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

[Alex Deucher]

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

Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/vi.c | 117 +++++++++++++++++++++++++++++++++-------
 1 file changed, 97 insertions(+), 20 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c
index 657de2a..82f72cd 100644
--- a/drivers/gpu/drm/amd/amdgpu/vi.c
+++ b/drivers/gpu/drm/amd/amdgpu/vi.c
@@ -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;
-- 
2.5.5