Skip to contentSkip to conversation

drm/amdgpu: add additional cached gca config variables

PATCH In review Alex Deucher 2 participants 5 messages Oct 14, 2016 – Oct 17, 2016 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/4 drm/amdgpu: add additional cached gca config variables

We need to cache some additional values to handle SR-IOV and PG.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h | 12 ++++++++++++
 1 file changed, 12 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +12 -0
index 7d30a8e..efbd9ef 100644
@@ -770,6 +770,16 @@ struct amdgpu_scratch {
 /*
  * GFX configurations
  */
+#define AMDGPU_GFX_MAX_SE 4
+#define AMDGPU_GFX_MAX_SH_PER_SE 2
+
+struct amdgpu_rb_config {
+	uint32_t rb_backend_disable;
+	uint32_t user_rb_backend_disable;
+	uint32_t raster_config;
+	uint32_t raster_config_1;
+};
+
 struct amdgpu_gca_config {
 	unsigned max_shader_engines;
 	unsigned max_tile_pipes;
@@ -798,6 +808,8 @@ struct amdgpu_gca_config {
 
 	uint32_t tile_mode_array[32];
 	uint32_t macrotile_mode_array[16];
+
+	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
 };
 
 struct amdgpu_cu_info {
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1476460893-15258-1-git-send-email-alexander.deucher-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 2/4 drm/amdgpu/gfx8: cache rb config values

Needed when for SR-IOV and when PG is enabled.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +15 -0
index ba36db8..e066441 100644
@@ -3694,6 +3694,21 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
 							num_rb_pipes);
 	}
 
+	/* cache the values for userspace */
+	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
+		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
+			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
+			adev->gfx.config.rb_config[i][j].rb_backend_disable =
+				RREG32(mmCC_RB_BACKEND_DISABLE);
+			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
+				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
+			adev->gfx.config.rb_config[i][j].raster_config =
+				RREG32(mmPA_SC_RASTER_CONFIG);
+			adev->gfx.config.rb_config[i][j].raster_config_1 =
+				RREG32(mmPA_SC_RASTER_CONFIG_1);
+		}
+	}
+	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
 	mutex_unlock(&adev->grbm_idx_mutex);
 }
 
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

PATCH 3/4 drm/amdgpu/gfx8: use cached raster config values in csb setup

In reply to a message not in the archive

Simplify the code and properly set the csb for harvest values.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 30 ++----------------------------
 1 file changed, 2 insertions(+), 28 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +2 -28
index e066441..8dc8f57 100644
@@ -1140,34 +1140,8 @@ static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
 			PACKET3_SET_CONTEXT_REG_START);
-	switch (adev->asic_type) {
-	case CHIP_TONGA:
-	case CHIP_POLARIS10:
-		buffer[count++] = cpu_to_le32(0x16000012);
-		buffer[count++] = cpu_to_le32(0x0000002A);
-		break;
-	case CHIP_POLARIS11:
-		buffer[count++] = cpu_to_le32(0x16000012);
-		buffer[count++] = cpu_to_le32(0x00000000);
-		break;
-	case CHIP_FIJI:
-		buffer[count++] = cpu_to_le32(0x3a00161a);
-		buffer[count++] = cpu_to_le32(0x0000002e);
-		break;
-	case CHIP_TOPAZ:
-	case CHIP_CARRIZO:
-		buffer[count++] = cpu_to_le32(0x00000002);
-		buffer[count++] = cpu_to_le32(0x00000000);
-		break;
-	case CHIP_STONEY:
-		buffer[count++] = cpu_to_le32(0x00000000);
-		buffer[count++] = cpu_to_le32(0x00000000);
-		break;
-	default:
-		buffer[count++] = cpu_to_le32(0x00000000);
-		buffer[count++] = cpu_to_le32(0x00000000);
-		break;
-	}
+	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
+	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
 
 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#4 lore raw

PATCH 4/4 drm/amdgpu: used cached gca values for vi_read_register (v2)

In reply to a message not in the archive

Using the cached values has less latency for bare metal and SR-IOV, and prevents reading back bogus values if the engine is powergated.

