The current kernel implementation, which grabs the idle VMID from pool when emitting the job may:
The back-to-back submission from one process could use different VMID.
The submission to different queues from single process could use different VMID
It works well in most case but cannot work for the SQ thread trace capture.
The VMID for the submission that set the {SQTT}_BASE, which refers to the address of the trace buffer, is stored in shader engine.
If the profiling application have to use different VMIDs to submit IBs in its life cycle:
Some trace is not captured since it actually uses different VMID to submit jobs.
Some part of captured trace may come from different application since they are accidentally uses the ownerâ??s VMID to submit jobs.
V2:
1. address Christian's comments:
a. drop context flags for tag process, instead, add vm ioctl.
b. change order of patches.
c. sync waiting only when vm flush needs.
2. address Alex's comments;
bump module version
V3:
address Jerry and Christian's comments.
and only reserve gfxhub vmid
v4:
address Jerry and Christian's comments.
fix some race condistions.
v5:
patch#5: peek_fence instead of get_fence.
fix potential context starved.
Chunming Zhou (6):
drm/amdgpu: add vm ioctl
drm/amdgpu: add reserved vmid field in vm struct v2
drm/amdgpu: reserve/unreserve vmid by vm ioctl v4
drm/amdgpu: add limitation for dedicated vm number v4
drm/amdgpu: implement grab reserved vmid V4
drm/amdgpu: bump module verion for reserved vmid
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 159 +++++++++++++++++++++++++++++++-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 6 ++
include/uapi/drm/amdgpu_drm.h | 22 +++++
5 files changed, 185 insertions(+), 6 deletions(-)
--
1.9.1
