Tom Rini | 10e4779 | 2018-05-06 17:58:06 -0400 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0+ |
Calvin Johnson | 781b838 | 2018-03-08 15:30:25 +0530 | [diff] [blame] | 2 | /* |
| 3 | * Copyright 2015-2016 Freescale Semiconductor, Inc. |
| 4 | * Copyright 2017 NXP |
Calvin Johnson | 781b838 | 2018-03-08 15:30:25 +0530 | [diff] [blame] | 5 | */ |
Simon Glass | 0f2af88 | 2020-05-10 11:40:05 -0600 | [diff] [blame] | 6 | #include <log.h> |
Simon Glass | 4dcacfc | 2020-05-10 11:40:13 -0600 | [diff] [blame] | 7 | #include <linux/bitops.h> |
Calvin Johnson | 781b838 | 2018-03-08 15:30:25 +0530 | [diff] [blame] | 8 | #include <net/pfe_eth/pfe_eth.h> |
| 9 | #include <net/pfe_eth/pfe/pfe_hw.h> |
| 10 | |
| 11 | static struct pe_info pe[MAX_PE]; |
| 12 | |
| 13 | /* |
| 14 | * Initializes the PFE library. |
| 15 | * Must be called before using any of the library functions. |
| 16 | */ |
| 17 | void pfe_lib_init(void) |
| 18 | { |
| 19 | int pfe_pe_id; |
| 20 | |
| 21 | for (pfe_pe_id = CLASS0_ID; pfe_pe_id <= CLASS_MAX_ID; pfe_pe_id++) { |
| 22 | pe[pfe_pe_id].dmem_base_addr = |
| 23 | (u32)CLASS_DMEM_BASE_ADDR(pfe_pe_id); |
| 24 | pe[pfe_pe_id].pmem_base_addr = |
| 25 | (u32)CLASS_IMEM_BASE_ADDR(pfe_pe_id); |
| 26 | pe[pfe_pe_id].pmem_size = (u32)CLASS_IMEM_SIZE; |
| 27 | pe[pfe_pe_id].mem_access_wdata = |
| 28 | (void *)CLASS_MEM_ACCESS_WDATA; |
| 29 | pe[pfe_pe_id].mem_access_addr = (void *)CLASS_MEM_ACCESS_ADDR; |
| 30 | pe[pfe_pe_id].mem_access_rdata = (void *)CLASS_MEM_ACCESS_RDATA; |
| 31 | } |
| 32 | |
| 33 | for (pfe_pe_id = TMU0_ID; pfe_pe_id <= TMU_MAX_ID; pfe_pe_id++) { |
| 34 | if (pfe_pe_id == TMU2_ID) |
| 35 | continue; |
| 36 | pe[pfe_pe_id].dmem_base_addr = |
| 37 | (u32)TMU_DMEM_BASE_ADDR(pfe_pe_id - TMU0_ID); |
| 38 | pe[pfe_pe_id].pmem_base_addr = |
| 39 | (u32)TMU_IMEM_BASE_ADDR(pfe_pe_id - TMU0_ID); |
| 40 | pe[pfe_pe_id].pmem_size = (u32)TMU_IMEM_SIZE; |
| 41 | pe[pfe_pe_id].mem_access_wdata = (void *)TMU_MEM_ACCESS_WDATA; |
| 42 | pe[pfe_pe_id].mem_access_addr = (void *)TMU_MEM_ACCESS_ADDR; |
| 43 | pe[pfe_pe_id].mem_access_rdata = (void *)TMU_MEM_ACCESS_RDATA; |
| 44 | } |
| 45 | } |
| 46 | |
| 47 | /* |
| 48 | * Writes a buffer to PE internal memory from the host |
| 49 | * through indirect access registers. |
| 50 | * |
| 51 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 52 | * ..., UTIL_ID) |
| 53 | * @param[in] mem_access_addr DMEM destination address (must be 32bit |
| 54 | * aligned) |
| 55 | * @param[in] src Buffer source address |
| 56 | * @param[in] len Number of bytes to copy |
| 57 | */ |
| 58 | static void pe_mem_memcpy_to32(int id, u32 mem_access_addr, const void *src, |
| 59 | unsigned int len) |
| 60 | { |
| 61 | u32 offset = 0, val, addr; |
| 62 | unsigned int len32 = len >> 2; |
| 63 | int i; |
| 64 | |
| 65 | addr = mem_access_addr | PE_MEM_ACCESS_WRITE | |
| 66 | PE_MEM_ACCESS_BYTE_ENABLE(0, 4); |
| 67 | |
| 68 | for (i = 0; i < len32; i++, offset += 4, src += 4) { |
| 69 | val = *(u32 *)src; |
| 70 | writel(cpu_to_be32(val), pe[id].mem_access_wdata); |
| 71 | writel(addr + offset, pe[id].mem_access_addr); |
| 72 | } |
| 73 | |
| 74 | len = (len & 0x3); |
| 75 | if (len) { |
| 76 | val = 0; |
| 77 | |
| 78 | addr = (mem_access_addr | PE_MEM_ACCESS_WRITE | |
| 79 | PE_MEM_ACCESS_BYTE_ENABLE(0, len)) + offset; |
| 80 | |
| 81 | for (i = 0; i < len; i++, src++) |
| 82 | val |= (*(u8 *)src) << (8 * i); |
| 83 | |
| 84 | writel(cpu_to_be32(val), pe[id].mem_access_wdata); |
| 85 | writel(addr, pe[id].mem_access_addr); |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | /* |
| 90 | * Writes a buffer to PE internal data memory (DMEM) from the host |
| 91 | * through indirect access registers. |
| 92 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 93 | * ..., UTIL_ID) |
| 94 | * @param[in] dst DMEM destination address (must be 32bit |
| 95 | * aligned) |
| 96 | * @param[in] src Buffer source address |
| 97 | * @param[in] len Number of bytes to copy |
| 98 | */ |
| 99 | static void pe_dmem_memcpy_to32(int id, u32 dst, const void *src, |
| 100 | unsigned int len) |
| 101 | { |
| 102 | pe_mem_memcpy_to32(id, pe[id].dmem_base_addr | dst | PE_MEM_ACCESS_DMEM, |
| 103 | src, len); |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * Writes a buffer to PE internal program memory (PMEM) from the host |
| 108 | * through indirect access registers. |
| 109 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 110 | * ..., TMU3_ID) |
| 111 | * @param[in] dst PMEM destination address (must be 32bit |
| 112 | * aligned) |
| 113 | * @param[in] src Buffer source address |
| 114 | * @param[in] len Number of bytes to copy |
| 115 | */ |
| 116 | static void pe_pmem_memcpy_to32(int id, u32 dst, const void *src, |
| 117 | unsigned int len) |
| 118 | { |
| 119 | pe_mem_memcpy_to32(id, pe[id].pmem_base_addr | (dst & (pe[id].pmem_size |
| 120 | - 1)) | PE_MEM_ACCESS_IMEM, src, len); |
| 121 | } |
| 122 | |
| 123 | /* |
| 124 | * Reads PE internal program memory (IMEM) from the host |
| 125 | * through indirect access registers. |
| 126 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 127 | * ..., TMU3_ID) |
| 128 | * @param[in] addr PMEM read address (must be aligned on size) |
| 129 | * @param[in] size Number of bytes to read (maximum 4, must not |
| 130 | * cross 32bit boundaries) |
| 131 | * @return the data read (in PE endianness, i.e BE). |
| 132 | */ |
| 133 | u32 pe_pmem_read(int id, u32 addr, u8 size) |
| 134 | { |
| 135 | u32 offset = addr & 0x3; |
| 136 | u32 mask = 0xffffffff >> ((4 - size) << 3); |
| 137 | u32 val; |
| 138 | |
| 139 | addr = pe[id].pmem_base_addr | ((addr & ~0x3) & (pe[id].pmem_size - 1)) |
| 140 | | PE_MEM_ACCESS_READ | PE_MEM_ACCESS_IMEM | |
| 141 | PE_MEM_ACCESS_BYTE_ENABLE(offset, size); |
| 142 | |
| 143 | writel(addr, pe[id].mem_access_addr); |
| 144 | val = be32_to_cpu(readl(pe[id].mem_access_rdata)); |
| 145 | |
| 146 | return (val >> (offset << 3)) & mask; |
| 147 | } |
| 148 | |
| 149 | /* |
| 150 | * Writes PE internal data memory (DMEM) from the host |
| 151 | * through indirect access registers. |
| 152 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 153 | * ..., UTIL_ID) |
| 154 | * @param[in] val Value to write (in PE endianness, i.e BE) |
| 155 | * @param[in] addr DMEM write address (must be aligned on size) |
| 156 | * @param[in] size Number of bytes to write (maximum 4, must not |
| 157 | * cross 32bit boundaries) |
| 158 | */ |
| 159 | void pe_dmem_write(int id, u32 val, u32 addr, u8 size) |
| 160 | { |
| 161 | u32 offset = addr & 0x3; |
| 162 | |
| 163 | addr = pe[id].dmem_base_addr | (addr & ~0x3) | PE_MEM_ACCESS_WRITE | |
| 164 | PE_MEM_ACCESS_DMEM | PE_MEM_ACCESS_BYTE_ENABLE(offset, size); |
| 165 | |
| 166 | /* Indirect access interface is byte swapping data being written */ |
| 167 | writel(cpu_to_be32(val << (offset << 3)), pe[id].mem_access_wdata); |
| 168 | writel(addr, pe[id].mem_access_addr); |
| 169 | } |
| 170 | |
| 171 | /* |
| 172 | * Reads PE internal data memory (DMEM) from the host |
| 173 | * through indirect access registers. |
| 174 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 175 | * ..., UTIL_ID) |
| 176 | * @param[in] addr DMEM read address (must be aligned on size) |
| 177 | * @param[in] size Number of bytes to read (maximum 4, must not |
| 178 | * cross 32bit boundaries) |
| 179 | * @return the data read (in PE endianness, i.e BE). |
| 180 | */ |
| 181 | u32 pe_dmem_read(int id, u32 addr, u8 size) |
| 182 | { |
| 183 | u32 offset = addr & 0x3; |
| 184 | u32 mask = 0xffffffff >> ((4 - size) << 3); |
| 185 | u32 val; |
| 186 | |
| 187 | addr = pe[id].dmem_base_addr | (addr & ~0x3) | PE_MEM_ACCESS_READ | |
| 188 | PE_MEM_ACCESS_DMEM | PE_MEM_ACCESS_BYTE_ENABLE(offset, size); |
| 189 | |
| 190 | writel(addr, pe[id].mem_access_addr); |
| 191 | |
| 192 | /* Indirect access interface is byte swapping data being read */ |
| 193 | val = be32_to_cpu(readl(pe[id].mem_access_rdata)); |
| 194 | |
| 195 | return (val >> (offset << 3)) & mask; |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * This function is used to write to CLASS internal bus peripherals (ccu, |
| 200 | * pe-lem) from the host |
| 201 | * through indirect access registers. |
| 202 | * @param[in] val value to write |
| 203 | * @param[in] addr Address to write to (must be aligned on size) |
| 204 | * @param[in] size Number of bytes to write (1, 2 or 4) |
| 205 | * |
| 206 | */ |
| 207 | static void class_bus_write(u32 val, u32 addr, u8 size) |
| 208 | { |
| 209 | u32 offset = addr & 0x3; |
| 210 | |
| 211 | writel((addr & CLASS_BUS_ACCESS_BASE_MASK), CLASS_BUS_ACCESS_BASE); |
| 212 | |
