Skip to contentSkip to conversation

SI power management fixes (v3)

PATCH In review Timur Kristóf 2 participants 12 messages Aug 28, 2025

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover SI power management fixes (v3)

This series fixes some power management issues on SI, addressing the feedback I got for the previous versions.

Additionally, the following improvements were made:

There is now a patch that increases the SMC timeout. This gets rid of some errors that happened when switching between power states. Apparently the SMC can sometimes take a long time.

Also removed one of the ad-hoc fixes from si_dpm because it didn't actually address anything, and edited that patch to focus on just solving the flickering on Oland.

And edited the si_upload_smc_data patch yet again to program the soft registers to zero when there are no displays connected.

Timur Kristóf (10):
drm/amdgpu: Power up UVD 3 for FW validation (v2) drm/amd/pm: Disable ULV even if unsupported (v3) drm/amd/pm: Increase SMC timeout on SI and warn (v3) drm/amd/pm: Fix si_upload_smc_data (v3)
drm/amd/pm: Adjust si_upload_smc_data register programming (v3) drm/amd/pm: Treat zero vblank time as too short in si_dpm (v3) drm/amd/pm: Disable MCLK switching with non-DC at 120 Hz+ (v2) drm/amd/pm: Disable SCLK switching on Oland with high pixel clocks

    (v3)

drm/amd/pm: Remove wm_low and wm_high fields from amdgpu_crtc (v2) drm/amd/pm: Print VCE clocks too in si_dpm (v3)

 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h     |  2 -
 drivers/gpu/drm/amd/amdgpu/dce_v10_0.c       |  3 +-
 drivers/gpu/drm/amd/amdgpu/dce_v11_0.c       |  3 +-
 drivers/gpu/drm/amd/amdgpu/dce_v6_0.c        |  1 -
 drivers/gpu/drm/amd/amdgpu/dce_v8_0.c        |  3 +-
 drivers/gpu/drm/amd/amdgpu/uvd_v3_1.c        | 29 +++++-
 drivers/gpu/drm/amd/pm/amdgpu_dpm_internal.c |  7 ++
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c   | 93 ++++++++++++++------
 drivers/gpu/drm/amd/pm/legacy-dpm/si_smc.c   | 26 +++++-
 9 files changed, 125 insertions(+), 42 deletions(-)
Signature · 1 line
2.51.0
#1 lore raw

PATCH 1/10 drm/amdgpu: Power up UVD 3 for FW validation (v2)

Unlike later versions, UVD 3 has firmware validation. For this to work, the UVD should be powered up correctly.

When DPM is enabled and the display clock is off, the SMU may choose a power state which doesn't power the UVD, which can result in failure to initialize UVD.

v2:
Add code comments to explain about the UVD power state and how UVD clock is turned on/off.

Fixes b38f3e80ecec ("drm amdgpu: SI UVD v3_1 (v2)")
---
 drivers/gpu/drm/amd/amdgpu/uvd_v3_1.c | 29 +++++++++++++++++++++++----
 1 file changed, 25 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/uvd_v3_1.c +25 -4
index 5dbaebb592b3..2e79a3afc774 100644
@@ -623,7 +623,22 @@ static void uvd_v3_1_enable_mgcg(struct amdgpu_device *adev,
  *
  * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
  *
- * Initialize the hardware, boot up the VCPU and do some testing
+ * Initialize the hardware, boot up the VCPU and do some testing.
+ *
+ * On SI, the UVD is meant to be used in a specific power state,
+ * or alternatively the driver can manually enable its clock.
+ * In amdgpu we use the dedicated UVD power state when DPM is enabled.
+ * Calling amdgpu_dpm_enable_uvd makes DPM select the UVD power state
+ * for the SMU and afterwards enables the UVD clock.
+ * This is automatically done by amdgpu_uvd_ring_begin_use when work
+ * is submitted to the UVD ring. Here, we have to call it manually
+ * in order to power up UVD before firmware validation.
+ *
+ * Note that we must not disable the UVD clock here, as that would
+ * cause the ring test to fail. However, UVD is powered off
+ * automatically after the ring test: amdgpu_uvd_ring_end_use calls
+ * the UVD idle work handler which will disable the UVD clock when
+ * all fences are signalled.
  */
 static int uvd_v3_1_hw_init(struct amdgpu_ip_block *ip_block)
 {
@@ -633,6 +648,15 @@ static int uvd_v3_1_hw_init(struct amdgpu_ip_block *ip_block)
 	int r;
 
 	uvd_v3_1_mc_resume(adev);
+	uvd_v3_1_enable_mgcg(adev, true);
+
+	/* Make sure UVD is powered during FW validation.
+	 * It's going to be automatically powered off after the ring test.
+	 */
+	if (adev->pm.dpm_enabled)
+		amdgpu_dpm_enable_uvd(adev, true);
+	else
+		amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
 
 	r = uvd_v3_1_fw_validate(adev);
 	if (r) {
@@ -640,9 +664,6 @@ static int uvd_v3_1_hw_init(struct amdgpu_ip_block *ip_block)
 		return r;
 	}
 
-	uvd_v3_1_enable_mgcg(adev, true);
-	amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
-
 	uvd_v3_1_start(adev);
 
 	r = amdgpu_ring_test_helper(ring);
Signature · 1 line
2.51.0
#2 lore raw

PATCH 2/10 drm/amd/pm: Disable ULV even if unsupported (v3)

In reply to Timur Kristóf

Always send PPSMC_MSG_DisableULV to the SMC, even if ULV mode is unsupported, to make sure it is properly turned off.

v3:
Simplify si_disable_ulv further.
Always check the return value of amdgpu_si_send_msg_to_smc.

Fixes 841686df9f7d ("drm/amdgpu: add SI DPM support (v4)")
---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +3 -7
index 52e732be59e3..e71070a23b91 100644
@@ -5637,14 +5637,10 @@ static int si_populate_smc_t(struct amdgpu_device *adev,
 
 static int si_disable_ulv(struct amdgpu_device *adev)
 {
-	struct si_power_info *si_pi = si_get_pi(adev);
-	struct si_ulv_param *ulv = &si_pi->ulv;
+	PPSMC_Result r;
 
-	if (ulv->supported)
-		return (amdgpu_si_send_msg_to_smc(adev, PPSMC_MSG_DisableULV) == PPSMC_Result_OK) ?
-			0 : -EINVAL;
-
-	return 0;
+	r = amdgpu_si_send_msg_to_smc(adev, PPSMC_MSG_DisableULV);
+	return (r == PPSMC_Result_OK) ? 0 : -EINVAL;
 }
 
 static bool si_is_state_ulv_compatible(struct amdgpu_device *adev,
Signature · 1 line
2.51.0
#3 lore raw

PATCH 3/10 drm/amd/pm: Increase SMC timeout on SI and warn (v3)

In reply to Timur Kristóf

The SMC can take an excessive amount of time to process some messages under some conditions.

Background:
Sending a message to the SMC works by writing the message into the mmSMC_MESSAGE_0 register and its optional parameter into the mmSMC_SCRATCH0, and then polling mmSMC_RESP_0. Previously the timeout was AMDGPU_MAX_USEC_TIMEOUT, ie. 100 ms.

Increase the timeout to 200 ms for all messages and to 1 sec for a few messages which I've observed to be especially slow:
PPSMC_MSG_NoForcedLevel
PPSMC_MSG_SetEnabledLevels
PPSMC_MSG_SetForcedLevels
PPSMC_MSG_DisableULV
PPSMC_MSG_SwitchToSwState

This fixes the following problems on Tahiti when switching from a lower clock power state to a higher clock state, such as when DC turns on a display which was previously turned off.

* si_restrict_performance_levels_before_switch would fail (if the user previously forced high clocks using sysfs)
* si_set_sw_state would fail (always)

It turns out that both of those failures were SMC timeouts and that the SMC actually didn't fail or hang, just needs more time to process those.

Add a warning when there is an SMC timeout to make it easier to identify this type of problem in the future.

---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_smc.c | 26 ++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/pm/legacy-dpm/si_smc.c +24 -2
index 4e65ab9e931c..281a5e377aee 100644
@@ -172,20 +172,42 @@ PPSMC_Result amdgpu_si_send_msg_to_smc(struct amdgpu_device *adev,
 {
 	u32 tmp;
 	int i;
+	int usec_timeout;
+
+	/* SMC seems to process some messages exceptionally slowly. */
+	switch (msg) {
+	case PPSMC_MSG_NoForcedLevel:
+	case PPSMC_MSG_SetEnabledLevels:
+	case PPSMC_MSG_SetForcedLevels:
+	case PPSMC_MSG_DisableULV:
+	case PPSMC_MSG_SwitchToSwState:
+		usec_timeout = 1000000; /* 1 sec */
+		break;
+	default:
+		usec_timeout = 200000; /* 200 ms */
+		break;
+	}
 
 	if (!amdgpu_si_is_smc_running(adev))
 		return PPSMC_Result_Failed;
 
 	WREG32(mmSMC_MESSAGE_0, msg);
 
-	for (i = 0; i < adev->usec_timeout; i++) {
+	for (i = 0; i < usec_timeout; i++) {
 		tmp = RREG32(mmSMC_RESP_0);
 		if (tmp != 0)
 			break;
 		udelay(1);
 	}
 
-	return (PPSMC_Result)RREG32(mmSMC_RESP_0);
+	tmp = RREG32(mmSMC_RESP_0);
+	if (tmp == 0) {
+		drm_warn(adev_to_drm(adev),
+			"%s timeout on message: %x (SMC_SCRATCH0: %x)\n",
+			__func__, msg, RREG32(mmSMC_SCRATCH0));
+	}
+
+	return (PPSMC_Result)tmp;
 }
 
 PPSMC_Result amdgpu_si_wait_for_smc_inactive(struct amdgpu_device *adev)
Signature · 1 line
2.51.0
#4 lore raw

PATCH 4/10 drm/amd/pm: Fix si_upload_smc_data (v3)

In reply to Timur Kristóf

The si_upload_smc_data function uses si_write_smc_soft_register to set some register values in the SMC, and expects the result to be PPSMC_Result_OK which is 1.

The PPSMC_Result_OK / PPSMC_Result_Failed values are used for checking the result of a command sent to the SMC. However, the si_write_smc_soft_register actually doesn't send any commands to the SMC and returns zero on success, so this check was incorrect.

Fix that by not checking the return value, just like other calls to si_write_smc_soft_register.

v3:
Additionally, when no display is plugged in, there is no need to restrict MCLK switching, so program the registers to zero.

Fixes 841686df9f7d ("drm/amdgpu: add SI DPM support (v4)")
---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 43 ++++++++++++----------
 1 file changed, 24 insertions(+), 19 deletions(-)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +24 -19
index e71070a23b91..6736c592dfdc 100644
@@ -5813,9 +5813,9 @@ static int si_upload_smc_data(struct amdgpu_device *adev)
 {
 	struct amdgpu_crtc *amdgpu_crtc = NULL;
 	int i;
-
-	if (adev->pm.dpm.new_active_crtc_count == 0)
-		return 0;
+	u32 crtc_index = 0;
+	u32 mclk_change_block_cp_min = 0;
+	u32 mclk_change_block_cp_max = 0;
 
 	for (i = 0; i < adev->mode_info.num_crtc; i++) {
 		if (adev->pm.dpm.new_active_crtcs & (1 << i)) {
@@ -5824,26 +5824,31 @@ static int si_upload_smc_data(struct amdgpu_device *adev)
 		}
 	}
 
-	if (amdgpu_crtc == NULL)
-		return 0;
+	/* When a display is plugged in, program these so that the SMC
+	 * performs MCLK switching when it doesn't cause flickering.
+	 * When no display is plugged in, there is no need to restrict
+	 * MCLK switching, so program them to zero.
+	 */
+	if (adev->pm.dpm.new_active_crtc_count && amdgpu_crtc) {
+		crtc_index = amdgpu_crtc->crtc_id;
 
-	if (amdgpu_crtc->line_time <= 0)
-		return 0;
+		if (amdgpu_crtc->line_time) {
+			mclk_change_block_cp_min = amdgpu_crtc->wm_high / amdgpu_crtc->line_time;
+			mclk_change_block_cp_max = amdgpu_crtc->wm_low / amdgpu_crtc->line_time;
+		}
+	}
 
-	if (si_write_smc_soft_register(adev,
-				       SI_SMC_SOFT_REGISTER_crtc_index,
-				       amdgpu_crtc->crtc_id) != PPSMC_Result_OK)
-		return 0;
+	si_write_smc_soft_register(adev,
+		SI_SMC_SOFT_REGISTER_crtc_index,
+		crtc_index);
 
-	if (si_write_smc_soft_register(adev,
-				       SI_SMC_SOFT_REGISTER_mclk_change_block_cp_min,
-				       amdgpu_crtc->wm_high / amdgpu_crtc->line_time) != PPSMC_Result_OK)
-		return 0;
+	si_write_smc_soft_register(adev,
+		SI_SMC_SOFT_REGISTER_mclk_change_block_cp_min,
+		mclk_change_block_cp_min);
 
-	if (si_write_smc_soft_register(adev,
-				       SI_SMC_SOFT_REGISTER_mclk_change_block_cp_max,
-				       amdgpu_crtc->wm_low / amdgpu_crtc->line_time) != PPSMC_Result_OK)
-		return 0;
+	si_write_smc_soft_register(adev,
+		SI_SMC_SOFT_REGISTER_mclk_change_block_cp_max,
+		mclk_change_block_cp_max);
 
 	return 0;
 }
Signature · 1 line
2.51.0
#5 lore raw

PATCH 5/10 drm/amd/pm: Adjust si_upload_smc_data register programming (v3)

In reply to Timur Kristóf

Based on some comments in dm_pp_display_configuration above the crtc_index and line_time fields, these values are programmed to the SMC to work around an SMC hang when it switches MCLK.

According to Alex, the Windows driver programs them to:
mclk_change_block_cp_min = 200 / line_time mclk_change_block_cp_max = 100 / line_time Let's use the same for the sake of consistency.

Previously we used the watermark values, but it seemed buggy as the code was mixing up low/high and A/B watermarks, and was not saving a low watermark value on DCE 6, so mclk_change_block_cp_max would be always zero previously.

Split this change off from the previous si_upload_smc_data to make it easier to bisect, in case it causes any issues.

Fixes 841686df9f7d ("drm/amdgpu: add SI DPM support (v4)")
---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +2 -2
index 6736c592dfdc..fb008c5980d6 100644
@@ -5833,8 +5833,8 @@ static int si_upload_smc_data(struct amdgpu_device *adev)
 		crtc_index = amdgpu_crtc->crtc_id;
 
 		if (amdgpu_crtc->line_time) {
-			mclk_change_block_cp_min = amdgpu_crtc->wm_high / amdgpu_crtc->line_time;
-			mclk_change_block_cp_max = amdgpu_crtc->wm_low / amdgpu_crtc->line_time;
+			mclk_change_block_cp_min = 200 / amdgpu_crtc->line_time;
+			mclk_change_block_cp_max = 100 / amdgpu_crtc->line_time;
 		}
 	}
 
Signature · 1 line
2.51.0
#6 lore raw

PATCH 6/10 drm/amd/pm: Treat zero vblank time as too short in si_dpm (v3)

In reply to Timur Kristóf

Some parts of the code base expect that MCLK switching is turned off when the vblank time is set to zero.

According to pp_pm_compute_clocks the non-DC code has issues with MCLK switching with refresh rates over 120 Hz.

v3:
Add code comment to explain this better.
Add an if statement instead of changing the switch_limit.

