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