1 /* Copyright (c) 2012-2017, The Linux Foundation. All rights reserved.
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License version 2 and
5 * only version 2 as published by the Free Software Foundation.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
14 #include "kgsl_sharedmem.h"
15 #include "kgsl_snapshot.h"
18 #include "adreno_pm4types.h"
20 #include "adreno_cp_parser.h"
21 #include "adreno_snapshot.h"
22 #include "adreno_a5xx.h"
24 #define VPC_MEMORY_BANKS 4
26 /* Maintain a list of the objects we see during parsing */
28 #define SNAPSHOT_OBJ_BUFSIZE 64
30 /* Used to print error message if an IB has too many objects in it */
31 static int ib_max_objs;
33 struct snapshot_rb_params {
34 struct kgsl_snapshot *snapshot;
35 struct adreno_ringbuffer *rb;
38 /* Keep track of how many bytes are frozen after a snapshot and tell the user */
39 static size_t snapshot_frozen_objsize;
41 static struct kgsl_snapshot_object objbuf[SNAPSHOT_OBJ_BUFSIZE];
43 /* Pointer to the next open entry in the object list */
44 static unsigned int objbufptr;
46 static inline int adreno_rb_ctxtswitch(struct adreno_device *adreno_dev,
49 return cmd[0] == cp_packet(adreno_dev, CP_NOP, 1) &&
50 cmd[1] == KGSL_CONTEXT_TO_MEM_IDENTIFIER;
53 /* Push a new buffer object onto the list */
54 void kgsl_snapshot_push_object(struct kgsl_process_private *process,
55 uint64_t gpuaddr, uint64_t dwords)
58 struct kgsl_mem_entry *entry;
64 * Sometimes IBs can be reused in the same dump. Because we parse from
65 * oldest to newest, if we come across an IB that has already been used,
66 * assume that it has been reused and update the list with the newest
70 for (index = 0; index < objbufptr; index++) {
71 if (objbuf[index].gpuaddr == gpuaddr &&
72 objbuf[index].entry->priv == process) {
74 objbuf[index].size = max_t(uint64_t,
81 if (objbufptr == SNAPSHOT_OBJ_BUFSIZE) {
82 KGSL_CORE_ERR("snapshot: too many snapshot objects\n");
86 entry = kgsl_sharedmem_find(process, gpuaddr);
88 KGSL_CORE_ERR("snapshot: Can't find entry for 0x%016llX\n",
93 if (!kgsl_gpuaddr_in_memdesc(&entry->memdesc, gpuaddr, dwords << 2)) {
94 KGSL_CORE_ERR("snapshot: Mem entry 0x%016llX is too small\n",
96 kgsl_mem_entry_put(entry);
100 /* Put it on the list of things to parse */
101 objbuf[objbufptr].gpuaddr = gpuaddr;
102 objbuf[objbufptr].size = dwords << 2;
103 objbuf[objbufptr++].entry = entry;
107 * Returns index of the specified object is already on the list of buffers
111 static int find_object(uint64_t gpuaddr, struct kgsl_process_private *process)
115 for (index = 0; index < objbufptr; index++) {
116 if (objbuf[index].gpuaddr == gpuaddr &&
117 objbuf[index].entry->priv == process)
124 * snapshot_freeze_obj_list() - Take a list of ib objects and freeze their
125 * memory for snapshot
126 * @snapshot: The snapshot data.
127 * @process: The process to which the IB belongs
128 * @ib_obj_list: List of the IB objects
130 * Returns 0 on success else error code
132 static int snapshot_freeze_obj_list(struct kgsl_snapshot *snapshot,
133 struct kgsl_process_private *process,
134 struct adreno_ib_object_list *ib_obj_list)
137 struct adreno_ib_object *ib_objs;
140 for (i = 0; i < ib_obj_list->num_objs; i++) {
145 ib_objs = &(ib_obj_list->obj_list[i]);
146 /* Make sure this object is not going to be saved statically */
147 for (index = 0; index < objbufptr; index++) {
148 if ((objbuf[index].gpuaddr <= ib_objs->gpuaddr) &&
149 ((objbuf[index].gpuaddr +
150 (objbuf[index].size)) >=
151 (ib_objs->gpuaddr + ib_objs->size)) &&
152 (objbuf[index].entry->priv == process)) {
159 temp_ret = kgsl_snapshot_get_object(snapshot,
160 process, ib_objs->gpuaddr,
162 ib_objs->snapshot_obj_type);
167 snapshot_frozen_objsize += temp_ret;
175 * We want to store the last executed IB1 and IB2 in the static region to ensure
176 * that we get at least some information out of the snapshot even if we can't
177 * access the dynamic data from the sysfs file. Push all other IBs on the
180 static inline void parse_ib(struct kgsl_device *device,
181 struct kgsl_snapshot *snapshot,
182 struct kgsl_process_private *process,
183 uint64_t gpuaddr, uint64_t dwords)
185 struct adreno_ib_object_list *ib_obj_list;
188 * Check the IB address - if it is either the last executed IB1
189 * then push it into the static blob otherwise put it in the dynamic
192 if (gpuaddr == snapshot->ib1base) {
193 kgsl_snapshot_push_object(process, gpuaddr, dwords);
197 if (kgsl_snapshot_have_object(snapshot, process,
198 gpuaddr, dwords << 2))
201 if (-E2BIG == adreno_ib_create_object_list(device, process,
202 gpuaddr, dwords, snapshot->ib2base,
207 kgsl_snapshot_add_ib_obj_list(snapshot, ib_obj_list);
211 static inline bool iommu_is_setstate_addr(struct kgsl_device *device,
212 uint64_t gpuaddr, uint64_t size)
214 struct kgsl_iommu *iommu = KGSL_IOMMU_PRIV(device);
216 if (kgsl_mmu_get_mmutype(device) != KGSL_MMU_TYPE_IOMMU)
219 return kgsl_gpuaddr_in_memdesc(&iommu->setstate, gpuaddr,
223 static void dump_all_ibs(struct kgsl_device *device,
224 struct adreno_ringbuffer *rb,
225 struct kgsl_snapshot *snapshot)
229 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
231 rbptr = rb->buffer_desc.hostptr;
233 for (index = 0; index < KGSL_RB_DWORDS;) {
235 if (adreno_cmd_is_ib(adreno_dev, rbptr[index])) {
239 if (ADRENO_LEGACY_PM4(adreno_dev)) {
240 ibaddr = rbptr[index + 1];
241 ibsize = rbptr[index + 2];
244 ibaddr = rbptr[index + 2];
245 ibaddr = ibaddr << 32 | rbptr[index + 1];
246 ibsize = rbptr[index + 3];
250 /* Don't parse known global IBs */
251 if (iommu_is_setstate_addr(device, ibaddr, ibsize))
254 if (kgsl_gpuaddr_in_memdesc(&adreno_dev->pwron_fixup,
258 parse_ib(device, snapshot, snapshot->process, ibaddr,
266 * snapshot_rb_ibs() - Dump rb data and capture the IB's in the RB as well
267 * @device: Pointer to a KGSL device
268 * @rb: The RB to dump
269 * @data: Pointer to memory where the RB data is to be dumped
270 * @snapshot: Pointer to information about the current snapshot being taken
272 static void snapshot_rb_ibs(struct kgsl_device *device,
273 struct adreno_ringbuffer *rb,
274 struct kgsl_snapshot *snapshot)
276 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
277 unsigned int rptr, *rbptr;
279 int parse_ibs = 0, ib_parse_start;
281 /* Get the current read pointers for the RB */
282 adreno_readreg(adreno_dev, ADRENO_REG_CP_RB_RPTR, &rptr);
285 * Figure out the window of ringbuffer data to dump. First we need to
286 * find where the last processed IB ws submitted. Start walking back
291 rbptr = rb->buffer_desc.hostptr;
297 if (ADRENO_LEGACY_PM4(adreno_dev))
298 index = KGSL_RB_DWORDS - 3;
300 index = KGSL_RB_DWORDS - 4;
302 /* We wrapped without finding what we wanted */
303 if (index < rb->wptr) {
309 if (adreno_cmd_is_ib(adreno_dev, rbptr[index])) {
310 if (ADRENO_LEGACY_PM4(adreno_dev)) {
311 if (rbptr[index + 1] == snapshot->ib1base)
316 ibaddr = rbptr[index + 2];
317 ibaddr = ibaddr << 32 | rbptr[index + 1];
318 if (ibaddr == snapshot->ib1base)
322 } while (index != rb->wptr);
325 * If the ib1 was not found, for example, if ib1base was restored
326 * incorrectly after preemption, then simply dump the entire
327 * ringbuffer along with all the IBs in the ringbuffer.
330 if (index == rb->wptr) {
331 dump_all_ibs(device, rb, snapshot);
336 * index points at the last submitted IB. We can only trust that the
337 * memory between the context switch and the hanging IB is valid, so
338 * the next step is to find the context switch before the submission
341 while (index != rb->wptr) {
345 index = KGSL_RB_DWORDS - 2;
348 * Wrapped without finding the context switch. This is
349 * harmless - we should still have enough data to dump a
353 if (index < rb->wptr) {
359 /* Break if the current packet is a context switch identifier */
360 if ((rbptr[index] == cp_packet(adreno_dev, CP_NOP, 1)) &&
361 (rbptr[index + 1] == KGSL_CONTEXT_TO_MEM_IDENTIFIER))
366 * Index represents the start of the window of interest. We will try
367 * to dump all buffers between here and the rptr
370 ib_parse_start = index;
373 * Loop through the RB, looking for indirect buffers and MMU pagetable
378 for (i = 0; i < KGSL_RB_DWORDS; i++) {
380 * Only parse IBs between the start and the rptr or the next
381 * context switch, whichever comes first
384 if (parse_ibs == 0 && index == ib_parse_start)
386 else if (index == rptr || adreno_rb_ctxtswitch(adreno_dev,
390 if (parse_ibs && adreno_cmd_is_ib(adreno_dev, rbptr[index])) {
394 if (ADRENO_LEGACY_PM4(adreno_dev)) {
395 ibaddr = rbptr[index + 1];
396 ibsize = rbptr[index + 2];
398 ibaddr = rbptr[index + 2];
399 ibaddr = ibaddr << 32 | rbptr[index + 1];
400 ibsize = rbptr[index + 3];
403 /* Don't parse known global IBs */
404 if (iommu_is_setstate_addr(device, ibaddr, ibsize))
407 if (kgsl_gpuaddr_in_memdesc(&adreno_dev->pwron_fixup,
411 parse_ib(device, snapshot, snapshot->process,
415 index = (index + 1) % KGSL_RB_DWORDS;
420 /* Snapshot the ringbuffer memory */
421 static size_t snapshot_rb(struct kgsl_device *device, u8 *buf,
422 size_t remain, void *priv)
424 struct kgsl_snapshot_rb_v2 *header = (struct kgsl_snapshot_rb_v2 *)buf;
425 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
426 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
427 struct snapshot_rb_params *snap_rb_params = priv;
428 struct kgsl_snapshot *snapshot = snap_rb_params->snapshot;
429 struct adreno_ringbuffer *rb = snap_rb_params->rb;
432 * Dump the entire ringbuffer - the parser can choose how much of it to
436 if (remain < KGSL_RB_SIZE + sizeof(*header)) {
437 KGSL_CORE_ERR("snapshot: Not enough memory for the rb section");
441 /* Write the sub-header for the section */
443 header->end = KGSL_RB_DWORDS;
444 header->wptr = rb->wptr;
445 header->rptr = adreno_get_rptr(rb);
446 header->rbsize = KGSL_RB_DWORDS;
447 header->count = KGSL_RB_DWORDS;
448 adreno_rb_readtimestamp(adreno_dev, rb, KGSL_TIMESTAMP_QUEUED,
449 &header->timestamp_queued);
450 adreno_rb_readtimestamp(adreno_dev, rb, KGSL_TIMESTAMP_RETIRED,
451 &header->timestamp_retired);
452 header->gpuaddr = rb->buffer_desc.gpuaddr;
455 if (rb == adreno_dev->cur_rb)
456 snapshot_rb_ibs(device, rb, snapshot);
458 /* Just copy the ringbuffer, there are no active IBs */
459 memcpy(data, rb->buffer_desc.hostptr, KGSL_RB_SIZE);
461 /* Return the size of the section */
462 return KGSL_RB_SIZE + sizeof(*header);
465 static int _count_mem_entries(int id, void *ptr, void *data)
478 static int _save_mem_entries(int id, void *ptr, void *data)
480 struct kgsl_mem_entry *entry = ptr;
481 struct mem_entry *m = (struct mem_entry *) data;
482 unsigned int index = id - 1;
484 m[index].gpuaddr = entry->memdesc.gpuaddr;
485 m[index].size = entry->memdesc.size;
486 m[index].type = kgsl_memdesc_get_memtype(&entry->memdesc);
491 static size_t snapshot_capture_mem_list(struct kgsl_device *device,
492 u8 *buf, size_t remain, void *priv)
494 struct kgsl_snapshot_mem_list_v2 *header =
495 (struct kgsl_snapshot_mem_list_v2 *)buf;
498 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
499 struct kgsl_process_private *process = priv;
501 /* we need a process to search! */
505 spin_lock(&process->mem_lock);
507 /* We need to know the number of memory objects that the process has */
508 idr_for_each(&process->mem_idr, _count_mem_entries, &num_mem);
513 if (remain < ((num_mem * sizeof(struct mem_entry)) + sizeof(*header))) {
514 KGSL_CORE_ERR("snapshot: Not enough memory for the mem list");
518 header->num_entries = num_mem;
519 header->ptbase = kgsl_mmu_pagetable_get_ttbr0(process->pagetable);
521 * Walk throught the memory list and store the
522 * tuples(gpuaddr, size, memtype) in snapshot
525 idr_for_each(&process->mem_idr, _save_mem_entries, data);
527 ret = sizeof(*header) + (num_mem * sizeof(struct mem_entry));
529 spin_unlock(&process->mem_lock);
533 struct snapshot_ib_meta {
534 struct kgsl_snapshot *snapshot;
535 struct kgsl_snapshot_object *obj;
542 void kgsl_snapshot_add_active_ib_obj_list(struct kgsl_device *device,
543 struct kgsl_snapshot *snapshot)
545 struct adreno_ib_object_list *ib_obj_list;
548 if (!snapshot->ib1dumped)
549 index = find_object(snapshot->ib1base, snapshot->process);
551 /* only do this for IB1 because the IB2's are part of IB1 objects */
552 if ((index != -ENOENT) &&
553 (snapshot->ib1base == objbuf[index].gpuaddr)) {
554 if (-E2BIG == adreno_ib_create_object_list(device,
555 objbuf[index].entry->priv,
556 objbuf[index].gpuaddr,
557 objbuf[index].size >> 2,
562 /* freeze the IB objects in the IB */
563 snapshot_freeze_obj_list(snapshot,
564 objbuf[index].entry->priv,
566 adreno_ib_destroy_obj_list(ib_obj_list);
569 /* Get the IB2 index from parsed object */
570 index = find_object(snapshot->ib2base, snapshot->process);
572 if (index != -ENOENT)
573 parse_ib(device, snapshot, snapshot->process,
574 snapshot->ib2base, objbuf[index].size >> 2);
579 * active_ib_is_parsed() - Checks if active ib is already parsed
580 * @gpuaddr: Active IB base address at the time of fault
581 * @size: Active IB size
582 * @process: The process to which the IB belongs
584 * Function returns true if the active is already is parsed
587 static bool active_ib_is_parsed(uint64_t gpuaddr, uint64_t size,
588 struct kgsl_process_private *process)
591 /* go through the static list for gpuaddr is in list or not */
592 for (index = 0; index < objbufptr; index++) {
593 if ((objbuf[index].gpuaddr <= gpuaddr) &&
594 ((objbuf[index].gpuaddr +
595 (objbuf[index].size)) >=
597 (objbuf[index].entry->priv == process))
602 /* Snapshot the memory for an indirect buffer */
603 static size_t snapshot_ib(struct kgsl_device *device, u8 *buf,
604 size_t remain, void *priv)
606 struct kgsl_snapshot_ib_v2 *header = (struct kgsl_snapshot_ib_v2 *)buf;
607 struct snapshot_ib_meta *meta = priv;
609 unsigned int *dst = (unsigned int *)(buf + sizeof(*header));
610 struct adreno_ib_object_list *ib_obj_list;
611 struct kgsl_snapshot *snapshot;
612 struct kgsl_snapshot_object *obj;
613 struct kgsl_memdesc *memdesc;
615 if (meta == NULL || meta->snapshot == NULL || meta->obj == NULL) {
616 KGSL_CORE_ERR("snapshot: bad metadata");
619 snapshot = meta->snapshot;
621 memdesc = &obj->entry->memdesc;
623 /* If size is zero get it from the medesc size */
625 obj->size = (memdesc->size - (obj->gpuaddr - memdesc->gpuaddr));
627 if (remain < (obj->size + sizeof(*header))) {
628 KGSL_CORE_ERR("snapshot: Not enough memory for the ib\n");
632 src = kgsl_gpuaddr_to_vaddr(memdesc, obj->gpuaddr);
635 "snapshot: Unable to map GPU memory object 0x%016llX into the kernel\n",
640 /* only do this for IB1 because the IB2's are part of IB1 objects */
641 if (meta->ib1base == obj->gpuaddr) {
643 snapshot->ib1dumped = active_ib_is_parsed(obj->gpuaddr,
644 obj->size, obj->entry->priv);
645 if (-E2BIG == adreno_ib_create_object_list(device,
647 obj->gpuaddr, obj->size >> 2,
652 /* freeze the IB objects in the IB */
653 snapshot_freeze_obj_list(snapshot,
656 adreno_ib_destroy_obj_list(ib_obj_list);
661 if (meta->ib2base == obj->gpuaddr)
662 snapshot->ib2dumped = active_ib_is_parsed(obj->gpuaddr,
663 obj->size, obj->entry->priv);
665 /* Write the sub-header for the section */
666 header->gpuaddr = obj->gpuaddr;
668 kgsl_mmu_pagetable_get_ttbr0(obj->entry->priv->pagetable);
669 header->size = obj->size >> 2;
671 /* Write the contents of the ib */
672 memcpy((void *)dst, (void *)src, (size_t) obj->size);
673 /* Write the contents of the ib */
675 return obj->size + sizeof(*header);
678 /* Dump another item on the current pending list */
679 static void dump_object(struct kgsl_device *device, int obj,
680 struct kgsl_snapshot *snapshot)
682 struct snapshot_ib_meta meta;
684 meta.snapshot = snapshot;
685 meta.obj = &objbuf[obj];
686 meta.ib1base = snapshot->ib1base;
687 meta.ib1size = snapshot->ib1size;
688 meta.ib2base = snapshot->ib2base;
689 meta.ib2size = snapshot->ib2size;
691 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_IB_V2,
692 snapshot, snapshot_ib, &meta);
693 if (objbuf[obj].entry) {
694 kgsl_memdesc_unmap(&(objbuf[obj].entry->memdesc));
695 kgsl_mem_entry_put(objbuf[obj].entry);
699 /* setup_fault process - Find kgsl_process_private struct that caused the fault
701 * Find the faulting process based what the dispatcher thinks happened and
702 * what the hardware is using for the current pagetable. The process struct
703 * will be used to look up GPU addresses that are encountered while parsing
706 static void setup_fault_process(struct kgsl_device *device,
707 struct kgsl_snapshot *snapshot,
708 struct kgsl_process_private *process)
710 u64 hw_ptbase, proc_ptbase;
712 if (process != NULL && !kgsl_process_private_get(process))
715 /* Get the physical address of the MMU pagetable */
716 hw_ptbase = kgsl_mmu_get_current_ttbr0(&device->mmu);
718 /* if we have an input process, make sure the ptbases match */
720 proc_ptbase = kgsl_mmu_pagetable_get_ttbr0(process->pagetable);
721 /* agreement! No need to check further */
722 if (hw_ptbase == proc_ptbase)
725 kgsl_process_private_put(process);
727 KGSL_CORE_ERR("snapshot: ptbase mismatch hw %llx sw %llx\n",
728 hw_ptbase, proc_ptbase);
731 /* try to find the right pagetable by walking the process list */
732 if (kgsl_mmu_is_perprocess(&device->mmu)) {
733 struct kgsl_process_private *tmp;
735 mutex_lock(&kgsl_driver.process_mutex);
736 list_for_each_entry(tmp, &kgsl_driver.process_list, list) {
739 pt_ttbr0 = kgsl_mmu_pagetable_get_ttbr0(tmp->pagetable);
740 if ((pt_ttbr0 == hw_ptbase)
741 && kgsl_process_private_get(tmp)) {
746 mutex_unlock(&kgsl_driver.process_mutex);
749 snapshot->process = process;
752 /* Snapshot a global memory buffer */
753 static size_t snapshot_global(struct kgsl_device *device, u8 *buf,
754 size_t remain, void *priv)
756 struct kgsl_memdesc *memdesc = priv;
758 struct kgsl_snapshot_gpu_object_v2 *header =
759 (struct kgsl_snapshot_gpu_object_v2 *)buf;
761 u8 *ptr = buf + sizeof(*header);
763 if (memdesc->size == 0)
766 if (remain < (memdesc->size + sizeof(*header))) {
767 KGSL_CORE_ERR("snapshot: Not enough memory for the memdesc\n");
771 if (memdesc->hostptr == NULL) {
772 KGSL_CORE_ERR("snapshot: no kernel mapping for global object 0x%016llX\n",
777 header->size = memdesc->size >> 2;
778 header->gpuaddr = memdesc->gpuaddr;
779 header->ptbase = MMU_DEFAULT_TTBR0(device);
780 header->type = SNAPSHOT_GPU_OBJECT_GLOBAL;
782 memcpy(ptr, memdesc->hostptr, memdesc->size);
784 return memdesc->size + sizeof(*header);
787 /* Snapshot IOMMU specific buffers */
788 static void adreno_snapshot_iommu(struct kgsl_device *device,
789 struct kgsl_snapshot *snapshot)
791 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
792 struct kgsl_iommu *iommu = KGSL_IOMMU_PRIV(device);
794 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_GPU_OBJECT_V2,
795 snapshot, snapshot_global, &iommu->setstate);
797 if (ADRENO_FEATURE(adreno_dev, ADRENO_PREEMPTION))
798 kgsl_snapshot_add_section(device,
799 KGSL_SNAPSHOT_SECTION_GPU_OBJECT_V2,
800 snapshot, snapshot_global, &iommu->smmu_info);
803 static void adreno_snapshot_ringbuffer(struct kgsl_device *device,
804 struct kgsl_snapshot *snapshot, struct adreno_ringbuffer *rb)
806 struct snapshot_rb_params params = {
807 .snapshot = snapshot,
814 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_RB_V2, snapshot,
815 snapshot_rb, ¶ms);
818 /* adreno_snapshot - Snapshot the Adreno GPU state
819 * @device - KGSL device to snapshot
820 * @snapshot - Pointer to the snapshot instance
821 * @context - context that caused the fault, if known by the driver
822 * This is a hook function called by kgsl_snapshot to snapshot the
823 * Adreno specific information for the GPU snapshot. In turn, this function
824 * calls the GPU specific snapshot function to get core specific information.
