1 /* 2 * Copyright 2014 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include <linux/delay.h> 25 #include <linux/kernel.h> 26 #include <linux/firmware.h> 27 #include <linux/module.h> 28 #include <linux/pci.h> 29 30 #include "amdgpu.h" 31 #include "amdgpu_gfx.h" 32 #include "vi.h" 33 #include "vi_structs.h" 34 #include "vid.h" 35 #include "amdgpu_ucode.h" 36 #include "amdgpu_atombios.h" 37 #include "atombios_i2c.h" 38 #include "clearstate_vi.h" 39 40 #include "gmc/gmc_8_2_d.h" 41 #include "gmc/gmc_8_2_sh_mask.h" 42 43 #include "oss/oss_3_0_d.h" 44 #include "oss/oss_3_0_sh_mask.h" 45 46 #include "bif/bif_5_0_d.h" 47 #include "bif/bif_5_0_sh_mask.h" 48 #include "gca/gfx_8_0_d.h" 49 #include "gca/gfx_8_0_enum.h" 50 #include "gca/gfx_8_0_sh_mask.h" 51 52 #include "dce/dce_10_0_d.h" 53 #include "dce/dce_10_0_sh_mask.h" 54 55 #include "smu/smu_7_1_3_d.h" 56 57 #include "ivsrcid/ivsrcid_vislands30.h" 58 59 #define GFX8_NUM_GFX_RINGS 1 60 #define GFX8_MEC_HPD_SIZE 4096 61 62 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001 63 #define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001 64 #define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002 65 #define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003 66 67 #define ARRAY_MODE(x) ((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT) 68 #define PIPE_CONFIG(x) ((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT) 69 #define TILE_SPLIT(x) ((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT) 70 #define MICRO_TILE_MODE_NEW(x) ((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT) 71 #define SAMPLE_SPLIT(x) ((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT) 72 #define BANK_WIDTH(x) ((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT) 73 #define BANK_HEIGHT(x) ((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT) 74 #define MACRO_TILE_ASPECT(x) ((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT) 75 #define NUM_BANKS(x) ((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT) 76 77 #define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK 0x00000001L 78 #define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK 0x00000002L 79 #define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK 0x00000004L 80 #define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK 0x00000008L 81 #define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK 0x00000010L 82 #define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK 0x00000020L 83 84 /* BPM SERDES CMD */ 85 #define SET_BPM_SERDES_CMD 1 86 #define CLE_BPM_SERDES_CMD 0 87 88 /* BPM Register Address*/ 89 enum { 90 BPM_REG_CGLS_EN = 0, /* Enable/Disable CGLS */ 91 BPM_REG_CGLS_ON, /* ON/OFF CGLS: shall be controlled by RLC FW */ 92 BPM_REG_CGCG_OVERRIDE, /* Set/Clear CGCG Override */ 93 BPM_REG_MGCG_OVERRIDE, /* Set/Clear MGCG Override */ 94 BPM_REG_FGCG_OVERRIDE, /* Set/Clear FGCG Override */ 95 BPM_REG_FGCG_MAX 96 }; 97 98 #define RLC_FormatDirectRegListLength 14 99 100 MODULE_FIRMWARE("amdgpu/carrizo_ce.bin"); 101 MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin"); 102 MODULE_FIRMWARE("amdgpu/carrizo_me.bin"); 103 MODULE_FIRMWARE("amdgpu/carrizo_mec.bin"); 104 MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin"); 105 MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin"); 106 107 MODULE_FIRMWARE("amdgpu/stoney_ce.bin"); 108 MODULE_FIRMWARE("amdgpu/stoney_pfp.bin"); 109 MODULE_FIRMWARE("amdgpu/stoney_me.bin"); 110 MODULE_FIRMWARE("amdgpu/stoney_mec.bin"); 111 MODULE_FIRMWARE("amdgpu/stoney_rlc.bin"); 112 113 MODULE_FIRMWARE("amdgpu/tonga_ce.bin"); 114 MODULE_FIRMWARE("amdgpu/tonga_pfp.bin"); 115 MODULE_FIRMWARE("amdgpu/tonga_me.bin"); 116 MODULE_FIRMWARE("amdgpu/tonga_mec.bin"); 117 MODULE_FIRMWARE("amdgpu/tonga_mec2.bin"); 118 MODULE_FIRMWARE("amdgpu/tonga_rlc.bin"); 119 120 MODULE_FIRMWARE("amdgpu/topaz_ce.bin"); 121 MODULE_FIRMWARE("amdgpu/topaz_pfp.bin"); 122 MODULE_FIRMWARE("amdgpu/topaz_me.bin"); 123 MODULE_FIRMWARE("amdgpu/topaz_mec.bin"); 124 MODULE_FIRMWARE("amdgpu/topaz_rlc.bin"); 125 126 MODULE_FIRMWARE("amdgpu/fiji_ce.bin"); 127 MODULE_FIRMWARE("amdgpu/fiji_pfp.bin"); 128 MODULE_FIRMWARE("amdgpu/fiji_me.bin"); 129 MODULE_FIRMWARE("amdgpu/fiji_mec.bin"); 130 MODULE_FIRMWARE("amdgpu/fiji_mec2.bin"); 131 MODULE_FIRMWARE("amdgpu/fiji_rlc.bin"); 132 133 MODULE_FIRMWARE("amdgpu/polaris10_ce.bin"); 134 MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin"); 135 MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin"); 136 MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin"); 137 MODULE_FIRMWARE("amdgpu/polaris10_me.bin"); 138 MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin"); 139 MODULE_FIRMWARE("amdgpu/polaris10_mec.bin"); 140 MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin"); 141 MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin"); 142 MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin"); 143 MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin"); 144 145 MODULE_FIRMWARE("amdgpu/polaris11_ce.bin"); 146 MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin"); 147 MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin"); 148 MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin"); 149 MODULE_FIRMWARE("amdgpu/polaris11_me.bin"); 150 MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin"); 151 MODULE_FIRMWARE("amdgpu/polaris11_mec.bin"); 152 MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin"); 153 MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin"); 154 MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin"); 155 MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin"); 156 157 MODULE_FIRMWARE("amdgpu/polaris12_ce.bin"); 158 MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin"); 159 MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin"); 160 MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin"); 161 MODULE_FIRMWARE("amdgpu/polaris12_me.bin"); 162 MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin"); 163 MODULE_FIRMWARE("amdgpu/polaris12_mec.bin"); 164 MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin"); 165 MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin"); 166 MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin"); 167 MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin"); 168 169 MODULE_FIRMWARE("amdgpu/vegam_ce.bin"); 170 MODULE_FIRMWARE("amdgpu/vegam_pfp.bin"); 171 MODULE_FIRMWARE("amdgpu/vegam_me.bin"); 172 MODULE_FIRMWARE("amdgpu/vegam_mec.bin"); 173 MODULE_FIRMWARE("amdgpu/vegam_mec2.bin"); 174 MODULE_FIRMWARE("amdgpu/vegam_rlc.bin"); 175 176 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] = 177 { 178 {mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0}, 179 {mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1}, 180 {mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2}, 181 {mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3}, 182 {mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4}, 183 {mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5}, 184 {mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6}, 185 {mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7}, 186 {mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8}, 187 {mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9}, 188 {mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10}, 189 {mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11}, 190 {mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12}, 191 {mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13}, 192 {mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14}, 193 {mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15} 194 }; 195 196 static const u32 golden_settings_tonga_a11[] = 197 { 198 mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208, 199 mmCB_HW_CONTROL_3, 0x00000040, 0x00000040, 200 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 201 mmGB_GPU_ID, 0x0000000f, 0x00000000, 202 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 203 mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc, 204 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 205 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c, 206 mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd, 207 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 208 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 209 mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, 210 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb, 211 mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b, 212 mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876, 213 mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, 214 }; 215 216 static const u32 tonga_golden_common_all[] = 217 { 218 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 219 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012, 220 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A, 221 mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003, 222 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 223 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 224 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 225 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF 226 }; 227 228 static const u32 tonga_mgcg_cgcg_init[] = 229 { 230 mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff, 231 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 232 mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 233 mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100, 234 mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100, 235 mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100, 236 mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100, 237 mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100, 238 mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100, 239 mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100, 240 mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100, 241 mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100, 242 mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100, 243 mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100, 244 mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100, 245 mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100, 246 mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100, 247 mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100, 248 mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100, 249 mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100, 250 mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100, 251 mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100, 252 mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100, 253 mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100, 254 mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100, 255 mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100, 256 mmTA_CGTT_CTRL, 0xffffffff, 0x00000100, 257 mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 258 mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 259 mmTD_CGTT_CTRL, 0xffffffff, 0x00000100, 260 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 261 mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000, 262 mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 263 mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007, 264 mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005, 265 mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 266 mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000, 267 mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 268 mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007, 269 mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005, 270 mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 271 mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000, 272 mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 273 mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007, 274 mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005, 275 mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 276 mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000, 277 mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 278 mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007, 279 mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005, 280 mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 281 mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000, 282 mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 283 mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007, 284 mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005, 285 mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 286 mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000, 287 mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 288 mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007, 289 mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005, 290 mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 291 mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000, 292 mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 293 mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007, 294 mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005, 295 mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 296 mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000, 297 mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 298 mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007, 299 mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005, 300 mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 301 mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200, 302 mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100, 303 mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c, 304 mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001, 305 }; 306 307 static const u32 golden_settings_vegam_a11[] = 308 { 309 mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208, 310 mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000, 311 mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040, 312 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 313 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 314 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 315 mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a, 316 mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e, 317 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, 318 mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c, 319 mmSQ_CONFIG, 0x07f80000, 0x01180000, 320 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 321 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 322 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7, 323 mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000, 324 mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054, 325 mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, 326 }; 327 328 static const u32 vegam_golden_common_all[] = 329 { 330 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 331 mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003, 332 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 333 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 334 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 335 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF, 336 }; 337 338 static const u32 golden_settings_polaris11_a11[] = 339 { 340 mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208, 341 mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000, 342 mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040, 343 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 344 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 345 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 346 mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012, 347 mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000, 348 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, 349 mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c, 350 mmSQ_CONFIG, 0x07f80000, 0x01180000, 351 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 352 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 353 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3, 354 mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000, 355 mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210, 356 mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, 357 }; 358 359 static const u32 polaris11_golden_common_all[] = 360 { 361 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 362 mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002, 363 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 364 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 365 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 366 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF, 367 }; 368 369 static const u32 golden_settings_polaris10_a11[] = 370 { 371 mmATC_MISC_CG, 0x000c0fc0, 0x000c0200, 372 mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208, 373 mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000, 374 mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040, 375 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 376 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 377 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 378 mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012, 379 mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a, 380 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, 381 mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c, 382 mmSQ_CONFIG, 0x07f80000, 0x07180000, 383 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 384 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 385 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7, 386 mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000, 387 mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, 388 }; 389 390 static const u32 polaris10_golden_common_all[] = 391 { 392 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 393 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012, 394 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A, 395 mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003, 396 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 397 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 398 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 399 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF, 400 }; 401 402 static const u32 fiji_golden_common_all[] = 403 { 404 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 405 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a, 406 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e, 407 mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003, 408 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 409 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 410 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 411 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF, 412 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 413 mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009, 414 }; 415 416 static const u32 golden_settings_fiji_a10[] = 417 { 418 mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040, 419 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 420 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 421 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 422 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, 423 mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd, 424 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 425 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 426 mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, 427 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff, 428 mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, 429 }; 430 431 static const u32 fiji_mgcg_cgcg_init[] = 432 { 433 mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff, 434 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 435 mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 436 mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100, 437 mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100, 438 mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100, 439 mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100, 440 mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100, 441 mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100, 442 mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100, 443 mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100, 444 mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100, 445 mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100, 446 mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100, 447 mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100, 448 mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100, 449 mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100, 450 mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100, 451 mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100, 452 mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100, 453 mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100, 454 mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100, 455 mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100, 456 mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100, 457 mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100, 458 mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100, 459 mmTA_CGTT_CTRL, 0xffffffff, 0x00000100, 460 mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 461 mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 462 mmTD_CGTT_CTRL, 0xffffffff, 0x00000100, 463 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 464 mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200, 465 mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100, 466 mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c, 467 mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001, 468 }; 469 470 static const u32 golden_settings_iceland_a11[] = 471 { 472 mmCB_HW_CONTROL_3, 0x00000040, 0x00000040, 473 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 474 mmDB_DEBUG3, 0xc0000000, 0xc0000000, 475 mmGB_GPU_ID, 0x0000000f, 0x00000000, 476 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 477 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 478 mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002, 479 mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000, 480 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c, 481 mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd, 482 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 483 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 484 mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, 485 mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1, 486 mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000, 487 mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010, 488 }; 489 490 static const u32 iceland_golden_common_all[] = 491 { 492 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 493 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002, 494 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000, 495 mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001, 496 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 497 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 498 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 499 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF 500 }; 501 502 static const u32 iceland_mgcg_cgcg_init[] = 503 { 504 mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff, 505 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 506 mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 507 mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100, 508 mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100, 509 mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100, 510 mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100, 511 mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100, 512 mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100, 513 mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100, 514 mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100, 515 mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100, 516 mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100, 517 mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100, 518 mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100, 519 mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100, 520 mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100, 521 mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100, 522 mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100, 523 mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100, 524 mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100, 525 mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100, 526 mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100, 527 mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100, 528 mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100, 529 mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100, 530 mmTA_CGTT_CTRL, 0xffffffff, 0x00000100, 531 mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 532 mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 533 mmTD_CGTT_CTRL, 0xffffffff, 0x00000100, 534 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 535 mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000, 536 mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 537 mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87, 538 mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005, 539 mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 540 mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000, 541 mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 542 mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007, 543 mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005, 544 mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 545 mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000, 546 mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 547 mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007, 548 mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005, 549 mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 550 mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000, 551 mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 552 mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007, 553 mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005, 554 mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 555 mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000, 556 mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 557 mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87, 558 mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005, 559 mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 560 mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000, 561 mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 562 mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007, 563 mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005, 564 mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 565 mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200, 566 mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100, 567 mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c, 568 }; 569 570 static const u32 cz_golden_settings_a11[] = 571 { 572 mmCB_HW_CONTROL_3, 0x00000040, 0x00000040, 573 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 574 mmGB_GPU_ID, 0x0000000f, 0x00000000, 575 mmPA_SC_ENHANCE, 0xffffffff, 0x00000001, 576 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 577 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c, 578 mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd, 579 mmTA_CNTL_AUX, 0x000f000f, 0x00010000, 580 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 581 mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, 582 mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3, 583 mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302 584 }; 585 586 static const u32 cz_golden_common_all[] = 587 { 588 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 589 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002, 590 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000, 591 mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001, 592 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 593 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 594 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 595 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF 596 }; 597 598 static const u32 cz_mgcg_cgcg_init[] = 599 { 600 mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff, 601 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 602 mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 603 mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100, 604 mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100, 605 mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100, 606 mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100, 607 mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100, 608 mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100, 609 mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100, 610 mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100, 611 mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100, 612 mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100, 613 mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100, 614 mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100, 615 mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100, 616 mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100, 617 mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100, 618 mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100, 619 mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100, 620 mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100, 621 mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100, 622 mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100, 623 mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100, 624 mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100, 625 mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100, 626 mmTA_CGTT_CTRL, 0xffffffff, 0x00000100, 627 mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 628 mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, 629 mmTD_CGTT_CTRL, 0xffffffff, 0x00000100, 630 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 631 mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000, 632 mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 633 mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007, 634 mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005, 635 mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 636 mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000, 637 mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 638 mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007, 639 mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005, 640 mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 641 mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000, 642 mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 643 mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007, 644 mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005, 645 mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 646 mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000, 647 mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 648 mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007, 649 mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005, 650 mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 651 mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000, 652 mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 653 mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007, 654 mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005, 655 mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 656 mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000, 657 mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 658 mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007, 659 mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005, 660 mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 661 mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000, 662 mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 663 mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007, 664 mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005, 665 mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 666 mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000, 667 mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002, 668 mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007, 669 mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005, 670 mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008, 671 mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200, 672 mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100, 673 mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f, 674 mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001, 675 }; 676 677 static const u32 stoney_golden_settings_a11[] = 678 { 679 mmDB_DEBUG2, 0xf00fffff, 0x00000400, 680 mmGB_GPU_ID, 0x0000000f, 0x00000000, 681 mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, 682 mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, 683 mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, 684 mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, 685 mmTCC_CTRL, 0x00100000, 0xf31fff7f, 686 mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, 687 mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1, 688 mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010, 689 }; 690 691 static const u32 stoney_golden_common_all[] = 692 { 693 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 694 mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000, 695 mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000, 696 mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001, 697 mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, 698 mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, 699 mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF, 700 mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF, 701 }; 702 703 static const u32 stoney_mgcg_cgcg_init[] = 704 { 705 mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, 706 mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f, 707 mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201, 708 mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201, 709 mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200, 710 }; 711 712 713 static const char * const sq_edc_source_names[] = { 714 "SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred", 715 "SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch", 716 "SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return", 717 "SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR", 718 "SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS", 719 "SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS", 720 "SQ_EDC_INFO_SOURCE_TA: EDC source is TA", 721 }; 722 723 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev); 724 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev); 725 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev); 726 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev); 727 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev); 728 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev); 729 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring); 730 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring); 731 732 #define CG_ACLK_CNTL__ACLK_DIVIDER_MASK 0x0000007fL 733 #define CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT 0x00000000L 734 735 static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev) 736 { 737 uint32_t data; 738 739 switch (adev->asic_type) { 740 case CHIP_TOPAZ: 741 amdgpu_device_program_register_sequence(adev, 742 iceland_mgcg_cgcg_init, 743 ARRAY_SIZE(iceland_mgcg_cgcg_init)); 744 amdgpu_device_program_register_sequence(adev, 745 golden_settings_iceland_a11, 746 ARRAY_SIZE(golden_settings_iceland_a11)); 747 amdgpu_device_program_register_sequence(adev, 748 iceland_golden_common_all, 749 ARRAY_SIZE(iceland_golden_common_all)); 750 break; 751 case CHIP_FIJI: 752 amdgpu_device_program_register_sequence(adev, 753 fiji_mgcg_cgcg_init, 754 ARRAY_SIZE(fiji_mgcg_cgcg_init)); 755 amdgpu_device_program_register_sequence(adev, 756 golden_settings_fiji_a10, 757 ARRAY_SIZE(golden_settings_fiji_a10)); 758 amdgpu_device_program_register_sequence(adev, 759 fiji_golden_common_all, 760 ARRAY_SIZE(fiji_golden_common_all)); 761 break; 762 763 case CHIP_TONGA: 764 amdgpu_device_program_register_sequence(adev, 765 tonga_mgcg_cgcg_init, 766 ARRAY_SIZE(tonga_mgcg_cgcg_init)); 767 amdgpu_device_program_register_sequence(adev, 768 golden_settings_tonga_a11, 769 ARRAY_SIZE(golden_settings_tonga_a11)); 770 amdgpu_device_program_register_sequence(adev, 771 tonga_golden_common_all, 772 ARRAY_SIZE(tonga_golden_common_all)); 773 break; 774 case CHIP_VEGAM: 775 amdgpu_device_program_register_sequence(adev, 776 golden_settings_vegam_a11, 777 ARRAY_SIZE(golden_settings_vegam_a11)); 778 amdgpu_device_program_register_sequence(adev, 779 vegam_golden_common_all, 780 ARRAY_SIZE(vegam_golden_common_all)); 781 break; 782 case CHIP_POLARIS11: 783 case CHIP_POLARIS12: 784 amdgpu_device_program_register_sequence(adev, 785 golden_settings_polaris11_a11, 786 ARRAY_SIZE(golden_settings_polaris11_a11)); 787 amdgpu_device_program_register_sequence(adev, 788 polaris11_golden_common_all, 789 ARRAY_SIZE(polaris11_golden_common_all)); 790 break; 791 case CHIP_POLARIS10: 792 amdgpu_device_program_register_sequence(adev, 793 golden_settings_polaris10_a11, 794 ARRAY_SIZE(golden_settings_polaris10_a11)); 795 amdgpu_device_program_register_sequence(adev, 796 polaris10_golden_common_all, 797 ARRAY_SIZE(polaris10_golden_common_all)); 798 data = RREG32_SMC(ixCG_ACLK_CNTL); 799 data &= ~CG_ACLK_CNTL__ACLK_DIVIDER_MASK; 800 data |= 0x18 << CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT; 801 WREG32_SMC(ixCG_ACLK_CNTL, data); 802 if ((adev->pdev->device == 0x67DF) && (adev->pdev->revision == 0xc7) && 803 ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) || 804 (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) || 805 (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1680))) { 806 amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD); 807 amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0); 808 } 809 break; 810 case CHIP_CARRIZO: 811 amdgpu_device_program_register_sequence(adev, 812 cz_mgcg_cgcg_init, 813 ARRAY_SIZE(cz_mgcg_cgcg_init)); 814 amdgpu_device_program_register_sequence(adev, 815 cz_golden_settings_a11, 816 ARRAY_SIZE(cz_golden_settings_a11)); 817 amdgpu_device_program_register_sequence(adev, 818 cz_golden_common_all, 819 ARRAY_SIZE(cz_golden_common_all)); 820 break; 821 case CHIP_STONEY: 822 amdgpu_device_program_register_sequence(adev, 823 stoney_mgcg_cgcg_init, 824 ARRAY_SIZE(stoney_mgcg_cgcg_init)); 825 amdgpu_device_program_register_sequence(adev, 826 stoney_golden_settings_a11, 827 ARRAY_SIZE(stoney_golden_settings_a11)); 828 amdgpu_device_program_register_sequence(adev, 829 stoney_golden_common_all, 830 ARRAY_SIZE(stoney_golden_common_all)); 831 break; 832 default: 833 break; 834 } 835 } 836 837 static void gfx_v8_0_scratch_init(struct amdgpu_device *adev) 838 { 839 adev->gfx.scratch.num_reg = 8; 840 adev->gfx.scratch.reg_base = mmSCRATCH_REG0; 841 adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1; 842 } 843 844 static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring) 845 { 846 struct amdgpu_device *adev = ring->adev; 847 uint32_t scratch; 848 uint32_t tmp = 0; 849 unsigned i; 850 int r; 851 852 r = amdgpu_gfx_scratch_get(adev, &scratch); 853 if (r) 854 return r; 855 856 WREG32(scratch, 0xCAFEDEAD); 857 r = amdgpu_ring_alloc(ring, 3); 858 if (r) 859 goto error_free_scratch; 860 861 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); 862 amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); 863 amdgpu_ring_write(ring, 0xDEADBEEF); 864 amdgpu_ring_commit(ring); 865 866 for (i = 0; i < adev->usec_timeout; i++) { 867 tmp = RREG32(scratch); 868 if (tmp == 0xDEADBEEF) 869 break; 870 udelay(1); 871 } 872 873 if (i >= adev->usec_timeout) 874 r = -ETIMEDOUT; 875 876 error_free_scratch: 877 amdgpu_gfx_scratch_free(adev, scratch); 878 return r; 879 } 880 881 static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) 882 { 883 struct amdgpu_device *adev = ring->adev; 884 struct amdgpu_ib ib; 885 struct dma_fence *f = NULL; 886 887 unsigned int index; 888 uint64_t gpu_addr; 889 uint32_t tmp; 890 long r; 891 892 r = amdgpu_device_wb_get(adev, &index); 893 if (r) 894 return r; 895 896 gpu_addr = adev->wb.gpu_addr + (index * 4); 897 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD); 898 memset(&ib, 0, sizeof(ib)); 899 r = amdgpu_ib_get(adev, NULL, 16, 900 AMDGPU_IB_POOL_DIRECT, &ib); 901 if (r) 902 goto err1; 903 904 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); 905 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; 906 ib.ptr[2] = lower_32_bits(gpu_addr); 907 ib.ptr[3] = upper_32_bits(gpu_addr); 908 ib.ptr[4] = 0xDEADBEEF; 909 ib.length_dw = 5; 910 911 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 912 if (r) 913 goto err2; 914 915 r = dma_fence_wait_timeout(f, false, timeout); 916 if (r == 0) { 917 r = -ETIMEDOUT; 918 goto err2; 919 } else if (r < 0) { 920 goto err2; 921 } 922 923 tmp = adev->wb.wb[index]; 924 if (tmp == 0xDEADBEEF) 925 r = 0; 926 else 927 r = -EINVAL; 928 929 err2: 930 amdgpu_ib_free(adev, &ib, NULL); 931 dma_fence_put(f); 932 err1: 933 amdgpu_device_wb_free(adev, index); 934 return r; 935 } 936 937 938 static void gfx_v8_0_free_microcode(struct amdgpu_device *adev) 939 { 940 release_firmware(adev->gfx.pfp_fw); 941 adev->gfx.pfp_fw = NULL; 942 release_firmware(adev->gfx.me_fw); 943 adev->gfx.me_fw = NULL; 944 release_firmware(adev->gfx.ce_fw); 945 adev->gfx.ce_fw = NULL; 946 release_firmware(adev->gfx.rlc_fw); 947 adev->gfx.rlc_fw = NULL; 948 release_firmware(adev->gfx.mec_fw); 949 adev->gfx.mec_fw = NULL; 950 if ((adev->asic_type != CHIP_STONEY) && 951 (adev->asic_type != CHIP_TOPAZ)) 952 release_firmware(adev->gfx.mec2_fw); 953 adev->gfx.mec2_fw = NULL; 954 955 kfree(adev->gfx.rlc.register_list_format); 956 } 957 958 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) 959 { 960 const char *chip_name; 961 char fw_name[30]; 962 int err; 963 struct amdgpu_firmware_info *info = NULL; 964 const struct common_firmware_header *header = NULL; 965 const struct gfx_firmware_header_v1_0 *cp_hdr; 966 const struct rlc_firmware_header_v2_0 *rlc_hdr; 967 unsigned int *tmp = NULL, i; 968 969 DRM_DEBUG("\n"); 970 971 switch (adev->asic_type) { 972 case CHIP_TOPAZ: 973 chip_name = "topaz"; 974 break; 975 case CHIP_TONGA: 976 chip_name = "tonga"; 977 break; 978 case CHIP_CARRIZO: 979 chip_name = "carrizo"; 980 break; 981 case CHIP_FIJI: 982 chip_name = "fiji"; 983 break; 984 case CHIP_STONEY: 985 chip_name = "stoney"; 986 break; 987 case CHIP_POLARIS10: 988 chip_name = "polaris10"; 989 break; 990 case CHIP_POLARIS11: 991 chip_name = "polaris11"; 992 break; 993 case CHIP_POLARIS12: 994 chip_name = "polaris12"; 995 break; 996 case CHIP_VEGAM: 997 chip_name = "vegam"; 998 break; 999 default: 1000 BUG(); 1001 } 1002 1003 if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) { 1004 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name); 1005 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev); 1006 if (err == -ENOENT) { 1007 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name); 1008 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev); 1009 } 1010 } else { 1011 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name); 1012 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev); 1013 } 1014 if (err) 1015 goto out; 1016 err = amdgpu_ucode_validate(adev->gfx.pfp_fw); 1017 if (err) 1018 goto out; 1019 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 1020 adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 1021 adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 1022 1023 if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) { 1024 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name); 1025 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev); 1026 if (err == -ENOENT) { 1027 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name); 1028 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev); 1029 } 1030 } else { 1031 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name); 1032 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev); 1033 } 1034 if (err) 1035 goto out; 1036 err = amdgpu_ucode_validate(adev->gfx.me_fw); 1037 if (err) 1038 goto out; 1039 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 1040 adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 1041 1042 adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 1043 1044 if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) { 1045 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name); 1046 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev); 1047 if (err == -ENOENT) { 1048 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name); 1049 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev); 1050 } 1051 } else { 1052 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name); 1053 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev); 1054 } 1055 if (err) 1056 goto out; 1057 err = amdgpu_ucode_validate(adev->gfx.ce_fw); 1058 if (err) 1059 goto out; 1060 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 1061 adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 1062 adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 1063 1064 /* 1065 * Support for MCBP/Virtualization in combination with chained IBs is 1066 * formal released on feature version #46 1067 */ 1068 if (adev->gfx.ce_feature_version >= 46 && 1069 adev->gfx.pfp_feature_version >= 46) { 1070 adev->virt.chained_ib_support = true; 1071 DRM_INFO("Chained IB support enabled!\n"); 1072 } else 1073 adev->virt.chained_ib_support = false; 1074 1075 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); 1076 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev); 1077 if (err) 1078 goto out; 1079 err = amdgpu_ucode_validate(adev->gfx.rlc_fw); 1080 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; 1081 adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version); 1082 adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version); 1083 1084 adev->gfx.rlc.save_and_restore_offset = 1085 le32_to_cpu(rlc_hdr->save_and_restore_offset); 1086 adev->gfx.rlc.clear_state_descriptor_offset = 1087 le32_to_cpu(rlc_hdr->clear_state_descriptor_offset); 1088 adev->gfx.rlc.avail_scratch_ram_locations = 1089 le32_to_cpu(rlc_hdr->avail_scratch_ram_locations); 1090 adev->gfx.rlc.reg_restore_list_size = 1091 le32_to_cpu(rlc_hdr->reg_restore_list_size); 1092 adev->gfx.rlc.reg_list_format_start = 1093 le32_to_cpu(rlc_hdr->reg_list_format_start); 1094 adev->gfx.rlc.reg_list_format_separate_start = 1095 le32_to_cpu(rlc_hdr->reg_list_format_separate_start); 1096 adev->gfx.rlc.starting_offsets_start = 1097 le32_to_cpu(rlc_hdr->starting_offsets_start); 1098 adev->gfx.rlc.reg_list_format_size_bytes = 1099 le32_to_cpu(rlc_hdr->reg_list_format_size_bytes); 1100 adev->gfx.rlc.reg_list_size_bytes = 1101 le32_to_cpu(rlc_hdr->reg_list_size_bytes); 1102 1103 adev->gfx.rlc.register_list_format = 1104 kmalloc(adev->gfx.rlc.reg_list_format_size_bytes + 1105 adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL); 1106 1107 if (!adev->gfx.rlc.register_list_format) { 1108 err = -ENOMEM; 1109 goto out; 1110 } 1111 1112 tmp = (unsigned int *)((uintptr_t)rlc_hdr + 1113 le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes)); 1114 for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++) 1115 adev->gfx.rlc.register_list_format[i] = le32_to_cpu(tmp[i]); 1116 1117 adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i; 1118 1119 tmp = (unsigned int *)((uintptr_t)rlc_hdr + 1120 le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes)); 1121 for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++) 1122 adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]); 1123 1124 if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) { 1125 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name); 1126 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev); 1127 if (err == -ENOENT) { 1128 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name); 1129 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev); 1130 } 1131 } else { 1132 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name); 1133 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev); 1134 } 1135 if (err) 1136 goto out; 1137 err = amdgpu_ucode_validate(adev->gfx.mec_fw); 1138 if (err) 1139 goto out; 1140 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 1141 adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 1142 adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 1143 1144 if ((adev->asic_type != CHIP_STONEY) && 1145 (adev->asic_type != CHIP_TOPAZ)) { 1146 if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) { 1147 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name); 1148 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); 1149 if (err == -ENOENT) { 1150 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); 1151 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); 1152 } 1153 } else { 1154 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); 1155 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); 1156 } 1157 if (!err) { 1158 err = amdgpu_ucode_validate(adev->gfx.mec2_fw); 1159 if (err) 1160 goto out; 1161 cp_hdr = (const struct gfx_firmware_header_v1_0 *) 1162 adev->gfx.mec2_fw->data; 1163 adev->gfx.mec2_fw_version = 1164 le32_to_cpu(cp_hdr->header.ucode_version); 1165 adev->gfx.mec2_feature_version = 1166 le32_to_cpu(cp_hdr->ucode_feature_version); 1167 } else { 1168 err = 0; 1169 adev->gfx.mec2_fw = NULL; 1170 } 1171 } 1172 1173 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP]; 1174 info->ucode_id = AMDGPU_UCODE_ID_CP_PFP; 1175 info->fw = adev->gfx.pfp_fw; 1176 header = (const struct common_firmware_header *)info->fw->data; 1177 adev->firmware.fw_size += 1178 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1179 1180 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME]; 1181 info->ucode_id = AMDGPU_UCODE_ID_CP_ME; 1182 info->fw = adev->gfx.me_fw; 1183 header = (const struct common_firmware_header *)info->fw->data; 1184 adev->firmware.fw_size += 1185 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1186 1187 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE]; 1188 info->ucode_id = AMDGPU_UCODE_ID_CP_CE; 1189 info->fw = adev->gfx.ce_fw; 1190 header = (const struct common_firmware_header *)info->fw->data; 1191 adev->firmware.fw_size += 1192 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1193 1194 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G]; 1195 info->ucode_id = AMDGPU_UCODE_ID_RLC_G; 1196 info->fw = adev->gfx.rlc_fw; 1197 header = (const struct common_firmware_header *)info->fw->data; 1198 adev->firmware.fw_size += 1199 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1200 1201 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1]; 1202 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1; 1203 info->fw = adev->gfx.mec_fw; 1204 header = (const struct common_firmware_header *)info->fw->data; 1205 adev->firmware.fw_size += 1206 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1207 1208 /* we need account JT in */ 1209 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 1210 adev->firmware.fw_size += 1211 ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE); 1212 1213 if (amdgpu_sriov_vf(adev)) { 1214 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE]; 1215 info->ucode_id = AMDGPU_UCODE_ID_STORAGE; 1216 info->fw = adev->gfx.mec_fw; 1217 adev->firmware.fw_size += 1218 ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE); 1219 } 1220 1221 if (adev->gfx.mec2_fw) { 1222 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2]; 1223 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2; 1224 info->fw = adev->gfx.mec2_fw; 1225 header = (const struct common_firmware_header *)info->fw->data; 1226 adev->firmware.fw_size += 1227 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 1228 } 1229 1230 out: 1231 if (err) { 1232 dev_err(adev->dev, 1233 "gfx8: Failed to load firmware \"%s\"\n", 1234 fw_name); 1235 release_firmware(adev->gfx.pfp_fw); 1236 adev->gfx.pfp_fw = NULL; 1237 release_firmware(adev->gfx.me_fw); 1238 adev->gfx.me_fw = NULL; 1239 release_firmware(adev->gfx.ce_fw); 1240 adev->gfx.ce_fw = NULL; 1241 release_firmware(adev->gfx.rlc_fw); 1242 adev->gfx.rlc_fw = NULL; 1243 release_firmware(adev->gfx.mec_fw); 1244 adev->gfx.mec_fw = NULL; 1245 release_firmware(adev->gfx.mec2_fw); 1246 adev->gfx.mec2_fw = NULL; 1247 } 1248 return err; 1249 } 1250 1251 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev, 1252 volatile u32 *buffer) 1253 { 1254 u32 count = 0, i; 1255 const struct cs_section_def *sect = NULL; 1256 const struct cs_extent_def *ext = NULL; 1257 1258 if (adev->gfx.rlc.cs_data == NULL) 1259 return; 1260 if (buffer == NULL) 1261 return; 1262 1263 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1264 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 1265 1266 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 1267 buffer[count++] = cpu_to_le32(0x80000000); 1268 buffer[count++] = cpu_to_le32(0x80000000); 1269 1270 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 1271 for (ext = sect->section; ext->extent != NULL; ++ext) { 1272 if (sect->id == SECT_CONTEXT) { 1273 buffer[count++] = 1274 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 1275 buffer[count++] = cpu_to_le32(ext->reg_index - 1276 PACKET3_SET_CONTEXT_REG_START); 1277 for (i = 0; i < ext->reg_count; i++) 1278 buffer[count++] = cpu_to_le32(ext->extent[i]); 1279 } else { 1280 return; 1281 } 1282 } 1283 } 1284 1285 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 1286 buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - 1287 PACKET3_SET_CONTEXT_REG_START); 1288 buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config); 1289 buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1); 1290 1291 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1292 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 1293 1294 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 1295 buffer[count++] = cpu_to_le32(0); 1296 } 1297 1298 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev) 1299 { 1300 if (adev->asic_type == CHIP_CARRIZO) 1301 return 5; 1302 else 1303 return 4; 1304 } 1305 1306 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev) 1307 { 1308 const struct cs_section_def *cs_data; 1309 int r; 1310 1311 adev->gfx.rlc.cs_data = vi_cs_data; 1312 1313 cs_data = adev->gfx.rlc.cs_data; 1314 1315 if (cs_data) { 1316 /* init clear state block */ 1317 r = amdgpu_gfx_rlc_init_csb(adev); 1318 if (r) 1319 return r; 1320 } 1321 1322 if ((adev->asic_type == CHIP_CARRIZO) || 1323 (adev->asic_type == CHIP_STONEY)) { 1324 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ 1325 r = amdgpu_gfx_rlc_init_cpt(adev); 1326 if (r) 1327 return r; 1328 } 1329 1330 /* init spm vmid with 0xf */ 1331 if (adev->gfx.rlc.funcs->update_spm_vmid) 1332 adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf); 1333 1334 return 0; 1335 } 1336 1337 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev) 1338 { 1339 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL); 1340 } 1341 1342 static int gfx_v8_0_mec_init(struct amdgpu_device *adev) 1343 { 1344 int r; 1345 u32 *hpd; 1346 size_t mec_hpd_size; 1347 1348 bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); 1349 1350 /* take ownership of the relevant compute queues */ 1351 amdgpu_gfx_compute_queue_acquire(adev); 1352 1353 mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE; 1354 if (mec_hpd_size) { 1355 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, 1356 AMDGPU_GEM_DOMAIN_VRAM, 1357 &adev->gfx.mec.hpd_eop_obj, 1358 &adev->gfx.mec.hpd_eop_gpu_addr, 1359 (void **)&hpd); 1360 if (r) { 1361 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r); 1362 return r; 1363 } 1364 1365 memset(hpd, 0, mec_hpd_size); 1366 1367 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj); 1368 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj); 1369 } 1370 1371 return 0; 1372 } 1373 1374 static const u32 vgpr_init_compute_shader[] = 1375 { 1376 0x7e000209, 0x7e020208, 1377 0x7e040207, 0x7e060206, 1378 0x7e080205, 0x7e0a0204, 1379 0x7e0c0203, 0x7e0e0202, 1380 0x7e100201, 0x7e120200, 1381 0x7e140209, 0x7e160208, 1382 0x7e180207, 0x7e1a0206, 1383 0x7e1c0205, 0x7e1e0204, 1384 0x7e200203, 0x7e220202, 1385 0x7e240201, 0x7e260200, 1386 0x7e280209, 0x7e2a0208, 1387 0x7e2c0207, 0x7e2e0206, 1388 0x7e300205, 0x7e320204, 1389 0x7e340203, 0x7e360202, 1390 0x7e380201, 0x7e3a0200, 1391 0x7e3c0209, 0x7e3e0208, 1392 0x7e400207, 0x7e420206, 1393 0x7e440205, 0x7e460204, 1394 0x7e480203, 0x7e4a0202, 1395 0x7e4c0201, 0x7e4e0200, 1396 0x7e500209, 0x7e520208, 1397 0x7e540207, 0x7e560206, 1398 0x7e580205, 0x7e5a0204, 1399 0x7e5c0203, 0x7e5e0202, 1400 0x7e600201, 0x7e620200, 1401 0x7e640209, 0x7e660208, 1402 0x7e680207, 0x7e6a0206, 1403 0x7e6c0205, 0x7e6e0204, 1404 0x7e700203, 0x7e720202, 1405 0x7e740201, 0x7e760200, 1406 0x7e780209, 0x7e7a0208, 1407 0x7e7c0207, 0x7e7e0206, 1408 0xbf8a0000, 0xbf810000, 1409 }; 1410 1411 static const u32 sgpr_init_compute_shader[] = 1412 { 1413 0xbe8a0100, 0xbe8c0102, 1414 0xbe8e0104, 0xbe900106, 1415 0xbe920108, 0xbe940100, 1416 0xbe960102, 0xbe980104, 1417 0xbe9a0106, 0xbe9c0108, 1418 0xbe9e0100, 0xbea00102, 1419 0xbea20104, 0xbea40106, 1420 0xbea60108, 0xbea80100, 1421 0xbeaa0102, 0xbeac0104, 1422 0xbeae0106, 0xbeb00108, 1423 0xbeb20100, 0xbeb40102, 1424 0xbeb60104, 0xbeb80106, 1425 0xbeba0108, 0xbebc0100, 1426 0xbebe0102, 0xbec00104, 1427 0xbec20106, 0xbec40108, 1428 0xbec60100, 0xbec80102, 1429 0xbee60004, 0xbee70005, 1430 0xbeea0006, 0xbeeb0007, 1431 0xbee80008, 0xbee90009, 1432 0xbefc0000, 0xbf8a0000, 1433 0xbf810000, 0x00000000, 1434 }; 1435 1436 static const u32 vgpr_init_regs[] = 1437 { 1438 mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff, 1439 mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */ 1440 mmCOMPUTE_NUM_THREAD_X, 256*4, 1441 mmCOMPUTE_NUM_THREAD_Y, 1, 1442 mmCOMPUTE_NUM_THREAD_Z, 1, 1443 mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */ 1444 mmCOMPUTE_PGM_RSRC2, 20, 1445 mmCOMPUTE_USER_DATA_0, 0xedcedc00, 1446 mmCOMPUTE_USER_DATA_1, 0xedcedc01, 1447 mmCOMPUTE_USER_DATA_2, 0xedcedc02, 1448 mmCOMPUTE_USER_DATA_3, 0xedcedc03, 1449 mmCOMPUTE_USER_DATA_4, 0xedcedc04, 1450 mmCOMPUTE_USER_DATA_5, 0xedcedc05, 1451 mmCOMPUTE_USER_DATA_6, 0xedcedc06, 1452 mmCOMPUTE_USER_DATA_7, 0xedcedc07, 1453 mmCOMPUTE_USER_DATA_8, 0xedcedc08, 1454 mmCOMPUTE_USER_DATA_9, 0xedcedc09, 1455 }; 1456 1457 static const u32 sgpr1_init_regs[] = 1458 { 1459 mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f, 1460 mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */ 1461 mmCOMPUTE_NUM_THREAD_X, 256*5, 1462 mmCOMPUTE_NUM_THREAD_Y, 1, 1463 mmCOMPUTE_NUM_THREAD_Z, 1, 1464 mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */ 1465 mmCOMPUTE_PGM_RSRC2, 20, 1466 mmCOMPUTE_USER_DATA_0, 0xedcedc00, 1467 mmCOMPUTE_USER_DATA_1, 0xedcedc01, 1468 mmCOMPUTE_USER_DATA_2, 0xedcedc02, 1469 mmCOMPUTE_USER_DATA_3, 0xedcedc03, 1470 mmCOMPUTE_USER_DATA_4, 0xedcedc04, 1471 mmCOMPUTE_USER_DATA_5, 0xedcedc05, 1472 mmCOMPUTE_USER_DATA_6, 0xedcedc06, 1473 mmCOMPUTE_USER_DATA_7, 0xedcedc07, 1474 mmCOMPUTE_USER_DATA_8, 0xedcedc08, 1475 mmCOMPUTE_USER_DATA_9, 0xedcedc09, 1476 }; 1477 1478 static const u32 sgpr2_init_regs[] = 1479 { 1480 mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0, 1481 mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, 1482 mmCOMPUTE_NUM_THREAD_X, 256*5, 1483 mmCOMPUTE_NUM_THREAD_Y, 1, 1484 mmCOMPUTE_NUM_THREAD_Z, 1, 1485 mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */ 1486 mmCOMPUTE_PGM_RSRC2, 20, 1487 mmCOMPUTE_USER_DATA_0, 0xedcedc00, 1488 mmCOMPUTE_USER_DATA_1, 0xedcedc01, 1489 mmCOMPUTE_USER_DATA_2, 0xedcedc02, 1490 mmCOMPUTE_USER_DATA_3, 0xedcedc03, 1491 mmCOMPUTE_USER_DATA_4, 0xedcedc04, 1492 mmCOMPUTE_USER_DATA_5, 0xedcedc05, 1493 mmCOMPUTE_USER_DATA_6, 0xedcedc06, 1494 mmCOMPUTE_USER_DATA_7, 0xedcedc07, 1495 mmCOMPUTE_USER_DATA_8, 0xedcedc08, 1496 mmCOMPUTE_USER_DATA_9, 0xedcedc09, 1497 }; 1498 1499 static const u32 sec_ded_counter_registers[] = 1500 { 1501 mmCPC_EDC_ATC_CNT, 1502 mmCPC_EDC_SCRATCH_CNT, 1503 mmCPC_EDC_UCODE_CNT, 1504 mmCPF_EDC_ATC_CNT, 1505 mmCPF_EDC_ROQ_CNT, 1506 mmCPF_EDC_TAG_CNT, 1507 mmCPG_EDC_ATC_CNT, 1508 mmCPG_EDC_DMA_CNT, 1509 mmCPG_EDC_TAG_CNT, 1510 mmDC_EDC_CSINVOC_CNT, 1511 mmDC_EDC_RESTORE_CNT, 1512 mmDC_EDC_STATE_CNT, 1513 mmGDS_EDC_CNT, 1514 mmGDS_EDC_GRBM_CNT, 1515 mmGDS_EDC_OA_DED, 1516 mmSPI_EDC_CNT, 1517 mmSQC_ATC_EDC_GATCL1_CNT, 1518 mmSQC_EDC_CNT, 1519 mmSQ_EDC_DED_CNT, 1520 mmSQ_EDC_INFO, 1521 mmSQ_EDC_SEC_CNT, 1522 mmTCC_EDC_CNT, 1523 mmTCP_ATC_EDC_GATCL1_CNT, 1524 mmTCP_EDC_CNT, 1525 mmTD_EDC_CNT 1526 }; 1527 1528 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) 1529 { 1530 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; 1531 struct amdgpu_ib ib; 1532 struct dma_fence *f = NULL; 1533 int r, i; 1534 u32 tmp; 1535 unsigned total_size, vgpr_offset, sgpr_offset; 1536 u64 gpu_addr; 1537 1538 /* only supported on CZ */ 1539 if (adev->asic_type != CHIP_CARRIZO) 1540 return 0; 1541 1542 /* bail if the compute ring is not ready */ 1543 if (!ring->sched.ready) 1544 return 0; 1545 1546 tmp = RREG32(mmGB_EDC_MODE); 1547 WREG32(mmGB_EDC_MODE, 0); 1548 1549 total_size = 1550 (((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; 1551 total_size += 1552 (((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; 1553 total_size += 1554 (((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; 1555 total_size = ALIGN(total_size, 256); 1556 vgpr_offset = total_size; 1557 total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256); 1558 sgpr_offset = total_size; 1559 total_size += sizeof(sgpr_init_compute_shader); 1560 1561 /* allocate an indirect buffer to put the commands in */ 1562 memset(&ib, 0, sizeof(ib)); 1563 r = amdgpu_ib_get(adev, NULL, total_size, 1564 AMDGPU_IB_POOL_DIRECT, &ib); 1565 if (r) { 1566 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); 1567 return r; 1568 } 1569 1570 /* load the compute shaders */ 1571 for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++) 1572 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i]; 1573 1574 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++) 1575 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i]; 1576 1577 /* init the ib length to 0 */ 1578 ib.length_dw = 0; 1579 1580 /* VGPR */ 1581 /* write the register state for the compute dispatch */ 1582 for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) { 1583 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 1584 ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START; 1585 ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1]; 1586 } 1587 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 1588 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8; 1589 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 1590 ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; 1591 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 1592 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 1593 1594 /* write dispatch packet */ 1595 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 1596 ib.ptr[ib.length_dw++] = 8; /* x */ 1597 ib.ptr[ib.length_dw++] = 1; /* y */ 1598 ib.ptr[ib.length_dw++] = 1; /* z */ 1599 ib.ptr[ib.length_dw++] = 1600 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 1601 1602 /* write CS partial flush packet */ 1603 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 1604 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 1605 1606 /* SGPR1 */ 1607 /* write the register state for the compute dispatch */ 1608 for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) { 1609 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 1610 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START; 1611 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1]; 1612 } 1613 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 1614 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 1615 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 1616 ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; 1617 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 1618 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 1619 1620 /* write dispatch packet */ 1621 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 1622 ib.ptr[ib.length_dw++] = 8; /* x */ 1623 ib.ptr[ib.length_dw++] = 1; /* y */ 1624 ib.ptr[ib.length_dw++] = 1; /* z */ 1625 ib.ptr[ib.length_dw++] = 1626 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 1627 1628 /* write CS partial flush packet */ 1629 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 1630 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 1631 1632 /* SGPR2 */ 1633 /* write the register state for the compute dispatch */ 1634 for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) { 1635 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 1636 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START; 1637 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1]; 1638 } 1639 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 1640 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 1641 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 1642 ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; 1643 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 1644 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 1645 1646 /* write dispatch packet */ 1647 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 1648 ib.ptr[ib.length_dw++] = 8; /* x */ 1649 ib.ptr[ib.length_dw++] = 1; /* y */ 1650 ib.ptr[ib.length_dw++] = 1; /* z */ 1651 ib.ptr[ib.length_dw++] = 1652 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 1653 1654 /* write CS partial flush packet */ 1655 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 1656 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 1657 1658 /* shedule the ib on the ring */ 1659 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 1660 if (r) { 1661 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r); 1662 goto fail; 1663 } 1664 1665 /* wait for the GPU to finish processing the IB */ 1666 r = dma_fence_wait(f, false); 1667 if (r) { 1668 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); 1669 goto fail; 1670 } 1671 1672 tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2); 1673 tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1); 1674 WREG32(mmGB_EDC_MODE, tmp); 1675 1676 tmp = RREG32(mmCC_GC_EDC_CONFIG); 1677 tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1; 1678 WREG32(mmCC_GC_EDC_CONFIG, tmp); 1679 1680 1681 /* read back registers to clear the counters */ 1682 for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++) 1683 RREG32(sec_ded_counter_registers[i]); 1684 1685 fail: 1686 amdgpu_ib_free(adev, &ib, NULL); 1687 dma_fence_put(f); 1688 1689 return r; 1690 } 1691 1692 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev) 1693 { 1694 u32 gb_addr_config; 1695 u32 mc_arb_ramcfg; 1696 u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; 1697 u32 tmp; 1698 int ret; 1699 1700 switch (adev->asic_type) { 1701 case CHIP_TOPAZ: 1702 adev->gfx.config.max_shader_engines = 1; 1703 adev->gfx.config.max_tile_pipes = 2; 1704 adev->gfx.config.max_cu_per_sh = 6; 1705 adev->gfx.config.max_sh_per_se = 1; 1706 adev->gfx.config.max_backends_per_se = 2; 1707 adev->gfx.config.max_texture_channel_caches = 2; 1708 adev->gfx.config.max_gprs = 256; 1709 adev->gfx.config.max_gs_threads = 32; 1710 adev->gfx.config.max_hw_contexts = 8; 1711 1712 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1713 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1714 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1715 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1716 gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN; 1717 break; 