From: "Christian König" <christian.koenig@amd.com>
To: Samuel Zhang <guoqing.zhang@amd.com>,
rafael@kernel.org, len.brown@intel.com, pavel@kernel.org,
alexander.deucher@amd.com, mario.limonciello@amd.com,
lijo.lazar@amd.com
Cc: victor.zhao@amd.com, haijun.chang@amd.com, Qing.Ma@amd.com,
amd-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org,
linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/3] drm/amdgpu: move GTT to SHM after eviction for hibernation
Date: Mon, 30 Jun 2025 13:54:18 +0200 [thread overview]
Message-ID: <ce04e266-6c3f-4256-aade-bafca8609ab3@amd.com> (raw)
In-Reply-To: <20250630104116.3050306-2-guoqing.zhang@amd.com>
On 30.06.25 12:41, Samuel Zhang wrote:
> When hibernate with data center dGPUs, huge number of VRAM BOs evicted
> to GTT and takes too much system memory. This will cause hibernation
> fail due to insufficient memory for creating the hibernation image.
>
> Move GTT BOs to shmem in KMD, then shmem to swap disk in kernel
> hibernation code to make room for hibernation image.
This should probably be two patches, one for TTM and then an amdgpu patch to forward the event.
>
> Signed-off-by: Samuel Zhang <guoqing.zhang@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 13 ++++++++++++-
> drivers/gpu/drm/ttm/ttm_resource.c | 18 ++++++++++++++++++
> include/drm/ttm/ttm_resource.h | 1 +
> 3 files changed, 31 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 4d57269c9ca8..5aede907a591 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -2889,6 +2889,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
> int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
> {
> struct ttm_resource_manager *man;
> + int r;
>
> switch (mem_type) {
> case TTM_PL_VRAM:
> @@ -2903,7 +2904,17 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
> return -EINVAL;
> }
>
> - return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
> + r = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
> + if (r) {
> + DRM_ERROR("Failed to evict memory type %d\n", mem_type);
> + return r;
> + }
> + if (adev->in_s4 && mem_type == TTM_PL_VRAM) {
> + r = ttm_resource_manager_swapout();
> + if (r)
> + DRM_ERROR("Failed to swap out, %d\n", r);
> + }
> + return r;
> }
>
> #if defined(CONFIG_DEBUG_FS)
> diff --git a/drivers/gpu/drm/ttm/ttm_resource.c b/drivers/gpu/drm/ttm/ttm_resource.c
> index fd41b56e2c66..07b1f5a5afc2 100644
> --- a/drivers/gpu/drm/ttm/ttm_resource.c
> +++ b/drivers/gpu/drm/ttm/ttm_resource.c
> @@ -534,6 +534,24 @@ void ttm_resource_manager_init(struct ttm_resource_manager *man,
> }
> EXPORT_SYMBOL(ttm_resource_manager_init);
>
> +int ttm_resource_manager_swapout(void)
This needs documentation, better placement and a better name.
First of all put it into ttm_device.c instead of the resource manager.
Then call it something like ttm_device_prepare_hibernation or similar.
> +{
> + struct ttm_operation_ctx ctx = {
> + .interruptible = false,
> + .no_wait_gpu = false,
> + .force_alloc = true
> + };
> + int ret;
> +
> + while (true) {
Make that:
do {
ret = ...
} while (ret > 0);
> + ret = ttm_global_swapout(&ctx, GFP_KERNEL);
> + if (ret <= 0)
> + break;
> + }
> + return ret;
It's rather pointless to return the number of swapped out pages.
Make that "return ret < 0 ? ret : 0;
Regards,
Christian.
> +}
> +EXPORT_SYMBOL(ttm_resource_manager_swapout);
> +
> /*
> * ttm_resource_manager_evict_all
> *
> diff --git a/include/drm/ttm/ttm_resource.h b/include/drm/ttm/ttm_resource.h
> index b873be9597e2..46181758068e 100644
> --- a/include/drm/ttm/ttm_resource.h
> +++ b/include/drm/ttm/ttm_resource.h
> @@ -463,6 +463,7 @@ void ttm_resource_manager_init(struct ttm_resource_manager *man,
>
> int ttm_resource_manager_evict_all(struct ttm_device *bdev,
> struct ttm_resource_manager *man);
> +int ttm_resource_manager_swapout(void);
>
> uint64_t ttm_resource_manager_usage(struct ttm_resource_manager *man);
> void ttm_resource_manager_debug(struct ttm_resource_manager *man,
next prev parent reply other threads:[~2025-06-30 11:54 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-06-30 10:41 [PATCH 0/3] reduce system memory requirement for hibernation Samuel Zhang
2025-06-30 10:41 ` [PATCH 1/3] drm/amdgpu: move GTT to SHM after eviction " Samuel Zhang
2025-06-30 11:54 ` Christian König [this message]
2025-07-01 8:18 ` Zhang, GuoQing (Sam)
2025-07-01 8:22 ` Christian König
2025-07-02 7:28 ` Samuel Zhang
2025-07-02 7:48 ` Christian König
2025-06-30 10:41 ` [PATCH 2/3] PM: hibernate: shrink shmem pages after dev_pm_ops.prepare() Samuel Zhang
2025-06-30 20:21 ` Rafael J. Wysocki
2025-06-30 10:41 ` [PATCH 3/3] drm/amdgpu: skip kfd resume_process for dev_pm_ops.thaw() Samuel Zhang
2025-06-30 11:58 ` Christian König
2025-07-01 8:03 ` Zhang, GuoQing (Sam)
2025-07-01 8:32 ` Christian König
2025-07-01 16:07 ` Alex Deucher
2025-07-02 7:23 ` Sam
2025-07-02 13:54 ` Alex Deucher
2025-07-02 14:07 ` Lazar, Lijo
2025-07-04 10:24 ` Zhang, GuoQing (Sam)
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=ce04e266-6c3f-4256-aade-bafca8609ab3@amd.com \
--to=christian.koenig@amd.com \
--cc=Qing.Ma@amd.com \
--cc=alexander.deucher@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=guoqing.zhang@amd.com \
--cc=haijun.chang@amd.com \
--cc=len.brown@intel.com \
--cc=lijo.lazar@amd.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pm@vger.kernel.org \
--cc=mario.limonciello@amd.com \
--cc=pavel@kernel.org \
--cc=rafael@kernel.org \
--cc=victor.zhao@amd.com \
/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.