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
hl_debugfs_i2c_read(struct hl_device * hdev,u8 i2c_bus,u8 i2c_addr,u8 i2c_reg,u8 i2c_len,u64 * val)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
hl_debugfs_i2c_write(struct hl_device * hdev,u8 i2c_bus,u8 i2c_addr,u8 i2c_reg,u8 i2c_len,u64 val)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
hl_debugfs_led_set(struct hl_device * hdev,u8 led,u8 state)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
command_buffers_show(struct seq_file * s,void * data)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
command_submission_show(struct seq_file * s,void * data)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 CTX ASID CS RefCnt Submitted Completed\n");
156 seq_puts(s, "------------------------------------------------------\n");
157 }
158 seq_printf(s,
159 " %llu %d %d %d %d\n",
160 cs->sequence, 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
command_submission_jobs_show(struct seq_file * s,void * data)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 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\n",
192 job->id, job->cs->sequence, job->cs->ctx->asid,
193 kref_read(&job->refcount), job->hw_queue_id);
194 else
195 seq_printf(s,
196 " %02d 0 %d %d %d\n",
197 job->id, HL_KERNEL_ASID_ID,
198 kref_read(&job->refcount), job->hw_queue_id);
199 }
200
201 spin_unlock(&dev_entry->cs_job_spinlock);
202
203 if (!first)
204 seq_puts(s, "\n");
205
206 return 0;
207 }
208
userptr_show(struct seq_file * s,void * data)209 static int userptr_show(struct seq_file *s, void *data)
210 {
211 struct hl_debugfs_entry *entry = s->private;
212 struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
213 struct hl_userptr *userptr;
214 char dma_dir[4][30] = {"DMA_BIDIRECTIONAL", "DMA_TO_DEVICE",
215 "DMA_FROM_DEVICE", "DMA_NONE"};
216 bool first = true;
217
218 spin_lock(&dev_entry->userptr_spinlock);
219
220 list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) {
221 if (first) {
222 first = false;
223 seq_puts(s, "\n");
224 seq_puts(s, " pid user virtual address size dma dir\n");
225 seq_puts(s, "----------------------------------------------------------\n");
226 }
227 seq_printf(s, " %-7d 0x%-14llx %-10llu %-30s\n",
228 userptr->pid, userptr->addr, userptr->size,
229 dma_dir[userptr->dir]);
230 }
231
232 spin_unlock(&dev_entry->userptr_spinlock);
233
234 if (!first)
235 seq_puts(s, "\n");
236
237 return 0;
238 }
239
vm_show(struct seq_file * s,void * data)240 static int vm_show(struct seq_file *s, void *data)
241 {
242 struct hl_debugfs_entry *entry = s->private;
243 struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
244 struct hl_vm_hw_block_list_node *lnode;
245 struct hl_ctx *ctx;
246 struct hl_vm *vm;
247 struct hl_vm_hash_node *hnode;
248 struct hl_userptr *userptr;
249 struct hl_vm_phys_pg_pack *phys_pg_pack = NULL;
250 struct hl_va_range *va_range;
251 struct hl_vm_va_block *va_block;
252 enum vm_type *vm_type;
253 bool once = true;
254 u64 j;
255 int i;
256
257 if (!dev_entry->hdev->mmu_enable)
258 return 0;
259
260 spin_lock(&dev_entry->ctx_mem_hash_spinlock);
261
262 list_for_each_entry(ctx, &dev_entry->ctx_mem_hash_list, debugfs_list) {
263 once = false;
264 seq_puts(s, "\n\n----------------------------------------------------");
265 seq_puts(s, "\n----------------------------------------------------\n\n");
266 seq_printf(s, "ctx asid: %u\n", ctx->asid);
267
268 seq_puts(s, "\nmappings:\n\n");
269 seq_puts(s, " virtual address size handle\n");
270 seq_puts(s, "----------------------------------------------------\n");
271 mutex_lock(&ctx->mem_hash_lock);
272 hash_for_each(ctx->mem_hash, i, hnode, node) {
273 vm_type = hnode->ptr;
274
275 if (*vm_type == VM_TYPE_USERPTR) {
276 userptr = hnode->ptr;
277 seq_printf(s,
278 " 0x%-14llx %-10llu\n",
279 hnode->vaddr, userptr->size);
280 } else {
281 phys_pg_pack = hnode->ptr;
282 seq_printf(s,
283 " 0x%-14llx %-10llu %-4u\n",
284 hnode->vaddr, phys_pg_pack->total_size,
285 phys_pg_pack->handle);
286 }
287 }
288 mutex_unlock(&ctx->mem_hash_lock);
289
290 if (ctx->asid != HL_KERNEL_ASID_ID &&
291 !list_empty(&ctx->hw_block_mem_list)) {
292 seq_puts(s, "\nhw_block mappings:\n\n");
293 seq_puts(s, " virtual address size HW block id\n");
294 seq_puts(s, "-------------------------------------------\n");
295 mutex_lock(&ctx->hw_block_list_lock);
296 list_for_each_entry(lnode, &ctx->hw_block_mem_list,
297 node) {
298 seq_printf(s,
299 " 0x%-14lx %-6u %-9u\n",
300 lnode->vaddr, lnode->size, 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 spin_unlock(&dev_entry->ctx_mem_hash_spinlock);
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
userptr_lookup_show(struct seq_file * s,void * data)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
userptr_lookup_write(struct file * file,const char __user * buf,size_t count,loff_t * f_pos)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
mmu_show(struct seq_file * s,void * data)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 (!hdev->mmu_enable)
437 return 0;
438
439 if (dev_entry->mmu_asid == HL_KERNEL_ASID_ID)
440 ctx = hdev->kernel_ctx;
441 else
442 ctx = hl_get_compute_ctx(hdev);
443
444 if (!ctx) {
445 dev_err(hdev->dev, "no ctx available\n");
446 return 0;
447 }
448
449 if (hl_mmu_get_tlb_info(ctx, virt_addr, &hops_info)) {
450 dev_err(hdev->dev, "virt addr 0x%llx is not mapped to phys addr\n",
451 virt_addr);
452 return 0;
453 }
454
455 hl_mmu_va_to_pa(ctx, virt_addr, &phys_addr);
456
457 if (hops_info.scrambled_vaddr &&
458 (dev_entry->mmu_addr != hops_info.scrambled_vaddr))
459 seq_printf(s,
460 "asid: %u, virt_addr: 0x%llx, scrambled virt_addr: 0x%llx,\nphys_addr: 0x%llx, scrambled_phys_addr: 0x%llx\n",
461 dev_entry->mmu_asid, dev_entry->mmu_addr,
462 hops_info.scrambled_vaddr,
463 hops_info.unscrambled_paddr, phys_addr);
464 else
465 seq_printf(s,
466 "asid: %u, virt_addr: 0x%llx, phys_addr: 0x%llx\n",
467 dev_entry->mmu_asid, dev_entry->mmu_addr, phys_addr);
468
469 for (i = 0 ; i < hops_info.used_hops ; i++) {
470 seq_printf(s, "hop%d_addr: 0x%llx\n",
