Skip to contentSkip to conversation

drm/amdgpu: implement per-process MES context

PATCH Superseded Zhu Lingshan 1 participant 1 message May 28

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 drm/amdgpu: implement per-process MES context

MES process context is a process-level page where process specific context is saved for MES scheduler.

However, current user-queue code path assigns fw_obj of a queue to MES process_context_addr when adding the queue to MES.

This means every new queue from the same process would replace the previous process context address with that queue's fw_obj address.
What's worse is, when user space frees a queue, its fw_obj will be freed as well, causing MES working on a NULL page pointer.

This issue leads to inconsistency and crash in the scheduler.

This commit allocates a process-level page for MES process contexts for a process other than queue-level

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c  |  6 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h  |  2 +
 drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 52 +++++++++++++++++-----
 3 files changed, 48 insertions(+), 12 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c +6 -0
index 38e310a8694d..951d5da850be 100644
@@ -1172,6 +1172,7 @@ int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr *userq_mgr, struct drm_file *f
 	xa_init_flags(&userq_mgr->userq_xa, XA_FLAGS_ALLOC);
 	userq_mgr->adev = adev;
 	userq_mgr->file = file_priv;
+	mutex_init(&userq_mgr->proc_ctx_lock);
 
 	INIT_DELAYED_WORK(&userq_mgr->resume_work, amdgpu_userq_restore_worker);
 	INIT_WORK(&userq_mgr->reset_work, amdgpu_userq_mgr_reset_work);
@@ -1225,6 +1226,11 @@ void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr)
 	 */
 	cancel_work_sync(&userq_mgr->reset_work);
 
+	amdgpu_bo_free_kernel(&userq_mgr->proc_ctx_obj.obj,
+			      &userq_mgr->proc_ctx_obj.gpu_addr,
+			      &userq_mgr->proc_ctx_obj.cpu_ptr);
+
+	mutex_destroy(&userq_mgr->proc_ctx_lock);
 	mutex_destroy(&userq_mgr->userq_mutex);
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h +2 -0
index 28cfc6682333..a5867ffe6988 100644
@@ -127,6 +127,8 @@ struct amdgpu_userq_mgr {
 	struct amdgpu_device		*adev;
 	struct delayed_work		resume_work;
 	struct drm_file			*file;
+	struct mutex			proc_ctx_lock;
+	struct amdgpu_userq_obj		proc_ctx_obj;
 
 	/**
 	 * @reset_work:
drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +40 -12

94 lines not on this page · raw message

Signature · 1 line
2.54.0