Fixes 841686df9f7d ("drm/amdgpu: add SI DPM support (v4)")
---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +7 -1
index fb008c5980d6..c11c4cc111df 100644
@@ -3085,7 +3085,13 @@ static bool si_dpm_vblank_too_short(void *handle)
 	/* we never hit the non-gddr5 limit so disable it */
 	u32 switch_limit = adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5 ? 450 : 0;
 
-	if (vblank_time < switch_limit)
+	/* Consider zero vblank time too short and disable MCLK switching.
+	 * Note that the vblank time is set to maximum when no displays are attached,
+	 * so we'll still enable MCLK switching in that case.
+	 */
+	if (vblank_time == 0)
+		return true;
+	else if (vblank_time < switch_limit)
 		return true;
 	else
 		return false;
Signature · 1 line
2.51.0
#7 lore raw

PATCH 7/10 drm/amd/pm: Disable MCLK switching with non-DC at 120 Hz+ (v2)

In reply to Timur Kristóf

According to pp_pm_compute_clocks the non-DC display code has "issues with mclk switching with refresh rates over 120 hz". The workaround is to disable MCLK switching in this case.

Do the same for legacy DPM.

Fixes 6ddbd37f1074 ("drm/amd/pm: optimize the amdgpu_pm_compute_clocks() implementations")
---
 drivers/gpu/drm/amd/pm/amdgpu_dpm_internal.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/amd/pm/amdgpu_dpm_internal.c +7 -0
index 42efe838fa85..2d2d2d5e6763 100644
@@ -66,6 +66,13 @@ u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev)
 					(amdgpu_crtc->v_border * 2));
 
 				vblank_time_us = vblank_in_pixels * 1000 / amdgpu_crtc->hw_mode.clock;
+
+				/* we have issues with mclk switching with
+				 * refresh rates over 120 hz on the non-DC code.
+				 */
+				if (drm_mode_vrefresh(&amdgpu_crtc->hw_mode) > 120)
+					vblank_time_us = 0;
+
 				break;
 			}
 		}
Signature · 1 line
2.51.0
#8 lore raw

PATCH 8/10 drm/amd/pm: Disable SCLK switching on Oland with high pixel clocks (v3)

In reply to Timur Kristóf

Port of commit 227545b9a08c ("drm/radeon/dpm: Disable sclk switching on Oland when two 4K 60Hz monitors are connected")

This is an ad-hoc DPM fix, necessary because we don't have proper bandwidth calculation for DCE 6.

We define "high pixelclock" for SI as higher than necessary for 4K 30Hz. For example, 4K 60Hz and 1080p 144Hz fall into this category.

When two high pixel clock displays are connected to Oland, additionally disable shader clock switching, which results in a higher voltage, thereby addressing some visible flickering.

v2:
Add more comments.
v3:
Split into two commits for easier review.

Fixes 841686df9f7d ("drm/amdgpu: add SI DPM support (v4)")
---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 31 ++++++++++++++++++++++
 1 file changed, 31 insertions(+)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +31 -0
index c11c4cc111df..4236700fc1ad 100644
@@ -3449,12 +3449,14 @@ static void si_apply_state_adjust_rules(struct amdgpu_device *adev,
 {
 	struct  si_ps *ps = si_get_ps(rps);
 	struct amdgpu_clock_and_voltage_limits *max_limits;
+	struct amdgpu_connector *conn;
 	bool disable_mclk_switching = false;
 	bool disable_sclk_switching = false;
 	u32 mclk, sclk;
 	u16 vddc, vddci, min_vce_voltage = 0;
 	u32 max_sclk_vddc, max_mclk_vddci, max_mclk_vddc;
 	u32 max_sclk = 0, max_mclk = 0;
+	u32 high_pixelclock_count = 0;
 	int i;
 
 	if (adev->asic_type == CHIP_HAINAN) {
@@ -3482,6 +3484,35 @@ static void si_apply_state_adjust_rules(struct amdgpu_device *adev,
 		}
 	}
 