| 213 | addr = (addr & ~CLASS_BUS_ACCESS_BASE_MASK) | PE_MEM_ACCESS_WRITE | |
| 214 | (size << 24); |
| 215 | |
| 216 | writel(cpu_to_be32(val << (offset << 3)), CLASS_BUS_ACCESS_WDATA); |
| 217 | writel(addr, CLASS_BUS_ACCESS_ADDR); |
| 218 | } |
| 219 | |
| 220 | /* |
| 221 | * Reads from CLASS internal bus peripherals (ccu, pe-lem) from the host |
| 222 | * through indirect access registers. |
| 223 | * @param[in] addr Address to read from (must be aligned on size) |
| 224 | * @param[in] size Number of bytes to read (1, 2 or 4) |
| 225 | * @return the read data |
| 226 | */ |
| 227 | static u32 class_bus_read(u32 addr, u8 size) |
| 228 | { |
| 229 | u32 offset = addr & 0x3; |
| 230 | u32 mask = 0xffffffff >> ((4 - size) << 3); |
| 231 | u32 val; |
| 232 | |
| 233 | writel((addr & CLASS_BUS_ACCESS_BASE_MASK), CLASS_BUS_ACCESS_BASE); |
| 234 | |
| 235 | addr = (addr & ~CLASS_BUS_ACCESS_BASE_MASK) | (size << 24); |
| 236 | |
| 237 | writel(addr, CLASS_BUS_ACCESS_ADDR); |
| 238 | val = be32_to_cpu(readl(CLASS_BUS_ACCESS_RDATA)); |
| 239 | |
| 240 | return (val >> (offset << 3)) & mask; |
| 241 | } |
| 242 | |
| 243 | /* |
| 244 | * Writes data to the cluster memory (PE_LMEM) |
| 245 | * @param[in] dst PE LMEM destination address (must be 32bit aligned) |
| 246 | * @param[in] src Buffer source address |
| 247 | * @param[in] len Number of bytes to copy |
| 248 | */ |
| 249 | static void class_pe_lmem_memcpy_to32(u32 dst, const void *src, |
| 250 | unsigned int len) |
| 251 | { |
| 252 | u32 len32 = len >> 2; |
| 253 | int i; |
| 254 | |
| 255 | for (i = 0; i < len32; i++, src += 4, dst += 4) |
| 256 | class_bus_write(*(u32 *)src, dst, 4); |
| 257 | |
| 258 | if (len & 0x2) { |
| 259 | class_bus_write(*(u16 *)src, dst, 2); |
| 260 | src += 2; |
| 261 | dst += 2; |
| 262 | } |
| 263 | |
| 264 | if (len & 0x1) { |
| 265 | class_bus_write(*(u8 *)src, dst, 1); |
| 266 | src++; |
| 267 | dst++; |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * Writes value to the cluster memory (PE_LMEM) |
| 273 | * @param[in] dst PE LMEM destination address (must be 32bit aligned) |
| 274 | * @param[in] val Value to write |
| 275 | * @param[in] len Number of bytes to write |
| 276 | */ |
| 277 | static void class_pe_lmem_memset(u32 dst, int val, unsigned int len) |
| 278 | { |
| 279 | u32 len32 = len >> 2; |
| 280 | int i; |
| 281 | |
| 282 | val = val | (val << 8) | (val << 16) | (val << 24); |
| 283 | |
| 284 | for (i = 0; i < len32; i++, dst += 4) |
| 285 | class_bus_write(val, dst, 4); |
| 286 | |
| 287 | if (len & 0x2) { |
| 288 | class_bus_write(val, dst, 2); |
| 289 | dst += 2; |
| 290 | } |
| 291 | |
| 292 | if (len & 0x1) { |
| 293 | class_bus_write(val, dst, 1); |
| 294 | dst++; |
| 295 | } |
| 296 | } |
| 297 | |
| 298 | /* |
| 299 | * Reads data from the cluster memory (PE_LMEM) |
| 300 | * @param[out] dst pointer to the source buffer data are copied to |
| 301 | * @param[in] len length in bytes of the amount of data to read |
| 302 | * from cluster memory |
| 303 | * @param[in] offset offset in bytes in the cluster memory where data are |
| 304 | * read from |
| 305 | */ |
| 306 | void pe_lmem_read(u32 *dst, u32 len, u32 offset) |
| 307 | { |
| 308 | u32 len32 = len >> 2; |
| 309 | int i = 0; |
| 310 | |
| 311 | for (i = 0; i < len32; dst++, i++, offset += 4) |
| 312 | *dst = class_bus_read(PE_LMEM_BASE_ADDR + offset, 4); |
| 313 | |
| 314 | if (len & 0x03) |
| 315 | *dst = class_bus_read(PE_LMEM_BASE_ADDR + offset, (len & 0x03)); |
| 316 | } |
| 317 | |
| 318 | /* |
| 319 | * Writes data to the cluster memory (PE_LMEM) |
| 320 | * @param[in] src pointer to the source buffer data are copied from |
| 321 | * @param[in] len length in bytes of the amount of data to write to the |
| 322 | * cluster memory |
| 323 | * @param[in] offset offset in bytes in the cluster memory where data are |
| 324 | * written to |
| 325 | */ |
| 326 | void pe_lmem_write(u32 *src, u32 len, u32 offset) |
| 327 | { |
| 328 | u32 len32 = len >> 2; |
| 329 | int i = 0; |
| 330 | |
| 331 | for (i = 0; i < len32; src++, i++, offset += 4) |
| 332 | class_bus_write(*src, PE_LMEM_BASE_ADDR + offset, 4); |
| 333 | |
| 334 | if (len & 0x03) |
| 335 | class_bus_write(*src, PE_LMEM_BASE_ADDR + offset, (len & |
| 336 | 0x03)); |
| 337 | } |
| 338 | |
| 339 | /* |
| 340 | * Loads an elf section into pmem |