1718 case CHIP_FIJI: 1719 adev->gfx.config.max_shader_engines = 4; 1720 adev->gfx.config.max_tile_pipes = 16; 1721 adev->gfx.config.max_cu_per_sh = 16; 1722 adev->gfx.config.max_sh_per_se = 1; 1723 adev->gfx.config.max_backends_per_se = 4; 1724 adev->gfx.config.max_texture_channel_caches = 16; 1725 adev->gfx.config.max_gprs = 256; 1726 adev->gfx.config.max_gs_threads = 32; 1727 adev->gfx.config.max_hw_contexts = 8; 1728 1729 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1730 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1731 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1732 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1733 gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; 1734 break; 1735 case CHIP_POLARIS11: 1736 case CHIP_POLARIS12: 1737 ret = amdgpu_atombios_get_gfx_info(adev); 1738 if (ret) 1739 return ret; 1740 adev->gfx.config.max_gprs = 256; 1741 adev->gfx.config.max_gs_threads = 32; 1742 adev->gfx.config.max_hw_contexts = 8; 1743 1744 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1745 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1746 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1747 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1748 gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN; 1749 break; 1750 case CHIP_POLARIS10: 1751 case CHIP_VEGAM: 1752 ret = amdgpu_atombios_get_gfx_info(adev); 1753 if (ret) 1754 return ret; 1755 adev->gfx.config.max_gprs = 256; 1756 adev->gfx.config.max_gs_threads = 32; 1757 adev->gfx.config.max_hw_contexts = 8; 1758 1759 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1760 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1761 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1762 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1763 gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; 1764 break; 1765 case CHIP_TONGA: 1766 adev->gfx.config.max_shader_engines = 4; 1767 adev->gfx.config.max_tile_pipes = 8; 1768 adev->gfx.config.max_cu_per_sh = 8; 1769 adev->gfx.config.max_sh_per_se = 1; 1770 adev->gfx.config.max_backends_per_se = 2; 1771 adev->gfx.config.max_texture_channel_caches = 8; 1772 adev->gfx.config.max_gprs = 256; 1773 adev->gfx.config.max_gs_threads = 32; 1774 adev->gfx.config.max_hw_contexts = 8; 1775 1776 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1777 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1778 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1779 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1780 gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; 1781 break; 1782 case CHIP_CARRIZO: 1783 adev->gfx.config.max_shader_engines = 1; 1784 adev->gfx.config.max_tile_pipes = 2; 1785 adev->gfx.config.max_sh_per_se = 1; 1786 adev->gfx.config.max_backends_per_se = 2; 1787 adev->gfx.config.max_cu_per_sh = 8; 1788 adev->gfx.config.max_texture_channel_caches = 2; 1789 adev->gfx.config.max_gprs = 256; 1790 adev->gfx.config.max_gs_threads = 32; 1791 adev->gfx.config.max_hw_contexts = 8; 1792 1793 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1794 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1795 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1796 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1797 gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN; 1798 break; 1799 case CHIP_STONEY: 1800 adev->gfx.config.max_shader_engines = 1; 1801 adev->gfx.config.max_tile_pipes = 2; 1802 adev->gfx.config.max_sh_per_se = 1; 1803 adev->gfx.config.max_backends_per_se = 1; 1804 adev->gfx.config.max_cu_per_sh = 3; 1805 adev->gfx.config.max_texture_channel_caches = 2; 1806 adev->gfx.config.max_gprs = 256; 1807 adev->gfx.config.max_gs_threads = 16; 1808 adev->gfx.config.max_hw_contexts = 8; 1809 1810 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1811 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1812 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1813 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1814 gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN; 1815 break; 1816 default: 1817 adev->gfx.config.max_shader_engines = 2; 1818 adev->gfx.config.max_tile_pipes = 4; 1819 adev->gfx.config.max_cu_per_sh = 2; 1820 adev->gfx.config.max_sh_per_se = 1; 1821 adev->gfx.config.max_backends_per_se = 2; 1822 adev->gfx.config.max_texture_channel_caches = 4; 1823 adev->gfx.config.max_gprs = 256; 1824 adev->gfx.config.max_gs_threads = 32; 1825 adev->gfx.config.max_hw_contexts = 8; 1826 1827 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1828 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1829 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1830 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1831 gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; 1832 break; 1833 } 1834 1835 adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); 1836 mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; 1837 1838 adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg, 1839 MC_ARB_RAMCFG, NOOFBANK); 1840 adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg, 1841 MC_ARB_RAMCFG, NOOFRANKS); 1842 1843 adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; 1844 adev->gfx.config.mem_max_burst_length_bytes = 256; 1845 if (adev->flags & AMD_IS_APU) { 1846 /* Get memory bank mapping mode. */ 1847 tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); 1848 dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); 1849 dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); 1850 1851 tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); 1852 dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); 1853 dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); 1854 1855 /* Validate settings in case only one DIMM installed. */ 1856 if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) 1857 dimm00_addr_map = 0; 1858 if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) 1859 dimm01_addr_map = 0; 1860 if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) 1861 dimm10_addr_map = 0; 1862 if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) 1863 dimm11_addr_map = 0; 1864 1865 /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ 1866 /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ 1867 if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) 1868 adev->gfx.config.mem_row_size_in_kb = 2; 1869 else 1870 adev->gfx.config.mem_row_size_in_kb = 1; 1871 } else { 1872 tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS); 1873 adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 1874 if (adev->gfx.config.mem_row_size_in_kb > 4) 1875 adev->gfx.config.mem_row_size_in_kb = 4; 1876 } 1877 1878 adev->gfx.config.shader_engine_tile_size = 32; 1879 adev->gfx.config.num_gpus = 1; 1880 adev->gfx.config.multi_gpu_tile_size = 64; 1881 1882 /* fix up row size */ 1883 switch (adev->gfx.config.mem_row_size_in_kb) { 1884 case 1: 1885 default: 1886 gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0); 1887 break; 1888 case 2: 1889 gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1); 1890 break; 1891 case 4: 1892 gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2); 1893 break; 1894 } 1895 adev->gfx.config.gb_addr_config = gb_addr_config; 1896 1897 return 0; 1898 } 1899 1900 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, 1901 int mec, int pipe, int queue) 1902 { 1903 int r; 1904 unsigned irq_type; 1905 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id]; 1906 unsigned int hw_prio; 1907 1908 ring = &adev->gfx.compute_ring[ring_id]; 1909 1910 /* mec0 is me1 */ 1911 ring->me = mec + 1; 1912 ring->pipe = pipe; 1913 ring->queue = queue; 1914 1915 ring->ring_obj = NULL; 1916 ring->use_doorbell = true; 1917 ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id; 1918 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr 1919 + (ring_id * GFX8_MEC_HPD_SIZE); 1920 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); 1921 1922 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP 1923 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec) 1924 + ring->pipe; 1925 1926 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring->pipe, 1927 ring->queue) ? 1928 AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_RING_PRIO_DEFAULT; 1929 /* type-2 packets are deprecated on MEC, use type-3 instead */ 1930 r = amdgpu_ring_init(adev, ring, 1024, 1931 &adev->gfx.eop_irq, irq_type, hw_prio); 1932 if (r) 1933 return r; 1934 1935 1936 return 0; 1937 } 1938 1939 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work); 1940 1941 static int gfx_v8_0_sw_init(void *handle) 1942 { 1943 int i, j, k, r, ring_id; 1944 struct amdgpu_ring *ring; 1945 struct amdgpu_kiq *kiq; 1946 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1947 1948 switch (adev->asic_type) { 1949 case CHIP_TONGA: 1950 case CHIP_CARRIZO: 1951 case CHIP_FIJI: 1952 case CHIP_POLARIS10: 1953 case CHIP_POLARIS11: 1954 case CHIP_POLARIS12: 1955 case CHIP_VEGAM: 1956 adev->gfx.mec.num_mec = 2; 1957 break; 1958 case CHIP_TOPAZ: 1959 case CHIP_STONEY: 1960 default: 1961 adev->gfx.mec.num_mec = 1; 1962 break; 1963 } 1964 1965 adev->gfx.mec.num_pipe_per_mec = 4; 1966 adev->gfx.mec.num_queue_per_pipe = 8; 1967 1968 /* EOP Event */ 1969 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq); 1970 if (r) 1971 return r; 1972 1973 /* Privileged reg */ 1974 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT, 1975 &adev->gfx.priv_reg_irq); 1976 if (r) 1977 return r; 1978 1979 /* Privileged inst */ 1980 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT, 1981 &adev->gfx.priv_inst_irq); 1982 if (r) 1983 return r; 1984 1985 /* Add CP EDC/ECC irq */ 1986 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR, 1987 &adev->gfx.cp_ecc_error_irq); 1988 if (r) 1989 return r; 1990 1991 /* SQ interrupts. */ 1992 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG, 1993 &adev->gfx.sq_irq); 1994 if (r) { 1995 DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r); 1996 return r; 1997 } 1998 1999 INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func); 2000 2001 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE; 2002 2003 gfx_v8_0_scratch_init(adev); 2004 2005 r = gfx_v8_0_init_microcode(adev); 2006 if (r) { 2007 DRM_ERROR("Failed to load gfx firmware!\n"); 2008 return r; 2009 } 2010 2011 r = adev->gfx.rlc.funcs->init(adev); 2012 if (r) { 2013 DRM_ERROR("Failed to init rlc BOs!\n"); 2014 return r; 2015 } 2016 2017 r = gfx_v8_0_mec_init(adev); 2018 if (r) { 2019 DRM_ERROR("Failed to init MEC BOs!\n"); 2020 return r; 2021 } 2022 2023 /* set up the gfx ring */ 2024 for (i = 0; i < adev->gfx.num_gfx_rings; i++) { 2025 ring = &adev->gfx.gfx_ring[i]; 2026 ring->ring_obj = NULL; 2027 sprintf(ring->name, "gfx"); 2028 /* no gfx doorbells on iceland */ 2029 if (adev->asic_type != CHIP_TOPAZ) { 2030 ring->use_doorbell = true; 2031 ring->doorbell_index = adev->doorbell_index.gfx_ring0; 2032 } 2033 2034 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, 2035 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, 2036 AMDGPU_RING_PRIO_DEFAULT); 2037 if (r) 2038 return r; 2039 } 2040 2041 2042 /* set up the compute queues - allocate horizontally across pipes */ 2043 ring_id = 0; 2044 for (i = 0; i < adev->gfx.mec.num_mec; ++i) { 2045 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) { 2046 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { 2047 if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j)) 2048 continue; 2049 2050 r = gfx_v8_0_compute_ring_init(adev, 2051 ring_id, 2052 i, k, j); 2053 if (r) 2054 return r; 2055 2056 ring_id++; 2057 } 2058 } 2059 } 2060 2061 r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE); 2062 if (r) { 2063 DRM_ERROR("Failed to init KIQ BOs!\n"); 2064 return r; 2065 } 2066 2067 kiq = &adev->gfx.kiq; 2068 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq); 2069 if (r) 2070 return r; 2071 2072 /* create MQD for all compute queues as well as KIQ for SRIOV case */ 2073 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation)); 2074 if (r) 2075 return r; 2076 2077 adev->gfx.ce_ram_size = 0x8000; 2078 2079 r = gfx_v8_0_gpu_early_init(adev); 2080 if (r) 2081 return r; 2082 2083 return 0; 2084 } 2085 2086 static int gfx_v8_0_sw_fini(void *handle) 2087 { 2088 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 2089 int i; 2090 2091 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 2092 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]); 2093 for (i = 0; i < adev->gfx.num_compute_rings; i++) 2094 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 2095 2096 amdgpu_gfx_mqd_sw_fini(adev); 2097 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring); 2098 amdgpu_gfx_kiq_fini(adev); 2099 2100 gfx_v8_0_mec_fini(adev); 2101 amdgpu_gfx_rlc_fini(adev); 2102 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, 2103 &adev->gfx.rlc.clear_state_gpu_addr, 2104 (void **)&adev->gfx.rlc.cs_ptr); 2105 if ((adev->asic_type == CHIP_CARRIZO) || 2106 (adev->asic_type == CHIP_STONEY)) { 2107 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, 2108 &adev->gfx.rlc.cp_table_gpu_addr, 2109 (void **)&adev->gfx.rlc.cp_table_ptr); 2110 } 2111 gfx_v8_0_free_microcode(adev); 2112 2113 return 0; 2114 } 2115 2116 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev) 2117 { 2118 uint32_t *modearray, *mod2array; 2119 const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array); 2120 const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array); 2121 u32 reg_offset; 2122 2123 modearray = adev->gfx.config.tile_mode_array; 2124 mod2array = adev->gfx.config.macrotile_mode_array; 2125 2126 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2127 modearray[reg_offset] = 0; 2128 2129 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 2130 mod2array[reg_offset] = 0; 2131 2132 switch (adev->asic_type) { 2133 case CHIP_TOPAZ: 2134 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2135 PIPE_CONFIG(ADDR_SURF_P2) | 2136 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2137 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2138 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2139 PIPE_CONFIG(ADDR_SURF_P2) | 2140 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2141 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2142 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2143 PIPE_CONFIG(ADDR_SURF_P2) | 2144 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2145 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2146 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2147 PIPE_CONFIG(ADDR_SURF_P2) | 2148 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2149 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2150 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2151 PIPE_CONFIG(ADDR_SURF_P2) | 2152 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2153 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2154 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2155 PIPE_CONFIG(ADDR_SURF_P2) | 2156 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2157 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2158 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2159 PIPE_CONFIG(ADDR_SURF_P2) | 2160 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2161 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2162 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2163 PIPE_CONFIG(ADDR_SURF_P2)); 2164 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2165 PIPE_CONFIG(ADDR_SURF_P2) | 2166 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2167 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2168 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2169 PIPE_CONFIG(ADDR_SURF_P2) | 2170 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2171 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2172 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2173 PIPE_CONFIG(ADDR_SURF_P2) | 2174 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2175 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2176 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2177 PIPE_CONFIG(ADDR_SURF_P2) | 2178 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2179 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2180 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2181 PIPE_CONFIG(ADDR_SURF_P2) | 2182 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2183 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2184 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 2185 PIPE_CONFIG(ADDR_SURF_P2) | 2186 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2187 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2188 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2189 PIPE_CONFIG(ADDR_SURF_P2) | 2190 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2191 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2192 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2193 PIPE_CONFIG(ADDR_SURF_P2) | 2194 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2195 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2196 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2197 PIPE_CONFIG(ADDR_SURF_P2) | 2198 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2199 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2200 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2201 PIPE_CONFIG(ADDR_SURF_P2) | 2202 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2203 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2204 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 2205 PIPE_CONFIG(ADDR_SURF_P2) | 2206 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2207 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2208 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2209 PIPE_CONFIG(ADDR_SURF_P2) | 2210 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2211 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2212 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2213 PIPE_CONFIG(ADDR_SURF_P2) | 2214 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2215 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2216 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 2217 PIPE_CONFIG(ADDR_SURF_P2) | 2218 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2219 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2220 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 2221 PIPE_CONFIG(ADDR_SURF_P2) | 2222 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2223 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2224 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2225 PIPE_CONFIG(ADDR_SURF_P2) | 2226 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2227 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2228 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2229 PIPE_CONFIG(ADDR_SURF_P2) | 2230 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2231 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2232 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2233 PIPE_CONFIG(ADDR_SURF_P2) | 2234 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2235 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2236 2237 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 2238 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2239 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2240 NUM_BANKS(ADDR_SURF_8_BANK)); 2241 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 2242 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2243 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2244 NUM_BANKS(ADDR_SURF_8_BANK)); 2245 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2246 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2247 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2248 NUM_BANKS(ADDR_SURF_8_BANK)); 2249 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2250 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2251 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2252 NUM_BANKS(ADDR_SURF_8_BANK)); 2253 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2254 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2255 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2256 NUM_BANKS(ADDR_SURF_8_BANK)); 2257 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2258 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2259 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2260 NUM_BANKS(ADDR_SURF_8_BANK)); 2261 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2262 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2263 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2264 NUM_BANKS(ADDR_SURF_8_BANK)); 2265 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 2266 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 2267 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2268 NUM_BANKS(ADDR_SURF_16_BANK)); 2269 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 2270 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2271 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2272 NUM_BANKS(ADDR_SURF_16_BANK)); 2273 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2274 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2275 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2276 NUM_BANKS(ADDR_SURF_16_BANK)); 2277 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2278 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2279 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2280 NUM_BANKS(ADDR_SURF_16_BANK)); 2281 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2282 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2283 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2284 NUM_BANKS(ADDR_SURF_16_BANK)); 2285 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2286 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2287 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2288 NUM_BANKS(ADDR_SURF_16_BANK)); 2289 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2290 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2291 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2292 NUM_BANKS(ADDR_SURF_8_BANK)); 2293 2294 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2295 if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && 2296 reg_offset != 23) 2297 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 2298 2299 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 2300 if (reg_offset != 7) 2301 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 2302 2303 break; 2304 case CHIP_FIJI: 2305 case CHIP_VEGAM: 2306 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2307 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2308 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2309 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2310 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2311 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2312 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2313 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2314 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2315 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2316 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2317 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2318 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2319 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2320 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2321 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2322 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2323 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2324 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2325 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2326 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2327 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2328 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2329 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2330 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2331 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2332 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2333 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2334 modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2335 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2336 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2337 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2338 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2339 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); 2340 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2341 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2342 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2343 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2344 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2345 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2346 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2347 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2348 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2349 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2350 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2351 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2352 modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2353 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2354 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2355 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2356 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2357 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2358 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2359 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2360 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2361 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2362 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2363 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2364 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 2365 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2366 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2367 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2368 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2369 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2370 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2371 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2372 modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2373 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2374 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2375 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2376 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2377 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2378 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2379 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2380 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2381 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2382 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2383 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2384 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2385 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2386 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2387 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2388 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 2389 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2390 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2391 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2392 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2393 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2394 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2395 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2396 modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2397 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2398 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2399 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2400 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2401 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2402 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2403 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2404 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 2405 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2406 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2407 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2408 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 2409 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2410 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2411 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2412 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2413 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2414 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2415 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2416 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2417 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2418 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2419 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2420 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2421 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 2422 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2423 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2424 modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2425 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2426 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2427 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2428 2429 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2430 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2431 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2432 NUM_BANKS(ADDR_SURF_8_BANK)); 2433 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2434 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2435 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2436 NUM_BANKS(ADDR_SURF_8_BANK)); 2437 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2438 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2439 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2440 NUM_BANKS(ADDR_SURF_8_BANK)); 2441 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2442 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2443 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2444 NUM_BANKS(ADDR_SURF_8_BANK)); 2445 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2446 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2447 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2448 NUM_BANKS(ADDR_SURF_8_BANK)); 2449 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2450 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2451 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2452 NUM_BANKS(ADDR_SURF_8_BANK)); 2453 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2454 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2455 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2456 NUM_BANKS(ADDR_SURF_8_BANK)); 2457 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2458 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 2459 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2460 NUM_BANKS(ADDR_SURF_8_BANK)); 2461 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2462 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2463 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2464 NUM_BANKS(ADDR_SURF_8_BANK)); 2465 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2466 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2467 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2468 NUM_BANKS(ADDR_SURF_8_BANK)); 2469 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2470 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2471 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2472 NUM_BANKS(ADDR_SURF_8_BANK)); 2473 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2474 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2475 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2476 NUM_BANKS(ADDR_SURF_8_BANK)); 2477 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2478 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2479 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2480 NUM_BANKS(ADDR_SURF_8_BANK)); 2481 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2482 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2483 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2484 NUM_BANKS(ADDR_SURF_4_BANK)); 2485 2486 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2487 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 2488 2489 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 2490 if (reg_offset != 7) 2491 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 2492 2493 break; 2494 case CHIP_TONGA: 2495 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2496 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2497 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2498 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2499 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2500 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2501 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2502 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2503 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2504 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2505 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2506 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2507 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2508 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2509 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2510 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2511 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2512 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2513 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2514 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2515 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2516 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2517 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2518 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2519 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2520 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2521 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2522 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2523 modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2524 