Used for communication with VFIO driver
(prep work for vfio-user, which will communicate over a socket)
Signed-off-by: John G Johnson <john.g.johnson@oracle.com>
---
include/hw/vfio/vfio-common.h | 24 ++++++++
hw/vfio/common.c | 128 ++++++++++++++++++++++++++++--------------
2 files changed, 110 insertions(+), 42 deletions(-)
diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
index e573f5a..953bc0f 100644
--- a/include/hw/vfio/vfio-common.h
+++ b/include/hw/vfio/vfio-common.h
@@ -75,6 +75,7 @@ typedef struct VFIOAddressSpace {
} VFIOAddressSpace;
struct VFIOGroup;
+typedef struct VFIOContainerIO VFIOContainerIO;
typedef struct VFIOContainer {
VFIOAddressSpace *space;
@@ -83,6 +84,7 @@ typedef struct VFIOContainer {
MemoryListener prereg_listener;
unsigned iommu_type;
Error *error;
+ VFIOContainerIO *io;
bool initialized;
bool dirty_pages_supported;
uint64_t dirty_pgsizes;
@@ -154,6 +156,28 @@ struct VFIODeviceOps {
int (*vfio_load_config)(VFIODevice *vdev, QEMUFile *f);
};
+#ifdef CONFIG_LINUX
+
+/*
+ * The next 2 ops vectors are how Devices and Containers
+ * communicate with the server. The default option is
+ * through ioctl() to the kernel VFIO driver, but vfio-user
+ * can use a socket to a remote process.
+ */
+
+struct VFIOContainerIO {
+ int (*dma_map)(VFIOContainer *container,
+ struct vfio_iommu_type1_dma_map *map);
+ int (*dma_unmap)(VFIOContainer *container,
+ struct vfio_iommu_type1_dma_unmap *unmap,
+ struct vfio_bitmap *bitmap);
+ int (*dirty_bitmap)(VFIOContainer *container,
+ struct vfio_iommu_type1_dirty_bitmap *bitmap,
+ struct vfio_iommu_type1_dirty_bitmap_get *range);
+};
+
+#endif /* CONFIG_LINUX */
+
typedef struct VFIOGroup {
int fd;
int groupid;
diff --git a/hw/vfio/common.c b/hw/vfio/common.c
index ace9562..9310a7f 100644
--- a/hw/vfio/common.c
+++ b/hw/vfio/common.c
@@ -58,6 +58,8 @@ static QLIST_HEAD(, VFIOAddressSpace) vfio_address_spaces =
static int vfio_kvm_device_fd = -1;
#endif
+static VFIOContainerIO vfio_cont_io_ioctl;
+
/*
* Common VFIO interrupt disable
*/
@@ -432,12 +434,12 @@ static int vfio_dma_unmap_bitmap(VFIOContainer *container,
goto unmap_exit;
}
- ret = ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, unmap);
+ ret = container->io->dma_unmap(container, unmap, bitmap);
if (!ret) {
cpu_physical_memory_set_dirty_lebitmap((unsigned long *)bitmap->data,
iotlb->translated_addr, pages);
} else {
- error_report("VFIO_UNMAP_DMA with DIRTY_BITMAP : %m");
+ error_report("VFIO_UNMAP_DMA with DIRTY_BITMAP : %s", strerror(-ret));
}
g_free(bitmap->data);
@@ -465,30 +467,7 @@ static int vfio_dma_unmap(VFIOContainer *container,
return vfio_dma_unmap_bitmap(container, iova, size, iotlb);
}
- while (ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, &unmap)) {
- /*
- * The type1 backend has an off-by-one bug in the kernel (71a7d3d78e3c
- * v4.15) where an overflow in its wrap-around check prevents us from
- * unmapping the last page of the address space. Test for the error
- * condition and re-try the unmap excluding the last page. The
- * expectation is that we've never mapped the last page anyway and this
- * unmap request comes via vIOMMU support which also makes it unlikely
- * that this page is used. This bug was introduced well after type1 v2
- * support was introduced, so we shouldn't need to test for v1. A fix
- * is queued for kernel v5.0 so this workaround can be removed once
- * affected kernels are sufficiently deprecated.
- */
- if (errno == EINVAL && unmap.size && !(unmap.iova + unmap.size) &&
- container->iommu_type == VFIO_TYPE1v2_IOMMU) {
- trace_vfio_dma_unmap_overflow_workaround();
- unmap.size -= 1ULL << ctz64(container->pgsizes);
- continue;
- }
- error_report("VFIO_UNMAP_DMA failed: %s", strerror(errno));
- return -errno;
- }
-
- return 0;
+ return container->io->dma_unmap(container, &unmap, NULL);
}
static int vfio_dma_map(VFIOContainer *container, hwaddr iova,
@@ -501,24 +480,18 @@ static int vfio_dma_map(VFIOContainer *container, hwaddr iova,
.iova = iova,
.size = size,
};
+ int ret;
if (!readonly) {
map.flags |= VFIO_DMA_MAP_FLAG_WRITE;
}
- /*
- * Try the mapping, if it fails with EBUSY, unmap the region and try
- * again. This shouldn't be necessary, but we sometimes see it in
- * the VGA ROM space.
- */
- if (ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0 ||
- (errno == EBUSY && vfio_dma_unmap(container, iova, size, NULL) == 0 &&
- ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0)) {
- return 0;
- }
+ ret = container->io->dma_map(container, &map);
- error_report("VFIO_MAP_DMA failed: %s", strerror(errno));
- return -errno;
+ if (ret < 0) {
+ error_report("VFIO_MAP_DMA failed: %s", strerror(-ret));
+ }
+ return ret;
}
static void vfio_host_win_add(VFIOContainer *container,
@@ -1263,10 +1236,10 @@ static void vfio_set_dirty_page_tracking(VFIOContainer *container, bool start)
dirty.flags = VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP;
}
- ret = ioctl(container->fd, VFIO_IOMMU_DIRTY_PAGES, &dirty);
+ ret = container->io->dirty_bitmap(container, &dirty, NULL);
if (ret) {
error_report("Failed to set dirty tracking flag 0x%x errno: %d",
- dirty.flags, errno);
+ dirty.flags, -ret);
}
}
@@ -1316,11 +1289,11 @@ static int vfio_get_dirty_bitmap(VFIOContainer *container, uint64_t iova,
goto err_out;
}
- ret = ioctl(container->fd, VFIO_IOMMU_DIRTY_PAGES, dbitmap);
+ ret = container->io->dirty_bitmap(container, dbitmap, range);
if (ret) {
error_report("Failed to get dirty bitmap for iova: 0x%"PRIx64
" size: 0x%"PRIx64" err: %d", (uint64_t)range->iova,
- (uint64_t)range->size, errno);
+ (uint64_t)range->size, -ret);
goto err_out;
}
@@ -2090,6 +2063,7 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as,
container->error = NULL;
container->dirty_pages_supported = false;
container->dma_max_mappings = 0;
+ container->io = &vfio_cont_io_ioctl;
QLIST_INIT(&container->giommu_list);
QLIST_INIT(&container->hostwin_list);
QLIST_INIT(&container->vrdl_list);
@@ -2626,3 +2600,73 @@ int vfio_eeh_as_op(AddressSpace *as, uint32_t op)
}
return vfio_eeh_container_op(container, op);
}
+
+/*
+ * Traditional ioctl() based io
+ */
+
+static int vfio_io_dma_map(VFIOContainer *container,
+ struct vfio_iommu_type1_dma_map *map)
+{
+
+ /*
+ * Try the mapping, if it fails with EBUSY, unmap the region and try
+ * again. This shouldn't be necessary, but we sometimes see it in
+ * the VGA ROM space.
+ */
+ if (ioctl(container->fd, VFIO_IOMMU_MAP_DMA, map) == 0 ||
+ (errno == EBUSY &&
+ vfio_dma_unmap(container, map->iova, map->size, NULL) == 0 &&
+ ioctl(container->fd, VFIO_IOMMU_MAP_DMA, map) == 0)) {
+ return 0;
+ }
+ return -errno;
+}
+
+static int vfio_io_dma_unmap(VFIOContainer *container,
+ struct vfio_iommu_type1_dma_unmap *unmap,
+ struct vfio_bitmap *bitmap)
+{
+
+ while (ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, unmap)) {
+ /*
+ * The type1 backend has an off-by-one bug in the kernel (71a7d3d78e3c
+ * v4.15) where an overflow in its wrap-around check prevents us from
+ * unmapping the last page of the address space. Test for the error
+ * condition and re-try the unmap excluding the last page. The
+ * expectation is that we've never mapped the last page anyway and this
+ * unmap request comes via vIOMMU support which also makes it unlikely
+ * that this page is used. This bug was introduced well after type1 v2
+ * support was introduced, so we shouldn't need to test for v1. A fix
+ * is queued for kernel v5.0 so this workaround can be removed once
+ * affected kernels are sufficiently deprecated.
+ */
+ if (errno == EINVAL && unmap->size && !(unmap->iova + unmap->size) &&
+ container->iommu_type == VFIO_TYPE1v2_IOMMU) {
+ trace_vfio_dma_unmap_overflow_workaround();
+ unmap->size -= 1ULL << ctz64(container->pgsizes);
+ continue;
+ }
+ error_report("VFIO_UNMAP_DMA failed: %s", strerror(errno));
+ return -errno;
+ }
+
+ return 0;
+}
+
+static int vfio_io_dirty_bitmap(VFIOContainer *container,
+ struct vfio_iommu_type1_dirty_bitmap *bitmap,
+ struct vfio_iommu_type1_dirty_bitmap_get *range)
+{
+ int ret;
+
+ ret = ioctl(container->fd, VFIO_IOMMU_DIRTY_PAGES, bitmap);
+
+ return ret < 0 ? -errno : ret;
+}
+
+static VFIOContainerIO vfio_cont_io_ioctl = {
+ .dma_map = vfio_io_dma_map,
+ .dma_unmap = vfio_io_dma_unmap,
+ .dirty_bitmap = vfio_io_dirty_bitmap,
+};
--
1.9.4On Wed, 1 Feb 2023 21:55:39 -0800
John Johnson <john.g.johnson@oracle.com> wrote:
> Used for communication with VFIO driver
> (prep work for vfio-user, which will communicate over a socket)
>
> Signed-off-by: John G Johnson <john.g.johnson@oracle.com>
> ---
> include/hw/vfio/vfio-common.h | 24 ++++++++
> hw/vfio/common.c | 128 ++++++++++++++++++++++++++++--------------
> 2 files changed, 110 insertions(+), 42 deletions(-)
>
> diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
> index e573f5a..953bc0f 100644
> --- a/include/hw/vfio/vfio-common.h
> +++ b/include/hw/vfio/vfio-common.h
> @@ -75,6 +75,7 @@ typedef struct VFIOAddressSpace {
> } VFIOAddressSpace;
>
> struct VFIOGroup;
> +typedef struct VFIOContainerIO VFIOContainerIO;
>
> typedef struct VFIOContainer {
> VFIOAddressSpace *space;
> @@ -83,6 +84,7 @@ typedef struct VFIOContainer {
> MemoryListener prereg_listener;
> unsigned iommu_type;
> Error *error;
> + VFIOContainerIO *io;
> bool initialized;
> bool dirty_pages_supported;
> uint64_t dirty_pgsizes;
> @@ -154,6 +156,28 @@ struct VFIODeviceOps {
> int (*vfio_load_config)(VFIODevice *vdev, QEMUFile *f);
> };
>
> +#ifdef CONFIG_LINUX
> +
> +/*
> + * The next 2 ops vectors are how Devices and Containers
> + * communicate with the server. The default option is
> + * through ioctl() to the kernel VFIO driver, but vfio-user
> + * can use a socket to a remote process.
> + */
> +
> +struct VFIOContainerIO {
> + int (*dma_map)(VFIOContainer *container,
> + struct vfio_iommu_type1_dma_map *map);
> + int (*dma_unmap)(VFIOContainer *container,
> + struct vfio_iommu_type1_dma_unmap *unmap,
> + struct vfio_bitmap *bitmap);
> + int (*dirty_bitmap)(VFIOContainer *container,
> + struct vfio_iommu_type1_dirty_bitmap *bitmap,
> + struct vfio_iommu_type1_dirty_bitmap_get *range);
> +};
> +
> +#endif /* CONFIG_LINUX */
> +
> typedef struct VFIOGroup {
> int fd;
> int groupid;
> diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> index ace9562..9310a7f 100644
> --- a/hw/vfio/common.c
> +++ b/hw/vfio/common.c
> @@ -58,6 +58,8 @@ static QLIST_HEAD(, VFIOAddressSpace) vfio_address_spaces =
> static int vfio_kvm_device_fd = -1;
> #endif
>
> +static VFIOContainerIO vfio_cont_io_ioctl;
> +
> /*
> * Common VFIO interrupt disable
> */
> @@ -432,12 +434,12 @@ static int vfio_dma_unmap_bitmap(VFIOContainer *container,
> goto unmap_exit;
> }
>
> - ret = ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, unmap);
> + ret = container->io->dma_unmap(container, unmap, bitmap);
The bitmap arg doesn't make a lot of sense here, its use doesn't come
around until vfio_user_dma_unmap() is added, but even then, it's
readily available at &unmap->data with validity determined by
unmap->flags. The eventual sanity test in vfio_user_dma_unmap() really
only seems to prove the arg is unnecessary. Thanks,
Alex
On Fri, 3 Feb 2023 15:21:09 -0700
Alex Williamson <alex.williamson@redhat.com> wrote:
