1 // SPDX-License-Identifier: GPL-2.0 2 3 /* 4 * Copyright 2016-2021 HabanaLabs, Ltd. 5 * All Rights Reserved. 6 */ 7 8 #include "habanalabs.h" 9 #include "../include/hw_ip/mmu/mmu_general.h" 10 11 #include <linux/pci.h> 12 #include <linux/uaccess.h> 13 #include <linux/vmalloc.h> 14 #include <linux/iommu.h> 15 16 #define MMU_ADDR_BUF_SIZE 40 17 #define MMU_ASID_BUF_SIZE 10 18 #define MMU_KBUF_SIZE (MMU_ADDR_BUF_SIZE + MMU_ASID_BUF_SIZE) 19 #define I2C_MAX_TRANSACTION_LEN 8 20 21 static struct dentry *hl_debug_root; 22 23 static int hl_debugfs_i2c_read(struct hl_device *hdev, u8 i2c_bus, u8 i2c_addr, 24 u8 i2c_reg, u8 i2c_len, u64 *val) 25 { 26 struct cpucp_packet pkt; 27 int rc; 28 29 if (!hl_device_operational(hdev, NULL)) 30 return -EBUSY; 31 32 if (i2c_len > I2C_MAX_TRANSACTION_LEN) { 33 dev_err(hdev->dev, "I2C transaction length %u, exceeds maximum of %u\n", 34 i2c_len, I2C_MAX_TRANSACTION_LEN); 35 return -EINVAL; 36 } 37 38 memset(&pkt, 0, sizeof(pkt)); 39 40 pkt.ctl = cpu_to_le32(CPUCP_PACKET_I2C_RD << 41 CPUCP_PKT_CTL_OPCODE_SHIFT); 42 pkt.i2c_bus = i2c_bus; 43 pkt.i2c_addr = i2c_addr; 44 pkt.i2c_reg = i2c_reg; 45 pkt.i2c_len = i2c_len; 46 47 rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt), 48 0, val); 49 if (rc) 50 dev_err(hdev->dev, "Failed to read from I2C, error %d\n", rc); 51 52 return rc; 53 } 54 55 static int hl_debugfs_i2c_write(struct hl_device *hdev, u8 i2c_bus, u8 i2c_addr, 56 u8 i2c_reg, u8 i2c_len, u64 val) 57 { 58 struct cpucp_packet pkt; 59 int rc; 60 61 if (!hl_device_operational(hdev, NULL)) 62 return -EBUSY; 63 64 if (i2c_len > I2C_MAX_TRANSACTION_LEN) { 65 dev_err(hdev->dev, "I2C transaction length %u, exceeds maximum of %u\n", 66 i2c_len, I2C_MAX_TRANSACTION_LEN); 67 return -EINVAL; 68 } 69 70 memset(&pkt, 0, sizeof(pkt)); 71 72 pkt.ctl = cpu_to_le32(CPUCP_PACKET_I2C_WR << 73 CPUCP_PKT_CTL_OPCODE_SHIFT); 74 pkt.i2c_bus = i2c_bus; 75 pkt.i2c_addr = i2c_addr; 76 pkt.i2c_reg = i2c_reg; 77 pkt.i2c_len = i2c_len; 78 pkt.value = cpu_to_le64(val); 79 80 rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt), 81 0, NULL); 82 83 if (rc) 84 dev_err(hdev->dev, "Failed to write to I2C, error %d\n", rc); 85 86 return rc; 87 } 88 89 static void hl_debugfs_led_set(struct hl_device *hdev, u8 led, u8 state) 90 { 91 struct cpucp_packet pkt; 92 int rc; 93 94 if (!hl_device_operational(hdev, NULL)) 95 return; 96 97 memset(&pkt, 0, sizeof(pkt)); 98 99 pkt.ctl = cpu_to_le32(CPUCP_PACKET_LED_SET << 100 CPUCP_PKT_CTL_OPCODE_SHIFT); 101 pkt.led_index = cpu_to_le32(led); 102 pkt.value = cpu_to_le64(state); 103 104 rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt), 105 0, NULL); 106 107 if (rc) 108 dev_err(hdev->dev, "Failed to set LED %d, error %d\n", led, rc); 109 } 110 111 static int command_buffers_show(struct seq_file *s, void *data) 112 { 113 struct hl_debugfs_entry *entry = s->private; 114 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 115 struct hl_cb *cb; 116 bool first = true; 117 118 spin_lock(&dev_entry->cb_spinlock); 119 120 list_for_each_entry(cb, &dev_entry->cb_list, debugfs_list) { 121 if (first) { 122 first = false; 123 seq_puts(s, "\n"); 124 seq_puts(s, " CB ID CTX ID CB size CB RefCnt mmap? CS counter\n"); 125 seq_puts(s, "---------------------------------------------------------------\n"); 126 } 127 seq_printf(s, 128 " %03llu %d 0x%08x %d %d %d\n", 129 cb->buf->handle, cb->ctx->asid, cb->size, 130 kref_read(&cb->buf->refcount), 131 atomic_read(&cb->buf->mmap), atomic_read(&cb->cs_cnt)); 132 } 133 134 spin_unlock(&dev_entry->cb_spinlock); 135 136 if (!first) 137 seq_puts(s, "\n"); 138 139 return 0; 140 } 141 142 static int command_submission_show(struct seq_file *s, void *data) 143 { 144 struct hl_debugfs_entry *entry = s->private; 145 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 146 struct hl_cs *cs; 147 bool first = true; 148 149 spin_lock(&dev_entry->cs_spinlock); 150 151 list_for_each_entry(cs, &dev_entry->cs_list, debugfs_list) { 152 if (first) { 153 first = false; 154 seq_puts(s, "\n"); 155 seq_puts(s, " CS ID CS TYPE CTX ASID CS RefCnt Submitted Completed\n"); 156 seq_puts(s, "----------------------------------------------------------------\n"); 157 } 158 seq_printf(s, 159 " %llu %d %d %d %d %d\n", 160 cs->sequence, cs->type, cs->ctx->asid, 161 kref_read(&cs->refcount), 162 cs->submitted, cs->completed); 163 } 164 165 spin_unlock(&dev_entry->cs_spinlock); 166 167 if (!first) 168 seq_puts(s, "\n"); 169 170 return 0; 171 } 172 173 static int command_submission_jobs_show(struct seq_file *s, void *data) 174 { 175 struct hl_debugfs_entry *entry = s->private; 176 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 177 struct hl_cs_job *job; 178 bool first = true; 179 180 spin_lock(&dev_entry->cs_job_spinlock); 181 182 list_for_each_entry(job, &dev_entry->cs_job_list, debugfs_list) { 183 if (first) { 184 first = false; 185 seq_puts(s, "\n"); 186 seq_puts(s, " JOB ID CS ID CS TYPE CTX ASID JOB RefCnt H/W Queue\n"); 187 seq_puts(s, "---------------------------------------------------------------\n"); 188 } 189 if (job->cs) 190 seq_printf(s, 191 " %02d %llu %d %d %d %d\n", 192 job->id, job->cs->sequence, job->cs->type, 193 job->cs->ctx->asid, kref_read(&job->refcount), 194 job->hw_queue_id); 195 else 196 seq_printf(s, 197 " %02d 0 0 %d %d %d\n", 198 job->id, HL_KERNEL_ASID_ID, 199 kref_read(&job->refcount), job->hw_queue_id); 200 } 201 202 spin_unlock(&dev_entry->cs_job_spinlock); 203 204 if (!first) 205 seq_puts(s, "\n"); 206 207 return 0; 208 } 209 210 static int userptr_show(struct seq_file *s, void *data) 211 { 212 struct hl_debugfs_entry *entry = s->private; 213 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 214 struct hl_userptr *userptr; 215 char dma_dir[4][30] = {"DMA_BIDIRECTIONAL", "DMA_TO_DEVICE", 216 "DMA_FROM_DEVICE", "DMA_NONE"}; 217 bool first = true; 218 219 spin_lock(&dev_entry->userptr_spinlock); 220 221 list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) { 222 if (first) { 223 first = false; 224 seq_puts(s, "\n"); 225 seq_puts(s, " pid user virtual address size dma dir\n"); 226 seq_puts(s, "----------------------------------------------------------\n"); 227 } 228 seq_printf(s, " %-7d 0x%-14llx %-10llu %-30s\n", 229 userptr->pid, userptr->addr, userptr->size, 230 dma_dir[userptr->dir]); 231 } 232 233 spin_unlock(&dev_entry->userptr_spinlock); 234 235 if (!first) 236 seq_puts(s, "\n"); 237 238 return 0; 239 } 240 241 static int vm_show(struct seq_file *s, void *data) 242 { 243 struct hl_debugfs_entry *entry = s->private; 244 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 245 struct hl_vm_hw_block_list_node *lnode; 246 struct hl_ctx *ctx; 247 struct hl_vm *vm; 248 struct hl_vm_hash_node *hnode; 249 struct hl_userptr *userptr; 250 struct hl_vm_phys_pg_pack *phys_pg_pack = NULL; 251 struct hl_va_range *va_range; 252 struct hl_vm_va_block *va_block; 253 enum vm_type *vm_type; 254 bool once = true; 255 u64 j; 256 int i; 257 258 mutex_lock(&dev_entry->ctx_mem_hash_mutex); 259 260 list_for_each_entry(ctx, &dev_entry->ctx_mem_hash_list, debugfs_list) { 261 once = false; 262 seq_puts(s, "\n\n----------------------------------------------------"); 263 seq_puts(s, "\n----------------------------------------------------\n\n"); 264 seq_printf(s, "ctx asid: %u\n", ctx->asid); 265 266 seq_puts(s, "\nmappings:\n\n"); 267 seq_puts(s, " virtual address size handle\n"); 268 seq_puts(s, "----------------------------------------------------\n"); 269 mutex_lock(&ctx->mem_hash_lock); 270 hash_for_each(ctx->mem_hash, i, hnode, node) { 271 vm_type = hnode->ptr; 272 273 if (*vm_type == VM_TYPE_USERPTR) { 274 userptr = hnode->ptr; 275 seq_printf(s, 276 " 0x%-14llx %-10llu\n", 277 hnode->vaddr, userptr->size); 278 } else { 279 phys_pg_pack = hnode->ptr; 280 seq_printf(s, 281 " 0x%-14llx %-10llu %-4u\n", 282 hnode->vaddr, phys_pg_pack->total_size, 283 phys_pg_pack->handle); 284 } 285 } 286 mutex_unlock(&ctx->mem_hash_lock); 287 288 if (ctx->asid != HL_KERNEL_ASID_ID && 289 !list_empty(&ctx->hw_block_mem_list)) { 290 seq_puts(s, "\nhw_block mappings:\n\n"); 291 seq_puts(s, 292 " virtual address block size mapped size HW block id\n"); 293 seq_puts(s, 294 "---------------------------------------------------------------\n"); 295 mutex_lock(&ctx->hw_block_list_lock); 296 list_for_each_entry(lnode, &ctx->hw_block_mem_list, node) { 297 seq_printf(s, 298 " 0x%-14lx %-6u %-6u %-9u\n", 299 lnode->vaddr, lnode->block_size, lnode->mapped_size, 300 lnode->id); 301 } 302 mutex_unlock(&ctx->hw_block_list_lock); 303 } 304 305 vm = &ctx->hdev->vm; 306 spin_lock(&vm->idr_lock); 307 308 if (!idr_is_empty(&vm->phys_pg_pack_handles)) 309 seq_puts(s, "\n\nallocations:\n"); 310 311 idr_for_each_entry(&vm->phys_pg_pack_handles, phys_pg_pack, i) { 312 if (phys_pg_pack->asid != ctx->asid) 313 continue; 314 315 seq_printf(s, "\nhandle: %u\n", phys_pg_pack->handle); 316 seq_printf(s, "page size: %u\n\n", 317 phys_pg_pack->page_size); 318 seq_puts(s, " physical address\n"); 319 seq_puts(s, "---------------------\n"); 320 for (j = 0 ; j < phys_pg_pack->npages ; j++) { 321 seq_printf(s, " 0x%-14llx\n", 322 phys_pg_pack->pages[j]); 323 } 324 } 325 spin_unlock(&vm->idr_lock); 326 327 } 328 329 mutex_unlock(&dev_entry->ctx_mem_hash_mutex); 330 331 ctx = hl_get_compute_ctx(dev_entry->hdev); 332 if (ctx) { 333 seq_puts(s, "\nVA ranges:\n\n"); 334 for (i = HL_VA_RANGE_TYPE_HOST ; i < HL_VA_RANGE_TYPE_MAX ; ++i) { 335 va_range = ctx->va_range[i]; 336 seq_printf(s, " va_range %d\n", i); 337 seq_puts(s, "---------------------\n"); 338 mutex_lock(&va_range->lock); 339 list_for_each_entry(va_block, &va_range->list, node) { 340 seq_printf(s, "%#16llx - %#16llx (%#llx)\n", 341 va_block->start, va_block->end, 342 va_block->size); 343 } 344 mutex_unlock(&va_range->lock); 345 seq_puts(s, "\n"); 346 } 347 hl_ctx_put(ctx); 348 } 349 350 if (!once) 351 seq_puts(s, "\n"); 352 353 return 0; 354 } 355 356 static int userptr_lookup_show(struct seq_file *s, void *data) 357 { 358 struct hl_debugfs_entry *entry = s->private; 359 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 360 struct scatterlist *sg; 361 struct hl_userptr *userptr; 362 bool first = true; 363 u64 total_npages, npages, sg_start, sg_end; 364 dma_addr_t dma_addr; 365 int i; 366 367 spin_lock(&dev_entry->userptr_spinlock); 368 369 list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) { 370 if (dev_entry->userptr_lookup >= userptr->addr && 371 dev_entry->userptr_lookup < userptr->addr + userptr->size) { 372 total_npages = 0; 373 for_each_sgtable_dma_sg(userptr->sgt, sg, i) { 374 npages = hl_get_sg_info(sg, &dma_addr); 375 sg_start = userptr->addr + 376 total_npages * PAGE_SIZE; 377 sg_end = userptr->addr + 378 (total_npages + npages) * PAGE_SIZE; 379 380 if (dev_entry->userptr_lookup >= sg_start && 381 dev_entry->userptr_lookup < sg_end) { 382 dma_addr += (dev_entry->userptr_lookup - 383 sg_start); 384 if (first) { 385 first = false; 386 seq_puts(s, "\n"); 387 seq_puts(s, " user virtual address dma address pid region start region size\n"); 388 seq_puts(s, "---------------------------------------------------------------------------------------\n"); 389 } 390 seq_printf(s, " 0x%-18llx 0x%-16llx %-8u 0x%-16llx %-12llu\n", 391 dev_entry->userptr_lookup, 392 (u64)dma_addr, userptr->pid, 393 userptr->addr, userptr->size); 394 } 395 total_npages += npages; 396 } 397 } 398 } 399 400 spin_unlock(&dev_entry->userptr_spinlock); 401 402 if (!first) 403 seq_puts(s, "\n"); 404 405 return 0; 406 } 407 408 static ssize_t userptr_lookup_write(struct file *file, const char __user *buf, 409 size_t count, loff_t *f_pos) 410 { 411 struct seq_file *s = file->private_data; 412 struct hl_debugfs_entry *entry = s->private; 413 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 414 ssize_t rc; 415 u64 value; 416 417 rc = kstrtoull_from_user(buf, count, 16, &value); 418 if (rc) 419 return rc; 420 421 dev_entry->userptr_lookup = value; 422 423 return count; 424 } 425 426 static int mmu_show(struct seq_file *s, void *data) 427 { 428 struct hl_debugfs_entry *entry = s->private; 429 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 430 struct hl_device *hdev = dev_entry->hdev; 431 struct hl_ctx *ctx; 432 struct hl_mmu_hop_info hops_info = {0}; 433 u64 virt_addr = dev_entry->mmu_addr, phys_addr; 434 int i; 435 436 if (dev_entry->mmu_asid == HL_KERNEL_ASID_ID) 437 ctx = hdev->kernel_ctx; 438 else 439 ctx = hl_get_compute_ctx(hdev); 440 441 if (!ctx) { 442 dev_err(hdev->dev, "no ctx available\n"); 443 return 0; 444 } 445 446 if (hl_mmu_get_tlb_info(ctx, virt_addr, &hops_info)) { 447 dev_err(hdev->dev, "virt addr 0x%llx is not mapped to phys addr\n", 448 virt_addr); 449 goto put_ctx; 450 } 451 452 hl_mmu_va_to_pa(ctx, virt_addr, &phys_addr); 453 454 if (hops_info.scrambled_vaddr && 455 (dev_entry->mmu_addr != hops_info.scrambled_vaddr)) 456 seq_printf(s, 457 "asid: %u, virt_addr: 0x%llx, scrambled virt_addr: 0x%llx,\nphys_addr: 0x%llx, scrambled_phys_addr: 0x%llx\n", 458 dev_entry->mmu_asid, dev_entry->mmu_addr, 459 hops_info.scrambled_vaddr, 460 hops_info.unscrambled_paddr, phys_addr); 461 else 462 seq_printf(s, 463 "asid: %u, virt_addr: 0x%llx, phys_addr: 0x%llx\n", 464 dev_entry->mmu_asid, dev_entry->mmu_addr, phys_addr); 465 