +    }
>  
>      if (dev->exp.sriov_pf.vf_user_created) {
>          uint16_t ven_id = pci_get_word(dev->config + PCI_VENDOR_ID);
> @@ -211,6 +213,8 @@ void pcie_sriov_pf_exit(PCIDevice *dev)
>              pci_config_set_device_id(dev->exp.sriov_pf.vf[i]->config, vf_dev_id);
>          }
>      } else {
> +        uint8_t *cfg = dev->config + dev->exp.sriov_cap;
> +
>          unparent_vfs(dev, pci_get_word(cfg + PCI_SRIOV_TOTAL_VF));
>      }
>  }
> -- 
> 2.51.0
>

Re: [PATCH] pcie_sriov: make pcie_sriov_pf_exit() safe on non-SR-IOV devices

Posted by Michael Tokarev 1 day ago

On 24.09.2025 20:51, Stefan Hajnoczi wrote:
> On Wed, Sep 24, 2025 at 11:51:53AM -0400, Stefan Hajnoczi wrote:
>> Commit 3f9cfaa92c96 ("virtio-pci: Implement SR-IOV PF") added an
>> unconditional call from virtio_pci_exit() to pcie_sriov_pf_exit().
>>
>> pcie_sriov_pf_exit() reads from the SR-IOV Capability in Configuration
>> Space:
>>
>>    uint8_t *cfg = dev->config + dev->exp.sriov_cap;
>>    ...
>>    unparent_vfs(dev, pci_get_word(cfg + PCI_SRIOV_TOTAL_VF));
>>                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> This results in undefined behavior when dev->exp.sriov_cap is 0 because
>> this is not an SR-IOV device. For example, unparent_vfs() segfaults when
>> total_vfs happens to be non-zero.
>>
>> Fix this by returning early from pcie_sriov_pf_exit() when
>> dev->exp.sriov_cap is 0 because this is not an SR-IOV device.
>>
>> Cc: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
>> Cc: Michael S. Tsirkin <mst@redhat.com>
>> Reported-by: Qing Wang <qinwang@redhat.com>
>> Buglink: https://issues.redhat.com/browse/RHEL-116443
>> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
>> ---
>>   hw/pci/pcie_sriov.c | 6 +++++-
>>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> CCing qemu-stable

Ping?  Can we apply this to the master branch, so I can pick it
up for the stable series?

Thanks,

/mjt