This is a note to let you know that I've just added the patch titled
drm/ttm: stop pooling cached NUMA pages v2
to the 6.8-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
drm-ttm-stop-pooling-cached-numa-pages-v2.patch
and it can be found in the queue-6.8 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
>From b6976f323a8687cc0d55bc92c2086fd934324ed5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <ckoenig.leichtzumerken@xxxxxxxxx>
Date: Mon, 15 Apr 2024 15:48:21 +0200
Subject: drm/ttm: stop pooling cached NUMA pages v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
From: Christian König <ckoenig.leichtzumerken@xxxxxxxxx>
commit b6976f323a8687cc0d55bc92c2086fd934324ed5 upstream.
We only pool write combined and uncached allocations because they
require extra overhead on allocation and release.
If we also pool cached NUMA it not only means some extra unnecessary
overhead, but also that under memory pressure it can happen that
pages from the wrong NUMA node enters the pool and are re-used
over and over again.
This can lead to performance reduction after running into memory
pressure.
v2: restructure and cleanup the code a bit from the internal hack to
test this.
Signed-off-by: Christian König <christian.koenig@xxxxxxx>
Fixes: 4482d3c94d7f ("drm/ttm: add NUMA node id to the pool")
CC: stable@xxxxxxxxxxxxxxx
Reviewed-by: Felix Kuehling <felix.kuehling@xxxxxxx>
Link: https://patchwork.freedesktop.org/patch/msgid/20240415134821.1919-1-christian.koenig@xxxxxxx
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/ttm/ttm_pool.c | 38 ++++++++++++++++++++++++++++----------
1 file changed, 28 insertions(+), 10 deletions(-)
--- a/drivers/gpu/drm/ttm/ttm_pool.c
+++ b/drivers/gpu/drm/ttm/ttm_pool.c
@@ -288,17 +288,23 @@ static struct ttm_pool_type *ttm_pool_se
enum ttm_caching caching,
unsigned int order)
{
- if (pool->use_dma_alloc || pool->nid != NUMA_NO_NODE)
+ if (pool->use_dma_alloc)
return &pool->caching[caching].orders[order];
#ifdef CONFIG_X86
switch (caching) {
case ttm_write_combined:
+ if (pool->nid != NUMA_NO_NODE)
+ return &pool->caching[caching].orders[order];
+
if (pool->use_dma32)
return &global_dma32_write_combined[order];
return &global_write_combined[order];
case ttm_uncached:
+ if (pool->nid != NUMA_NO_NODE)
+ return &pool->caching[caching].orders[order];
+
if (pool->use_dma32)
return &global_dma32_uncached[order];
@@ -566,11 +572,17 @@ void ttm_pool_init(struct ttm_pool *pool
pool->use_dma_alloc = use_dma_alloc;
pool->use_dma32 = use_dma32;
- if (use_dma_alloc || nid != NUMA_NO_NODE) {
- for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i)
- for (j = 0; j < NR_PAGE_ORDERS; ++j)
- ttm_pool_type_init(&pool->caching[i].orders[j],
- pool, i, j);
+ for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) {
+ for (j = 0; j < NR_PAGE_ORDERS; ++j) {
+ struct ttm_pool_type *pt;
+
+ /* Initialize only pool types which are actually used */
+ pt = ttm_pool_select_type(pool, i, j);
+ if (pt != &pool->caching[i].orders[j])
+ continue;
+
+ ttm_pool_type_init(pt, pool, i, j);
+ }
}
}
EXPORT_SYMBOL(ttm_pool_init);
@@ -599,10 +611,16 @@ void ttm_pool_fini(struct ttm_pool *pool
{
unsigned int i, j;
- if (pool->use_dma_alloc || pool->nid != NUMA_NO_NODE) {
- for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i)
- for (j = 0; j < NR_PAGE_ORDERS; ++j)
- ttm_pool_type_fini(&pool->caching[i].orders[j]);
+ for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) {
+ for (j = 0; j < NR_PAGE_ORDERS; ++j) {
+ struct ttm_pool_type *pt;
+
+ pt = ttm_pool_select_type(pool, i, j);
+ if (pt != &pool->caching[i].orders[j])
+ continue;
+
+ ttm_pool_type_fini(pt);
+ }
}
/* We removed the pool types from the LRU, but we need to also make sure
Patches currently in stable-queue which might be from ckoenig.leichtzumerken@xxxxxxxxx are
queue-6.8/drm-ttm-stop-pooling-cached-numa-pages-v2.patch
