drm/amd: Invert APU check for amdgpu_device_evict_resources()
Subsystems (2)
PATCH v5 1/2 drm/amd: Invert APU check for amdgpu_device_evict_resources()
From: Mario Limonciello <[email protected]>
Resource eviction isn't needed for s3 or s2idle on APUs, but should be run for S4. As amdgpu_device_evict_resources() will be called by prepare notifier adjust logic so that APUs only cover S4.
--
v2:
* New patch
--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -2
@@ -4735,8 +4735,8 @@ static int amdgpu_device_evict_resources(struct amdgpu_device *adev) {int ret; -/* No need to evict vram on APUs for suspend to ram or s2idle */ -if ((adev->in_s3 || adev->in_s0ix) && (adev->flags &AMD_IS_APU )) +/* No need to evict vram on APUs unless going to S4 */ +if (!adev->in_s4 && (adev->flags &AMD_IS_APU ))return 0 ; ret =amdgpu_ttm_evict_resources (adev,TTM_PL_VRAM );
Signature · 1 line
2.43.0
PATCH v5 2/2 drm/amd: Add Suspend/Hibernate notification callback support
From: Mario Limonciello <[email protected]>
As part of the suspend sequence VRAM needs to be evicted on dGPUs. In order to make suspend/resume more reliable we moved this into the pmops prepare() callback so that the suspend sequence would fail but the system could remain operational under high memory usage suspend.
Another class of issues exist though where due to memory fragementation there isn't a large enough contiguous space and swap isn't accessible.
Add support for a suspend/hibernate notification callback that could evict VRAM before tasks are frozen. This should allow paging out to swap if necessary.
---
v5:
* Keep s4 set all the way through prepare.
v4:
* Make non fatal, drop patch 3
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 46 +++++++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 1 - 3 files changed, 46 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
@@ -879,6 +879,7 @@ struct amdgpu_device {bool need_swiotlb;bool accel_working;struct notifier_block acpi_nb; +struct notifier_block pm_nb;struct amdgpu_i2c_chan *i2c_bus[AMDGPU_MAX_I2C_BUS ];struct debugfs_blob_wrapper debugfs_vbios_blob;struct debugfs_blob_wrapper debugfs_discovery_blob;
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +45 -1
76 lines not on this page · raw message
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0 -1
@@ -2552,7 +2552,6 @@ static int amdgpu_pmops_freeze(struct device *dev)struct amdgpu_device *adev =drm_to_adev (drm_dev);int r; - adev->in_s4 = true; r =amdgpu_device_suspend (drm_dev, true); adev->in_s4 = false;if (r)
Signature · 1 line
2.43.0
Re: [PATCH v5 2/2] drm/amd: Add Suspend/Hibernate notification callback support
Quoted 14 lines · Mario Limonciello
From: Mario Limonciello <[email protected]>
As part of the suspend sequence VRAM needs to be evicted on dGPUs. In order to make suspend/resume more reliable we moved this into the pmops
prepare()callback so that the suspend sequence would fail but the system could remain operational under high memory usage suspend.Another class of issues exist though where due to memory fragementation there isn't a large enough contiguous space and swap isn't accessible.
Add support for a suspend/hibernate notification callback that could evict VRAM before tasks are frozen. This should allow paging out to swap if necessary.
Quoted 5 lines
Series is -
Thanks,
Lijo
Quoted 120 lines
---
v5:
* Keep s4 set all the way through prepare.
v4:
* Make non fatal, drop patch 3--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 46 +++++++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 1 - 3 files changed, 46 insertions(+), 2 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-0@@ -879,6 +879,7 @@ struct amdgpu_device {bool need_swiotlb;bool accel_working;struct notifier_block acpi_nb; +struct notifier_block pm_nb;struct amdgpu_i2c_chan *i2c_bus[AMDGPU_MAX_I2C_BUS ];struct debugfs_blob_wrapper debugfs_vbios_blob;struct debugfs_blob_wrapper debugfs_discovery_blob;drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +45
-1@@ -191,6 +191,8 @@ void amdgpu_set_init_level(struct amdgpu_device *adev, }static inline void amdgpu_device_stop_pending_resets (struct amdgpu_device *adev); +static int amdgpu_device_pm_notifier (struct notifier_block *nb,unsigned long mode, +void *data);/** * DOC: pcie_replay_count @@ -4553,6 +4555,11 @@ int amdgpu_device_init(struct amdgpu_device *adev,amdgpu_device_check_iommu_direct_map (adev); + adev->pm_nb.notifier_call = amdgpu_device_pm_notifier; + r =register_pm_notifier (&adev->pm_nb); +if (r) + goto failed; +return 0 ; release_ras_con:@@ -4617,6 +4624,8 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)drain_workqueue (adev->mman.bdev.wq); adev->shutdown = true; +unregister_pm_notifier (&adev->pm_nb); +/* make sure IB test finished before entering exclusive mode * to avoid preemption on IB test */ @@ -4748,6 +4757,41 @@ static int amdgpu_device_evict_resources(struct amdgpu_device *adev)/* * Suspend & resume. */ +/** +* amdgpu_device_pm_notifier - Notification block for Suspend/Hibernate events +* @nb: notifier block +* @mode: suspend mode +* @data: data +* +* This function is called when the system is about to suspend or hibernate. +* It is used to evict resources from the device before the system goes to +* sleep while there is still access to swap. +*/ +static int amdgpu_device_pm_notifier (struct notifier_block *nb,unsigned long mode, +void *data) +{ + struct amdgpu_device *adev = container_of(nb, struct amdgpu_device, pm_nb); +int r; + +switch (mode) { +case PM_HIBERNATION_PREPARE : + adev->in_s4 = true; + fallthrough; +case PM_SUSPEND_PREPARE : + r =amdgpu_device_evict_resources (adev); +/* +* This is considered non-fatal at this time because +* amdgpu_device_prepare() will also fatally evict resources. +* See https://gitlab.freedesktop.org/drm/amd/-/issues/3781 +*/ +if (r) +drm_warn (adev_to_drm (adev),"Failed to evict resources, freeze active processes if problems occur: %d\n" , r); +break ; + } + +return NOTIFY_DONE ; +} +/** * amdgpu_device_prepare - prepare for device suspend * @@ -4787,7 +4831,7 @@ int amdgpu_device_prepare(struct drm_device *dev)return 0 ; unprepare: - adev->in_s0ix = adev->in_s3 = false; + adev->in_s0ix = adev->in_s3 = adev->in_s4 = false;return r; }drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0
-1@@ -2552,7 +2552,6 @@ static int amdgpu_pmops_freeze(struct device *dev)struct amdgpu_device *adev =drm_to_adev (drm_dev);int r; - adev->in_s4 = true; r =amdgpu_device_suspend (drm_dev, true); adev->in_s4 = false;if (r)