Re: [PATCH 1/3] drm/amdgpu: Add new api to switch on/off power profile mode

From: Yadav, Arvind
Date: Thu Aug 17 2023 - 03:59:19 EST



On 8/14/2023 8:28 PM, Shashank Sharma wrote:
Hey Arvind,

On 14/08/2023 09:34, Arvind Yadav wrote:
This patch adds a function which will allow to
change the GPU power profile based on a submitted job.
This can optimize the power performance when the
workload is on.

Cc: Shashank Sharma <shashank.sharma@xxxxxxx>
Cc: Christian Koenig <christian.koenig@xxxxxxx>
Cc: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Arvind Yadav <Arvind.Yadav@xxxxxxx>
---
  drivers/gpu/drm/amd/amdgpu/Makefile           |   2 +-
  drivers/gpu/drm/amd/amdgpu/amdgpu.h           |   3 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |   2 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c  | 156 ++++++++++++++++++
  drivers/gpu/drm/amd/include/amdgpu_workload.h |  44 +++++
  5 files changed, 206 insertions(+), 1 deletion(-)
  create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
  create mode 100644 drivers/gpu/drm/amd/include/amdgpu_workload.h

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 415a7fa395c4..6a9e187d61e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -60,7 +60,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
      amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
      amdgpu_fw_attestation.o amdgpu_securedisplay.o \
      amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o \
-    amdgpu_ring_mux.o
+    amdgpu_ring_mux.o amdgpu_workload.o
    amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
  diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 02b827785e39..1939fa1af8a6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -107,6 +107,7 @@
  #include "amdgpu_fdinfo.h"
  #include "amdgpu_mca.h"
  #include "amdgpu_ras.h"
+#include "amdgpu_workload.h"
    #define MAX_GPU_INSTANCE        16
  @@ -1050,6 +1051,8 @@ struct amdgpu_device {
        bool                            job_hang;
      bool                            dc_enabled;
+
+    struct amdgpu_smu_workload    smu_workload;
  };
    static inline struct amdgpu_device *drm_to_adev(struct drm_device *ddev)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 5c7d40873ee2..0ec18b8fe29f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3672,6 +3672,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
        INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func);
  +    amdgpu_smu_workload_init(adev);
+
      adev->gfx.gfx_off_req_count = 1;
      adev->gfx.gfx_off_residency = 0;
      adev->gfx.gfx_off_entrycount = 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
new file mode 100644
index 000000000000..ce0339d75c12
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
@@ -0,0 +1,156 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2023 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 "amdgpu.h"
+
+/* 100 millsecond timeout */
+#define SMU_IDLE_TIMEOUT    msecs_to_jiffies(100)
+
+static enum PP_SMC_POWER_PROFILE
+ring_to_power_profile(uint32_t ring_type)
+{
+    switch (ring_type) {
+    case AMDGPU_RING_TYPE_GFX:
+        return PP_SMC_POWER_PROFILE_FULLSCREEN3D;
+    case AMDGPU_RING_TYPE_COMPUTE:
+        return PP_SMC_POWER_PROFILE_COMPUTE;
+    case AMDGPU_RING_TYPE_UVD:
+    case AMDGPU_RING_TYPE_VCE:
+    case AMDGPU_RING_TYPE_UVD_ENC:
+    case AMDGPU_RING_TYPE_VCN_DEC:
+    case AMDGPU_RING_TYPE_VCN_ENC:
+    case AMDGPU_RING_TYPE_VCN_JPEG:
+        return PP_SMC_POWER_PROFILE_VIDEO;
+    default:
+        return PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
+    }
+}
+
+static void
+amdgpu_power_profile_set(struct amdgpu_device *adev,
+             enum PP_SMC_POWER_PROFILE profile)
This function expects the caller to hold the smu_workload_mutex, may be we should document it.
+{
+    int ret = amdgpu_dpm_switch_power_profile(adev, profile, true);
I think we should pass this return value to caller instead of keeping the function void.
Noted,
+
+    if (ret == 0) {
+        /* Set the bit for the submitted workload profile */
+        adev->smu_workload.submit_workload_status |= (1 << profile);
+ atomic_inc(&adev->smu_workload.power_profile_ref[profile]);
+    } else {
+        DRM_ERROR("Failed to set power profile, error %d\n", ret);
This can be a warning instead of error.
Noted,
+    }
+
+}
+
+static void
+amdgpu_power_profile_clear(struct amdgpu_device *adev,
+               enum PP_SMC_POWER_PROFILE profile)
+{
+    int ret = amdgpu_dpm_switch_power_profile(adev, profile, false);
same for return value here as well.
Noted,