471 i, hops_info.hop_info[i].hop_addr);
472 seq_printf(s, "hop%d_pte_addr: 0x%llx\n",
473 i, hops_info.hop_info[i].hop_pte_addr);
474 seq_printf(s, "hop%d_pte: 0x%llx\n",
475 i, hops_info.hop_info[i].hop_pte_val);
476 }
477
478 return 0;
479 }
480
mmu_asid_va_write(struct file * file,const char __user * buf,size_t count,loff_t * f_pos)481 static ssize_t mmu_asid_va_write(struct file *file, const char __user *buf,
482 size_t count, loff_t *f_pos)
483 {
484 struct seq_file *s = file->private_data;
485 struct hl_debugfs_entry *entry = s->private;
486 struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
487 struct hl_device *hdev = dev_entry->hdev;
488 char kbuf[MMU_KBUF_SIZE];
489 char *c;
490 ssize_t rc;
491
492 if (!hdev->mmu_enable)
493 return count;
494
495 if (count > sizeof(kbuf) - 1)
496 goto err;
497 if (copy_from_user(kbuf, buf, count))
498 goto err;
499 kbuf[count] = 0;
500
501 c = strchr(kbuf, ' ');
502 if (!c)
503 goto err;
504 *c = '\0';
505
506 rc = kstrtouint(kbuf, 10, &dev_entry->mmu_asid);
507 if (rc)
508 goto err;
509
510 if (strncmp(c+1, "0x", 2))
511 goto err;
512 rc = kstrtoull(c+3, 16, &dev_entry->mmu_addr);
513 if (rc)
514 goto err;
515
516 return count;
517
518 err:
519 dev_err(hdev->dev, "usage: echo <asid> <0xaddr> > mmu\n");
520
521 return -EINVAL;
522 }
523
engines_show(struct seq_file * s,void * data)524 static int engines_show(struct seq_file *s, void *data)
525 {
526 struct hl_debugfs_entry *entry = s->private;
527 struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
528 struct hl_device *hdev = dev_entry->hdev;
529
530 if (hdev->reset_info.in_reset) {
531 dev_warn_ratelimited(hdev->dev,
532 "Can't check device idle during reset\n");
533 return 0;
534 }
535
536 hdev->asic_funcs->is_device_idle(hdev, NULL, 0, s);
537
538 return 0;
539 }
540
hl_memory_scrub(struct file * f,const char __user * buf,size_t count,loff_t * ppos)541 static ssize_t hl_memory_scrub(struct file *f, const char __user *buf,
542 size_t count, loff_t *ppos)
543 {
544 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
545 struct hl_device *hdev = entry->hdev;
546 u64 val = entry->memory_scrub_val;
547 int rc;
548
549 if (!hl_device_operational(hdev, NULL)) {
550 dev_warn_ratelimited(hdev->dev, "Can't scrub memory, device is not operational\n");
551 return -EIO;
552 }
553
554 mutex_lock(&hdev->fpriv_list_lock);
555 if (hdev->is_compute_ctx_active) {
556 mutex_unlock(&hdev->fpriv_list_lock);
557 dev_err(hdev->dev, "can't scrub dram, context exist\n");
558 return -EBUSY;
559 }
560 hdev->is_in_dram_scrub = true;
561 mutex_unlock(&hdev->fpriv_list_lock);
562
563 rc = hdev->asic_funcs->scrub_device_dram(hdev, val);
564
565 mutex_lock(&hdev->fpriv_list_lock);
566 hdev->is_in_dram_scrub = false;
567 mutex_unlock(&hdev->fpriv_list_lock);
568
569 if (rc)
570 return rc;
571 return count;
572 }
573
hl_is_device_va(struct hl_device * hdev,u64 addr)574 static bool hl_is_device_va(struct hl_device *hdev, u64 addr)
575 {
576 struct asic_fixed_properties *prop = &hdev->asic_prop;
577
578 if (!hdev->mmu_enable)
579 goto out;
580
581 if (prop->dram_supports_virtual_memory &&
582 (addr >= prop->dmmu.start_addr && addr < prop->dmmu.end_addr))
583 return true;
584
585 if (addr >= prop->pmmu.start_addr &&
586 addr < prop->pmmu.end_addr)
587 return true;
588
589 if (addr >= prop->pmmu_huge.start_addr &&
590 addr < prop->pmmu_huge.end_addr)
591 return true;
592 out:
593 return false;
594 }
595
hl_is_device_internal_memory_va(struct hl_device * hdev,u64 addr,u32 size)596 static bool hl_is_device_internal_memory_va(struct hl_device *hdev, u64 addr,
597 u32 size)
598 {
599 struct asic_fixed_properties *prop = &hdev->asic_prop;
600 u64 dram_start_addr, dram_end_addr;
601
602 if (!hdev->mmu_enable)
603 return false;
604
605 if (prop->dram_supports_virtual_memory) {
606 dram_start_addr = prop->dmmu.start_addr;
607 dram_end_addr = prop->dmmu.end_addr;
608 } else {
609 dram_start_addr = prop->dram_base_address;
610 dram_end_addr = prop->dram_end_address;
611 }
612
613 if (hl_mem_area_inside_range(addr, size, dram_start_addr,
614 dram_end_addr))
615 return true;
616
617 if (hl_mem_area_inside_range(addr, size, prop->sram_base_address,
618 prop->sram_end_address))
619 return true;
620
621 return false;
622 }
623
device_va_to_pa(struct hl_device * hdev,u64 virt_addr,u32 size,u64 * phys_addr)624 static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size,
625 u64 *phys_addr)
626 {
627 struct hl_vm_phys_pg_pack *phys_pg_pack;
628 struct hl_ctx *ctx;
629 struct hl_vm_hash_node *hnode;
630 u64 end_address, range_size;
631 struct hl_userptr *userptr;
632 enum vm_type *vm_type;
633 bool valid = false;
634 int i, rc = 0;
635
636 ctx = hl_get_compute_ctx(hdev);
637
638 if (!ctx) {
639 dev_err(hdev->dev, "no ctx available\n");
640 return -EINVAL;
641 }
642
643 /* Verify address is mapped */
644 mutex_lock(&ctx->mem_hash_lock);
645 hash_for_each(ctx->mem_hash, i, hnode, node) {
646 vm_type = hnode->ptr;
647
648 if (*vm_type == VM_TYPE_USERPTR) {
649 userptr = hnode->ptr;
650 range_size = userptr->size;
651 } else {
652 phys_pg_pack = hnode->ptr;
653 range_size = phys_pg_pack->total_size;
654 }
655
656 end_address = virt_addr + size;
657 if ((virt_addr >= hnode->vaddr) &&
658 (end_address <= hnode->vaddr + range_size)) {
659 valid = true;
660 break;
661 }
662 }
663 mutex_unlock(&ctx->mem_hash_lock);
664
665 if (!valid) {
666 dev_err(hdev->dev,
667 "virt addr 0x%llx is not mapped\n",
668 virt_addr);
669 return -EINVAL;
670 }
671
672 rc = hl_mmu_va_to_pa(ctx, virt_addr, phys_addr);
673 if (rc) {
674 dev_err(hdev->dev,
675 "virt addr 0x%llx is not mapped to phys addr\n",
676 virt_addr);
677 rc = -EINVAL;
678 }
679
680 return rc;
681 }
682
hl_access_dev_mem_by_region(struct hl_device * hdev,u64 addr,u64 * val,enum debugfs_access_type acc_type,bool * found)683 static int hl_access_dev_mem_by_region(struct hl_device *hdev, u64 addr,
684 u64 *val, enum debugfs_access_type acc_type, bool *found)
685 {
686 size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ?
687 sizeof(u64) : sizeof(u32);
688 struct pci_mem_region *mem_reg;
689 int i;
690
691 for (i = 0; i < PCI_REGION_NUMBER; i++) {
692 mem_reg = &hdev->pci_mem_region[i];
693 if (!mem_reg->used)
694 continue;
695 if (addr >= mem_reg->region_base &&
696 addr <= mem_reg->region_base + mem_reg->region_size - acc_size) {
697 *found = true;
698 return hdev->asic_funcs->access_dev_mem(hdev, mem_reg, i,
699 addr, val, acc_type);
700 }
701 }
702 return 0;
703 }
704
hl_access_host_mem(struct hl_device * hdev,u64 addr,u64 * val,enum debugfs_access_type acc_type)705 static void hl_access_host_mem(struct hl_device *hdev, u64 addr, u64 *val,
706 enum debugfs_access_type acc_type)
707 {
708 struct asic_fixed_properties *prop = &hdev->asic_prop;
709 u64 offset = prop->device_dma_offset_for_host_access;
710
711 switch (acc_type) {
712 case DEBUGFS_READ32:
713 *val = *(u32 *) phys_to_virt(addr - offset);
714 break;
715 case DEBUGFS_WRITE32:
716 *(u32 *) phys_to_virt(addr - offset) = *val;
717 break;
718 case DEBUGFS_READ64:
719 *val = *(u64 *) phys_to_virt(addr - offset);
720 break;
721 case DEBUGFS_WRITE64:
722 *(u64 *) phys_to_virt(addr - offset) = *val;
723 break;
724 default:
725 dev_err(hdev->dev, "hostmem access-type %d id not supported\n", acc_type);
726 break;
727 }
728 }
729
hl_access_mem(struct hl_device * hdev,u64 addr,u64 * val,enum debugfs_access_type acc_type)730 static int hl_access_mem(struct hl_device *hdev, u64 addr, u64 *val,
731 enum debugfs_access_type acc_type)
732 {
733 size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ?
734 sizeof(u64) : sizeof(u32);
735 u64 host_start = hdev->asic_prop.host_base_address;
736 u64 host_end = hdev->asic_prop.host_end_address;
737 bool user_address, found = false;
738 int rc;
739
740 user_address = hl_is_device_va(hdev, addr);
741 if (user_address) {
742 rc = device_va_to_pa(hdev, addr, acc_size, &addr);
743 if (rc)
744 return rc;
745 }
746
747 rc = hl_access_dev_mem_by_region(hdev, addr, val, acc_type, &found);
748 if (rc) {
749 dev_err(hdev->dev,
750 "Failed reading addr %#llx from dev mem (%d)\n",
751 addr, rc);
752 return rc;
753 }
754
755 if (found)
756 return 0;
757
758 if (!user_address || device_iommu_mapped(&hdev->pdev->dev)) {
759 rc = -EINVAL;
760 goto err;
761 }
762
763 if (addr >= host_start && addr <= host_end - acc_size) {
764 hl_access_host_mem(hdev, addr, val, acc_type);
765 } else {
766 rc = -EINVAL;
767 goto err;
768 }
769
770 return 0;
771 err:
772 dev_err(hdev->dev, "invalid addr %#llx\n", addr);
773 return rc;
774 }
775
hl_data_read32(struct file * f,char __user * buf,size_t count,loff_t * ppos)776 static ssize_t hl_data_read32(struct file *f, char __user *buf,
777 size_t count, loff_t *ppos)
778 {
779 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
780 struct hl_device *hdev = entry->hdev;
781 u64 value64, addr = entry->addr;
782 char tmp_buf[32];
783 ssize_t rc;
784 u32 val;
785
786 if (hdev->reset_info.in_reset) {
787 dev_warn_ratelimited(hdev->dev, "Can't read during reset\n");
788 return 0;
789 }
790
791 if (*ppos)
792 return 0;
793
794 rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_READ32);
795 if (rc)
796 return rc;
797
798 val = value64; /* downcast back to 32 */
799
800 sprintf(tmp_buf, "0x%08x\n", val);
801 return simple_read_from_buffer(buf, count, ppos, tmp_buf,
802 strlen(tmp_buf));
803 }
804
hl_data_write32(struct file * f,const char __user * buf,size_t count,loff_t * ppos)805 static ssize_t hl_data_write32(struct file *f, const char __user *buf,
806 size_t count, loff_t *ppos)
807 {
808 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
809 struct hl_device *hdev = entry->hdev;
810 u64 value64, addr = entry->addr;
811 u32 value;
812 ssize_t rc;
813
814 if (hdev->reset_info.in_reset) {
815 dev_warn_ratelimited(hdev->dev, "Can't write during reset\n");
816 return 0;
817 }
818
819 rc = kstrtouint_from_user(buf, count, 16, &value);
820 if (rc)
821 return rc;
822
823 value64 = value;
824 rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_WRITE32);
825 if (rc)
826 return rc;
827
828 return count;
829 }
830
hl_data_read64(struct file * f,char __user * buf,size_t count,loff_t * ppos)831 static ssize_t hl_data_read64(struct file *f, char __user *buf,
832 size_t count, loff_t *ppos)
833 {
834 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
835 struct hl_device *hdev = entry->hdev;
836 u64 addr = entry->addr;
837 char tmp_buf[32];
838 ssize_t rc;
839 u64 val;
840
841 if (hdev->reset_info.in_reset) {
842 dev_warn_ratelimited(hdev->dev, "Can't read during reset\n");
843 return 0;
844 }
845
846 if (*ppos)
847 return 0;
848
849 rc = hl_access_mem(hdev, addr, &val, DEBUGFS_READ64);
850 if (rc)
851 return rc;
852
853 sprintf(tmp_buf, "0x%016llx\n", val);
854 return simple_read_from_buffer(buf, count, ppos, tmp_buf,
855 strlen(tmp_buf));
856 }
857
hl_data_write64(struct file * f,const char __user * buf,size_t count,loff_t * ppos)858 static ssize_t hl_data_write64(struct file *f, const char __user *buf,
859 size_t count, loff_t *ppos)
860 {
861 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
862 struct hl_device *hdev = entry->hdev;
863 u64 addr = entry->addr;
864 u64 value;
865 ssize_t rc;
866
867 if (hdev->reset_info.in_reset) {
868 dev_warn_ratelimited(hdev->dev, "Can't write during reset\n");
869 return 0;
870 }
871
872 rc = kstrtoull_from_user(buf, count, 16, &value);
873 if (rc)
874 return rc;
875
876 rc = hl_access_mem(hdev, addr, &value, DEBUGFS_WRITE64);
877 if (rc)
878 return rc;
879
880 return count;
881 }
882
hl_dma_size_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)883 static ssize_t hl_dma_size_write(struct file *f, const char __user *buf,
884 size_t count, loff_t *ppos)
885 {
886 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
887 struct hl_device *hdev = entry->hdev;
888 u64 addr = entry->addr;
889 ssize_t rc;
890 u32 size;
891
892 if (hdev->reset_info.in_reset) {
893 dev_warn_ratelimited(hdev->dev, "Can't DMA during reset\n");
894 return 0;
895 }
896 rc = kstrtouint_from_user(buf, count, 16, &size);
897 if (rc)
898 return rc;
899
900 if (!size) {
901 dev_err(hdev->dev, "DMA read failed. size can't be 0\n");
902 return -EINVAL;
903 }
904
905 if (size > SZ_128M) {
906 dev_err(hdev->dev,
907 "DMA read failed. size can't be larger than 128MB\n");
908 return -EINVAL;
909 }
910
911 if (!hl_is_device_internal_memory_va(hdev, addr, size)) {
912 dev_err(hdev->dev,
913 "DMA read failed. Invalid 0x%010llx + 0x%08x\n",
914 addr, size);
915 return -EINVAL;
916 }
917
918 /* Free the previous allocation, if there was any */
919 entry->data_dma_blob_desc.size = 0;
920 vfree(entry->data_dma_blob_desc.data);
921
922 entry->data_dma_blob_desc.data = vmalloc(size);
923 if (!entry->data_dma_blob_desc.data)
924 return -ENOMEM;
925
926 rc = hdev->asic_funcs->debugfs_read_dma(hdev, addr, size,
927 entry->data_dma_blob_desc.data);
928 if (rc) {
929 dev_err(hdev->dev, "Failed to DMA from 0x%010llx\n", addr);
930 vfree(entry->data_dma_blob_desc.data);
931 entry->data_dma_blob_desc.data = NULL;
932 return -EIO;
933 }
934
935 entry->data_dma_blob_desc.size = size;
936
937 return count;
938 }
939
hl_monitor_dump_trigger(struct file * f,const char __user * buf,size_t count,loff_t * ppos)940 static ssize_t hl_monitor_dump_trigger(struct file *f, const char __user *buf,
941 size_t count, loff_t *ppos)
942 {
943 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
944 struct hl_device *hdev = entry->hdev;
945 u32 size, trig;
946 ssize_t rc;
947
948 if (hdev->reset_info.in_reset) {
949 dev_warn_ratelimited(hdev->dev, "Can't dump monitors during reset\n");
950 return 0;
951 }
952 rc = kstrtouint_from_user(buf, count, 10, &trig);
953 if (rc)
954 return rc;
955
956 if (trig != 1) {
957 dev_err(hdev->dev, "Must write 1 to trigger monitor dump\n");
958 return -EINVAL;
959 }
960
961 size = sizeof(struct cpucp_monitor_dump);
962
963 /* Free the previous allocation, if there was any */
964 entry->mon_dump_blob_desc.size = 0;
965 vfree(entry->mon_dump_blob_desc.data);
966
967 entry->mon_dump_blob_desc.data = vmalloc(size);
968 if (!entry->mon_dump_blob_desc.data)
969 return -ENOMEM;
970
971 rc = hdev->asic_funcs->get_monitor_dump(hdev, entry->mon_dump_blob_desc.data);
972 if (rc) {
973 dev_err(hdev->dev, "Failed to dump monitors\n");
974 vfree(entry->mon_dump_blob_desc.data);
975 entry->mon_dump_blob_desc.data = NULL;
976 return -EIO;
977 }
978
979 entry->mon_dump_blob_desc.size = size;
980
981 return count;
982 }
983
hl_get_power_state(struct file * f,char __user * buf,size_t count,loff_t * ppos)984 static ssize_t hl_get_power_state(struct file *f, char __user *buf,
985 size_t count, loff_t *ppos)
986 {
987 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
988 struct hl_device *hdev = entry->hdev;
989 char tmp_buf[200];
990 int i;
991
992 if (*ppos)
993 return 0;
994
995 if (hdev->pdev->current_state == PCI_D0)
996 i = 1;
997 else if (hdev->pdev->current_state == PCI_D3hot)
998 i = 2;
999 else
1000 i = 3;
1001
1002 sprintf(tmp_buf,
1003 "current power state: %d\n1 - D0\n2 - D3hot\n3 - Unknown\n", i);
1004 return simple_read_from_buffer(buf, count, ppos, tmp_buf,
1005 strlen(tmp_buf));
1006 }
1007
hl_set_power_state(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1008 static ssize_t hl_set_power_state(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 value;
1014 ssize_t rc;
1015
1016 rc = kstrtouint_from_user(buf, count, 10, &value);
1017 if (rc)
1018 return rc;
1019
1020 if (value == 1) {
1021 pci_set_power_state(hdev->pdev, PCI_D0);
1022 pci_restore_state(hdev->pdev);
1023 rc = pci_enable_device(hdev->pdev);
1024 if (rc < 0)
1025 return rc;
1026 } else if (value == 2) {
1027 pci_save_state(hdev->pdev);
1028 pci_disable_device(hdev->pdev);
1029 pci_set_power_state(hdev->pdev, PCI_D3hot);
1030 } else {
1031 dev_dbg(hdev->dev, "invalid power state value %u\n", value);
1032 return -EINVAL;
1033 }
1034
1035 return count;
1036 }
1037
hl_i2c_data_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1038 static ssize_t hl_i2c_data_read(struct file *f, char __user *buf,
1039 size_t count, loff_t *ppos)
1040 {
1041 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1042 struct hl_device *hdev = entry->hdev;
1043 char tmp_buf[32];
1044 u64 val;
1045 ssize_t rc;
1046
1047 if (*ppos)
1048 return 0;
1049
1050 rc = hl_debugfs_i2c_read(hdev, entry->i2c_bus, entry->i2c_addr,
1051 entry->i2c_reg, entry->i2c_len, &val);
1052 if (rc) {
1053 dev_err(hdev->dev,
1054 "Failed to read from I2C bus %d, addr %d, reg %d, len %d\n",
1055 entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len);
1056 return rc;
1057 }
1058
1059 sprintf(tmp_buf, "%#02llx\n", val);
1060 rc = simple_read_from_buffer(buf, count, ppos, tmp_buf,
1061 strlen(tmp_buf));
1062
1063 return rc;
1064 }
1065
hl_i2c_data_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1066 static ssize_t hl_i2c_data_write(struct file *f, const char __user *buf,
1067 size_t count, loff_t *ppos)
1068 {
1069 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1070 struct hl_device *hdev = entry->hdev;
1071 u64 value;
1072 ssize_t rc;
1073
1074 rc = kstrtou64_from_user(buf, count, 16, &value);
1075 if (rc)
1076 return rc;
1077
1078 rc = hl_debugfs_i2c_write(hdev, entry->i2c_bus, entry->i2c_addr,
1079 entry->i2c_reg, entry->i2c_len, value);
1080 if (rc) {
1081 dev_err(hdev->dev,
1082 "Failed to write %#02llx to I2C bus %d, addr %d, reg %d, len %d\n",
1083 value, entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len);
1084 return rc;
1085 }
1086
1087 return count;
1088 }
1089
hl_led0_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1090 static ssize_t hl_led0_write(struct file *f, const char __user *buf,
1091 size_t count, loff_t *ppos)
1092 {
1093 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1094 struct hl_device *hdev = entry->hdev;
1095 u32 value;
1096 ssize_t rc;
1097
1098 rc = kstrtouint_from_user(buf, count, 10, &value);
1099 if (rc)
1100 return rc;
1101
1102 value = value ? 1 : 0;
1103
1104 hl_debugfs_led_set(hdev, 0, value);
1105
1106 return count;
1107 }
1108
hl_led1_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1109 static ssize_t hl_led1_write(struct file *f, const char __user *buf,
1110 size_t count, loff_t *ppos)
1111 {
1112 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1113 struct hl_device *hdev = entry->hdev;
1114 u32 value;
1115 ssize_t rc;
1116
1117 rc = kstrtouint_from_user(buf, count, 10, &value);
1118 if (rc)
1119 return rc;
1120
1121 value = value ? 1 : 0;
1122
1123 hl_debugfs_led_set(hdev, 1, value);
1124
1125 return count;
1126 }
1127
hl_led2_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1128 static ssize_t hl_led2_write(struct file *f, const char __user *buf,
1129 size_t count, loff_t *ppos)
1130 {
1131 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1132 struct hl_device *hdev = entry->hdev;
1133 u32 value;
1134 ssize_t rc;
1135
1136 rc = kstrtouint_from_user(buf, count, 10, &value);
1137 if (rc)
1138 return rc;
1139
1140 value = value ? 1 : 0;
1141
1142 hl_debugfs_led_set(hdev, 2, value);
1143
1144 return count;
1145 }
1146
hl_device_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1147 static ssize_t hl_device_read(struct file *f, char __user *buf,
1148 size_t count, loff_t *ppos)
1149 {
1150 static const char *help =
1151 "Valid values: disable, enable, suspend, resume, cpu_timeout\n";
1152 return simple_read_from_buffer(buf, count, ppos, help, strlen(help));
1153 }
1154
hl_device_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1155 static ssize_t hl_device_write(struct file *f, const char __user *buf,
1156 size_t count, loff_t *ppos)
1157 {
1158 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1159 struct hl_device *hdev = entry->hdev;
1160 char data[30] = {0};
1161
1162 /* don't allow partial writes */
1163 if (*ppos != 0)
1164 return 0;
1165
1166 simple_write_to_buffer(data, 29, ppos, buf, count);
1167
1168 if (strncmp("disable", data, strlen("disable")) == 0) {
1169 hdev->disabled = true;
1170 } else if (strncmp("enable", data, strlen("enable")) == 0) {
1171 hdev->disabled = false;
1172 } else if (strncmp("suspend", data, strlen("suspend")) == 0) {
1173 hdev->asic_funcs->suspend(hdev);
1174 } else if (strncmp("resume", data, strlen("resume")) == 0) {
1175 hdev->asic_funcs->resume(hdev);
1176 } else if (strncmp("cpu_timeout", data, strlen("cpu_timeout")) == 0) {
1177 hdev->device_cpu_disabled = true;
1178 } else {
1179 dev_err(hdev->dev,
1180 "Valid values: disable, enable, suspend, resume, cpu_timeout\n");
1181 count = -EINVAL;
1182 }
1183
1184 return count;
1185 }
1186
hl_clk_gate_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1187 static ssize_t hl_clk_gate_read(struct file *f, char __user *buf,
1188 size_t count, loff_t *ppos)
1189 {
1190 return 0;
1191 }
1192
hl_clk_gate_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1193 static ssize_t hl_clk_gate_write(struct file *f, const char __user *buf,
1194 size_t count, loff_t *ppos)
1195 {
1196 return count;
1197 }
1198
hl_stop_on_err_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1199 static ssize_t hl_stop_on_err_read(struct file *f, char __user *buf,
1200 size_t count, loff_t *ppos)
1201 {
1202 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1203 struct hl_device *hdev = entry->hdev;
1204 char tmp_buf[200];
1205 ssize_t rc;
1206
1207 if (!hdev->asic_prop.configurable_stop_on_err)
1208 return -EOPNOTSUPP;
1209
1210 if (*ppos)
1211 return 0;
1212
1213 sprintf(tmp_buf, "%d\n", hdev->stop_on_err);
1214 rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf,
1215 strlen(tmp_buf) + 1);
1216
1217 return rc;
1218 }
1219
hl_stop_on_err_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1220 static ssize_t hl_stop_on_err_write(struct file *f, const char __user *buf,
1221 size_t count, loff_t *ppos)
1222 {
1223 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1224 struct hl_device *hdev = entry->hdev;
1225 u32 value;
1226 ssize_t rc;
1227
1228 if (!hdev->asic_prop.configurable_stop_on_err)
1229 return -EOPNOTSUPP;
1230
1231 if (hdev->reset_info.in_reset) {
1232 dev_warn_ratelimited(hdev->dev,
1233 "Can't change stop on error during reset\n");
1234 return 0;
1235 }
1236
1237 rc = kstrtouint_from_user(buf, count, 10, &value);
1238 if (rc)
1239 return rc;
1240
1241 hdev->stop_on_err = value ? 1 : 0;
1242
1243 hl_device_reset(hdev, 0);
1244
1245 return count;
1246 }
1247
hl_security_violations_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1248 static ssize_t hl_security_violations_read(struct file *f, char __user *buf,
1249 size_t count, loff_t *ppos)
1250 {
1251 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1252 struct hl_device *hdev = entry->hdev;
1253
1254 hdev->asic_funcs->ack_protection_bits_errors(hdev);
1255
1256 return 0;
1257 }
1258
hl_state_dump_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1259 static ssize_t hl_state_dump_read(struct file *f, char __user *buf,