| 341 | * Code needs to be at least 16bit aligned and only PROGBITS sections are |
| 342 | * supported |
| 343 | * |
| 344 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, ..., |
| 345 | * TMU3_ID) |
| 346 | * @param[in] data pointer to the elf firmware |
| 347 | * @param[in] shdr pointer to the elf section header |
| 348 | */ |
| 349 | static int pe_load_pmem_section(int id, const void *data, Elf32_Shdr *shdr) |
| 350 | { |
| 351 | u32 offset = be32_to_cpu(shdr->sh_offset); |
| 352 | u32 addr = be32_to_cpu(shdr->sh_addr); |
| 353 | u32 size = be32_to_cpu(shdr->sh_size); |
| 354 | u32 type = be32_to_cpu(shdr->sh_type); |
| 355 | |
| 356 | if (((unsigned long)(data + offset) & 0x3) != (addr & 0x3)) { |
| 357 | printf( |
| 358 | "%s: load address(%x) and elf file address(%lx) don't have the same alignment\n", |
| 359 | __func__, addr, (unsigned long)data + offset); |
| 360 | |
| 361 | return -1; |
| 362 | } |
| 363 | |
| 364 | if (addr & 0x1) { |
| 365 | printf("%s: load address(%x) is not 16bit aligned\n", |
| 366 | __func__, addr); |
| 367 | return -1; |
| 368 | } |
| 369 | |
| 370 | if (size & 0x1) { |
| 371 | printf("%s: load size(%x) is not 16bit aligned\n", __func__, |
| 372 | size); |
| 373 | return -1; |
| 374 | } |
| 375 | |
| 376 | debug("pmem pe%d @%x len %d\n", id, addr, size); |
| 377 | switch (type) { |
| 378 | case SHT_PROGBITS: |
| 379 | pe_pmem_memcpy_to32(id, addr, data + offset, size); |
| 380 | break; |
| 381 | |
| 382 | default: |
| 383 | printf("%s: unsupported section type(%x)\n", __func__, type); |
| 384 | return -1; |
| 385 | } |
| 386 | |
| 387 | return 0; |
| 388 | } |
| 389 | |
| 390 | /* |
| 391 | * Loads an elf section into dmem |
| 392 | * Data needs to be at least 32bit aligned, NOBITS sections are correctly |
| 393 | * initialized to 0 |
| 394 | * |
| 395 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 396 | * ..., UTIL_ID) |
| 397 | * @param[in] data pointer to the elf firmware |
| 398 | * @param[in] shdr pointer to the elf section header |
| 399 | */ |
| 400 | static int pe_load_dmem_section(int id, const void *data, Elf32_Shdr *shdr) |
| 401 | { |
| 402 | u32 offset = be32_to_cpu(shdr->sh_offset); |
| 403 | u32 addr = be32_to_cpu(shdr->sh_addr); |
| 404 | u32 size = be32_to_cpu(shdr->sh_size); |
| 405 | u32 type = be32_to_cpu(shdr->sh_type); |
| 406 | u32 size32 = size >> 2; |
| 407 | int i; |
| 408 | |
| 409 | if (((unsigned long)(data + offset) & 0x3) != (addr & 0x3)) { |
| 410 | printf( |
| 411 | "%s: load address(%x) and elf file address(%lx) don't have the same alignment\n", |
| 412 | __func__, addr, (unsigned long)data + offset); |
| 413 | |
| 414 | return -1; |
| 415 | } |
| 416 | |
| 417 | if (addr & 0x3) { |
| 418 | printf("%s: load address(%x) is not 32bit aligned\n", |
| 419 | __func__, addr); |
| 420 | return -1; |
| 421 | } |
| 422 | |
| 423 | switch (type) { |
| 424 | case SHT_PROGBITS: |
| 425 | debug("dmem pe%d @%x len %d\n", id, addr, size); |
| 426 | pe_dmem_memcpy_to32(id, addr, data + offset, size); |
| 427 | break; |
| 428 | |
| 429 | case SHT_NOBITS: |
| 430 | debug("dmem zero pe%d @%x len %d\n", id, addr, size); |
| 431 | for (i = 0; i < size32; i++, addr += 4) |
| 432 | pe_dmem_write(id, 0, addr, 4); |
| 433 | |
| 434 | if (size & 0x3) |
| 435 | pe_dmem_write(id, 0, addr, size & 0x3); |
| 436 | |
| 437 | break; |
| 438 | |
| 439 | default: |
| 440 | printf("%s: unsupported section type(%x)\n", __func__, type); |
| 441 | return -1; |
| 442 | } |
| 443 | |
| 444 | return 0; |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | * Loads an elf section into DDR |
| 449 | * Data needs to be at least 32bit aligned, NOBITS sections are correctly |
| 450 | * initialized to 0 |
| 451 | * |
| 452 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 453 | * ..., UTIL_ID) |
| 454 | * @param[in] data pointer to the elf firmware |
| 455 | * @param[in] shdr pointer to the elf section header |
| 456 | */ |
| 457 | static int pe_load_ddr_section(int id, const void *data, Elf32_Shdr *shdr) |
| 458 | { |
| 459 | u32 offset = be32_to_cpu(shdr->sh_offset); |
| 460 | u32 addr = be32_to_cpu(shdr->sh_addr); |
| 461 | u32 size = be32_to_cpu(shdr->sh_size); |
| 462 | u32 type = be32_to_cpu(shdr->sh_type); |
| 463 | u32 flags = be32_to_cpu(shdr->sh_flags); |
| 464 | |
| 465 | switch (type) { |
