From: Felix Kuehling <felix.kuehling@gmail.com>
To: amd-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Cc: Alex Sierra <alex.sierra@amd.com>
Subject: [PATCH 06/10] drm/amdkfd: skip invalid pages during migrations
Date: Thu, 27 May 2021 16:56:02 -0400 [thread overview]
Message-ID: <20210527205606.2660-6-Felix.Kuehling@amd.com> (raw)
In-Reply-To: <20210527205606.2660-1-Felix.Kuehling@amd.com>
From: Alex Sierra <alex.sierra@amd.com>
Invalid pages can be the result of pages that have been migrated
already due to copy-on-write procedure or pages that were never
migrated to VRAM in first place. This is not an issue anymore,
as pranges now support mixed memory domains (CPU/GPU).
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
---
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 38 +++++++++++-------------
1 file changed, 18 insertions(+), 20 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index b298aa8dea4d..6fd68528c425 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -419,7 +419,6 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
size_t size;
void *buf;
int r = -ENOMEM;
- int retry = 0;
memset(&migrate, 0, sizeof(migrate));
migrate.vma = vma;
@@ -438,7 +437,6 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
migrate.dst = migrate.src + npages;
scratch = (dma_addr_t *)(migrate.dst + npages);
-retry:
r = migrate_vma_setup(&migrate);
if (r) {
pr_debug("failed %d prepare migrate svms 0x%p [0x%lx 0x%lx]\n",
@@ -446,17 +444,9 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
goto out_free;
}
if (migrate.cpages != npages) {
- pr_debug("collect 0x%lx/0x%llx pages, retry\n", migrate.cpages,
+ pr_debug("Partial migration. 0x%lx/0x%llx pages can be migrated\n",
+ migrate.cpages,
npages);
- migrate_vma_finalize(&migrate);
- if (retry++ >= 3) {
- r = -ENOMEM;
- pr_debug("failed %d migrate svms 0x%p [0x%lx 0x%lx]\n",
- r, prange->svms, prange->start, prange->last);
- goto out_free;
- }
-
- goto retry;
}
if (migrate.cpages) {
@@ -547,9 +537,8 @@ static void svm_migrate_page_free(struct page *page)
static int
svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
struct migrate_vma *migrate, struct dma_fence **mfence,
- dma_addr_t *scratch)
+ dma_addr_t *scratch, uint64_t npages)
{
- uint64_t npages = migrate->cpages;
struct device *dev = adev->dev;
uint64_t *src;
dma_addr_t *dst;
@@ -566,15 +555,23 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
src = (uint64_t *)(scratch + npages);
dst = scratch;
- for (i = 0, j = 0; i < npages; i++, j++, addr += PAGE_SIZE) {
+ for (i = 0, j = 0; i < npages; i++, addr += PAGE_SIZE) {
struct page *spage;
spage = migrate_pfn_to_page(migrate->src[i]);
- if (!spage) {
- pr_debug("failed get spage svms 0x%p [0x%lx 0x%lx]\n",
+ if (!spage || !is_zone_device_page(spage)) {
+ pr_debug("invalid page. Could be in CPU already svms 0x%p [0x%lx 0x%lx]\n",
prange->svms, prange->start, prange->last);
- r = -ENOMEM;
- goto out_oom;
+ if (j) {
+ r = svm_migrate_copy_memory_gart(adev, dst + i - j,
+ src + i - j, j,
+ FROM_VRAM_TO_RAM,
+ mfence);
+ if (r)
+ goto out_oom;
+ j = 0;
+ }
+ continue;
}
src[i] = svm_migrate_addr(adev, spage);
if (i > 0 && src[i] != src[i - 1] + PAGE_SIZE) {
@@ -607,6 +604,7 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
migrate->dst[i] = migrate_pfn(page_to_pfn(dpage));
migrate->dst[i] |= MIGRATE_PFN_LOCKED;
+ j++;
}
r = svm_migrate_copy_memory_gart(adev, dst + i - j, src + i - j, j,
@@ -664,7 +662,7 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
if (migrate.cpages) {
r = svm_migrate_copy_to_ram(adev, prange, &migrate, &mfence,
- scratch);
+ scratch, npages);
migrate_vma_pages(&migrate);
svm_migrate_copy_done(adev, mfence);
migrate_vma_finalize(&migrate);
--
2.31.1
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
next prev parent reply other threads:[~2021-05-27 20:56 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-27 20:55 [PATCH 01/10] drm/amdkfd: device pgmap owner at the svm migrate init Felix Kuehling
2021-05-27 20:55 ` [PATCH 02/10] drm/amdkfd: add owner ref param to get hmm pages Felix Kuehling
2021-05-27 20:55 ` [PATCH 03/10] drm/amdkfd: set owner ref to svm range prefault Felix Kuehling
2021-05-27 20:56 ` [PATCH 04/10] drm/amdgpu: get owner ref in validate and map Felix Kuehling
2021-05-27 20:56 ` [PATCH 05/10] drm/amdkfd: classify and map mixed svm range pages in GPU Felix Kuehling
2021-05-27 20:56 ` Felix Kuehling [this message]
2021-05-27 20:56 ` [PATCH 07/10] drm/amdkfd: skip migration for pages already in VRAM Felix Kuehling
2021-05-27 20:56 ` [PATCH 08/10] drm/amdkfd: add invalid pages debug at vram migration Felix Kuehling
2021-05-27 20:56 ` [PATCH 09/10] drm/amdkfd: partially actual_loc removed Felix Kuehling
2021-05-27 20:56 ` [PATCH 10/10] drm/amdkfd: protect svm_bo ref in case prange has forked Felix Kuehling
-- strict thread matches above, loose matches on Subject: below --
2021-06-21 16:04 [PATCH 01/10] drm/amdkfd: device pgmap owner at the svm migrate init Alex Sierra
2021-06-21 16:04 ` [PATCH 06/10] drm/amdkfd: skip invalid pages during migrations Alex Sierra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210527205606.2660-6-Felix.Kuehling@amd.com \
--to=felix.kuehling@gmail.com \
--cc=alex.sierra@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=dri-devel@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox