On 2022-03-09 12:41, David Yat Sin wrote:
Set dmabuf handle to invalid for BOs that cannot be accessed using SDMA
during checkpoint/restore.
Signed-off-by: David Yat Sin <david.yatsin@xxxxxxx>
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 8 ++++++--
include/uapi/linux/kfd_ioctl.h | 2 ++
2 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index e1e2362841f8..1ffa976ad318 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1767,7 +1767,9 @@ static int criu_checkpoint_bos(struct kfd_process *p,
&bo_bucket->dmabuf_fd);
if (ret)
goto exit;
- }
+ } else
+ bo_bucket->dmabuf_fd = KFD_INVALID_FD;
Minor nit-pick: It would be better to use {} around the else-branch for
consistency with the if-branch. Same below.
Ideally, this should have been part of the patch that bumped the KFD
version to 1.8. Alex, is there a way to squash this when you send this
in a pull-request for drm-next? Maybe if we create the commit with "git
commit --fixup" you can let auto-squash handle it.
Other than that, the patch looks good to me.
Regards,
Felix
+
if (bo_bucket->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL)
bo_bucket->offset = KFD_MMAP_TYPE_DOORBELL |
KFD_MMAP_GPU_ID(pdd->dev->id);
@@ -2219,7 +2221,9 @@ static int criu_restore_bo(struct kfd_process *p,
&bo_bucket->dmabuf_fd);
if (ret)
return ret;
- }
+ } else
+ bo_bucket->dmabuf_fd = KFD_INVALID_FD;
+
return 0;
}
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index eb9ff85f8556..42975e940758 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -196,6 +196,8 @@ struct kfd_ioctl_dbg_wave_control_args {
__u32 buf_size_in_bytes; /*including gpu_id and buf_size */
};
+#define KFD_INVALID_FD 0xffffffff
+
/* Matching HSA_EVENTTYPE */
#define KFD_IOC_EVENT_SIGNAL 0
#define KFD_IOC_EVENT_NODECHANGE 1