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2525 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2526 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2527 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2528 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16)); 2529 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2530 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2531 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2532 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2533 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2534 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2535 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2536 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2537 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2538 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2539 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2540 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2541 modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2542 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2543 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2544 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2545 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2546 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2547 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2548 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2549 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2550 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2551 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2552 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2553 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 2554 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2555 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2556 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2557 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2558 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2559 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2560 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2561 modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2562 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2563 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2564 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2565 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2566 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2567 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2568 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2569 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2570 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2571 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2572 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2573 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2574 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2575 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2576 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2577 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 2578 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2579 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2580 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2581 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2582 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2583 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2584 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2585 modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2586 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2587 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2588 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2589 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2590 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2591 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2592 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2593 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 2594 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2595 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2596 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2597 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 2598 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2599 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2600 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2601 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2602 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2603 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2604 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2605 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2606 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2607 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2608 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2609 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2610 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2611 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2612 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2613 modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2614 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2615 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2616 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2617 2618 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2619 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2620 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2621 NUM_BANKS(ADDR_SURF_16_BANK)); 2622 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2623 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2624 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2625 NUM_BANKS(ADDR_SURF_16_BANK)); 2626 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2627 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2628 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2629 NUM_BANKS(ADDR_SURF_16_BANK)); 2630 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2631 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2632 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2633 NUM_BANKS(ADDR_SURF_16_BANK)); 2634 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2635 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2636 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2637 NUM_BANKS(ADDR_SURF_16_BANK)); 2638 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2639 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2640 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2641 NUM_BANKS(ADDR_SURF_16_BANK)); 2642 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2643 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2644 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2645 NUM_BANKS(ADDR_SURF_16_BANK)); 2646 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2647 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 2648 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2649 NUM_BANKS(ADDR_SURF_16_BANK)); 2650 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2651 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2652 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2653 NUM_BANKS(ADDR_SURF_16_BANK)); 2654 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2655 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2656 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2657 NUM_BANKS(ADDR_SURF_16_BANK)); 2658 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2659 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2660 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2661 NUM_BANKS(ADDR_SURF_16_BANK)); 2662 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2663 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2664 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2665 NUM_BANKS(ADDR_SURF_8_BANK)); 2666 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2667 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2668 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2669 NUM_BANKS(ADDR_SURF_4_BANK)); 2670 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2671 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2672 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2673 NUM_BANKS(ADDR_SURF_4_BANK)); 2674 2675 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2676 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 2677 2678 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 2679 if (reg_offset != 7) 2680 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 2681 2682 break; 2683 case CHIP_POLARIS11: 2684 case CHIP_POLARIS12: 2685 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2686 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2687 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2688 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2689 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2690 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2691 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2692 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2693 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2694 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2695 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2696 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2697 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2698 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2699 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2700 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2701 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2702 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2703 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2704 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2705 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2706 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2707 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2708 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2709 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2710 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2711 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2712 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2713 modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2714 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2715 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2716 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2717 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2718 PIPE_CONFIG(ADDR_SURF_P4_16x16)); 2719 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2720 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2721 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2722 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2723 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2724 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2725 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2726 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2727 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2728 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2729 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2730 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2731 modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2732 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2733 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2734 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2735 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2736 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2737 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2738 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2739 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2740 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2741 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2742 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2743 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 2744 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2745 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2746 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2747 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2748 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2749 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2750 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2751 modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2752 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2753 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2754 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2755 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2756 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2757 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2758 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2759 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2760 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2761 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2762 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2763 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2764 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2765 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2766 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2767 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 2768 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2769 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2770 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2771 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2772 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2773 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2774 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2775 modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2776 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2777 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2778 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2779 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2780 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2781 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2782 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2783 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 2784 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2785 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2786 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2787 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 2788 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2789 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2790 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2791 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2792 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2793 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2794 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2795 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2796 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2797 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2798 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2799 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2800 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2801 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2802 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2803 modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2804 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2805 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2806 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2807 2808 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2809 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2810 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2811 NUM_BANKS(ADDR_SURF_16_BANK)); 2812 2813 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2814 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2815 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2816 NUM_BANKS(ADDR_SURF_16_BANK)); 2817 2818 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2819 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2820 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2821 NUM_BANKS(ADDR_SURF_16_BANK)); 2822 2823 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2824 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2825 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2826 NUM_BANKS(ADDR_SURF_16_BANK)); 2827 2828 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2829 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2830 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2831 NUM_BANKS(ADDR_SURF_16_BANK)); 2832 2833 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2834 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2835 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2836 NUM_BANKS(ADDR_SURF_16_BANK)); 2837 2838 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2839 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2840 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2841 NUM_BANKS(ADDR_SURF_16_BANK)); 2842 2843 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2844 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 2845 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2846 NUM_BANKS(ADDR_SURF_16_BANK)); 2847 2848 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2849 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2850 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2851 NUM_BANKS(ADDR_SURF_16_BANK)); 2852 2853 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2854 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2855 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2856 NUM_BANKS(ADDR_SURF_16_BANK)); 2857 2858 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2859 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2860 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 2861 NUM_BANKS(ADDR_SURF_16_BANK)); 2862 2863 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2864 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2865 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2866 NUM_BANKS(ADDR_SURF_16_BANK)); 2867 2868 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2869 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2870 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 2871 NUM_BANKS(ADDR_SURF_8_BANK)); 2872 2873 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2874 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2875 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 2876 NUM_BANKS(ADDR_SURF_4_BANK)); 2877 2878 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2879 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 2880 2881 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 2882 if (reg_offset != 7) 2883 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 2884 2885 break; 2886 case CHIP_POLARIS10: 2887 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2888 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2889 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2890 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2891 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2892 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2893 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2894 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2895 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2896 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2897 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2898 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2899 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2900 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2901 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2902 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2903 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2904 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2905 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2906 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2907 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2908 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2909 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2910 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2911 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2912 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2913 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2914 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2915 modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2916 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2917 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 2918 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 2919 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2920 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16)); 2921 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2922 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2923 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2924 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2925 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2926 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2927 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2928 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2929 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2930 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2931 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2932 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2933 modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2934 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2935 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 2936 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2937 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2938 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2939 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2940 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2941 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2942 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2943 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2944 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2945 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 2946 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2947 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2948 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2949 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2950 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2951 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2952 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2953 modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 2954 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2955 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2956 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 2957 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2958 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2959 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2960 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2961 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 2962 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2963 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2964 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2965 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2966 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2967 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2968 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2969 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 2970 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2971 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2972 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2973 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2974 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2975 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2976 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2977 modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 2978 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 2979 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2980 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2981 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 2982 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2983 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 2984 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2985 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 2986 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2987 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2988 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2989 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 2990 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2991 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 2992 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 2993 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2994 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2995 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 2996 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 2997 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2998 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 2999 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3000 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3001 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3002 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | 3003 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3004 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3005 modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3006 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 3007 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3008 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3009 3010 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3011 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3012 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3013 NUM_BANKS(ADDR_SURF_16_BANK)); 3014 3015 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3016 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3017 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3018 NUM_BANKS(ADDR_SURF_16_BANK)); 3019 3020 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3021 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3022 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3023 NUM_BANKS(ADDR_SURF_16_BANK)); 3024 3025 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3026 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3027 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3028 NUM_BANKS(ADDR_SURF_16_BANK)); 3029 3030 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3031 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3032 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3033 NUM_BANKS(ADDR_SURF_16_BANK)); 3034 3035 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3036 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3037 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 3038 NUM_BANKS(ADDR_SURF_16_BANK)); 3039 3040 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3041 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3042 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 3043 NUM_BANKS(ADDR_SURF_16_BANK)); 3044 3045 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3046 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 3047 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3048 NUM_BANKS(ADDR_SURF_16_BANK)); 3049 3050 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3051 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3052 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3053 NUM_BANKS(ADDR_SURF_16_BANK)); 3054 3055 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3056 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3057 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3058 NUM_BANKS(ADDR_SURF_16_BANK)); 3059 3060 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3061 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3062 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3063 NUM_BANKS(ADDR_SURF_16_BANK)); 3064 3065 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3066 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3067 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 3068 NUM_BANKS(ADDR_SURF_8_BANK)); 3069 3070 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3071 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3072 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 3073 NUM_BANKS(ADDR_SURF_4_BANK)); 3074 3075 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3076 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3077 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 3078 NUM_BANKS(ADDR_SURF_4_BANK)); 3079 3080 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 3081 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 3082 3083 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 3084 if (reg_offset != 7) 3085 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 3086 3087 break; 3088 case CHIP_STONEY: 3089 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3090 PIPE_CONFIG(ADDR_SURF_P2) | 3091 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 3092 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3093 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3094 PIPE_CONFIG(ADDR_SURF_P2) | 3095 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 3096 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3097 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3098 PIPE_CONFIG(ADDR_SURF_P2) | 3099 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 3100 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3101 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3102 PIPE_CONFIG(ADDR_SURF_P2) | 3103 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 3104 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3105 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3106 PIPE_CONFIG(ADDR_SURF_P2) | 3107 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3108 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3109 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3110 PIPE_CONFIG(ADDR_SURF_P2) | 3111 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3112 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3113 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3114 PIPE_CONFIG(ADDR_SURF_P2) | 3115 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3116 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3117 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 3118 PIPE_CONFIG(ADDR_SURF_P2)); 3119 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3120 PIPE_CONFIG(ADDR_SURF_P2) | 3121 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3122 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3123 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3124 PIPE_CONFIG(ADDR_SURF_P2) | 3125 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3126 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3127 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3128 PIPE_CONFIG(ADDR_SURF_P2) | 3129 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3130 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3131 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3132 PIPE_CONFIG(ADDR_SURF_P2) | 3133 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3134 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3135 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3136 PIPE_CONFIG(ADDR_SURF_P2) | 3137 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3138 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3139 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 3140 PIPE_CONFIG(ADDR_SURF_P2) | 3141 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3142 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3143 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3144 PIPE_CONFIG(ADDR_SURF_P2) | 3145 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3146 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3147 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 3148 PIPE_CONFIG(ADDR_SURF_P2) | 3149 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3150 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3151 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 3152 PIPE_CONFIG(ADDR_SURF_P2) | 3153 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3154 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3155 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 3156 PIPE_CONFIG(ADDR_SURF_P2) | 3157 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3158 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3159 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 3160 PIPE_CONFIG(ADDR_SURF_P2) | 3161 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3162 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3163 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 3164 PIPE_CONFIG(ADDR_SURF_P2) | 3165 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3166 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3167 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 3168 PIPE_CONFIG(ADDR_SURF_P2) | 3169 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3170 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3171 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 3172 PIPE_CONFIG(ADDR_SURF_P2) | 3173 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3174 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3175 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 3176 PIPE_CONFIG(ADDR_SURF_P2) | 3177 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3178 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3179 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3180 PIPE_CONFIG(ADDR_SURF_P2) | 3181 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3182 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3183 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3184 