On Thu, Mar 12, 2015 at 12:33:36PM +1030, Rusty Russell wrote:> "Michael S. Tsirkin" <mst at redhat.com> writes: > > Going over the virtio mmio code, I noticed that it doesn't correctly > > return device config values in LE format when using virtio 1.0. > > Borrow code from virtio_pci_modern to do this correctly. > > AFAICT, it doesn't need to. The endian correction is done by the > callers. > > The only reason that virtio_pci_modern() does it is because readl() etc > do endian conversion, and we don't want them to. And the PCI part of > the spec says to use "natural" accessors, so we don't do byte-at-a-time. > > Cheers, > Rusty.You are right, the endina-ness is not an issue, so the commit log I wrote is wrong, but I still think the patch is required, because MMIO spec says the same as PCI. The driver MUST only use 32 bit wide and aligned reads and writes to access the control registers described in table 4.1. For the device-specific configuration space, the driver MUST use 8 bit wide accesses for 8 bit wide fields, 16 bit wide and aligned accesses for 16 bit wide fields and 32 bit wide and aligned accesses for 32 and 64 bit wide fields. Here's a better commit log: ---> virtio_mmio: fix access width for mmio Going over the virtio mmio code, I noticed that it doesn't correctly access modern device config values using "natural" accessors: it uses readb to get/set them byte by byte, while the virtio 1.0 spec explicitly states: 4.2.2.2 Driver Requirements: MMIO Device Register Layout ... The driver MUST only use 32 bit wide and aligned reads and writes to access the control registers described in table 4.1. For the device-specific configuration space, the driver MUST use 8 bit wide accesses for 8 bit wide fields, 16 bit wide and aligned accesses for 16 bit wide fields and 32 bit wide and aligned accesses for 32 and 64 bit wide fields. Borrow code from virtio_pci_modern to do this correctly. Signed-off-by: Michael S. Tsirkin <mst at redhat.com> Makes sense now, right? Want me to repost or can you just tweak the commit log?> > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > > --- > > > > Note: untested: QEMU doesn't support virtio 1.0 for virtio-mmio. > > Pawel, could you please confirm this patch makes sense? > > > > drivers/virtio/virtio_mmio.c | 79 +++++++++++++++++++++++++++++++++++++++----- > > 1 file changed, 71 insertions(+), 8 deletions(-) > > > > diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c > > index cad5698..0375456 100644 > > --- a/drivers/virtio/virtio_mmio.c > > +++ b/drivers/virtio/virtio_mmio.c > > @@ -156,22 +156,85 @@ static void vm_get(struct virtio_device *vdev, unsigned offset, > > void *buf, unsigned len) > > { > > struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev); > > - u8 *ptr = buf; > > - int i; > > + void __iomem *base = vm_dev->base + VIRTIO_MMIO_CONFIG; > > + u8 b; > > + __le16 w; > > + __le32 l; > > > > - for (i = 0; i < len; i++) > > - ptr[i] = readb(vm_dev->base + VIRTIO_MMIO_CONFIG + offset + i); > > + if (vm_dev->version == 1) { > > + u8 *ptr = buf; > > + int i; > > + > > + for (i = 0; i < len; i++) > > + ptr[i] = readb(base + offset + i); > > + return; > > + } > > + > > + switch (len) { > > + case 1: > > + b = readb(base + offset); > > + memcpy(buf, &b, sizeof b); > > + break; > > + case 2: > > + w = cpu_to_le16(readw(base + offset)); > > + memcpy(buf, &w, sizeof w); > > + break; > > + case 4: > > + l = cpu_to_le32(readl(base + offset)); > > + memcpy(buf, &l, sizeof l); > > + break; > > + case 8: > > + l = cpu_to_le32(readl(base + offset)); > > + memcpy(buf, &l, sizeof l); > > + l = cpu_to_le32(ioread32(base + offset + sizeof l)); > > + memcpy(buf + sizeof l, &l, sizeof