From: Felix Kuehling <felix.kuehling@amd.com>
To: Philip Yang <Philip.Yang@amd.com>, amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH 2/4] drm/amdkfd: handle partial migration cpages 0
Date: Wed, 6 Oct 2021 17:36:51 -0400 [thread overview]
Message-ID: <b080e4ac-7b95-bb16-6989-85085b05335e@amd.com> (raw)
In-Reply-To: <20211006143210.4954-2-Philip.Yang@amd.com>
Am 2021-10-06 um 10:32 a.m. schrieb Philip Yang:
> migrate_vma_setup may return cpages 0, means 0 page can be migrated,
> treat this as error case to skip the rest of migration steps, and don't
> change prange actual loc, to avoid warning message "VRAM BO missing
> during validation".
>
> Signed-off-by: Philip Yang <Philip.Yang@amd.com>
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 48 ++++++++++++++----------
> 1 file changed, 29 insertions(+), 19 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> index 069422337cf7..9b68e3e8f2a1 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> @@ -409,20 +409,25 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
> r, prange->svms, prange->start, prange->last);
> goto out_free;
> }
> - if (migrate.cpages != npages) {
> - pr_debug("Partial migration. 0x%lx/0x%llx pages can be migrated\n",
> - migrate.cpages,
> - npages);
> - }
>
> - if (migrate.cpages) {
> - r = svm_migrate_copy_to_vram(adev, prange, &migrate, &mfence,
> - scratch);
> - migrate_vma_pages(&migrate);
> - svm_migrate_copy_done(adev, mfence);
> - migrate_vma_finalize(&migrate);
> + if (migrate.cpages != npages)
> + pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n",
> + migrate.cpages, npages);
> + else
> + pr_debug("0x%lx pages migrated\n", migrate.cpages);
> +
> + if (!migrate.cpages) {
> + pr_debug("failed collect migrate sys pages [0x%lx 0x%lx]\n",
> + prange->start, prange->last);
> + r = -ENOMEM;
I think just returning an error here is incorrect. This error gets
handled in svm_migrate_ram_to_vram and prevents the following VMAs from
migrating as well (if the range spans multiple VMAs).
Maybe return the number of pages migrated, if successful. Then the
caller can add up all the successful migrations and update
prange->actual_loc only if the total is > 0.
Regards,
Felix
> + goto out_free;
> }
>
> + r = svm_migrate_copy_to_vram(adev, prange, &migrate, &mfence, scratch);
> + migrate_vma_pages(&migrate);
> + svm_migrate_copy_done(adev, mfence);
> + migrate_vma_finalize(&migrate);
> +
> svm_range_dma_unmap(adev->dev, scratch, 0, npages);
> svm_range_free_dma_mappings(prange);
>
> @@ -636,19 +641,24 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
> goto out_free;
> }
>
> - pr_debug("cpages %ld\n", migrate.cpages);
> + if (migrate.cpages != npages)
> + pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n",
> + migrate.cpages, npages);
> + else
> + pr_debug("0x%lx pages migrated\n", migrate.cpages);
>
> - if (migrate.cpages) {
> - r = svm_migrate_copy_to_ram(adev, prange, &migrate, &mfence,
> - scratch, npages);
> - migrate_vma_pages(&migrate);
> - svm_migrate_copy_done(adev, mfence);
> - migrate_vma_finalize(&migrate);
> - } else {
> + if (!migrate.cpages) {
> pr_debug("failed collect migrate device pages [0x%lx 0x%lx]\n",
> prange->start, prange->last);
> + r = -ENOMEM;
> + goto out_free;
> }
>
> + r = svm_migrate_copy_to_ram(adev, prange, &migrate, &mfence,
> + scratch, npages);
> + migrate_vma_pages(&migrate);
> + svm_migrate_copy_done(adev, mfence);
> + migrate_vma_finalize(&migrate);
> svm_range_dma_unmap(adev->dev, scratch, 0, npages);
>
> out_free:
next prev parent reply other threads:[~2021-10-06 21:36 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-06 14:32 [PATCH 1/4] drm/amdkfd: ratelimited svm debug messages Philip Yang
2021-10-06 14:32 ` [PATCH 2/4] drm/amdkfd: handle partial migration cpages 0 Philip Yang
2021-10-06 21:36 ` Felix Kuehling [this message]
2021-10-06 14:32 ` [PATCH 3/4] drm/amdkfd: avoid svm conflicting with userptr address Philip Yang
2021-10-06 17:22 ` Felix Kuehling
2021-10-06 18:09 ` philip yang
2021-10-06 18:23 ` Felix Kuehling
2021-10-06 19:57 ` philip yang
2021-10-06 14:32 ` [PATCH 4/4] drm/amdkfd: create svm unregister range not overlap with userptr Philip Yang
2021-10-06 20:55 ` [PATCH 1/4] drm/amdkfd: ratelimited svm debug messages Felix Kuehling
2021-10-08 19:28 ` philip yang
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=b080e4ac-7b95-bb16-6989-85085b05335e@amd.com \
--to=felix.kuehling@amd.com \
--cc=Philip.Yang@amd.com \
--cc=amd-gfx@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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.