PIPE_CONFIG(ADDR_SURF_P2) | 3185 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3186 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3187 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3188 PIPE_CONFIG(ADDR_SURF_P2) | 3189 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3190 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3191 3192 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3193 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3194 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3195 NUM_BANKS(ADDR_SURF_8_BANK)); 3196 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3197 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3198 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3199 NUM_BANKS(ADDR_SURF_8_BANK)); 3200 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3201 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3202 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3203 NUM_BANKS(ADDR_SURF_8_BANK)); 3204 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3205 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3206 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3207 NUM_BANKS(ADDR_SURF_8_BANK)); 3208 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3209 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3210 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3211 NUM_BANKS(ADDR_SURF_8_BANK)); 3212 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3213 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3214 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3215 NUM_BANKS(ADDR_SURF_8_BANK)); 3216 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3217 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3218 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3219 NUM_BANKS(ADDR_SURF_8_BANK)); 3220 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 3221 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 3222 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3223 NUM_BANKS(ADDR_SURF_16_BANK)); 3224 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 3225 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3226 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3227 NUM_BANKS(ADDR_SURF_16_BANK)); 3228 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 3229 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3230 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3231 NUM_BANKS(ADDR_SURF_16_BANK)); 3232 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 3233 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3234 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3235 NUM_BANKS(ADDR_SURF_16_BANK)); 3236 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3237 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3238 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3239 NUM_BANKS(ADDR_SURF_16_BANK)); 3240 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3241 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3242 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3243 NUM_BANKS(ADDR_SURF_16_BANK)); 3244 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3245 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3246 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3247 NUM_BANKS(ADDR_SURF_8_BANK)); 3248 3249 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 3250 if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && 3251 reg_offset != 23) 3252 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 3253 3254 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 3255 if (reg_offset != 7) 3256 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 3257 3258 break; 3259 default: 3260 dev_warn(adev->dev, 3261 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n", 3262 adev->asic_type); 3263 fallthrough; 3264 3265 case CHIP_CARRIZO: 3266 modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3267 PIPE_CONFIG(ADDR_SURF_P2) | 3268 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 3269 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3270 modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3271 PIPE_CONFIG(ADDR_SURF_P2) | 3272 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 3273 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3274 modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3275 PIPE_CONFIG(ADDR_SURF_P2) | 3276 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 3277 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3278 modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3279 PIPE_CONFIG(ADDR_SURF_P2) | 3280 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 3281 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3282 modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3283 PIPE_CONFIG(ADDR_SURF_P2) | 3284 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3285 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3286 modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3287 PIPE_CONFIG(ADDR_SURF_P2) | 3288 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3289 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3290 modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3291 PIPE_CONFIG(ADDR_SURF_P2) | 3292 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 3293 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 3294 modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 3295 PIPE_CONFIG(ADDR_SURF_P2)); 3296 modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3297 PIPE_CONFIG(ADDR_SURF_P2) | 3298 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3299 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3300 modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3301 PIPE_CONFIG(ADDR_SURF_P2) | 3302 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3303 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3304 modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3305 PIPE_CONFIG(ADDR_SURF_P2) | 3306 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 3307 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3308 modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3309 PIPE_CONFIG(ADDR_SURF_P2) | 3310 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3311 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3312 modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3313 PIPE_CONFIG(ADDR_SURF_P2) | 3314 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3315 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3316 modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 3317 PIPE_CONFIG(ADDR_SURF_P2) | 3318 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3319 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3320 modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3321 PIPE_CONFIG(ADDR_SURF_P2) | 3322 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3323 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3324 modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 3325 PIPE_CONFIG(ADDR_SURF_P2) | 3326 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3327 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3328 modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 3329 PIPE_CONFIG(ADDR_SURF_P2) | 3330 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3331 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3332 modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 3333 PIPE_CONFIG(ADDR_SURF_P2) | 3334 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3335 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3336 modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 3337 PIPE_CONFIG(ADDR_SURF_P2) | 3338 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3339 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3340 modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 3341 PIPE_CONFIG(ADDR_SURF_P2) | 3342 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3343 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3344 modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 3345 PIPE_CONFIG(ADDR_SURF_P2) | 3346 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 3347 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3348 modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 3349 PIPE_CONFIG(ADDR_SURF_P2) | 3350 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3351 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3352 modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 3353 PIPE_CONFIG(ADDR_SURF_P2) | 3354 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 3355 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 3356 modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 3357 PIPE_CONFIG(ADDR_SURF_P2) | 3358 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3359 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3360 modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 3361 PIPE_CONFIG(ADDR_SURF_P2) | 3362 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3363 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 3364 modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 3365 PIPE_CONFIG(ADDR_SURF_P2) | 3366 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 3367 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 3368 3369 mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3370 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3371 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3372 NUM_BANKS(ADDR_SURF_8_BANK)); 3373 mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3374 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3375 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3376 NUM_BANKS(ADDR_SURF_8_BANK)); 3377 mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3378 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3379 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3380 NUM_BANKS(ADDR_SURF_8_BANK)); 3381 mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3382 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3383 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3384 NUM_BANKS(ADDR_SURF_8_BANK)); 3385 mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3386 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3387 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3388 NUM_BANKS(ADDR_SURF_8_BANK)); 3389 mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3390 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3391 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3392 NUM_BANKS(ADDR_SURF_8_BANK)); 3393 mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3394 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3395 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3396 NUM_BANKS(ADDR_SURF_8_BANK)); 3397 mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 3398 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 3399 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3400 NUM_BANKS(ADDR_SURF_16_BANK)); 3401 mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 3402 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3403 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3404 NUM_BANKS(ADDR_SURF_16_BANK)); 3405 mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 3406 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 3407 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3408 NUM_BANKS(ADDR_SURF_16_BANK)); 3409 mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 3410 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3411 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3412 NUM_BANKS(ADDR_SURF_16_BANK)); 3413 mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3414 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 3415 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3416 NUM_BANKS(ADDR_SURF_16_BANK)); 3417 mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3418 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3419 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 3420 NUM_BANKS(ADDR_SURF_16_BANK)); 3421 mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 3422 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 3423 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 3424 NUM_BANKS(ADDR_SURF_8_BANK)); 3425 3426 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 3427 if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && 3428 reg_offset != 23) 3429 WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); 3430 3431 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 3432 if (reg_offset != 7) 3433 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); 3434 3435 break; 3436 } 3437 } 3438 3439 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev, 3440 u32 se_num, u32 sh_num, u32 instance) 3441 { 3442 u32 data; 3443 3444 if (instance == 0xffffffff) 3445 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1); 3446 else 3447 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 3448 3449 if (se_num == 0xffffffff) 3450 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1); 3451 else 3452 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num); 3453 3454 if (sh_num == 0xffffffff) 3455 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1); 3456 else 3457 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num); 3458 3459 WREG32(mmGRBM_GFX_INDEX, data); 3460 } 3461 3462 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev, 3463 u32 me, u32 pipe, u32 q, u32 vm) 3464 { 3465 vi_srbm_select(adev, me, pipe, q, vm); 3466 } 3467 3468 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev) 3469 { 3470 u32 data, mask; 3471 3472 data = RREG32(mmCC_RB_BACKEND_DISABLE) | 3473 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 3474 3475 data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE); 3476 3477 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se / 3478 adev->gfx.config.max_sh_per_se); 3479 3480 return (~data) & mask; 3481 } 3482 3483 static void 3484 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1) 3485 { 3486 switch (adev->asic_type) { 3487 case CHIP_FIJI: 3488 case CHIP_VEGAM: 3489 *rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) | 3490 RB_XSEL2(1) | PKR_MAP(2) | 3491 PKR_XSEL(1) | PKR_YSEL(1) | 3492 SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3); 3493 *rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) | 3494 SE_PAIR_YSEL(2); 3495 break; 3496 case CHIP_TONGA: 3497 case CHIP_POLARIS10: 3498 *rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) | 3499 SE_XSEL(1) | SE_YSEL(1); 3500 *rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) | 3501 SE_PAIR_YSEL(2); 3502 break; 3503 case CHIP_TOPAZ: 3504 case CHIP_CARRIZO: 3505 *rconf |= RB_MAP_PKR0(2); 3506 *rconf1 |= 0x0; 3507 break; 3508 case CHIP_POLARIS11: 3509 case CHIP_POLARIS12: 3510 *rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) | 3511 SE_XSEL(1) | SE_YSEL(1); 3512 *rconf1 |= 0x0; 3513 break; 3514 case CHIP_STONEY: 3515 *rconf |= 0x0; 3516 *rconf1 |= 0x0; 3517 break; 3518 default: 3519 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 3520 break; 3521 } 3522 } 3523 3524 static void 3525 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev, 3526 u32 raster_config, u32 raster_config_1, 3527 unsigned rb_mask, unsigned num_rb) 3528 { 3529 unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1); 3530 unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1); 3531 unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2); 3532 unsigned rb_per_se = num_rb / num_se; 3533 unsigned se_mask[4]; 3534 unsigned se; 3535 3536 se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask; 3537 se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask; 3538 se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask; 3539 se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask; 3540 3541 WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4)); 3542 WARN_ON(!(sh_per_se == 1 || sh_per_se == 2)); 3543 WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2)); 3544 3545 if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) || 3546 (!se_mask[2] && !se_mask[3]))) { 3547 raster_config_1 &= ~SE_PAIR_MAP_MASK; 3548 3549 if (!se_mask[0] && !se_mask[1]) { 3550 raster_config_1 |= 3551 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3); 3552 } else { 3553 raster_config_1 |= 3554 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0); 3555 } 3556 } 3557 3558 for (se = 0; se < num_se; se++) { 3559 unsigned raster_config_se = raster_config; 3560 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se); 3561 unsigned pkr1_mask = pkr0_mask << rb_per_pkr; 3562 int idx = (se / 2) * 2; 3563 3564 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) { 3565 raster_config_se &= ~SE_MAP_MASK; 3566 3567 if (!se_mask[idx]) { 3568 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3); 3569 } else { 3570 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0); 3571 } 3572 } 3573 3574 pkr0_mask &= rb_mask; 3575 pkr1_mask &= rb_mask; 3576 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) { 3577 raster_config_se &= ~PKR_MAP_MASK; 3578 3579 if (!pkr0_mask) { 3580 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3); 3581 } else { 3582 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0); 3583 } 3584 } 3585 3586 if (rb_per_se >= 2) { 3587 unsigned rb0_mask = 1 << (se * rb_per_se); 3588 unsigned rb1_mask = rb0_mask << 1; 3589 3590 rb0_mask &= rb_mask; 3591 rb1_mask &= rb_mask; 3592 if (!rb0_mask || !rb1_mask) { 3593 raster_config_se &= ~RB_MAP_PKR0_MASK; 3594 3595 if (!rb0_mask) { 3596 raster_config_se |= 3597 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3); 3598 } else { 3599 raster_config_se |= 3600 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0); 3601 } 3602 } 3603 3604 if (rb_per_se > 2) { 3605 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr); 3606 rb1_mask = rb0_mask << 1; 3607 rb0_mask &= rb_mask; 3608 rb1_mask &= rb_mask; 3609 if (!rb0_mask || !rb1_mask) { 3610 raster_config_se &= ~RB_MAP_PKR1_MASK; 3611 3612 if (!rb0_mask) { 3613 raster_config_se |= 3614 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3); 3615 } else { 3616 raster_config_se |= 3617 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0); 3618 } 3619 } 3620 } 3621 } 3622 3623 /* GRBM_GFX_INDEX has a different offset on VI */ 3624 gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff); 3625 WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se); 3626 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1); 3627 } 3628 3629 /* GRBM_GFX_INDEX has a different offset on VI */ 3630 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3631 } 3632 3633 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) 3634 { 3635 int i, j; 3636 u32 data; 3637 u32 raster_config = 0, raster_config_1 = 0; 3638 u32 active_rbs = 0; 3639 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / 3640 adev->gfx.config.max_sh_per_se; 3641 unsigned num_rb_pipes; 3642 3643 mutex_lock(&adev->grbm_idx_mutex); 3644 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3645 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3646 gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff); 3647 data = gfx_v8_0_get_rb_active_bitmap(adev); 3648 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * 3649 rb_bitmap_width_per_sh); 3650 } 3651 } 3652 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3653 3654 adev->gfx.config.backend_enable_mask = active_rbs; 3655 adev->gfx.config.num_rbs = hweight32(active_rbs); 3656 3657 num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se * 3658 adev->gfx.config.max_shader_engines, 16); 3659 3660 gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1); 3661 3662 if (!adev->gfx.config.backend_enable_mask || 3663 adev->gfx.config.num_rbs >= num_rb_pipes) { 3664 WREG32(mmPA_SC_RASTER_CONFIG, raster_config); 3665 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1); 3666 } else { 3667 gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1, 3668 adev->gfx.config.backend_enable_mask, 3669 num_rb_pipes); 3670 } 3671 3672 /* cache the values for userspace */ 3673 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3674 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3675 gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff); 3676 adev->gfx.config.rb_config[i][j].rb_backend_disable = 3677 RREG32(mmCC_RB_BACKEND_DISABLE); 3678 adev->gfx.config.rb_config[i][j].user_rb_backend_disable = 3679 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 3680 adev->gfx.config.rb_config[i][j].raster_config = 3681 RREG32(mmPA_SC_RASTER_CONFIG); 3682 adev->gfx.config.rb_config[i][j].raster_config_1 = 3683 RREG32(mmPA_SC_RASTER_CONFIG_1); 3684 } 3685 } 3686 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3687 mutex_unlock(&adev->grbm_idx_mutex); 3688 } 3689 3690 #define DEFAULT_SH_MEM_BASES (0x6000) 3691 /** 3692 * gfx_v8_0_init_compute_vmid - gart enable 3693 * 3694 * @adev: amdgpu_device pointer 3695 * 3696 * Initialize compute vmid sh_mem registers 3697 * 3698 */ 3699 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev) 3700 { 3701 int i; 3702 uint32_t sh_mem_config; 3703 uint32_t sh_mem_bases; 3704 3705 /* 3706 * Configure apertures: 3707 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB) 3708 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB) 3709 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB) 3710 */ 3711 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16); 3712 3713 sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 << 3714 SH_MEM_CONFIG__ADDRESS_MODE__SHIFT | 3715 SH_MEM_ALIGNMENT_MODE_UNALIGNED << 3716 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT | 3717 MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT | 3718 SH_MEM_CONFIG__PRIVATE_ATC_MASK; 3719 3720 mutex_lock(&adev->srbm_mutex); 3721 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 3722 vi_srbm_select(adev, 0, 0, 0, i); 3723 /* CP and shaders */ 3724 WREG32(mmSH_MEM_CONFIG, sh_mem_config); 3725 WREG32(mmSH_MEM_APE1_BASE, 1); 3726 WREG32(mmSH_MEM_APE1_LIMIT, 0); 3727 WREG32(mmSH_MEM_BASES, sh_mem_bases); 3728 } 3729 vi_srbm_select(adev, 0, 0, 0, 0); 3730 mutex_unlock(&adev->srbm_mutex); 3731 3732 /* Initialize all compute VMIDs to have no GDS, GWS, or OA 3733 acccess. These should be enabled by FW for target VMIDs. */ 3734 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 3735 WREG32(amdgpu_gds_reg_offset[i].mem_base, 0); 3736 WREG32(amdgpu_gds_reg_offset[i].mem_size, 0); 3737 WREG32(amdgpu_gds_reg_offset[i].gws, 0); 3738 WREG32(amdgpu_gds_reg_offset[i].oa, 0); 3739 } 3740 } 3741 3742 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev) 3743 { 3744 int vmid; 3745 3746 /* 3747 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA 3748 * access. Compute VMIDs should be enabled by FW for target VMIDs, 3749 * the driver can enable them for graphics. VMID0 should maintain 3750 * access so that HWS firmware can save/restore entries. 3751 */ 3752 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) { 3753 WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0); 3754 WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0); 3755 WREG32(amdgpu_gds_reg_offset[vmid].gws, 0); 3756 WREG32(amdgpu_gds_reg_offset[vmid].oa, 0); 3757 } 3758 } 3759 3760 static void gfx_v8_0_config_init(struct amdgpu_device *adev) 3761 { 3762 switch (adev->asic_type) { 3763 default: 3764 adev->gfx.config.double_offchip_lds_buf = 1; 3765 break; 3766 case CHIP_CARRIZO: 3767 case CHIP_STONEY: 3768 adev->gfx.config.double_offchip_lds_buf = 0; 3769 break; 3770 } 3771 } 3772 3773 static void gfx_v8_0_constants_init(struct amdgpu_device *adev) 3774 { 3775 u32 tmp, sh_static_mem_cfg; 3776 int i; 3777 3778 WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF); 3779 WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 3780 WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 3781 WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config); 3782 3783 gfx_v8_0_tiling_mode_table_init(adev); 3784 gfx_v8_0_setup_rb(adev); 3785 gfx_v8_0_get_cu_info(adev); 3786 gfx_v8_0_config_init(adev); 3787 3788 /* XXX SH_MEM regs */ 3789 /* where to put LDS, scratch, GPUVM in FSA64 space */ 3790 sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG, 3791 SWIZZLE_ENABLE, 1); 3792 sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG, 3793 ELEMENT_SIZE, 1); 3794 sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG, 3795 INDEX_STRIDE, 3); 3796 WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg); 3797 3798 mutex_lock(&adev->srbm_mutex); 3799 for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) { 3800 vi_srbm_select(adev, 0, 0, 0, i); 3801 /* CP and shaders */ 3802 if (i == 0) { 3803 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC); 3804 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC); 3805 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, 3806 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 3807 WREG32(mmSH_MEM_CONFIG, tmp); 3808 WREG32(mmSH_MEM_BASES, 0); 3809 } else { 3810 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC); 3811 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC); 3812 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, 3813 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 3814 WREG32(mmSH_MEM_CONFIG, tmp); 3815 tmp = adev->gmc.shared_aperture_start >> 48; 3816 WREG32(mmSH_MEM_BASES, tmp); 3817 } 3818 3819 WREG32(mmSH_MEM_APE1_BASE, 1); 3820 WREG32(mmSH_MEM_APE1_LIMIT, 0); 3821 } 3822 vi_srbm_select(adev, 0, 0, 0, 0); 3823 mutex_unlock(&adev->srbm_mutex); 3824 3825 gfx_v8_0_init_compute_vmid(adev); 3826 gfx_v8_0_init_gds_vmid(adev); 3827 3828 mutex_lock(&adev->grbm_idx_mutex); 3829 /* 3830 * making sure that the following register writes will be broadcasted 3831 * to all the shaders 3832 */ 3833 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3834 3835 WREG32(mmPA_SC_FIFO_SIZE, 3836 (adev->gfx.config.sc_prim_fifo_size_frontend << 3837 PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) | 3838 (adev->gfx.config.sc_prim_fifo_size_backend << 3839 PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) | 3840 (adev->gfx.config.sc_hiz_tile_fifo_size << 3841 PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) | 3842 (adev->gfx.config.sc_earlyz_tile_fifo_size << 3843 PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)); 3844 3845 tmp = RREG32(mmSPI_ARB_PRIORITY); 3846 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2); 3847 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2); 3848 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2); 3849 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2); 3850 WREG32(mmSPI_ARB_PRIORITY, tmp); 3851 3852 mutex_unlock(&adev->grbm_idx_mutex); 3853 3854 } 3855 3856 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev) 3857 { 3858 u32 i, j, k; 3859 u32 mask; 3860 3861 mutex_lock(&adev->grbm_idx_mutex); 3862 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3863 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3864 gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff); 3865 for (k = 0; k < adev->usec_timeout; k++) { 3866 if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0) 3867 break; 3868 udelay(1); 3869 } 3870 if (k == adev->usec_timeout) { 3871 gfx_v8_0_select_se_sh(adev, 0xffffffff, 3872 0xffffffff, 0xffffffff); 3873 mutex_unlock(&adev->grbm_idx_mutex); 3874 DRM_INFO("Timeout wait for RLC serdes %u,%u\n", 3875 i, j); 3876 return; 3877 } 3878 } 3879 } 3880 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3881 mutex_unlock(&adev->grbm_idx_mutex); 3882 3883 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK | 3884 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK | 3885 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK | 3886 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK; 3887 for (k = 0; k < adev->usec_timeout; k++) { 3888 if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) 3889 break; 3890 udelay(1); 3891 } 3892 } 3893 3894 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 3895 bool enable) 3896 { 3897 u32 tmp = RREG32(mmCP_INT_CNTL_RING0); 3898 3899 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0); 3900 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0); 3901 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0); 3902 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0); 3903 3904 WREG32(mmCP_INT_CNTL_RING0, tmp); 3905 } 3906 3907 static void gfx_v8_0_init_csb(struct amdgpu_device *adev) 3908 { 3909 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr); 3910 /* csib */ 3911 WREG32(mmRLC_CSIB_ADDR_HI, 3912 adev->gfx.rlc.clear_state_gpu_addr >> 32); 3913 WREG32(mmRLC_CSIB_ADDR_LO, 3914 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc); 3915 WREG32(mmRLC_CSIB_LENGTH, 3916 adev->gfx.rlc.clear_state_size); 3917 } 3918 3919 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format, 3920 int ind_offset, 3921 int list_size, 3922 int *unique_indices, 3923 int *indices_count, 3924 int max_indices, 3925 int *ind_start_offsets, 3926 int *offset_count, 3927 int max_offset) 3928 { 3929 int indices; 3930 bool new_entry = true; 3931 3932 for (; ind_offset < list_size; ind_offset++) { 3933 3934 if (new_entry) { 3935 new_entry = false; 3936 ind_start_offsets[*offset_count] = ind_offset; 3937 *offset_count = *offset_count + 1; 3938 BUG_ON(*offset_count >= max_offset); 3939 } 3940 3941 if (register_list_format[ind_offset] == 0xFFFFFFFF) { 3942 new_entry = true; 3943 continue; 3944 } 3945 3946 ind_offset += 2; 3947 3948 /* look for the matching indice */ 3949 for (indices = 0; 3950 indices < *indices_count; 3951 indices++) { 3952 if (unique_indices[indices] == 3953 register_list_format[ind_offset]) 3954 break; 3955 } 3956 3957 if (indices >= *indices_count) { 3958 unique_indices[*indices_count] = 3959 register_list_format[ind_offset]; 3960 indices = *indices_count; 3961 *indices_count = *indices_count + 1; 3962 BUG_ON(*indices_count >= max_indices); 3963 } 3964 3965 register_list_format[ind_offset] = indices; 3966 } 3967 } 3968 3969 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev) 3970 { 3971 int i, temp, data; 3972 int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0}; 3973 int indices_count = 0; 3974 int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; 3975 int offset_count = 0; 3976 3977 int list_size; 3978 unsigned int *register_list_format = 3979 kmemdup(adev->gfx.rlc.register_list_format, 3980 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL); 3981 if (!register_list_format) 3982 return -ENOMEM; 3983 3984 gfx_v8_0_parse_ind_reg_list(register_list_format, 3985 RLC_FormatDirectRegListLength, 3986 adev->gfx.rlc.reg_list_format_size_bytes >> 2, 3987 unique_indices, 3988 &indices_count, 3989 ARRAY_SIZE(unique_indices), 3990 indirect_start_offsets, 3991 &offset_count, 3992 ARRAY_SIZE(indirect_start_offsets)); 3993 3994 /* save and restore list */ 3995 WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1); 3996 3997 WREG32(mmRLC_SRM_ARAM_ADDR, 0); 3998 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++) 3999 WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]); 4000 4001 /* indirect list */ 4002 WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start); 4003 for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++) 4004 WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]); 4005 4006 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2; 4007 list_size = list_size >> 1; 4008 WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size); 4009 WREG32(mmRLC_GPM_SCRATCH_DATA, list_size); 4010 4011 /* starting offsets starts */ 4012 WREG32(mmRLC_GPM_SCRATCH_ADDR, 4013 adev->gfx.rlc.starting_offsets_start); 4014 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++) 4015 WREG32(mmRLC_GPM_SCRATCH_DATA, 4016 indirect_start_offsets[i]); 4017 4018 /* unique indices */ 4019 temp = mmRLC_SRM_INDEX_CNTL_ADDR_0; 4020 data = mmRLC_SRM_INDEX_CNTL_DATA_0; 4021 for (i = 0; i < ARRAY_SIZE(unique_indices); i++) { 4022 if (unique_indices[i] != 0) { 4023 WREG32(temp + i, unique_indices[i] & 0x3FFFF); 4024 WREG32(data + i, unique_indices[i] >> 20); 4025 } 4026 } 4027 kfree(register_list_format); 4028 4029 return 0; 4030 } 4031 4032 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev) 4033 { 4034 WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1); 4035 } 4036 4037 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev) 4038 { 4039 uint32_t data; 4040 4041 WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60); 4042 4043 data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10); 4044 data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10); 4045 data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10); 4046 data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10); 4047 WREG32(mmRLC_PG_DELAY, data); 4048 4049 WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3); 4050 WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0); 4051 4052 } 4053 4054 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev, 4055 bool enable) 4056 { 4057 WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0); 4058 } 4059 4060 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev, 4061 bool enable) 4062 { 4063 WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0); 4064 } 4065 4066 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable) 4067 { 4068 WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1); 4069 } 4070 4071 static void gfx_v8_0_init_pg(struct amdgpu_device *adev) 4072 { 4073 if ((adev->asic_type == CHIP_CARRIZO) || 4074 (adev->asic_type == CHIP_STONEY)) { 4075 gfx_v8_0_init_csb(adev); 4076 gfx_v8_0_init_save_restore_list(adev); 4077 gfx_v8_0_enable_save_restore_machine(adev); 4078 WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8); 4079 gfx_v8_0_init_power_gating(adev); 4080 WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask); 4081 } else if ((adev->asic_type == CHIP_POLARIS11) || 4082 (adev->asic_type == CHIP_POLARIS12) || 4083 (adev->asic_type == CHIP_VEGAM)) { 4084 gfx_v8_0_init_csb(adev); 4085 gfx_v8_0_init_save_restore_list(adev); 4086 gfx_v8_0_enable_save_restore_machine(adev); 4087 gfx_v8_0_init_power_gating(adev); 4088 } 4089 4090 } 4091 4092 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev) 4093 { 4094 WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0); 4095 4096 gfx_v8_0_enable_gui_idle_interrupt(adev, false); 4097 gfx_v8_0_wait_for_rlc_serdes(adev); 4098 } 4099 4100 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev) 4101 { 4102 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 4103 udelay(50); 4104 4105 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0); 4106 udelay(50); 4107 } 4108 4109 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev) 4110 { 4111 WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1); 4112 4113 /* carrizo do enable cp interrupt after cp inited */ 4114 if (!