466 for (i = 0 ; i < hops_info.used_hops ; i++) { 467 seq_printf(s, "hop%d_addr: 0x%llx\n", 468 i, hops_info.hop_info[i].hop_addr); 469 seq_printf(s, "hop%d_pte_addr: 0x%llx\n", 470 i, hops_info.hop_info[i].hop_pte_addr); 471 seq_printf(s, "hop%d_pte: 0x%llx\n", 472 i, hops_info.hop_info[i].hop_pte_val); 473 } 474 475 put_ctx: 476 if (dev_entry->mmu_asid != HL_KERNEL_ASID_ID) 477 hl_ctx_put(ctx); 478 479 return 0; 480 } 481 482 static ssize_t mmu_asid_va_write(struct file *file, const char __user *buf, 483 size_t count, loff_t *f_pos) 484 { 485 struct seq_file *s = file->private_data; 486 struct hl_debugfs_entry *entry = s->private; 487 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 488 struct hl_device *hdev = dev_entry->hdev; 489 char kbuf[MMU_KBUF_SIZE]; 490 char *c; 491 ssize_t rc; 492 493 if (count > sizeof(kbuf) - 1) 494 goto err; 495 if (copy_from_user(kbuf, buf, count)) 496 goto err; 497 kbuf[count] = 0; 498 499 c = strchr(kbuf, ' '); 500 if (!c) 501 goto err; 502 *c = '\0'; 503 504 rc = kstrtouint(kbuf, 10, &dev_entry->mmu_asid); 505 if (rc) 506 goto err; 507 508 if (strncmp(c+1, "0x", 2)) 509 goto err; 510 rc = kstrtoull(c+3, 16, &dev_entry->mmu_addr); 511 if (rc) 512 goto err; 513 514 return count; 515 516 err: 517 dev_err(hdev->dev, "usage: echo <asid> <0xaddr> > mmu\n"); 518 519 return -EINVAL; 520 } 521 522 static int mmu_ack_error(struct seq_file *s, void *data) 523 { 524 struct hl_debugfs_entry *entry = s->private; 525 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 526 struct hl_device *hdev = dev_entry->hdev; 527 int rc; 528 529 if (!dev_entry->mmu_cap_mask) { 530 dev_err(hdev->dev, "mmu_cap_mask is not set\n"); 531 goto err; 532 } 533 534 rc = hdev->asic_funcs->ack_mmu_errors(hdev, dev_entry->mmu_cap_mask); 535 if (rc) 536 goto err; 537 538 return 0; 539 err: 540 return -EINVAL; 541 } 542 543 static ssize_t mmu_ack_error_value_write(struct file *file, 544 const char __user *buf, 545 size_t count, loff_t *f_pos) 546 { 547 struct seq_file *s = file->private_data; 548 struct hl_debugfs_entry *entry = s->private; 549 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 550 struct hl_device *hdev = dev_entry->hdev; 551 char kbuf[MMU_KBUF_SIZE]; 552 ssize_t rc; 553 554 if (count > sizeof(kbuf) - 1) 555 goto err; 556 557 if (copy_from_user(kbuf, buf, count)) 558 goto err; 559 560 kbuf[count] = 0; 561 562 if (strncmp(kbuf, "0x", 2)) 563 goto err; 564 565 rc = kstrtoull(kbuf, 16, &dev_entry->mmu_cap_mask); 566 if (rc) 567 goto err; 568 569 return count; 570 err: 571 dev_err(hdev->dev, "usage: echo <0xmmu_cap_mask > > mmu_error\n"); 572 573 return -EINVAL; 574 } 575 576 static int engines_show(struct seq_file *s, void *data) 577 { 578 struct hl_debugfs_entry *entry = s->private; 579 struct hl_dbg_device_entry *dev_entry = entry->dev_entry; 580 struct hl_device *hdev = dev_entry->hdev; 581 struct engines_data eng_data; 582 583 if (hdev->reset_info.in_reset) { 584 dev_warn_ratelimited(hdev->dev, 585 "Can't check device idle during reset\n"); 586 return 0; 587 } 588 589 eng_data.actual_size = 0; 590 eng_data.allocated_buf_size = HL_ENGINES_DATA_MAX_SIZE; 591 eng_data.buf = vmalloc(eng_data.allocated_buf_size); 592 if (!eng_data.buf) 593 return -ENOMEM; 594 595 hdev->asic_funcs->is_device_idle(hdev, NULL, 0, &eng_data); 596 597 if (eng_data.actual_size > eng_data.allocated_buf_size) { 598 dev_err(hdev->dev, 599 "Engines data size (%d Bytes) is bigger than allocated size (%u Bytes)\n", 600 eng_data.actual_size, eng_data.allocated_buf_size); 601 vfree(eng_data.buf); 602 return -ENOMEM; 603 } 604 605 seq_write(s, eng_data.buf, eng_data.actual_size); 606 607 vfree(eng_data.buf); 608 609 return 0; 610 } 611 612 static ssize_t hl_memory_scrub(struct file *f, const char __user *buf, 613 size_t count, loff_t *ppos) 614 { 615 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 616 struct hl_device *hdev = entry->hdev; 617 u64 val = hdev->memory_scrub_val; 618 int rc; 619 620 if (!hl_device_operational(hdev, NULL)) { 621 dev_warn_ratelimited(hdev->dev, "Can't scrub memory, device is not operational\n"); 622 return -EIO; 623 } 624 625 mutex_lock(&hdev->fpriv_list_lock); 626 if (hdev->is_compute_ctx_active) { 627 mutex_unlock(&hdev->fpriv_list_lock); 628 dev_err(hdev->dev, "can't scrub dram, context exist\n"); 629 return -EBUSY; 630 } 631 hdev->is_in_dram_scrub = true; 632 mutex_unlock(&hdev->fpriv_list_lock); 633 634 rc = hdev->asic_funcs->scrub_device_dram(hdev, val); 635 636 mutex_lock(&hdev->fpriv_list_lock); 637 hdev->is_in_dram_scrub = false; 638 mutex_unlock(&hdev->fpriv_list_lock); 639 640 if (rc) 641 return rc; 642 return count; 643 } 644 645 static bool hl_is_device_va(struct hl_device *hdev, u64 addr) 646 { 647 struct asic_fixed_properties *prop = &hdev->asic_prop; 648 649 if (prop->dram_supports_virtual_memory && 650 (addr >= prop->dmmu.start_addr && addr < prop->dmmu.end_addr)) 651 return true; 652 653 if (addr >= prop->pmmu.start_addr && 654 addr < prop->pmmu.end_addr) 655 return true; 656 657 if (addr >= prop->pmmu_huge.start_addr && 658 addr < prop->pmmu_huge.end_addr) 659 return true; 660 661 return false; 662 } 663 664 static bool hl_is_device_internal_memory_va(struct hl_device *hdev, u64 addr, 665 u32 size) 666 { 667 struct asic_fixed_properties *prop = &hdev->asic_prop; 668 u64 dram_start_addr, dram_end_addr; 669 670 if (prop->dram_supports_virtual_memory) { 671 dram_start_addr = prop->dmmu.start_addr; 672 dram_end_addr = prop->dmmu.end_addr; 673 } else { 674 dram_start_addr = prop->dram_base_address; 675 dram_end_addr = prop->dram_end_address; 676 } 677 678 if (hl_mem_area_inside_range(addr, size, dram_start_addr, 679 dram_end_addr)) 680 return true; 681 682 if (hl_mem_area_inside_range(addr, size, prop->sram_base_address, 683 prop->sram_end_address)) 684 return true; 685 686 return false; 687 } 688 689 static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size, 690 u64 *phys_addr) 691 { 692 struct hl_vm_phys_pg_pack *phys_pg_pack; 693 struct hl_ctx *ctx; 694 struct hl_vm_hash_node *hnode; 695 u64 end_address, range_size; 696 struct hl_userptr *userptr; 697 enum vm_type *vm_type; 698 bool valid = false; 699 int i, rc = 0; 700 701 ctx = hl_get_compute_ctx(hdev); 702 703 if (!ctx) { 704 dev_err(hdev->dev, "no ctx available\n"); 705 return -EINVAL; 706 } 707 708 /* Verify address is mapped */ 709 mutex_lock(&ctx->mem_hash_lock); 710 hash_for_each(ctx->mem_hash, i, hnode, node) { 711 vm_type = hnode->ptr; 712 713 if (*vm_type == VM_TYPE_USERPTR) { 714 userptr = hnode->ptr; 715 range_size = userptr->size; 716 } else { 717 phys_pg_pack = hnode->ptr; 718 range_size = phys_pg_pack->total_size; 719 } 720 721 end_address = virt_addr + size; 722 if ((virt_addr >= hnode->vaddr) && 723 (end_address <= hnode->vaddr + range_size)) { 724 valid = true; 725 break; 726 } 727 } 728 mutex_unlock(&ctx->mem_hash_lock); 729 730 if (!valid) { 731 dev_err(hdev->dev, 732 "virt addr 0x%llx is not mapped\n", 733 virt_addr); 734 rc = -EINVAL; 735 goto put_ctx; 736 } 737 738 rc = hl_mmu_va_to_pa(ctx, virt_addr, phys_addr); 739 if (rc) { 740 dev_err(hdev->dev, 741 "virt addr 0x%llx is not mapped to phys addr\n", 742 virt_addr); 743 rc = -EINVAL; 744 } 745 746 put_ctx: 747 hl_ctx_put(ctx); 748 749 return rc; 750 } 751 752 static int hl_access_dev_mem_by_region(struct hl_device *hdev, u64 addr, 753 u64 *val, enum debugfs_access_type acc_type, bool *found) 754 { 755 size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ? 756 sizeof(u64) : sizeof(u32); 757 struct pci_mem_region *mem_reg; 758 int i; 759 760 for (i = 0; i < PCI_REGION_NUMBER; i++) { 761 mem_reg = &hdev->pci_mem_region[i]; 762 if (!mem_reg->used) 763 continue; 764 if (addr >= mem_reg->region_base && 765 addr <= mem_reg->region_base + mem_reg->region_size - acc_size) { 766 *found = true; 767 return hdev->asic_funcs->access_dev_mem(hdev, i, addr, val, acc_type); 768 } 769 } 770 return 0; 771 } 772 773 static void hl_access_host_mem(struct hl_device *hdev, u64 addr, u64 *val, 774 enum debugfs_access_type acc_type) 775 { 776 struct asic_fixed_properties *prop = &hdev->asic_prop; 777 u64 offset = prop->device_dma_offset_for_host_access; 778 779 switch (acc_type) { 780 case DEBUGFS_READ32: 781 *val = *(u32 *) phys_to_virt(addr - offset); 782 break; 783 case DEBUGFS_WRITE32: 784 *(u32 *) phys_to_virt(addr - offset) = *val; 785 break; 786 case DEBUGFS_READ64: 787 *val = *(u64 *) phys_to_virt(addr - offset); 788 break; 789 case DEBUGFS_WRITE64: 790 *(u64 *) phys_to_virt(addr - offset) = *val; 791 break; 792 default: 793 dev_err(hdev->dev, "hostmem access-type %d id not supported\n", acc_type); 794 break; 795 } 796 } 797 798 static int hl_access_mem(struct hl_device *hdev, u64 addr, u64 *val, 799 enum debugfs_access_type acc_type) 800 { 801 size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ? 802 sizeof(u64) : sizeof(u32); 803 u64 host_start = hdev->asic_prop.host_base_address; 804 u64 host_end = hdev->asic_prop.host_end_address; 805 bool user_address, found = false; 806 int rc; 807 808 user_address = hl_is_device_va(hdev, addr); 809 if (user_address) { 810 rc = device_va_to_pa(hdev, addr, acc_size, &addr); 811 if (rc) 812 return rc; 813 } 814 815 rc = hl_access_dev_mem_by_region(hdev, addr, val, acc_type, &found); 816 if (rc) { 817 dev_err(hdev->dev, 818 "Failed reading addr %#llx from dev mem (%d)\n", 819 addr, rc); 820 return rc; 821 } 822 823 if (found) 824 return 0; 825 826 if (!user_address || device_iommu_mapped(&hdev->pdev->dev)) { 827 rc = -EINVAL; 828 goto err; 829 } 830 831 if (addr >= host_start && addr <= host_end - acc_size) { 832 hl_access_host_mem(hdev, addr, val, acc_type); 833 } else { 834 rc = -EINVAL; 835 goto err; 836 } 837 838 return 0; 839 err: 840 dev_err(hdev->dev, "invalid addr %#llx\n", addr); 841 return rc; 842 } 843 844 static ssize_t hl_data_read32(struct file *f, char __user *buf, 845 size_t count, loff_t *ppos) 846 { 847 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 848 struct hl_device *hdev = entry->hdev; 849 u64 value64, addr = entry->addr; 850 char tmp_buf[32]; 851 ssize_t rc; 852 u32 val; 853 854 if (hdev->reset_info.in_reset) { 855 dev_warn_ratelimited(hdev->dev, "Can't read during reset\n"); 856 return 0; 857 } 858 859 if (*ppos) 860 return 0; 861 862 rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_READ32); 863 if (rc) 864 return rc; 865 866 val = value64; /* downcast back to 32 */ 867 868 sprintf(tmp_buf, "0x%08x\n", val); 869 return simple_read_from_buffer(buf, count, ppos, tmp_buf, 870 strlen(tmp_buf)); 871 } 872 873 static ssize_t hl_data_write32(struct file *f, const char __user *buf, 874 size_t count, loff_t *ppos) 875 { 876 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 877 struct hl_device *hdev = entry->hdev; 878 u64 value64, addr = entry->addr; 879 u32 value; 880 ssize_t rc; 881 882 if (hdev->reset_info.in_reset) { 883 dev_warn_ratelimited(hdev->dev, "Can't write during reset\n"); 884 return 0; 885 } 886 887 rc = kstrtouint_from_user(buf, count, 16, &value); 888 if (rc) 889 return rc; 890 891 value64 = value; 892 rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_WRITE32); 893 if (rc) 894 return rc; 895 896 return count; 897 } 898 899 static ssize_t hl_data_read64(struct file *f, char __user *buf, 900 size_t count, loff_t *ppos) 901 { 902 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 903 struct hl_device *hdev = entry->hdev; 904 u64 addr = entry->addr; 905 char tmp_buf[32]; 906 ssize_t rc; 907 u64 val; 908 909 if (hdev->reset_info.in_reset) { 910 dev_warn_ratelimited(hdev->dev, "Can't read during reset\n"); 911 return 0; 912 } 913 914 if (*ppos) 915 return 0; 916 917 rc = hl_access_mem(hdev, addr, &val, DEBUGFS_READ64); 918 if (rc) 919 return rc; 920 921 sprintf(tmp_buf, "0x%016llx\n", val); 922 return simple_read_from_buffer(buf, count, ppos, tmp_buf, 923 strlen(tmp_buf)); 924 } 925 926 static ssize_t hl_data_write64(struct file *f, const char __user *buf, 927 size_t count, loff_t *ppos) 928 { 929 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 930 struct hl_device *hdev = entry->hdev; 931 u64 addr = entry->addr; 932 u64 value; 933 ssize_t rc; 934 935 if (hdev->reset_info.in_reset) { 936 dev_warn_ratelimited(hdev->dev, "Can't write during reset\n"); 937 return 0; 938 } 939 940 rc = kstrtoull_from_user(buf, count, 16, &value); 941 if (rc) 942 return rc; 943 944 rc = hl_access_mem(hdev, addr, &value, DEBUGFS_WRITE64); 945 if (rc) 946 return rc; 947 948 return count; 949 } 950 951 static ssize_t hl_dma_size_write(struct file *f, const char __user *buf, 952 size_t count, loff_t *ppos) 953 { 954 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 955 struct hl_device *hdev = entry->hdev; 956 u64 addr = entry->addr; 957 ssize_t rc; 958 u32 size; 959 960 if (hdev->reset_info.in_reset) { 961 dev_warn_ratelimited(hdev->dev, "Can't DMA during reset\n"); 962 return 0; 963 } 964 rc = kstrtouint_from_user(buf, count, 16, &size); 965 if (rc) 966 return rc; 967 968 if (!size) { 969 dev_err(hdev->dev, "DMA read failed. size can't be 0\n"); 970 return -EINVAL; 971 } 972 973 if (size > SZ_128M) { 974 dev_err(hdev->dev, 975 "DMA read failed. size can't be larger than 128MB\n"); 976 return -EINVAL; 977 } 978 979 if (!hl_is_device_internal_memory_va(hdev, addr, size)) { 