l); > > + break; > > + default: > > + BUG(); > > + } > > } > > > > static void vm_set(struct virtio_device *vdev, unsigned offset, > > const void *buf, unsigned len) > > { > > struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev); > > - const u8 *ptr = buf; > > - int i; > > + void __iomem *base = vm_dev->base + VIRTIO_MMIO_CONFIG; > > + u8 b; > > + __le16 w; > > + __le32 l; > > > > - for (i = 0; i < len; i++) > > - writeb(ptr[i], vm_dev->base + VIRTIO_MMIO_CONFIG + offset + i); > > + if (vm_dev->version == 1) { > > + const u8 *ptr = buf; > > + int i; > > + > > + for (i = 0; i < len; i++) > > + writeb(ptr[i], base + offset + i); > > + > > + return; > > + } > > + > > + switch (len) { > > + case 1: > > + memcpy(&b, buf, sizeof b); > > + writeb(b, base + offset); > > + break; > > + case 2: > > + memcpy(&w, buf, sizeof w); > > + writew(le16_to_cpu(w), base + offset); > > + break; > > + case 4: > > + memcpy(&l, buf, sizeof l); > > + writel(le32_to_cpu(l), base + offset); > > + break; > > + case 8: > > + memcpy(&l, buf, sizeof l); > > + writel(le32_to_cpu(l), base + offset); > > + memcpy(&l, buf + sizeof l, sizeof l); > > + writel(le32_to_cpu(l), base + offset + sizeof l); > > + break; > > + default: > > + BUG(); > > + } > > } > > > > static u8 vm_get_status(struct virtio_device *vdev) > > -- > > MST
"Michael S. Tsirkin" <mst at redhat.com> writes:> On Thu, Mar 12, 2015 at 12:33:36PM +1030, Rusty Russell wrote: >> "Michael S. Tsirkin" <mst at redhat.com> writes: >> > Going over the virtio mmio code, I noticed that it doesn't correctly >> > return device config values in LE format when using virtio 1.0. >> > Borrow code from virtio_pci_modern to do this correctly. >> >> AFAICT, it doesn't need to. The endian correction is done by the >> callers. >> >> The only reason that virtio_pci_modern() does it is because readl() etc >> do endian conversion, and we don't want them to. And the PCI part of >> the spec says to use "natural" accessors, so we don't do byte-at-a-time. >> >> Cheers, >> Rusty. > > You are right, the endina-ness is not an issue, so the commit log I > wrote is wrong, but I still think the patch is required, because MMIO > spec says the same as PCI. > The driver MUST only use 32 bit wide and aligned reads and writes to > access the control registers described > in table 4.1. For the device-specific configuration space, the driver > MUST use 8 bit wide accesses for 8 bit > wide fields, 16 bit wide and aligned accesses for 16 bit wide fields and > 32 bit wide and aligned accesses for > 32 and 64 bit wide fields.Ah, good point. I'm sure Pawel is on a beach somewhere sipping cocktails, so I'll apply this immediately (with your updated commit message) to my pending-rebases and give him the weekend to Ack. Thanks, Rusty.> > > Here's a better commit log: > > ---> > > virtio_mmio: fix access width for mmio > > Going over the virtio mmio code, I noticed that it doesn't correctly > access modern device config values using "natural" accessors: it uses > readb to get/set them byte by byte, while the virtio 1.0 spec explicitly states: > > 4.2.2.2 Driver Requirements: MMIO Device Register Layout > > ... > > The driver MUST only use 32 bit wide and aligned reads and writes to > access the control registers described in table 4.1. > For the device-specific configuration space, the driver MUST use > 8 bit wide accesses for 8 bit wide fields, 16 bit wide and aligned > accesses for 16 bit wide fields and 32 bit wide and aligned accesses for > 32 and 64 bit wide fields. > > Borrow code from virtio_pci_modern to do this correctly. > > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > > Makes sense now, right? > Want me to repost or can you just tweak the commit log? > > >> > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> >> > --- >> > >> > Note: untested: QEMU doesn't support virtio 1.0 for virtio-mmio. >> > Pawel, could you please confirm this patch makes sense? >> > >> > drivers/virtio/virtio_mmio.c | 79 +++++++++++++++++++++++++++++++++++++++----- >> > 1 file changed, 71 insertions(+), 8 deletions(-) >> > >> > diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c >> > index cad5698..0375456 100644 >> > --- a/drivers/virtio/virtio_mmio.c >> > +++ b/drivers/virtio/virtio_mmio.c >> > @@ -156,22 +156,85 @@ static void vm_get(struct virtio_device *vdev, unsigned offset, >> > void *buf, unsigned len) >> > { >> > struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev); >> > - u8 *ptr = buf; >> > - int i; >> > + void __iomem *base = vm_dev->base + VIRTIO_MMIO_CONFIG; >> > + u8 b; >> > + __le16 w; >> > + __le32 l; >> > >> > - for (i = 0; i < len; i++) >> > - ptr[i] = readb(vm_dev->base + VIRTIO_MMIO_CONFIG + offset + i); >> > + if (vm_dev->version == 1) { >> > + u8 *ptr = buf; >> > + int i; >> > + >> > + for (i = 0; i < len; i++) >> > + ptr[i] = readb(base + offset + i); >> > + return; >> > + } >> > + >> > + switch (len) { >> > + case 1: >> > + b = readb(base + offset); >> > + memcpy(buf, &b, sizeof b); >> > + break; >> > + case 2: >> > + w = cpu_to_le16(readw(base + offset)); >> > + memcpy(buf, &w, sizeof w); >> > + break; >> > + case 4: >> > + l = cpu_to_le32(readl(base + offset)); >> > + memcpy(buf, &l, sizeof l); >> > + break; >> > + case 8: >> > + l = cpu_to_le32(readl(base + offset)); >> > + memcpy(buf, &l, sizeof l); >> > + l = cpu_to_le32(ioread32(base + offset + sizeof l)); >> > + memcpy(buf + sizeof l, &l, sizeof l); >> > + break; >> > + default: >> > + BUG(); >> > + } >> > } >> > >> > static void vm_set(struct virtio_device *vdev, unsigned offset, >> > const void *buf, unsigned len) >> > { >> > struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev); >> > - const u8 *ptr = buf; >> > - int i; >> > + void __iomem *base = vm_dev->base + VIRTIO_MMIO_CONFIG; >> > + u8 b; >> > + __le16 w; >> > + __le32 l; >> > >> > - for (i = 0; i < len; i++) >> > - writeb(ptr[i], vm_dev->base + VIRTIO_MMIO_CONFIG + offset + i); >> > + if (vm_dev->version == 1) { >> > + const u8 *ptr = buf; >> > + int i; >> > + >> > + for (i = 0; i < len; i++) >> > + writeb(ptr[i], base + offset + i); >> > + >> > + return; >> > + } >> > + >> > + switch (len) { >> > + case 1: >> > + memcpy(&b, buf, sizeof b); >> > + writeb(b, base + offset); >> > + break; >> > + case 2: >> > + memcpy(&w, buf, sizeof w); >> > + writew(le16_to_cpu(w), base + offset); >> > + break; >> > + case 4: >> > + memcpy(&l, buf, sizeof l); >> > + writel(le32_to_cpu(l), base + offset); >> > + break; >> > + case 8: >> > + memcpy(&l, buf, sizeof l); >> > + writel(le32_to_cpu(l), base + offset); >> > + memcpy(&l, buf + sizeof l, sizeof l); >> > + writel(le32_to_cpu(l), base + offset + sizeof l); >> > + break; >> > + default: >> > + BUG(); >> > + } >> > } >> > >> > static u8 vm_get_status(struct virtio_device *vdev) >> > -- >> > MST
On Fri, 2015-03-13 at 01:22 +0000, Rusty Russell wrote:> I'm sure Pawel is on a beach somewhere sipping cocktails,Ehm... There's still a some more prohibition for me (granted, by choice, just to share the "pain" ;-), so it wasn't exactly> so I'll apply > this immediately (with your updated commit message) to my > pending-rebases and give him the weekend to Ack.Not that it matters any more, but a belated Acked-by: Pawel Moll <pawel.moll at arm.com> Thanks, Michael, for fixing this! (although, I must admit, I haven't tested it yet - all in right time :-) Pawel