1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2013 Red Hat 4 * Author: Rob Clark <robdclark@gmail.com> 5 * 6 * Copyright (c) 2014,2017, 2019 The Linux Foundation. All rights reserved. 7 */ 8 9 #ifndef __ADRENO_GPU_H__ 10 #define __ADRENO_GPU_H__ 11 12 #include <linux/firmware.h> 13 #include <linux/iopoll.h> 14 15 #include "msm_gpu.h" 16 17 #include "adreno_common.xml.h" 18 #include "adreno_pm4.xml.h" 19 20 #define REG_ADRENO_DEFINE(_offset, _reg) [_offset] = (_reg) + 1 21 #define REG_SKIP ~0 22 #define REG_ADRENO_SKIP(_offset) [_offset] = REG_SKIP 23 24 extern bool snapshot_debugbus; 25 26 /** 27 * adreno_regs: List of registers that are used in across all 28 * 3D devices. Each device type has different offset value for the same 29 * register, so an array of register offsets are declared for every device 30 * and are indexed by the enumeration values defined in this enum 31 */ 32 enum adreno_regs { 33 REG_ADRENO_CP_RB_BASE, 34 REG_ADRENO_CP_RB_BASE_HI, 35 REG_ADRENO_CP_RB_RPTR_ADDR, 36 REG_ADRENO_CP_RB_RPTR_ADDR_HI, 37 REG_ADRENO_CP_RB_RPTR, 38 REG_ADRENO_CP_RB_WPTR, 39 REG_ADRENO_CP_RB_CNTL, 40 REG_ADRENO_REGISTER_MAX, 41 }; 42 43 enum { 44 ADRENO_FW_PM4 = 0, 45 ADRENO_FW_SQE = 0, /* a6xx */ 46 ADRENO_FW_PFP = 1, 47 ADRENO_FW_GMU = 1, /* a6xx */ 48 ADRENO_FW_GPMU = 2, 49 ADRENO_FW_MAX, 50 }; 51 52 enum adreno_quirks { 53 ADRENO_QUIRK_TWO_PASS_USE_WFI = 1, 54 ADRENO_QUIRK_FAULT_DETECT_MASK = 2, 55 ADRENO_QUIRK_LMLOADKILL_DISABLE = 3, 56 }; 57 58 struct adreno_rev { 59 uint8_t core; 60 uint8_t major; 61 uint8_t minor; 62 uint8_t patchid; 63 }; 64 65 #define ADRENO_REV(core, major, minor, patchid) \ 66 ((struct adreno_rev){ core, major, minor, patchid }) 67 68 struct adreno_gpu_funcs { 69 struct msm_gpu_funcs base; 70 int (*get_timestamp)(struct msm_gpu *gpu, uint64_t *value); 71 }; 72 73 struct adreno_reglist { 74 u32 offset; 75 u32 value; 76 }; 77 78 extern const struct adreno_reglist a630_hwcg[], a640_hwcg[], a650_hwcg[]; 79 80 struct adreno_info { 81 struct adreno_rev rev; 82 uint32_t revn; 83 const char *name; 84 const char *fw[ADRENO_FW_MAX]; 85 uint32_t gmem; 86 enum adreno_quirks quirks; 87 struct msm_gpu *(*init)(struct drm_device *dev); 88 const char *zapfw; 89 u32 inactive_period; 90 const struct adreno_reglist *hwcg; 91 }; 92 93 const struct adreno_info *adreno_info(struct adreno_rev rev); 94 95 struct adreno_gpu { 96 struct msm_gpu base; 97 struct adreno_rev rev; 98 const struct adreno_info *info; 99 uint32_t gmem; /* actual gmem size */ 100 uint32_t revn; /* numeric revision name */ 101 const struct adreno_gpu_funcs *funcs; 102 103 /* interesting register offsets to dump: */ 104 const unsigned int *registers; 105 106 /* 107 * Are we loading fw from legacy path? Prior to addition 108 * of gpu firmware to linux-firmware, the fw files were 109 * placed in toplevel firmware directory, following qcom's 110 * android kernel. But linux-firmware preferred they be 111 * placed in a 'qcom' subdirectory. 112 * 113 * For backwards compatibility, we try first to load from 114 * the new path, using request_firmware_direct() to avoid 115 * any potential timeout waiting for usermode helper, then 116 * fall back to the old path (with direct load). And 117 * finally fall back to request_firmware() with the new 118 * path to allow the usermode helper. 