| 466 | case SHT_PROGBITS: |
| 467 | debug("ddr pe%d @%x len %d\n", id, addr, size); |
| 468 | if (flags & SHF_EXECINSTR) { |
| 469 | if (id <= CLASS_MAX_ID) { |
| 470 | /* DO the loading only once in DDR */ |
| 471 | if (id == CLASS0_ID) { |
| 472 | debug( |
| 473 | "%s: load address(%x) and elf file address(%lx) rcvd\n" |
| 474 | , __func__, addr, |
| 475 | (unsigned long)data + offset); |
| 476 | if (((unsigned long)(data + offset) |
| 477 | & 0x3) != (addr & 0x3)) { |
| 478 | printf( |
| 479 | "%s: load address(%x) and elf file address(%lx) don't have the same alignment\n", |
| 480 | __func__, addr, |
| 481 | (unsigned long)data + |
| 482 | offset); |
| 483 | |
| 484 | return -1; |
| 485 | } |
| 486 | |
| 487 | if (addr & 0x1) { |
| 488 | printf( |
| 489 | "%s: load address(%x) is not 16bit aligned\n" |
| 490 | , __func__, addr); |
| 491 | return -1; |
| 492 | } |
| 493 | |
| 494 | if (size & 0x1) { |
| 495 | printf( |
| 496 | "%s: load length(%x) is not 16bit aligned\n" |
| 497 | , __func__, size); |
| 498 | return -1; |
| 499 | } |
| 500 | |
| 501 | memcpy((void *)DDR_PFE_TO_VIRT(addr), |
| 502 | data + offset, size); |
| 503 | } |
| 504 | } else { |
| 505 | printf( |
| 506 | "%s: unsupported ddr section type(%x) for PE(%d)\n" |
| 507 | , __func__, type, id); |
| 508 | return -1; |
| 509 | } |
| 510 | |
| 511 | } else { |
| 512 | memcpy((void *)DDR_PFE_TO_VIRT(addr), data + offset, |
| 513 | size); |
| 514 | } |
| 515 | |
| 516 | break; |
| 517 | |
| 518 | case SHT_NOBITS: |
| 519 | debug("ddr zero pe%d @%x len %d\n", id, addr, size); |
| 520 | memset((void *)DDR_PFE_TO_VIRT(addr), 0, size); |
| 521 | |
| 522 | break; |
| 523 | |
| 524 | default: |
| 525 | printf("%s: unsupported section type(%x)\n", __func__, type); |
| 526 | return -1; |
| 527 | } |
| 528 | |
| 529 | return 0; |
| 530 | } |
| 531 | |
| 532 | /* |
| 533 | * Loads an elf section into pe lmem |
| 534 | * Data needs to be at least 32bit aligned, NOBITS sections are correctly |
| 535 | * initialized to 0 |
| 536 | * |
| 537 | * @param[in] id PE identification (CLASS0_ID,..., CLASS5_ID) |
| 538 | * @param[in] data pointer to the elf firmware |
| 539 | * @param[in] shdr pointer to the elf section header |
| 540 | */ |
| 541 | static int pe_load_pe_lmem_section(int id, const void *data, Elf32_Shdr *shdr) |
| 542 | { |
| 543 | u32 offset = be32_to_cpu(shdr->sh_offset); |
| 544 | u32 addr = be32_to_cpu(shdr->sh_addr); |
| 545 | u32 size = be32_to_cpu(shdr->sh_size); |
| 546 | u32 type = be32_to_cpu(shdr->sh_type); |
| 547 | |
| 548 | if (id > CLASS_MAX_ID) { |
| 549 | printf("%s: unsupported pe-lmem section type(%x) for PE(%d)\n", |
| 550 | __func__, type, id); |
| 551 | return -1; |
| 552 | } |
| 553 | |
| 554 | if (((unsigned long)(data + offset) & 0x3) != (addr & 0x3)) { |
| 555 | printf( |
| 556 | "%s: load address(%x) and elf file address(%lx) don't have the same alignment\n", |
| 557 | __func__, addr, (unsigned long)data + offset); |
| 558 | |
| 559 | return -1; |
| 560 | } |
| 561 | |
| 562 | if (addr & 0x3) { |
| 563 | printf("%s: load address(%x) is not 32bit aligned\n", |
| 564 | __func__, addr); |
| 565 | return -1; |
| 566 | } |
| 567 | |
| 568 | debug("lmem pe%d @%x len %d\n", id, addr, size); |
| 569 | |
| 570 | switch (type) { |
| 571 | case SHT_PROGBITS: |
| 572 | class_pe_lmem_memcpy_to32(addr, data + offset, size); |
| 573 | break; |
| 574 | |
| 575 | case SHT_NOBITS: |
| 576 | class_pe_lmem_memset(addr, 0, size); |
| 577 | break; |
| 578 | |
| 579 | default: |
| 580 | printf("%s: unsupported section type(%x)\n", __func__, type); |
| 581 | return -1; |
| 582 | } |
| 583 | |
| 584 | return 0; |
| 585 | } |
| 586 | |
| 587 | /* |
| 588 | * Loads an elf section into a PE |
| 589 | * For now only supports loading a section to dmem (all PE's), pmem (class and |
| 590 | * tmu PE's), DDDR (util PE code) |
| 591 | * @param[in] id PE identification (CLASS0_ID, ..., TMU0_ID, |
| 592 | * ..., UTIL_ID) |
| 593 | * @param[in] data pointer to the elf firmware |
| 594 | * @param[in] shdr pointer to the elf section header |
| 595 | */ |
| 596 | int pe_load_elf_section(int id, const void *data, Elf32_Shdr *shdr) |
| 597 | { |
| 598 | u32 addr = be32_to_cpu(shdr->sh_addr); |
| 599 | u32 size = be32_to_cpu(shdr->sh_size); |
| 600 | |
| 601 | if (IS_DMEM(addr, size)) |
