nv-tegra-mirror/linux-nv-oot
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nv-oot.git synced 2025-12-22 17:25:35 +03:00
Code Issues Packages Projects Releases Wiki Activity

video: tegra: nvmap: Gather pages and then flush cache

Browse Source 
After a deeper analysis, it was found that multithreaded cache flush is
not required. Only gathering the pages and then doing cache flush is
enough to get the perf improvement, the cache flush time of each page is
almost half in this case as compared with not gathering the pages. So it
means there is continuous cache eviction is happening. So, while
allocating big pages from allocator, instead of allocating one big page
and flushing it, allocate all required big pages and then flush pages
at last.

Bug 4628529

Change-Id: I4d8113010a8c82a1cab124e26b14e8a3f5791cd1
Signed-off-by: Ketan Patil <ketanp@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3325002
Reviewed-by: Pritesh Raithatha <praithatha@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Ajay Nandakumar Mannargudi <anandakumarm@nvidia.com>
This commit is contained in:
Ketan Patil
2025-03-24 13:59:32 +00:00
committed by Jon Hunter
parent 5bb431b8df
commit 838bb12196
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
drivers/video/tegra/nvmap/nvmap_alloc.c
View File

@@ -105,6 +105,7 @@ static int handle_page_alloc(struct nvmap_client *client,
u64 result;
size_t tot_size;
#ifdef CONFIG_ARM64_4K_PAGES
int start_index = 0;
#ifdef NVMAP_CONFIG_PAGE_POOLS
int pages_per_big_pg = NVMAP_PP_BIG_PAGE_SIZE >> PAGE_SHIFT;
#else
@@ -137,6 +138,7 @@ static int handle_page_alloc(struct nvmap_client *client,
pages_per_big_pg = nvmap_dev->pool->pages_per_big_pg;
#endif
/* Try to allocate big pages from page allocator */
start_index = page_index;
for (i = page_index;
i < nr_page && pages_per_big_pg > 1 && (nr_page - i) >= pages_per_big_pg;
i += pages_per_big_pg, page_index += pages_per_big_pg) {
@@ -155,9 +157,10 @@ static int handle_page_alloc(struct nvmap_client *client,
for (idx = 0; idx < pages_per_big_pg; idx++)
pages[i + idx] = nth_page(page, idx);
nvmap_clean_cache(&pages[i], pages_per_big_pg);
}
/* Perform cache clean for the pages allocated from page allocator */
nvmap_clean_cache(&pages[start_index], page_index - start_index);
if (check_add_overflow(nvmap_big_page_allocs, (u64)page_index, &result))
goto fail;