1260 size_t count, loff_t *ppos)
1261 {
1262 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1263 ssize_t rc;
1264
1265 down_read(&entry->state_dump_sem);
1266 if (!entry->state_dump[entry->state_dump_head])
1267 rc = 0;
1268 else
1269 rc = simple_read_from_buffer(
1270 buf, count, ppos,
1271 entry->state_dump[entry->state_dump_head],
1272 strlen(entry->state_dump[entry->state_dump_head]));
1273 up_read(&entry->state_dump_sem);
1274
1275 return rc;
1276 }
1277
hl_state_dump_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1278 static ssize_t hl_state_dump_write(struct file *f, const char __user *buf,
1279 size_t count, loff_t *ppos)
1280 {
1281 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1282 struct hl_device *hdev = entry->hdev;
1283 ssize_t rc;
1284 u32 size;
1285 int i;
1286
1287 rc = kstrtouint_from_user(buf, count, 10, &size);
1288 if (rc)
1289 return rc;
1290
1291 if (size <= 0 || size >= ARRAY_SIZE(entry->state_dump)) {
1292 dev_err(hdev->dev, "Invalid number of dumps to skip\n");
1293 return -EINVAL;
1294 }
1295
1296 if (entry->state_dump[entry->state_dump_head]) {
1297 down_write(&entry->state_dump_sem);
1298 for (i = 0; i < size; ++i) {
1299 vfree(entry->state_dump[entry->state_dump_head]);
1300 entry->state_dump[entry->state_dump_head] = NULL;
1301 if (entry->state_dump_head > 0)
1302 entry->state_dump_head--;
1303 else
1304 entry->state_dump_head =
1305 ARRAY_SIZE(entry->state_dump) - 1;
1306 }
1307 up_write(&entry->state_dump_sem);
1308 }
1309
1310 return count;
1311 }
1312
hl_timeout_locked_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)1313 static ssize_t hl_timeout_locked_read(struct file *f, char __user *buf,
1314 size_t count, loff_t *ppos)
1315 {
1316 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1317 struct hl_device *hdev = entry->hdev;
1318 char tmp_buf[200];
1319 ssize_t rc;
1320
1321 if (*ppos)
1322 return 0;
1323
1324 sprintf(tmp_buf, "%d\n",
1325 jiffies_to_msecs(hdev->timeout_jiffies) / 1000);
1326 rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf,
1327 strlen(tmp_buf) + 1);
1328
1329 return rc;
1330 }
1331
hl_timeout_locked_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)1332 static ssize_t hl_timeout_locked_write(struct file *f, const char __user *buf,
1333 size_t count, loff_t *ppos)
1334 {
1335 struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1336 struct hl_device *hdev = entry->hdev;
1337 u32 value;
1338 ssize_t rc;
1339
1340 rc = kstrtouint_from_user(buf, count, 10, &value);
1341 if (rc)
1342 return rc;
1343
1344 if (value)
1345 hdev->timeout_jiffies = msecs_to_jiffies(value * 1000);
1346 else
1347 hdev->timeout_jiffies = MAX_SCHEDULE_TIMEOUT;
1348
1349 return count;
1350 }
1351
1352 static const struct file_operations hl_mem_scrub_fops = {
1353 .owner = THIS_MODULE,
1354 .write = hl_memory_scrub,
1355 };
1356
1357 static const struct file_operations hl_data32b_fops = {
1358 .owner = THIS_MODULE,
1359 .read = hl_data_read32,
1360 .write = hl_data_write32
1361 };
1362
1363 static const struct file_operations hl_data64b_fops = {
1364 .owner = THIS_MODULE,
1365 .read = hl_data_read64,
1366 .write = hl_data_write64
1367 };
1368
1369 static const struct file_operations hl_dma_size_fops = {
1370 .owner = THIS_MODULE,
1371 .write = hl_dma_size_write
1372 };
1373
1374 static const struct file_operations hl_monitor_dump_fops = {
1375 .owner = THIS_MODULE,
1376 .write = hl_monitor_dump_trigger
1377 };
1378
1379 static const struct file_operations hl_i2c_data_fops = {
1380 .owner = THIS_MODULE,
1381 .read = hl_i2c_data_read,
1382 .write = hl_i2c_data_write
1383 };
1384
1385 static const struct file_operations hl_power_fops = {
1386 .owner = THIS_MODULE,
1387 .read = hl_get_power_state,
1388 .write = hl_set_power_state
1389 };
1390
1391 static const struct file_operations hl_led0_fops = {
1392 .owner = THIS_MODULE,
1393 .write = hl_led0_write
1394 };
1395
1396 static const struct file_operations hl_led1_fops = {
1397 .owner = THIS_MODULE,
1398 .write = hl_led1_write
1399 };
1400
1401 static const struct file_operations hl_led2_fops = {
1402 .owner = THIS_MODULE,
1403 .write = hl_led2_write
1404 };
1405
1406 static const struct file_operations hl_device_fops = {
1407 .owner = THIS_MODULE,
1408 .read = hl_device_read,
1409 .write = hl_device_write
1410 };
1411
1412 static const struct file_operations hl_clk_gate_fops = {
1413 .owner = THIS_MODULE,
1414 .read = hl_clk_gate_read,
1415 .write = hl_clk_gate_write
1416 };
1417
1418 static const struct file_operations hl_stop_on_err_fops = {
1419 .owner = THIS_MODULE,
1420 .read = hl_stop_on_err_read,
1421 .write = hl_stop_on_err_write
1422 };
1423
1424 static const struct file_operations hl_security_violations_fops = {
1425 .owner = THIS_MODULE,
1426 .read = hl_security_violations_read
1427 };
1428
1429 static const struct file_operations hl_state_dump_fops = {
1430 .owner = THIS_MODULE,
1431 .read = hl_state_dump_read,
1432 .write = hl_state_dump_write
1433 };
1434
1435 static const struct file_operations hl_timeout_locked_fops = {
1436 .owner = THIS_MODULE,
1437 .read = hl_timeout_locked_read,
1438 .write = hl_timeout_locked_write
1439 };
1440
1441 static const struct hl_info_list hl_debugfs_list[] = {
1442 {"command_buffers", command_buffers_show, NULL},
1443 {"command_submission", command_submission_show, NULL},
1444 {"command_submission_jobs", command_submission_jobs_show, NULL},
1445 {"userptr", userptr_show, NULL},
1446 {"vm", vm_show, NULL},
1447 {"userptr_lookup", userptr_lookup_show, userptr_lookup_write},
1448 {"mmu", mmu_show, mmu_asid_va_write},
1449 {"engines", engines_show, NULL}
1450 };
1451
hl_debugfs_open(struct inode * inode,struct file * file)1452 static int hl_debugfs_open(struct inode *inode, struct file *file)
1453 {
1454 struct hl_debugfs_entry *node = inode->i_private;
1455
1456 return single_open(file, node->info_ent->show, node);
1457 }
1458
hl_debugfs_write(struct file * file,const char __user * buf,size_t count,loff_t * f_pos)1459 static ssize_t hl_debugfs_write(struct file *file, const char __user *buf,
1460 size_t count, loff_t *f_pos)
1461 {
1462 struct hl_debugfs_entry *node = file->f_inode->i_private;
1463
1464 if (node->info_ent->write)
1465 return node->info_ent->write(file, buf, count, f_pos);
1466 else
1467 return -EINVAL;
1468
1469 }
1470
1471 static const struct file_operations hl_debugfs_fops = {
1472 .owner = THIS_MODULE,
1473 .open = hl_debugfs_open,
1474 .read = seq_read,
1475 .write = hl_debugfs_write,
1476 .llseek = seq_lseek,
1477 .release = single_release,
1478 };
1479
hl_debugfs_add_device(struct hl_device * hdev)1480 void hl_debugfs_add_device(struct hl_device *hdev)
1481 {
1482 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1483 int count = ARRAY_SIZE(hl_debugfs_list);
1484 struct hl_debugfs_entry *entry;
1485 int i;
1486
1487 dev_entry->hdev = hdev;
1488 dev_entry->entry_arr = kmalloc_array(count,
1489 sizeof(struct hl_debugfs_entry),
1490 GFP_KERNEL);
1491 if (!dev_entry->entry_arr)
1492 return;
1493
1494 dev_entry->data_dma_blob_desc.size = 0;
1495 dev_entry->data_dma_blob_desc.data = NULL;
1496 dev_entry->mon_dump_blob_desc.size = 0;
1497 dev_entry->mon_dump_blob_desc.data = NULL;
1498
1499 INIT_LIST_HEAD(&dev_entry->file_list);
1500 INIT_LIST_HEAD(&dev_entry->cb_list);
1501 INIT_LIST_HEAD(&dev_entry->cs_list);
1502 INIT_LIST_HEAD(&dev_entry->cs_job_list);
1503 INIT_LIST_HEAD(&dev_entry->userptr_list);
1504 INIT_LIST_HEAD(&dev_entry->ctx_mem_hash_list);
1505 mutex_init(&dev_entry->file_mutex);
1506 init_rwsem(&dev_entry->state_dump_sem);
1507 spin_lock_init(&dev_entry->cb_spinlock);
1508 spin_lock_init(&dev_entry->cs_spinlock);
1509 spin_lock_init(&dev_entry->cs_job_spinlock);
1510 spin_lock_init(&dev_entry->userptr_spinlock);
1511 spin_lock_init(&dev_entry->ctx_mem_hash_spinlock);
1512
1513 dev_entry->root = debugfs_create_dir(dev_name(hdev->dev),
1514 hl_debug_root);
1515
1516 debugfs_create_x64("memory_scrub_val",
1517 0644,
1518 dev_entry->root,
1519 &dev_entry->memory_scrub_val);
1520
1521 debugfs_create_file("memory_scrub",
1522 0200,
1523 dev_entry->root,
1524 dev_entry,
1525 &hl_mem_scrub_fops);
1526
1527 debugfs_create_x64("addr",
1528 0644,
1529 dev_entry->root,
1530 &dev_entry->addr);
1531
1532 debugfs_create_file("data32",
1533 0644,
1534 dev_entry->root,
1535 dev_entry,
1536 &hl_data32b_fops);
1537
1538 debugfs_create_file("data64",
1539 0644,
1540 dev_entry->root,
1541 dev_entry,
1542 &hl_data64b_fops);
1543
1544 debugfs_create_file("set_power_state",
1545 0200,
1546 dev_entry->root,
1547 dev_entry,
1548 &hl_power_fops);
1549
1550 debugfs_create_u8("i2c_bus",
1551 0644,
1552 dev_entry->root,
1553 &dev_entry->i2c_bus);
1554
1555 debugfs_create_u8("i2c_addr",
1556 0644,
1557 dev_entry->root,
1558 &dev_entry->i2c_addr);
1559
1560 debugfs_create_u8("i2c_reg",
1561 0644,
1562 dev_entry->root,
1563 &dev_entry->i2c_reg);
1564
1565 debugfs_create_u8("i2c_len",
1566 0644,
1567 dev_entry->root,
1568 &dev_entry->i2c_len);
1569
1570 debugfs_create_file("i2c_data",
1571 0644,
1572 dev_entry->root,
1573 dev_entry,
1574 &hl_i2c_data_fops);
1575
1576 debugfs_create_file("led0",
1577 0200,
1578 dev_entry->root,
1579 dev_entry,
1580 &hl_led0_fops);
1581
1582 debugfs_create_file("led1",
1583 0200,
1584 dev_entry->root,
1585 dev_entry,
1586 &hl_led1_fops);
1587
1588 debugfs_create_file("led2",
1589 0200,
1590 dev_entry->root,
1591 dev_entry,
1592 &hl_led2_fops);
1593
1594 debugfs_create_file("device",
1595 0200,
1596 dev_entry->root,
1597 dev_entry,
1598 &hl_device_fops);
1599
1600 debugfs_create_file("clk_gate",
1601 0200,
1602 dev_entry->root,
1603 dev_entry,
1604 &hl_clk_gate_fops);
1605
1606 debugfs_create_file("stop_on_err",
1607 0644,
1608 dev_entry->root,
1609 dev_entry,
1610 &hl_stop_on_err_fops);
1611
1612 debugfs_create_file("dump_security_violations",
1613 0644,
1614 dev_entry->root,
1615 dev_entry,
1616 &hl_security_violations_fops);
1617
1618 debugfs_create_file("dma_size",
1619 0200,
1620 dev_entry->root,
1621 dev_entry,
1622 &hl_dma_size_fops);
1623
1624 debugfs_create_blob("data_dma",
1625 0400,
1626 dev_entry->root,
1627 &dev_entry->data_dma_blob_desc);
1628
1629 debugfs_create_file("monitor_dump_trig",
1630 0200,
1631 dev_entry->root,
1632 dev_entry,
1633 &hl_monitor_dump_fops);
1634
1635 debugfs_create_blob("monitor_dump",
1636 0400,
1637 dev_entry->root,
1638 &dev_entry->mon_dump_blob_desc);
1639
1640 debugfs_create_x8("skip_reset_on_timeout",
1641 0644,
1642 dev_entry->root,
1643 &hdev->reset_info.skip_reset_on_timeout);
1644
1645 debugfs_create_file("state_dump",
1646 0600,
1647 dev_entry->root,
1648 dev_entry,
1649 &hl_state_dump_fops);
1650
1651 debugfs_create_file("timeout_locked",
1652 0644,
1653 dev_entry->root,
1654 dev_entry,
1655 &hl_timeout_locked_fops);
1656
1657 for (i = 0, entry = dev_entry->entry_arr ; i < count ; i++, entry++) {
1658 debugfs_create_file(hl_debugfs_list[i].name,
1659 0444,
1660 dev_entry->root,
1661 entry,
1662 &hl_debugfs_fops);
1663 entry->info_ent = &hl_debugfs_list[i];
1664 entry->dev_entry = dev_entry;
1665 }
1666 }
1667
hl_debugfs_remove_device(struct hl_device * hdev)1668 void hl_debugfs_remove_device(struct hl_device *hdev)
1669 {