| 602 | return pe_load_dmem_section(id, data, shdr); |
| 603 | else if (IS_PMEM(addr, size)) |
| 604 | return pe_load_pmem_section(id, data, shdr); |
| 605 | else if (IS_PFE_LMEM(addr, size)) |
| 606 | return 0; |
| 607 | else if (IS_PHYS_DDR(addr, size)) |
| 608 | return pe_load_ddr_section(id, data, shdr); |
| 609 | else if (IS_PE_LMEM(addr, size)) |
| 610 | return pe_load_pe_lmem_section(id, data, shdr); |
| 611 | |
| 612 | printf("%s: unsupported memory range(%x)\n", __func__, addr); |
| 613 | |
| 614 | return 0; |
| 615 | } |
| 616 | |
| 617 | /**************************** BMU ***************************/ |
| 618 | /* |
| 619 | * Resets a BMU block. |
| 620 | * @param[in] base BMU block base address |
| 621 | */ |
| 622 | static inline void bmu_reset(void *base) |
| 623 | { |
| 624 | writel(CORE_SW_RESET, base + BMU_CTRL); |
| 625 | |
| 626 | /* Wait for self clear */ |
| 627 | while (readl(base + BMU_CTRL) & CORE_SW_RESET) |
| 628 | ; |
| 629 | } |
| 630 | |
| 631 | /* |
| 632 | * Enabled a BMU block. |
| 633 | * @param[in] base BMU block base address |
| 634 | */ |
| 635 | void bmu_enable(void *base) |
| 636 | { |
| 637 | writel(CORE_ENABLE, base + BMU_CTRL); |
| 638 | } |
| 639 | |
| 640 | /* |
| 641 | * Disables a BMU block. |
| 642 | * @param[in] base BMU block base address |
| 643 | */ |
| 644 | static inline void bmu_disable(void *base) |
| 645 | { |
| 646 | writel(CORE_DISABLE, base + BMU_CTRL); |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * Sets the configuration of a BMU block. |
| 651 | * @param[in] base BMU block base address |
| 652 | * @param[in] cfg BMU configuration |
| 653 | */ |
| 654 | static inline void bmu_set_config(void *base, struct bmu_cfg *cfg) |
| 655 | { |
| 656 | writel(cfg->baseaddr, base + BMU_UCAST_BASE_ADDR); |
| 657 | writel(cfg->count & 0xffff, base + BMU_UCAST_CONFIG); |
| 658 | writel(cfg->size & 0xffff, base + BMU_BUF_SIZE); |
| 659 | |
| 660 | /* Interrupts are never used */ |
| 661 | writel(0x0, base + BMU_INT_ENABLE); |
| 662 | } |
| 663 | |
| 664 | /* |
| 665 | * Initializes a BMU block. |
| 666 | * @param[in] base BMU block base address |
| 667 | * @param[in] cfg BMU configuration |
| 668 | */ |
| 669 | void bmu_init(void *base, struct bmu_cfg *cfg) |
| 670 | { |
| 671 | bmu_disable(base); |
| 672 | |
| 673 | bmu_set_config(base, cfg); |
| 674 | |
| 675 | bmu_reset(base); |
| 676 | } |
| 677 | |
| 678 | /**************************** GPI ***************************/ |
| 679 | /* |
| 680 | * Resets a GPI block. |
| 681 | * @param[in] base GPI base address |
| 682 | */ |
| 683 | static inline void gpi_reset(void *base) |
| 684 | { |
| 685 | writel(CORE_SW_RESET, base + GPI_CTRL); |
| 686 | } |
| 687 | |
| 688 | /* |
| 689 | * Enables a GPI block. |
| 690 | * @param[in] base GPI base address |
| 691 | */ |
| 692 | void gpi_enable(void *base) |
| 693 | { |
| 694 | writel(CORE_ENABLE, base + GPI_CTRL); |
| 695 | } |
| 696 | |
| 697 | /* |
| 698 | * Disables a GPI block. |
| 699 | * @param[in] base GPI base address |
| 700 | */ |
| 701 | void gpi_disable(void *base) |
| 702 | { |
| 703 | writel(CORE_DISABLE, base + GPI_CTRL); |
| 704 | } |
| 705 | |
| 706 | /* |
| 707 | * Sets the configuration of a GPI block. |
| 708 | * @param[in] base GPI base address |
| 709 | * @param[in] cfg GPI configuration |
| 710 | */ |
| 711 | static inline void gpi_set_config(void *base, struct gpi_cfg *cfg) |
| 712 | { |
| 713 | writel(CBUS_VIRT_TO_PFE(BMU1_BASE_ADDR + BMU_ALLOC_CTRL), base |
| 714 | + GPI_LMEM_ALLOC_ADDR); |
| 715 | writel(CBUS_VIRT_TO_PFE(BMU1_BASE_ADDR + BMU_FREE_CTRL), base |
| 716 | + GPI_LMEM_FREE_ADDR); |
| 717 | writel(CBUS_VIRT_TO_PFE(BMU2_BASE_ADDR + BMU_ALLOC_CTRL), base |
| 718 | + GPI_DDR_ALLOC_ADDR); |
| 719 | writel(CBUS_VIRT_TO_PFE(BMU2_BASE_ADDR + BMU_FREE_CTRL), base |
| 720 | + GPI_DDR_FREE_ADDR); |
| 721 | writel(CBUS_VIRT_TO_PFE(CLASS_INQ_PKTPTR), base + GPI_CLASS_ADDR); |
| 722 | writel(DDR_HDR_SIZE, base + GPI_DDR_DATA_OFFSET); |
| 723 | writel(LMEM_HDR_SIZE, base + GPI_LMEM_DATA_OFFSET); |
| 724 | writel(0, base + GPI_LMEM_SEC_BUF_DATA_OFFSET); |
| 725 | writel(0, base + GPI_DDR_SEC_BUF_DATA_OFFSET); |
| 726 | writel((DDR_HDR_SIZE << 16) | LMEM_HDR_SIZE, base + GPI_HDR_SIZE); |
| 727 | writel((DDR_BUF_SIZE << 16) | LMEM_BUF_SIZE, base + GPI_BUF_SIZE); |
| 728 | |
