Re: [RFC PATCH 1/2] drm/amdgpu/UAPI: add new PROFILE IOCTL

[Alex Deucher <alexdeucher@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 13613 bytes --]

On Fri, Dec 3, 2021 at 7:15 AM Christian König
<ckoenig.leichtzumerken@gmail.com> wrote:
>
> Am 02.12.21 um 20:19 schrieb Alex Deucher:
> > This adds a new IOCTL currently used to implement querying
> > and setting the stable power state for GPU profiling.  The
> > stable pstates use fixed clocks and disable certain power
> > features in order to get accurate pipeline profiling.
> >
> > Currently this is handled via sysfs, and that is still
> > available, but this makes it easier for applications
> > to utilize.  Note that the power state is global so
> > setting it will affect all applications.  There are currently
> > no checks in place to prevent multiple applications from
> > using this interface, but it doesn't make sense to do
> > profiling while you have multiple applications running in the
> > first place, so it's up to the user to ensure this in order
> > to get good results.
> >
> > This patch add an interface to query what profiling mode is
> > currently active and to set enable a profiling mode.
>
> First of all I wouldn't call this profiling mode. Profiling is the use
> case, but performance and power management are what is controlled here.
>

Ok.

> Then we already have functionality for process and context priority
> override for the SW scheduler in amdgpu_sched.c. I think we should add
> this functionality there.

I'm not sure I follow.  Do you mean to integrate this with the
rendering context?  That was what my original patch did (attached).
Or were you thinking it would be better to make this part of the sched
ioctl?

Alex


>
> Christian.
>
> >
> > Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/Makefile         |   2 +-
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c     |   2 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_profile.c | 112 ++++++++++++++++++++
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_profile.h |  30 ++++++
> >   include/uapi/drm/amdgpu_drm.h               |  28 +++++
> >   5 files changed, 173 insertions(+), 1 deletion(-)
> >   create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_profile.c
> >   create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_profile.h
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
> > index 7fedbb725e17..4cf5bf637a9e 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/Makefile
> > +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
> > @@ -58,7 +58,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
> >       amdgpu_vm_sdma.o amdgpu_discovery.o amdgpu_ras_eeprom.o amdgpu_nbio.o \
> >       amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
> >       amdgpu_fw_attestation.o amdgpu_securedisplay.o amdgpu_hdp.o \
> > -     amdgpu_eeprom.o amdgpu_mca.o
> > +     amdgpu_eeprom.o amdgpu_mca.o amdgpu_profile.o
> >
> >   amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> > index bc1355c6248d..0e27f9673f8f 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> > @@ -46,6 +46,7 @@
> >   #include "amdgpu_sched.h"
> >   #include "amdgpu_fdinfo.h"
> >   #include "amdgpu_amdkfd.h"
> > +#include "amdgpu_profile.h"
> >
> >   #include "amdgpu_ras.h"
> >   #include "amdgpu_xgmi.h"
> > @@ -2467,6 +2468,7 @@ const struct drm_ioctl_desc amdgpu_ioctls_kms[] = {
> >       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
> >       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
> >       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
> > +     DRM_IOCTL_DEF_DRV(AMDGPU_PROFILE, amdgpu_profile_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
> >   };
> >
> >   static const struct drm_driver amdgpu_kms_driver = {
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.c
> > new file mode 100644
> > index 000000000000..94fe408e810f
> > --- /dev/null
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.c
> > @@ -0,0 +1,112 @@
> > +/*
> > + * Copyright 2021 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/amdgpu_drm.h>
> > +#include "amdgpu.h"
> > +
> > +/**
> > + * amdgpu_profile_ioctl - Manages settings for profiling.
> > + *
> > + * @dev: drm device pointer
> > + * @data: drm_amdgpu_vm
> > + * @filp: drm file pointer
> > + *
> > + * Returns:
> > + * 0 for success, -errno for errors.
> > + */
> > +int amdgpu_profile_ioctl(struct drm_device *dev, void *data,
> > +                      struct drm_file *filp)
> > +{
> > +     union drm_amdgpu_profile *args = data;
> > +     struct amdgpu_device *adev = drm_to_adev(dev);
> > +     const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
> > +     enum amd_dpm_forced_level current_level, requested_level;
> > +     int r;
> > +
> > +     if (pp_funcs->get_performance_level)
> > +             current_level = amdgpu_dpm_get_performance_level(adev);
> > +     else
> > +             current_level = adev->pm.dpm.forced_level;
> > +
> > +     switch (args->in.op) {
> > +     case AMDGPU_PROFILE_OP_GET_STABLE_PSTATE:
> > +             if (args->in.flags)
> > +                     return -EINVAL;
> > +             switch (current_level) {
> > +             case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD:
> > +                     args->out.flags = AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_STANDARD;
> > +                     break;
> > +             case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK:
> > +                     args->out.flags = AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_SCLK;
> > +                     break;
> > +             case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK:
> > +                     args->out.flags = AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_MCLK;
> > +                     break;
> > +             case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
> > +                     args->out.flags = AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_PEAK;
> > +                     break;
> > +             default:
> > +                     args->out.flags = AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_NONE;
> > +                     break;
> > +             }
> > +             break;
> > +     case AMDGPU_PROFILE_OP_SET_STABLE_PSTATE:
> > +             if (args->in.flags & ~AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MASK)
> > +                     return -EINVAL;
> > +             switch (args->in.flags & AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MASK) {
> > +             case AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_STANDARD:
> > +                     requested_level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
> > +                     break;
> > +             case AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_SCLK:
> > +                     requested_level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
> > +                     break;
> > +             case AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_MCLK:
> > +                     requested_level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
> > +                     break;
> > +             case AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_PEAK:
> > +                     requested_level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
> > +                     break;
> > +             case AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_NONE:
> > +                     requested_level = AMD_DPM_FORCED_LEVEL_AUTO;
> > +                     break;
> > +             default:
> > +                     return -EINVAL;
> > +             }
> > +
> > +             if ((current_level != requested_level) && pp_funcs->force_performance_level) {
> > +                     mutex_lock(&adev->pm.mutex);
> > +                     r = amdgpu_dpm_force_performance_level(adev, requested_level);
> > +                     if (!r)
> > +                             adev->pm.dpm.forced_level = requested_level;
> > +                     mutex_unlock(&adev->pm.mutex);
> > +                     if (r)
> > +                             return r;
> > +             }
> > +             break;
> > +     default:
> > +             return -EINVAL;
> > +     }
> > +
> > +     return 0;
> > +}
> > +
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.h
> > new file mode 100644
> > index 000000000000..cd1c597bae11
> > --- /dev/null
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_profile.h
> > @@ -0,0 +1,30 @@
> > +/*
> > + * Copyright 2021 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_PROFILE_H__
> > +#define __AMDGPU_PROFILE_H__
> > +
> > +int amdgpu_profile_ioctl(struct drm_device *dev, void *data,
> > +                      struct drm_file *filp);
> > +
> > +#endif
> > diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
> > index 26e45fc5eb1a..b6edf4a826f9 100644
> > --- a/include/uapi/drm/amdgpu_drm.h
> > +++ b/include/uapi/drm/amdgpu_drm.h
> > @@ -54,6 +54,7 @@ extern "C" {
> >   #define DRM_AMDGPU_VM                       0x13
> >   #define DRM_AMDGPU_FENCE_TO_HANDLE  0x14
> >   #define DRM_AMDGPU_SCHED            0x15
> > +#define DRM_AMDGPU_PROFILE           0x16
> >
> >   #define DRM_IOCTL_AMDGPU_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create)
> >   #define DRM_IOCTL_AMDGPU_GEM_MMAP   DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap)
> > @@ -71,6 +72,7 @@ extern "C" {
> >   #define DRM_IOCTL_AMDGPU_VM         DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_VM, union drm_amdgpu_vm)
> >   #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
> >   #define DRM_IOCTL_AMDGPU_SCHED              DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
> > +#define DRM_IOCTL_AMDGPU_PROFILE     DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_PROFILE, union drm_amdgpu_profile)
> >
> >   /**
> >    * DOC: memory domains
> > @@ -1120,6 +1122,32 @@ struct drm_amdgpu_info_video_caps {
> >       struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
> >   };
> >
> > +/* profile ioctl */
> > +#define AMDGPU_PROFILE_OP_GET_STABLE_PSTATE  1
> > +#define AMDGPU_PROFILE_OP_SET_STABLE_PSTATE  2
> > +
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MASK      0xf
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_NONE      0
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_STANDARD  1
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_SCLK  2
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_MIN_MCLK  3
> > +#define AMDGPU_PROFILE_FLAGS_STABLE_PSTATE_PEAK      4
> > +
> > +struct drm_amdgpu_profile_in {
> > +     /** AMDGPU_PROFILE_OP_* */
> > +     __u32   op;
> > +     __u32   flags;
> > +};
> > +
> > +struct drm_amdgpu_profile_out {
> > +     __u64   flags;
> > +};
> > +
> > +union drm_amdgpu_profile {
> > +     struct drm_amdgpu_profile_in in;
> > +     struct drm_amdgpu_profile_out out;
> > +};
> > +
> >   /*
> >    * Supported GPU families
> >    */
>

[-- Attachment #2: original.patch --]
[-- Type: text/x-patch, Size: 7474 bytes --]

commit 03ce38b4f5fc9d43820c669cba7bd8c03e3bdc02
Author: Alex Deucher <alexander.deucher@amd.com>
Date:   Wed Dec 1 16:17:13 2021 -0500

    drm/amdgpu/UAPI: add new CTX OP for setting profile modes
    
    Add a new CTX ioctl operation to set profile modes.  When
    creating traces for tools like RGP or using SPM or doing
    performance profiling, it's required to enable a special
    stable profiling power state on the GPU.  These profiling
    states set fixed clocks and disable certain other power
    features like powergating which may impact the results.
    
    Historically, these profiles were enabled via sysfs, but
    this adds an interface to enable it via the CTX ioctl
    from the application.  Since the power state is global
    only one application can set it at a time, so if multiple
    applications try and use it only the first will get it,
    the ioctl will return -EBUSY for others.
    
    Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index c5cfe2926ca1..95b8908bd2e5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1097,6 +1097,9 @@ struct amdgpu_device {
 
 	struct amdgpu_reset_control     *reset_cntl;
 	uint32_t                        ip_versions[MAX_HWIP][HWIP_MAX_INSTANCE];
+
+	struct mutex                    pstate_profile_ctx_lock;
+	struct amdgpu_ctx               *pstate_profile_ctx;
 };
 
 static inline struct amdgpu_device *drm_to_adev(struct drm_device *ddev)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 468003583b2a..bf8a5939e444 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -237,6 +237,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	ctx->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 	ctx->init_priority = priority;
 	ctx->override_priority = AMDGPU_CTX_PRIORITY_UNSET;
+	ctx->pstate_profile = AMDGPU_CTX_PSTATE_PROFILE_NONE;
 
 	return 0;
 }
@@ -255,6 +256,67 @@ static void amdgpu_ctx_fini_entity(struct amdgpu_ctx_entity *entity)
 	kfree(entity);
 }
 
+static int amdgpu_ctx_do_set_pstate_profile(struct amdgpu_ctx *ctx,
+					    u32 pstate_profile)
+{
+	struct amdgpu_device *adev = ctx->adev;
+	const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
+	enum amd_dpm_forced_level level, current_level;
+	int r = 0;
+
+	if (!ctx)
+		return -EINVAL;
+
+	mutex_lock(&adev->pstate_profile_ctx_lock);
+	if (adev->pstate_profile_ctx && adev->pstate_profile_ctx != ctx) {
+		r = -EBUSY;
+		goto done;
+	}
+
+	switch (pstate_profile) {
+	case AMDGPU_CTX_PSTATE_PROFILE_NONE:
+		level = AMD_DPM_FORCED_LEVEL_AUTO;
+		break;
+	case AMDGPU_CTX_PSTATE_PROFILE_STANDARD:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
+		break;
+	case AMDGPU_CTX_PSTATE_PROFILE_MIN_SCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
+		break;
+	case AMDGPU_CTX_PSTATE_PROFILE_MIN_MCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
+		break;
+	case AMDGPU_CTX_PSTATE_PROFILE_PEAK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
+		break;
+	default:
+		r = -EINVAL;
+		goto done;
+	}
+
+	if (pp_funcs->get_performance_level)
+		current_level = amdgpu_dpm_get_performance_level(adev);
+	else
+		current_level = adev->pm.dpm.forced_level;
+
+	if ((current_level != level) && pp_funcs->force_performance_level) {
+		mutex_lock(&adev->pm.mutex);
+		r = amdgpu_dpm_force_performance_level(adev, level);
+		if (!r)
+			adev->pm.dpm.forced_level = level;
+		mutex_unlock(&adev->pm.mutex);
+	}
+
+	if (level == AMD_DPM_FORCED_LEVEL_AUTO)
+		adev->pstate_profile_ctx = NULL;
+	else
+		adev->pstate_profile_ctx = ctx;
+done:
+	mutex_unlock(&adev->pstate_profile_ctx_lock);
+
+	return r;
+}
+
 static void amdgpu_ctx_fini(struct kref *ref)
 {
 	struct amdgpu_ctx *ctx = container_of(ref, struct amdgpu_ctx, refcount);
@@ -270,7 +332,7 @@ static void amdgpu_ctx_fini(struct kref *ref)
 			ctx->entities[i][j] = NULL;
 		}
 	}
-
+	amdgpu_ctx_do_set_pstate_profile(ctx, AMDGPU_CTX_PSTATE_PROFILE_NONE);
 	mutex_destroy(&ctx->lock);
 	kfree(ctx);
 }
@@ -467,11 +529,38 @@ static int amdgpu_ctx_query2(struct amdgpu_device *adev,
 	return 0;
 }
 
