Skip to contentSkip to conversation

drm/amdgpu/pm: move additional logic into amdgpu_dpm_force_performance_level

PATCH In review Alex Deucher 2 participants 4 messages Jan 10, 2022 – Jan 11, 2022

Subsystems (4)
lore
Share & tools
lore

Alex Deucher

author maintainer
#0 lore raw

PATCH 1/3 drm/amdgpu/pm: move additional logic into amdgpu_dpm_force_performance_level

This is part of the forced performance level. Move it from the sysfs handler into amdgpu_dpm_force_performance_level.

---
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c | 61 ++++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/pm/amdgpu_pm.c  | 50 -----------------------
 2 files changed, 59 insertions(+), 52 deletions(-)
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +59 -2
index 68d2e80a673b..728b6e10f302 100644
@@ -692,16 +692,25 @@ void amdgpu_dpm_set_power_state(struct amdgpu_device *adev,
 		amdgpu_dpm_compute_clocks(adev);
 }
 
-enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
+static enum amd_dpm_forced_level amdgpu_dpm_get_performance_level_locked(struct amdgpu_device *adev)
 {
 	const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
 	enum amd_dpm_forced_level level;
 
-	mutex_lock(&adev->pm.mutex);
 	if (pp_funcs->get_performance_level)
 		level = pp_funcs->get_performance_level(adev->powerplay.pp_handle);
 	else
 		level = adev->pm.dpm.forced_level;
+
+	return level;
+}
+
+enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
+{
+	enum amd_dpm_forced_level level;
+
+	mutex_lock(&adev->pm.mutex);
+	level = amdgpu_dpm_get_performance_level_locked(adev);
 	mutex_unlock(&adev->pm.mutex);
 
 	return level;
@@ -711,6 +720,11 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
 				       enum amd_dpm_forced_level level)
 {
 	const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
+	enum amd_dpm_forced_level current_level;
+	uint32_t profile_mode_mask = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD |
+					AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK |
+					AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK |
+					AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
 	int ret = 0;
 
 	if (!pp_funcs->force_performance_level)
@@ -723,6 +737,49 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
 		goto out;
 	}
 
+	current_level = amdgpu_dpm_get_performance_level_locked(adev);
+	if (current_level == level) {
+		ret = 0;
+		goto out;
+	}
+
+	if (adev->asic_type == CHIP_RAVEN) {
+		if (!(adev->apu_flags & AMD_APU_IS_RAVEN2)) {
+			if (current_level != AMD_DPM_FORCED_LEVEL_MANUAL &&
+			    level == AMD_DPM_FORCED_LEVEL_MANUAL)
+				amdgpu_gfx_off_ctrl(adev, false);
+			else if (current_level == AMD_DPM_FORCED_LEVEL_MANUAL &&
+				 level != AMD_DPM_FORCED_LEVEL_MANUAL)
+				amdgpu_gfx_off_ctrl(adev, true);
+		}
+	}
+
+	if (!(current_level & profile_mode_mask) &&
+	    (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (!(current_level & profile_mode_mask) &&
+	      (level & profile_mode_mask)) {
+		/* enter UMD Pstate */
+		amdgpu_device_ip_set_powergating_state(adev,
+						       AMD_IP_BLOCK_TYPE_GFX,
+						       AMD_PG_STATE_UNGATE);
+		amdgpu_device_ip_set_clockgating_state(adev,
+						       AMD_IP_BLOCK_TYPE_GFX,
+						       AMD_CG_STATE_UNGATE);
+	} else if ((current_level & profile_mode_mask) &&
+		    !(level & profile_mode_mask)) {
+		/* exit UMD Pstate */
+		amdgpu_device_ip_set_clockgating_state(adev,
+						       AMD_IP_BLOCK_TYPE_GFX,
+						       AMD_CG_STATE_GATE);
+		amdgpu_device_ip_set_powergating_state(adev,
+						       AMD_IP_BLOCK_TYPE_GFX,
+						       AMD_PG_STATE_GATE);
+	}
+
 	if (pp_funcs->force_performance_level(adev->powerplay.pp_handle,
 					      level))
 		ret = -EINVAL;
drivers/gpu/drm/amd/pm/amdgpu_pm.c +0 -50

62 lines not on this page · raw message

Signature · 1 line
2.33.1
author maintainer
#1 lore raw

PATCH 2/3 drm/amdgpu/UAPI: add new CTX OP to get/set stable pstates

Add a new CTX ioctl operation to set stable pstates for profiling. 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 profiling pstates 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. The sysfs interface will override whatever has been set by this interface.