v2: fix typo in tile idx calculation

---
 drivers/gpu/drm/amd/amdgpu/vi.c | 117 +++++++++++++++++++++++++++++++++-------
 1 file changed, 97 insertions(+), 20 deletions(-)
drivers/gpu/drm/amd/amdgpu/vi.c +97 -20
index 657de2a..82f72cd 100644
@@ -556,21 +556,100 @@ static const struct amdgpu_allowed_register_entry vi_allowed_read_registers[] =
 	{mmPA_SC_RASTER_CONFIG_1, false, true},
 };
 
-static uint32_t vi_read_indexed_register(struct amdgpu_device *adev, u32 se_num,
-					 u32 sh_num, u32 reg_offset)
-{
-	uint32_t val;
+static uint32_t vi_get_register_value(struct amdgpu_device *adev,
+				      bool indexed, u32 se_num,
+				      u32 sh_num, u32 reg_offset)
+{
+	if (indexed) {
+		uint32_t val;
+		unsigned se_idx = (se_num == 0xffffffff) ? 0 : se_num;
+		unsigned sh_idx = (sh_num == 0xffffffff) ? 0 : sh_num;
+
+		switch (reg_offset) {
+		case mmCC_RB_BACKEND_DISABLE:
+			return adev->gfx.config.rb_config[se_idx][sh_idx].rb_backend_disable;
+		case mmGC_USER_RB_BACKEND_DISABLE:
+			return adev->gfx.config.rb_config[se_idx][sh_idx].user_rb_backend_disable;
+		case mmPA_SC_RASTER_CONFIG:
+			return adev->gfx.config.rb_config[se_idx][sh_idx].raster_config;
+		case mmPA_SC_RASTER_CONFIG_1:
+			return adev->gfx.config.rb_config[se_idx][sh_idx].raster_config_1;
+		}
 
-	mutex_lock(&adev->grbm_idx_mutex);
-	if (se_num != 0xffffffff || sh_num != 0xffffffff)
-		amdgpu_gfx_select_se_sh(adev, se_num, sh_num, 0xffffffff);
+		mutex_lock(&adev->grbm_idx_mutex);
+		if (se_num != 0xffffffff || sh_num != 0xffffffff)
+			amdgpu_gfx_select_se_sh(adev, se_num, sh_num, 0xffffffff);
 
-	val = RREG32(reg_offset);
+		val = RREG32(reg_offset);
 
-	if (se_num != 0xffffffff || sh_num != 0xffffffff)
-		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-	mutex_unlock(&adev->grbm_idx_mutex);
-	return val;
+		if (se_num != 0xffffffff || sh_num != 0xffffffff)
+			amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+		mutex_unlock(&adev->grbm_idx_mutex);
+		return val;
+	} else {
+		unsigned idx;
+
+		switch (reg_offset) {
+		case mmGB_ADDR_CONFIG:
+			return adev->gfx.config.gb_addr_config;
+		case mmMC_ARB_RAMCFG:
+			return adev->gfx.config.mc_arb_ramcfg;
+		case mmGB_TILE_MODE0:
+		case mmGB_TILE_MODE1:
+		case mmGB_TILE_MODE2:
+		case mmGB_TILE_MODE3:
+		case mmGB_TILE_MODE4:
+		case mmGB_TILE_MODE5:
+		case mmGB_TILE_MODE6:
+		case mmGB_TILE_MODE7:
+		case mmGB_TILE_MODE8:
+		case mmGB_TILE_MODE9:
+		case mmGB_TILE_MODE10:
+		case mmGB_TILE_MODE11:
+		case mmGB_TILE_MODE12:
+		case mmGB_TILE_MODE13:
+		case mmGB_TILE_MODE14:
+		case mmGB_TILE_MODE15:
+		case mmGB_TILE_MODE16:
+		case mmGB_TILE_MODE17:
+		case mmGB_TILE_MODE18:
+		case mmGB_TILE_MODE19:
+		case mmGB_TILE_MODE20:
+		case mmGB_TILE_MODE21:
+		case mmGB_TILE_MODE22:
+		case mmGB_TILE_MODE23:
+		case mmGB_TILE_MODE24:
+		case mmGB_TILE_MODE25:
+		case mmGB_TILE_MODE26:
+		case mmGB_TILE_MODE27:
+		case mmGB_TILE_MODE28:
+		case mmGB_TILE_MODE29:
+		case mmGB_TILE_MODE30:
+		case mmGB_TILE_MODE31:
+			idx = (reg_offset - mmGB_TILE_MODE0);
+			return adev->gfx.config.tile_mode_array[idx];
+		case mmGB_MACROTILE_MODE0:
+		case mmGB_MACROTILE_MODE1:
+		case mmGB_MACROTILE_MODE2:
+		case mmGB_MACROTILE_MODE3:
+		case mmGB_MACROTILE_MODE4:
+		case mmGB_MACROTILE_MODE5:
+		case mmGB_MACROTILE_MODE6:
+		case mmGB_MACROTILE_MODE7:
+		case mmGB_MACROTILE_MODE8:
+		case mmGB_MACROTILE_MODE9:
+		case mmGB_MACROTILE_MODE10:
+		case mmGB_MACROTILE_MODE11:
+		case mmGB_MACROTILE_MODE12:
+		case mmGB_MACROTILE_MODE13:
+		case mmGB_MACROTILE_MODE14:
+		case mmGB_MACROTILE_MODE15:
+			idx = (reg_offset - mmGB_MACROTILE_MODE0);
+			return adev->gfx.config.macrotile_mode_array[idx];
+		default:
+			return RREG32(reg_offset);
+		}
+	}
 }
 
