On 5/20/25 17:03, John Levon wrote:
For vfio-user, each region has its own fd rather than sharing
vbasedev's. Add the necessary plumbing to support this. For vfio
backends with a shared fd, initialize region->fd to the shared one.
Signed-off-by: John Levon <john.le...@nutanix.com>
---
include/hw/vfio/vfio-device.h | 7 +++++--
include/hw/vfio/vfio-region.h | 1 +
hw/vfio/device.c | 28 ++++++++++++++++++++++++++--
hw/vfio/region.c | 7 ++++++-
4 files changed, 38 insertions(+), 5 deletions(-)
diff --git a/include/hw/vfio/vfio-device.h b/include/hw/vfio/vfio-device.h
index 923f9cd116..5cb817fd6a 100644
--- a/include/hw/vfio/vfio-device.h
+++ b/include/hw/vfio/vfio-device.h
@@ -66,6 +66,7 @@ typedef struct VFIODevice {
OnOffAuto enable_migration;
OnOffAuto migration_multifd_transfer;
bool migration_events;
+ bool use_region_fds;
VFIODeviceOps *ops;
VFIODeviceIOOps *io_ops;
unsigned int num_irqs;
@@ -84,6 +85,7 @@ typedef struct VFIODevice {
VFIOIOASHwpt *hwpt;
QLIST_ENTRY(VFIODevice) hwpt_next;
struct vfio_region_info **reginfo;
+ int *region_fds;
} VFIODevice;
struct VFIODeviceOps {
@@ -172,10 +174,11 @@ struct VFIODeviceIOOps {
/**
* @get_region_info
*
- * Fill in @info with information on the region given by @info->index.
+ * Fill in @info (and optionally @fd) with information on the region given
+ * by @info->index.
The whole VFIODeviceIOOps struct needs better documentation. The arguments
are missing.
*/
int (*get_region_info)(VFIODevice *vdev,
- struct vfio_region_info *info);
+ struct vfio_region_info *info, int *fd);
/**
* @get_irq_info
diff --git a/include/hw/vfio/vfio-region.h b/include/hw/vfio/vfio-region.h
index cbffb26962..80e83b23fd 100644
--- a/include/hw/vfio/vfio-region.h
+++ b/include/hw/vfio/vfio-region.h
@@ -29,6 +29,7 @@ typedef struct VFIORegion {
uint32_t nr_mmaps;
VFIOMmap *mmaps;
uint8_t nr; /* cache the region number for debug */
+ int fd; /* fd to mmap() region */
Could you split this change ? I am not sure it is needed.
} VFIORegion;
diff --git a/hw/vfio/device.c b/hw/vfio/device.c
index d0068086ae..41db403992 100644
--- a/hw/vfio/device.c
+++ b/hw/vfio/device.c
@@ -226,6 +226,7 @@ int vfio_device_get_region_info(VFIODevice *vbasedev, int
index,
struct vfio_region_info **info)
{
size_t argsz = sizeof(struct vfio_region_info);
+ int fd = -1;
int ret;
/* check cache */
@@ -240,7 +241,7 @@ int vfio_device_get_region_info(VFIODevice *vbasedev, int
index,
retry:
(*info)->argsz = argsz;
- ret = vbasedev->io_ops->get_region_info(vbasedev, *info);
+ ret = vbasedev->io_ops->get_region_info(vbasedev, *info, &fd);
if (ret != 0) {
g_free(*info);
*info = NULL;
@@ -251,11 +252,19 @@ retry:
argsz = (*info)->argsz;
*info = g_realloc(*info, argsz);
+ if (fd != -1) {
+ close(fd);
+ fd = -1;
+ }
+
goto retry;
}
/* fill cache */
vbasedev->reginfo[index] = *info;
+ if (vbasedev->region_fds != NULL) {
+ vbasedev->region_fds[index] = fd;
+ }
return 0;
}
@@ -360,6 +369,7 @@ void vfio_device_init(VFIODevice *vbasedev, int type,
VFIODeviceOps *ops,
vbasedev->io_ops = &vfio_device_io_ops_ioctl;
vbasedev->dev = dev;
vbasedev->fd = -1;
+ vbasedev->use_region_fds = false;
I wish we could avoid this bool. I have no idea yet.
vbasedev->ram_block_discard_allowed = ram_discard;
}
@@ -470,6 +480,9 @@ void vfio_device_prepare(VFIODevice *vbasedev,
VFIOContainerBase *bcontainer,
vbasedev->reginfo = g_new0(struct vfio_region_info *,
vbasedev->num_regions);
+ if (vbasedev->use_region_fds) {
+ vbasedev->region_fds = g_new0(int, vbasedev->num_regions);
+ }
}
void vfio_device_unprepare(VFIODevice *vbasedev)
@@ -478,9 +491,17 @@ void vfio_device_unprepare(VFIODevice *vbasedev)
for (i = 0; i < vbasedev->num_regions; i++) {
g_free(vbasedev->reginfo[i]);
+ if (vbasedev->region_fds != NULL && vbasedev->region_fds[i] != -1) {
+ close(vbasedev->region_fds[i]);
+ }
+
}
g_free(vbasedev->reginfo);
vbasedev->reginfo = NULL;
+ if (vbasedev->region_fds != NULL) {
g_free is NULL safe. No need to test for it. g_clear_pointer() is a nice
helper too.
+ g_free(vbasedev->region_fds);
+ vbasedev->region_fds = NULL;
+ }
QLIST_REMOVE(vbasedev, container_next);
QLIST_REMOVE(vbasedev, global_next);
@@ -502,10 +523,13 @@ static int vfio_device_io_device_feature(VFIODevice
*vbasedev,
}
static int vfio_device_io_get_region_info(VFIODevice *vbasedev,
- struct vfio_region_info *info)
+ struct vfio_region_info *info,
+ int *fd)
{
int ret;
+ *fd = -1;
+
ret = ioctl(vbasedev->fd, VFIO_DEVICE_GET_REGION_INFO, info);
return ret < 0 ? -errno : ret;
diff --git a/hw/vfio/region.c b/hw/vfio/region.c
index 34752c3f65..3c93da91d8 100644
--- a/hw/vfio/region.c
+++ b/hw/vfio/region.c
@@ -200,6 +200,11 @@ int vfio_region_setup(Object *obj, VFIODevice *vbasedev,
VFIORegion *region,
region->size = info->size;
region->fd_offset = info->offset;
region->nr = index;
+ if (vbasedev->region_fds != NULL) {
+ region->fd = vbasedev->region_fds[index];
+ } else {
+ region->fd = vbasedev->fd;
+ }
if (region->size) {
region->mem = g_new0(MemoryRegion, 1);
@@ -278,7 +283,7 @@ int vfio_region_mmap(VFIORegion *region)
region->mmaps[i].mmap = mmap(map_align, region->mmaps[i].size, prot,
MAP_SHARED | MAP_FIXED,
- region->vbasedev->fd,
+ region->fd,
Would this work ?
vbasedev->region_fds ? vbasedev->region_fds[region->nr] :
vbasedev->fd,
Thanks,
C.
region->fd_offset +
region->mmaps[i].offset);
if (region->mmaps[i].mmap == MAP_FAILED) {