Mesa MR: https://gitlab.freedesktop.org/mesa/drm/-/merge_requests/207

v2: don't default r = 0;
v3: rebase on Evan's PM cleanup

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    | 129 ++++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h    |   1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |   1 +
 drivers/gpu/drm/amd/pm/amdgpu_pm.c         |   5 +
 drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h    |   3 +
 include/uapi/drm/amdgpu_drm.h              |  17 ++-
 6 files changed, 153 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +127 -2
index 468003583b2a..2c929fa40379 100644
@@ -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->stable_pstate = AMDGPU_CTX_STABLE_PSTATE_NONE;
 
 	return 0;
 }
@@ -255,6 +256,86 @@ static void amdgpu_ctx_fini_entity(struct amdgpu_ctx_entity *entity)
 	kfree(entity);
 }
 
+static int amdgpu_ctx_get_stable_pstate(struct amdgpu_ctx *ctx,
+					u32 *stable_pstate)
+{
+	struct amdgpu_device *adev = ctx->adev;
+	enum amd_dpm_forced_level current_level;
+
+	if (!ctx)
+		return -EINVAL;
+
+	current_level = amdgpu_dpm_get_performance_level(adev);
+
+	switch (current_level) {
+	case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_STANDARD;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_PEAK;
+		break;
+	default:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_NONE;
+		break;
+	}
+	return 0;
+}
+
+static int amdgpu_ctx_set_stable_pstate(struct amdgpu_ctx *ctx,
+					u32 stable_pstate)
+{
+	struct amdgpu_device *adev = ctx->adev;
+	enum amd_dpm_forced_level level;
+	int r;
+
+	if (!ctx)
+		return -EINVAL;
+
+	mutex_lock(&adev->pm.stable_pstate_ctx_lock);
+	if (adev->pm.stable_pstate_ctx && adev->pm.stable_pstate_ctx != ctx) {
+		r = -EBUSY;
+		goto done;
+	}
+
+	switch (stable_pstate) {
+	case AMDGPU_CTX_STABLE_PSTATE_NONE:
+		level = AMD_DPM_FORCED_LEVEL_AUTO;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_STANDARD:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_PEAK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
+		break;
+	default:
+		r = -EINVAL;
+		goto done;
+	}
+
+	r = amdgpu_dpm_force_performance_level(adev, level);
+
+	if (level == AMD_DPM_FORCED_LEVEL_AUTO)
+		adev->pm.stable_pstate_ctx = NULL;
+	else
+		adev->pm.stable_pstate_ctx = ctx;
+done:
+	mutex_unlock(&adev->pm.stable_pstate_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 +351,7 @@ static void amdgpu_ctx_fini(struct kref *ref)
 			ctx->entities[i][j] = NULL;
 		}
 	}
-
+	amdgpu_ctx_set_stable_pstate(ctx, AMDGPU_CTX_STABLE_PSTATE_NONE);
 	mutex_destroy(&ctx->lock);
 	kfree(ctx);
 }
@@ -467,11 +548,41 @@ static int amdgpu_ctx_query2(struct amdgpu_device *adev,
 	return 0;
 }
 
+
+
+static int amdgpu_ctx_stable_pstate(struct amdgpu_device *adev,
+				    struct amdgpu_fpriv *fpriv, uint32_t id,
+				    bool set, u32 *stable_pstate)
+{
+	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;
+	}
+
+	if (set)
+		r = amdgpu_ctx_set_stable_pstate(ctx, *stable_pstate);
+	else
+		r = amdgpu_ctx_get_stable_pstate(ctx, stable_pstate);
+
+	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, stable_pstate;
 	int32_t priority;
 
 	union drm_amdgpu_ctx *args = data;
@@ -500,6 +611,20 @@ 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_GET_STABLE_PSTATE:
+		if (args->in.flags)
+			return -EINVAL;
+		r = amdgpu_ctx_stable_pstate(adev, fpriv, id, false, &stable_pstate);
+		args->out.pstate.flags = stable_pstate;
+		break;
+	case AMDGPU_CTX_OP_SET_STABLE_PSTATE:
+		if (args->in.flags & ~AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK)
+			return -EINVAL;
+		stable_pstate = args->in.flags & AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK;
+		if (stable_pstate > AMDGPU_CTX_STABLE_PSTATE_PEAK)
+			return -EINVAL;
+		r = amdgpu_ctx_stable_pstate(adev, fpriv, id, true, &stable_pstate);
+		break;
 	default:
 		return -EINVAL;
 	}
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h +1 -0
index a44b8b8ed39c..142f2f87d44c 100644
@@ -53,6 +53,7 @@ struct amdgpu_ctx {
 	atomic_t			guilty;
 	unsigned long			ras_counter_ce;
 	unsigned long			ras_counter_ue;
+	uint32_t			stable_pstate;
 };
 
 struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -0
index 257593bd1c40..3229381bf3fe 100644
@@ -3497,6 +3497,7 @@ 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->pm.stable_pstate_ctx_lock);
 
 	r = amdgpu_device_init_apu_flags(adev);
 	if (r)
drivers/gpu/drm/amd/pm/amdgpu_pm.c +5 -0
index d2823aaeca09..7731dbbcb288 100644
@@ -336,11 +336,16 @@ static ssize_t amdgpu_set_power_dpm_force_performance_level(struct device *dev,
 		return ret;
 	}
 