| 729 | writel(((cfg->lmem_rtry_cnt << 16) | (GPI_DDR_BUF_EN << 1) | |
| 730 | GPI_LMEM_BUF_EN), base + GPI_RX_CONFIG); |
| 731 | writel(cfg->tmlf_txthres, base + GPI_TMLF_TX); |
| 732 | writel(cfg->aseq_len, base + GPI_DTX_ASEQ); |
| 733 | |
| 734 | /*Make GPI AXI transactions non-bufferable */ |
| 735 | writel(0x1, base + GPI_AXI_CTRL); |
| 736 | } |
| 737 | |
| 738 | /* |
| 739 | * Initializes a GPI block. |
| 740 | * @param[in] base GPI base address |
| 741 | * @param[in] cfg GPI configuration |
| 742 | */ |
| 743 | void gpi_init(void *base, struct gpi_cfg *cfg) |
| 744 | { |
| 745 | gpi_reset(base); |
| 746 | |
| 747 | gpi_disable(base); |
| 748 | |
| 749 | gpi_set_config(base, cfg); |
| 750 | } |
| 751 | |
| 752 | /**************************** CLASSIFIER ***************************/ |
| 753 | /* |
| 754 | * Resets CLASSIFIER block. |
| 755 | */ |
| 756 | static inline void class_reset(void) |
| 757 | { |
| 758 | writel(CORE_SW_RESET, CLASS_TX_CTRL); |
| 759 | } |
| 760 | |
| 761 | /* |
| 762 | * Enables all CLASS-PE's cores. |
| 763 | */ |
| 764 | void class_enable(void) |
| 765 | { |
| 766 | writel(CORE_ENABLE, CLASS_TX_CTRL); |
| 767 | } |
| 768 | |
| 769 | /* |
| 770 | * Disables all CLASS-PE's cores. |
| 771 | */ |
| 772 | void class_disable(void) |
| 773 | { |
| 774 | writel(CORE_DISABLE, CLASS_TX_CTRL); |
| 775 | } |
| 776 | |
| 777 | /* |
| 778 | * Sets the configuration of the CLASSIFIER block. |
| 779 | * @param[in] cfg CLASSIFIER configuration |
| 780 | */ |
| 781 | static inline void class_set_config(struct class_cfg *cfg) |
| 782 | { |
| 783 | if (PLL_CLK_EN == 0) { |
| 784 | /* Clock ratio: for 1:1 the value is 0 */ |
| 785 | writel(0x0, CLASS_PE_SYS_CLK_RATIO); |
| 786 | } else { |
| 787 | /* Clock ratio: for 1:2 the value is 1 */ |
| 788 | writel(0x1, CLASS_PE_SYS_CLK_RATIO); |
| 789 | } |
| 790 | writel((DDR_HDR_SIZE << 16) | LMEM_HDR_SIZE, CLASS_HDR_SIZE); |
| 791 | writel(LMEM_BUF_SIZE, CLASS_LMEM_BUF_SIZE); |
| 792 | writel(CLASS_ROUTE_ENTRY_SIZE(CLASS_ROUTE_SIZE) | |
| 793 | CLASS_ROUTE_HASH_SIZE(cfg->route_table_hash_bits), |
| 794 | CLASS_ROUTE_HASH_ENTRY_SIZE); |
| 795 | writel(HASH_CRC_PORT_IP | QB2BUS_LE, CLASS_ROUTE_MULTI); |
| 796 | |
| 797 | writel(cfg->route_table_baseaddr, CLASS_ROUTE_TABLE_BASE); |
| 798 | memset((void *)DDR_PFE_TO_VIRT(cfg->route_table_baseaddr), 0, |
| 799 | ROUTE_TABLE_SIZE); |
| 800 | |
| 801 | writel(CLASS_PE0_RO_DM_ADDR0_VAL, CLASS_PE0_RO_DM_ADDR0); |
| 802 | writel(CLASS_PE0_RO_DM_ADDR1_VAL, CLASS_PE0_RO_DM_ADDR1); |
| 803 | writel(CLASS_PE0_QB_DM_ADDR0_VAL, CLASS_PE0_QB_DM_ADDR0); |
| 804 | writel(CLASS_PE0_QB_DM_ADDR1_VAL, CLASS_PE0_QB_DM_ADDR1); |
| 805 | writel(CBUS_VIRT_TO_PFE(TMU_PHY_INQ_PKTPTR), CLASS_TM_INQ_ADDR); |
| 806 | |
| 807 | writel(23, CLASS_AFULL_THRES); |
| 808 | writel(23, CLASS_TSQ_FIFO_THRES); |
| 809 | |
| 810 | writel(24, CLASS_MAX_BUF_CNT); |
| 811 | writel(24, CLASS_TSQ_MAX_CNT); |
| 812 | |
| 813 | /*Make Class AXI transactions non-bufferable */ |
| 814 | writel(0x1, CLASS_AXI_CTRL); |
| 815 | |
| 816 | /*Make Util AXI transactions non-bufferable */ |
| 817 | /*Util is disabled in U-boot, do it from here */ |
| 818 | writel(0x1, UTIL_AXI_CTRL); |
| 819 | } |
| 820 | |
| 821 | /* |
| 822 | * Initializes CLASSIFIER block. |
| 823 | * @param[in] cfg CLASSIFIER configuration |
| 824 | */ |
| 825 | void class_init(struct class_cfg *cfg) |
| 826 | { |
| 827 | class_reset(); |
| 828 | |
| 829 | class_disable(); |
| 830 | |
| 831 | class_set_config(cfg); |
| 832 | } |
| 833 | |
| 834 | /**************************** TMU ***************************/ |
| 835 | /* |
| 836 | * Enables TMU-PE cores. |
| 837 | * @param[in] pe_mask TMU PE mask |
| 838 | */ |
| 839 | void tmu_enable(u32 pe_mask) |
| 840 | { |
| 841 | writel(readl(TMU_TX_CTRL) | (pe_mask & 0xF), TMU_TX_CTRL); |
| 842 | } |
| 843 | |
| 844 | /* |
| 845 | * Disables TMU cores. |
| 846 | * @param[in] pe_mask TMU PE mask |
| 847 | */ |
| 848 | void tmu_disable(u32 pe_mask) |
| 849 | { |
| 850 | writel(readl(TMU_TX_CTRL) & ~(pe_mask & 0xF), TMU_TX_CTRL); |
| 851 | } |
| 852 | |
| 853 | /* |
| 854 | * Initializes TMU block. |
| 855 | * @param[in] cfg TMU configuration |
| 856 | */ |
| 857 | void tmu_init(struct tmu_cfg *cfg) |
| 858 | { |
| 859 | int q, phyno; |
| 860 | |
| 861 | /* keep in soft reset */ |
