The idea behind this patch is to delay the freeing of PT entry objects
until the TLB flush is done.
This patch:
- Adds a tlb_flush_waitlist in amdgpu_vm_update_params which will keep the
objects that need to be freed after tlb_flush.
- Adds PT entries in this list in amdgpu_vm_ptes_update after finding
the PT entry.
- Changes functionality of amdgpu_vm_pt_free_dfs from (df_search + free)
to simply freeing of the BOs, also renames it to
amdgpu_vm_pt_free_list to reflect this same.
- Exports function amdgpu_vm_pt_free_list to be called directly.
- Calls amdgpu_vm_pt_free_list directly from amdgpu_vm_update_range.
V2: rebase
V4: Addressed review comments from Christian
- add only locked PTEs entries in TLB flush waitlist.
- do not create a separate function for list flush.
- do not create a new lock for TLB flush.
- there is no need to wait on tlb_flush_fence exclusively.
V5: Addressed review comments from Christian
- change the amdgpu_vm_pt_free_dfs's functionality to simple freeing
of the objects and rename it.
- add all the PTE objects in params->tlb_flush_waitlist
- let amdgpu_vm_pt_free_root handle the freeing of BOs independently
- call amdgpu_vm_pt_free_list directly
V6: Rebase
V7: Rebase
Cc: Christian König <Christian.Koenig@xxxxxxx>
Cc: Alex Deucher <alexander.deucher@xxxxxxx>
Cc: Felix Kuehling <felix.kuehling@xxxxxxx>
Cc: Rajneesh Bhardwaj <rajneesh.bhardwaj@xxxxxxx>
Acked-by: Felix Kuehling <felix.kuehling@xxxxxxx>
Acked-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@xxxxxxx>
Tested-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@xxxxxxx>
Signed-off-by: Shashank Sharma <shashank.sharma@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 7 +++
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 53 +++++++++++++----------
3 files changed, 40 insertions(+), 25 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 26f1c3359642..eaa402f99fe0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -977,6 +977,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
params.unlocked = unlocked;
params.needs_flush = flush_tlb;
params.allow_override = allow_override;
+ INIT_LIST_HEAD(¶ms.tlb_flush_waitlist);
/* Implicitly sync to command submissions in the same VM before
* unmapping. Sync to moving fences before mapping.
@@ -1062,8 +1063,10 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
if (r)
goto error_unlock;
- if (params.needs_flush)
+ if (params.needs_flush) {
r = amdgpu_vm_tlb_flush(¶ms, fence);
+ amdgpu_vm_pt_free_list(adev, ¶ms);
+ }
error_unlock:
amdgpu_vm_eviction_unlock(vm);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index b0a4fe683352..54d7da396de0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -266,6 +266,11 @@ struct amdgpu_vm_update_params {
* to be overridden for NUMA local memory.
*/
bool allow_override;
+
+ /**
+ * @tlb_flush_waitlist: temporary storage for BOs until tlb_flush
+ */
+ struct list_head tlb_flush_waitlist;
};
struct amdgpu_vm_update_funcs {
@@ -547,6 +552,8 @@ int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params,
uint64_t start, uint64_t end,
uint64_t dst, uint64_t flags);
void amdgpu_vm_pt_free_work(struct work_struct *work);
+void amdgpu_vm_pt_free_list(struct amdgpu_device *adev,
+ struct amdgpu_vm_update_params *params);
#if defined(CONFIG_DEBUG_FS)
void amdgpu_debugfs_vm_bo_info(struct amdgpu_vm *vm, struct seq_file *m);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
index 601df0ce8290..440dc8c581fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
@@ -622,40 +622,30 @@ void amdgpu_vm_pt_free_work(struct work_struct *work)
}
/**
- * amdgpu_vm_pt_free_dfs - free PD/PT levels
+ * amdgpu_vm_pt_free_list - free PD/PT levels
*
* @adev: amdgpu device structure
- * @vm: amdgpu vm structure
- * @start: optional cursor where to start freeing PDs/PTs
- * @unlocked: vm resv unlock status
+ * @params: see amdgpu_vm_update_params definition
*
- * Free the page directory or page table level and all sub levels.
+ * Free the page directory objects saved in the flush list
*/
-static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- struct amdgpu_vm_pt_cursor *start,
- bool unlocked)
+void amdgpu_vm_pt_free_list(struct amdgpu_device *adev,
+ struct amdgpu_vm_update_params *params)
{
- struct amdgpu_vm_pt_cursor cursor;
- struct amdgpu_vm_bo_base *entry;
+ struct amdgpu_vm_bo_base *entry, *next;
+ struct amdgpu_vm *vm = params->vm;
+ bool unlocked = params->unlocked;
if (unlocked) {
spin_lock(&vm->status_lock);
- for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry)
- list_move(&entry->vm_status, &vm->pt_freed);
-
- if (start)
- list_move(&start->entry->vm_status, &vm->pt_freed);
+ list_splice_init(&vm->pt_freed, ¶ms->tlb_flush_waitlist);
spin_unlock(&vm->status_lock);
schedule_work(&vm->pt_free_work);
return;
}
- for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry)
+ list_for_each_entry_safe(entry, next, ¶ms->tlb_flush_waitlist, vm_status)
amdgpu_vm_pt_free(entry);
-
- if (start)
- amdgpu_vm_pt_free(start->entry);
}
/**
@@ -667,7 +657,11 @@ static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev,
*/
void amdgpu_vm_pt_free_root(struct amdgpu_device *adev, struct amdgpu_vm *vm)
{
- amdgpu_vm_pt_free_dfs(adev, vm, NULL, false);
+ struct amdgpu_vm_pt_cursor cursor;
+ struct amdgpu_vm_bo_base *entry;
+
+ for_each_amdgpu_vm_pt_dfs_safe(adev, vm, NULL, cursor, entry)
+ amdgpu_vm_pt_free(entry);
}
/**
@@ -972,10 +966,21 @@ int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params,
while (cursor.pfn < frag_start) {
/* Make sure previous mapping is freed */
if (cursor.entry->bo) {
+ struct amdgpu_vm_pt_cursor seek;
+ struct amdgpu_vm_bo_base *entry;
+
params->needs_flush = true;
- amdgpu_vm_pt_free_dfs(adev, params->vm,
- &cursor,
- params->unlocked);
+ spin_lock(¶ms->vm->status_lock);
+ for_each_amdgpu_vm_pt_dfs_safe(adev, params->vm, &cursor,
+ seek, entry) {
+ list_move(&entry->vm_status,
+ ¶ms->tlb_flush_waitlist);
+ }
+
+ /* enter start node now */
+ list_move(&cursor.entry->vm_status,
+ ¶ms->tlb_flush_waitlist);
+ spin_unlock(¶ms->vm->status_lock);
}
amdgpu_vm_pt_next(adev, &cursor);
}
--
2.43.2
