Merge tag 'amd-drm-next-6.4-2023-04-14' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-6.4-2023-04-14:

amdgpu:
- S4 fixes for APUs
- GFX11 fixes
- Misc code cleanups
- DCN 3.2 fixes
- DCN 3.1.4 fixes
- FPO/FAMS work to improve display power savings
- DP fixes
- UMC 8.10 code cleanup
- SDMA v4 fix
- GPU clock counter fixes
- SMU 13 fixes
- Sdma v6 invalidation fix for preemption
- RAS fixes
- S0ix fix
- GC 9.4.3 updates

amdkfd:
- Fix user pointers with IOMMU
- Fix coherency flag handling

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230414204609.7942-1-alexander.deucher@amd.com

5 files changed, 87 insertions, 21 deletions

diff --git a/drivers/gpu/drm/amd/pm/amdgpu_pm.c b/drivers/gpu/drm/amd/pm/amdgpu_pm.c
index e011041e3ec6..58c2246918fd 100644
--- a/drivers/gpu/drm/amd/pm/amdgpu_pm.c
+++ b/drivers/gpu/drm/amd/pm/amdgpu_pm.c
@@ -3395,7 +3395,7 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj,
 	      attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr)) /* can't manage state */
 		effective_mode &= ~S_IWUSR;
 
-	/* not implemented yet for GC 10.3.1 APUs */
+	/* In the case of APUs, this is only implemented on Vangogh */
 	if (((adev->family == AMDGPU_FAMILY_SI) ||
 	     ((adev->flags & AMD_IS_APU) && (gc_ver != IP_VERSION(10, 3, 1)))) &&
 	    (attr == &sensor_dev_attr_power1_cap_max.dev_attr.attr ||
@@ -3404,7 +3404,7 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj,
 	     attr == &sensor_dev_attr_power1_cap_default.dev_attr.attr))
 		return 0;
 
-	/* not implemented yet for APUs having <= GC 9.3.0 */
+	/* not implemented yet for APUs having < GC 9.3.0 (Renoir) */
 	if (((adev->family == AMDGPU_FAMILY_SI) ||
 	     ((adev->flags & AMD_IS_APU) && (gc_ver < IP_VERSION(9, 3, 0)))) &&
 	    (attr == &sensor_dev_attr_power1_average.dev_attr.attr))
diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 056ac2b512eb..5633c5797e85 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -1712,8 +1712,6 @@ static int smu_display_configuration_change(void *handle,
 					    const struct amd_pp_display_configuration *display_config)
 {
 	struct smu_context *smu = handle;
-	int index = 0;
-	int num_of_active_display = 0;
 
 	if (!smu->pm_enabled || !smu->adev->pm.dpm_enabled)
 		return -EOPNOTSUPP;
@@ -1724,11 +1722,6 @@ static int smu_display_configuration_change(void *handle,
 	smu_set_min_dcef_deep_sleep(smu,
 				    display_config->min_dcef_deep_sleep_set_clk / 100);
 
-	for (index = 0; index < display_config->num_path_including_non_display; index++) {
-		if (display_config->displays[index].controller_id != 0)
-			num_of_active_display++;
-	}
-
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
index 0ef37837b164..7944ce80e5c3 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
+++ b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
@@ -62,6 +62,12 @@
 #define CTF_OFFSET_HOTSPOT		5
 #define CTF_OFFSET_MEM			5
 
+static const int pmfw_decoded_link_speed[5] = {1, 2, 3, 4, 5};
+static const int pmfw_decoded_link_width[7] = {0, 1, 2, 4, 8, 12, 16};
+
+#define DECODE_GEN_SPEED(gen_speed_idx)		(pmfw_decoded_link_speed[gen_speed_idx])
+#define DECODE_LANE_WIDTH(lane_width_idx)	(pmfw_decoded_link_width[lane_width_idx])
+
 struct smu_13_0_max_sustainable_clocks {
 	uint32_t display_clock;
 	uint32_t phy_clock;
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
index e9766fe5656e..09405ef1e3c8 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
@@ -1144,8 +1144,8 @@ static int smu_v13_0_0_print_clk_levels(struct smu_context *smu,
 					(pcie_table->pcie_lane[i] == 5) ? "x12" :
 					(pcie_table->pcie_lane[i] == 6) ? "x16" : "",
 					pcie_table->clk_freq[i],
-					((gen_speed - 1) == pcie_table->pcie_gen[i]) &&
-					(lane_width == link_width[pcie_table->pcie_lane[i]]) ?
+					(gen_speed == DECODE_GEN_SPEED(pcie_table->pcie_gen[i])) &&
+					(lane_width == DECODE_LANE_WIDTH(link_width[pcie_table->pcie_lane[i]])) ?
 					"*" : "");
 		break;
 
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
index 1b2c82449f20..3d9ff46706fb 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
@@ -575,6 +575,14 @@ static int smu_v13_0_7_set_default_dpm_table(struct smu_context *smu)
 						     dpm_table);
 		if (ret)
 			return ret;
+
+		if (skutable->DriverReportedClocks.GameClockAc &&
+			(dpm_table->dpm_levels[dpm_table->count - 1].value >
+			skutable->DriverReportedClocks.GameClockAc)) {
+			dpm_table->dpm_levels[dpm_table->count - 1].value =
+				skutable->DriverReportedClocks.GameClockAc;
+			dpm_table->max = skutable->DriverReportedClocks.GameClockAc;
+		}
 	} else {
 		dpm_table->count = 1;
 		dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.gfxclk / 100;
@@ -828,6 +836,57 @@ static int smu_v13_0_7_get_smu_metrics_data(struct smu_context *smu,
 	return ret;
 }
 