+
+    if (ret == 0) {
+         /* Clear the bit for the submitted workload profile */
+        adev->smu_workload.submit_workload_status &= ~(1 << profile);
+    } else
+        DRM_ERROR("Failed to clear power profile, error %d\n", ret);
+
+}
+
+static void amdgpu_smu_idle_work_handler(struct work_struct *work)
+{
+
+    struct amdgpu_smu_workload *wl = container_of(work,
+                              struct amdgpu_smu_workload,
+                              smu_delayed_work.work);
+    struct amdgpu_device *adev = wl->adev;
+    bool reschedule = false;
+
+    mutex_lock(&adev->smu_workload.workload_lock);
+    for (int index  = fls(adev->smu_workload.submit_workload_status);
This can be kept outside the for() for better readability and alignment.
Noted,
+         index >= 0; index--) {
+        if (!atomic_read(&adev->smu_workload.power_profile_ref[index]) &&
+            adev->smu_workload.submit_workload_status & (1 << index)) {
+            amdgpu_power_profile_clear(adev, index);
+        } else if (atomic_read(&adev->smu_workload.power_profile_ref[index]))
+            reschedule = true;
+    }
+

This block can be re-arranged a bit for better readability, pls consider:

for () {

    atomic_t val = atomic_read(&adev->smu_workload.power_profile_ref[index];

    if (val) {

        reschedule = true;

        break;

    } else {

        if (adev->smu_workload.submit_workload_status & (1 << index))

            amdgpu_power_profile_clear(adev, index);

    }

}

Noted,
+    if (reschedule)
+ schedule_delayed_work(&adev->smu_workload.smu_delayed_work,
+                      SMU_IDLE_TIMEOUT);
pls check the return value of work
Noted,
+
+    mutex_unlock(&adev->smu_workload.workload_lock);
+}
+
+void amdgpu_put_workload_profile(struct amdgpu_device *adev,
+                 uint32_t ring_type)
+{
+
+    enum PP_SMC_POWER_PROFILE profile = ring_to_power_profile(ring_type);
+
+    if (profile == PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT)
+        return;
+
+    mutex_lock(&adev->smu_workload.workload_lock);
+ atomic_dec(&adev->smu_workload.power_profile_ref[profile]);
+ schedule_delayed_work(&adev->smu_workload.smu_delayed_work, SMU_IDLE_TIMEOUT);
+    mutex_unlock(&adev->smu_workload.workload_lock);
+}
+
+void amdgpu_set_workload_profile(struct amdgpu_device *adev,
+                 uint32_t ring_type)
I would prefer if you can split this patch into two, one just to set profile, other to clear profile and schedule work.
Noted,
+{
+    enum PP_SMC_POWER_PROFILE profile = ring_to_power_profile(ring_type);
+
+    if (profile == PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT)
+        return;
+
+    mutex_lock(&adev->smu_workload.workload_lock);
+ cancel_delayed_work_sync(&adev->smu_workload.smu_delayed_work);
Please check the return value here and proceed only when we were able to cancel successfully.
Noted,
+
+    amdgpu_power_profile_set(adev, profile);
+
+    /* Clear the already finished jobs of higher power profile*/
+    for (int index = fls(adev->smu_workload.submit_workload_status);
+         index > profile; index--) {
+        if (!atomic_read(&adev->smu_workload.power_profile_ref[index]) &&
+            adev->smu_workload.submit_workload_status & (1 << index)) {
+            amdgpu_power_profile_clear(adev, index);
+        }
+    }
+
+    mutex_unlock(&adev->smu_workload.workload_lock);
+}
+
+void amdgpu_smu_workload_init(struct amdgpu_device *adev)
+{
+    struct amdgpu_smu_workload wl;
+
+    wl.adev = adev;
+    wl.submit_workload_status = 0;
+    adev->smu_workload = wl;

Why do we need variable wl at all, which is a local variable of the stack ? You can just do:
Noted,

adev->smu_workload.adev = adev;
adev->smu_workload.submit_workload_status = 0;

+
+    mutex_init(&adev->smu_workload.workload_lock);
+ INIT_DELAYED_WORK(&adev->smu_workload.smu_delayed_work, amdgpu_smu_idle_work_handler);

Are we missing the respective amdgpu_smu_workload_fini which will destroy the mutex ?

Noted,

Regards,

~Arvind

- Shashank

+}
diff --git a/drivers/gpu/drm/amd/include/amdgpu_workload.h b/drivers/gpu/drm/amd/include/amdgpu_workload.h
new file mode 100644
index 000000000000..09804c3d2869
--- /dev/null
+++ b/drivers/gpu/drm/amd/include/amdgpu_workload.h
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright 2023 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.
+ *
+ */
+
+#ifndef _AMDGPU_WORKLOAD_H_
+#define _AMDGPU_WORKLOAD_H_
+
+struct amdgpu_smu_workload {
+    struct amdgpu_device    *adev;
+    struct mutex        workload_lock;
+    struct delayed_work    smu_delayed_work;
+    uint32_t        submit_workload_status;
+    atomic_t power_profile_ref[PP_SMC_POWER_PROFILE_COUNT];
+};
+
+void amdgpu_set_workload_profile(struct amdgpu_device *adev,
+                 uint32_t ring_type);
+
+void amdgpu_put_workload_profile(struct amdgpu_device *adev,
+                 uint32_t ring_type);
+
+void amdgpu_smu_workload_init(struct amdgpu_device *adev);
+
+#endif