 static int vi_read_register(struct amdgpu_device *adev, u32 se_num,
@@ -605,10 +684,9 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num,
 			if (reg_offset != asic_register_entry->reg_offset)
 				continue;
 			if (!asic_register_entry->untouched)
-				*value = asic_register_entry->grbm_indexed ?
-					vi_read_indexed_register(adev, se_num,
-								 sh_num, reg_offset) :
-					RREG32(reg_offset);
+				*value = vi_get_register_value(adev,
+							       asic_register_entry->grbm_indexed,
+							       se_num, sh_num, reg_offset);
 			return 0;
 		}
 	}
@@ -618,10 +696,9 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num,
 			continue;
 
 		if (!vi_allowed_read_registers[i].untouched)
-			*value = vi_allowed_read_registers[i].grbm_indexed ?
-				vi_read_indexed_register(adev, se_num,
-							 sh_num, reg_offset) :
-				RREG32(reg_offset);
+			*value = vi_get_register_value(adev,
+						       vi_allowed_read_registers[i].grbm_indexed,
+						       se_num, sh_num, reg_offset);
 		return 0;
 	}
 	return -EINVAL;
Signature · 6 lines
2.5.5

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Reviewed-by
#5 lore raw
In reply to a message not in the archive

Am 14.10.2016 um 18:01 schrieb Alex Deucher:

Quoted 4 lines

We need to cache some additional values to handle SR-IOV and PG.

Quoted 26 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h | 12 ++++++++++++
  1 file changed, 12 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +10 -0
index 7d30a8e..efbd9ef 100644
@@ -770,6 +770,16 @@ struct amdgpu_scratch {
  /*
   * GFX configurations
   */
+#define AMDGPU_GFX_MAX_SE 4
+#define AMDGPU_GFX_MAX_SH_PER_SE 2
+
+struct amdgpu_rb_config {
+	uint32_t rb_backend_disable;
+	uint32_t user_rb_backend_disable;
+	uint32_t raster_config;
+	uint32_t raster_config_1;
+};
+
  struct amdgpu_gca_config {
  	unsigned max_shader_engines;
  	unsigned max_tile_pipes;

drivers/gpu/drm/amd/amdgpu/amdgpu.h struct amdgpu_gca_config { · line 815

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
@@ -798,6 +808,8 @@ struct amdgpu_gca_config {
 
  	uint32_t tile_mode_array[32];
  	uint32_t macrotile_mode_array[16];
+
+	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
  };
 
  struct amdgpu_cu_info {

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx