On 6/6/2025 3:38 PM, Emily Deng wrote:
> For the suspend and resume process, exclusive access is not required.
> Therefore, it can be moved out of the full access section to reduce the
> duration of exclusive access.
>
> v3:
> Move suspend processes before hardware fini.
> Remove twice call for bare metal.
>
> v4:
> Refine code
>
> Signed-off-by: Emily Deng <emily.d...@amd.com>
Acked-by: Lijo Lazar <lijo.la...@amd.com>
cc: Mukul/Harish to take a look.
Thanks,
Lijo
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 24 ++++++--
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 25 ++++++--
> .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c | 4 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 +++-
> drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c | 4 +-
> drivers/gpu/drm/amd/amdkfd/kfd_device.c | 57 ++++++++++++-------
> 6 files changed, 89 insertions(+), 36 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> index d8ac4b1051a8..fe282b855734 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> @@ -248,18 +248,34 @@ void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
> kgd2kfd_interrupt(adev->kfd.dev, ih_ring_entry);
> }
>
> -void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool run_pm)
> +void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool suspend_proc)
> {
> if (adev->kfd.dev)
> - kgd2kfd_suspend(adev->kfd.dev, run_pm);
> + kgd2kfd_suspend(adev->kfd.dev, suspend_proc);
> }
>
> -int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool run_pm)
> +int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool resume_proc)
> {
> int r = 0;
>
> if (adev->kfd.dev)
> - r = kgd2kfd_resume(adev->kfd.dev, run_pm);
> + r = kgd2kfd_resume(adev->kfd.dev, resume_proc);
> +
> + return r;
> +}
> +
> +void amdgpu_amdkfd_suspend_process(struct amdgpu_device *adev)
> +{
> + if (adev->kfd.dev)
> + kgd2kfd_suspend_process(adev->kfd.dev);
> +}
> +
> +int amdgpu_amdkfd_resume_process(struct amdgpu_device *adev)
> +{
> + int r = 0;
> +
> + if (adev->kfd.dev)
> + r = kgd2kfd_resume_process(adev->kfd.dev);
>
> return r;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> index b6ca41859b53..05ecba2a85eb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> @@ -154,8 +154,10 @@ struct amdkfd_process_info {
> int amdgpu_amdkfd_init(void);
> void amdgpu_amdkfd_fini(void);
>
> -void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool run_pm);
> -int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool run_pm);
> +void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool suspend_proc);
> +int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool resume_proc);
> +void amdgpu_amdkfd_suspend_process(struct amdgpu_device *adev);
> +int amdgpu_amdkfd_resume_process(struct amdgpu_device *adev);
> void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
> const void *ih_ring_entry);
> void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);
> @@ -411,8 +413,10 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device
> *adev, bool vf);
> bool kgd2kfd_device_init(struct kfd_dev *kfd,
> const struct kgd2kfd_shared_resources *gpu_resources);
> void kgd2kfd_device_exit(struct kfd_dev *kfd);
> -void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm);
> -int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm);
> +void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc);
> +int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc);
> +void kgd2kfd_suspend_process(struct kfd_dev *kfd);
> +int kgd2kfd_resume_process(struct kfd_dev *kfd);
> int kgd2kfd_pre_reset(struct kfd_dev *kfd,
> struct amdgpu_reset_context *reset_context);
> int kgd2kfd_post_reset(struct kfd_dev *kfd);
> @@ -454,11 +458,20 @@ static inline void kgd2kfd_device_exit(struct kfd_dev
> *kfd)
> {
> }
>
> -static inline void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
> +static inline void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc)
> {
> }
>
> -static inline int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
> +static inline int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc)
> +{
> + return 0;
> +}
> +
> +static inline void kgd2kfd_suspend_process(struct kfd_dev *kfd, bool
> suspend_proc
> +{
> +}
> +
> +static inline int kgd2kfd_resume_process(struct kfd_dev *kfd, bool
> resume_proc)
> {
> return 0;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> index ffbaa8bc5eea..1105a09e55dc 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> @@ -320,7 +320,7 @@ static void set_barrier_auto_waitcnt(struct amdgpu_device
> *adev, bool enable_wai
> if (!down_read_trylock(&adev->reset_domain->sem))
> return;
>
> - amdgpu_amdkfd_suspend(adev, false);
> + amdgpu_amdkfd_suspend(adev, true);
>
> if (suspend_resume_compute_scheduler(adev, true))
> goto out;
> @@ -333,7 +333,7 @@ static void set_barrier_auto_waitcnt(struct amdgpu_device
> *adev, bool enable_wai
> out:
> suspend_resume_compute_scheduler(adev, false);
>
> - amdgpu_amdkfd_resume(adev, false);
> + amdgpu_amdkfd_resume(adev, true);
>
> up_read(&adev->reset_domain->sem);