+	mutex_lock(&adev->pm.stable_pstate_ctx_lock);
 	if (amdgpu_dpm_force_performance_level(adev, level)) {
 		pm_runtime_mark_last_busy(ddev->dev);
 		pm_runtime_put_autosuspend(ddev->dev);
+		mutex_unlock(&adev->pm.stable_pstate_ctx_lock);
 		return -EINVAL;
 	}
+	/* override whatever a user ctx may have set */
+	adev->pm.stable_pstate_ctx = NULL;
+	mutex_unlock(&adev->pm.stable_pstate_ctx_lock);
 
 	pm_runtime_mark_last_busy(ddev->dev);
 	pm_runtime_put_autosuspend(ddev->dev);
drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h +3 -0
index ba857ca75392..64e54d0daa15 100644
@@ -328,6 +328,9 @@ struct amdgpu_pm {
 	uint32_t		smu_debug_mask;
 
 	bool			pp_force_state_enabled;
+
+	struct mutex            stable_pstate_ctx_lock;
+	struct amdgpu_ctx       *stable_pstate_ctx;
 };
 
 u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev);
include/uapi/drm/amdgpu_drm.h +16 -1
index 0b94ec7b73e7..7f01f9830bf8 100644
@@ -206,6 +206,8 @@ 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_GET_STABLE_PSTATE	5
+#define AMDGPU_CTX_OP_SET_STABLE_PSTATE	6
 
 /* GPU reset status */
 #define AMDGPU_CTX_NO_RESET		0
@@ -238,10 +240,18 @@ union drm_amdgpu_bo_list {
 #define AMDGPU_CTX_PRIORITY_HIGH        512
 #define AMDGPU_CTX_PRIORITY_VERY_HIGH   1023
 
+/* select a stable profiling pstate for perfmon tools */
+#define AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK  0xf
+#define AMDGPU_CTX_STABLE_PSTATE_NONE  0
+#define AMDGPU_CTX_STABLE_PSTATE_STANDARD  1
+#define AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK  2
+#define AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK  3
+#define AMDGPU_CTX_STABLE_PSTATE_PEAK  4
+
 struct drm_amdgpu_ctx_in {
 	/** AMDGPU_CTX_OP_* */
 	__u32	op;
-	/** For future use, no flags defined so far */
+	/** Flags */
 	__u32	flags;
 	__u32	ctx_id;
 	/** AMDGPU_CTX_PRIORITY_* */
@@ -262,6 +272,11 @@ union drm_amdgpu_ctx_out {
 			/** Reset status since the last call of the ioctl. */
 			__u32	reset_status;
 		} state;
+
+		struct {
+			__u32	flags;
+			__u32	_pad;
+		} pstate;
 };
 
 union drm_amdgpu_ctx {
Signature · 1 line
2.33.1
author maintainer
#2 lore raw

PATCH 3/3 drm/amdgpu: bump driver version for new CTX OP to set/get stable pstates

In reply to Alex Deucher

So mesa and tools know when this is available.

Mesa MR: https://gitlab.freedesktop.org/mesa/drm/-/merge_requests/207

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +2 -1
index d3ef99d7d117..5e3de111ae0e 100644
@@ -99,9 +99,10 @@
  * - 3.42.0 - Add 16bpc fixed point display support
  * - 3.43.0 - Add device hot plug/unplug support
  * - 3.44.0 - DCN3 supports DCC independent block settings: !64B && 128B, 64B && 128B
+ * - 3.45.0 - Add context ioctl stable pstate interface
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	44
+#define KMS_DRIVER_MINOR	45
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit;
Signature · 1 line
2.33.1
#3 lore raw

RE: [PATCH 3/3] drm/amdgpu: bump driver version for new CTX OP to set/get stable pstates

[AMD Official Use Only]

Series is reviewed-by: Evan Quan <[email protected]>

Quoted 46 lines

-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Alex Deucher
Sent: Tuesday, January 11, 2022 4:20 AM
To: [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: [PATCH 3/3] drm/amdgpu: bump driver version for new CTX OP to set/get stable pstates

So mesa and tools know when this is available.

Mesa MR:
https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitla
b.freedesktop.org%2Fmesa%2Fdrm%2F- %2Fmerge_requests%2F207&amp;data=04%7C01%7Cevan.quan%40amd.co m%7C2c0626cabecf4fe0056008d9d476afce%7C3dd8961fe4884e608e11a82d99 4e183d%7C0%7C0%7C637774428494652078%7CUnknown%7CTWFpbGZsb3d8 eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3 D%7C3000&amp;sdata=YqpsYhXD24uIFjntEAkeWvEMPAuCP42kbh02aqEPHO 4%3D&amp;reserved=0

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0 -0
index d3ef99d7d117..5e3de111ae0e 100644
@@ -99,9 +99,10 @@
  * - 3.42.0 - Add 16bpc fixed point display support
  * - 3.43.0 - Add device hot plug/unplug support
  * - 3.44.0 - DCN3 supports DCC independent block settings: !64B && 128B,

64B && 128B
+ * - 3.45.0 - Add context ioctl stable pstate interface */
#define KMS_DRIVER_MAJOR 3

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1 -2
-#define KMS_DRIVER_MINOR	44
+#define KMS_DRIVER_MINOR	45
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit;
--

2.33.1