Message ID | 1473851788-31504-1-git-send-email-will.deacon@arm.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Wed, Sep 14, 2016 at 12:16:28PM +0100, Will Deacon wrote: > Legacy virtio defines the virtqueue base using a 32-bit PFN field, with > a read-only register indicating a fixed page size of 4k. > > This can cause problems for DMA allocators that allocate top down from > the DMA mask, which is set to 64 bits. In this case, the addresses are > silently truncated to 44-bit, leading to IOMMU faults, failure to read > from the queue or data corruption. > > This patch restricts the DMA mask for legacy PCI virtio devices to > 44 bits, which matches the specification. > > Cc: Andy Lutomirski <luto@kernel.org> > Cc: Michael S. Tsirkin <mst@redhat.com> > Cc: Benjamin Serebrin <serebrin@google.com> > Signed-off-by: Will Deacon <will.deacon@arm.com> > --- > drivers/virtio/virtio_pci_legacy.c | 9 +++++++-- > 1 file changed, 7 insertions(+), 2 deletions(-) > > diff --git a/drivers/virtio/virtio_pci_legacy.c b/drivers/virtio/virtio_pci_legacy.c > index 8c4e61783441..efb3f5dff4b7 100644 > --- a/drivers/virtio/virtio_pci_legacy.c > +++ b/drivers/virtio/virtio_pci_legacy.c > @@ -212,12 +212,17 @@ int virtio_pci_legacy_probe(struct virtio_pci_device *vp_dev) > return -ENODEV; > } > > - rc = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64)); > + /* > + * The virtio ring base address is expressed as a 32-bit PFN, with a > + * page size of 1 << VIRTIO_PCI_QUEUE_ADDR_SHIFT. > + */ > + rc = dma_set_mask_and_coherent(&pci_dev->dev, > + DMA_BIT_MASK(32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT)); So why not limit just the coherent mask, as I suggested in a comment on v1? > if (rc) > rc = dma_set_mask_and_coherent(&pci_dev->dev, > DMA_BIT_MASK(32)); > if (rc) > - dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n"); > + dev_warn(&pci_dev->dev, "Failed to enable %d-bit or 32-bit DMA. Trying to continue, but this might not work.\n", 32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT); I'd rather we split this. > > rc = pci_request_region(pci_dev, 0, "virtio-pci-legacy"); > if (rc) > -- > 2.1.4 -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Michael, On Wed, Sep 14, 2016 at 03:42:25PM +0300, Michael S. Tsirkin wrote: > On Wed, Sep 14, 2016 at 12:16:28PM +0100, Will Deacon wrote: > > Legacy virtio defines the virtqueue base using a 32-bit PFN field, with > > a read-only register indicating a fixed page size of 4k. > > > > This can cause problems for DMA allocators that allocate top down from > > the DMA mask, which is set to 64 bits. In this case, the addresses are > > silently truncated to 44-bit, leading to IOMMU faults, failure to read > > from the queue or data corruption. > > > > This patch restricts the DMA mask for legacy PCI virtio devices to > > 44 bits, which matches the specification. > > > > Cc: Andy Lutomirski <luto@kernel.org> > > Cc: Michael S. Tsirkin <mst@redhat.com> > > Cc: Benjamin Serebrin <serebrin@google.com> > > Signed-off-by: Will Deacon <will.deacon@arm.com> > > --- > > drivers/virtio/virtio_pci_legacy.c | 9 +++++++-- > > 1 file changed, 7 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/virtio/virtio_pci_legacy.c b/drivers/virtio/virtio_pci_legacy.c > > index 8c4e61783441..efb3f5dff4b7 100644 > > --- a/drivers/virtio/virtio_pci_legacy.c > > +++ b/drivers/virtio/virtio_pci_legacy.c > > @@ -212,12 +212,17 @@ int virtio_pci_legacy_probe(struct virtio_pci_device *vp_dev) > > return -ENODEV; > > } > > > > - rc = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64)); > > + /* > > + * The virtio ring base address is expressed as a 32-bit PFN, with a > > + * page size of 1 << VIRTIO_PCI_QUEUE_ADDR_SHIFT. > > + */ > > + rc = dma_set_mask_and_coherent(&pci_dev->dev, > > + DMA_BIT_MASK(32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT)); > > So why not limit just the coherent mask, as I suggested in a comment on v1? Sorry, I completely missed that suggestion. I'll roll a v3 with that included, although I'll rejig things a bit to make the error handling a bit more bearable. > > if (rc) > > rc = dma_set_mask_and_coherent(&pci_dev->dev, > > DMA_BIT_MASK(32)); > > if (rc) > > - dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n"); > > + dev_warn(&pci_dev->dev, "Failed to enable %d-bit or 32-bit DMA. Trying to continue, but this might not work.\n", 32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT); > > > I'd rather we split this. I'll actually just leave like it was in the first place if we're trying a 64-bit streaming mask. Stay tuned, Will -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/virtio/virtio_pci_legacy.c b/drivers/virtio/virtio_pci_legacy.c index 8c4e61783441..efb3f5dff4b7 100644 --- a/drivers/virtio/virtio_pci_legacy.c +++ b/drivers/virtio/virtio_pci_legacy.c @@ -212,12 +212,17 @@ int virtio_pci_legacy_probe(struct virtio_pci_device *vp_dev) return -ENODEV; } - rc = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64)); + /* + * The virtio ring base address is expressed as a 32-bit PFN, with a + * page size of 1 << VIRTIO_PCI_QUEUE_ADDR_SHIFT. + */ + rc = dma_set_mask_and_coherent(&pci_dev->dev, + DMA_BIT_MASK(32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT)); if (rc) rc = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(32)); if (rc) - dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n"); + dev_warn(&pci_dev->dev, "Failed to enable %d-bit or 32-bit DMA. Trying to continue, but this might not work.\n", 32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT); rc = pci_request_region(pci_dev, 0, "virtio-pci-legacy"); if (rc)
Legacy virtio defines the virtqueue base using a 32-bit PFN field, with a read-only register indicating a fixed page size of 4k. This can cause problems for DMA allocators that allocate top down from the DMA mask, which is set to 64 bits. In this case, the addresses are silently truncated to 44-bit, leading to IOMMU faults, failure to read from the queue or data corruption. This patch restricts the DMA mask for legacy PCI virtio devices to 44 bits, which matches the specification. Cc: Andy Lutomirski <luto@kernel.org> Cc: Michael S. Tsirkin <mst@redhat.com> Cc: Benjamin Serebrin <serebrin@google.com> Signed-off-by: Will Deacon <will.deacon@arm.com> --- drivers/virtio/virtio_pci_legacy.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)