Re: [PATCH 2/2] pci-assign: Fix MSI-X registration

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2011-09-22 05:12, Alex Williamson wrote:
> Commit c4525754 added a capability check for KVM_CAP_DEVICE_MSIX,
> which is unfortunately not exposed, resulting in MSIX never
> being listed as a capability. 

Oops. Should we fix this nevertheless in the kernel?

> This breaks anything depending on
> MSIX, such as igbvf.  Since we can't specifically check for MSIX
> support and KVM_CAP_ASSIGN_DEV_IRQ indicates more than just MSI,
> let's just revert c4525754 and replace it with a sanity check that
> we need KVM_CAP_ASSIGN_DEV_IRQ if the device supports any kind of
> interrupt (which is still mostly paranoia).
> 
> Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
> ---
> 
>  hw/device-assignment.c |   13 +++++++++----
>  1 files changed, 9 insertions(+), 4 deletions(-)
> 
> diff --git a/hw/device-assignment.c b/hw/device-assignment.c
> index 93913b3..b5bde68 100644
> --- a/hw/device-assignment.c
> +++ b/hw/device-assignment.c
> @@ -1189,8 +1189,7 @@ static int assigned_device_pci_cap_init(PCIDevice *pci_dev)
>  
>      /* Expose MSI capability
>       * MSI capability is the 1st capability in capability config */
> -    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSI, 0);
> -    if (pos != 0 && kvm_check_extension(kvm_state, KVM_CAP_ASSIGN_DEV_IRQ)) {
> +    if ((pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSI, 0))) {
>          dev->cap.available |= ASSIGNED_DEVICE_CAP_MSI;
>          /* Only 32-bit/no-mask currently supported */
>          if ((ret = pci_add_capability(pci_dev, PCI_CAP_ID_MSI, pos, 10)) < 0) {
> @@ -1211,8 +1210,7 @@ static int assigned_device_pci_cap_init(PCIDevice *pci_dev)
>          pci_set_word(pci_dev->wmask + pos + PCI_MSI_DATA_32, 0xffff);
>      }
>      /* Expose MSI-X capability */
> -    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSIX, 0);
> -    if (pos != 0 && kvm_check_extension(kvm_state, KVM_CAP_DEVICE_MSIX)) {
> +    if ((pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSIX, 0))) {
>          int bar_nr;
>          uint32_t msix_table_entry;
>  
> @@ -1606,6 +1604,13 @@ static int assigned_initfn(struct PCIDevice *pci_dev)
>      if (assigned_device_pci_cap_init(pci_dev) < 0)
>          goto out;
>  
> +    if (!kvm_check_extension(kvm_state, KVM_CAP_ASSIGN_DEV_IRQ) &&
> +        (dev->cap.available & ASSIGNED_DEVICE_CAP_MSIX ||
> +         dev->cap.available & ASSIGNED_DEVICE_CAP_MSI ||
> +         assigned_dev_pci_read_byte(pci_dev, PCI_INTERRUPT_PIN) != 0)) {
> +        goto out;
> +    }
> +

That's not equivalent as it needlessly prevents IRQ support in the
absence of KVM_CAP_ASSIGN_DEV_IRQ.

Let's just fix the core issue and replace the test for
KVM_CAP_DEVICE_MSIX with a test call of KVM_ASSIGN_SET_MSIX_NR, passing
in a NULL struct. If it returns -EFAULT, the IOCTL is known and MSIX is
supported.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux