Re: [PATCH v2 23/23] OvmfPkg/Virtio: define VIRITO_F_IOMMU_PLATFORM feature bit

[Laszlo Ersek <lersek@redhat.com>]

On 08/14/17 13:36, Brijesh Singh wrote:
> This feature indicates that the device is behind an IOMMU that translates
> bus addresses from the device into physical addresses in memory.  If this
> feature bit is set to 0, then the device emits physical addresses which
> are not translated further, even though an IOMMU may be present.
> see [1] for more infromation
> 
> [1] https://lists.oasis-open.org/archives/virtio-dev/201610/msg00121.html
> 
> In previous patches, we have implemented IOMMU-like member functions in
> VIRTIO_DEVICE_PROTOCOL to translate the physical address to bus address
> and virtio drivers are updated to use those member functions. We do not
> need to do anything special when VIRTIO_F_IOMMU_PLATFORM bit is present
> hence treat it in parallel with VIRTIO_F_VERSION_1.
> 
> Suggested-by: Laszlo Ersek <lersek@redhat.com>
> Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> Cc: Jordan Justen <jordan.l.justen@intel.com>
> Cc: Tom Lendacky <thomas.lendacky@amd.com>
> Cc: Laszlo Ersek <lersek@redhat.com>
> Contributed-under: TianoCore Contribution Agreement 1.1
> Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
> ---
>  OvmfPkg/Include/IndustryStandard/Virtio10.h | 5 +++++
>  OvmfPkg/VirtioBlkDxe/VirtioBlk.c            | 3 ++-
>  OvmfPkg/VirtioNetDxe/SnpInitialize.c        | 3 ++-
>  OvmfPkg/VirtioRngDxe/VirtioRng.c            | 2 +-
>  OvmfPkg/VirtioScsiDxe/VirtioScsi.c          | 3 ++-
>  5 files changed, 12 insertions(+), 4 deletions(-)
> 
> diff --git a/OvmfPkg/Include/IndustryStandard/Virtio10.h b/OvmfPkg/Include/IndustryStandard/Virtio10.h
> index 4c9b62a3cf59..c62429d94432 100644
> --- a/OvmfPkg/Include/IndustryStandard/Virtio10.h
> +++ b/OvmfPkg/Include/IndustryStandard/Virtio10.h
> @@ -83,4 +83,9 @@ typedef struct {
>  //
>  #define VIRTIO_F_VERSION_1 BIT32
>  
> +//
> +// VirtIo 1.0 reserved iommu platform feature bits
> +//
> +#define VIRTIO_F_IOMMU_PLATFORM BIT33
> +
>  #endif // _VIRTIO_1_0_H_

(1) The comment is not necessary. Namely, the comment that you see just
above the VIRTIO_F_VERSION_1 macro:

//
// VirtIo 1.0 reserved (device-independent) feature bits
//

is meant as the introduction for the *full set* of such bits.

Given that VIRTIO_F_IOMMU_PLATFORM is just another bit in that set,
please just add it right below VIRTIO_F_VERSION_1.

(2) Please update the whitespace for VIRTIO_F_VERSION_1 as well, so that
the replacement texts (BIT32, BIT33) line up nicely.

(3) Please split this change out to a separate patch.

> diff --git a/OvmfPkg/VirtioBlkDxe/VirtioBlk.c b/OvmfPkg/VirtioBlkDxe/VirtioBlk.c
> index 57baceb20a19..a78ec0df9c7b 100644
> --- a/OvmfPkg/VirtioBlkDxe/VirtioBlk.c
> +++ b/OvmfPkg/VirtioBlkDxe/VirtioBlk.c
> @@ -773,7 +773,8 @@ VirtioBlkInit (
>    }
>  
>    Features &= VIRTIO_BLK_F_BLK_SIZE | VIRTIO_BLK_F_TOPOLOGY | VIRTIO_BLK_F_RO |
> -              VIRTIO_BLK_F_FLUSH | VIRTIO_F_VERSION_1;
> +              VIRTIO_BLK_F_FLUSH | VIRTIO_F_VERSION_1 |
> +              VIRTIO_F_IOMMU_PLATFORM;
>  
>    //
>    // In virtio-1.0, feature negotiation is expected to complete before queue

(4) The modifications for this file (VirtioBlk.c) are incomplete.