+	/* We define "high pixelclock" for SI as higher than necessary for 4K 30Hz.
+	 * For example, 4K 60Hz and 1080p 144Hz fall into this category.
+	 * Find number of such displays connected.
+	 */
+	for (i = 0; i < adev->mode_info.num_crtc; i++) {
+		if (!(adev->pm.dpm.new_active_crtcs & (1 << i)) ||
+			!adev->mode_info.crtcs[i]->enabled)
+			continue;
+
+		conn = to_amdgpu_connector(adev->mode_info.crtcs[i]->connector);
+
+		if (conn->pixelclock_for_modeset > 297000)
+			high_pixelclock_count++;
+	}
+
+	/* These are some ad-hoc fixes to some issues observed with SI GPUs.
+	 * They are necessary because we don't have something like dce_calcs
+	 * for these GPUs to calculate bandwidth requirements.
+	 */
+	if (high_pixelclock_count) {
+		/* On Oland, we observe some flickering when two 4K 60Hz
+		 * displays are connected, possibly because voltage is too low.
+		 * Raise the voltage by requiring a higher SCLK.
+		 * (Voltage cannot be adjusted independently without also SCLK.)
+		 */
+		if (high_pixelclock_count > 1 && adev->asic_type == CHIP_OLAND)
+			disable_sclk_switching = true;
+	}
+
 	if (rps->vce_active) {
 		rps->evclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].evclk;
 		rps->ecclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].ecclk;
Signature · 1 line
2.51.0
#9 lore raw

PATCH 9/10 drm/amd/pm: Remove wm_low and wm_high fields from amdgpu_crtc (v2)

In reply to Timur Kristóf

These fields were only used by si_dpm and are not necessary anymore. They also may have been incorrect because:
- wm_high was set to the LOW_WATERMARK field of watermark A.
- wm_low was not set on DCE 6 and was always zero.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h | 2 --
 drivers/gpu/drm/amd/amdgpu/dce_v10_0.c   | 3 +--
 drivers/gpu/drm/amd/amdgpu/dce_v11_0.c   | 3 +--
 drivers/gpu/drm/amd/amdgpu/dce_v6_0.c    | 1 -
 drivers/gpu/drm/amd/amdgpu/dce_v8_0.c    | 3 +--
 5 files changed, 3 insertions(+), 9 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +0 -2
index 6da4f946cac0..20460cfd09bc 100644
@@ -496,8 +496,6 @@ struct amdgpu_crtc {
 	struct drm_connector *connector;
 	/* for dpm */
 	u32 line_time;
-	u32 wm_low;
-	u32 wm_high;
 	u32 lb_vblank_lead_lines;
 	struct drm_display_mode hw_mode;
 	/* for virtual dce */
drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +1 -2
index bf7c22f81cda..c7b104222a97 100644
@@ -1141,8 +1141,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev,
 
 	/* save values for DPM */
 	amdgpu_crtc->line_time = line_time;
-	amdgpu_crtc->wm_high = latency_watermark_a;
-	amdgpu_crtc->wm_low = latency_watermark_b;
+
 	/* Save number of lines the linebuffer leads before the scanout */
 	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
drivers/gpu/drm/amd/amdgpu/dce_v11_0.c +1 -2
index 47e05783c4a0..fe8a4f70414a 100644
@@ -1173,8 +1173,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev,
 
 	/* save values for DPM */
 	amdgpu_crtc->line_time = line_time;
-	amdgpu_crtc->wm_high = latency_watermark_a;
-	amdgpu_crtc->wm_low = latency_watermark_b;
+
 	/* Save number of lines the linebuffer leads before the scanout */
 	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
drivers/gpu/drm/amd/amdgpu/dce_v6_0.c +0 -1
index 276c025c4c03..8ab1f08afcef 100644
@@ -1034,7 +1034,6 @@ static void dce_v6_0_program_watermarks(struct amdgpu_device *adev,
 
 	/* save values for DPM */
 	amdgpu_crtc->line_time = line_time;
-	amdgpu_crtc->wm_high = latency_watermark_a;
 