980 dev_err(hdev->dev, 981 "DMA read failed. Invalid 0x%010llx + 0x%08x\n", 982 addr, size); 983 return -EINVAL; 984 } 985 986 /* Free the previous allocation, if there was any */ 987 entry->data_dma_blob_desc.size = 0; 988 vfree(entry->data_dma_blob_desc.data); 989 990 entry->data_dma_blob_desc.data = vmalloc(size); 991 if (!entry->data_dma_blob_desc.data) 992 return -ENOMEM; 993 994 rc = hdev->asic_funcs->debugfs_read_dma(hdev, addr, size, 995 entry->data_dma_blob_desc.data); 996 if (rc) { 997 dev_err(hdev->dev, "Failed to DMA from 0x%010llx\n", addr); 998 vfree(entry->data_dma_blob_desc.data); 999 entry->data_dma_blob_desc.data = NULL; 1000 return -EIO; 1001 } 1002 1003 entry->data_dma_blob_desc.size = size; 1004 1005 return count; 1006 } 1007 1008 static ssize_t hl_monitor_dump_trigger(struct file *f, const char __user *buf, 1009 size_t count, loff_t *ppos) 1010 { 1011 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1012 struct hl_device *hdev = entry->hdev; 1013 u32 size, trig; 1014 ssize_t rc; 1015 1016 if (hdev->reset_info.in_reset) { 1017 dev_warn_ratelimited(hdev->dev, "Can't dump monitors during reset\n"); 1018 return 0; 1019 } 1020 rc = kstrtouint_from_user(buf, count, 10, &trig); 1021 if (rc) 1022 return rc; 1023 1024 if (trig != 1) { 1025 dev_err(hdev->dev, "Must write 1 to trigger monitor dump\n"); 1026 return -EINVAL; 1027 } 1028 1029 size = sizeof(struct cpucp_monitor_dump); 1030 1031 /* Free the previous allocation, if there was any */ 1032 entry->mon_dump_blob_desc.size = 0; 1033 vfree(entry->mon_dump_blob_desc.data); 1034 1035 entry->mon_dump_blob_desc.data = vmalloc(size); 1036 if (!entry->mon_dump_blob_desc.data) 1037 return -ENOMEM; 1038 1039 rc = hdev->asic_funcs->get_monitor_dump(hdev, entry->mon_dump_blob_desc.data); 1040 if (rc) { 1041 dev_err(hdev->dev, "Failed to dump monitors\n"); 1042 vfree(entry->mon_dump_blob_desc.data); 1043 entry->mon_dump_blob_desc.data = NULL; 1044 return -EIO; 1045 } 1046 1047 entry->mon_dump_blob_desc.size = size; 1048 1049 return count; 1050 } 1051 1052 static ssize_t hl_get_power_state(struct file *f, char __user *buf, 1053 size_t count, loff_t *ppos) 1054 { 1055 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1056 struct hl_device *hdev = entry->hdev; 1057 char tmp_buf[200]; 1058 int i; 1059 1060 if (*ppos) 1061 return 0; 1062 1063 if (hdev->pdev->current_state == PCI_D0) 1064 i = 1; 1065 else if (hdev->pdev->current_state == PCI_D3hot) 1066 i = 2; 1067 else 1068 i = 3; 1069 1070 sprintf(tmp_buf, 1071 "current power state: %d\n1 - D0\n2 - D3hot\n3 - Unknown\n", i); 1072 return simple_read_from_buffer(buf, count, ppos, tmp_buf, 1073 strlen(tmp_buf)); 1074 } 1075 1076 static ssize_t hl_set_power_state(struct file *f, const char __user *buf, 1077 size_t count, loff_t *ppos) 1078 { 1079 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1080 struct hl_device *hdev = entry->hdev; 1081 u32 value; 1082 ssize_t rc; 1083 1084 rc = kstrtouint_from_user(buf, count, 10, &value); 1085 if (rc) 1086 return rc; 1087 1088 if (value == 1) { 1089 pci_set_power_state(hdev->pdev, PCI_D0); 1090 pci_restore_state(hdev->pdev); 1091 rc = pci_enable_device(hdev->pdev); 1092 if (rc < 0) 1093 return rc; 1094 } else if (value == 2) { 1095 pci_save_state(hdev->pdev); 1096 pci_disable_device(hdev->pdev); 1097 pci_set_power_state(hdev->pdev, PCI_D3hot); 1098 } else { 1099 dev_dbg(hdev->dev, "invalid power state value %u\n", value); 1100 return -EINVAL; 1101 } 1102 1103 return count; 1104 } 1105 1106 static ssize_t hl_i2c_data_read(struct file *f, char __user *buf, 1107 size_t count, loff_t *ppos) 1108 { 1109 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1110 struct hl_device *hdev = entry->hdev; 1111 char tmp_buf[32]; 1112 u64 val; 1113 ssize_t rc; 1114 1115 if (*ppos) 1116 return 0; 1117 1118 rc = hl_debugfs_i2c_read(hdev, entry->i2c_bus, entry->i2c_addr, 1119 entry->i2c_reg, entry->i2c_len, &val); 1120 if (rc) { 1121 dev_err(hdev->dev, 1122 "Failed to read from I2C bus %d, addr %d, reg %d, len %d\n", 1123 entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len); 1124 return rc; 1125 } 1126 1127 sprintf(tmp_buf, "%#02llx\n", val); 1128 rc = simple_read_from_buffer(buf, count, ppos, tmp_buf, 1129 strlen(tmp_buf)); 1130 1131 return rc; 1132 } 1133 1134 static ssize_t hl_i2c_data_write(struct file *f, const char __user *buf, 1135 size_t count, loff_t *ppos) 1136 { 1137 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1138 struct hl_device *hdev = entry->hdev; 1139 u64 value; 1140 ssize_t rc; 1141 1142 rc = kstrtou64_from_user(buf, count, 16, &value); 1143 if (rc) 1144 return rc; 1145 1146 rc = hl_debugfs_i2c_write(hdev, entry->i2c_bus, entry->i2c_addr, 1147 entry->i2c_reg, entry->i2c_len, value); 1148 if (rc) { 1149 dev_err(hdev->dev, 1150 "Failed to write %#02llx to I2C bus %d, addr %d, reg %d, len %d\n", 1151 value, entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len); 1152 return rc; 1153 } 1154 1155 return count; 1156 } 1157 1158 static ssize_t hl_led0_write(struct file *f, const char __user *buf, 1159 size_t count, loff_t *ppos) 1160 { 1161 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1162 struct hl_device *hdev = entry->hdev; 1163 u32 value; 1164 ssize_t rc; 1165 1166 rc = kstrtouint_from_user(buf, count, 10, &value); 1167 if (rc) 1168 return rc; 1169 1170 value = value ? 1 : 0; 1171 1172 hl_debugfs_led_set(hdev, 0, value); 1173 1174 return count; 1175 } 1176 1177 static ssize_t hl_led1_write(struct file *f, const char __user *buf, 1178 size_t count, loff_t *ppos) 1179 { 1180 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1181 struct hl_device *hdev = entry->hdev; 1182 u32 value; 1183 ssize_t rc; 1184 1185 rc = kstrtouint_from_user(buf, count, 10, &value); 1186 if (rc) 1187 return rc; 1188 1189 value = value ? 1 : 0; 1190 1191 hl_debugfs_led_set(hdev, 1, value); 1192 1193 return count; 1194 } 1195 1196 static ssize_t hl_led2_write(struct file *f, const char __user *buf, 1197 size_t count, loff_t *ppos) 1198 { 1199 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1200 struct hl_device *hdev = entry->hdev; 1201 u32 value; 1202 ssize_t rc; 1203 1204 rc = kstrtouint_from_user(buf, count, 10, &value); 1205 if (rc) 1206 return rc; 1207 1208 value = value ? 