119 */ 120 enum { 121 FW_LOCATION_UNKNOWN = 0, 122 FW_LOCATION_NEW, /* /lib/firmware/qcom/$fwfile */ 123 FW_LOCATION_LEGACY, /* /lib/firmware/$fwfile */ 124 FW_LOCATION_HELPER, 125 } fwloc; 126 127 /* firmware: */ 128 const struct firmware *fw[ADRENO_FW_MAX]; 129 130 /* 131 * Register offsets are different between some GPUs. 132 * GPU specific offsets will be exported by GPU specific 133 * code (a3xx_gpu.c) and stored in this common location. 134 */ 135 const unsigned int *reg_offsets; 136 }; 137 #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base) 138 139 struct adreno_ocmem { 140 struct ocmem *ocmem; 141 unsigned long base; 142 void *hdl; 143 }; 144 145 /* platform config data (ie. from DT, or pdata) */ 146 struct adreno_platform_config { 147 struct adreno_rev rev; 148 }; 149 150 #define ADRENO_IDLE_TIMEOUT msecs_to_jiffies(1000) 151 152 #define spin_until(X) ({ \ 153 int __ret = -ETIMEDOUT; \ 154 unsigned long __t = jiffies + ADRENO_IDLE_TIMEOUT; \ 155 do { \ 156 if (X) { \ 157 __ret = 0; \ 158 break; \ 159 } \ 160 } while (time_before(jiffies, __t)); \ 161 __ret; \ 162 }) 163 164 static inline bool adreno_is_a2xx(struct adreno_gpu *gpu) 165 { 166 return (gpu->revn < 300); 167 } 168 169 static inline bool adreno_is_a20x(struct adreno_gpu *gpu) 170 { 171 return (gpu->revn < 210); 172 } 173 174 static inline bool adreno_is_a225(struct adreno_gpu *gpu) 175 { 176 return gpu->revn == 225; 177 } 178 179 static inline bool adreno_is_a3xx(struct adreno_gpu *gpu) 180 { 181 return (gpu->revn >= 300) && (gpu->revn < 400); 182 } 183 184 static inline bool adreno_is_a305(struct adreno_gpu *gpu) 185 { 186 return gpu->revn == 305; 187 } 188 189 static inline bool adreno_is_a306(struct adreno_gpu *gpu) 190 { 191 /* yes, 307, because a305c is 306 */ 192 return gpu->revn == 307; 193 } 194 195 static inline bool adreno_is_a320(struct adreno_gpu *gpu) 196 { 197 return gpu->revn == 320; 198 } 199 200 static inline bool adreno_is_a330(struct adreno_gpu *gpu) 201 { 202 return gpu->revn == 330; 203 } 204 205 static inline bool adreno_is_a330v2(struct adreno_gpu *gpu) 206 { 207 return adreno_is_a330(gpu) && (gpu->rev.patchid > 0); 208 } 209 210 static inline bool adreno_is_a4xx(struct adreno_gpu *gpu) 211 { 212 return (gpu->revn >= 400) && (gpu->revn < 500); 213 } 214 215 static inline int adreno_is_a405(struct adreno_gpu *gpu) 216 { 217 return gpu->revn == 405; 218 } 219 220 static inline int adreno_is_a420(struct adreno_gpu *gpu) 221 { 222 return gpu->revn == 420; 223 } 224 225 static inline int adreno_is_a430(struct adreno_gpu *gpu) 226 { 227 return gpu->revn == 430; 228 } 229 230 static inline int adreno_is_a510(struct adreno_gpu *gpu) 231 { 232 return gpu->revn == 510; 233 } 234 235 static inline int adreno_is_a530(struct adreno_gpu *gpu) 236 { 237 return gpu->revn == 530; 238 } 239 240 static inline int adreno_is_a540(struct adreno_gpu *gpu) 241 { 242 return gpu->revn == 540; 243 } 244 245 static inline int adreno_is_a618(struct adreno_gpu *gpu) 246 { 247 return gpu->revn == 618; 248 } 249 250 static inline int adreno_is_a630(struct adreno_gpu *gpu) 251 { 252 return gpu->revn == 630; 253 } 254 255 static inline int adreno_is_a640(struct adreno_gpu *gpu) 256 { 257 return gpu->revn == 640; 258 } 259 260 static inline int adreno_is_a650(struct adreno_gpu *gpu) 261 { 262 return gpu->revn == 650; 263 } 264 265 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value); 