Skip to contentSkip to conversation

drm/i915: clear the shadow batch

PATCH Superseded Matthew Auld 3 participants 5 messages Dec 24, 2020

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/i915: clear the shadow batch

The shadow batch is an internal object, which doesn't have any page clearing, and since the batch_len can be smaller than the object, we should take care to clear it.

Testcase: igt/gen9_exec_parse/shadow-peek

Fixes 4f7af1948abc ("drm/i915: Support ro ppgtt mapped cmdparser shadow buffers")
---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/i915_cmd_parser.c +7 -0
index 8d88402387bd..ff3a0b8ccdd5 100644
@@ -1147,6 +1147,13 @@ static u32 *copy_batch(struct drm_i915_gem_object *dst_obj,
 	if (IS_ERR(dst))
 		return dst;
 
+	if (length < dst_obj->base.size) {
+		memset32(dst + length, 0,
+			 (dst_obj->base.size - length) / sizeof(u32));
+		__i915_gem_object_flush_map(dst_obj, length,
+					    dst_obj->base.size - length);
+	}
+
 	ret = i915_gem_object_pin_pages(src_obj);
 	if (ret) {
 		i915_gem_object_unpin_map(dst_obj);
Signature · 6 lines
2.26.2

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

PATCH 2/2 drm/i915: clear the gpu reloc batch

The reloc batch is short lived but can exist in the user visible ppGTT, and since it's backed by an internal object, which lacks page clearing, we should take care to clear it upfront.

---
 drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +3 -0
index 0cf9e79325a8..a4ecd4b4e874 100644
@@ -1295,6 +1295,9 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 		goto err_pool;
 	}
 
+	memset64((void*)cmd, 0, pool->obj->base.size / sizeof(u64));
+	i915_gem_object_flush_map(pool->obj);
+
 	batch = i915_vma_instance(pool->obj, vma->vm, NULL);
 	if (IS_ERR(batch)) {
 		err = PTR_ERR(batch);
Signature · 6 lines
2.26.2

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [Intel-gfx] [PATCH 2/2] drm/i915: clear the gpu reloc batch

Quoting Matthew Auld (2020-12-24 14:34:55)

Quoted 14 lines

The reloc batch is short lived but can exist in the user visible ppGTT, and since it's backed by an internal object, which lacks page clearing, we should take care to clear it upfront.

---
 drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +0 -0
index 0cf9e79325a8..a4ecd4b4e874 100644

drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c static int __reloc_gpu_alloc(struct i915_execbuffer *eb, · line 1298

Quoted 4 lines
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +1 -0
@@ -1295,6 +1295,9 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
                goto err_pool;
        }
 
+       memset64((void*)cmd, 0, pool->obj->base.size / sizeof(u64));

Might as well make it memset32 to avoid the cast. I'm willing to bet it's in the noise by that point on 4096+ object.

drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c static int __reloc_gpu_alloc(struct i915_execbuffer *eb, · line 1299

Quoted 1 line
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +1 -0
+       i915_gem_object_flush_map(pool->obj);

We really don't want to have to flush it twice. Remove the limit on the flush_map before submission.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw
In reply to Matthew Auld

Quoting Matthew Auld (2020-12-24 14:34:54)

Quoted 16 lines

The shadow batch is an internal object, which doesn't have any page clearing, and since the batch_len can be smaller than the object, we should take care to clear it.

Testcase: igt/gen9_exec_parse/shadow-peek

Fixes 4f7af1948abc ("drm/i915: Support ro ppgtt mapped cmdparser shadow buffers")
---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/i915_cmd_parser.c +0 -0
index 8d88402387bd..ff3a0b8ccdd5 100644

drivers/gpu/drm/i915/i915_cmd_parser.c static u32 *copy_batch(struct drm_i915_gem_object *dst_obj, · line 1155

Quoted 9 lines
drivers/gpu/drm/i915/i915_cmd_parser.c +6 -0
@@ -1147,6 +1147,13 @@ static u32 *copy_batch(struct drm_i915_gem_object *dst_obj,
        if (IS_ERR(dst))
                return dst;
 
+       if (length < dst_obj->base.size) {
+               memset32(dst + length, 0,
+                        (dst_obj->base.size - length) / sizeof(u32));
+               __i915_gem_object_flush_map(dst_obj, length,
+                                           dst_obj->base.size - length);
+       }

I feel like we might as well remove the shadow_needs_clflush() and just do a i915_gem_object_flush_map(shadow->obj) after parsing.

Then for simple chronological ordering, we can place this at the end of copy_batch

memset32(dst + length, 0, (dst_obj->base.size - length) / sizeof(u32));

The cost of one more function call along here? Unnoticable. -Chris
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

[Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [1/2] drm/i915: clear the shadow batch

In reply to Matthew Auld
== Series Details ==

== Series Details ==

Series: series starting with [1/2] drm/i915: clear the shadow batch
URL : https://patchwork.freedesktop.org/series/85204/
State : failure

== Summary ==

Applying: drm/i915: clear the shadow batch
Applying: drm/i915: clear the gpu reloc batch
Using index info to reconstruct a base tree...
M drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
Falling back to patching base and 3-way merge...
Auto-merging drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch=diff' to see the failed patch Patch failed at 0002 drm/i915: clear the gpu reloc batch
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx