Re: [Qemu-devel] [PATCH v2] pci: fix pci_requester_id()

[Michael S. Tsirkin]

On Mon, May 16, 2016 at 03:13:20PM +0800, Peter Xu wrote:
> The original pci_requester_id() is not really getting requester ID, but
> only BDF. It will not work if we enable IOMMU IR with SID verifications
> when with PCI bridges.  Renaming it to pci_get_bdf().
> 
> Meanwhile, we provide the correct implementation to get requester
> ID. VT-d spec 5.1.1 is a good reference to go, though it talks only
> about interrupt delivery, the rule works exactly the same for
> non-interrupt cases.
> 
> Currently, there are three use cases for pci_requester_id():
> 
> - PCIX status bits: here we need BDF only, not requester ID. Replacing
>   with pci_get_bdf().
> - PCIe Error injection and MSI delivery: for both these cases, we are
>   looking for requester IDs. Kept untouched to leverage the new impl.
> 
> Signed-off-by: Peter Xu <address@hidden>
> ---
>  hw/i386/kvm/pci-assign.c |  2 +-
>  hw/pci/pci.c             | 29 +++++++++++++++++++++++++++++
>  include/hw/pci/pci.h     |  7 +++++--
>  3 files changed, 35 insertions(+), 3 deletions(-)
> 
> diff --git a/hw/i386/kvm/pci-assign.c b/hw/i386/kvm/pci-assign.c
> index bf425a2..c40ab36 100644
> --- a/hw/i386/kvm/pci-assign.c
> +++ b/hw/i386/kvm/pci-assign.c
> @@ -1481,7 +1481,7 @@ static int assigned_device_pci_cap_init(PCIDevice 
> *pci_dev, Error **errp)
>           * error bits, leave the rest. */
>          status = pci_get_long(pci_dev->config + pos + PCI_X_STATUS);
>          status &= ~(PCI_X_STATUS_BUS | PCI_X_STATUS_DEVFN);
> -        status |= pci_requester_id(pci_dev);
> +        status |= pci_get_bdf(pci_dev);
>          status &= ~(PCI_X_STATUS_SPL_DISC | PCI_X_STATUS_UNX_SPL |
>                      PCI_X_STATUS_SPL_ERR);
>          pci_set_long(pci_dev->config + pos + PCI_X_STATUS, status);
> diff --git a/hw/pci/pci.c b/hw/pci/pci.c
> index bb605ef..7430715 100644
> --- a/hw/pci/pci.c
> +++ b/hw/pci/pci.c
> @@ -2498,6 +2498,35 @@ PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
>      }
>  }
>  
> +/* Parse bridges up to the root complex and get final Requester ID
> + * for this device.  For PCIe-all topology, this works exactly as
> + * pci_get_bdf() does. However, several tricks are required for
> + * legacy PCI topology and PCIe-to-PCI bridges, to be better aligned
> + * with spec. */
> +uint16_t pci_requester_id(PCIDevice *dev)
> +{
> +    uint8_t bus_n;
> +    uint16_t result = pci_get_bdf(dev);

add empty line here pls.

> +    while ((bus_n = pci_bus_num(dev->bus))) {
> +        /* We are under PCI/PCIe bridges */
> +        dev = dev->bus->parent_dev;
> +        if (pci_is_express(dev)) {
> +            if (pcie_cap_get_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE) {
> +                /* When we pass through PCIe-to-PCI/PCIX bridges, we
> +                 * override the requester ID using secondary bus
> +                 * number with zeroed devfn (pcie-to-pci bridge spec
> +                 * chap 2.3). */
> +                result = PCI_BUILD_BDF(bus_n, 0);
> +            }
> +        } else {


> +            /* Legacy PCI bus, override requester ID with the
> +             * bridge's BDF upstream. */

Don't document what is done, pls document why.
Please add an explanation why this is the right thing to do.

> +            result = pci_get_bdf(dev);

Won't dev be NULL for a root bus?

> +        }
> +    }
> +    return result;
> +}
> +
>  static const TypeInfo pci_device_type_info = {
>      .name = TYPE_PCI_DEVICE,
>      .parent = TYPE_DEVICE,

OK but this is used on data path by msi_send_message and
doing scans there isn't nice.
Can we store a pointer to device who's requester id is used instead?


> diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h
> index ef6ba51..351266c 100644
> --- a/include/hw/pci/pci.h
> +++ b/include/hw/pci/pci.h
> @@ -15,6 +15,7 @@
>  #define PCI_DEVFN(slot, func)   ((((slot) & 0x1f) << 3) | ((func) & 0x07))
>  #define PCI_SLOT(devfn)         (((devfn) >> 3) & 0x1f)
>  #define PCI_FUNC(devfn)         ((devfn) & 0x07)
> +#define PCI_BUILD_BDF(bus, devfn)     ((bus << 8) | (devfn))
>  #define PCI_SLOT_MAX            32
>  #define PCI_FUNC_MAX            8
>  
> @@ -685,11 +686,13 @@ static inline uint32_t pci_config_size(const PCIDevice 
> *d)
>      return pci_is_express(d) ? PCIE_CONFIG_SPACE_SIZE : 
> PCI_CONFIG_SPACE_SIZE;
>  }
>  
> -static inline uint16_t pci_requester_id(PCIDevice *dev)
> +static inline uint16_t pci_get_bdf(PCIDevice *dev)
>  {
> -    return (pci_bus_num(dev->bus) << 8) | dev->devfn;
> +    return PCI_BUILD_BDF(pci_bus_num(dev->bus), dev->devfn);
>  }
>  
> +uint16_t pci_requester_id(PCIDevice *dev);
> +
>  /* DMA access functions */
>  static inline AddressSpace *pci_get_address_space(PCIDevice *dev)
>  {
> -- 
> 2.4.11