| 862 | writel(SW_RESET, TMU_CTRL); |
| 863 | |
| 864 | /*Make Class AXI transactions non-bufferable */ |
| 865 | writel(0x1, TMU_AXI_CTRL); |
| 866 | |
| 867 | /* enable EMAC PHY ports */ |
| 868 | writel(0x3, TMU_SYS_GENERIC_CONTROL); |
| 869 | |
| 870 | writel(750, TMU_INQ_WATERMARK); |
| 871 | |
| 872 | writel(CBUS_VIRT_TO_PFE(EGPI1_BASE_ADDR + GPI_INQ_PKTPTR), |
| 873 | TMU_PHY0_INQ_ADDR); |
| 874 | writel(CBUS_VIRT_TO_PFE(EGPI2_BASE_ADDR + GPI_INQ_PKTPTR), |
| 875 | TMU_PHY1_INQ_ADDR); |
| 876 | |
| 877 | writel(CBUS_VIRT_TO_PFE(HGPI_BASE_ADDR + GPI_INQ_PKTPTR), |
| 878 | TMU_PHY3_INQ_ADDR); |
| 879 | writel(CBUS_VIRT_TO_PFE(HIF_NOCPY_RX_INQ0_PKTPTR), TMU_PHY4_INQ_ADDR); |
| 880 | writel(CBUS_VIRT_TO_PFE(UTIL_INQ_PKTPTR), TMU_PHY5_INQ_ADDR); |
| 881 | writel(CBUS_VIRT_TO_PFE(BMU2_BASE_ADDR + BMU_FREE_CTRL), |
| 882 | TMU_BMU_INQ_ADDR); |
| 883 | |
| 884 | /* enabling all 10 schedulers [9:0] of each TDQ */ |
| 885 | writel(0x3FF, TMU_TDQ0_SCH_CTRL); |
| 886 | writel(0x3FF, TMU_TDQ1_SCH_CTRL); |
| 887 | writel(0x3FF, TMU_TDQ3_SCH_CTRL); |
| 888 | |
| 889 | if (PLL_CLK_EN == 0) { |
| 890 | /* Clock ratio: for 1:1 the value is 0 */ |
| 891 | writel(0x0, TMU_PE_SYS_CLK_RATIO); |
| 892 | } else { |
| 893 | /* Clock ratio: for 1:2 the value is 1 */ |
| 894 | writel(0x1, TMU_PE_SYS_CLK_RATIO); |
| 895 | } |
| 896 | |
| 897 | /* Extra packet pointers will be stored from this address onwards */ |
| 898 | debug("TMU_LLM_BASE_ADDR %x\n", cfg->llm_base_addr); |
| 899 | writel(cfg->llm_base_addr, TMU_LLM_BASE_ADDR); |
| 900 | |
| 901 | debug("TMU_LLM_QUE_LEN %x\n", cfg->llm_queue_len); |
| 902 | writel(cfg->llm_queue_len, TMU_LLM_QUE_LEN); |
| 903 | |
| 904 | writel(5, TMU_TDQ_IIFG_CFG); |
| 905 | writel(DDR_BUF_SIZE, TMU_BMU_BUF_SIZE); |
| 906 | |
| 907 | writel(0x0, TMU_CTRL); |
| 908 | |
| 909 | /* MEM init */ |
| 910 | writel(MEM_INIT, TMU_CTRL); |
| 911 | |
| 912 | while (!(readl(TMU_CTRL) & MEM_INIT_DONE)) |
| 913 | ; |
| 914 | |
| 915 | /* LLM init */ |
| 916 | writel(LLM_INIT, TMU_CTRL); |
| 917 | |
| 918 | while (!(readl(TMU_CTRL) & LLM_INIT_DONE)) |
| 919 | ; |
| 920 | |
| 921 | /* set up each queue for tail drop */ |
| 922 | for (phyno = 0; phyno < 4; phyno++) { |
| 923 | if (phyno == 2) |
| 924 | continue; |
| 925 | for (q = 0; q < 16; q++) { |
| 926 | u32 qmax; |
| 927 | |
| 928 | writel((phyno << 8) | q, TMU_TEQ_CTRL); |
| 929 | writel(BIT(22), TMU_TEQ_QCFG); |
| 930 | |
| 931 | if (phyno == 3) |
| 932 | qmax = DEFAULT_TMU3_QDEPTH; |
| 933 | else |
| 934 | qmax = (q == 0) ? DEFAULT_Q0_QDEPTH : |
| 935 | DEFAULT_MAX_QDEPTH; |
| 936 | |
| 937 | writel(qmax << 18, TMU_TEQ_HW_PROB_CFG2); |
| 938 | writel(qmax >> 14, TMU_TEQ_HW_PROB_CFG3); |
| 939 | } |
| 940 | } |
| 941 | writel(0x05, TMU_TEQ_DISABLE_DROPCHK); |
| 942 | writel(0, TMU_CTRL); |
| 943 | } |
| 944 | |
| 945 | /**************************** HIF ***************************/ |
| 946 | /* |
| 947 | * Enable hif tx DMA and interrupt |
| 948 | */ |
| 949 | void hif_tx_enable(void) |
| 950 | { |
| 951 | writel(HIF_CTRL_DMA_EN, HIF_TX_CTRL); |
| 952 | } |
| 953 | |
| 954 | /* |
| 955 | * Disable hif tx DMA and interrupt |
| 956 | */ |
| 957 | void hif_tx_disable(void) |
| 958 | { |
| 959 | u32 hif_int; |
| 960 | |
| 961 | writel(0, HIF_TX_CTRL); |
| 962 | |
| 963 | hif_int = readl(HIF_INT_ENABLE); |
| 964 | hif_int &= HIF_TXPKT_INT_EN; |
| 965 | writel(hif_int, HIF_INT_ENABLE); |
| 966 | } |
| 967 | |
| 968 | /* |
| 969 | * Enable hif rx DMA and interrupt |
| 970 | */ |
| 971 | void hif_rx_enable(void) |
| 972 | { |
| 973 | writel((HIF_CTRL_DMA_EN | HIF_CTRL_BDP_CH_START_WSTB), HIF_RX_CTRL); |
| 974 | } |
| 975 | |
| 976 | /* |
| 977 | * Disable hif rx DMA and interrupt |
| 978 | */ |
| 979 | void hif_rx_disable(void) |
| 980 | { |
| 981 | u32 hif_int; |
| 982 | |
| 983 | writel(0, HIF_RX_CTRL); |
| 984 | |
| 985 | hif_int = readl(HIF_INT_ENABLE); |
| 986 | hif_int &= HIF_RXPKT_INT_EN; |
| 987 | writel(hif_int, HIF_INT_ENABLE); |
| 988 | } |
| 989 | |
| 990 | /* |
| 991 | * Initializes HIF copy block. |
| 992 | */ |
| 993 | void hif_init(void) |
| 994 | { |
| 995 | /* Initialize HIF registers */ |
| 996 | writel(HIF_RX_POLL_CTRL_CYCLE << 16 | HIF_TX_POLL_CTRL_CYCLE, |
| 997 | HIF_POLL_CTRL); |
| 998 | /* Make HIF AXI transactions non-bufferable */ |
| 999 | writel(0x1, HIF_AXI_CTRL); |
| 1000 | } |