+static int smu_v13_0_7_get_dpm_ultimate_freq(struct smu_context *smu,
+					     enum smu_clk_type clk_type,
+					     uint32_t *min,
+					     uint32_t *max)
+{
+	struct smu_13_0_dpm_context *dpm_context =
+		smu->smu_dpm.dpm_context;
+	struct smu_13_0_dpm_table *dpm_table;
+
+	switch (clk_type) {
+	case SMU_MCLK:
+	case SMU_UCLK:
+		/* uclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.uclk_table;
+		break;
+	case SMU_GFXCLK:
+	case SMU_SCLK:
+		/* gfxclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.gfx_table;
+		break;
+	case SMU_SOCCLK:
+		/* socclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.soc_table;
+		break;
+	case SMU_FCLK:
+		/* fclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.fclk_table;
+		break;
+	case SMU_VCLK:
+	case SMU_VCLK1:
+		/* vclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.vclk_table;
+		break;
+	case SMU_DCLK:
+	case SMU_DCLK1:
+		/* dclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.dclk_table;
+		break;
+	default:
+		dev_err(smu->adev->dev, "Unsupported clock type!\n");
+		return -EINVAL;
+	}
+
+	if (min)
+		*min = dpm_table->min;
+	if (max)
+		*max = dpm_table->max;
+
+	return 0;
+}
+
 static int smu_v13_0_7_read_sensor(struct smu_context *smu,
 				   enum amd_pp_sensors sensor,
 				   void *data,
@@ -1074,8 +1133,8 @@ static int smu_v13_0_7_print_clk_levels(struct smu_context *smu,
 					(pcie_table->pcie_lane[i] == 5) ? "x12" :
 					(pcie_table->pcie_lane[i] == 6) ? "x16" : "",
 					pcie_table->clk_freq[i],
-					(gen_speed == pcie_table->pcie_gen[i]) &&
-					(lane_width == pcie_table->pcie_lane[i]) ?
+					(gen_speed == DECODE_GEN_SPEED(pcie_table->pcie_gen[i])) &&
+					(lane_width == DECODE_LANE_WIDTH(pcie_table->pcie_lane[i])) ?
 					"*" : "");
 		break;
 
@@ -1329,9 +1388,17 @@ static int smu_v13_0_7_populate_umd_state_clk(struct smu_context *smu)
 				&dpm_context->dpm_tables.fclk_table;
 	struct smu_umd_pstate_table *pstate_table =
 				&smu->pstate_table;
+	struct smu_table_context *table_context = &smu->smu_table;
+	PPTable_t *pptable = table_context->driver_pptable;
+	DriverReportedClocks_t driver_clocks =
+		pptable->SkuTable.DriverReportedClocks;
 
 	pstate_table->gfxclk_pstate.min = gfx_table->min;
-	pstate_table->gfxclk_pstate.peak = gfx_table->max;
+	if (driver_clocks.GameClockAc &&
+		(driver_clocks.GameClockAc < gfx_table->max))
+		pstate_table->gfxclk_pstate.peak = driver_clocks.GameClockAc;
+	else
+		pstate_table->gfxclk_pstate.peak = gfx_table->max;
 
 	pstate_table->uclk_pstate.min = mem_table->min;
 	pstate_table->uclk_pstate.peak = mem_table->max;
@@ -1348,12 +1415,12 @@ static int smu_v13_0_7_populate_umd_state_clk(struct smu_context *smu)
 	pstate_table->fclk_pstate.min = fclk_table->min;
 	pstate_table->fclk_pstate.peak = fclk_table->max;
 
-	/*
-	 * For now, just use the mininum clock frequency.
-	 * TODO: update them when the real pstate settings available
-	 */
-	pstate_table->gfxclk_pstate.standard = gfx_table->min;
-	pstate_table->uclk_pstate.standard = mem_table->min;
+	if (driver_clocks.BaseClockAc &&
+		driver_clocks.BaseClockAc < gfx_table->max)
+		pstate_table->gfxclk_pstate.standard = driver_clocks.BaseClockAc;
+	else
+		pstate_table->gfxclk_pstate.standard = gfx_table->max;
+	pstate_table->uclk_pstate.standard = mem_table->max;
 	pstate_table->socclk_pstate.standard = soc_table->min;
 	pstate_table->vclk_pstate.standard = vclk_table->min;
 	pstate_table->dclk_pstate.standard = dclk_table->min;
@@ -1678,7 +1745,7 @@ static const struct pptable_funcs smu_v13_0_7_ppt_funcs = {
 	.dpm_set_jpeg_enable = smu_v13_0_set_jpeg_enable,
 	.init_pptable_microcode = smu_v13_0_init_pptable_microcode,
 	.populate_umd_state_clk = smu_v13_0_7_populate_umd_state_clk,
-	.get_dpm_ultimate_freq = smu_v13_0_get_dpm_ultimate_freq,
+	.get_dpm_ultimate_freq = smu_v13_0_7_get_dpm_ultimate_freq,
 	.get_vbios_bootup_values = smu_v13_0_get_vbios_bootup_values,
 	.read_sensor = smu_v13_0_7_read_sensor,
 	.feature_is_enabled = smu_cmn_feature_is_enabled,