1 : 0; 1209 1210 hl_debugfs_led_set(hdev, 2, value); 1211 1212 return count; 1213 } 1214 1215 static ssize_t hl_device_read(struct file *f, char __user *buf, 1216 size_t count, loff_t *ppos) 1217 { 1218 static const char *help = 1219 "Valid values: disable, enable, suspend, resume, cpu_timeout\n"; 1220 return simple_read_from_buffer(buf, count, ppos, help, strlen(help)); 1221 } 1222 1223 static ssize_t hl_device_write(struct file *f, const char __user *buf, 1224 size_t count, loff_t *ppos) 1225 { 1226 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1227 struct hl_device *hdev = entry->hdev; 1228 char data[30] = {0}; 1229 1230 /* don't allow partial writes */ 1231 if (*ppos != 0) 1232 return 0; 1233 1234 simple_write_to_buffer(data, 29, ppos, buf, count); 1235 1236 if (strncmp("disable", data, strlen("disable")) == 0) { 1237 hdev->disabled = true; 1238 } else if (strncmp("enable", data, strlen("enable")) == 0) { 1239 hdev->disabled = false; 1240 } else if (strncmp("suspend", data, strlen("suspend")) == 0) { 1241 hdev->asic_funcs->suspend(hdev); 1242 } else if (strncmp("resume", data, strlen("resume")) == 0) { 1243 hdev->asic_funcs->resume(hdev); 1244 } else if (strncmp("cpu_timeout", data, strlen("cpu_timeout")) == 0) { 1245 hdev->device_cpu_disabled = true; 1246 } else { 1247 dev_err(hdev->dev, 1248 "Valid values: disable, enable, suspend, resume, cpu_timeout\n"); 1249 count = -EINVAL; 1250 } 1251 1252 return count; 1253 } 1254 1255 static ssize_t hl_clk_gate_read(struct file *f, char __user *buf, 1256 size_t count, loff_t *ppos) 1257 { 1258 return 0; 1259 } 1260 1261 static ssize_t hl_clk_gate_write(struct file *f, const char __user *buf, 1262 size_t count, loff_t *ppos) 1263 { 1264 return count; 1265 } 1266 1267 static ssize_t hl_stop_on_err_read(struct file *f, char __user *buf, 1268 size_t count, loff_t *ppos) 1269 { 1270 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1271 struct hl_device *hdev = entry->hdev; 1272 char tmp_buf[200]; 1273 ssize_t rc; 1274 1275 if (!hdev->asic_prop.configurable_stop_on_err) 1276 return -EOPNOTSUPP; 1277 1278 if (*ppos) 1279 return 0; 1280 1281 sprintf(tmp_buf, "%d\n", hdev->stop_on_err); 1282 rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf, 1283 strlen(tmp_buf) + 1); 1284 1285 return rc; 1286 } 1287 1288 static ssize_t hl_stop_on_err_write(struct file *f, const char __user *buf, 1289 size_t count, loff_t *ppos) 1290 { 1291 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1292 struct hl_device *hdev = entry->hdev; 1293 u32 value; 1294 ssize_t rc; 1295 1296 if (!hdev->asic_prop.configurable_stop_on_err) 1297 return -EOPNOTSUPP; 1298 1299 if (hdev->reset_info.in_reset) { 1300 dev_warn_ratelimited(hdev->dev, 1301 "Can't change stop on error during reset\n"); 1302 return 0; 1303 } 1304 1305 rc = kstrtouint_from_user(buf, count, 10, &value); 1306 if (rc) 1307 return rc; 1308 1309 hdev->stop_on_err = value ? 1 : 0; 1310 1311 hl_device_reset(hdev, 0); 1312 1313 return count; 1314 } 1315 1316 static ssize_t hl_security_violations_read(struct file *f, char __user *buf, 1317 size_t count, loff_t *ppos) 1318 { 1319 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1320 struct hl_device *hdev = entry->hdev; 1321 1322 hdev->asic_funcs->ack_protection_bits_errors(hdev); 1323 1324 return 0; 1325 } 1326 1327 static ssize_t hl_state_dump_read(struct file *f, char __user *buf, 1328 size_t count, loff_t *ppos) 1329 { 1330 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1331 ssize_t rc; 1332 1333 down_read(&entry->state_dump_sem); 1334 if (!entry->state_dump[entry->state_dump_head]) 1335 rc = 0; 1336 else 1337 rc = simple_read_from_buffer( 1338 buf, count, ppos, 1339 entry->state_dump[entry->state_dump_head], 1340 strlen(entry->state_dump[entry->state_dump_head])); 1341 up_read(&entry->state_dump_sem); 1342 1343 return rc; 1344 } 1345 1346 static ssize_t hl_state_dump_write(struct file *f, const char __user *buf, 1347 size_t count, loff_t *ppos) 1348 { 1349 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1350 struct hl_device *hdev = entry->hdev; 1351 ssize_t rc; 1352 u32 size; 1353 int i; 1354 1355 rc = kstrtouint_from_user(buf, count, 10, &size); 1356 if (rc) 1357 return rc; 1358 1359 if (size <= 0 || size >= ARRAY_SIZE(entry->state_dump)) { 1360 dev_err(hdev->dev, "Invalid number of dumps to skip\n"); 1361 return -EINVAL; 1362 } 1363 1364 if (entry->state_dump[entry->state_dump_head]) { 1365 down_write(&entry->state_dump_sem); 1366 for (i = 0; i < size; ++i) { 1367 vfree(entry->state_dump[entry->state_dump_head]); 1368 entry->state_dump[entry->state_dump_head] = NULL; 1369 if (entry->state_dump_head > 0) 1370 entry->state_dump_head--; 1371 else 1372 entry->state_dump_head = 1373 ARRAY_SIZE(entry->state_dump) - 1; 1374 } 1375 up_write(&entry->state_dump_sem); 1376 } 1377 1378 return count; 1379 } 1380 1381 static ssize_t hl_timeout_locked_read(struct file *f, char __user *buf, 1382 size_t count, loff_t *ppos) 1383 { 1384 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1385 struct hl_device *hdev = entry->hdev; 1386 char tmp_buf[200]; 1387 ssize_t rc; 1388 1389 if (*ppos) 1390 return 0; 1391 1392 sprintf(tmp_buf, "%d\n", 1393 jiffies_to_msecs(hdev->timeout_jiffies) / 1000); 1394 rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf, 1395 strlen(tmp_buf) + 1); 1396 1397 return rc; 1398 } 1399 1400 static ssize_t hl_timeout_locked_write(struct file *f, const char __user *buf, 1401 size_t count, loff_t *ppos) 1402 { 1403 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1404 struct hl_device *hdev = entry->hdev; 1405 u32 value; 1406 ssize_t rc; 1407 1408 rc = kstrtouint_from_user(buf, count, 10, &value); 1409 if (rc) 1410 return rc; 1411 1412 if (value) 1413 hdev->timeout_jiffies = msecs_to_jiffies(value * 1000); 1414 else 1415 hdev->timeout_jiffies = MAX_SCHEDULE_TIMEOUT; 1416 1417 return count; 1418 } 1419 1420 static ssize_t hl_check_razwi_happened(struct file *f, char __user *buf, 1421 size_t count, loff_t *ppos) 1422 { 1423 struct hl_dbg_device_entry *entry = file_inode(f)->i_private; 1424 struct hl_device *hdev = entry->hdev; 1425 1426 hdev->asic_funcs->check_if_razwi_happened(hdev); 1427 1428 return 0; 1429 } 1430 1431 static const struct file_operations hl_mem_scrub_fops = { 1432 .owner = THIS_MODULE, 1433 .write = hl_memory_scrub, 1434 }; 1435 1436 static const struct file_operations hl_data32b_fops = { 1437 .owner = THIS_MODULE, 1438 .read = hl_data_read32, 1439 .write = hl_data_write32 1440 }; 1441 1442 static const struct file_operations hl_data64b_fops = { 1443 .owner = THIS_MODULE, 1444 .read = hl_data_read64, 1445 .write = hl_data_write64 1446 }; 1447 1448 static const struct file_operations hl_dma_size_fops = { 1449 .owner = THIS_MODULE, 1450 .write = hl_dma_size_write 1451 }; 1452 1453 static const struct file_operations hl_monitor_dump_fops = { 1454 .owner = THIS_MODULE, 1455 .write = hl_monitor_dump_trigger 1456 }; 1457 1458 static const struct