+
+
+static int amdgpu_ctx_set_pstate_profile(struct amdgpu_device *adev,
+					 struct amdgpu_fpriv *fpriv, uint32_t id,
+					 u32 pstate_profile)
+{
+	struct amdgpu_ctx *ctx;
+	struct amdgpu_ctx_mgr *mgr;
+	int r;
+
+	if (!fpriv)
+		return -EINVAL;
+
+	mgr = &fpriv->ctx_mgr;
+	mutex_lock(&mgr->lock);
+	ctx = idr_find(&mgr->ctx_handles, id);
+	if (!ctx) {
+		mutex_unlock(&mgr->lock);
+		return -EINVAL;
+	}
+
+	r = amdgpu_ctx_do_set_pstate_profile(ctx, pstate_profile);
+
+	mutex_unlock(&mgr->lock);
+	return r;
+}
+
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp)
 {
 	int r;
-	uint32_t id;
+	uint32_t id, pstate_profile;
 	int32_t priority;
 
 	union drm_amdgpu_ctx *args = data;
@@ -500,6 +589,14 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 	case AMDGPU_CTX_OP_QUERY_STATE2:
 		r = amdgpu_ctx_query2(adev, fpriv, id, &args->out);
 		break;
+	case AMDGPU_CTX_OP_SET_PSTATE_PROFILE:
+		if (args->in.flags & ~AMDGPU_CTX_PSTATE_PROFILE_FLAGS_MASK)
+			return -EINVAL;
+		pstate_profile = args->in.flags & AMDGPU_CTX_PSTATE_PROFILE_FLAGS_MASK;
+		if (pstate_profile > AMDGPU_CTX_PSTATE_PROFILE_PEAK)
+			return -EINVAL;
+		r = amdgpu_ctx_set_pstate_profile(adev, fpriv, id, pstate_profile);
+		break;
 	default:
 		return -EINVAL;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
index a44b8b8ed39c..7007ec9450dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
@@ -53,6 +53,7 @@ struct amdgpu_ctx {
 	atomic_t			guilty;
 	unsigned long			ras_counter_ce;
 	unsigned long			ras_counter_ue;
+	uint32_t			pstate_profile;
 };
 
 struct amdgpu_ctx_mgr {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index a1c14466f23d..c72e6153239a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3452,6 +3452,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	init_rwsem(&adev->reset_sem);
 	mutex_init(&adev->psp.mutex);
 	mutex_init(&adev->notifier_lock);
+	mutex_init(&adev->pstate_profile_ctx_lock);
 
 	r = amdgpu_device_init_apu_flags(adev);
 	if (r)
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index b9d68339103d..a4c2f7dc2f77 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -206,6 +206,7 @@ union drm_amdgpu_bo_list {
 #define AMDGPU_CTX_OP_FREE_CTX	2
 #define AMDGPU_CTX_OP_QUERY_STATE	3
 #define AMDGPU_CTX_OP_QUERY_STATE2	4
+#define AMDGPU_CTX_OP_SET_PSTATE_PROFILE	5
 
 /* GPU reset status */
 #define AMDGPU_CTX_NO_RESET		0
@@ -238,6 +239,14 @@ union drm_amdgpu_bo_list {
 #define AMDGPU_CTX_PRIORITY_HIGH        512
 #define AMDGPU_CTX_PRIORITY_VERY_HIGH   1023
 
+/* select a profiling pstate for perfmon tools */
+#define AMDGPU_CTX_PSTATE_PROFILE_FLAGS_MASK  0xf
+#define AMDGPU_CTX_PSTATE_PROFILE_NONE  0
+#define AMDGPU_CTX_PSTATE_PROFILE_STANDARD  1
+#define AMDGPU_CTX_PSTATE_PROFILE_MIN_SCLK  2
+#define AMDGPU_CTX_PSTATE_PROFILE_MIN_MCLK  3
+#define AMDGPU_CTX_PSTATE_PROFILE_PEAK  4
+
 struct drm_amdgpu_ctx_in {
 	/** AMDGPU_CTX_OP_* */
 	__u32	op;