Skip to contentSkip to conversation

FAILED: patch "[PATCH] drm/amdgpu: revert "use more than 64KB fragment size if" failed to apply to 4.8-stable tree

PATCH [email protected] 1 participant 1 message Oct 28, 2016

Subsystems (3)
Stable branch DRM drivers Radeon and amdgpu DRM drivers
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 4.8-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

Quoted 1 line

From 8036617e92e3fad49eef9bbe868b661c58249aff Mon Sep 17 00:00:00 2001

From: =?UTF-8?q?Christian=20K=C3=B6nig?= <[email protected]>
Date: Tue, 4 Oct 2016 13:39:43 +0200
Subject: [PATCH] drm/amdgpu: revert "use more than 64KB fragment size if
possible"
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This reverts commit 1dcd32fb9c54334ec948a0f18174a748d6b14364.

The block size is indeed an equal match, so this can cause performance regressions.

Reviewed-by Alex Deucher <[email protected]>
Signed-off-by Christian König <[email protected]>
Signed-off-by Alex Deucher <[email protected]>
Cc [email protected]
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -8
index bc4b22c6fc08..06f24322e7c3 100644
@@ -878,13 +878,13 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params	*params,
 	 * allocation size to the fragment size.
 	 */
 
-	const uint64_t frag_align = 1 << AMDGPU_LOG2_PAGES_PER_FRAG;
+	/* SI and newer are optimized for 64KB */
+	uint64_t frag_flags = AMDGPU_PTE_FRAG(AMDGPU_LOG2_PAGES_PER_FRAG);
+	uint64_t frag_align = 1 << AMDGPU_LOG2_PAGES_PER_FRAG;
 
 	uint64_t frag_start = ALIGN(start, frag_align);
 	uint64_t frag_end = end & ~(frag_align - 1);
 
-	uint32_t frag;
-
 	/* system pages are non continuously */
 	if (params->src || !(flags & AMDGPU_PTE_VALID) ||
 	    (frag_start >= frag_end)) {

@@ -893,10 +893,6 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params	*params,
 		return;
 	}
 
-	/* use more than 64KB fragment size if possible */
-	frag = lower_32_bits(frag_start | frag_end);
-	frag = likely(frag) ? __ffs(frag) : 31;
-
 	/* handle the 4K area at the beginning */
 	if (start != frag_start) {
 		amdgpu_vm_update_ptes(params, vm, start, frag_start,

@@ -906,7 +902,7 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params	*params,
 
 	/* handle the area in the middle */
 	amdgpu_vm_update_ptes(params, vm, frag_start, frag_end, dst,
-			      flags | AMDGPU_PTE_FRAG(frag));
+			      flags | frag_flags);
 
 	/* handle the 4K area at the end */
 	if (frag_end != end) {