1670 struct hl_dbg_device_entry *entry = &hdev->hl_debugfs;
1671 int i;
1672
1673 debugfs_remove_recursive(entry->root);
1674
1675 mutex_destroy(&entry->file_mutex);
1676
1677 vfree(entry->data_dma_blob_desc.data);
1678 vfree(entry->mon_dump_blob_desc.data);
1679
1680 for (i = 0; i < ARRAY_SIZE(entry->state_dump); ++i)
1681 vfree(entry->state_dump[i]);
1682
1683 kfree(entry->entry_arr);
1684 }
1685
hl_debugfs_add_file(struct hl_fpriv * hpriv)1686 void hl_debugfs_add_file(struct hl_fpriv *hpriv)
1687 {
1688 struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs;
1689
1690 mutex_lock(&dev_entry->file_mutex);
1691 list_add(&hpriv->debugfs_list, &dev_entry->file_list);
1692 mutex_unlock(&dev_entry->file_mutex);
1693 }
1694
hl_debugfs_remove_file(struct hl_fpriv * hpriv)1695 void hl_debugfs_remove_file(struct hl_fpriv *hpriv)
1696 {
1697 struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs;
1698
1699 mutex_lock(&dev_entry->file_mutex);
1700 list_del(&hpriv->debugfs_list);
1701 mutex_unlock(&dev_entry->file_mutex);
1702 }
1703
hl_debugfs_add_cb(struct hl_cb * cb)1704 void hl_debugfs_add_cb(struct hl_cb *cb)
1705 {
1706 struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs;
1707
1708 spin_lock(&dev_entry->cb_spinlock);
1709 list_add(&cb->debugfs_list, &dev_entry->cb_list);
1710 spin_unlock(&dev_entry->cb_spinlock);
1711 }
1712
hl_debugfs_remove_cb(struct hl_cb * cb)1713 void hl_debugfs_remove_cb(struct hl_cb *cb)
1714 {
1715 struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs;
1716
1717 spin_lock(&dev_entry->cb_spinlock);
1718 list_del(&cb->debugfs_list);
1719 spin_unlock(&dev_entry->cb_spinlock);
1720 }
1721
hl_debugfs_add_cs(struct hl_cs * cs)1722 void hl_debugfs_add_cs(struct hl_cs *cs)
1723 {
1724 struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs;
1725
1726 spin_lock(&dev_entry->cs_spinlock);
1727 list_add(&cs->debugfs_list, &dev_entry->cs_list);
1728 spin_unlock(&dev_entry->cs_spinlock);
1729 }
1730
hl_debugfs_remove_cs(struct hl_cs * cs)1731 void hl_debugfs_remove_cs(struct hl_cs *cs)
1732 {
1733 struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs;
1734
1735 spin_lock(&dev_entry->cs_spinlock);
1736 list_del(&cs->debugfs_list);
1737 spin_unlock(&dev_entry->cs_spinlock);
1738 }
1739
hl_debugfs_add_job(struct hl_device * hdev,struct hl_cs_job * job)1740 void hl_debugfs_add_job(struct hl_device *hdev, struct hl_cs_job *job)
1741 {
1742 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1743
1744 spin_lock(&dev_entry->cs_job_spinlock);
1745 list_add(&job->debugfs_list, &dev_entry->cs_job_list);
1746 spin_unlock(&dev_entry->cs_job_spinlock);
1747 }
1748
hl_debugfs_remove_job(struct hl_device * hdev,struct hl_cs_job * job)1749 void hl_debugfs_remove_job(struct hl_device *hdev, struct hl_cs_job *job)
1750 {
1751 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1752
1753 spin_lock(&dev_entry->cs_job_spinlock);
1754 list_del(&job->debugfs_list);
1755 spin_unlock(&dev_entry->cs_job_spinlock);
1756 }
1757
hl_debugfs_add_userptr(struct hl_device * hdev,struct hl_userptr * userptr)1758 void hl_debugfs_add_userptr(struct hl_device *hdev, struct hl_userptr *userptr)
1759 {
1760 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1761
1762 spin_lock(&dev_entry->userptr_spinlock);
1763 list_add(&userptr->debugfs_list, &dev_entry->userptr_list);
1764 spin_unlock(&dev_entry->userptr_spinlock);
1765 }
1766
hl_debugfs_remove_userptr(struct hl_device * hdev,struct hl_userptr * userptr)1767 void hl_debugfs_remove_userptr(struct hl_device *hdev,
1768 struct hl_userptr *userptr)
1769 {
1770 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1771
1772 spin_lock(&dev_entry->userptr_spinlock);
1773 list_del(&userptr->debugfs_list);
1774 spin_unlock(&dev_entry->userptr_spinlock);
1775 }
1776
hl_debugfs_add_ctx_mem_hash(struct hl_device * hdev,struct hl_ctx * ctx)1777 void hl_debugfs_add_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx)
1778 {
1779 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1780
1781 spin_lock(&dev_entry->ctx_mem_hash_spinlock);
1782 list_add(&ctx->debugfs_list, &dev_entry->ctx_mem_hash_list);
1783 spin_unlock(&dev_entry->ctx_mem_hash_spinlock);
1784 }
1785
hl_debugfs_remove_ctx_mem_hash(struct hl_device * hdev,struct hl_ctx * ctx)1786 void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx)
1787 {
1788 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1789
1790 spin_lock(&dev_entry->ctx_mem_hash_spinlock);
1791 list_del(&ctx->debugfs_list);
1792 spin_unlock(&dev_entry->ctx_mem_hash_spinlock);
1793 }
1794
1795 /**
1796 * hl_debugfs_set_state_dump - register state dump making it accessible via
1797 * debugfs
1798 * @hdev: pointer to the device structure
1799 * @data: the actual dump data
1800 * @length: the length of the data
1801 */
hl_debugfs_set_state_dump(struct hl_device * hdev,char * data,unsigned long length)1802 void hl_debugfs_set_state_dump(struct hl_device *hdev, char *data,
1803 unsigned long length)
1804 {
1805 struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1806
1807 down_write(&dev_entry->state_dump_sem);
1808
1809 dev_entry->state_dump_head = (dev_entry->state_dump_head + 1) %
1810 ARRAY_SIZE(dev_entry->state_dump);
1811 vfree(dev_entry->state_dump[dev_entry->state_dump_head]);
1812 dev_entry->state_dump[dev_entry->state_dump_head] = data;
1813
1814 up_write(&dev_entry->state_dump_sem);
1815 }
1816
hl_debugfs_init(void)1817 void __init hl_debugfs_init(void)
1818 {
1819 hl_debug_root = debugfs_create_dir("habanalabs", NULL);
1820 }
1821
hl_debugfs_fini(void)1822 void hl_debugfs_fini(void)
1823 {
1824 debugfs_remove_recursive(hl_debug_root);
1825 }
1826