xref: /linux/drivers/accel/habanalabs/common/debugfs.c (revision 223981db9bafb80f558162c148f261e2ff043dbe)
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 int hl_debugfs_i2c_read(struct hl_device *hdev, u8 i2c_bus, u8 i2c_addr,
22 				u8 i2c_reg, u8 i2c_len, u64 *val)
23 {
24 	struct cpucp_packet pkt;
25 	int rc;
26 
27 	if (!hl_device_operational(hdev, NULL))
28 		return -EBUSY;
29 
30 	if (i2c_len > I2C_MAX_TRANSACTION_LEN) {
31 		dev_err(hdev->dev, "I2C transaction length %u, exceeds maximum of %u\n",
32 				i2c_len, I2C_MAX_TRANSACTION_LEN);
33 		return -EINVAL;
34 	}
35 
36 	memset(&pkt, 0, sizeof(pkt));
37 
38 	pkt.ctl = cpu_to_le32(CPUCP_PACKET_I2C_RD <<
39 				CPUCP_PKT_CTL_OPCODE_SHIFT);
40 	pkt.i2c_bus = i2c_bus;
41 	pkt.i2c_addr = i2c_addr;
42 	pkt.i2c_reg = i2c_reg;
43 	pkt.i2c_len = i2c_len;
44 
45 	rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt),
46 						0, val);
47 	if (rc)
48 		dev_err(hdev->dev, "Failed to read from I2C, error %d\n", rc);
49 
50 	return rc;
51 }
52 
53 static int hl_debugfs_i2c_write(struct hl_device *hdev, u8 i2c_bus, u8 i2c_addr,
54 				u8 i2c_reg, u8 i2c_len, u64 val)
55 {
56 	struct cpucp_packet pkt;
57 	int rc;
58 
59 	if (!hl_device_operational(hdev, NULL))
60 		return -EBUSY;
61 
62 	if (i2c_len > I2C_MAX_TRANSACTION_LEN) {
63 		dev_err(hdev->dev, "I2C transaction length %u, exceeds maximum of %u\n",
64 				i2c_len, I2C_MAX_TRANSACTION_LEN);
65 		return -EINVAL;
66 	}
67 
68 	memset(&pkt, 0, sizeof(pkt));
69 
70 	pkt.ctl = cpu_to_le32(CPUCP_PACKET_I2C_WR <<
71 				CPUCP_PKT_CTL_OPCODE_SHIFT);
72 	pkt.i2c_bus = i2c_bus;
73 	pkt.i2c_addr = i2c_addr;
74 	pkt.i2c_reg = i2c_reg;
75 	pkt.i2c_len = i2c_len;
76 	pkt.value = cpu_to_le64(val);
77 
78 	rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt),
79 						0, NULL);
80 
81 	if (rc)
82 		dev_err(hdev->dev, "Failed to write to I2C, error %d\n", rc);
83 
84 	return rc;
85 }
86 
87 static void hl_debugfs_led_set(struct hl_device *hdev, u8 led, u8 state)
88 {
89 	struct cpucp_packet pkt;
90 	int rc;
91 
92 	if (!hl_device_operational(hdev, NULL))
93 		return;
94 
95 	memset(&pkt, 0, sizeof(pkt));
96 
97 	pkt.ctl = cpu_to_le32(CPUCP_PACKET_LED_SET <<
98 				CPUCP_PKT_CTL_OPCODE_SHIFT);
99 	pkt.led_index = cpu_to_le32(led);
100 	pkt.value = cpu_to_le64(state);
101 
102 	rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt),
103 						0, NULL);
104 
105 	if (rc)
106 		dev_err(hdev->dev, "Failed to set LED %d, error %d\n", led, rc);
107 }
108 
109 static int command_buffers_show(struct seq_file *s, void *data)
110 {
111 	struct hl_debugfs_entry *entry = s->private;
112 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
113 	struct hl_cb *cb;
114 	bool first = true;
115 
116 	spin_lock(&dev_entry->cb_spinlock);
117 
118 	list_for_each_entry(cb, &dev_entry->cb_list, debugfs_list) {
119 		if (first) {
120 			first = false;
121 			seq_puts(s, "\n");
122 			seq_puts(s, " CB ID   CTX ID   CB size    CB RefCnt    mmap?   CS counter\n");
123 			seq_puts(s, "---------------------------------------------------------------\n");
124 		}
125 		seq_printf(s,
126 			"   %03llu        %d    0x%08x      %d          %d          %d\n",
127 			cb->buf->handle, cb->ctx->asid, cb->size,
128 			kref_read(&cb->buf->refcount),
129 			atomic_read(&cb->buf->mmap), atomic_read(&cb->cs_cnt));
130 	}
131 
132 	spin_unlock(&dev_entry->cb_spinlock);
133 
134 	if (!first)
135 		seq_puts(s, "\n");
136 
137 	return 0;
138 }
139 
140 static int command_submission_show(struct seq_file *s, void *data)
141 {
142 	struct hl_debugfs_entry *entry = s->private;
143 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
144 	struct hl_cs *cs;
145 	bool first = true;
146 
147 	spin_lock(&dev_entry->cs_spinlock);
148 
149 	list_for_each_entry(cs, &dev_entry->cs_list, debugfs_list) {
150 		if (first) {
151 			first = false;
152 			seq_puts(s, "\n");
153 			seq_puts(s, " CS ID   CS TYPE   CTX ASID   CS RefCnt   Submitted    Completed\n");
154 			seq_puts(s, "----------------------------------------------------------------\n");
155 		}
156 		seq_printf(s,
157 			"   %llu        %d          %d          %d           %d            %d\n",
158 			cs->sequence, cs->type, cs->ctx->asid,
159 			kref_read(&cs->refcount),
160 			cs->submitted, cs->completed);
161 	}
162 
163 	spin_unlock(&dev_entry->cs_spinlock);
164 
165 	if (!first)
166 		seq_puts(s, "\n");
167 
168 	return 0;
169 }
170 
171 static int command_submission_jobs_show(struct seq_file *s, void *data)
172 {
173 	struct hl_debugfs_entry *entry = s->private;
174 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
175 	struct hl_cs_job *job;
176 	bool first = true;
177 
178 	spin_lock(&dev_entry->cs_job_spinlock);
179 
180 	list_for_each_entry(job, &dev_entry->cs_job_list, debugfs_list) {
181 		if (first) {
182 			first = false;
183 			seq_puts(s, "\n");
184 			seq_puts(s, " JOB ID   CS ID    CS TYPE    CTX ASID   JOB RefCnt   H/W Queue\n");
185 			seq_puts(s, "---------------------------------------------------------------\n");
186 		}
187 		if (job->cs)
188 			seq_printf(s,
189 				"   %02d      %llu        %d        %d          %d           %d\n",
190 				job->id, job->cs->sequence, job->cs->type,
191 				job->cs->ctx->asid, kref_read(&job->refcount),
192 				job->hw_queue_id);
193 		else
194 			seq_printf(s,
195 				"   %02d      0        0        %d          %d           %d\n",
196 				job->id, HL_KERNEL_ASID_ID,
197 				kref_read(&job->refcount), job->hw_queue_id);
198 	}
199 
200 	spin_unlock(&dev_entry->cs_job_spinlock);
201 
202 	if (!first)
203 		seq_puts(s, "\n");
204 
205 	return 0;
206 }
207 
208 static int userptr_show(struct seq_file *s, void *data)
209 {
210 	struct hl_debugfs_entry *entry = s->private;
211 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
212 	struct hl_userptr *userptr;
213 	char dma_dir[4][30] = {"DMA_BIDIRECTIONAL", "DMA_TO_DEVICE",
214 				"DMA_FROM_DEVICE", "DMA_NONE"};
215 	bool first = true;
216 
217 	spin_lock(&dev_entry->userptr_spinlock);
218 
219 	list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) {
220 		if (first) {
221 			first = false;
222 			seq_puts(s, "\n");
223 			seq_puts(s, " pid      user virtual address     size             dma dir\n");
224 			seq_puts(s, "----------------------------------------------------------\n");
225 		}
226 		seq_printf(s, " %-7d  0x%-14llx      %-10llu    %-30s\n",
227 				userptr->pid, userptr->addr, userptr->size,
228 				dma_dir[userptr->dir]);
229 	}
230 
231 	spin_unlock(&dev_entry->userptr_spinlock);
232 
233 	if (!first)
234 		seq_puts(s, "\n");
235 
236 	return 0;
237 }
238 
239 static int vm_show(struct seq_file *s, void *data)
240 {
241 	struct hl_debugfs_entry *entry = s->private;
242 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
243 	struct hl_vm_hw_block_list_node *lnode;
244 	struct hl_ctx *ctx;
245 	struct hl_vm *vm;
246 	struct hl_vm_hash_node *hnode;
247 	struct hl_userptr *userptr;
248 	struct hl_vm_phys_pg_pack *phys_pg_pack = NULL;
249 	struct hl_va_range *va_range;
250 	struct hl_vm_va_block *va_block;
251 	enum vm_type *vm_type;
252 	bool once = true;
253 	u64 j;
254 	int i;
255 
256 	mutex_lock(&dev_entry->ctx_mem_hash_mutex);
257 
258 	list_for_each_entry(ctx, &dev_entry->ctx_mem_hash_list, debugfs_list) {
259 		once = false;
260 		seq_puts(s, "\n\n----------------------------------------------------");
261 		seq_puts(s, "\n----------------------------------------------------\n\n");
262 		seq_printf(s, "ctx asid: %u\n", ctx->asid);
263 
264 		seq_puts(s, "\nmappings:\n\n");
265 		seq_puts(s, "    virtual address        size          handle\n");
266 		seq_puts(s, "----------------------------------------------------\n");
267 		mutex_lock(&ctx->mem_hash_lock);
268 		hash_for_each(ctx->mem_hash, i, hnode, node) {
269 			vm_type = hnode->ptr;
270 
271 			if (*vm_type == VM_TYPE_USERPTR) {
272 				userptr = hnode->ptr;
273 				seq_printf(s,
274 					"    0x%-14llx      %-10llu\n",
275 					hnode->vaddr, userptr->size);
276 			} else {
277 				phys_pg_pack = hnode->ptr;
278 				seq_printf(s,
279 					"    0x%-14llx      %-10llu       %-4u\n",
280 					hnode->vaddr, phys_pg_pack->total_size,
281 					phys_pg_pack->handle);
282 			}
283 		}
284 		mutex_unlock(&ctx->mem_hash_lock);
285 
286 		if (ctx->asid != HL_KERNEL_ASID_ID &&
287 		    !list_empty(&ctx->hw_block_mem_list)) {
288 			seq_puts(s, "\nhw_block mappings:\n\n");
289 			seq_puts(s,
290 				"    virtual address    block size    mapped size    HW block id\n");
291 			seq_puts(s,
292 				"---------------------------------------------------------------\n");
293 			mutex_lock(&ctx->hw_block_list_lock);
294 			list_for_each_entry(lnode, &ctx->hw_block_mem_list, node) {
295 				seq_printf(s,
296 					"    0x%-14lx   %-6u        %-6u             %-9u\n",
297 					lnode->vaddr, lnode->block_size, lnode->mapped_size,
298 					lnode->id);
299 			}
300 			mutex_unlock(&ctx->hw_block_list_lock);
301 		}
302 
303 		vm = &ctx->hdev->vm;
304 		spin_lock(&vm->idr_lock);
305 
306 		if (!idr_is_empty(&vm->phys_pg_pack_handles))
307 			seq_puts(s, "\n\nallocations:\n");
308 
309 		idr_for_each_entry(&vm->phys_pg_pack_handles, phys_pg_pack, i) {
310 			if (phys_pg_pack->asid != ctx->asid)
311 				continue;
312 
313 			seq_printf(s, "\nhandle: %u\n", phys_pg_pack->handle);
314 			seq_printf(s, "page size: %u\n\n",
315 						phys_pg_pack->page_size);
316 			seq_puts(s, "   physical address\n");
317 			seq_puts(s, "---------------------\n");
318 			for (j = 0 ; j < phys_pg_pack->npages ; j++) {
319 				seq_printf(s, "    0x%-14llx\n",
320 						phys_pg_pack->pages[j]);
321 			}
322 		}
323 		spin_unlock(&vm->idr_lock);
324 
325 	}
326 
327 	mutex_unlock(&dev_entry->ctx_mem_hash_mutex);
328 
329 	ctx = hl_get_compute_ctx(dev_entry->hdev);
330 	if (ctx) {
331 		seq_puts(s, "\nVA ranges:\n\n");
332 		for (i = HL_VA_RANGE_TYPE_HOST ; i < HL_VA_RANGE_TYPE_MAX ; ++i) {
333 			va_range = ctx->va_range[i];
334 			seq_printf(s, "   va_range %d\n", i);
335 			seq_puts(s, "---------------------\n");
336 			mutex_lock(&va_range->lock);
337 			list_for_each_entry(va_block, &va_range->list, node) {
338 				seq_printf(s, "%#16llx - %#16llx (%#llx)\n",
339 					   va_block->start, va_block->end,
340 					   va_block->size);
341 			}
342 			mutex_unlock(&va_range->lock);
343 			seq_puts(s, "\n");
344 		}
345 		hl_ctx_put(ctx);
346 	}
347 
348 	if (!once)
349 		seq_puts(s, "\n");
350 
351 	return 0;
352 }
353 
354 static int userptr_lookup_show(struct seq_file *s, void *data)
355 {
356 	struct hl_debugfs_entry *entry = s->private;
357 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
358 	struct scatterlist *sg;
359 	struct hl_userptr *userptr;
360 	bool first = true;
361 	u64 total_npages, npages, sg_start, sg_end;
362 	dma_addr_t dma_addr;
363 	int i;
364 
365 	spin_lock(&dev_entry->userptr_spinlock);
366 
367 	list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) {
368 		if (dev_entry->userptr_lookup >= userptr->addr &&
369 		dev_entry->userptr_lookup < userptr->addr + userptr->size) {
370 			total_npages = 0;
371 			for_each_sgtable_dma_sg(userptr->sgt, sg, i) {
372 				npages = hl_get_sg_info(sg, &dma_addr);
373 				sg_start = userptr->addr +
374 					total_npages * PAGE_SIZE;
375 				sg_end = userptr->addr +
376 					(total_npages + npages) * PAGE_SIZE;
377 
378 				if (dev_entry->userptr_lookup >= sg_start &&
379 				    dev_entry->userptr_lookup < sg_end) {
380 					dma_addr += (dev_entry->userptr_lookup -
381 							sg_start);
382 					if (first) {
383 						first = false;
384 						seq_puts(s, "\n");
385 						seq_puts(s, " user virtual address         dma address       pid        region start     region size\n");
386 						seq_puts(s, "---------------------------------------------------------------------------------------\n");
387 					}
388 					seq_printf(s, " 0x%-18llx  0x%-16llx  %-8u  0x%-16llx %-12llu\n",
389 						dev_entry->userptr_lookup,
390 						(u64)dma_addr, userptr->pid,
391 						userptr->addr, userptr->size);
392 				}
393 				total_npages += npages;
394 			}
395 		}
396 	}
397 
398 	spin_unlock(&dev_entry->userptr_spinlock);
399 
400 	if (!first)
401 		seq_puts(s, "\n");
402 
403 	return 0;
404 }
405 
406 static ssize_t userptr_lookup_write(struct file *file, const char __user *buf,
407 		size_t count, loff_t *f_pos)
408 {
409 	struct seq_file *s = file->private_data;
410 	struct hl_debugfs_entry *entry = s->private;
411 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
412 	ssize_t rc;
413 	u64 value;
414 
415 	rc = kstrtoull_from_user(buf, count, 16, &value);
416 	if (rc)
417 		return rc;
418 
419 	dev_entry->userptr_lookup = value;
420 
421 	return count;
422 }
423 
424 static int mmu_show(struct seq_file *s, void *data)
425 {
426 	struct hl_debugfs_entry *entry = s->private;
427 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
428 	struct hl_device *hdev = dev_entry->hdev;
429 	struct hl_ctx *ctx;
430 	struct hl_mmu_hop_info hops_info = {0};
431 	u64 virt_addr = dev_entry->mmu_addr, phys_addr;
432 	int i;
433 
434 	if (dev_entry->mmu_asid == HL_KERNEL_ASID_ID)
435 		ctx = hdev->kernel_ctx;
436 	else
437 		ctx = hl_get_compute_ctx(hdev);
438 
439 	if (!ctx) {
440 		dev_err(hdev->dev, "no ctx available\n");
441 		return 0;
442 	}
443 
444 	if (hl_mmu_get_tlb_info(ctx, virt_addr, &hops_info)) {
445 		dev_err(hdev->dev, "virt addr 0x%llx is not mapped to phys addr\n",
446 				virt_addr);
447 		goto put_ctx;
448 	}
449 
450 	hl_mmu_va_to_pa(ctx, virt_addr, &phys_addr);
451 
452 	if (hops_info.scrambled_vaddr &&
453 		(dev_entry->mmu_addr != hops_info.scrambled_vaddr))
454 		seq_printf(s,
455 			"asid: %u, virt_addr: 0x%llx, scrambled virt_addr: 0x%llx,\nphys_addr: 0x%llx, scrambled_phys_addr: 0x%llx\n",
456 			dev_entry->mmu_asid, dev_entry->mmu_addr,
457 			hops_info.scrambled_vaddr,
458 			hops_info.unscrambled_paddr, phys_addr);
459 	else
460 		seq_printf(s,
461 			"asid: %u, virt_addr: 0x%llx, phys_addr: 0x%llx\n",
462 			dev_entry->mmu_asid, dev_entry->mmu_addr, phys_addr);
463 
464 	for (i = 0 ; i < hops_info.used_hops ; i++) {
465 		seq_printf(s, "hop%d_addr: 0x%llx\n",
466 				i, hops_info.hop_info[i].hop_addr);
467 		seq_printf(s, "hop%d_pte_addr: 0x%llx\n",
468 				i, hops_info.hop_info[i].hop_pte_addr);
469 		seq_printf(s, "hop%d_pte: 0x%llx\n",
470 				i, hops_info.hop_info[i].hop_pte_val);
471 	}
472 
473 put_ctx:
474 	if (dev_entry->mmu_asid != HL_KERNEL_ASID_ID)
475 		hl_ctx_put(ctx);
476 
477 	return 0;
478 }
479 
480 static ssize_t mmu_asid_va_write(struct file *file, const char __user *buf,
481 		size_t count, loff_t *f_pos)
482 {
483 	struct seq_file *s = file->private_data;
484 	struct hl_debugfs_entry *entry = s->private;
485 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
486 	struct hl_device *hdev = dev_entry->hdev;
487 	char kbuf[MMU_KBUF_SIZE];
488 	char *c;
489 	ssize_t rc;
490 
491 	if (count > sizeof(kbuf) - 1)
492 		goto err;
493 	if (copy_from_user(kbuf, buf, count))
494 		goto err;
495 	kbuf[count] = 0;
496 
497 	c = strchr(kbuf, ' ');
498 	if (!c)
499 		goto err;
500 	*c = '\0';
501 
502 	rc = kstrtouint(kbuf, 10, &dev_entry->mmu_asid);
503 	if (rc)
504 		goto err;
505 
506 	if (strncmp(c+1, "0x", 2))
507 		goto err;
508 	rc = kstrtoull(c+3, 16, &dev_entry->mmu_addr);
509 	if (rc)
510 		goto err;
511 
512 	return count;
513 
514 err:
515 	dev_err(hdev->dev, "usage: echo <asid> <0xaddr> > mmu\n");
516 
517 	return -EINVAL;
518 }
519 
520 static int mmu_ack_error(struct seq_file *s, void *data)
521 {
522 	struct hl_debugfs_entry *entry = s->private;
523 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
524 	struct hl_device *hdev = dev_entry->hdev;
525 	int rc;
526 
527 	if (!dev_entry->mmu_cap_mask) {
528 		dev_err(hdev->dev, "mmu_cap_mask is not set\n");
529 		goto err;
530 	}
531 
532 	rc = hdev->asic_funcs->ack_mmu_errors(hdev, dev_entry->mmu_cap_mask);
533 	if (rc)
534 		goto err;
535 
536 	return 0;
537 err:
538 	return -EINVAL;
539 }
540 
541 static ssize_t mmu_ack_error_value_write(struct file *file,
542 		const char __user *buf,
543 		size_t count, loff_t *f_pos)
544 {
545 	struct seq_file *s = file->private_data;
546 	struct hl_debugfs_entry *entry = s->private;
547 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
548 	struct hl_device *hdev = dev_entry->hdev;
549 	char kbuf[MMU_KBUF_SIZE];
550 	ssize_t rc;
551 
552 	if (count > sizeof(kbuf) - 1)
553 		goto err;
554 
555 	if (copy_from_user(kbuf, buf, count))
556 		goto err;
557 
558 	kbuf[count] = 0;
559 
560 	if (strncmp(kbuf, "0x", 2))
561 		goto err;
562 
563 	rc = kstrtoull(kbuf, 16, &dev_entry->mmu_cap_mask);
564 	if (rc)
565 		goto err;
566 
567 	return count;
568 err:
569 	dev_err(hdev->dev, "usage: echo <0xmmu_cap_mask > > mmu_error\n");
570 
571 	return -EINVAL;
572 }
573 
574 static int engines_show(struct seq_file *s, void *data)
575 {
576 	struct hl_debugfs_entry *entry = s->private;
577 	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
578 	struct hl_device *hdev = dev_entry->hdev;
579 	struct engines_data eng_data;
580 
581 	if (hdev->reset_info.in_reset) {
582 		dev_warn_ratelimited(hdev->dev,
583 				"Can't check device idle during reset\n");
584 		return 0;
585 	}
586 
587 	eng_data.actual_size = 0;
588 	eng_data.allocated_buf_size = HL_ENGINES_DATA_MAX_SIZE;
589 	eng_data.buf = vmalloc(eng_data.allocated_buf_size);
590 	if (!eng_data.buf)
591 		return -ENOMEM;
592 
593 	hdev->asic_funcs->is_device_idle(hdev, NULL, 0, &eng_data);
594 
595 	if (eng_data.actual_size > eng_data.allocated_buf_size) {
596 		dev_err(hdev->dev,
597 				"Engines data size (%d Bytes) is bigger than allocated size (%u Bytes)\n",
598 				eng_data.actual_size, eng_data.allocated_buf_size);
599 		vfree(eng_data.buf);
600 		return -ENOMEM;
601 	}
602 
603 	seq_write(s, eng_data.buf, eng_data.actual_size);
604 
605 	vfree(eng_data.buf);
606 
607 	return 0;
608 }
609 
610 static ssize_t hl_memory_scrub(struct file *f, const char __user *buf,
611 					size_t count, loff_t *ppos)
612 {
613 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
614 	struct hl_device *hdev = entry->hdev;
615 	u64 val = hdev->memory_scrub_val;
616 	int rc;
617 
618 	if (!hl_device_operational(hdev, NULL)) {
619 		dev_warn_ratelimited(hdev->dev, "Can't scrub memory, device is not operational\n");
620 		return -EIO;
621 	}
622 
623 	mutex_lock(&hdev->fpriv_list_lock);
624 	if (hdev->is_compute_ctx_active) {
625 		mutex_unlock(&hdev->fpriv_list_lock);
626 		dev_err(hdev->dev, "can't scrub dram, context exist\n");
627 		return -EBUSY;
628 	}
629 	hdev->is_in_dram_scrub = true;
630 	mutex_unlock(&hdev->fpriv_list_lock);
631 
632 	rc = hdev->asic_funcs->scrub_device_dram(hdev, val);
633 
634 	mutex_lock(&hdev->fpriv_list_lock);
635 	hdev->is_in_dram_scrub = false;
636 	mutex_unlock(&hdev->fpriv_list_lock);
637 
638 	if (rc)
639 		return rc;
640 	return count;
641 }
642 
643 static bool hl_is_device_va(struct hl_device *hdev, u64 addr)
644 {
645 	struct asic_fixed_properties *prop = &hdev->asic_prop;
646 
647 	if (prop->dram_supports_virtual_memory &&
648 		(addr >= prop->dmmu.start_addr && addr < prop->dmmu.end_addr))
649 		return true;
650 
651 	if (addr >= prop->pmmu.start_addr &&
652 		addr < prop->pmmu.end_addr)
653 		return true;
654 
655 	if (addr >= prop->pmmu_huge.start_addr &&
656 		addr < prop->pmmu_huge.end_addr)
657 		return true;
658 
659 	return false;
660 }
661 
662 static bool hl_is_device_internal_memory_va(struct hl_device *hdev, u64 addr,
663 						u32 size)
664 {
665 	struct asic_fixed_properties *prop = &hdev->asic_prop;
666 	u64 dram_start_addr, dram_end_addr;
667 
668 	if (prop->dram_supports_virtual_memory) {
669 		dram_start_addr = prop->dmmu.start_addr;
670 		dram_end_addr = prop->dmmu.end_addr;
671 	} else {
672 		dram_start_addr = prop->dram_base_address;
673 		dram_end_addr = prop->dram_end_address;
674 	}
675 
676 	if (hl_mem_area_inside_range(addr, size, dram_start_addr,
677 					dram_end_addr))
678 		return true;
679 
680 	if (hl_mem_area_inside_range(addr, size, prop->sram_base_address,
681 					prop->sram_end_address))
682 		return true;
683 
684 	return false;
685 }
686 
687 static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size,
688 			u64 *phys_addr)
689 {
690 	struct hl_vm_phys_pg_pack *phys_pg_pack;
691 	struct hl_ctx *ctx;
692 	struct hl_vm_hash_node *hnode;
693 	u64 end_address, range_size;
694 	struct hl_userptr *userptr;
695 	enum vm_type *vm_type;
696 	bool valid = false;
697 	int i, rc = 0;
698 
699 	ctx = hl_get_compute_ctx(hdev);
700 
701 	if (!ctx) {
702 		dev_err(hdev->dev, "no ctx available\n");
703 		return -EINVAL;
704 	}
705 
706 	/* Verify address is mapped */
707 	mutex_lock(&ctx->mem_hash_lock);
708 	hash_for_each(ctx->mem_hash, i, hnode, node) {
709 		vm_type = hnode->ptr;
710 
711 		if (*vm_type == VM_TYPE_USERPTR) {
712 			userptr = hnode->ptr;
713 			range_size = userptr->size;
714 		} else {
715 			phys_pg_pack = hnode->ptr;
716 			range_size = phys_pg_pack->total_size;
717 		}
718 
719 		end_address = virt_addr + size;
720 		if ((virt_addr >= hnode->vaddr) &&
721 				(end_address <= hnode->vaddr + range_size)) {
722 			valid = true;
723 			break;
724 		}
725 	}
726 	mutex_unlock(&ctx->mem_hash_lock);
727 
728 	if (!valid) {
729 		dev_err(hdev->dev,
730 			"virt addr 0x%llx is not mapped\n",
731 			virt_addr);
732 		rc = -EINVAL;
733 		goto put_ctx;
734 	}
735 
736 	rc = hl_mmu_va_to_pa(ctx, virt_addr, phys_addr);
737 	if (rc) {
738 		dev_err(hdev->dev,
739 			"virt addr 0x%llx is not mapped to phys addr\n",
740 			virt_addr);
741 		rc = -EINVAL;
742 	}
743 
744 put_ctx:
745 	hl_ctx_put(ctx);
746 
747 	return rc;
748 }
749 
750 static int hl_access_dev_mem_by_region(struct hl_device *hdev, u64 addr,
751 		u64 *val, enum debugfs_access_type acc_type, bool *found)
752 {
753 	size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ?
754 		sizeof(u64) : sizeof(u32);
755 	struct pci_mem_region *mem_reg;
756 	int i;
757 
758 	for (i = 0; i < PCI_REGION_NUMBER; i++) {
759 		mem_reg = &hdev->pci_mem_region[i];
760 		if (!mem_reg->used)
761 			continue;
762 		if (addr >= mem_reg->region_base &&
763 			addr <= mem_reg->region_base + mem_reg->region_size - acc_size) {
764 			*found = true;
765 			return hdev->asic_funcs->access_dev_mem(hdev, i, addr, val, acc_type);
766 		}
767 	}
768 	return 0;
769 }
770 
771 static void hl_access_host_mem(struct hl_device *hdev, u64 addr, u64 *val,
772 		enum debugfs_access_type acc_type)
773 {
774 	struct asic_fixed_properties *prop = &hdev->asic_prop;
775 	u64 offset = prop->device_dma_offset_for_host_access;
776 
777 	switch (acc_type) {
778 	case DEBUGFS_READ32:
779 		*val = *(u32 *) phys_to_virt(addr - offset);
780 		break;
781 	case DEBUGFS_WRITE32:
782 		*(u32 *) phys_to_virt(addr - offset) = *val;
783 		break;
784 	case DEBUGFS_READ64:
785 		*val = *(u64 *) phys_to_virt(addr - offset);
786 		break;
787 	case DEBUGFS_WRITE64:
788 		*(u64 *) phys_to_virt(addr - offset) = *val;
789 		break;
790 	default:
791 		dev_err(hdev->dev, "hostmem access-type %d id not supported\n", acc_type);
792 		break;
793 	}
794 }
795 
796 static int hl_access_mem(struct hl_device *hdev, u64 addr, u64 *val,
797 				enum debugfs_access_type acc_type)
798 {
799 	size_t acc_size = (acc_type == DEBUGFS_READ64 || acc_type == DEBUGFS_WRITE64) ?
800 		sizeof(u64) : sizeof(u32);
801 	u64 host_start = hdev->asic_prop.host_base_address;
802 	u64 host_end = hdev->asic_prop.host_end_address;
803 	bool user_address, found = false;
804 	int rc;
805 
806 	user_address = hl_is_device_va(hdev, addr);
807 	if (user_address) {
808 		rc = device_va_to_pa(hdev, addr, acc_size, &addr);
809 		if (rc)
810 			return rc;
811 	}
812 
813 	rc = hl_access_dev_mem_by_region(hdev, addr, val, acc_type, &found);
814 	if (rc) {
815 		dev_err(hdev->dev,
816 			"Failed reading addr %#llx from dev mem (%d)\n",
817 			addr, rc);
818 		return rc;
819 	}
820 
821 	if (found)
822 		return 0;
823 
824 	if (!user_address || device_iommu_mapped(&hdev->pdev->dev)) {
825 		rc = -EINVAL;
826 		goto err;
827 	}
828 
829 	if (addr >= host_start && addr <= host_end - acc_size) {
830 		hl_access_host_mem(hdev, addr, val, acc_type);
831 	} else {
832 		rc = -EINVAL;
833 		goto err;
834 	}
835 
836 	return 0;
837 err:
838 	dev_err(hdev->dev, "invalid addr %#llx\n", addr);
839 	return rc;
840 }
841 
842 static ssize_t hl_data_read32(struct file *f, char __user *buf,
843 					size_t count, loff_t *ppos)
844 {
845 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
846 	struct hl_device *hdev = entry->hdev;
847 	u64 value64, addr = entry->addr;
848 	char tmp_buf[32];
849 	ssize_t rc;
850 	u32 val;
851 
852 	if (hdev->reset_info.in_reset) {
853 		dev_warn_ratelimited(hdev->dev, "Can't read during reset\n");
854 		return 0;
855 	}
856 
857 	if (*ppos)
858 		return 0;
859 
860 	rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_READ32);
861 	if (rc)
862 		return rc;
863 
864 	val = value64; /* downcast back to 32 */
865 
866 	sprintf(tmp_buf, "0x%08x\n", val);
867 	return simple_read_from_buffer(buf, count, ppos, tmp_buf,
868 			strlen(tmp_buf));
869 }
870 
871 static ssize_t hl_data_write32(struct file *f, const char __user *buf,
872 					size_t count, loff_t *ppos)
873 {
874 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
875 	struct hl_device *hdev = entry->hdev;
876 	u64 value64, addr = entry->addr;
877 	u32 value;
878 	ssize_t rc;
879 
880 	if (hdev->reset_info.in_reset) {
881 		dev_warn_ratelimited(hdev->dev, "Can't write during reset\n");
882 		return 0;
883 	}
884 
885 	rc = kstrtouint_from_user(buf, count, 16, &value);
886 	if (rc)
887 		return rc;
888 
889 	value64 = value;
890 	rc = hl_access_mem(hdev, addr, &value64, DEBUGFS_WRITE32);
891 	if (rc)
892 		return rc;
893 
894 	return count;
895 }
896 
897 static ssize_t hl_data_read64(struct file *f, char __user *buf,
898 					size_t count, loff_t *ppos)
899 {
900 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
901 	struct hl_device *hdev = entry->hdev;
902 	u64 addr = entry->addr;
903 	char tmp_buf[32];
904 	ssize_t rc;
905 	u64 val;
906 
907 	if (hdev->reset_info.in_reset) {
908 		dev_warn_ratelimited(hdev->dev, "Can't read during reset\n");
909 		return 0;
910 	}
911 
912 	if (*ppos)
913 		return 0;
914 
915 	rc = hl_access_mem(hdev, addr, &val, DEBUGFS_READ64);
916 	if (rc)
917 		return rc;
918 
919 	sprintf(tmp_buf, "0x%016llx\n", val);
920 	return simple_read_from_buffer(buf, count, ppos, tmp_buf,
921 			strlen(tmp_buf));
922 }
923 
924 static ssize_t hl_data_write64(struct file *f, const char __user *buf,
925 					size_t count, loff_t *ppos)
926 {
927 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
928 	struct hl_device *hdev = entry->hdev;
929 	u64 addr = entry->addr;
930 	u64 value;
931 	ssize_t rc;
932 
933 	if (hdev->reset_info.in_reset) {
934 		dev_warn_ratelimited(hdev->dev, "Can't write during reset\n");
935 		return 0;
936 	}
937 
938 	rc = kstrtoull_from_user(buf, count, 16, &value);
939 	if (rc)
940 		return rc;
941 
942 	rc = hl_access_mem(hdev, addr, &value, DEBUGFS_WRITE64);
943 	if (rc)
944 		return rc;
945 
946 	return count;
947 }
948 
949 static ssize_t hl_dma_size_write(struct file *f, const char __user *buf,
950 					size_t count, loff_t *ppos)
951 {
952 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
953 	struct hl_device *hdev = entry->hdev;
954 	u64 addr = entry->addr;
955 	ssize_t rc;
956 	u32 size;
957 
958 	if (hdev->reset_info.in_reset) {
959 		dev_warn_ratelimited(hdev->dev, "Can't DMA during reset\n");
960 		return 0;
961 	}
962 	rc = kstrtouint_from_user(buf, count, 16, &size);
963 	if (rc)
964 		return rc;
965 
966 	if (!size) {
967 		dev_err(hdev->dev, "DMA read failed. size can't be 0\n");
968 		return -EINVAL;
969 	}
970 
971 	if (size > SZ_128M) {
972 		dev_err(hdev->dev,
973 			"DMA read failed. size can't be larger than 128MB\n");
974 		return -EINVAL;
975 	}
976 
977 	if (!hl_is_device_internal_memory_va(hdev, addr, size)) {
978 		dev_err(hdev->dev,
979 			"DMA read failed. Invalid 0x%010llx + 0x%08x\n",
980 			addr, size);
981 		return -EINVAL;
982 	}
983 
984 	/* Free the previous allocation, if there was any */
985 	entry->data_dma_blob_desc.size = 0;
986 	vfree(entry->data_dma_blob_desc.data);
987 
988 	entry->data_dma_blob_desc.data = vmalloc(size);
989 	if (!entry->data_dma_blob_desc.data)
990 		return -ENOMEM;
991 
992 	rc = hdev->asic_funcs->debugfs_read_dma(hdev, addr, size,
993 						entry->data_dma_blob_desc.data);
994 	if (rc) {
995 		dev_err(hdev->dev, "Failed to DMA from 0x%010llx\n", addr);
996 		vfree(entry->data_dma_blob_desc.data);
997 		entry->data_dma_blob_desc.data = NULL;
998 		return -EIO;
999 	}
1000 
1001 	entry->data_dma_blob_desc.size = size;
1002 
1003 	return count;
1004 }
1005 
1006 static ssize_t hl_monitor_dump_trigger(struct file *f, const char __user *buf,
1007 					size_t count, loff_t *ppos)
1008 {
1009 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1010 	struct hl_device *hdev = entry->hdev;
1011 	u32 size, trig;
1012 	ssize_t rc;
1013 
1014 	if (hdev->reset_info.in_reset) {
1015 		dev_warn_ratelimited(hdev->dev, "Can't dump monitors during reset\n");
1016 		return 0;
1017 	}
1018 	rc = kstrtouint_from_user(buf, count, 10, &trig);
1019 	if (rc)
1020 		return rc;
1021 
1022 	if (trig != 1) {
1023 		dev_err(hdev->dev, "Must write 1 to trigger monitor dump\n");
1024 		return -EINVAL;
1025 	}
1026 
1027 	size = sizeof(struct cpucp_monitor_dump);
1028 
1029 	/* Free the previous allocation, if there was any */
1030 	entry->mon_dump_blob_desc.size = 0;
1031 	vfree(entry->mon_dump_blob_desc.data);
1032 
1033 	entry->mon_dump_blob_desc.data = vmalloc(size);
1034 	if (!entry->mon_dump_blob_desc.data)
1035 		return -ENOMEM;
1036 
1037 	rc = hdev->asic_funcs->get_monitor_dump(hdev, entry->mon_dump_blob_desc.data);
1038 	if (rc) {
1039 		dev_err(hdev->dev, "Failed to dump monitors\n");
1040 		vfree(entry->mon_dump_blob_desc.data);
1041 		entry->mon_dump_blob_desc.data = NULL;
1042 		return -EIO;
1043 	}
1044 
1045 	entry->mon_dump_blob_desc.size = size;
1046 
1047 	return count;
1048 }
1049 
1050 static ssize_t hl_get_power_state(struct file *f, char __user *buf,
1051 		size_t count, loff_t *ppos)
1052 {
1053 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1054 	struct hl_device *hdev = entry->hdev;
1055 	char tmp_buf[200];
1056 	int i;
1057 
1058 	if (*ppos)
1059 		return 0;
1060 
1061 	if (hdev->pdev->current_state == PCI_D0)
1062 		i = 1;
1063 	else if (hdev->pdev->current_state == PCI_D3hot)
1064 		i = 2;
1065 	else
1066 		i = 3;
1067 
1068 	sprintf(tmp_buf,
1069 		"current power state: %d\n1 - D0\n2 - D3hot\n3 - Unknown\n", i);
1070 	return simple_read_from_buffer(buf, count, ppos, tmp_buf,
1071 			strlen(tmp_buf));
1072 }
1073 
1074 static ssize_t hl_set_power_state(struct file *f, const char __user *buf,
1075 					size_t count, loff_t *ppos)
1076 {
1077 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1078 	struct hl_device *hdev = entry->hdev;
1079 	u32 value;
1080 	ssize_t rc;
1081 
1082 	rc = kstrtouint_from_user(buf, count, 10, &value);
1083 	if (rc)
1084 		return rc;
1085 
1086 	if (value == 1) {
1087 		pci_set_power_state(hdev->pdev, PCI_D0);
1088 		pci_restore_state(hdev->pdev);
1089 		rc = pci_enable_device(hdev->pdev);
1090 		if (rc < 0)
1091 			return rc;
1092 	} else if (value == 2) {
1093 		pci_save_state(hdev->pdev);
1094 		pci_disable_device(hdev->pdev);
1095 		pci_set_power_state(hdev->pdev, PCI_D3hot);
1096 	} else {
1097 		dev_dbg(hdev->dev, "invalid power state value %u\n", value);
1098 		return -EINVAL;
1099 	}
1100 
1101 	return count;
1102 }
1103 
1104 static ssize_t hl_i2c_data_read(struct file *f, char __user *buf,
1105 					size_t count, loff_t *ppos)
1106 {
1107 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1108 	struct hl_device *hdev = entry->hdev;
1109 	char tmp_buf[32];
1110 	u64 val;
1111 	ssize_t rc;
1112 
1113 	if (*ppos)
1114 		return 0;
1115 
1116 	rc = hl_debugfs_i2c_read(hdev, entry->i2c_bus, entry->i2c_addr,
1117 			entry->i2c_reg, entry->i2c_len, &val);
1118 	if (rc) {
1119 		dev_err(hdev->dev,
1120 			"Failed to read from I2C bus %d, addr %d, reg %d, len %d\n",
1121 			entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len);
1122 		return rc;
1123 	}
1124 
1125 	sprintf(tmp_buf, "%#02llx\n", val);
1126 	rc = simple_read_from_buffer(buf, count, ppos, tmp_buf,
1127 			strlen(tmp_buf));
1128 
1129 	return rc;
1130 }
1131 
1132 static ssize_t hl_i2c_data_write(struct file *f, const char __user *buf,
1133 					size_t count, loff_t *ppos)
1134 {
1135 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1136 	struct hl_device *hdev = entry->hdev;
1137 	u64 value;
1138 	ssize_t rc;
1139 
1140 	rc = kstrtou64_from_user(buf, count, 16, &value);
1141 	if (rc)
1142 		return rc;
1143 
1144 	rc = hl_debugfs_i2c_write(hdev, entry->i2c_bus, entry->i2c_addr,
1145 			entry->i2c_reg, entry->i2c_len, value);
1146 	if (rc) {
1147 		dev_err(hdev->dev,
1148 			"Failed to write %#02llx to I2C bus %d, addr %d, reg %d, len %d\n",
1149 			value, entry->i2c_bus, entry->i2c_addr, entry->i2c_reg, entry->i2c_len);
1150 		return rc;
1151 	}
1152 
1153 	return count;
1154 }
1155 
1156 static ssize_t hl_led0_write(struct file *f, const char __user *buf,
1157 					size_t count, loff_t *ppos)
1158 {
1159 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1160 	struct hl_device *hdev = entry->hdev;
1161 	u32 value;
1162 	ssize_t rc;
1163 
1164 	rc = kstrtouint_from_user(buf, count, 10, &value);
1165 	if (rc)
1166 		return rc;
1167 
1168 	value = value ? 1 : 0;
1169 
1170 	hl_debugfs_led_set(hdev, 0, value);
1171 
1172 	return count;
1173 }
1174 
1175 static ssize_t hl_led1_write(struct file *f, const char __user *buf,
1176 					size_t count, loff_t *ppos)
1177 {
1178 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1179 	struct hl_device *hdev = entry->hdev;
1180 	u32 value;
1181 	ssize_t rc;
1182 
1183 	rc = kstrtouint_from_user(buf, count, 10, &value);
1184 	if (rc)
1185 		return rc;
1186 
1187 	value = value ? 1 : 0;
1188 
1189 	hl_debugfs_led_set(hdev, 1, value);
1190 
1191 	return count;
1192 }
1193 
1194 static ssize_t hl_led2_write(struct file *f, const char __user *buf,
1195 					size_t count, loff_t *ppos)
1196 {
1197 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1198 	struct hl_device *hdev = entry->hdev;
1199 	u32 value;
1200 	ssize_t rc;
1201 
1202 	rc = kstrtouint_from_user(buf, count, 10, &value);
1203 	if (rc)
1204 		return rc;
1205 
1206 	value = value ? 1 : 0;
1207 
1208 	hl_debugfs_led_set(hdev, 2, value);
1209 
1210 	return count;
1211 }
1212 
1213 static ssize_t hl_device_read(struct file *f, char __user *buf,
1214 					size_t count, loff_t *ppos)
1215 {
1216 	static const char *help =
1217 		"Valid values: disable, enable, suspend, resume, cpu_timeout\n";
1218 	return simple_read_from_buffer(buf, count, ppos, help, strlen(help));
1219 }
1220 
1221 static ssize_t hl_device_write(struct file *f, const char __user *buf,
1222 				     size_t count, loff_t *ppos)
1223 {
1224 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1225 	struct hl_device *hdev = entry->hdev;
1226 	char data[30] = {0};
1227 
1228 	/* don't allow partial writes */
1229 	if (*ppos != 0)
1230 		return 0;
1231 
1232 	simple_write_to_buffer(data, 29, ppos, buf, count);
1233 
1234 	if (strncmp("disable", data, strlen("disable")) == 0) {
1235 		hdev->disabled = true;
1236 	} else if (strncmp("enable", data, strlen("enable")) == 0) {
1237 		hdev->disabled = false;
1238 	} else if (strncmp("suspend", data, strlen("suspend")) == 0) {
1239 		hdev->asic_funcs->suspend(hdev);
1240 	} else if (strncmp("resume", data, strlen("resume")) == 0) {
1241 		hdev->asic_funcs->resume(hdev);
1242 	} else if (strncmp("cpu_timeout", data, strlen("cpu_timeout")) == 0) {
1243 		hdev->device_cpu_disabled = true;
1244 	} else {
1245 		dev_err(hdev->dev,
1246 			"Valid values: disable, enable, suspend, resume, cpu_timeout\n");
1247 		count = -EINVAL;
1248 	}
1249 
1250 	return count;
1251 }
1252 
1253 static ssize_t hl_clk_gate_read(struct file *f, char __user *buf,
1254 					size_t count, loff_t *ppos)
1255 {
1256 	return 0;
1257 }
1258 
1259 static ssize_t hl_clk_gate_write(struct file *f, const char __user *buf,
1260 				     size_t count, loff_t *ppos)
1261 {
1262 	return count;
1263 }
1264 
1265 static ssize_t hl_stop_on_err_read(struct file *f, char __user *buf,
1266 					size_t count, loff_t *ppos)
1267 {
1268 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1269 	struct hl_device *hdev = entry->hdev;
1270 	char tmp_buf[200];
1271 	ssize_t rc;
1272 
1273 	if (!hdev->asic_prop.configurable_stop_on_err)
1274 		return -EOPNOTSUPP;
1275 
1276 	if (*ppos)
1277 		return 0;
1278 
1279 	sprintf(tmp_buf, "%d\n", hdev->stop_on_err);
1280 	rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf,
1281 			strlen(tmp_buf) + 1);
1282 
1283 	return rc;
1284 }
1285 
1286 static ssize_t hl_stop_on_err_write(struct file *f, const char __user *buf,
1287 				     size_t count, loff_t *ppos)
1288 {
1289 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1290 	struct hl_device *hdev = entry->hdev;
1291 	u32 value;
1292 	ssize_t rc;
1293 
1294 	if (!hdev->asic_prop.configurable_stop_on_err)
1295 		return -EOPNOTSUPP;
1296 
1297 	if (hdev->reset_info.in_reset) {
1298 		dev_warn_ratelimited(hdev->dev,
1299 				"Can't change stop on error during reset\n");
1300 		return 0;
1301 	}
1302 
1303 	rc = kstrtouint_from_user(buf, count, 10, &value);
1304 	if (rc)
1305 		return rc;
1306 
1307 	hdev->stop_on_err = value ? 1 : 0;
1308 
1309 	hl_device_reset(hdev, 0);
1310 
1311 	return count;
1312 }
1313 
1314 static ssize_t hl_security_violations_read(struct file *f, char __user *buf,
1315 					size_t count, loff_t *ppos)
1316 {
1317 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1318 	struct hl_device *hdev = entry->hdev;
1319 
1320 	hdev->asic_funcs->ack_protection_bits_errors(hdev);
1321 
1322 	return 0;
1323 }
1324 
1325 static ssize_t hl_state_dump_read(struct file *f, char __user *buf,
1326 					size_t count, loff_t *ppos)
1327 {
1328 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1329 	ssize_t rc;
1330 
1331 	down_read(&entry->state_dump_sem);
1332 	if (!entry->state_dump[entry->state_dump_head])
1333 		rc = 0;
1334 	else
1335 		rc = simple_read_from_buffer(
1336 			buf, count, ppos,
1337 			entry->state_dump[entry->state_dump_head],
1338 			strlen(entry->state_dump[entry->state_dump_head]));
1339 	up_read(&entry->state_dump_sem);
1340 
1341 	return rc;
1342 }
1343 
1344 static ssize_t hl_state_dump_write(struct file *f, const char __user *buf,
1345 					size_t count, loff_t *ppos)
1346 {
1347 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1348 	struct hl_device *hdev = entry->hdev;
1349 	ssize_t rc;
1350 	u32 size;
1351 	int i;
1352 
1353 	rc = kstrtouint_from_user(buf, count, 10, &size);
1354 	if (rc)
1355 		return rc;
1356 
1357 	if (size <= 0 || size >= ARRAY_SIZE(entry->state_dump)) {
1358 		dev_err(hdev->dev, "Invalid number of dumps to skip\n");
1359 		return -EINVAL;
1360 	}
1361 
1362 	if (entry->state_dump[entry->state_dump_head]) {
1363 		down_write(&entry->state_dump_sem);
1364 		for (i = 0; i < size; ++i) {
1365 			vfree(entry->state_dump[entry->state_dump_head]);
1366 			entry->state_dump[entry->state_dump_head] = NULL;
1367 			if (entry->state_dump_head > 0)
1368 				entry->state_dump_head--;
1369 			else
1370 				entry->state_dump_head =
1371 					ARRAY_SIZE(entry->state_dump) - 1;
1372 		}
1373 		up_write(&entry->state_dump_sem);
1374 	}
1375 
1376 	return count;
1377 }
1378 
1379 static ssize_t hl_timeout_locked_read(struct file *f, char __user *buf,
1380 					size_t count, loff_t *ppos)
1381 {
1382 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1383 	struct hl_device *hdev = entry->hdev;
1384 	char tmp_buf[200];
1385 	ssize_t rc;
1386 
1387 	if (*ppos)
1388 		return 0;
1389 
1390 	sprintf(tmp_buf, "%d\n",
1391 		jiffies_to_msecs(hdev->timeout_jiffies) / 1000);
1392 	rc = simple_read_from_buffer(buf, strlen(tmp_buf) + 1, ppos, tmp_buf,
1393 			strlen(tmp_buf) + 1);
1394 
1395 	return rc;
1396 }
1397 
1398 static ssize_t hl_timeout_locked_write(struct file *f, const char __user *buf,
1399 				     size_t count, loff_t *ppos)
1400 {
1401 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1402 	struct hl_device *hdev = entry->hdev;
1403 	u32 value;
1404 	ssize_t rc;
1405 
1406 	rc = kstrtouint_from_user(buf, count, 10, &value);
1407 	if (rc)
1408 		return rc;
1409 
1410 	if (value)
1411 		hdev->timeout_jiffies = msecs_to_jiffies(value * 1000);
1412 	else
1413 		hdev->timeout_jiffies = MAX_SCHEDULE_TIMEOUT;
1414 
1415 	return count;
1416 }
1417 
1418 static ssize_t hl_check_razwi_happened(struct file *f, char __user *buf,
1419 					size_t count, loff_t *ppos)
1420 {
1421 	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
1422 	struct hl_device *hdev = entry->hdev;
1423 
1424 	hdev->asic_funcs->check_if_razwi_happened(hdev);
1425 
1426 	return 0;
1427 }
1428 
1429 static const struct file_operations hl_mem_scrub_fops = {
1430 	.owner = THIS_MODULE,
1431 	.write = hl_memory_scrub,
1432 };
1433 
1434 static const struct file_operations hl_data32b_fops = {
1435 	.owner = THIS_MODULE,
1436 	.read = hl_data_read32,
1437 	.write = hl_data_write32
1438 };
1439 
1440 static const struct file_operations hl_data64b_fops = {
1441 	.owner = THIS_MODULE,
1442 	.read = hl_data_read64,
1443 	.write = hl_data_write64
1444 };
1445 
1446 static const struct file_operations hl_dma_size_fops = {
1447 	.owner = THIS_MODULE,
1448 	.write = hl_dma_size_write
1449 };
1450 
1451 static const struct file_operations hl_monitor_dump_fops = {
1452 	.owner = THIS_MODULE,
1453 	.write = hl_monitor_dump_trigger
1454 };
1455 
1456 static const struct file_operations hl_i2c_data_fops = {
1457 	.owner = THIS_MODULE,
1458 	.read = hl_i2c_data_read,
1459 	.write = hl_i2c_data_write
1460 };
1461 
1462 static const struct file_operations hl_power_fops = {
1463 	.owner = THIS_MODULE,
1464 	.read = hl_get_power_state,
1465 	.write = hl_set_power_state
1466 };
1467 
1468 static const struct file_operations hl_led0_fops = {
1469 	.owner = THIS_MODULE,
1470 	.write = hl_led0_write
1471 };
1472 
1473 static const struct file_operations hl_led1_fops = {
1474 	.owner = THIS_MODULE,
1475 	.write = hl_led1_write
1476 };
1477 
1478 static const struct file_operations hl_led2_fops = {
1479 	.owner = THIS_MODULE,
1480 	.write = hl_led2_write
1481 };
1482 
1483 static const struct file_operations hl_device_fops = {
1484 	.owner = THIS_MODULE,
1485 	.read = hl_device_read,
1486 	.write = hl_device_write
1487 };
1488 
1489 static const struct file_operations hl_clk_gate_fops = {
1490 	.owner = THIS_MODULE,
1491 	.read = hl_clk_gate_read,
1492 	.write = hl_clk_gate_write
1493 };
1494 
1495 static const struct file_operations hl_stop_on_err_fops = {
1496 	.owner = THIS_MODULE,
1497 	.read = hl_stop_on_err_read,
1498 	.write = hl_stop_on_err_write
1499 };
1500 
1501 static const struct file_operations hl_security_violations_fops = {
1502 	.owner = THIS_MODULE,
1503 	.read = hl_security_violations_read
1504 };
1505 
1506 static const struct file_operations hl_state_dump_fops = {
1507 	.owner = THIS_MODULE,
1508 	.read = hl_state_dump_read,
1509 	.write = hl_state_dump_write
1510 };
1511 
1512 static const struct file_operations hl_timeout_locked_fops = {
1513 	.owner = THIS_MODULE,
1514 	.read = hl_timeout_locked_read,
1515 	.write = hl_timeout_locked_write
1516 };
1517 
1518 static const struct file_operations hl_razwi_check_fops = {
1519 	.owner = THIS_MODULE,
1520 	.read = hl_check_razwi_happened
1521 };
1522 
1523 static const struct hl_info_list hl_debugfs_list[] = {
1524 	{"command_buffers", command_buffers_show, NULL},
1525 	{"command_submission", command_submission_show, NULL},
1526 	{"command_submission_jobs", command_submission_jobs_show, NULL},
1527 	{"userptr", userptr_show, NULL},
1528 	{"vm", vm_show, NULL},
1529 	{"userptr_lookup", userptr_lookup_show, userptr_lookup_write},
1530 	{"mmu", mmu_show, mmu_asid_va_write},
1531 	{"mmu_error", mmu_ack_error, mmu_ack_error_value_write},
1532 	{"engines", engines_show, NULL},
1533 };
1534 
1535 static int hl_debugfs_open(struct inode *inode, struct file *file)
1536 {
1537 	struct hl_debugfs_entry *node = inode->i_private;
1538 
1539 	return single_open(file, node->info_ent->show, node);
1540 }
1541 
1542 static ssize_t hl_debugfs_write(struct file *file, const char __user *buf,
1543 		size_t count, loff_t *f_pos)
1544 {
1545 	struct hl_debugfs_entry *node = file->f_inode->i_private;
1546 
1547 	if (node->info_ent->write)
1548 		return node->info_ent->write(file, buf, count, f_pos);
1549 	else
1550 		return -EINVAL;
1551 
1552 }
1553 
1554 static const struct file_operations hl_debugfs_fops = {
1555 	.owner = THIS_MODULE,
1556 	.open = hl_debugfs_open,
1557 	.read = seq_read,
1558 	.write = hl_debugfs_write,
1559 	.llseek = seq_lseek,
1560 	.release = single_release,
1561 };
1562 
1563 static void add_secured_nodes(struct hl_dbg_device_entry *dev_entry, struct dentry *root)
1564 {
1565 	debugfs_create_u8("i2c_bus",
1566 				0644,
1567 				root,
1568 				&dev_entry->i2c_bus);
1569 
1570 	debugfs_create_u8("i2c_addr",
1571 				0644,
1572 				root,
1573 				&dev_entry->i2c_addr);
1574 
1575 	debugfs_create_u8("i2c_reg",
1576 				0644,
1577 				root,
1578 				&dev_entry->i2c_reg);
1579 
1580 	debugfs_create_u8("i2c_len",
1581 				0644,
1582 				root,
1583 				&dev_entry->i2c_len);
1584 
1585 	debugfs_create_file("i2c_data",
1586 				0644,
1587 				root,
1588 				dev_entry,
1589 				&hl_i2c_data_fops);
1590 
1591 	debugfs_create_file("led0",
1592 				0200,
1593 				root,
1594 				dev_entry,
1595 				&hl_led0_fops);
1596 
1597 	debugfs_create_file("led1",
1598 				0200,
1599 				root,
1600 				dev_entry,
1601 				&hl_led1_fops);
1602 
1603 	debugfs_create_file("led2",
1604 				0200,
1605 				root,
1606 				dev_entry,
1607 				&hl_led2_fops);
1608 }
1609 
1610 static void add_files_to_device(struct hl_device *hdev, struct hl_dbg_device_entry *dev_entry,
1611 				struct dentry *root)
1612 {
1613 	int count = ARRAY_SIZE(hl_debugfs_list);
1614 	struct hl_debugfs_entry *entry;
1615 	int i;
1616 
1617 	debugfs_create_x64("memory_scrub_val",
1618 				0644,
1619 				root,
1620 				&hdev->memory_scrub_val);
1621 
1622 	debugfs_create_file("memory_scrub",
1623 				0200,
1624 				root,
1625 				dev_entry,
1626 				&hl_mem_scrub_fops);
1627 
1628 	debugfs_create_x64("addr",
1629 				0644,
1630 				root,
1631 				&dev_entry->addr);
1632 
1633 	debugfs_create_file("data32",
1634 				0644,
1635 				root,
1636 				dev_entry,
1637 				&hl_data32b_fops);
1638 
1639 	debugfs_create_file("data64",
1640 				0644,
1641 				root,
1642 				dev_entry,
1643 				&hl_data64b_fops);
1644 
1645 	debugfs_create_file("set_power_state",
1646 				0200,
1647 				root,
1648 				dev_entry,
1649 				&hl_power_fops);
1650 
1651 	debugfs_create_file("device",
1652 				0200,
1653 				root,
1654 				dev_entry,
1655 				&hl_device_fops);
1656 
1657 	debugfs_create_file("clk_gate",
1658 				0200,
1659 				root,
1660 				dev_entry,
1661 				&hl_clk_gate_fops);
1662 
1663 	debugfs_create_file("stop_on_err",
1664 				0644,
1665 				root,
1666 				dev_entry,
1667 				&hl_stop_on_err_fops);
1668 
1669 	debugfs_create_file("dump_security_violations",
1670 				0644,
1671 				root,
1672 				dev_entry,
1673 				&hl_security_violations_fops);
1674 
1675 	debugfs_create_file("dump_razwi_events",
1676 				0644,
1677 				root,
1678 				dev_entry,
1679 				&hl_razwi_check_fops);
1680 
1681 	debugfs_create_file("dma_size",
1682 				0200,
1683 				root,
1684 				dev_entry,
1685 				&hl_dma_size_fops);
1686 
1687 	debugfs_create_blob("data_dma",
1688 				0400,
1689 				root,
1690 				&dev_entry->data_dma_blob_desc);
1691 
1692 	debugfs_create_file("monitor_dump_trig",
1693 				0200,
1694 				root,
1695 				dev_entry,
1696 				&hl_monitor_dump_fops);
1697 
1698 	debugfs_create_blob("monitor_dump",
1699 				0400,
1700 				root,
1701 				&dev_entry->mon_dump_blob_desc);
1702 
1703 	debugfs_create_x8("skip_reset_on_timeout",
1704 				0644,
1705 				root,
1706 				&hdev->reset_info.skip_reset_on_timeout);
1707 
1708 	debugfs_create_file("state_dump",
1709 				0600,
1710 				root,
1711 				dev_entry,
1712 				&hl_state_dump_fops);
1713 
1714 	debugfs_create_file("timeout_locked",
1715 				0644,
1716 				root,
1717 				dev_entry,
1718 				&hl_timeout_locked_fops);
1719 
1720 	debugfs_create_u32("device_release_watchdog_timeout",
1721 				0644,
1722 				root,
1723 				&hdev->device_release_watchdog_timeout_sec);
1724 
1725 	for (i = 0, entry = dev_entry->entry_arr ; i < count ; i++, entry++) {
1726 		debugfs_create_file(hl_debugfs_list[i].name,
1727 					0444,
1728 					root,
1729 					entry,
1730 					&hl_debugfs_fops);
1731 		entry->info_ent = &hl_debugfs_list[i];
1732 		entry->dev_entry = dev_entry;
1733 	}
1734 }
1735 
1736 int hl_debugfs_device_init(struct hl_device *hdev)
1737 {
1738 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1739 	int count = ARRAY_SIZE(hl_debugfs_list);
1740 
1741 	dev_entry->hdev = hdev;
1742 	dev_entry->entry_arr = kmalloc_array(count, sizeof(struct hl_debugfs_entry), GFP_KERNEL);
1743 	if (!dev_entry->entry_arr)
1744 		return -ENOMEM;
1745 
1746 	dev_entry->data_dma_blob_desc.size = 0;
1747 	dev_entry->data_dma_blob_desc.data = NULL;
1748 	dev_entry->mon_dump_blob_desc.size = 0;
1749 	dev_entry->mon_dump_blob_desc.data = NULL;
1750 
1751 	INIT_LIST_HEAD(&dev_entry->file_list);
1752 	INIT_LIST_HEAD(&dev_entry->cb_list);
1753 	INIT_LIST_HEAD(&dev_entry->cs_list);
1754 	INIT_LIST_HEAD(&dev_entry->cs_job_list);
1755 	INIT_LIST_HEAD(&dev_entry->userptr_list);
1756 	INIT_LIST_HEAD(&dev_entry->ctx_mem_hash_list);
1757 	mutex_init(&dev_entry->file_mutex);
1758 	init_rwsem(&dev_entry->state_dump_sem);
1759 	spin_lock_init(&dev_entry->cb_spinlock);
1760 	spin_lock_init(&dev_entry->cs_spinlock);
1761 	spin_lock_init(&dev_entry->cs_job_spinlock);
1762 	spin_lock_init(&dev_entry->userptr_spinlock);
1763 	mutex_init(&dev_entry->ctx_mem_hash_mutex);
1764 
1765 	return 0;
1766 }
1767 
1768 void hl_debugfs_device_fini(struct hl_device *hdev)
1769 {
1770 	struct hl_dbg_device_entry *entry = &hdev->hl_debugfs;
1771 	int i;
1772 
1773 	mutex_destroy(&entry->ctx_mem_hash_mutex);
1774 	mutex_destroy(&entry->file_mutex);
1775 
1776 	vfree(entry->data_dma_blob_desc.data);
1777 	vfree(entry->mon_dump_blob_desc.data);
1778 
1779 	for (i = 0; i < ARRAY_SIZE(entry->state_dump); ++i)
1780 		vfree(entry->state_dump[i]);
1781 
1782 	kfree(entry->entry_arr);
1783 }
1784 
1785 void hl_debugfs_add_device(struct hl_device *hdev)
1786 {
1787 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1788 
1789 	dev_entry->root = hdev->drm.accel->debugfs_root;
1790 
1791 	add_files_to_device(hdev, dev_entry, dev_entry->root);
1792 
1793 	if (!hdev->asic_prop.fw_security_enabled)
1794 		add_secured_nodes(dev_entry, dev_entry->root);
1795 }
1796 
1797 void hl_debugfs_add_file(struct hl_fpriv *hpriv)
1798 {
1799 	struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs;
1800 
1801 	mutex_lock(&dev_entry->file_mutex);
1802 	list_add(&hpriv->debugfs_list, &dev_entry->file_list);
1803 	mutex_unlock(&dev_entry->file_mutex);
1804 }
1805 
1806 void hl_debugfs_remove_file(struct hl_fpriv *hpriv)
1807 {
1808 	struct hl_dbg_device_entry *dev_entry = &hpriv->hdev->hl_debugfs;
1809 
1810 	mutex_lock(&dev_entry->file_mutex);
1811 	list_del(&hpriv->debugfs_list);
1812 	mutex_unlock(&dev_entry->file_mutex);
1813 }
1814 
1815 void hl_debugfs_add_cb(struct hl_cb *cb)
1816 {
1817 	struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs;
1818 
1819 	spin_lock(&dev_entry->cb_spinlock);
1820 	list_add(&cb->debugfs_list, &dev_entry->cb_list);
1821 	spin_unlock(&dev_entry->cb_spinlock);
1822 }
1823 
1824 void hl_debugfs_remove_cb(struct hl_cb *cb)
1825 {
1826 	struct hl_dbg_device_entry *dev_entry = &cb->hdev->hl_debugfs;
1827 
1828 	spin_lock(&dev_entry->cb_spinlock);
1829 	list_del(&cb->debugfs_list);
1830 	spin_unlock(&dev_entry->cb_spinlock);
1831 }
1832 
1833 void hl_debugfs_add_cs(struct hl_cs *cs)
1834 {
1835 	struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs;
1836 
1837 	spin_lock(&dev_entry->cs_spinlock);
1838 	list_add(&cs->debugfs_list, &dev_entry->cs_list);
1839 	spin_unlock(&dev_entry->cs_spinlock);
1840 }
1841 
1842 void hl_debugfs_remove_cs(struct hl_cs *cs)
1843 {
1844 	struct hl_dbg_device_entry *dev_entry = &cs->ctx->hdev->hl_debugfs;
1845 
1846 	spin_lock(&dev_entry->cs_spinlock);
1847 	list_del(&cs->debugfs_list);
1848 	spin_unlock(&dev_entry->cs_spinlock);
1849 }
1850 
1851 void hl_debugfs_add_job(struct hl_device *hdev, struct hl_cs_job *job)
1852 {
1853 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1854 
1855 	spin_lock(&dev_entry->cs_job_spinlock);
1856 	list_add(&job->debugfs_list, &dev_entry->cs_job_list);
1857 	spin_unlock(&dev_entry->cs_job_spinlock);
1858 }
1859 
1860 void hl_debugfs_remove_job(struct hl_device *hdev, struct hl_cs_job *job)
1861 {
1862 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1863 
1864 	spin_lock(&dev_entry->cs_job_spinlock);
1865 	list_del(&job->debugfs_list);
1866 	spin_unlock(&dev_entry->cs_job_spinlock);
1867 }
1868 
1869 void hl_debugfs_add_userptr(struct hl_device *hdev, struct hl_userptr *userptr)
1870 {
1871 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1872 
1873 	spin_lock(&dev_entry->userptr_spinlock);
1874 	list_add(&userptr->debugfs_list, &dev_entry->userptr_list);
1875 	spin_unlock(&dev_entry->userptr_spinlock);
1876 }
1877 
1878 void hl_debugfs_remove_userptr(struct hl_device *hdev,
1879 				struct hl_userptr *userptr)
1880 {
1881 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1882 
1883 	spin_lock(&dev_entry->userptr_spinlock);
1884 	list_del(&userptr->debugfs_list);
1885 	spin_unlock(&dev_entry->userptr_spinlock);
1886 }
1887 
1888 void hl_debugfs_add_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx)
1889 {
1890 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1891 
1892 	mutex_lock(&dev_entry->ctx_mem_hash_mutex);
1893 	list_add(&ctx->debugfs_list, &dev_entry->ctx_mem_hash_list);
1894 	mutex_unlock(&dev_entry->ctx_mem_hash_mutex);
1895 }
1896 
1897 void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx)
1898 {
1899 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1900 
1901 	mutex_lock(&dev_entry->ctx_mem_hash_mutex);
1902 	list_del(&ctx->debugfs_list);
1903 	mutex_unlock(&dev_entry->ctx_mem_hash_mutex);
1904 }
1905 
1906 /**
1907  * hl_debugfs_set_state_dump - register state dump making it accessible via
1908  *                             debugfs
1909  * @hdev: pointer to the device structure
1910  * @data: the actual dump data
1911  * @length: the length of the data
1912  */
1913 void hl_debugfs_set_state_dump(struct hl_device *hdev, char *data,
1914 					unsigned long length)
1915 {
1916 	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
1917 
1918 	down_write(&dev_entry->state_dump_sem);
1919 
1920 	dev_entry->state_dump_head = (dev_entry->state_dump_head + 1) %
1921 					ARRAY_SIZE(dev_entry->state_dump);
1922 	vfree(dev_entry->state_dump[dev_entry->state_dump_head]);
1923 	dev_entry->state_dump[dev_entry->state_dump_head] = data;
1924 
1925 	up_write(&dev_entry->state_dump_sem);
1926 }
1927