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 5289400879ec..e8b500c266c1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -3524,7 +3524,7 @@ static int amdgpu_device_ip_fini_early(struct
> amdgpu_device *adev)
> amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
> amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
>
> - amdgpu_amdkfd_suspend(adev, false);
> + amdgpu_amdkfd_suspend(adev, true);
> amdgpu_userq_suspend(adev);
>
> /* Workaround for ASICs need to disable SMC first */
> @@ -5061,6 +5061,8 @@ int amdgpu_device_suspend(struct drm_device *dev, bool
> notify_clients)
> adev->in_suspend = true;
>
> if (amdgpu_sriov_vf(adev)) {
> + if (!adev->in_s0ix && !adev->in_runpm)
> + amdgpu_amdkfd_suspend_process(adev);
> amdgpu_virt_fini_data_exchange(adev);
> r = amdgpu_virt_request_full_gpu(adev, false);
> if (r)
> @@ -5080,7 +5082,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool
> notify_clients)
> amdgpu_device_ip_suspend_phase1(adev);
>
> if (!adev->in_s0ix) {
> - amdgpu_amdkfd_suspend(adev, adev->in_runpm);
> + amdgpu_amdkfd_suspend(adev, !amdgpu_sriov_vf(adev) &&
> !adev->in_runpm);
> amdgpu_userq_suspend(adev);
> }
>
> @@ -5178,7 +5180,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool
> notify_clients)
> }
>
> if (!adev->in_s0ix) {
> - r = amdgpu_amdkfd_resume(adev, adev->in_runpm);
> + r = amdgpu_amdkfd_resume(adev, !amdgpu_sriov_vf(adev) &&
> !adev->in_runpm);
> if (r)
> goto exit;
>
> @@ -5197,6 +5199,9 @@ int amdgpu_device_resume(struct drm_device *dev, bool
> notify_clients)
> if (amdgpu_sriov_vf(adev)) {
> amdgpu_virt_init_data_exchange(adev);
> amdgpu_virt_release_full_gpu(adev, true);
> +
> + if (!adev->in_s0ix && !r && !adev->in_runpm)
> + r = amdgpu_amdkfd_resume_process(adev);
> }
>
> if (r)
> diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
> b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
> index 9c169112a5e7..48e2d67273f7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
> +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
> @@ -1676,9 +1676,9 @@ static int sdma_v4_4_2_reset_queue(struct amdgpu_ring
> *ring, unsigned int vmid)
> if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
> return -EOPNOTSUPP;
>
> - amdgpu_amdkfd_suspend(adev, false);
> + amdgpu_amdkfd_suspend(adev, true);
> r = amdgpu_sdma_reset_engine(adev, id);
> - amdgpu_amdkfd_resume(adev, false);
> + amdgpu_amdkfd_resume(adev, true);
>
> return r;
> }
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> index bf0854bd5555..6a832f3c2518 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> @@ -971,7 +971,7 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd,
> kfd_smi_event_update_gpu_reset(node, false, reset_context);
> }
>
> - kgd2kfd_suspend(kfd, false);
> + kgd2kfd_suspend(kfd, true);
>
> for (i = 0; i < kfd->num_nodes; i++)
> kfd_signal_reset_event(kfd->nodes[i]);
> @@ -1019,7 +1019,7 @@ bool kfd_is_locked(void)
> return (kfd_locked > 0);
> }
>
> -void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
> +void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc)
> {
> struct kfd_node *node;
> int i;
> @@ -1027,14 +1027,8 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
> if (!kfd->init_complete)
> return;
>
> - /* for runtime suspend, skip locking kfd */
> - if (!run_pm) {
> - mutex_lock(&kfd_processes_mutex);
> - /* For first KFD device suspend all the KFD processes */
> - if (++kfd_locked == 1)
> - kfd_suspend_all_processes();
> - mutex_unlock(&kfd_processes_mutex);
> - }
> + if (suspend_proc)
> + kgd2kfd_suspend_process(kfd);
>
> for (i = 0; i < kfd->num_nodes; i++) {
> node = kfd->nodes[i];
> @@ -1042,7 +1036,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
> }
> }
>
> -int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
> +int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc)
> {
> int ret, i;
>
> @@ -1055,14 +1049,39 @@ int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
> return ret;
> }
>
> - /* for runtime resume, skip unlocking kfd */
> - if (!run_pm) {
> - mutex_lock(&kfd_processes_mutex);
> - if (--kfd_locked == 0)
> - ret = kfd_resume_all_processes();
> - WARN_ONCE(kfd_locked < 0, "KFD suspend / resume ref. error");
> - mutex_unlock(&kfd_processes_mutex);
> - }
> + if (resume_proc)
> + ret = kgd2kfd_resume_process(kfd);
> +
> + return ret;
> +}
> +
> +void kgd2kfd_suspend_process(struct kfd_dev *kfd)
> +{
> + struct kfd_node *node;
> + int i;
> +
> + if (!kfd->init_complete)
> + return;
> +
> + mutex_lock(&kfd_processes_mutex);
> + /* For first KFD device suspend all the KFD processes */
> + if (++kfd_locked == 1)
> + kfd_suspend_all_processes();
> + mutex_unlock(&kfd_processes_mutex);
> +}
> +
> +int kgd2kfd_resume_process(struct kfd_dev *kfd)
> +{
> + int ret, i;
> +
> + if (!kfd->init_complete)
> + return 0;
> +
> + mutex_lock(&kfd_processes_mutex);
> + if (--kfd_locked == 0)
> + ret = kfd_resume_all_processes();
> + WARN_ONCE(kfd_locked < 0, "KFD suspend / resume ref. error");
> + mutex_unlock(&kfd_processes_mutex);
>
> return ret;
> }
