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 extern bool snapshot_debugbus; 21 extern bool allow_vram_carveout; 22 23 enum { 24 ADRENO_FW_PM4 = 0, 25 ADRENO_FW_SQE = 0, /* a6xx */ 26 ADRENO_FW_PFP = 1, 27 ADRENO_FW_GMU = 1, /* a6xx */ 28 ADRENO_FW_GPMU = 2, 29 ADRENO_FW_MAX, 30 }; 31 32 enum adreno_quirks { 33 ADRENO_QUIRK_TWO_PASS_USE_WFI = 1, 34 ADRENO_QUIRK_FAULT_DETECT_MASK = 2, 35 ADRENO_QUIRK_LMLOADKILL_DISABLE = 3, 36 }; 37 38 struct adreno_rev { 39 uint8_t core; 40 uint8_t major; 41 uint8_t minor; 42 uint8_t patchid; 43 }; 44 45 #define ANY_ID 0xff 46 47 #define ADRENO_REV(core, major, minor, patchid) \ 48 ((struct adreno_rev){ core, major, minor, patchid }) 49 50 struct adreno_gpu_funcs { 51 struct msm_gpu_funcs base; 52 int (*get_timestamp)(struct msm_gpu *gpu, uint64_t *value); 53 }; 54 55 struct adreno_reglist { 56 u32 offset; 57 u32 value; 58 }; 59 60 extern const struct adreno_reglist a630_hwcg[], a640_hwcg[], a650_hwcg[], a660_hwcg[]; 61 62 struct adreno_info { 63 struct adreno_rev rev; 64 uint32_t revn; 65 const char *name; 66 const char *fw[ADRENO_FW_MAX]; 67 uint32_t gmem; 68 enum adreno_quirks quirks; 69 struct msm_gpu *(*init)(struct drm_device *dev); 70 const char *zapfw; 71 u32 inactive_period; 72 const struct adreno_reglist *hwcg; 73 }; 74 75 const struct adreno_info *adreno_info(struct adreno_rev rev); 76 77 struct adreno_gpu { 78 struct msm_gpu base; 79 struct adreno_rev rev; 80 const struct adreno_info *info; 81 uint32_t gmem; /* actual gmem size */ 82 uint32_t revn; /* numeric revision name */ 83 uint16_t speedbin; 84 const struct adreno_gpu_funcs *funcs; 85 86 /* interesting register offsets to dump: */ 87 const unsigned int *registers; 88 89 /* 90 * Are we loading fw from legacy path? Prior to addition 91 * of gpu firmware to linux-firmware, the fw files were 92 * placed in toplevel firmware directory, following qcom's 93 * android kernel. But linux-firmware preferred they be 94 * placed in a 'qcom' subdirectory. 95 * 96 * For backwards compatibility, we try first to load from 97 * the new path, using request_firmware_direct() to avoid 98 * any potential timeout waiting for usermode helper, then 99 * fall back to the old path (with direct load). And 100 * finally fall back to request_firmware() with the new 101 * path to allow the usermode helper. 102 */ 103 enum { 104 FW_LOCATION_UNKNOWN = 0, 105 FW_LOCATION_NEW, /* /lib/firmware/qcom/$fwfile */ 106 FW_LOCATION_LEGACY, /* /lib/firmware/$fwfile */ 107 FW_LOCATION_HELPER, 108 } fwloc; 109 110 /* firmware: */ 111 const struct firmware *fw[ADRENO_FW_MAX]; 112 113 /* 114 * Register offsets are different between some GPUs. 115 * GPU specific offsets will be exported by GPU specific 116 * code (a3xx_gpu.c) and stored in this common location. 117 */ 118 const unsigned int *reg_offsets; 119 }; 120 #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base) 121 122 struct adreno_ocmem { 123 struct ocmem *ocmem; 124 unsigned long base; 125 void *hdl; 126 }; 127 128 /* platform config data (ie. from DT, or pdata) */ 129 struct adreno_platform_config { 130 struct adreno_rev rev; 131 }; 132 133 #define ADRENO_IDLE_TIMEOUT msecs_to_jiffies(1000) 134 135 #define spin_until(X) ({ \ 136 int __ret = -ETIMEDOUT; \ 137 unsigned long __t = jiffies + ADRENO_IDLE_TIMEOUT; \ 138 do { \ 139 if (X) { \ 140 __ret = 0; \ 141 break; \ 142 } \ 143 } while (time_before(jiffies, __t)); \ 144 __ret; \ 145 }) 146 147 bool adreno_cmp_rev(struct adreno_rev rev1, struct adreno_rev rev2); 148 149 static inline bool adreno_is_a2xx(struct adreno_gpu *gpu) 150 { 151 return (gpu->revn < 300); 152 } 153 154 static inline bool adreno_is_a20x(struct adreno_gpu *gpu) 155 { 156 return (gpu->revn < 210); 157 } 158 159 static inline bool adreno_is_a225(struct adreno_gpu *gpu) 160 { 161 return gpu->revn == 225; 162 } 163 164 static inline bool adreno_is_a305(struct adreno_gpu *gpu) 165 { 166 return gpu->revn == 305; 167 } 168 169 static inline bool adreno_is_a306(struct adreno_gpu *gpu) 170 { 171 /* yes, 307, because a305c is 306 */ 172 return gpu->revn == 307; 173 } 174 175 static inline bool adreno_is_a320(struct adreno_gpu *gpu) 176 { 177 return gpu->revn == 320; 178 } 179 180 static inline bool adreno_is_a330(struct adreno_gpu *gpu) 181 { 182 return gpu->revn == 330; 183 } 184 185 static inline bool adreno_is_a330v2(struct adreno_gpu *gpu) 186 { 187 return adreno_is_a330(gpu) && (gpu->rev.patchid > 0); 188 } 189 190 static inline int adreno_is_a405(struct adreno_gpu *gpu) 191 { 192 return gpu->revn == 405; 193 } 194 195 static inline int adreno_is_a420(struct adreno_gpu *gpu) 196 { 197 return gpu->revn == 420; 198 } 199 200 static inline int adreno_is_a430(struct adreno_gpu *gpu) 201 { 202 return gpu->revn == 430; 203 } 204 205 static inline int adreno_is_a506(struct adreno_gpu *gpu) 206 { 207 return gpu->revn == 506; 208 } 209 210 static inline int adreno_is_a508(struct adreno_gpu *gpu) 211 { 212 return gpu->revn == 508; 213 } 214 215 static inline int adreno_is_a509(struct adreno_gpu *gpu) 216 { 217 return gpu->revn == 509; 218 } 219 220 static inline int adreno_is_a510(struct adreno_gpu *gpu) 221 { 222 return gpu->revn == 510; 223 } 224 225 static inline int adreno_is_a512(struct adreno_gpu *gpu) 226 { 227 return gpu->revn == 512; 228 } 229 230 static inline int adreno_is_a530(struct adreno_gpu *gpu) 231 { 232 return gpu->revn == 530; 233 } 234 235 static inline int adreno_is_a540(struct adreno_gpu *gpu) 236 { 237 return gpu->revn == 540; 238 } 239 240 static inline int adreno_is_a618(struct adreno_gpu *gpu) 241 { 242 return gpu->revn == 618; 243 } 244 245 static inline int adreno_is_a630(struct adreno_gpu *gpu) 246 { 247 return gpu->revn == 630; 248 } 249 250 static inline int adreno_is_a640_family(struct adreno_gpu *gpu) 251 { 252 return (gpu->revn == 640) || (gpu->revn == 680); 253 } 254 255 static inline int adreno_is_a650(struct adreno_gpu *gpu) 256 { 257 return gpu->revn == 650; 258 } 259 260 static inline int adreno_is_7c3(struct adreno_gpu *gpu) 261 { 262 /* The order of args is important here to handle ANY_ID correctly */ 263 return adreno_cmp_rev(ADRENO_REV(6, 3, 5, ANY_ID), gpu->rev); 264 } 265 266 static inline int adreno_is_a660(struct adreno_gpu *gpu) 267 { 268 return gpu->revn == 660; 269 } 270 271 static inline int adreno_is_a660_family(struct adreno_gpu *gpu) 272 { 273 return adreno_is_a660(gpu) || adreno_is_7c3(gpu); 274 } 275 276 /* check for a650, a660, or any derivatives */ 277 static inline int adreno_is_a650_family(struct adreno_gpu *gpu) 278 { 279 return gpu->revn == 650 || gpu->revn == 620 || 280 adreno_is_a660_family(gpu); 281 } 282 283 int adreno_get_param(struct msm_gpu *gpu, struct msm_file_private *ctx, 284 uint32_t param, uint64_t *value); 285 int adreno_set_param(struct msm_gpu *gpu, struct msm_file_private *ctx, 286 uint32_t param, uint64_t value); 287 const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu, 288 const char *fwname); 289 struct drm_gem_object *adreno_fw_create_bo(struct msm_gpu *gpu, 290 const struct firmware *fw, u64 *iova); 291 int adreno_hw_init(struct msm_gpu *gpu); 292 void adreno_recover(struct msm_gpu *gpu); 293 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring, u32 reg); 294 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 295 #if defined(CONFIG_DEBUG_FS) || defined(CONFIG_DEV_COREDUMP) 296 void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state, 297 struct drm_printer *p); 298 #endif 299 void adreno_dump_info(struct msm_gpu *gpu); 300 void adreno_dump(struct msm_gpu *gpu); 301 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords); 302 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu); 303 304 int adreno_gpu_ocmem_init(struct device *dev, struct adreno_gpu *adreno_gpu, 305 struct adreno_ocmem *ocmem); 306 void adreno_gpu_ocmem_cleanup(struct adreno_ocmem *ocmem); 307 308 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, 309 struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs, 310 int nr_rings); 311 void adreno_gpu_cleanup(struct adreno_gpu *gpu); 312 int adreno_load_fw(struct adreno_gpu *adreno_gpu); 313 314 void adreno_gpu_state_destroy(struct msm_gpu_state *state); 315 316 int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state); 317 int adreno_gpu_state_put(struct msm_gpu_state *state); 318 void adreno_show_object(struct drm_printer *p, void **ptr, int len, 319 bool *encoded); 320 321 /* 322 * Common helper function to initialize the default address space for arm-smmu 323 * attached targets 324 */ 325 struct msm_gem_address_space * 326 adreno_iommu_create_address_space(struct msm_gpu *gpu, 327 struct platform_device *pdev); 328 329 void adreno_set_llc_attributes(struct iommu_domain *iommu); 330 331 int adreno_read_speedbin(struct device *dev, u32 *speedbin); 332 333 /* 334 * For a5xx and a6xx targets load the zap shader that is used to pull the GPU 335 * out of secure mode 336 */ 337 int adreno_zap_shader_load(struct msm_gpu *gpu, u32 pasid); 338 339 /* ringbuffer helpers (the parts that are adreno specific) */ 340 341 static inline void 342 OUT_PKT0(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 343 { 344 adreno_wait_ring(ring, cnt+1); 345 OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF)); 346 } 347 348 /* no-op packet: */ 349 static inline void 350 OUT_PKT2(struct msm_ringbuffer *ring) 351 { 352 adreno_wait_ring(ring, 1); 353 OUT_RING(ring, CP_TYPE2_PKT); 354 } 355 356 static inline void 357 OUT_PKT3(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 358 { 359 adreno_wait_ring(ring, cnt+1); 360 OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8)); 361 } 362 363 static inline u32 PM4_PARITY(u32 val) 364 { 365 return (0x9669 >> (0xF & (val ^ 366 (val >> 4) ^ (val >> 8) ^ (val >> 12) ^ 367 (val >> 16) ^ ((val) >> 20) ^ (val >> 24) ^ 368 (val >> 28)))) & 1; 369 } 370 371 /* Maximum number of values that can be executed for one opcode */ 372 #define TYPE4_MAX_PAYLOAD 127 373 374 #define PKT4(_reg, _cnt) \ 375 (CP_TYPE4_PKT | ((_cnt) << 0) | (PM4_PARITY((_cnt)) << 7) | \ 376 (((_reg) & 0x3FFFF) << 8) | (PM4_PARITY((_reg)) << 27)) 377 378 static inline void 379 OUT_PKT4(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 380 { 381 adreno_wait_ring(ring, cnt + 1); 382 OUT_RING(ring, PKT4(regindx, cnt)); 383 } 384 385 static inline void 386 OUT_PKT7(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 387 { 388 adreno_wait_ring(ring, cnt + 1); 389 OUT_RING(ring, CP_TYPE7_PKT | (cnt << 0) | (PM4_PARITY(cnt) << 15) | 390 ((opcode & 0x7F) << 16) | (PM4_PARITY(opcode) << 23)); 391 } 392 393 struct msm_gpu *a2xx_gpu_init(struct drm_device *dev); 394 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev); 395 struct msm_gpu *a4xx_gpu_init(struct drm_device *dev); 396 struct msm_gpu *a5xx_gpu_init(struct drm_device *dev); 397 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev); 398 399 static inline uint32_t get_wptr(struct msm_ringbuffer *ring) 400 { 401 return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2); 402 } 403 404 /* 405 * Given a register and a count, return a value to program into 406 * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len 407 * registers starting at _reg. 408 * 409 * The register base needs to be a multiple of the length. If it is not, the 410 * hardware will quietly mask off the bits for you and shift the size. For 411 * example, if you intend the protection to start at 0x07 for a length of 4 412 * (0x07-0x0A) the hardware will actually protect (0x04-0x07) which might 413 * expose registers you intended to protect! 414 */ 415 #define ADRENO_PROTECT_RW(_reg, _len) \ 416 ((1 << 30) | (1 << 29) | \ 417 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 418 419 /* 420 * Same as above, but allow reads over the range. For areas of mixed use (such 421 * as performance counters) this allows us to protect a much larger range with a 422 * single register 423 */ 424 #define ADRENO_PROTECT_RDONLY(_reg, _len) \ 425 ((1 << 29) \ 426 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 427 428 429 #define gpu_poll_timeout(gpu, addr, val, cond, interval, timeout) \ 430 readl_poll_timeout((gpu)->mmio + ((addr) << 2), val, cond, \ 431 interval, timeout) 432 433 #endif /* __ADRENO_GPU_H__ */ 434