Namely, in message
<http://mid.mail-archive.com/841bec5f-6f6e-8b1f-25ba-0fd37a915b72@redhat.com>,
points (3) and (5.3), I requested that

> [...] VIRTIO_F_IOMMU_PLATFORM should be treated in OVMF *precisely* in
> parallel with VIRTIO_F_VERSION_1.

Therefore please locate *all* occurrences of VIRTIO_F_VERSION_1, and act
accordingly. In particular, in addition to the above location, we also have:

  //
  // step 5 -- Report understood features.
  //
  if (Dev->VirtIo->Revision < VIRTIO_SPEC_REVISION (1, 0, 0)) {
    Features &= ~(UINT64)VIRTIO_F_VERSION_1;
    Status = Dev->VirtIo->SetGuestFeatures (Dev->VirtIo, Features);
    if (EFI_ERROR (Status)) {
      goto ReleaseQueue;
    }
  }

which should now be changed to:

    Features &= ~(UINT64)(VIRTIO_F_VERSION_1 | VIRTIO_F_IOMMU_PLATFORM);

Same for all the other drivers.

(5) Please split each VIRTIO_F_IOMMU_PLATFORM change to a separate
patch. (That is, one patch per driver.)

- OvmfPkg/Virtio10: define VIRITO_F_IOMMU_PLATFORM feature bit
- OvmfPkg/VirtioBlkDxe: negotiate VIRITO_F_IOMMU_PLATFORM
- OvmfPkg/VirtioNetDxe: negotiate VIRITO_F_IOMMU_PLATFORM
- ...

Thanks,
Laszlo

> diff --git a/OvmfPkg/VirtioNetDxe/SnpInitialize.c b/OvmfPkg/VirtioNetDxe/SnpInitialize.c
> index 9bac19f8e3d1..b389f6243661 100644
> --- a/OvmfPkg/VirtioNetDxe/SnpInitialize.c
> +++ b/OvmfPkg/VirtioNetDxe/SnpInitialize.c
> @@ -536,7 +536,8 @@ VirtioNetInitialize (
>    ASSERT (Dev->Snm.MediaPresentSupported ==
>      !!(Features & VIRTIO_NET_F_STATUS));
>  
> -  Features &= VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | VIRTIO_F_VERSION_1;
> +  Features &= VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | VIRTIO_F_VERSION_1 |
> +              VIRTIO_F_IOMMU_PLATFORM;
>  
>    //
>    // In virtio-1.0, feature negotiation is expected to complete before queue
> diff --git a/OvmfPkg/VirtioRngDxe/VirtioRng.c b/OvmfPkg/VirtioRngDxe/VirtioRng.c
> index b19d90901430..81d8c7facc10 100644
> --- a/OvmfPkg/VirtioRngDxe/VirtioRng.c
> +++ b/OvmfPkg/VirtioRngDxe/VirtioRng.c
> @@ -273,7 +273,7 @@ VirtioRngInit (
>      goto Failed;
>    }
>  
> -  Features &= VIRTIO_F_VERSION_1;
> +  Features &= VIRTIO_F_VERSION_1 | VIRTIO_F_IOMMU_PLATFORM;
>  
>    //
>    // In virtio-1.0, feature negotiation is expected to complete before queue
> diff --git a/OvmfPkg/VirtioScsiDxe/VirtioScsi.c b/OvmfPkg/VirtioScsiDxe/VirtioScsi.c
> index 65e9bda0827a..286b005c994c 100644
> --- a/OvmfPkg/VirtioScsiDxe/VirtioScsi.c
> +++ b/OvmfPkg/VirtioScsiDxe/VirtioScsi.c
> @@ -903,7 +903,8 @@ VirtioScsiInit (
>      goto Failed;
>    }
>  
> -  Features &= VIRTIO_SCSI_F_INOUT | VIRTIO_F_VERSION_1;
> +  Features &= VIRTIO_SCSI_F_INOUT | VIRTIO_F_VERSION_1 |
> +              VIRTIO_F_IOMMU_PLATFORM;
>  
>    //
>    // In virtio-1.0, feature negotiation is expected to complete before queue
>