1 /* 2 * Copyright (C) 2013 Red Hat 3 * Author: Rob Clark <robdclark@gmail.com> 4 * 5 * Copyright (c) 2014,2017 The Linux Foundation. All rights reserved. 6 * 7 * This program is free software; you can redistribute it and/or modify it 8 * under the terms of the GNU General Public License version 2 as published by 9 * the Free Software Foundation. 10 * 11 * This program is distributed in the hope that it will be useful, but WITHOUT 12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 14 * more details. 15 * 16 * You should have received a copy of the GNU General Public License along with 17 * this program. If not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #ifndef __ADRENO_GPU_H__ 21 #define __ADRENO_GPU_H__ 22 23 #include <linux/firmware.h> 24 25 #include "msm_gpu.h" 26 27 #include "adreno_common.xml.h" 28 #include "adreno_pm4.xml.h" 29 30 #define REG_ADRENO_DEFINE(_offset, _reg) [_offset] = (_reg) + 1 31 #define REG_SKIP ~0 32 #define REG_ADRENO_SKIP(_offset) [_offset] = REG_SKIP 33 34 /** 35 * adreno_regs: List of registers that are used in across all 36 * 3D devices. Each device type has different offset value for the same 37 * register, so an array of register offsets are declared for every device 38 * and are indexed by the enumeration values defined in this enum 39 */ 40 enum adreno_regs { 41 REG_ADRENO_CP_RB_BASE, 42 REG_ADRENO_CP_RB_BASE_HI, 43 REG_ADRENO_CP_RB_RPTR_ADDR, 44 REG_ADRENO_CP_RB_RPTR_ADDR_HI, 45 REG_ADRENO_CP_RB_RPTR, 46 REG_ADRENO_CP_RB_WPTR, 47 REG_ADRENO_CP_RB_CNTL, 48 REG_ADRENO_REGISTER_MAX, 49 }; 50 51 enum adreno_quirks { 52 ADRENO_QUIRK_TWO_PASS_USE_WFI = 1, 53 ADRENO_QUIRK_FAULT_DETECT_MASK = 2, 54 }; 55 56 struct adreno_rev { 57 uint8_t core; 58 uint8_t major; 59 uint8_t minor; 60 uint8_t patchid; 61 }; 62 63 #define ADRENO_REV(core, major, minor, patchid) \ 64 ((struct adreno_rev){ core, major, minor, patchid }) 65 66 struct adreno_gpu_funcs { 67 struct msm_gpu_funcs base; 68 int (*get_timestamp)(struct msm_gpu *gpu, uint64_t *value); 69 }; 70 71 struct adreno_info { 72 struct adreno_rev rev; 73 uint32_t revn; 74 const char *name; 75 const char *pm4fw, *pfpfw; 76 const char *gpmufw; 77 uint32_t gmem; 78 enum adreno_quirks quirks; 79 struct msm_gpu *(*init)(struct drm_device *dev); 80 const char *zapfw; 81 }; 82 83 const struct adreno_info *adreno_info(struct adreno_rev rev); 84 85 struct adreno_gpu { 86 struct msm_gpu base; 87 struct adreno_rev rev; 88 const struct adreno_info *info; 89 uint32_t gmem; /* actual gmem size */ 90 uint32_t revn; /* numeric revision name */ 91 const struct adreno_gpu_funcs *funcs; 92 93 /* interesting register offsets to dump: */ 94 const unsigned int *registers; 95 96 /* 97 * Are we loading fw from legacy path? Prior to addition 98 * of gpu firmware to linux-firmware, the fw files were 99 * placed in toplevel firmware directory, following qcom's 100 * android kernel. But linux-firmware preferred they be 101 * placed in a 'qcom' subdirectory. 102 * 103 * For backwards compatibility, we try first to load from 104 * the new path, using request_firmware_direct() to avoid 105 * any potential timeout waiting for usermode helper, then 106 * fall back to the old path (with direct load). And 107 * finally fall back to request_firmware() with the new 108 * path to allow the usermode helper. 109 */ 110 enum { 111 FW_LOCATION_UNKNOWN = 0, 112 FW_LOCATION_NEW, /* /lib/firmware/qcom/$fwfile */ 113 FW_LOCATION_LEGACY, /* /lib/firmware/$fwfile */ 114 FW_LOCATION_HELPER, 115 } fwloc; 116 117 /* firmware: */ 118 const struct firmware *pm4, *pfp; 119 120 /* 121 * Register offsets are different between some GPUs. 122 * GPU specific offsets will be exported by GPU specific 123 * code (a3xx_gpu.c) and stored in this common location. 124 */ 125 const unsigned int *reg_offsets; 126 }; 127 #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base) 128 129 /* platform config data (ie. from DT, or pdata) */ 130 struct adreno_platform_config { 131 struct adreno_rev rev; 132 }; 133 134 #define ADRENO_IDLE_TIMEOUT msecs_to_jiffies(1000) 135 136 #define spin_until(X) ({ \ 137 int __ret = -ETIMEDOUT; \ 138 unsigned long __t = jiffies + ADRENO_IDLE_TIMEOUT; \ 139 do { \ 140 if (X) { \ 141 __ret = 0; \ 142 break; \ 143 } \ 144 } while (time_before(jiffies, __t)); \ 145 __ret; \ 146 }) 147 148 149 static inline bool adreno_is_a3xx(struct adreno_gpu *gpu) 150 { 151 return (gpu->revn >= 300) && (gpu->revn < 400); 152 } 153 154 static inline bool adreno_is_a305(struct adreno_gpu *gpu) 155 { 156 return gpu->revn == 305; 157 } 158 159 static inline bool adreno_is_a306(struct adreno_gpu *gpu) 160 { 161 /* yes, 307, because a305c is 306 */ 162 return gpu->revn == 307; 163 } 164 165 static inline bool adreno_is_a320(struct adreno_gpu *gpu) 166 { 167 return gpu->revn == 320; 168 } 169 170 static inline bool adreno_is_a330(struct adreno_gpu *gpu) 171 { 172 return gpu->revn == 330; 173 } 174 175 static inline bool adreno_is_a330v2(struct adreno_gpu *gpu) 176 { 177 return adreno_is_a330(gpu) && (gpu->rev.patchid > 0); 178 } 179 180 static inline bool adreno_is_a4xx(struct adreno_gpu *gpu) 181 { 182 return (gpu->revn >= 400) && (gpu->revn < 500); 183 } 184 185 static inline int adreno_is_a420(struct adreno_gpu *gpu) 186 { 187 return gpu->revn == 420; 188 } 189 190 static inline int adreno_is_a430(struct adreno_gpu *gpu) 191 { 192 return gpu->revn == 430; 193 } 194 195 static inline int adreno_is_a530(struct adreno_gpu *gpu) 196 { 197 return gpu->revn == 530; 198 } 199 200 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value); 201 const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu, 202 const char *fwname); 203 int adreno_hw_init(struct msm_gpu *gpu); 204 void adreno_recover(struct msm_gpu *gpu); 205 void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, 206 struct msm_file_private *ctx); 207 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 208 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 209 #ifdef CONFIG_DEBUG_FS 210 void adreno_show(struct msm_gpu *gpu, struct seq_file *m); 211 #endif 212 void adreno_dump_info(struct msm_gpu *gpu); 213 void adreno_dump(struct msm_gpu *gpu); 214 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords); 215 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu); 216 217 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, 218 struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs, 219 int nr_rings); 220 void adreno_gpu_cleanup(struct adreno_gpu *gpu); 221 222 223 /* ringbuffer helpers (the parts that are adreno specific) */ 224 225 static inline void 226 OUT_PKT0(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 227 { 228 adreno_wait_ring(ring, cnt+1); 229 OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF)); 230 } 231 232 /* no-op packet: */ 233 static inline void 234 OUT_PKT2(struct msm_ringbuffer *ring) 235 { 236 adreno_wait_ring(ring, 1); 237 OUT_RING(ring, CP_TYPE2_PKT); 238 } 239 240 static inline void 241 OUT_PKT3(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 242 { 243 adreno_wait_ring(ring, cnt+1); 244 OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8)); 245 } 246 247 static inline u32 PM4_PARITY(u32 val) 248 { 249 return (0x9669 >> (0xF & (val ^ 250 (val >> 4) ^ (val >> 8) ^ (val >> 12) ^ 251 (val >> 16) ^ ((val) >> 20) ^ (val >> 24) ^ 252 (val >> 28)))) & 1; 253 } 254 255 /* Maximum number of values that can be executed for one opcode */ 256 #define TYPE4_MAX_PAYLOAD 127 257 258 #define PKT4(_reg, _cnt) \ 259 (CP_TYPE4_PKT | ((_cnt) << 0) | (PM4_PARITY((_cnt)) << 7) | \ 260 (((_reg) & 0x3FFFF) << 8) | (PM4_PARITY((_reg)) << 27)) 261 262 static inline void 263 OUT_PKT4(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) 264 { 265 adreno_wait_ring(ring, cnt + 1); 266 OUT_RING(ring, PKT4(regindx, cnt)); 267 } 268 269 static inline void 270 OUT_PKT7(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) 271 { 272 adreno_wait_ring(ring, cnt + 1); 273 OUT_RING(ring, CP_TYPE7_PKT | (cnt << 0) | (PM4_PARITY(cnt) << 15) | 274 ((opcode & 0x7F) << 16) | (PM4_PARITY(opcode) << 23)); 275 } 276 277 /* 278 * adreno_reg_check() - Checks the validity of a register enum 279 * @gpu: Pointer to struct adreno_gpu 280 * @offset_name: The register enum that is checked 281 */ 282 static inline bool adreno_reg_check(struct adreno_gpu *gpu, 283 enum adreno_regs offset_name) 284 { 285 if (offset_name >= REG_ADRENO_REGISTER_MAX || 286 !gpu->reg_offsets[offset_name]) { 287 BUG(); 288 } 289 290 /* 291 * REG_SKIP is a special value that tell us that the register in 292 * question isn't implemented on target but don't trigger a BUG(). This 293 * is used to cleanly implement adreno_gpu_write64() and 294 * adreno_gpu_read64() in a generic fashion 295 */ 296 if (gpu->reg_offsets[offset_name] == REG_SKIP) 297 return false; 298 299 return true; 300 } 301 302 static inline u32 adreno_gpu_read(struct adreno_gpu *gpu, 303 enum adreno_regs offset_name) 304 { 305 u32 reg = gpu->reg_offsets[offset_name]; 306 u32 val = 0; 307 if(adreno_reg_check(gpu,offset_name)) 308 val = gpu_read(&gpu->base, reg - 1); 309 return val; 310 } 311 312 static inline void adreno_gpu_write(struct adreno_gpu *gpu, 313 enum adreno_regs offset_name, u32 data) 314 { 315 u32 reg = gpu->reg_offsets[offset_name]; 316 if(adreno_reg_check(gpu, offset_name)) 317 gpu_write(&gpu->base, reg - 1, data); 318 } 319 320 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev); 321 struct msm_gpu *a4xx_gpu_init(struct drm_device *dev); 322 struct msm_gpu *a5xx_gpu_init(struct drm_device *dev); 323 324 static inline void adreno_gpu_write64(struct adreno_gpu *gpu, 325 enum adreno_regs lo, enum adreno_regs hi, u64 data) 326 { 327 adreno_gpu_write(gpu, lo, lower_32_bits(data)); 328 adreno_gpu_write(gpu, hi, upper_32_bits(data)); 329 } 330 331 static inline uint32_t get_wptr(struct msm_ringbuffer *ring) 332 { 333 return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2); 334 } 335 336 /* 337 * Given a register and a count, return a value to program into 338 * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len 339 * registers starting at _reg. 340 * 341 * The register base needs to be a multiple of the length. If it is not, the 342 * hardware will quietly mask off the bits for you and shift the size. For 343 * example, if you intend the protection to start at 0x07 for a length of 4 344 * (0x07-0x0A) the hardware will actually protect (0x04-0x07) which might 345 * expose registers you intended to protect! 346 */ 347 #define ADRENO_PROTECT_RW(_reg, _len) \ 348 ((1 << 30) | (1 << 29) | \ 349 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 350 351 /* 352 * Same as above, but allow reads over the range. For areas of mixed use (such 353 * as performance counters) this allows us to protect a much larger range with a 354 * single register 355 */ 356 #define ADRENO_PROTECT_RDONLY(_reg, _len) \ 357 ((1 << 29) \ 358 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF)) 359 360 #endif /* __ADRENO_GPU_H__ */ 361