> On Wed, 1 Feb 2023 21:55:39 -0800
> John Johnson <john.g.johnson@oracle.com> wrote:
>
> > Used for communication with VFIO driver
> > (prep work for vfio-user, which will communicate over a socket)
> >
> > Signed-off-by: John G Johnson <john.g.johnson@oracle.com>
> > ---
> > include/hw/vfio/vfio-common.h | 24 ++++++++
> > hw/vfio/common.c | 128 ++++++++++++++++++++++++++++--------------
> > 2 files changed, 110 insertions(+), 42 deletions(-)
> >
> > diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
> > index e573f5a..953bc0f 100644
> > --- a/include/hw/vfio/vfio-common.h
> > +++ b/include/hw/vfio/vfio-common.h
> > @@ -75,6 +75,7 @@ typedef struct VFIOAddressSpace {
> > } VFIOAddressSpace;
> >
> > struct VFIOGroup;
> > +typedef struct VFIOContainerIO VFIOContainerIO;
> >
> > typedef struct VFIOContainer {
> > VFIOAddressSpace *space;
> > @@ -83,6 +84,7 @@ typedef struct VFIOContainer {
> > MemoryListener prereg_listener;
> > unsigned iommu_type;
> > Error *error;
> > + VFIOContainerIO *io;
> > bool initialized;
> > bool dirty_pages_supported;
> > uint64_t dirty_pgsizes;
> > @@ -154,6 +156,28 @@ struct VFIODeviceOps {
> > int (*vfio_load_config)(VFIODevice *vdev, QEMUFile *f);
> > };
> >
> > +#ifdef CONFIG_LINUX
> > +
> > +/*
> > + * The next 2 ops vectors are how Devices and Containers
> > + * communicate with the server. The default option is
> > + * through ioctl() to the kernel VFIO driver, but vfio-user
> > + * can use a socket to a remote process.
> > + */
> > +
> > +struct VFIOContainerIO {
> > + int (*dma_map)(VFIOContainer *container,
> > + struct vfio_iommu_type1_dma_map *map);
> > + int (*dma_unmap)(VFIOContainer *container,
> > + struct vfio_iommu_type1_dma_unmap *unmap,
> > + struct vfio_bitmap *bitmap);
> > + int (*dirty_bitmap)(VFIOContainer *container,
> > + struct vfio_iommu_type1_dirty_bitmap *bitmap,
> > + struct vfio_iommu_type1_dirty_bitmap_get *range);
> > +};
> > +
> > +#endif /* CONFIG_LINUX */
> > +
> > typedef struct VFIOGroup {
> > int fd;
> > int groupid;
> > diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> > index ace9562..9310a7f 100644
> > --- a/hw/vfio/common.c
> > +++ b/hw/vfio/common.c
> > @@ -58,6 +58,8 @@ static QLIST_HEAD(, VFIOAddressSpace) vfio_address_spaces =
> > static int vfio_kvm_device_fd = -1;
> > #endif
> >
> > +static VFIOContainerIO vfio_cont_io_ioctl;
> > +
> > /*
> > * Common VFIO interrupt disable
> > */
> > @@ -432,12 +434,12 @@ static int vfio_dma_unmap_bitmap(VFIOContainer *container,
> > goto unmap_exit;
> > }
> >
> > - ret = ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, unmap);
> > + ret = container->io->dma_unmap(container, unmap, bitmap);
>
> The bitmap arg doesn't make a lot of sense here, its use doesn't come
> around until vfio_user_dma_unmap() is added, but even then, it's
> readily available at &unmap->data with validity determined by
> unmap->flags. The eventual sanity test in vfio_user_dma_unmap() really
> only seems to prove the arg is unnecessary. Thanks,
The same comment really applies to .dirty_bitmap() with respect to the
range arg, but I think it's also important to note that both of these
are all but deprecated interfaces for the kernel. The migration
series[1] is pre-enabling these kernel interfaces to be removed by
adding support to QEMU to dirty all pages when support isn't present,
then we replace it with what we expect to be a longer term solution
with device dirty tracking support[2]. All the more reason not to make
these part of the internal API shared by kernel and user versions.
Thanks,
Alex
[1]https://lore.kernel.org/all/20230116141135.12021-1-avihaih@nvidia.com/
[2]https://lore.kernel.org/all/20230126184948.10478-1-avihaih@nvidia.com/
© 2016 - 2026 Red Hat, Inc.