-----Original Message-----
From: Sharma, Shashank <Shashank.Sharma@xxxxxxx>
Sent: Tuesday, September 27, 2022 3:30 PM
To: Quan, Evan <Evan.Quan@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Somalapuram,
Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig, Christian
<Christian.Koenig@xxxxxxx>
Subject: Re: [PATCH v3 2/5] drm/amdgpu: add new functions to set GPU
power profile
Hello Evan,
On 9/27/2022 4:14 AM, Quan, Evan wrote:
[AMD Official Use Only - General]
-----Original Message-----
From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of
Shashank Sharma
Sent: Tuesday, September 27, 2022 5:40 AM
To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Somalapuram,
Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig, Christian
<Christian.Koenig@xxxxxxx>; Sharma, Shashank
<Shashank.Sharma@xxxxxxx>
Subject: [PATCH v3 2/5] drm/amdgpu: add new functions to set GPU
power profile
This patch adds new functions which will allow a user to change the
GPU power profile based a GPU workload hint flag.
Cc: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Shashank Sharma <shashank.sharma@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
.../gpu/drm/amd/amdgpu/amdgpu_ctx_workload.c | 97
+++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +
.../gpu/drm/amd/include/amdgpu_ctx_workload.h | 54 +++++++++++
drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h | 5 +
5 files changed, 158 insertions(+), 1 deletion(-)
create mode 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx_workload.c
create mode 100644
drivers/gpu/drm/amd/include/amdgpu_ctx_workload.h
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile
b/drivers/gpu/drm/amd/amdgpu/Makefile
index 5a283d12f8e1..34679c657ecc 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -50,7 +50,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
atombios_dp.o amdgpu_afmt.o amdgpu_trace_points.o \
atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
amdgpu_dma_buf.o amdgpu_vm.o amdgpu_vm_pt.o amdgpu_ib.o
amdgpu_pll.o \
- amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
+ amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o
amdgpu_ctx_workload.o amdgpu_sync.o \
amdgpu_gtt_mgr.o amdgpu_preempt_mgr.o amdgpu_vram_mgr.o
amdgpu_virt.o \
amdgpu_atomfirmware.o amdgpu_vf_error.o amdgpu_sched.o \
amdgpu_debugfs.o amdgpu_ids.o amdgpu_gmc.o \ diff --git
a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx_workload.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx_workload.c
new file mode 100644
index 000000000000..a11cf29bc388
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx_workload.c
@@ -0,0 +1,97 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person
+obtaining a
+ * copy of this software and associated documentation files (the
"Software"),
+ * to deal in the Software without restriction, including without
+ limitation
+ * the rights to use, copy, modify, merge, publish, distribute,
+ sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom
+ the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
KIND,
EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN
NO
EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM,
DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR
THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include <drm/drm.h>
+#include "kgd_pp_interface.h"
+#include "amdgpu_ctx_workload.h"
+
+static enum PP_SMC_POWER_PROFILE
+amdgpu_workload_to_power_profile(uint32_t hint) {
+ switch (hint) {
+ case AMDGPU_CTX_WORKLOAD_HINT_NONE:
+ default:
+ return PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
+
+ case AMDGPU_CTX_WORKLOAD_HINT_3D:
+ return PP_SMC_POWER_PROFILE_FULLSCREEN3D;
+ case AMDGPU_CTX_WORKLOAD_HINT_VIDEO:
+ return PP_SMC_POWER_PROFILE_VIDEO;
+ case AMDGPU_CTX_WORKLOAD_HINT_VR:
+ return PP_SMC_POWER_PROFILE_VR;
+ case AMDGPU_CTX_WORKLOAD_HINT_COMPUTE:
+ return PP_SMC_POWER_PROFILE_COMPUTE;
+ }
+}
+
+int amdgpu_set_workload_profile(struct amdgpu_device *adev,
+ uint32_t hint)
+{
+ int ret = 0;
+ enum PP_SMC_POWER_PROFILE profile =
+ amdgpu_workload_to_power_profile(hint);
+
+ if (adev->pm.workload_mode == hint)
+ return 0;
+
+ mutex_lock(&adev->pm.smu_workload_lock);
+
+ if (adev->pm.workload_mode == hint)
+ goto unlock;
[Quan, Evan] This seems redundant with code above. I saw you dropped
this in Patch4.
But I kind of feel this should be the one which needs to be kept.
Yes, this shuffle happened during the rebase-testing of V3, will update this.
+
+ ret = amdgpu_dpm_switch_power_profile(adev, profile, 1);
+ if (!ret)
+ adev->pm.workload_mode = hint;
+ atomic_inc(&adev->pm.workload_switch_ref);
+
+unlock:
+ mutex_unlock(&adev->pm.smu_workload_lock);
+ return ret;
+}
+
+int amdgpu_clear_workload_profile(struct amdgpu_device *adev,
+ uint32_t hint)
+{
+ int ret = 0;
+ enum PP_SMC_POWER_PROFILE profile =
+ amdgpu_workload_to_power_profile(hint);
+
+ if (hint == AMDGPU_CTX_WORKLOAD_HINT_NONE)
+ return 0;
+
+ /* Do not reset GPU power profile if another reset is coming */
+ if (atomic_dec_return(&adev->pm.workload_switch_ref) > 0)
+ return 0;
+
+ mutex_lock(&adev->pm.smu_workload_lock);
+
+ if (adev->pm.workload_mode != hint)
+ goto unlock;
+
+ ret = amdgpu_dpm_switch_power_profile(adev, profile, 0);
+ if (!ret)
+ adev->pm.workload_mode =
AMDGPU_CTX_WORKLOAD_HINT_NONE;
+
+unlock:
+ mutex_unlock(&adev->pm.smu_workload_lock);
+ return ret;
+}
[Quan, Evan] Instead of setting to
AMDGPU_CTX_WORKLOAD_HINT_NONE, better to reset it back to original
workload profile mode.
That can make it compatible with existing sysfs interface which has similar
functionality for setting workload profile mode.
This API is specifically written to remove any workload profile applied, hense
named as "clear_workload_profile" and the intention is reset. As you can see
in the next patch, the work profile is being set from the job_run and reset
again once the job execution is done.
If there is another set() in progress, the reference counter takes care of that.
So I would like to keep it this way.