file_operations hl_i2c_data_fops = { 1459 .owner = THIS_MODULE, 1460 .read = hl_i2c_data_read, 1461 .write = hl_i2c_data_write 1462 }; 1463 1464 static const struct file_operations hl_power_fops = { 1465 .owner = THIS_MODULE, 1466 .read = hl_get_power_state, 1467 .write = hl_set_power_state 1468 }; 1469 1470 static const struct file_operations hl_led0_fops = { 1471 .owner = THIS_MODULE, 1472 .write = hl_led0_write 1473 }; 1474 1475 static const struct file_operations hl_led1_fops = { 1476 .owner = THIS_MODULE, 1477 .write = hl_led1_write 1478 }; 1479 1480 static const struct file_operations hl_led2_fops = { 1481 .owner = THIS_MODULE, 1482 .write = hl_led2_write 1483 }; 1484 1485 static const struct file_operations hl_device_fops = { 1486 .owner = THIS_MODULE, 1487 .read = hl_device_read, 1488 .write = hl_device_write 1489 }; 1490 1491 static const struct file_operations hl_clk_gate_fops = { 1492 .owner = THIS_MODULE, 1493 .read = hl_clk_gate_read, 1494 .write = hl_clk_gate_write 1495 }; 1496 1497 static const struct file_operations hl_stop_on_err_fops = { 1498 .owner = THIS_MODULE, 1499 .read = hl_stop_on_err_read, 1500 .write = hl_stop_on_err_write 1501 }; 1502 1503 static const struct file_operations hl_security_violations_fops = { 1504 .owner = THIS_MODULE, 1505 .read = hl_security_violations_read 1506 }; 1507 1508 static const struct file_operations hl_state_dump_fops = { 1509 .owner = THIS_MODULE, 1510 .read = hl_state_dump_read, 1511 .write = hl_state_dump_write 1512 }; 1513 1514 static const struct file_operations hl_timeout_locked_fops = { 1515 .owner = THIS_MODULE, 1516 .read = hl_timeout_locked_read, 1517 .write = hl_timeout_locked_write 1518 }; 1519 1520 static const struct file_operations hl_razwi_check_fops = { 1521 .owner = THIS_MODULE, 1522 .read = hl_check_razwi_happened 1523 }; 1524 1525 static const struct hl_info_list hl_debugfs_list[] = { 1526 {"command_buffers", command_buffers_show, NULL}, 1527 {"command_submission", command_submission_show, NULL}, 1528 {"command_submission_jobs", command_submission_jobs_show, NULL}, 1529 {"userptr", userptr_show, NULL}, 1530 {"vm", vm_show, NULL}, 1531 {"userptr_lookup", userptr_lookup_show, userptr_lookup_write}, 1532 {"mmu", mmu_show, mmu_asid_va_write}, 1533 {"mmu_error", mmu_ack_error, mmu_ack_error_value_write}, 1534 {"engines", engines_show, NULL}, 1535 }; 1536 1537 static int hl_debugfs_open(struct inode *inode, struct file *file) 1538 { 1539 struct hl_debugfs_entry *node = inode->i_private; 1540 1541 return single_open(file, node->info_ent->show, node); 1542 } 1543 1544 static ssize_t hl_debugfs_write(struct file *file, const char __user *buf, 1545 size_t count, loff_t *f_pos) 1546 { 1547 struct hl_debugfs_entry *node = file->f_inode->i_private; 1548 1549 if (node->info_ent->write) 1550 return node->info_ent->write(file, buf, count, f_pos); 1551 else 1552 return -EINVAL; 1553 1554 } 1555 1556 static const struct file_operations hl_debugfs_fops = { 1557 .owner = THIS_MODULE, 1558 .open = hl_debugfs_open, 1559 .read = seq_read, 1560 .write = hl_debugfs_write, 1561 .llseek = seq_lseek, 1562 .release = single_release, 1563 }; 1564 1565 static void add_secured_nodes(struct hl_dbg_device_entry *dev_entry, struct dentry *root) 1566 { 1567 debugfs_create_u8("i2c_bus", 1568 0644, 1569 root, 1570 &dev_entry->i2c_bus); 1571 1572 debugfs_create_u8("i2c_addr", 1573 0644, 1574 root, 1575 &dev_entry->i2c_addr); 1576 1577 debugfs_create_u8("i2c_reg", 1578 0644, 1579 root, 1580 &dev_entry->i2c_reg); 1581 1582 debugfs_create_u8("i2c_len", 1583 0644, 1584 root, 1585 &dev_entry->i2c_len); 1586 1587 debugfs_create_file("i2c_data", 1588 0644, 1589 root, 1590 dev_entry, 1591 &hl_i2c_data_fops); 1592 1593 debugfs_create_file("led0", 1594 0200, 1595 root, 1596 dev_entry, 1597 &hl_led0_fops); 1598 1599 debugfs_create_file("led1", 1600 0200, 1601 root, 1602 dev_entry, 1603 &hl_led1_fops); 1604 1605 debugfs_create_file("led2", 1606 0200, 1607 root, 1608 dev_entry, 1609 &hl_led2_fops); 1610 } 1611 1612 static void add_files_to_device(struct hl_device *hdev, struct hl_dbg_device_entry *dev_entry, 1613 struct dentry *root) 1614 { 1615 int count = ARRAY_SIZE(hl_debugfs_list); 1616 struct hl_debugfs_entry *entry; 1617 int i; 1618 1619 debugfs_create_x64("memory_scrub_val", 1620 0644, 1621 root, 1622 &hdev->memory_scrub_val); 1623 1624 debugfs_create_file("memory_scrub", 1625 0200, 1626 root, 1627 dev_entry, 1628 &hl_mem_scrub_fops); 1629 1630 debugfs_create_x64("addr", 1631 0644, 1632 root, 1633 &dev_entry->addr); 1634 1635 debugfs_create_file("data32", 1636 0644, 1637 root, 1638 dev_entry, 1639 &hl_data32b_fops); 1640 1641 debugfs_create_file("data64", 1642 0644, 1643 root, 1644 dev_entry, 1645 &hl_data64b_fops); 1646 1647 debugfs_create_file("set_power_state", 1648 0200, 1649 root, 1650 dev_entry, 1651 &hl_power_fops); 1652 1653 debugfs_create_file("device", 1654 0200, 1655 root, 1656 dev_entry, 1657 &hl_device_fops); 1658 1659 debugfs_create_file("clk_gate", 1660 0200, 1661 root, 1662 dev_entry, 1663 &hl_clk_gate_fops); 1664 1665 debugfs_create_file("stop_on_err", 1666 0644, 1667 root, 1668 dev_entry, 1669 &hl_stop_on_err_fops); 1670 1671 debugfs_create_file("dump_security_violations", 1672 0644, 1673 root, 1674 dev_entry, 1675 &hl_security_violations_fops); 1676 1677 debugfs_create_file("dump_razwi_events", 1678 0644, 1679 root, 1680 dev_entry, 1681 &hl_razwi_check_fops); 1682 1683 debugfs_create_file("dma_size", 1684 0200, 1685 root, 1686 dev_entry, 1687 &hl_dma_size_fops); 1688 1689 debugfs_create_blob("data_dma", 1690 0400, 1691 root, 1692 &dev_entry->data_dma_blob_desc); 1693 1694 debugfs_create_file("monitor_dump_trig", 1695 0200, 1696 root, 1697 dev_entry, 1698 &hl_monitor_dump_fops); 1699 1700 debugfs_create_blob("monitor_dump", 1701 0400, 1702 root, 1703 &dev_entry->mon_dump_blob_desc); 1704 1705 debugfs_create_x8("skip_reset_on_timeout", 1706 0644, 1707 root, 1708 &hdev->reset_info.skip_reset_on_timeout); 1709 1710 debugfs_create_file("state_dump", 1711 0600, 1712 root, 1713 dev_entry, 1714 &hl_state_dump_fops); 1715 1716 debugfs_create_file("timeout_locked", 1717 0644, 1718 root, 1719 dev_entry, 1720 &hl_timeout_locked_fops); 1721 1722 debugfs_create_u32("device_release_watchdog_timeout", 1723 0644, 1724 root, 1725 &hdev->device_release_watchdog_timeout_sec); 1726 1727 for (i = 0, entry = dev_entry->entry_arr ; i < count ; i++, entry++) { 1728 debugfs_create_file(hl_debugfs_list[i].name, 1729 0444, 1730 root, 1731 entry, 1732 &hl_debugfs_fops); 1733 entry->info_ent = &hl_debugfs_list[i]; 1734 entry->dev_entry = dev_entry; 1735 } 1736 } 1737 1738 int