 	/* Save number of lines the linebuffer leads before the scanout */
 	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
drivers/gpu/drm/amd/amdgpu/dce_v8_0.c +1 -2
index e62ccf9eb73d..47ed5f1840fc 100644
@@ -1096,8 +1096,7 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev,
 
 	/* save values for DPM */
 	amdgpu_crtc->line_time = line_time;
-	amdgpu_crtc->wm_high = latency_watermark_a;
-	amdgpu_crtc->wm_low = latency_watermark_b;
+
 	/* Save number of lines the linebuffer leads before the scanout */
 	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
Signature · 1 line
2.51.0
#10 lore raw

PATCH 10/10 drm/amd/pm: Print VCE clocks too in si_dpm (v3)

In reply to Timur Kristóf

They are part of a power state too and should be printed alongside the rest of the data from the power state.

---
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +1 -0
index 4236700fc1ad..6595a611ce6e 100644
@@ -7992,6 +7992,7 @@ static void si_dpm_print_power_state(void *handle,
 	amdgpu_dpm_dbg_print_class_info(adev, rps->class, rps->class2);
 	amdgpu_dpm_dbg_print_cap_info(adev, rps->caps);
 	drm_dbg(adev_to_drm(adev), "\tuvd    vclk: %d dclk: %d\n", rps->vclk, rps->dclk);
+	drm_dbg(adev_to_drm(adev), "\tvce    evclk: %d ecclk: %d\n", rps->evclk, rps->ecclk);
 	for (i = 0; i < ps->performance_level_count; i++) {
 		pl = &ps->performance_levels[i];
 		drm_dbg(adev_to_drm(adev), "\t\tpower level %d    sclk: %u mclk: %u vddc: %u vddci: %u pcie gen: %u\n",
Signature · 1 line
2.51.0
Reviewed-by on the v1 series
#11 lore raw
In reply to Timur Kristóf
Quoted 28 lines · Timur Kristóf

This series fixes some power management issues on SI, addressing the feedback I got for the previous versions.

Additionally, the following improvements were made:

There is now a patch that increases the SMC timeout. This gets rid of some errors that happened when switching between power states. Apparently the SMC can sometimes take a long time.

Also removed one of the ad-hoc fixes from si_dpm because it didn't actually address anything, and edited that patch to focus on just solving the flickering on Oland.

And edited the si_upload_smc_data patch yet again to program the soft registers to zero when there are no displays connected.

Timur Kristóf (10):
drm/amdgpu: Power up UVD 3 for FW validation (v2) drm/amd/pm: Disable ULV even if unsupported (v3) drm/amd/pm: Increase SMC timeout on SI and warn (v3) drm/amd/pm: Fix si_upload_smc_data (v3)
drm/amd/pm: Adjust si_upload_smc_data register programming (v3) drm/amd/pm: Treat zero vblank time as too short in si_dpm (v3) drm/amd/pm: Disable MCLK switching with non-DC at 120 Hz+ (v2) drm/amd/pm: Disable SCLK switching on Oland with high pixel clocks

Quoted 3 lines
    (v3)

drm/amd/pm: Remove wm_low and wm_high fields from amdgpu_crtc (v2) drm/amd/pm: Print VCE clocks too in si_dpm (v3)

Series is:

And applied.

Thanks!

Alex

Quoted 15 lines
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h     |  2 -
 drivers/gpu/drm/amd/amdgpu/dce_v10_0.c       |  3 +-
 drivers/gpu/drm/amd/amdgpu/dce_v11_0.c       |  3 +-
 drivers/gpu/drm/amd/amdgpu/dce_v6_0.c        |  1 -
 drivers/gpu/drm/amd/amdgpu/dce_v8_0.c        |  3 +-
 drivers/gpu/drm/amd/amdgpu/uvd_v3_1.c        | 29 +++++-
 drivers/gpu/drm/amd/pm/amdgpu_dpm_internal.c |  7 ++
 drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c   | 93 ++++++++++++++------
 drivers/gpu/drm/amd/pm/legacy-dpm/si_smc.c   | 26 +++++-
 9 files changed, 125 insertions(+), 42 deletions(-)

--
2.51.0