(adev->flags & AMD_IS_APU)) 4115 gfx_v8_0_enable_gui_idle_interrupt(adev, true); 4116 4117 udelay(50); 4118 } 4119 4120 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev) 4121 { 4122 if (amdgpu_sriov_vf(adev)) { 4123 gfx_v8_0_init_csb(adev); 4124 return 0; 4125 } 4126 4127 adev->gfx.rlc.funcs->stop(adev); 4128 adev->gfx.rlc.funcs->reset(adev); 4129 gfx_v8_0_init_pg(adev); 4130 adev->gfx.rlc.funcs->start(adev); 4131 4132 return 0; 4133 } 4134 4135 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 4136 { 4137 u32 tmp = RREG32(mmCP_ME_CNTL); 4138 4139 if (enable) { 4140 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0); 4141 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0); 4142 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0); 4143 } else { 4144 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1); 4145 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1); 4146 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1); 4147 } 4148 WREG32(mmCP_ME_CNTL, tmp); 4149 udelay(50); 4150 } 4151 4152 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev) 4153 { 4154 u32 count = 0; 4155 const struct cs_section_def *sect = NULL; 4156 const struct cs_extent_def *ext = NULL; 4157 4158 /* begin clear state */ 4159 count += 2; 4160 /* context control state */ 4161 count += 3; 4162 4163 for (sect = vi_cs_data; sect->section != NULL; ++sect) { 4164 for (ext = sect->section; ext->extent != NULL; ++ext) { 4165 if (sect->id == SECT_CONTEXT) 4166 count += 2 + ext->reg_count; 4167 else 4168 return 0; 4169 } 4170 } 4171 /* pa_sc_raster_config/pa_sc_raster_config1 */ 4172 count += 4; 4173 /* end clear state */ 4174 count += 2; 4175 /* clear state */ 4176 count += 2; 4177 4178 return count; 4179 } 4180 4181 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev) 4182 { 4183 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0]; 4184 const struct cs_section_def *sect = NULL; 4185 const struct cs_extent_def *ext = NULL; 4186 int r, i; 4187 4188 /* init the CP */ 4189 WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1); 4190 WREG32(mmCP_ENDIAN_SWAP, 0); 4191 WREG32(mmCP_DEVICE_ID, 1); 4192 4193 gfx_v8_0_cp_gfx_enable(adev, true); 4194 4195 r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4); 4196 if (r) { 4197 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 4198 return r; 4199 } 4200 4201 /* clear state buffer */ 4202 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 4203 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 4204 4205 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 4206 amdgpu_ring_write(ring, 0x80000000); 4207 amdgpu_ring_write(ring, 0x80000000); 4208 4209 for (sect = vi_cs_data; sect->section != NULL; ++sect) { 4210 for (ext = sect->section; ext->extent != NULL; ++ext) { 4211 if (sect->id == SECT_CONTEXT) { 4212 amdgpu_ring_write(ring, 4213 PACKET3(PACKET3_SET_CONTEXT_REG, 4214 ext->reg_count)); 4215 amdgpu_ring_write(ring, 4216 ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 4217 for (i = 0; i < ext->reg_count; i++) 4218 amdgpu_ring_write(ring, ext->extent[i]); 4219 } 4220 } 4221 } 4222 4223 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 4224 amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 4225 amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config); 4226 amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1); 4227 4228 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 4229 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 4230 4231 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 4232 amdgpu_ring_write(ring, 0); 4233 4234 /* init the CE partitions */ 4235 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 4236 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 4237 amdgpu_ring_write(ring, 0x8000); 4238 amdgpu_ring_write(ring, 0x8000); 4239 4240 amdgpu_ring_commit(ring); 4241 4242 return 0; 4243 } 4244 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring) 4245 { 4246 u32 tmp; 4247 /* no gfx doorbells on iceland */ 4248 if (adev->asic_type == CHIP_TOPAZ) 4249 return; 4250 4251 tmp = RREG32(mmCP_RB_DOORBELL_CONTROL); 4252 4253 if (ring->use_doorbell) { 4254 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 4255 DOORBELL_OFFSET, ring->doorbell_index); 4256 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 4257 DOORBELL_HIT, 0); 4258 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 4259 DOORBELL_EN, 1); 4260 } else { 4261 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0); 4262 } 4263 4264 WREG32(mmCP_RB_DOORBELL_CONTROL, tmp); 4265 4266 if (adev->flags & AMD_IS_APU) 4267 return; 4268 4269 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER, 4270 DOORBELL_RANGE_LOWER, 4271 adev->doorbell_index.gfx_ring0); 4272 WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp); 4273 4274 WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, 4275 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); 4276 } 4277 4278 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev) 4279 { 4280 struct amdgpu_ring *ring; 4281 u32 tmp; 4282 u32 rb_bufsz; 4283 u64 rb_addr, rptr_addr, wptr_gpu_addr; 4284 4285 /* Set the write pointer delay */ 4286 WREG32(mmCP_RB_WPTR_DELAY, 0); 4287 4288 /* set the RB to use vmid 0 */ 4289 WREG32(mmCP_RB_VMID, 0); 4290 4291 /* Set ring buffer size */ 4292 ring = &adev->gfx.gfx_ring[0]; 4293 rb_bufsz = order_base_2(ring->ring_size / 8); 4294 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz); 4295 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2); 4296 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3); 4297 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1); 4298 #ifdef __BIG_ENDIAN 4299 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1); 4300 #endif 4301 WREG32(mmCP_RB0_CNTL, tmp); 4302 4303 /* Initialize the ring buffer's read and write pointers */ 4304 WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK); 4305 ring->wptr = 0; 4306 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 4307 4308 /* set the wb address wether it's enabled or not */ 4309 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 4310 WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr)); 4311 WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 4312 4313 wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); 4314 WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr)); 4315 WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr)); 4316 mdelay(1); 4317 WREG32(mmCP_RB0_CNTL, tmp); 4318 4319 rb_addr = ring->gpu_addr >> 8; 4320 WREG32(mmCP_RB0_BASE, rb_addr); 4321 WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr)); 4322 4323 gfx_v8_0_set_cpg_door_bell(adev, ring); 4324 /* start the ring */ 4325 amdgpu_ring_clear_ring(ring); 4326 gfx_v8_0_cp_gfx_start(adev); 4327 ring->sched.ready = true; 4328 4329 return 0; 4330 } 4331 4332 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) 4333 { 4334 if (enable) { 4335 WREG32(mmCP_MEC_CNTL, 0); 4336 } else { 4337 WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); 4338 adev->gfx.kiq.ring.sched.ready = false; 4339 } 4340 udelay(50); 4341 } 4342 4343 /* KIQ functions */ 4344 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring) 4345 { 4346 uint32_t tmp; 4347 struct amdgpu_device *adev = ring->adev; 4348 4349 /* tell RLC which is KIQ queue */ 4350 tmp = RREG32(mmRLC_CP_SCHEDULERS); 4351 tmp &= 0xffffff00; 4352 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue); 4353 WREG32(mmRLC_CP_SCHEDULERS, tmp); 4354 tmp |= 0x80; 4355 WREG32(mmRLC_CP_SCHEDULERS, tmp); 4356 } 4357 4358 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev) 4359 { 4360 struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring; 4361 uint64_t queue_mask = 0; 4362 int r, i; 4363 4364 for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) { 4365 if (!test_bit(i, adev->gfx.mec.queue_bitmap)) 4366 continue; 4367 4368 /* This situation may be hit in the future if a new HW 4369 * generation exposes more than 64 queues. If so, the 4370 * definition of queue_mask needs updating */ 4371 if (WARN_ON(i >= (sizeof(queue_mask)*8))) { 4372 DRM_ERROR("Invalid KCQ enabled: %d\n", i); 4373 break; 4374 } 4375 4376 queue_mask |= (1ull << i); 4377 } 4378 4379 r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8); 4380 if (r) { 4381 DRM_ERROR("Failed to lock KIQ (%d).\n", r); 4382 return r; 4383 } 4384 /* set resources */ 4385 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6)); 4386 amdgpu_ring_write(kiq_ring, 0); /* vmid_mask:0 queue_type:0 (KIQ) */ 4387 amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask)); /* queue mask lo */ 4388 amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask)); /* queue mask hi */ 4389 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */ 4390 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */ 4391 amdgpu_ring_write(kiq_ring, 0); /* oac mask */ 4392 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */ 4393 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4394 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; 4395 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj); 4396 uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); 4397 4398 /* map queues */ 4399 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5)); 4400 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/ 4401 amdgpu_ring_write(kiq_ring, 4402 PACKET3_MAP_QUEUES_NUM_QUEUES(1)); 4403 amdgpu_ring_write(kiq_ring, 4404 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) | 4405 PACKET3_MAP_QUEUES_QUEUE(ring->queue) | 4406 PACKET3_MAP_QUEUES_PIPE(ring->pipe) | 4407 PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */ 4408 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr)); 4409 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr)); 4410 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr)); 4411 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); 4412 } 4413 4414 amdgpu_ring_commit(kiq_ring); 4415 4416 return 0; 4417 } 4418 4419 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req) 4420 { 4421 int i, r = 0; 4422 4423 if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) { 4424 WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req); 4425 for (i = 0; i < adev->usec_timeout; i++) { 4426 if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK)) 4427 break; 4428 udelay(1); 4429 } 4430 if (i == adev->usec_timeout) 4431 r = -ETIMEDOUT; 4432 } 4433 WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0); 4434 WREG32(mmCP_HQD_PQ_RPTR, 0); 4435 WREG32(mmCP_HQD_PQ_WPTR, 0); 4436 4437 return r; 4438 } 4439 4440 static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd) 4441 { 4442 struct amdgpu_device *adev = ring->adev; 4443 4444 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 4445 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring->pipe, 4446 ring->queue)) { 4447 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH; 4448 mqd->cp_hqd_queue_priority = 4449 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM; 4450 } 4451 } 4452 } 4453 4454 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring) 4455 { 4456 struct amdgpu_device *adev = ring->adev; 4457 struct vi_mqd *mqd = ring->mqd_ptr; 4458 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr; 4459 uint32_t tmp; 4460 4461 mqd->header = 0xC0310800; 4462 mqd->compute_pipelinestat_enable = 0x00000001; 4463 mqd->compute_static_thread_mgmt_se0 = 0xffffffff; 4464 mqd->compute_static_thread_mgmt_se1 = 0xffffffff; 4465 mqd->compute_static_thread_mgmt_se2 = 0xffffffff; 4466 mqd->compute_static_thread_mgmt_se3 = 0xffffffff; 4467 mqd->compute_misc_reserved = 0x00000003; 4468 mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr 4469 + offsetof(struct vi_mqd_allocation, dynamic_cu_mask)); 4470 mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr 4471 + offsetof(struct vi_mqd_allocation, dynamic_cu_mask)); 4472 eop_base_addr = ring->eop_gpu_addr >> 8; 4473 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr; 4474 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr); 4475 4476 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 4477 tmp = RREG32(mmCP_HQD_EOP_CONTROL); 4478 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE, 4479 (order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1)); 4480 4481 mqd->cp_hqd_eop_control = tmp; 4482 4483 /* enable doorbell? */ 4484 tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL), 4485 CP_HQD_PQ_DOORBELL_CONTROL, 4486 DOORBELL_EN, 4487 ring->use_doorbell ? 1 : 0); 4488 4489 mqd->cp_hqd_pq_doorbell_control = tmp; 4490 4491 /* set the pointer to the MQD */ 4492 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc; 4493 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr); 4494 4495 /* set MQD vmid to 0 */ 4496 tmp = RREG32(mmCP_MQD_CONTROL); 4497 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0); 4498 mqd->cp_mqd_control = tmp; 4499 4500 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 4501 hqd_gpu_addr = ring->gpu_addr >> 8; 4502 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr; 4503 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr); 4504 4505 /* set up the HQD, this is similar to CP_RB0_CNTL */ 4506 tmp = RREG32(mmCP_HQD_PQ_CONTROL); 4507 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, 4508 (order_base_2(ring->ring_size / 4) - 1)); 4509 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, 4510 ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); 4511 #ifdef __BIG_ENDIAN 4512 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); 4513 #endif 4514 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0); 4515 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0); 4516 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1); 4517 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1); 4518 mqd->cp_hqd_pq_control = tmp; 4519 4520 /* set the wb address whether it's enabled or not */ 4521 wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 4522 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc; 4523 mqd->cp_hqd_pq_rptr_report_addr_hi = 4524 upper_32_bits(wb_gpu_addr) & 0xffff; 4525 4526 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 4527 wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); 4528 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc; 4529 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff; 4530 4531 tmp = 0; 4532 /* enable the doorbell if requested */ 4533 if (ring->use_doorbell) { 4534 tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL); 4535 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 4536 DOORBELL_OFFSET, ring->doorbell_index); 4537 4538 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 4539 DOORBELL_EN, 1); 4540 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 4541 DOORBELL_SOURCE, 0); 4542 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 4543 DOORBELL_HIT, 0); 4544 } 4545 4546 mqd->cp_hqd_pq_doorbell_control = tmp; 4547 4548 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 4549 ring->wptr = 0; 4550 mqd->cp_hqd_pq_wptr = ring->wptr; 4551 mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR); 4552 4553 /* set the vmid for the queue */ 4554 mqd->cp_hqd_vmid = 0; 4555 4556 tmp = RREG32(mmCP_HQD_PERSISTENT_STATE); 4557 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53); 4558 mqd->cp_hqd_persistent_state = tmp; 4559 4560 /* set MTYPE */ 4561 tmp = RREG32(mmCP_HQD_IB_CONTROL); 4562 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3); 4563 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3); 4564 mqd->cp_hqd_ib_control = tmp; 4565 4566 tmp = RREG32(mmCP_HQD_IQ_TIMER); 4567 tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3); 4568 mqd->cp_hqd_iq_timer = tmp; 4569 4570 tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL); 4571 tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3); 4572 mqd->cp_hqd_ctx_save_control = tmp; 4573 4574 /* defaults */ 4575 mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR); 4576 mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR); 4577 mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO); 4578 mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI); 4579 mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET); 4580 mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE); 4581 mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET); 4582 mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE); 4583 mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS); 4584 mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR); 4585 mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM); 4586 mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES); 4587 4588 /* set static priority for a queue/ring */ 4589 gfx_v8_0_mqd_set_priority(ring, mqd); 4590 mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM); 4591 4592 /* map_queues packet doesn't need activate the queue, 4593 * so only kiq need set this field. 4594 */ 4595 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) 4596 mqd->cp_hqd_active = 1; 4597 4598 return 0; 4599 } 4600 4601 static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev, 4602 struct vi_mqd *mqd) 4603 { 4604 uint32_t mqd_reg; 4605 uint32_t *mqd_data; 4606 4607 /* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */ 4608 mqd_data = &mqd->cp_mqd_base_addr_lo; 4609 4610 /* disable wptr polling */ 4611 WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0); 4612 4613 /* program all HQD registers */ 4614 for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++) 4615 WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]); 4616 4617 /* Tonga errata: EOP RPTR/WPTR should be left unmodified. 4618 * This is safe since EOP RPTR==WPTR for any inactive HQD 4619 * on ASICs that do not support context-save. 4620 * EOP writes/reads can start anywhere in the ring. 4621 */ 4622 if (adev->asic_type != CHIP_TONGA) { 4623 WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr); 4624 WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr); 4625 WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem); 4626 } 4627 4628 for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++) 4629 WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]); 4630 4631 /* activate the HQD */ 4632 for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++) 4633 WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]); 4634 4635 return 0; 4636 } 4637 4638 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring) 4639 { 4640 struct amdgpu_device *adev = ring->adev; 4641 struct vi_mqd *mqd = ring->mqd_ptr; 4642 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS; 4643 4644 gfx_v8_0_kiq_setting(ring); 4645 4646 if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */ 4647 /* reset MQD to a clean status */ 4648 if (adev->gfx.mec.mqd_backup[mqd_idx]) 4649 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation)); 4650 4651 /* reset ring buffer */ 4652 ring->wptr = 0; 4653 amdgpu_ring_clear_ring(ring); 4654 mutex_lock(&adev->srbm_mutex); 4655 vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 4656 gfx_v8_0_mqd_commit(adev, mqd); 4657 vi_srbm_select(adev, 0, 0, 0, 0); 4658 mutex_unlock(&adev->srbm_mutex); 4659 } else { 4660 memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation)); 4661 ((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 4662 ((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 4663 mutex_lock(&adev->srbm_mutex); 4664 vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 4665 gfx_v8_0_mqd_init(ring); 4666 gfx_v8_0_mqd_commit(adev, mqd); 4667 vi_srbm_select(adev, 0, 0, 0, 0); 4668 mutex_unlock(&adev->srbm_mutex); 4669 4670 if (adev->gfx.mec.mqd_backup[mqd_idx]) 4671 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation)); 4672 } 4673 4674 return 0; 4675 } 4676 4677 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring) 4678 { 4679 struct amdgpu_device *adev = ring->adev; 4680 struct vi_mqd *mqd = ring->mqd_ptr; 4681 int mqd_idx = ring - &adev->gfx.compute_ring[0]; 4682 4683 if (!amdgpu_in_reset(adev) && !adev->in_suspend) { 4684 memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation)); 4685 ((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 4686 ((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 4687 mutex_lock(&adev->srbm_mutex); 4688 vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 4689 gfx_v8_0_mqd_init(ring); 4690 vi_srbm_select(adev, 0, 0, 0, 0); 4691 mutex_unlock(&adev->srbm_mutex); 4692 4693 if (adev->gfx.mec.mqd_backup[mqd_idx]) 4694 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation)); 4695 } else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */ 4696 /* reset MQD to a clean status */ 4697 if (adev->gfx.mec.mqd_backup[mqd_idx]) 4698 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation)); 4699 /* reset ring buffer */ 4700 ring->wptr = 0; 4701 amdgpu_ring_clear_ring(ring); 4702 } else { 4703 amdgpu_ring_clear_ring(ring); 4704 } 4705 return 0; 4706 } 4707 4708 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev) 4709 { 4710 if (adev->asic_type > CHIP_TONGA) { 4711 WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2); 4712 WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2); 4713 } 4714 /* enable doorbells */ 4715 WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1); 4716 } 4717 4718 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev) 4719 { 4720 struct amdgpu_ring *ring; 4721 int r; 4722 4723 ring = &adev->gfx.kiq.ring; 4724 4725 r = amdgpu_bo_reserve(ring->mqd_obj, false); 4726 if (unlikely(r != 0)) 4727 return r; 4728 4729 r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr); 4730 if (unlikely(r != 0)) 4731 return r; 4732 4733 gfx_v8_0_kiq_init_queue(ring); 4734 amdgpu_bo_kunmap(ring->mqd_obj); 4735 ring->mqd_ptr = NULL; 4736 amdgpu_bo_unreserve(ring->mqd_obj); 4737 ring->sched.ready = true; 4738 return 0; 4739 } 4740 4741 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev) 4742 { 4743 struct amdgpu_ring *ring = NULL; 4744 int r = 0, i; 4745 4746 gfx_v8_0_cp_compute_enable(adev, true); 4747 4748 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4749 ring = &adev->gfx.compute_ring[i]; 4750 4751 r = amdgpu_bo_reserve(ring->mqd_obj, false); 4752 if (unlikely(r != 0)) 4753 goto done; 4754 r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr); 4755 if (!r) { 4756 r = gfx_v8_0_kcq_init_queue(ring); 4757 amdgpu_bo_kunmap(ring->mqd_obj); 4758 ring->mqd_ptr = NULL; 4759 } 4760 amdgpu_bo_unreserve(ring->mqd_obj); 4761 if (r) 4762 goto done; 4763 } 4764 4765 gfx_v8_0_set_mec_doorbell_range(adev); 4766 4767 r = gfx_v8_0_kiq_kcq_enable(adev); 4768 if (r) 4769 goto done; 4770 4771 done: 4772 return r; 4773 } 4774 4775 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev) 4776 { 4777 int r, i; 4778 struct amdgpu_ring *ring; 4779 4780 /* collect all the ring_tests here, gfx, kiq, compute */ 4781 ring = &adev->gfx.gfx_ring[0]; 4782 r = amdgpu_ring_test_helper(ring); 4783 if (r) 4784 return r; 4785 4786 ring = &adev->gfx.kiq.ring; 4787 r = amdgpu_ring_test_helper(ring); 4788 if (r) 4789 return r; 4790 4791 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4792 ring = &adev->gfx.compute_ring[i]; 4793 amdgpu_ring_test_helper(ring); 4794 } 4795 4796 return 0; 4797 } 4798 4799 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev) 4800 { 4801 int r; 4802 4803 if (!(adev->flags & AMD_IS_APU)) 4804 gfx_v8_0_enable_gui_idle_interrupt(adev, false); 4805 4806 r = gfx_v8_0_kiq_resume(adev); 4807 if (r) 4808 return r; 4809 4810 r = gfx_v8_0_cp_gfx_resume(adev); 4811 if (r) 4812 return r; 4813 4814 r = gfx_v8_0_kcq_resume(adev); 4815 if (r) 4816 return r; 4817 4818 r = gfx_v8_0_cp_test_all_rings(adev); 4819 if (r) 4820 return r; 4821 4822 gfx_v8_0_enable_gui_idle_interrupt(adev, true); 4823 4824 return 0; 4825 } 4826 4827 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable) 4828 { 4829 gfx_v8_0_cp_gfx_enable(adev, enable); 4830 gfx_v8_0_cp_compute_enable(adev, enable); 4831 } 4832 4833 static int gfx_v8_0_hw_init(void *handle) 4834 { 4835 int r; 4836 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4837 4838 gfx_v8_0_init_golden_registers(adev); 4839 gfx_v8_0_constants_init(adev); 4840 4841 r = adev->gfx.rlc.funcs->resume(adev); 4842 if (r) 4843 return r; 4844 4845 r = gfx_v8_0_cp_resume(adev); 4846 4847 return r; 4848 } 4849 4850 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev) 4851 { 4852 int r, i; 4853 struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring; 4854 4855 r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings); 4856 if (r) 4857 DRM_ERROR("Failed to lock KIQ (%d).\n", r); 4858 4859 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4860 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; 4861 4862 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4)); 4863 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 4864 PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */ 4865 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) | 4866 PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) | 4867 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1)); 4868 amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index)); 4869 amdgpu_ring_write(kiq_ring, 0); 4870 amdgpu_ring_write(kiq_ring, 0); 4871 amdgpu_ring_write(kiq_ring, 0); 4872 } 4873 r = amdgpu_ring_test_helper(kiq_ring); 4874 if (r) 4875 DRM_ERROR("KCQ disable failed\n"); 4876 4877 return r; 4878 } 4879 4880 static bool gfx_v8_0_is_idle(void *handle) 4881 { 4882 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4883 4884 if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE) 4885 || RREG32(mmGRBM_STATUS2) != 0x8) 4886 return false; 4887 else 4888 return true; 4889 } 4890 4891 static bool gfx_v8_0_rlc_is_idle(void *handle) 4892 { 4893 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4894 4895 if (RREG32(mmGRBM_STATUS2) != 0x8) 4896 return false; 4897 else 4898 return true; 4899 } 4900 4901 static int gfx_v8_0_wait_for_rlc_idle(void *handle) 4902 { 4903 unsigned int i; 4904 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4905 4906 for (i = 0; i < adev->usec_timeout; i++) { 4907 if (gfx_v8_0_rlc_is_idle(handle)) 4908 return 0; 4909 4910 udelay(1); 4911 } 4912 return -ETIMEDOUT; 4913 } 4914 4915 static int gfx_v8_0_wait_for_idle(void *handle) 4916 { 4917 unsigned int i; 4918 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4919 4920 for (i = 0; i < adev->usec_timeout; i++) { 4921 if (gfx_v8_0_is_idle(handle)) 4922 return 0; 4923 4924 udelay(1); 4925 } 4926 return -ETIMEDOUT; 4927 } 4928 4929 static int gfx_v8_0_hw_fini(void *handle) 4930 { 4931 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4932 4933 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); 4934 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); 4935 4936 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0); 4937 4938 amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0); 4939 4940 /* disable KCQ to avoid CPC touch memory not valid anymore */ 4941 gfx_v8_0_kcq_disable(adev); 4942 4943 if (amdgpu_sriov_vf(adev)) { 4944 pr_debug("For SRIOV client, shouldn't do anything.\n"); 4945 return 0; 4946 } 4947 amdgpu_gfx_rlc_enter_safe_mode(adev); 4948 if (!gfx_v8_0_wait_for_idle(adev)) 4949 gfx_v8_0_cp_enable(adev, false); 4950 else 4951 pr_err("cp is busy, skip halt cp\n"); 4952 if (!gfx_v8_0_wait_for_rlc_idle(adev)) 4953 adev->gfx.rlc.funcs->stop(adev); 4954 else 4955 pr_err("rlc is busy, skip halt rlc\n"); 4956 amdgpu_gfx_rlc_exit_safe_mode(adev); 4957 4958 return 0; 4959 } 4960 4961 static int gfx_v8_0_suspend(void *handle) 4962 { 4963 return gfx_v8_0_hw_fini(handle); 4964 } 4965 4966 static int gfx_v8_0_resume(void *handle) 4967 { 4968 return gfx_v8_0_hw_init(handle); 4969 } 4970 4971 static bool gfx_v8_0_check_soft_reset(void *handle) 4972 { 4973 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4974 u32 grbm_soft_reset = 0, srbm_soft_reset = 0; 4975 u32 tmp; 4976 4977 /* GRBM_STATUS */ 4978 tmp = RREG32(mmGRBM_STATUS); 4979 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | 4980 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | 4981 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | 4982 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | 4983 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | 4984 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK | 4985 GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) { 4986 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 4987 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 4988 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 4989 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1); 4990 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, 4991 SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1); 4992 } 4993 4994 /* GRBM_STATUS2 */ 4995 tmp = RREG32(mmGRBM_STATUS2); 4996 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY)) 4997 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 4998 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 4999 5000 if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) || 5001 REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) || 5002 REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) { 5003 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, 5004 SOFT_RESET_CPF, 1); 5005 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, 5006 SOFT_RESET_CPC, 1); 5007 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, 5008 SOFT_RESET_CPG, 1); 5009 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, 5010 SOFT_RESET_GRBM, 1); 5011 } 5012 5013 /* SRBM_STATUS */ 5014 tmp = RREG32(mmSRBM_STATUS); 5015 if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING)) 5016 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, 5017 SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1); 5018 if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY)) 5019 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, 5020 SRBM_SOFT_RESET, SOFT_RESET_SEM, 1); 5021 5022 if (grbm_soft_reset || srbm_soft_reset) { 5023 adev->gfx.grbm_soft_reset = grbm_soft_reset; 5024 adev->gfx.srbm_soft_reset = srbm_soft_reset; 5025 return true; 5026 } else { 5027 adev->gfx.grbm_soft_reset = 0; 5028 adev->gfx.srbm_soft_reset = 0; 5029 return false; 5030 } 5031 } 5032 5033 static int gfx_v8_0_pre_soft_reset(void *handle) 5034 { 5035 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5036 u32 grbm_soft_reset = 0; 5037 5038 if ((!adev->gfx.grbm_soft_reset) && 5039 (!adev->gfx.srbm_soft_reset)) 5040 return 0; 5041 5042 grbm_soft_reset = adev->gfx.grbm_soft_reset; 5043 5044 /* stop the rlc */ 5045 adev->gfx.rlc.funcs->stop(adev); 5046 5047 if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || 5048 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX)) 5049 /* Disable GFX parsing/prefetching */ 5050 gfx_v8_0_cp_gfx_enable(adev, false); 5051 5052 if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || 5053 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) || 5054 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) || 5055 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) { 5056 int i; 5057 5058 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5059 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; 5060 5061 mutex_lock(&adev->srbm_mutex); 5062 vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 5063 gfx_v8_0_deactivate_hqd(adev, 2); 5064 vi_srbm_select(adev, 0, 0, 0, 0); 5065 mutex_unlock(&adev->srbm_mutex); 5066 } 5067 /* Disable MEC parsing/prefetching */ 5068 gfx_v8_0_cp_compute_enable(adev, false); 5069 } 5070 5071 return 0; 5072 } 5073 5074 static int gfx_v8_0_soft_reset(void *handle) 5075 { 5076 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5077 u32 grbm_soft_reset = 0, srbm_soft_reset = 0; 5078 u32 tmp; 5079 5080 if ((!adev->gfx.grbm_soft_reset) && 5081 (!adev->gfx.srbm_soft_reset)) 5082 return 0; 5083 5084 grbm_soft_reset = adev->gfx.grbm_soft_reset; 5085 srbm_soft_reset = adev->gfx.srbm_soft_reset; 5086 5087 if (grbm_soft_reset || srbm_soft_reset) { 5088 tmp = RREG32(mmGMCON_DEBUG); 5089 tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1); 5090 tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1); 5091 WREG32(mmGMCON_DEBUG, tmp); 5092 udelay(50); 5093 } 5094 5095 if (grbm_soft_reset) { 5096 tmp = RREG32(mmGRBM_SOFT_RESET); 5097 tmp |= grbm_soft_reset; 5098 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 5099 WREG32(mmGRBM_SOFT_RESET, tmp); 5100 tmp = RREG32(mmGRBM_SOFT_RESET); 5101 5102 udelay(50); 5103 5104 tmp &= ~grbm_soft_reset; 5105 WREG32(mmGRBM_SOFT_RESET, tmp); 5106 tmp = RREG32(mmGRBM_SOFT_RESET); 5107 } 5108 5109 if (srbm_soft_reset) { 5110 tmp = RREG32(mmSRBM_SOFT_RESET); 5111 tmp |= srbm_soft_reset; 5112 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); 5113 WREG32(mmSRBM_SOFT_RESET, tmp); 5114 tmp = RREG32(mmSRBM_SOFT_RESET); 5115 5116 udelay(50); 5117 5118 tmp &= ~srbm_soft_reset; 5119 WREG32(mmSRBM_SOFT_RESET, tmp); 5120 tmp = RREG32(mmSRBM_SOFT_RESET); 5121 } 5122 5123 if (grbm_soft_reset || srbm_soft_reset) { 5124 tmp = RREG32(mmGMCON_DEBUG); 5125 tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0); 5126 tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0); 5127 WREG32(mmGMCON_DEBUG, tmp); 5128 } 5129 5130 /* Wait a little for things to settle down */ 5131 udelay(50); 5132 5133 return 0; 5134 } 5135 5136 static int gfx_v8_0_post_soft_reset(void *handle) 5137 { 5138 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5139 u32 grbm_soft_reset = 0; 5140 5141 if ((!adev->gfx.grbm_soft_reset) && 5142 (!adev->gfx.srbm_soft_reset)) 5143 return 0; 5144 5145 grbm_soft_reset = adev->gfx.grbm_soft_reset; 5146 5147 if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || 5148 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) || 5149 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) || 5150 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) { 5151 int i; 5152 5153 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5154 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; 5155 5156 mutex_lock(&adev->srbm_mutex); 5157 vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 5158 gfx_v8_0_deactivate_hqd(adev, 2); 5159 vi_srbm_select(adev, 0, 0, 0, 0); 5160 mutex_unlock(&adev->srbm_mutex); 5161 } 5162 gfx_v8_0_kiq_resume(adev); 5163 gfx_v8_0_kcq_resume(adev); 5164 } 5165 5166 if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || 5167 REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX)) 5168 gfx_v8_0_cp_gfx_resume(adev); 5169 5170 gfx_v8_0_cp_test_all_rings(adev); 5171 5172 adev->gfx.rlc.funcs->start(adev); 5173 5174 return 0; 5175 } 5176 5177 /** 5178 * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot 5179 * 5180 * @adev: amdgpu_device pointer 5181 * 5182 * Fetches a GPU clock counter snapshot. 5183 * Returns the 64 bit clock counter snapshot. 5184 */ 5185 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev) 5186 { 5187 uint64_t clock; 5188 5189 mutex_lock(&adev->gfx.gpu_clock_mutex); 5190 WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 5191 clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) | 5192 ((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 5193 mutex_unlock(&adev->gfx.gpu_clock_mutex); 5194 return clock; 5195 } 5196 5197 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring, 5198 uint32_t vmid, 5199 uint32_t gds_base, uint32_t gds_size, 5200 uint32_t gws_base, uint32_t gws_size, 5201 uint32_t oa_base, uint32_t oa_size) 5202 { 5203 /* GDS Base */ 5204 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5205 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5206 WRITE_DATA_DST_SEL(0))); 5207 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base); 5208 amdgpu_ring_write(ring, 0); 5209 amdgpu_ring_write(ring, gds_base); 5210 5211 /* GDS Size */ 5212 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5213 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5214 WRITE_DATA_DST_SEL(0))); 5215 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size); 5216 amdgpu_ring_write(ring, 0); 5217 amdgpu_ring_write(ring, gds_size); 5218 5219 /* GWS */ 5220 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5221 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5222 WRITE_DATA_DST_SEL(0))); 5223 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws); 5224 amdgpu_ring_write(ring, 0); 5225 amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base); 5226 5227 /* OA */ 5228 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5229 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5230 WRITE_DATA_DST_SEL(0))); 5231 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa); 5232 amdgpu_ring_write(ring, 0); 5233 amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base)); 5234 } 5235 5236 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address) 5237 { 5238 WREG32(mmSQ_IND_INDEX, 5239 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 5240 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 5241 (address << SQ_IND_INDEX__INDEX__SHIFT) | 5242 (SQ_IND_INDEX__FORCE_READ_MASK)); 5243 return RREG32(mmSQ_IND_DATA); 5244 } 5245 5246 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd, 5247 uint32_t wave, uint32_t thread, 5248 uint32_t regno, uint32_t num, uint32_t *out) 5249 { 5250 WREG32(mmSQ_IND_INDEX, 5251 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 5252 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 5253 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 5254 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 5255 (SQ_IND_INDEX__FORCE_READ_MASK) | 5256 (SQ_IND_INDEX__AUTO_INCR_MASK)); 5257 while (num--) 5258 *(out++) = RREG32(mmSQ_IND_DATA); 5259 } 5260 5261 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields) 5262 { 5263 /* type 0 wave data */ 5264 dst[(*no_fields)++] = 0; 5265 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS); 5266 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 5267 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 5268 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 5269 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 5270 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 5271 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 5272 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 5273 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 5274 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 5275 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 5276 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 5277 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO); 5278 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI); 5279 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO); 5280 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI); 5281 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 5282 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 5283 } 5284 5285 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd, 5286 uint32_t wave, uint32_t start, 5287 uint32_t size, uint32_t *dst) 5288 { 5289 wave_read_regs( 5290 adev, simd, wave, 0, 5291 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 5292 } 5293 5294 5295 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = { 5296 .get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter, 5297 .select_se_sh = &gfx_v8_0_select_se_sh, 5298 .read_wave_data = &gfx_v8_0_read_wave_data, 5299 .read_wave_sgprs = &gfx_v8_0_read_wave_sgprs, 5300 .select_me_pipe_q = &gfx_v8_0_select_me_pipe_q 5301 }; 5302 5303 static int gfx_v8_0_early_init(void *handle) 5304 { 5305 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5306 5307 adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS; 5308 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 5309 AMDGPU_MAX_COMPUTE_RINGS); 5310 adev->gfx.funcs = &gfx_v8_0_gfx_funcs; 5311 gfx_v8_0_set_ring_funcs(adev); 5312 gfx_v8_0_set_irq_funcs(adev); 5313 gfx_v8_0_set_gds_init(adev); 5314 gfx_v8_0_set_rlc_funcs(adev); 5315 5316 return 0; 5317 } 5318 5319 static int gfx_v8_0_late_init(void *handle) 5320 { 5321 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5322 int r; 5323 5324 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); 5325 if (r) 5326 return r; 5327 5328 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); 5329 if (r) 5330 return r; 5331 5332 /* requires IBs so do in late init after IB pool is initialized */ 5333 r = gfx_v8_0_do_edc_gpr_workarounds(adev); 5334 if (r) 5335 return r; 5336 5337 r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0); 5338 if (r) { 5339 DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r); 5340 return r; 5341 } 5342 5343 r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0); 5344 if (r) { 5345 DRM_ERROR( 5346 "amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n", 5347 r); 5348 return r; 5349 } 5350 5351 return 0; 5352 } 5353 5354 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev, 5355 bool enable) 5356 { 5357 if ((adev->asic_type == CHIP_POLARIS11) || 5358 (adev->asic_type == CHIP_POLARIS12) || 5359 (adev->asic_type == CHIP_VEGAM)) 5360 /* Send msg to SMU via Powerplay */ 5361 amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable); 5362 5363 WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0); 5364 } 5365 5366 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev, 5367 bool enable) 5368 { 5369 WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0); 5370 } 5371 5372 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev, 5373 bool enable) 5374 { 5375 WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0); 5376 } 5377 5378 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev, 5379 bool enable) 5380 { 5381 WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0); 5382 } 5383 5384 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev, 5385 bool enable) 5386 { 5387 WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0); 5388 5389 /* Read any GFX register to wake up GFX. */ 5390 if (!enable) 5391 RREG32(mmDB_RENDER_CONTROL); 5392 } 5393 5394 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev, 5395 bool enable) 5396 { 5397 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) { 5398 cz_enable_gfx_cg_power_gating(adev, true); 5399 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE) 5400 cz_enable_gfx_pipeline_power_gating(adev, true); 5401 } else { 5402 cz_enable_gfx_cg_power_gating(adev, false); 5403 cz_enable_gfx_pipeline_power_gating(adev, false); 5404 } 5405 } 5406 5407 static int gfx_v8_0_set_powergating_state(void *handle, 5408 enum amd_powergating_state state) 5409 { 5410 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5411 bool enable = (state == AMD_PG_STATE_GATE); 5412 5413 if (amdgpu_sriov_vf(adev)) 5414 return 0; 5415 5416 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG | 5417 AMD_PG_SUPPORT_RLC_SMU_HS | 5418 AMD_PG_SUPPORT_CP | 5419 AMD_PG_SUPPORT_GFX_DMG)) 5420 amdgpu_gfx_rlc_enter_safe_mode(adev); 5421 switch (adev->asic_type) { 5422 case CHIP_CARRIZO: 5423 case CHIP_STONEY: 5424 5425 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) { 5426 cz_enable_sck_slow_down_on_power_up(adev, true); 5427 cz_enable_sck_slow_down_on_power_down(adev, true); 5428 } else { 5429 cz_enable_sck_slow_down_on_power_up(adev, false); 5430 cz_enable_sck_slow_down_on_power_down(adev, false); 5431 } 5432 if (adev->pg_flags & AMD_PG_SUPPORT_CP) 5433 cz_enable_cp_power_gating(adev, true); 5434 else 5435 cz_enable_cp_power_gating(adev, false); 5436 5437 cz_update_gfx_cg_power_gating(adev, enable); 5438 5439 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable) 5440 gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true); 5441 else 5442 gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false); 5443 5444 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable) 5445 gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true); 5446 else 5447 gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false); 5448 break; 5449 case CHIP_POLARIS11: 5450 case CHIP_POLARIS12: 5451 case CHIP_VEGAM: 5452 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable) 5453 gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true); 5454 else 5455 gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false); 5456 5457 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable) 5458 gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true); 5459 else 5460 gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false); 5461 5462 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable) 5463 polaris11_enable_gfx_quick_mg_power_gating(adev, true); 5464 else 5465 polaris11_enable_gfx_quick_mg_power_gating(adev, false); 5466 break; 5467 default: 5468 break; 5469 } 5470 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG | 5471 AMD_PG_SUPPORT_RLC_SMU_HS | 5472 AMD_PG_SUPPORT_CP | 5473 AMD_PG_SUPPORT_GFX_DMG)) 5474 amdgpu_gfx_rlc_exit_safe_mode(adev); 5475 return 0; 5476 } 5477 5478 static void gfx_v8_0_get_clockgating_state(void *handle, u32 *flags) 5479 { 5480 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5481 int data; 5482 5483 if (amdgpu_sriov_vf(adev)) 5484 *flags = 0; 5485 5486 /* AMD_CG_SUPPORT_GFX_MGCG */ 5487 data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5488 if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK)) 5489 *flags |= AMD_CG_SUPPORT_GFX_MGCG; 5490 5491 /* AMD_CG_SUPPORT_GFX_CGLG */ 5492 data = RREG32(mmRLC_CGCG_CGLS_CTRL); 5493 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK) 5494 *flags |= AMD_CG_SUPPORT_GFX_CGCG; 5495 5496 /* AMD_CG_SUPPORT_GFX_CGLS */ 5497 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK) 5498 *flags |= AMD_CG_SUPPORT_GFX_CGLS; 5499 5500 /* AMD_CG_SUPPORT_GFX_CGTS */ 5501 data = RREG32(mmCGTS_SM_CTRL_REG); 5502 if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK)) 5503 *flags |= AMD_CG_SUPPORT_GFX_CGTS; 5504 5505 /* AMD_CG_SUPPORT_GFX_CGTS_LS */ 5506 if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK)) 5507 *flags |= AMD_CG_SUPPORT_GFX_CGTS_LS; 5508 5509 /* AMD_CG_SUPPORT_GFX_RLC_LS */ 5510 data = RREG32(mmRLC_MEM_SLP_CNTL); 5511 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) 5512 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS; 5513 5514 /* AMD_CG_SUPPORT_GFX_CP_LS */ 5515 data = RREG32(mmCP_MEM_SLP_CNTL); 5516 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) 5517 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS; 5518 } 5519 5520 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev, 5521 uint32_t reg_addr, uint32_t cmd) 5522 { 5523 uint32_t data; 5524 5525 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 5526 5527 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff); 5528 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff); 5529 5530 data = RREG32(mmRLC_SERDES_WR_CTRL); 5531 if (adev->asic_type == CHIP_STONEY) 5532 data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK | 5533 RLC_SERDES_WR_CTRL__READ_COMMAND_MASK | 5534 RLC_SERDES_WR_CTRL__P1_SELECT_MASK | 5535 RLC_SERDES_WR_CTRL__P2_SELECT_MASK | 5536 RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK | 5537 RLC_SERDES_WR_CTRL__POWER_DOWN_MASK | 5538 RLC_SERDES_WR_CTRL__POWER_UP_MASK | 5539 RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK | 5540 RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK); 5541 else 5542 data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK | 5543 RLC_SERDES_WR_CTRL__READ_COMMAND_MASK | 5544 RLC_SERDES_WR_CTRL__P1_SELECT_MASK | 5545 RLC_SERDES_WR_CTRL__P2_SELECT_MASK | 5546 RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK | 5547 RLC_SERDES_WR_CTRL__POWER_DOWN_MASK | 5548 RLC_SERDES_WR_CTRL__POWER_UP_MASK | 5549 RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK | 5550 RLC_SERDES_WR_CTRL__BPM_DATA_MASK | 5551 RLC_SERDES_WR_CTRL__REG_ADDR_MASK | 5552 RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK); 5553 data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK | 5554 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) | 5555 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) | 5556 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT)); 5557 5558 WREG32(mmRLC_SERDES_WR_CTRL, data); 5559 } 5560 5561 #define MSG_ENTER_RLC_SAFE_MODE 1 5562 #define MSG_EXIT_RLC_SAFE_MODE 0 5563 #define RLC_GPR_REG2__REQ_MASK 0x00000001 5564 #define RLC_GPR_REG2__REQ__SHIFT 0 5565 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001 5566 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e 5567 5568 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev) 5569 { 5570 uint32_t rlc_setting; 5571 5572 rlc_setting = RREG32(mmRLC_CNTL); 5573 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) 5574 return false; 5575 5576 return true; 5577 } 5578 5579 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev) 5580 { 5581 uint32_t data; 5582 unsigned i; 5583 data = RREG32(mmRLC_CNTL); 5584 data |= RLC_SAFE_MODE__CMD_MASK; 5585 data &= ~RLC_SAFE_MODE__MESSAGE_MASK; 5586 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); 5587 WREG32(mmRLC_SAFE_MODE, data); 5588 5589 /* wait for RLC_SAFE_MODE */ 5590 for (i = 0; i < adev->usec_timeout; i++) { 5591 if ((RREG32(mmRLC_GPM_STAT) & 5592 (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | 5593 RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) == 5594 (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | 5595 RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) 5596 break; 5597 udelay(1); 5598 } 5599 for (i = 0; i < adev->usec_timeout; i++) { 5600 if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) 5601 break; 5602 udelay(1); 5603 } 5604 } 5605 5606 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev) 5607 { 5608 uint32_t data; 5609 unsigned i; 5610 5611 data = RREG32(mmRLC_CNTL); 5612 data |= RLC_SAFE_MODE__CMD_MASK; 5613 data &= ~RLC_SAFE_MODE__MESSAGE_MASK; 5614 WREG32(mmRLC_SAFE_MODE, data); 5615 5616 for (i = 0; i < adev->usec_timeout; i++) { 5617 if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) 5618 break; 5619 udelay(1); 5620 } 5621 } 5622 5623 static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid) 5624 { 5625 u32 data; 5626 5627 if (amdgpu_sriov_is_pp_one_vf(adev)) 5628 data = RREG32_NO_KIQ(mmRLC_SPM_VMID); 5629 else 5630 data = RREG32(mmRLC_SPM_VMID); 5631 5632 data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK; 5633 data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT; 5634 5635 if (amdgpu_sriov_is_pp_one_vf(adev)) 5636 WREG32_NO_KIQ(mmRLC_SPM_VMID, data); 5637 else 5638 WREG32(mmRLC_SPM_VMID, data); 5639 } 5640 5641 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = { 5642 .is_rlc_enabled = gfx_v8_0_is_rlc_enabled, 5643 .set_safe_mode = gfx_v8_0_set_safe_mode, 5644 .unset_safe_mode = gfx_v8_0_unset_safe_mode, 5645 .init = gfx_v8_0_rlc_init, 5646 .get_csb_size = gfx_v8_0_get_csb_size, 5647 .get_csb_buffer = gfx_v8_0_get_csb_buffer, 5648 .get_cp_table_num = gfx_v8_0_cp_jump_table_num, 5649 .resume = gfx_v8_0_rlc_resume, 5650 .stop = gfx_v8_0_rlc_stop, 5651 .reset = gfx_v8_0_rlc_reset, 5652 .start = gfx_v8_0_rlc_start, 5653 .update_spm_vmid = gfx_v8_0_update_spm_vmid 5654 }; 5655 5656 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev, 5657 bool enable) 5658 { 5659 uint32_t temp, data; 5660 5661 amdgpu_gfx_rlc_enter_safe_mode(adev); 5662 5663 /* It is disabled by HW by default */ 5664 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 5665 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 5666 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) 5667 /* 1 - RLC memory Light sleep */ 5668 WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1); 5669 5670 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) 5671 WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1); 5672 } 5673 5674 /* 3 - RLC_CGTT_MGCG_OVERRIDE */ 5675 temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5676 if (adev->flags & AMD_IS_APU) 5677 data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK | 5678 RLC_CGTT_MGCG_OVERRIDE__RLC_MASK | 5679 RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK); 5680 else 5681 data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK | 5682 RLC_CGTT_MGCG_OVERRIDE__RLC_MASK | 5683 RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK | 5684 RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK); 5685 5686 if (temp != data) 5687 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 5688 5689 /* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5690 gfx_v8_0_wait_for_rlc_serdes(adev); 5691 5692 /* 5 - clear mgcg override */ 5693 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD); 5694 5695 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) { 5696 /* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */ 5697 temp = data = RREG32(mmCGTS_SM_CTRL_REG); 5698 data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK); 5699 data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT); 5700 data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK; 5701 data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK; 5702 if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) && 5703 (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS)) 5704 data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK; 5705 data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK; 5706 data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT); 5707 if (temp != data) 5708 WREG32(mmCGTS_SM_CTRL_REG, data); 5709 } 5710 udelay(50); 5711 5712 /* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5713 gfx_v8_0_wait_for_rlc_serdes(adev); 5714 } else { 5715 /* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */ 5716 temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5717 data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK | 5718 RLC_CGTT_MGCG_OVERRIDE__RLC_MASK | 5719 RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK | 5720 RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK); 5721 if (temp != data) 5722 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 5723 5724 /* 2 - disable MGLS in RLC */ 5725 data = RREG32(mmRLC_MEM_SLP_CNTL); 5726 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { 5727 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 5728 WREG32(mmRLC_MEM_SLP_CNTL, data); 5729 } 5730 5731 /* 3 - disable MGLS in CP */ 5732 data = RREG32(mmCP_MEM_SLP_CNTL); 5733 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 5734 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 5735 WREG32(mmCP_MEM_SLP_CNTL, data); 5736 } 5737 5738 /* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */ 5739 temp = data = RREG32(mmCGTS_SM_CTRL_REG); 5740 data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK | 5741 CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK); 5742 if (temp != data) 5743 WREG32(mmCGTS_SM_CTRL_REG, data); 5744 5745 /* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5746 gfx_v8_0_wait_for_rlc_serdes(adev); 5747 5748 /* 6 - set mgcg override */ 5749 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD); 5750 5751 udelay(50); 5752 5753 /* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5754 gfx_v8_0_wait_for_rlc_serdes(adev); 5755 } 5756 5757 amdgpu_gfx_rlc_exit_safe_mode(adev); 5758 } 5759 5760 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, 5761 bool enable) 5762 { 5763 uint32_t temp, temp1, data, data1; 5764 5765 temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL); 5766 5767 amdgpu_gfx_rlc_enter_safe_mode(adev); 5768 5769 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 5770 temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5771 data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK; 5772 if (temp1 != data1) 5773 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); 5774 5775 /* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5776 gfx_v8_0_wait_for_rlc_serdes(adev); 5777 5778 /* 2 - clear cgcg override */ 5779 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD); 5780 5781 /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5782 gfx_v8_0_wait_for_rlc_serdes(adev); 5783 5784 /* 3 - write cmd to set CGLS */ 5785 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD); 5786 5787 /* 4 - enable cgcg */ 5788 data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 5789 5790 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) { 5791 /* enable cgls*/ 5792 data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 5793 5794 temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5795 data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK; 5796 5797 if (temp1 != data1) 5798 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); 5799 } else { 5800 data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 5801 } 5802 5803 if (temp != data) 5804 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 5805 5806 /* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/ 5807 * Cmp_busy/GFX_Idle interrupts 5808 */ 5809 gfx_v8_0_enable_gui_idle_interrupt(adev, true); 5810 } else { 5811 /* disable cntx_empty_int_enable & GFX Idle interrupt */ 5812 gfx_v8_0_enable_gui_idle_interrupt(adev, false); 5813 5814 /* TEST CGCG */ 5815 temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 5816 data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK | 5817 RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK); 5818 if (temp1 != data1) 5819 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); 5820 5821 /* read gfx register to wake up cgcg */ 5822 RREG32(mmCB_CGTT_SCLK_CTRL); 5823 RREG32(mmCB_CGTT_SCLK_CTRL); 5824 RREG32(mmCB_CGTT_SCLK_CTRL); 5825 RREG32(mmCB_CGTT_SCLK_CTRL); 5826 5827 /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5828 gfx_v8_0_wait_for_rlc_serdes(adev); 5829 5830 /* write cmd to Set CGCG Overrride */ 5831 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD); 5832 5833 /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ 5834 gfx_v8_0_wait_for_rlc_serdes(adev); 5835 5836 /* write cmd to Clear CGLS */ 5837 gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD); 5838 5839 /* disable cgcg, cgls should be disabled too. */ 5840 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | 5841 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 5842 if (temp != data) 5843 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 5844 /* enable interrupts again for PG */ 5845 gfx_v8_0_enable_gui_idle_interrupt(adev, true); 5846 } 5847 5848 gfx_v8_0_wait_for_rlc_serdes(adev); 5849 5850 amdgpu_gfx_rlc_exit_safe_mode(adev); 5851 } 5852 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev, 5853 bool enable) 5854 { 5855 if (enable) { 5856 /* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS) 5857 * === MGCG + MGLS + TS(CG/LS) === 5858 */ 5859 gfx_v8_0_update_medium_grain_clock_gating(adev, enable); 5860 gfx_v8_0_update_coarse_grain_clock_gating(adev, enable); 5861 } else { 5862 /* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS) 5863 * === CGCG + CGLS === 5864 */ 5865 gfx_v8_0_update_coarse_grain_clock_gating(adev, enable); 5866 gfx_v8_0_update_medium_grain_clock_gating(adev, enable); 5867 } 5868 return 0; 5869 } 5870 5871 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev, 5872 enum amd_clockgating_state state) 5873 { 5874 uint32_t msg_id, pp_state = 0; 5875 uint32_t pp_support_state = 0; 5876 5877 if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) { 5878 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) { 5879 pp_support_state = PP_STATE_SUPPORT_LS; 5880 pp_state = PP_STATE_LS; 5881 } 5882 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) { 5883 pp_support_state |= PP_STATE_SUPPORT_CG; 5884 pp_state |= PP_STATE_CG; 5885 } 5886 if (state == AMD_CG_STATE_UNGATE) 5887 pp_state = 0; 5888 5889 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5890 PP_BLOCK_GFX_CG, 5891 pp_support_state, 5892 pp_state); 5893 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5894 } 5895 5896 if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) { 5897 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 5898 pp_support_state = PP_STATE_SUPPORT_LS; 5899 pp_state = PP_STATE_LS; 5900 } 5901 5902 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) { 5903 pp_support_state |= PP_STATE_SUPPORT_CG; 5904 pp_state |= PP_STATE_CG; 5905 } 5906 5907 if (state == AMD_CG_STATE_UNGATE) 5908 pp_state = 0; 5909 5910 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5911 PP_BLOCK_GFX_MG, 5912 pp_support_state, 5913 pp_state); 5914 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5915 } 5916 5917 return 0; 5918 } 5919 5920 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev, 5921 enum amd_clockgating_state state) 5922 { 5923 5924 uint32_t msg_id, pp_state = 0; 5925 uint32_t pp_support_state = 0; 5926 5927 if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) { 5928 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) { 5929 pp_support_state = PP_STATE_SUPPORT_LS; 5930 pp_state = PP_STATE_LS; 5931 } 5932 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) { 5933 pp_support_state |= PP_STATE_SUPPORT_CG; 5934 pp_state |= PP_STATE_CG; 5935 } 5936 if (state == AMD_CG_STATE_UNGATE) 5937 pp_state = 0; 5938 5939 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5940 PP_BLOCK_GFX_CG, 5941 pp_support_state, 5942 pp_state); 5943 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5944 } 5945 5946 if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) { 5947 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) { 5948 pp_support_state = PP_STATE_SUPPORT_LS; 5949 pp_state = PP_STATE_LS; 5950 } 5951 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) { 5952 pp_support_state |= PP_STATE_SUPPORT_CG; 5953 pp_state |= PP_STATE_CG; 5954 } 5955 if (state == AMD_CG_STATE_UNGATE) 5956 pp_state = 0; 5957 5958 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5959 PP_BLOCK_GFX_3D, 5960 pp_support_state, 5961 pp_state); 5962 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5963 } 5964 5965 if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) { 5966 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 5967 pp_support_state = PP_STATE_SUPPORT_LS; 5968 pp_state = PP_STATE_LS; 5969 } 5970 5971 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) { 5972 pp_support_state |= PP_STATE_SUPPORT_CG; 5973 pp_state |= PP_STATE_CG; 5974 } 5975 5976 if (state == AMD_CG_STATE_UNGATE) 5977 pp_state = 0; 5978 5979 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5980 PP_BLOCK_GFX_MG, 5981 pp_support_state, 5982 pp_state); 5983 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5984 } 5985 5986 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) { 5987 pp_support_state = PP_STATE_SUPPORT_LS; 5988 5989 if (state == AMD_CG_STATE_UNGATE) 5990 pp_state = 0; 5991 else 5992 pp_state = PP_STATE_LS; 5993 5994 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 5995 PP_BLOCK_GFX_RLC, 5996 pp_support_state, 5997 pp_state); 5998 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 5999 } 6000 6001 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 6002 pp_support_state = PP_STATE_SUPPORT_LS; 6003 6004 if (state == AMD_CG_STATE_UNGATE) 6005 pp_state = 0; 6006 else 6007 pp_state = PP_STATE_LS; 6008 msg_id = PP_CG_MSG_ID(PP_GROUP_GFX, 6009 PP_BLOCK_GFX_CP, 6010 pp_support_state, 6011 pp_state); 6012 amdgpu_dpm_set_clockgating_by_smu(adev, msg_id); 6013 } 6014 6015 return 0; 6016 } 6017 6018 static int gfx_v8_0_set_clockgating_state(void *handle, 6019 enum amd_clockgating_state state) 6020 { 6021 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 6022 6023 if (amdgpu_sriov_vf(adev)) 6024 return 0; 6025 6026 switch (adev->asic_type) { 6027 case CHIP_FIJI: 6028 case CHIP_CARRIZO: 6029 case CHIP_STONEY: 6030 gfx_v8_0_update_gfx_clock_gating(adev, 6031 state == AMD_CG_STATE_GATE); 6032 break; 6033 case CHIP_TONGA: 6034 gfx_v8_0_tonga_update_gfx_clock_gating(adev, state); 6035 break; 6036 case CHIP_POLARIS10: 6037 case CHIP_POLARIS11: 6038 case CHIP_POLARIS12: 6039 case CHIP_VEGAM: 6040 gfx_v8_0_polaris_update_gfx_clock_gating(adev, state); 6041 break; 6042 default: 6043 break; 6044 } 6045 return 0; 6046 } 6047 6048 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring) 6049 { 6050 return ring->adev->wb.wb[ring->rptr_offs]; 6051 } 6052 6053 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring) 6054 { 6055 struct amdgpu_device *adev = ring->adev; 6056 6057 if (ring->use_doorbell) 6058 /* XXX check if swapping is necessary on BE */ 6059 return ring->adev->wb.wb[ring->wptr_offs]; 6060 else 6061 return RREG32(mmCP_RB0_WPTR); 6062 } 6063 6064 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 6065 { 6066 struct amdgpu_device *adev = ring->adev; 6067 6068 if (ring->use_doorbell) { 6069 /* XXX check if swapping is necessary on BE */ 6070 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); 6071 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 6072 } else { 6073 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 6074 (void)RREG32(mmCP_RB0_WPTR); 6075 } 6076 } 6077 6078 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 6079 { 6080 u32 ref_and_mask, reg_mem_engine; 6081 6082 if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) || 6083 (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) { 6084 switch (ring->me) { 6085 case 1: 6086 ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe; 6087 break; 6088 case 2: 6089 ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe; 6090 break; 6091 default: 6092 return; 6093 } 6094 reg_mem_engine = 0; 6095 } else { 6096 ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK; 6097 reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */ 6098 } 6099 6100 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 6101 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */ 6102 WAIT_REG_MEM_FUNCTION(3) | /* == */ 6103 reg_mem_engine)); 6104 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ); 6105 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE); 6106 amdgpu_ring_write(ring, ref_and_mask); 6107 amdgpu_ring_write(ring, ref_and_mask); 6108 amdgpu_ring_write(ring, 0x20); /* poll interval */ 6109 } 6110 6111 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring) 6112 { 6113 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 6114 amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) | 6115 EVENT_INDEX(4)); 6116 6117 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 6118 amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) | 6119 EVENT_INDEX(0)); 6120 } 6121 6122 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, 6123 struct amdgpu_job *job, 6124 struct amdgpu_ib *ib, 6125 uint32_t flags) 6126 { 6127 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 6128 u32 header, control = 0; 6129 6130 if (ib->flags & AMDGPU_IB_FLAG_CE) 6131 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 6132 else 6133 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 6134 6135 control |= ib->length_dw | (vmid << 24); 6136 6137 if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) { 6138 control |= INDIRECT_BUFFER_PRE_ENB(1); 6139 6140 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid) 6141 gfx_v8_0_ring_emit_de_meta(ring); 6142 } 6143 6144 amdgpu_ring_write(ring, header); 6145 amdgpu_ring_write(ring, 6146 #ifdef __BIG_ENDIAN 6147 (2 << 0) | 6148 #endif 6149 (ib->gpu_addr & 0xFFFFFFFC)); 6150 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 6151 amdgpu_ring_write(ring, control); 6152 } 6153 6154 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, 6155 struct amdgpu_job *job, 6156 struct amdgpu_ib *ib, 6157 uint32_t flags) 6158 { 6159 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 6160 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); 6161 6162 /* Currently, there is a high possibility to get wave ID mismatch 6163 * between ME and GDS, leading to a hw deadlock, because ME generates 6164 * different wave IDs than the GDS expects. This situation happens 6165 * randomly when at least 5 compute pipes use GDS ordered append. 6166 * The wave IDs generated by ME are also wrong after suspend/resume. 6167 * Those are probably bugs somewhere else in the kernel driver. 6168 * 6169 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and 6170 * GDS to 0 for this ring (me/pipe). 6171 */ 6172 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { 6173 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 6174 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START); 6175 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); 6176 } 6177 6178 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 6179 amdgpu_ring_write(ring, 6180 #ifdef __BIG_ENDIAN 6181 (2 << 0) | 6182 #endif 6183 (ib->gpu_addr & 0xFFFFFFFC)); 6184 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 6185 amdgpu_ring_write(ring, control); 6186 } 6187 6188 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr, 6189 u64 seq, unsigned flags) 6190 { 6191 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 6192 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 6193 6194 /* Workaround for cache flush problems. First send a dummy EOP 6195 * event down the pipe with seq one below. 6196 */ 6197 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 6198 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 6199 EOP_TC_ACTION_EN | 6200 EOP_TC_WB_ACTION_EN | 6201 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 6202 EVENT_INDEX(5))); 6203 amdgpu_ring_write(ring, addr & 0xfffffffc); 6204 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 6205 DATA_SEL(1) | INT_SEL(0)); 6206 amdgpu_ring_write(ring, lower_32_bits(seq - 1)); 6207 amdgpu_ring_write(ring, upper_32_bits(seq - 1)); 6208 6209 /* Then send the real EOP event down the pipe: 6210 * EVENT_WRITE_EOP - flush caches, send int */ 6211 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 6212 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 6213 EOP_TC_ACTION_EN | 6214 EOP_TC_WB_ACTION_EN | 6215 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 6216 EVENT_INDEX(5))); 6217 amdgpu_ring_write(ring, addr & 0xfffffffc); 6218 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 6219 DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 6220 amdgpu_ring_write(ring, lower_32_bits(seq)); 6221 amdgpu_ring_write(ring, upper_32_bits(seq)); 6222 6223 } 6224 6225 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 6226 { 6227 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 6228 uint32_t seq = ring->fence_drv.sync_seq; 6229 uint64_t addr = ring->fence_drv.gpu_addr; 6230 6231 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 6232 amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ 6233 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 6234 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ 6235 amdgpu_ring_write(ring, addr & 0xfffffffc); 6236 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); 6237 amdgpu_ring_write(ring, seq); 6238 amdgpu_ring_write(ring, 0xffffffff); 6239 amdgpu_ring_write(ring, 4); /* poll interval */ 6240 } 6241 6242 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 6243 unsigned vmid, uint64_t pd_addr) 6244 { 6245 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 6246 6247 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 6248 6249 /* wait for the invalidate to complete */ 6250 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 6251 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */ 6252 WAIT_REG_MEM_FUNCTION(0) | /* always */ 6253 WAIT_REG_MEM_ENGINE(0))); /* me */ 6254 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); 6255 amdgpu_ring_write(ring, 0); 6256 amdgpu_ring_write(ring, 0); /* ref */ 6257 amdgpu_ring_write(ring, 0); /* mask */ 6258 amdgpu_ring_write(ring, 0x20); /* poll interval */ 6259 6260 /* compute doesn't have PFP */ 6261 if (usepfp) { 6262 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 6263 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 6264 amdgpu_ring_write(ring, 0x0); 6265 } 6266 } 6267 6268 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring) 6269 { 6270 return ring->adev->wb.wb[ring->wptr_offs]; 6271 } 6272 6273 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 6274 { 6275 struct amdgpu_device *adev = ring->adev; 6276 6277 /* XXX check if swapping is necessary on BE */ 6278 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); 6279 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 6280 } 6281 6282 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring, 6283 u64 addr, u64 seq, 6284 unsigned flags) 6285 { 6286 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 6287 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 6288 6289 /* RELEASE_MEM - flush caches, send int */ 6290 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5)); 6291 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 6292 EOP_TC_ACTION_EN | 6293 EOP_TC_WB_ACTION_EN | 6294 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 6295 EVENT_INDEX(5))); 6296 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 6297 amdgpu_ring_write(ring, addr & 0xfffffffc); 6298 amdgpu_ring_write(ring, upper_32_bits(addr)); 6299 amdgpu_ring_write(ring, lower_32_bits(seq)); 6300 amdgpu_ring_write(ring, upper_32_bits(seq)); 6301 } 6302 6303 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr, 6304 u64 seq, unsigned int flags) 6305 { 6306 /* we only allocate 32bit for each seq wb address */ 6307 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 6308 6309 /* write fence seq to the "addr" */ 6310 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 6311 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 6312 WRITE_DATA_DST_SEL(5) | WR_CONFIRM)); 6313 amdgpu_ring_write(ring, lower_32_bits(addr)); 6314 amdgpu_ring_write(ring, upper_32_bits(addr)); 6315 amdgpu_ring_write(ring, lower_32_bits(seq)); 6316 6317 if (flags & AMDGPU_FENCE_FLAG_INT) { 6318 /* set register to trigger INT */ 6319 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 6320 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 6321 WRITE_DATA_DST_SEL(0) | WR_CONFIRM)); 6322 amdgpu_ring_write(ring, mmCPC_INT_STATUS); 6323 amdgpu_ring_write(ring, 0); 6324 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */ 6325 } 6326 } 6327 6328 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring) 6329 { 6330 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 6331 amdgpu_ring_write(ring, 0); 6332 } 6333 6334 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 6335 { 6336 uint32_t dw2 = 0; 6337 6338 if (amdgpu_sriov_vf(ring->adev)) 6339 gfx_v8_0_ring_emit_ce_meta(ring); 6340 6341 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */ 6342 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 6343 gfx_v8_0_ring_emit_vgt_flush(ring); 6344 /* set load_global_config & load_global_uconfig */ 6345 dw2 |= 0x8001; 6346 /* set load_cs_sh_regs */ 6347 dw2 |= 0x01000000; 6348 /* set load_per_context_state & load_gfx_sh_regs for GFX */ 6349 dw2 |= 0x10002; 6350 6351 /* set load_ce_ram if preamble presented */ 6352 if (AMDGPU_PREAMBLE_IB_PRESENT & flags) 6353 dw2 |= 0x10000000; 6354 } else { 6355 /* still load_ce_ram if this is the first time preamble presented 6356 * although there is no context switch happens. 6357 */ 6358 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags) 6359 dw2 |= 0x10000000; 6360 } 6361 6362 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 6363 amdgpu_ring_write(ring, dw2); 6364 amdgpu_ring_write(ring, 0); 6365 } 6366 6367 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring) 6368 { 6369 unsigned ret; 6370 6371 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3)); 6372 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr)); 6373 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr)); 6374 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */ 6375 ret = ring->wptr & ring->buf_mask; 6376 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */ 6377 return ret; 6378 } 6379 6380 static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset) 6381 { 6382 unsigned cur; 6383 6384 BUG_ON(offset > ring->buf_mask); 6385 BUG_ON(ring->ring[offset] != 0x55aa55aa); 6386 6387 cur = (ring->wptr & ring->buf_mask) - 1; 6388 if (likely(cur > offset)) 6389 ring->ring[offset] = cur - offset; 6390 else 6391 ring->ring[offset] = (ring->ring_size >> 2) - offset + cur; 6392 } 6393 6394 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg, 6395 uint32_t reg_val_offs) 6396 { 6397 struct amdgpu_device *adev = ring->adev; 6398 6399 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 6400 amdgpu_ring_write(ring, 0 | /* src: register*/ 6401 (5 << 8) | /* dst: memory */ 6402 (1 << 20)); /* write confirm */ 6403 amdgpu_ring_write(ring, reg); 6404 amdgpu_ring_write(ring, 0); 6405 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 6406 reg_val_offs * 4)); 6407 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 6408 reg_val_offs * 4)); 6409 } 6410 6411 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, 6412 uint32_t val) 6413 { 6414 uint32_t cmd; 6415 6416 switch (ring->funcs->type) { 6417 case AMDGPU_RING_TYPE_GFX: 6418 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM; 6419 break; 6420 case AMDGPU_RING_TYPE_KIQ: 6421 cmd = 1 << 16; /* no inc addr */ 6422 break; 6423 default: 6424 cmd = WR_CONFIRM; 6425 break; 6426 } 6427 6428 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 6429 amdgpu_ring_write(ring, cmd); 6430 amdgpu_ring_write(ring, reg); 6431 amdgpu_ring_write(ring, 0); 6432 amdgpu_ring_write(ring, val); 6433 } 6434 6435 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid) 6436 { 6437 struct amdgpu_device *adev = ring->adev; 6438 uint32_t value = 0; 6439 6440 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); 6441 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); 6442 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); 6443 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); 6444 WREG32(mmSQ_CMD, value); 6445 } 6446 6447 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 6448 enum amdgpu_interrupt_state state) 6449 { 6450 WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE, 6451 state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1); 6452 } 6453 6454 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 6455 int me, int pipe, 6456 enum amdgpu_interrupt_state state) 6457 { 6458 u32 mec_int_cntl, mec_int_cntl_reg; 6459 6460 /* 6461 * amdgpu controls only the first MEC. That's why this function only 6462 * handles the setting of interrupts for this specific MEC. All other 6463 * pipes' interrupts are set by amdkfd. 6464 */ 6465 6466 if (me == 1) { 6467 switch (pipe) { 6468 case 0: 6469 mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL; 6470 break; 6471 case 1: 6472 mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL; 6473 break; 6474 case 2: 6475 mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL; 6476 break; 6477 case 3: 6478 mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL; 6479 break; 6480 default: 6481 DRM_DEBUG("invalid pipe %d\n", pipe); 6482 return; 6483 } 6484 } else { 6485 DRM_DEBUG("invalid me %d\n", me); 6486 return; 6487 } 6488 6489 switch (state) { 6490 case AMDGPU_IRQ_STATE_DISABLE: 6491 mec_int_cntl = RREG32(mec_int_cntl_reg); 6492 mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 6493 WREG32(mec_int_cntl_reg, mec_int_cntl); 6494 break; 6495 case AMDGPU_IRQ_STATE_ENABLE: 6496 mec_int_cntl = RREG32(mec_int_cntl_reg); 6497 mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 6498 WREG32(mec_int_cntl_reg, mec_int_cntl); 6499 break; 6500 default: 6501 break; 6502 } 6503 } 6504 6505 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 6506 struct amdgpu_irq_src *source, 6507 unsigned type, 6508 enum amdgpu_interrupt_state state) 6509 { 6510 WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE, 6511 state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1); 6512 6513 return 0; 6514 } 6515 6516 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 6517 struct amdgpu_irq_src *source, 6518 unsigned type, 6519 enum amdgpu_interrupt_state state) 6520 { 6521 WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE, 6522 state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1); 6523 6524 return 0; 6525 } 6526 6527 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev, 6528 struct amdgpu_irq_src *src, 6529 unsigned type, 6530 enum amdgpu_interrupt_state state) 6531 { 6532 switch (type) { 6533 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP: 6534 gfx_v8_0_set_gfx_eop_interrupt_state(adev, state); 6535 break; 6536 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 6537 gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state); 6538 break; 6539 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 6540 gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state); 6541 break; 6542 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP: 6543 gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state); 6544 break; 6545 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP: 6546 gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state); 6547 break; 6548 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP: 6549 gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state); 6550 break; 6551 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP: 6552 gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state); 6553 break; 6554 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP: 6555 gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state); 6556 break; 6557 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP: 6558 gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state); 6559 break; 6560 default: 6561 break; 6562 } 6563 return 0; 6564 } 6565 6566 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev, 6567 struct amdgpu_irq_src *source, 6568 unsigned int type, 6569 enum amdgpu_interrupt_state state) 6570 { 6571 int enable_flag; 6572 6573 switch (state) { 6574 case AMDGPU_IRQ_STATE_DISABLE: 6575 enable_flag = 0; 6576 break; 6577 6578 case AMDGPU_IRQ_STATE_ENABLE: 6579 enable_flag = 1; 6580 break; 6581 6582 default: 6583 return -EINVAL; 6584 } 6585 6586 WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag); 6587 WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag); 6588 WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag); 6589 WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag); 6590 WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag); 6591 WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6592 enable_flag); 6593 WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6594 enable_flag); 6595 WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6596 enable_flag); 6597 WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6598 enable_flag); 6599 WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6600 enable_flag); 6601 WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6602 enable_flag); 6603 WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6604 enable_flag); 6605 WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, 6606 enable_flag); 6607 6608 return 0; 6609 } 6610 6611 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev, 6612 struct amdgpu_irq_src *source, 6613 unsigned int type, 6614 enum amdgpu_interrupt_state state) 6615 { 6616 int enable_flag; 6617 6618 switch (state) { 6619 case AMDGPU_IRQ_STATE_DISABLE: 6620 enable_flag = 1; 6621 break; 6622 6623 case AMDGPU_IRQ_STATE_ENABLE: 6624 enable_flag = 0; 6625 break; 6626 6627 default: 6628 return -EINVAL; 6629 } 6630 6631 WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL, 6632 enable_flag); 6633 6634 return 0; 6635 } 6636 6637 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev, 6638 struct amdgpu_irq_src *source, 6639 struct amdgpu_iv_entry *entry) 6640 { 6641 int i; 6642 u8 me_id, pipe_id, queue_id; 6643 struct amdgpu_ring *ring; 6644 6645 DRM_DEBUG("IH: CP EOP\n"); 6646 me_id = (entry->ring_id & 0x0c) >> 2; 6647 pipe_id = (entry->ring_id & 0x03) >> 0; 6648 queue_id = (entry->ring_id & 0x70) >> 4; 6649 6650 switch (me_id) { 6651 case 0: 6652 amdgpu_fence_process(&adev->gfx.gfx_ring[0]); 6653 break; 6654 case 1: 6655 case 2: 6656 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 6657 ring = &adev->gfx.compute_ring[i]; 6658 /* Per-queue interrupt is supported for MEC starting from VI. 6659 * The interrupt can only be enabled/disabled per pipe instead of per queue. 6660 */ 6661 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id)) 6662 amdgpu_fence_process(ring); 6663 } 6664 break; 6665 } 6666 return 0; 6667 } 6668 6669 static void gfx_v8_0_fault(struct amdgpu_device *adev, 6670 struct amdgpu_iv_entry *entry) 6671 { 6672 u8 me_id, pipe_id, queue_id; 6673 struct amdgpu_ring *ring; 6674 int i; 6675 6676 me_id = (entry->ring_id & 0x0c) >> 2; 6677 pipe_id = (entry->ring_id & 0x03) >> 0; 6678 queue_id = (entry->ring_id & 0x70) >> 4; 6679 6680 switch (me_id) { 6681 case 0: 6682 drm_sched_fault(&adev->gfx.gfx_ring[0].sched); 6683 break; 6684 case 1: 6685 case 2: 6686 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 6687 ring = &adev->gfx.compute_ring[i]; 6688 if (ring->me == me_id && ring->pipe == pipe_id && 6689 ring->queue == queue_id) 6690 drm_sched_fault(&ring->sched); 6691 } 6692 break; 6693 } 6694 } 6695 6696 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev, 6697 struct amdgpu_irq_src *source, 6698 struct amdgpu_iv_entry *entry) 6699 { 6700 DRM_ERROR("Illegal register access in command stream\n"); 6701 gfx_v8_0_fault(adev, entry); 6702 return 0; 6703 } 6704 6705 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev, 6706 struct amdgpu_irq_src *source, 6707 struct amdgpu_iv_entry *entry) 6708 { 6709 DRM_ERROR("Illegal instruction in command stream\n"); 6710 gfx_v8_0_fault(adev, entry); 6711 return 0; 6712 } 6713 6714 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev, 6715 struct amdgpu_irq_src *source, 6716 struct amdgpu_iv_entry *entry) 6717 { 6718 DRM_ERROR("CP EDC/ECC error detected."); 6719 return 0; 6720 } 6721 6722 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data) 6723 { 6724 u32 enc, se_id, sh_id, cu_id; 6725 char type[20]; 6726 int sq_edc_source = -1; 6727 6728 enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING); 6729 se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID); 6730 6731 switch (enc) { 6732 case 0: 6733 DRM_INFO("SQ general purpose intr detected:" 6734 "se_id %d, immed_overflow %d, host_reg_overflow %d," 6735 "host_cmd_overflow %d, cmd_timestamp %d," 6736 "reg_timestamp %d, thread_trace_buff_full %d," 6737 "wlt %d, thread_trace %d.\n", 6738 se_id, 6739 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW), 6740 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW), 6741 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW), 6742 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP), 6743 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP), 6744 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL), 6745 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT), 6746 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE) 6747 ); 6748 break; 6749 case 1: 6750 case 2: 6751 6752 cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID); 6753 sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID); 6754 6755 /* 6756 * This function can be called either directly from ISR 6757 * or from BH in which case we can access SQ_EDC_INFO 6758 * instance 6759 */ 6760 if (in_task()) { 6761 mutex_lock(&adev->grbm_idx_mutex); 6762 gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id); 6763 6764 sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE); 6765 6766 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 6767 mutex_unlock(&adev->grbm_idx_mutex); 6768 } 6769 6770 if (enc == 1) 6771 sprintf(type, "instruction intr"); 6772 else 6773 sprintf(type, "EDC/ECC error"); 6774 6775 DRM_INFO( 6776 "SQ %s detected: " 6777 "se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d " 6778 "trap %s, sq_ed_info.source %s.\n", 6779 type, se_id, sh_id, cu_id, 6780 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID), 6781 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID), 6782 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID), 6783 REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false", 6784 (sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable" 6785 ); 6786 break; 6787 default: 6788 DRM_ERROR("SQ invalid encoding type\n."); 6789 } 6790 } 6791 6792 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work) 6793 { 6794 6795 struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work); 6796 struct sq_work *sq_work = container_of(work, struct sq_work, work); 6797 6798 gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data); 6799 } 6800 6801 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev, 6802 struct amdgpu_irq_src *source, 6803 struct amdgpu_iv_entry *entry) 6804 { 6805 unsigned ih_data = entry->src_data[0]; 6806 6807 /* 6808 * Try to submit work so SQ_EDC_INFO can be accessed from 6809 * BH. If previous work submission hasn't finished yet 6810 * just print whatever info is possible directly from the ISR. 6811 */ 6812 if (work_pending(&adev->gfx.sq_work.work)) { 6813 gfx_v8_0_parse_sq_irq(adev, ih_data); 6814 } else { 6815 adev->gfx.sq_work.ih_data = ih_data; 6816 schedule_work(&adev->gfx.sq_work.work); 6817 } 6818 6819 return 0; 6820 } 6821 6822 static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring) 6823 { 6824 amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 6825 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 6826 PACKET3_TC_ACTION_ENA | 6827 PACKET3_SH_KCACHE_ACTION_ENA | 6828 PACKET3_SH_ICACHE_ACTION_ENA | 6829 PACKET3_TC_WB_ACTION_ENA); /* CP_COHER_CNTL */ 6830 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 6831 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 6832 amdgpu_ring_write(ring, 0x0000000A); /* poll interval */ 6833 } 6834 6835 static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring) 6836 { 6837 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5)); 6838 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 6839 PACKET3_TC_ACTION_ENA | 6840 PACKET3_SH_KCACHE_ACTION_ENA | 6841 PACKET3_SH_ICACHE_ACTION_ENA | 6842 PACKET3_TC_WB_ACTION_ENA); /* CP_COHER_CNTL */ 6843 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 6844 amdgpu_ring_write(ring, 0xff); /* CP_COHER_SIZE_HI */ 6845 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 6846 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */ 6847 amdgpu_ring_write(ring, 0x0000000A); /* poll interval */ 6848 } 6849 6850 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = { 6851 .name = "gfx_v8_0", 6852 .early_init = gfx_v8_0_early_init, 6853 .late_init = gfx_v8_0_late_init, 6854 .sw_init = gfx_v8_0_sw_init, 6855 .sw_fini = gfx_v8_0_sw_fini, 6856 .hw_init = gfx_v8_0_hw_init, 6857 .hw_fini = gfx_v8_0_hw_fini, 6858 .suspend = gfx_v8_0_suspend, 6859 .resume = gfx_v8_0_resume, 6860 .is_idle = gfx_v8_0_is_idle, 6861 .wait_for_idle = gfx_v8_0_wait_for_idle, 6862 .check_soft_reset = gfx_v8_0_check_soft_reset, 6863 .pre_soft_reset = gfx_v8_0_pre_soft_reset, 6864 .soft_reset = gfx_v8_0_soft_reset, 6865 .post_soft_reset = gfx_v8_0_post_soft_reset, 6866 .set_clockgating_state = gfx_v8_0_set_clockgating_state, 6867 .set_powergating_state = gfx_v8_0_set_powergating_state, 6868 .get_clockgating_state = gfx_v8_0_get_clockgating_state, 6869 }; 6870 6871 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = { 6872 .type = AMDGPU_RING_TYPE_GFX, 6873 .align_mask = 0xff, 6874 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6875 .support_64bit_ptrs = false, 6876 .get_rptr = gfx_v8_0_ring_get_rptr, 6877 .get_wptr = gfx_v8_0_ring_get_wptr_gfx, 6878 .set_wptr = gfx_v8_0_ring_set_wptr_gfx, 6879 .emit_frame_size = /* maximum 215dw if count 16 IBs in */ 6880 5 + /* COND_EXEC */ 6881 7 + /* PIPELINE_SYNC */ 6882 VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */ 6883 12 + /* FENCE for VM_FLUSH */ 6884 20 + /* GDS switch */ 6885 4 + /* double SWITCH_BUFFER, 6886 the first COND_EXEC jump to the place just 6887 prior to this double SWITCH_BUFFER */ 6888 5 + /* COND_EXEC */ 6889 7 + /* HDP_flush */ 6890 4 + /* VGT_flush */ 6891 14 + /* CE_META */ 6892 31 + /* DE_META */ 6893 3 + /* CNTX_CTRL */ 6894 5 + /* HDP_INVL */ 6895 12 + 12 + /* FENCE x2 */ 6896 2 + /* SWITCH_BUFFER */ 6897 5, /* SURFACE_SYNC */ 6898 .emit_ib_size = 4, /* gfx_v8_0_ring_emit_ib_gfx */ 6899 .emit_ib = gfx_v8_0_ring_emit_ib_gfx, 6900 .emit_fence = gfx_v8_0_ring_emit_fence_gfx, 6901 .emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync, 6902 .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, 6903 .emit_gds_switch = gfx_v8_0_ring_emit_gds_switch, 6904 .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, 6905 .test_ring = gfx_v8_0_ring_test_ring, 6906 .test_ib = gfx_v8_0_ring_test_ib, 6907 .insert_nop = amdgpu_ring_insert_nop, 6908 .pad_ib = amdgpu_ring_generic_pad_ib, 6909 .emit_switch_buffer = gfx_v8_ring_emit_sb, 6910 .emit_cntxcntl = gfx_v8_ring_emit_cntxcntl, 6911 .init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec, 6912 .patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec, 6913 .emit_wreg = gfx_v8_0_ring_emit_wreg, 6914 .soft_recovery = gfx_v8_0_ring_soft_recovery, 6915 .emit_mem_sync = gfx_v8_0_emit_mem_sync, 6916 }; 6917 6918 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { 6919 .type = AMDGPU_RING_TYPE_COMPUTE, 6920 .align_mask = 0xff, 6921 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6922 .support_64bit_ptrs = false, 6923 .get_rptr = gfx_v8_0_ring_get_rptr, 6924 .get_wptr = gfx_v8_0_ring_get_wptr_compute, 6925 .set_wptr = gfx_v8_0_ring_set_wptr_compute, 6926 .emit_frame_size = 6927 20 + /* gfx_v8_0_ring_emit_gds_switch */ 6928 7 + /* gfx_v8_0_ring_emit_hdp_flush */ 6929 5 + /* hdp_invalidate */ 6930 7 + /* gfx_v8_0_ring_emit_pipeline_sync */ 6931 VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */ 6932 7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */ 6933 7, /* gfx_v8_0_emit_mem_sync_compute */ 6934 .emit_ib_size = 7, /* gfx_v8_0_ring_emit_ib_compute */ 6935 .emit_ib = gfx_v8_0_ring_emit_ib_compute, 6936 .emit_fence = gfx_v8_0_ring_emit_fence_compute, 6937 .emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync, 6938 .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, 6939 .emit_gds_switch = gfx_v8_0_ring_emit_gds_switch, 6940 .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, 6941 .test_ring = gfx_v8_0_ring_test_ring, 6942 .test_ib = gfx_v8_0_ring_test_ib, 6943 .insert_nop = amdgpu_ring_insert_nop, 6944 .pad_ib = amdgpu_ring_generic_pad_ib, 6945 .emit_wreg = gfx_v8_0_ring_emit_wreg, 6946 .emit_mem_sync = gfx_v8_0_emit_mem_sync_compute, 6947 }; 6948 6949 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = { 6950 .type = AMDGPU_RING_TYPE_KIQ, 6951 .align_mask = 0xff, 6952 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6953 .support_64bit_ptrs = false, 6954 .get_rptr = gfx_v8_0_ring_get_rptr, 6955 .get_wptr = gfx_v8_0_ring_get_wptr_compute, 6956 .set_wptr = gfx_v8_0_ring_set_wptr_compute, 6957 .emit_frame_size = 6958 20 + /* gfx_v8_0_ring_emit_gds_switch */ 6959 7 + /* gfx_v8_0_ring_emit_hdp_flush */ 6960 5 + /* hdp_invalidate */ 6961 7 + /* gfx_v8_0_ring_emit_pipeline_sync */ 6962 17 + /* gfx_v8_0_ring_emit_vm_flush */ 6963 7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */ 6964 .emit_ib_size = 7, /* gfx_v8_0_ring_emit_ib_compute */ 6965 .emit_fence = gfx_v8_0_ring_emit_fence_kiq, 6966 .test_ring = gfx_v8_0_ring_test_ring, 6967 .insert_nop = amdgpu_ring_insert_nop, 6968 .pad_ib = amdgpu_ring_generic_pad_ib, 6969 .emit_rreg = gfx_v8_0_ring_emit_rreg, 6970 .emit_wreg = gfx_v8_0_ring_emit_wreg, 6971 }; 6972 6973 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev) 6974 { 6975 int i; 6976 6977 adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq; 6978 6979 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 6980 adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx; 6981 6982 for (i = 0; i < adev->gfx.num_compute_rings; i++) 6983 adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute; 6984 } 6985 6986 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = { 6987 .set = gfx_v8_0_set_eop_interrupt_state, 6988 .process = gfx_v8_0_eop_irq, 6989 }; 6990 6991 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = { 6992 .set = gfx_v8_0_set_priv_reg_fault_state, 6993 .process = gfx_v8_0_priv_reg_irq, 6994 }; 6995 6996 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = { 6997 .set = gfx_v8_0_set_priv_inst_fault_state, 6998 .process = gfx_v8_0_priv_inst_irq, 6999 }; 7000 7001 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = { 7002 .set = gfx_v8_0_set_cp_ecc_int_state, 7003 .process = gfx_v8_0_cp_ecc_error_irq, 7004 }; 7005 7006 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = { 7007 .set = gfx_v8_0_set_sq_int_state, 7008 .process = gfx_v8_0_sq_irq, 7009 }; 7010 7011 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev) 7012 { 7013 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 7014 adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs; 7015 7016 adev->gfx.priv_reg_irq.num_types = 1; 7017 adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs; 7018 7019 adev->gfx.priv_inst_irq.num_types = 1; 7020 adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs; 7021 7022 adev->gfx.cp_ecc_error_irq.num_types = 1; 7023 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs; 7024 7025 adev->gfx.sq_irq.num_types = 1; 7026 adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs; 7027 } 7028 7029 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev) 7030 { 7031 adev->gfx.rlc.funcs = &iceland_rlc_funcs; 7032 } 7033 7034 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev) 7035 { 7036 /* init asci gds info */ 7037 adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE); 7038 adev->gds.gws_size = 64; 7039 adev->gds.oa_size = 16; 7040 adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID); 7041 } 7042 7043 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 7044 u32 bitmap) 7045 { 7046 u32 data; 7047 7048 if (!bitmap) 7049 return; 7050 7051 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 7052 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 7053 7054 WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data); 7055 } 7056 7057 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev) 7058 { 7059 u32 data, mask; 7060 7061 data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) | 7062 RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); 7063 7064 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 7065 7066 return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask; 7067 } 7068 7069 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev) 7070 { 7071 int i, j, k, counter, active_cu_number = 0; 7072 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 7073 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info; 7074 unsigned disable_masks[4 * 2]; 7075 u32 ao_cu_num; 7076 7077 memset(cu_info, 0, sizeof(*cu_info)); 7078 7079 if (adev->flags & AMD_IS_APU) 7080 ao_cu_num = 2; 7081 else 7082 ao_cu_num = adev->gfx.config.max_cu_per_sh; 7083 7084 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2); 7085 7086 mutex_lock(&adev->grbm_idx_mutex); 7087 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 7088 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 7089 mask = 1; 7090 ao_bitmap = 0; 7091 counter = 0; 7092 gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff); 7093 if (i < 4 && j < 2) 7094 gfx_v8_0_set_user_cu_inactive_bitmap( 7095 adev, disable_masks[i * 2 + j]); 7096 bitmap = gfx_v8_0_get_cu_active_bitmap(adev); 7097 cu_info->bitmap[i][j] = bitmap; 7098 7099 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { 7100 if (bitmap & mask) { 7101 if (counter < ao_cu_num) 7102 ao_bitmap |= mask; 7103 counter ++; 7104 } 7105 mask <<= 1; 7106 } 7107 active_cu_number += counter; 7108 if (i < 2 && j < 2) 7109 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 7110 cu_info->ao_cu_bitmap[i][j] = ao_bitmap; 7111 } 7112 } 7113 gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 7114 mutex_unlock(&adev->grbm_idx_mutex); 7115 7116 cu_info->number = active_cu_number; 7117 cu_info->ao_cu_mask = ao_cu_mask; 7118 cu_info->simd_per_cu = NUM_SIMD_PER_CU; 7119 cu_info->max_waves_per_simd = 10; 7120 cu_info->max_scratch_slots_per_cu = 32; 7121 cu_info->wave_front_size = 64; 7122 cu_info->lds_size = 64; 7123 } 7124 7125 const struct amdgpu_ip_block_version gfx_v8_0_ip_block = 7126 { 7127 .type = AMD_IP_BLOCK_TYPE_GFX, 7128 .major = 8, 7129 .minor = 0, 7130 .rev = 0, 7131 .funcs = &gfx_v8_0_ip_funcs, 7132 }; 7133 7134 const struct amdgpu_ip_block_version gfx_v8_1_ip_block = 7135 { 7136 .type = AMD_IP_BLOCK_TYPE_GFX, 7137 .major = 8, 7138 .minor = 1, 7139 .rev = 0, 7140 .funcs = &gfx_v8_0_ip_funcs, 7141 }; 7142 7143 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring) 7144 { 7145 uint64_t ce_payload_addr; 7146 int cnt_ce; 7147 union { 7148 struct vi_ce_ib_state regular; 7149 struct vi_ce_ib_state_chained_ib chained; 7150 } ce_payload = {}; 7151 7152 if (ring->adev->virt.chained_ib_support) { 7153 ce_payload_addr = amdgpu_csa_vaddr(ring->adev) + 7154 offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload); 7155 cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2; 7156 } else { 7157 ce_payload_addr = amdgpu_csa_vaddr(ring->adev) + 7158 offsetof(struct vi_gfx_meta_data, ce_payload); 7159 cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2; 7160 } 7161 7162 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce)); 7163 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) | 7164 WRITE_DATA_DST_SEL(8) | 7165 WR_CONFIRM) | 7166 WRITE_DATA_CACHE_POLICY(0)); 7167 amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr)); 7168 amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr)); 7169 amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2); 7170 } 7171 7172 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring) 7173 { 7174 uint64_t de_payload_addr, gds_addr, csa_addr; 7175 int cnt_de; 7176 union { 7177 struct vi_de_ib_state regular; 7178 struct vi_de_ib_state_chained_ib chained; 7179 } de_payload = {}; 7180 7181 csa_addr = amdgpu_csa_vaddr(ring->adev); 7182 gds_addr = csa_addr + 4096; 7183 if (ring->adev->virt.chained_ib_support) { 7184 de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr); 7185 de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr); 7186 de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload); 7187 cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2; 7188 } else { 7189 de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr); 7190 de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr); 7191 de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload); 7192 cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2; 7193 } 7194 7195 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de)); 7196 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 7197 WRITE_DATA_DST_SEL(8) | 7198 WR_CONFIRM) | 7199 WRITE_DATA_CACHE_POLICY(0)); 7200 amdgpu_ring_write(ring, lower_32_bits(de_payload_addr)); 7201 amdgpu_ring_write(ring, upper_32_bits(de_payload_addr)); 7202 amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2); 7203 } 7204