266 const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu, 267 const char *fwname); 268 struct drm_gem_object *adreno_fw_create_bo(struct msm_gpu *gpu, 269 const struct firmware *fw, u64 *iova); 270 int adreno_hw_init(struct msm_gpu *gpu); 271 void adreno_recover(struct msm_gpu *gpu); 272 void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, 273 struct msm_file_private *ctx); 274 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 275 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 276 #if defined(CONFIG_DEBUG_FS) || defined(CONFIG_DEV_COREDUMP) 277 void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state, 278 struct drm_printer *p); 279 #endif 280 void adreno_dump_info(struct msm_gpu *gpu); 281 void adreno_dump(struct msm_gpu *gpu); 282 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords); 283 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu); 284 285 int adreno_gpu_ocmem_init(struct device *dev, struct adreno_gpu *adreno_gpu, 286 struct adreno_ocmem *ocmem); 287 void adreno_gpu_ocmem_cleanup(struct adreno_ocmem *ocmem); 288 289 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, 290 struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs, 291 int nr_rings); 292 void adreno_gpu_cleanup(struct adreno_gpu *gpu); 293 int adreno_load_fw(struct adreno_gpu *adreno_gpu); 294 295 void adreno_gpu_state_destroy(struct msm_gpu_state *state); 296 297 int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state); 298 int adreno_gpu_state_put(struct msm_gpu_state *state); 299 300 /* 301 * Common helper function to initialize the default address space for arm-smmu 302 * attached targets 303 */ 304 struct msm_gem_address_space * 305 adreno_iommu_create_address_space(struct msm_gpu *gpu, 306 struct platform_device *pdev); 307 308 /* 309 * For a5xx and a6xx targets load the zap shader that is used to pull the GPU 310 * out of secure mode 311 */ 312 int adreno_zap_shader_load(struct msm_gpu *gpu, u32 pasid); 313 314 /* ringbuffer helpers (the parts that are adreno specific) */ 315 316 static inline void 317 OUT_PKT0(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 318 { 319 adreno_wait_ring(ring, cnt+1); 320 OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF)); 321 } 322 323 /* no-op packet: */ 324 static inline void 325 OUT_PKT2(struct msm_ringbuffer *ring) 326 { 327 adreno_wait_ring(ring, 1); 328 OUT_RING(ring, CP_TYPE2_PKT); 329 } 330 331 static inline void 332 OUT_PKT3(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 333 { 334 adreno_wait_ring(ring, cnt+1); 335 OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8)); 336 } 337 338 static inline u32 PM4_PARITY(u32 val) 339 { 340 return (0x9669 >> (0xF & (val ^ 341 (val >> 4) ^ (val >> 8) ^ (val >> 12) ^ 342 (val >> 16) ^ ((val) >> 20) ^ (val >> 24) ^ 343 (val >> 28)))) & 1; 344 } 345 346 /* Maximum number of values that can be executed for one opcode */ 347 #define TYPE4_MAX_PAYLOAD 127 348 349 #define PKT4(_reg, _cnt) \ 350 (CP_TYPE4_PKT | ((_cnt) << 0) | (PM4_PARITY((_cnt)) << 7) | \ 351 (((_reg) & 0x3FFFF) << 8) | (PM4_PARITY((_reg)) << 27)) 352 353 static inline void 354 OUT_PKT4(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 355 { 356 adreno_wait_ring(ring, cnt + 1); 357 OUT_RING(ring, PKT4(regindx, cnt)); 358 } 359 360 static inline void 361 OUT_PKT7(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 362 { 363 adreno_wait_ring(ring, cnt + 1); 364 OUT_RING(ring, CP_TYPE7_PKT | (cnt << 0) | (PM4_PARITY(cnt) << 15) | 365 ((opcode & 0x7F) << 16) | (PM4_PARITY(opcode) << 23)); 366 } 367 368 /* 369 * adreno_reg_check() - Checks the validity of a register enum 370 * @gpu: Pointer to struct adreno_gpu 371 * @offset_name: The register enum that is checked 372 */ 373 static inline bool adreno_reg_check(struct adreno_gpu *gpu, 374 enum adreno_regs offset_name) 375 { 376 BUG_ON(offset_name >= REG_ADRENO_REGISTER_MAX || !gpu->reg_offsets[offset_name]); 377 378 /* 379 * REG_SKIP is a special value that tell us that the register in 380 * question isn't implemented on target but don't trigger a BUG(). This 381 * is used to cleanly implement adreno_gpu_write64() and 382 * adreno_gpu_read64() in a generic fashion 383 */ 384 if (gpu->reg_offsets[offset_name] == REG_SKIP) 385 return false; 386 387 return true; 388 } 389 390 static inline u32 adreno_gpu_read(struct adreno_gpu *gpu, 391 enum adreno_regs offset_name) 392 { 393 u32 reg = gpu->reg_offsets[offset_name]; 394 u32 val = 0; 395 if(adreno_reg_check(gpu,offset_name)) 396 val = gpu_read(&gpu->base, reg - 1); 397 return val; 398 } 399 400 static inline void adreno_gpu_write(struct adreno_gpu *gpu, 401 enum adreno_regs offset_name, u32 data) 402 { 403 u32 reg = gpu->reg_offsets[offset_name]; 404 if(adreno_reg_check(gpu, offset_name)) 405 gpu_write(&gpu->base, reg - 1, data); 406 } 407 408 struct msm_gpu *a2xx_gpu_init(struct drm_device *dev); 409 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev); 410 struct msm_gpu *a4xx_gpu_init(struct drm_device *dev); 411 struct msm_gpu *a5xx_gpu_init(struct drm_device *dev); 412 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev); 413 414 static inline void adreno_gpu_write64(struct adreno_gpu *gpu, 415 enum adreno_regs lo, enum adreno_regs hi, u64 data) 416 { 417 adreno_gpu_write(gpu, lo, lower_32_bits(data)); 418 adreno_gpu_write(gpu, hi, upper_32_bits(data)); 419 } 420 421 static inline uint32_t get_wptr(struct msm_ringbuffer *ring) 422 { 423 return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2); 424 } 425 426 /* 427 * Given a register and a count, return a value to program into 428 * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len 429 * registers starting at _reg. 430 * 431 * The register base needs to be a multiple of the length. If it is not, the 432 * hardware will quietly mask off the bits for you and shift the size. For 433 * example, if you intend the protection to start at 0x07 for a length of 4 434 * (0x07-0x0A) the hardware will actually protect (0x04-0x07) which might 435 * expose registers you intended to protect! 436 */ 437 #define ADRENO_PROTECT_RW(_reg, _len) \ 438 ((1 << 30) | (1 << 29) | \ 439 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 440 441 /* 442 * Same as above, but allow reads over the range. For areas of mixed use (such 443 * as performance counters) this allows us to protect a much larger range with a 444 * single register 445 */ 446 #define ADRENO_PROTECT_RDONLY(_reg, _len) \ 447 ((1 << 29) \ 448 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 449 450 451 #define gpu_poll_timeout(gpu, addr, val, cond, interval, timeout) \ 452 readl_poll_timeout((gpu)->mmio + ((addr) << 2), val, cond, \ 453 interval, timeout) 454 455 #endif /* __ADRENO_GPU_H__ */ 456