hl_debugfs_device_init(struct hl_device *hdev) 1739 { 1740 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1741 int count = ARRAY_SIZE(hl_debugfs_list); 1742 1743 dev_entry->hdev = hdev; 1744 dev_entry->entry_arr = kmalloc_array(count, sizeof(struct hl_debugfs_entry), GFP_KERNEL); 1745 if (!dev_entry->entry_arr) 1746 return -ENOMEM; 1747 1748 dev_entry->data_dma_blob_desc.size = 0; 1749 dev_entry->data_dma_blob_desc.data = NULL; 1750 dev_entry->mon_dump_blob_desc.size = 0; 1751 dev_entry->mon_dump_blob_desc.data = NULL; 1752 1753 INIT_LIST_HEAD(&dev_entry->file_list); 1754 INIT_LIST_HEAD(&dev_entry->cb_list); 1755 INIT_LIST_HEAD(&dev_entry->cs_list); 1756 INIT_LIST_HEAD(&dev_entry->cs_job_list); 1757 INIT_LIST_HEAD(&dev_entry->userptr_list); 1758 INIT_LIST_HEAD(&dev_entry->ctx_mem_hash_list); 1759 mutex_init(&dev_entry->file_mutex); 1760 init_rwsem(&dev_entry->state_dump_sem); 1761 spin_lock_init(&dev_entry->cb_spinlock); 1762 spin_lock_init(&dev_entry->cs_spinlock); 1763 spin_lock_init(&dev_entry->cs_job_spinlock); 1764 spin_lock_init(&dev_entry->userptr_spinlock); 1765 mutex_init(&dev_entry->ctx_mem_hash_mutex); 1766 1767 return 0; 1768 } 1769 1770 void hl_debugfs_device_fini(struct hl_device *hdev) 1771 { 1772 struct hl_dbg_device_entry *entry = &hdev->hl_debugfs; 1773 int i; 1774 1775 mutex_destroy(&entry->ctx_mem_hash_mutex); 1776 mutex_destroy(&entry->file_mutex); 1777 1778 vfree(entry->data_dma_blob_desc.data); 1779 vfree(entry->mon_dump_blob_desc.data); 1780 1781 for (i = 0; i < ARRAY_SIZE(entry->state_dump); ++i) 1782 vfree(entry->state_dump[i]); 1783 1784 kfree(entry->entry_arr); 1785 } 1786 1787 void hl_debugfs_add_device(struct hl_device *hdev) 1788 { 1789 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1790 1791 dev_entry->root = debugfs_create_dir(dev_name(hdev->dev), hl_debug_root); 1792 1793 add_files_to_device(hdev, dev_entry, dev_entry->root); 1794 if (!hdev->asic_prop.fw_security_enabled) 1795 add_secured_nodes(dev_entry, dev_entry->root); 1796 } 1797 1798 void hl_debugfs_remove_device(struct hl_device *hdev) 1799 { 1800 struct hl_dbg_device_entry *entry = &hdev->hl_debugfs; 1801 1802 debugfs_remove_recursive(entry->root); 1803 } 1804 1805 void hl_debugfs_add_file(struct hl_fpriv *hpriv) 1806 { 1807 struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs; 1808 1809 mutex_lock(&dev_entry->file_mutex); 1810 list_add(&hpriv->debugfs_list, &dev_entry->file_list); 1811 mutex_unlock(&dev_entry->file_mutex); 1812 } 1813 1814 void hl_debugfs_remove_file(struct hl_fpriv *hpriv) 1815 { 1816 struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs; 1817 1818 mutex_lock(&dev_entry->file_mutex); 1819 list_del(&hpriv->debugfs_list); 1820 mutex_unlock(&dev_entry->file_mutex); 1821 } 1822 1823 void hl_debugfs_add_cb(struct hl_cb *cb) 1824 { 1825 struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs; 1826 1827 spin_lock(&dev_entry->cb_spinlock); 1828 list_add(&cb->debugfs_list, &dev_entry->cb_list); 1829 spin_unlock(&dev_entry->cb_spinlock); 1830 } 1831 1832 void hl_debugfs_remove_cb(struct hl_cb *cb) 1833 { 1834 struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs; 1835 1836 spin_lock(&dev_entry->cb_spinlock); 1837 list_del(&cb->debugfs_list); 1838 spin_unlock(&dev_entry->cb_spinlock); 1839 } 1840 1841 void hl_debugfs_add_cs(struct hl_cs *cs) 1842 { 1843 struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs; 1844 1845 spin_lock(&dev_entry->cs_spinlock); 1846 list_add(&cs->debugfs_list, &dev_entry->cs_list); 1847 spin_unlock(&dev_entry->cs_spinlock); 1848 } 1849 1850 void hl_debugfs_remove_cs(struct hl_cs *cs) 1851 { 1852 struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs; 1853 1854 spin_lock(&dev_entry->cs_spinlock); 1855 list_del(&cs->debugfs_list); 1856 spin_unlock(&dev_entry->cs_spinlock); 1857 } 1858 1859 void hl_debugfs_add_job(struct hl_device *hdev, struct hl_cs_job *job) 1860 { 1861 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1862 1863 spin_lock(&dev_entry->cs_job_spinlock); 1864 list_add(&job->debugfs_list, &dev_entry->cs_job_list); 1865 spin_unlock(&dev_entry->cs_job_spinlock); 1866 } 1867 1868 void hl_debugfs_remove_job(struct hl_device *hdev, struct hl_cs_job *job) 1869 { 1870 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1871 1872 spin_lock(&dev_entry->cs_job_spinlock); 1873 list_del(&job->debugfs_list); 1874 spin_unlock(&dev_entry->cs_job_spinlock); 1875 } 1876 1877 void hl_debugfs_add_userptr(struct hl_device *hdev, struct hl_userptr *userptr) 1878 { 1879 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1880 1881 spin_lock(&dev_entry->userptr_spinlock); 1882 list_add(&userptr->debugfs_list, &dev_entry->userptr_list); 1883 spin_unlock(&dev_entry->userptr_spinlock); 1884 } 1885 1886 void hl_debugfs_remove_userptr(struct hl_device *hdev, 1887 struct hl_userptr *userptr) 1888 { 1889 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1890 1891 spin_lock(&dev_entry->userptr_spinlock); 1892 list_del(&userptr->debugfs_list); 1893 spin_unlock(&dev_entry->userptr_spinlock); 1894 } 1895 1896 void hl_debugfs_add_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx) 1897 { 1898 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1899 1900 mutex_lock(&dev_entry->ctx_mem_hash_mutex); 1901 list_add(&ctx->debugfs_list, &dev_entry->ctx_mem_hash_list); 1902 mutex_unlock(&dev_entry->ctx_mem_hash_mutex); 1903 } 1904 1905 void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx) 1906 { 1907 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1908 1909 mutex_lock(&dev_entry->ctx_mem_hash_mutex); 1910 list_del(&ctx->debugfs_list); 1911 mutex_unlock(&dev_entry->ctx_mem_hash_mutex); 1912 } 1913 1914 /** 1915 * hl_debugfs_set_state_dump - register state dump making it accessible via 1916 * debugfs 1917 * @hdev: pointer to the device structure 1918 * @data: the actual dump data 1919 * @length: the length of the data 1920 */ 1921 void hl_debugfs_set_state_dump(struct hl_device *hdev, char *data, 1922 unsigned long length) 1923 { 1924 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs; 1925 1926 down_write(&dev_entry->state_dump_sem); 1927 1928 dev_entry->state_dump_head = (dev_entry->state_dump_head + 1) % 1929 ARRAY_SIZE(dev_entry->state_dump); 1930 vfree(dev_entry->state_dump[dev_entry->state_dump_head]); 1931 dev_entry->state_dump[dev_entry->state_dump_head] = data; 1932 1933 up_write(&dev_entry->state_dump_sem); 1934 } 1935 1936 void __init hl_debugfs_init(void) 1937 { 1938 hl_debug_root = debugfs_create_dir("habanalabs", NULL); 1939 } 1940 1941 void hl_debugfs_fini(void) 1942 { 1943 debugfs_remove_recursive(hl_debug_root); 1944 } 1945