826 void adreno_snapshot(struct kgsl_device *device, struct kgsl_snapshot *snapshot,
827 struct kgsl_context *context)
830 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
831 struct adreno_gpudev *gpudev = ADRENO_GPU_DEVICE(adreno_dev);
834 /* Reset the list of objects */
837 snapshot_frozen_objsize = 0;
839 setup_fault_process(device, snapshot,
840 context ? context->proc_priv : NULL);
842 adreno_readreg64(adreno_dev, ADRENO_REG_CP_IB1_BASE,
843 ADRENO_REG_CP_IB1_BASE_HI, &snapshot->ib1base);
844 adreno_readreg(adreno_dev, ADRENO_REG_CP_IB1_BUFSZ, &snapshot->ib1size);
845 adreno_readreg64(adreno_dev, ADRENO_REG_CP_IB2_BASE,
846 ADRENO_REG_CP_IB2_BASE_HI, &snapshot->ib2base);
847 adreno_readreg(adreno_dev, ADRENO_REG_CP_IB2_BUFSZ, &snapshot->ib2size);
849 snapshot->ib1dumped = false;
850 snapshot->ib2dumped = false;
852 adreno_snapshot_ringbuffer(device, snapshot, adreno_dev->cur_rb);
854 /* Dump the prev ringbuffer */
855 if (adreno_dev->prev_rb != adreno_dev->cur_rb)
856 adreno_snapshot_ringbuffer(device, snapshot,
857 adreno_dev->prev_rb);
859 if ((adreno_dev->next_rb != adreno_dev->prev_rb) &&
860 (adreno_dev->next_rb != adreno_dev->cur_rb))
861 adreno_snapshot_ringbuffer(device, snapshot,
862 adreno_dev->next_rb);
864 /* Add GPU specific sections - registers mainly, but other stuff too */
865 if (gpudev->snapshot)
866 gpudev->snapshot(adreno_dev, snapshot);
868 /* Dump selected global buffers */
869 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_GPU_OBJECT_V2,
870 snapshot, snapshot_global, &device->memstore);
872 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_GPU_OBJECT_V2,
873 snapshot, snapshot_global,
874 &adreno_dev->pwron_fixup);
876 if (kgsl_mmu_get_mmutype(device) == KGSL_MMU_TYPE_IOMMU)
877 adreno_snapshot_iommu(device, snapshot);
880 * Add a section that lists (gpuaddr, size, memtype) tuples of the
883 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_MEMLIST_V2,
884 snapshot, snapshot_capture_mem_list, snapshot->process);
886 * Make sure that the last IB1 that was being executed is dumped.
887 * Since this was the last IB1 that was processed, we should have
888 * already added it to the list during the ringbuffer parse but we
889 * want to be double plus sure.
890 * The problem is that IB size from the register is the unprocessed size
891 * of the buffer not the original size, so if we didn't catch this
892 * buffer being directly used in the RB, then we might not be able to
893 * dump the whole thing. Print a warning message so we can try to
894 * figure how often this really happens.
897 if (-ENOENT == find_object(snapshot->ib1base, snapshot->process) &&
899 kgsl_snapshot_push_object(snapshot->process, snapshot->ib1base,
902 "CP_IB1_BASE not found in the ringbuffer.Dumping %x dwords of the buffer.\n",
907 * Add the last parsed IB2 to the list. The IB2 should be found as we
908 * parse the objects below, but we try to add it to the list first, so
909 * it too can be parsed. Don't print an error message in this case - if
910 * the IB2 is found during parsing, the list will be updated with the
914 if (-ENOENT == find_object(snapshot->ib2base, snapshot->process)) {
915 kgsl_snapshot_push_object(snapshot->process, snapshot->ib2base,
920 * Go through the list of found objects and dump each one. As the IBs
921 * are parsed, more objects might be found, and objbufptr will increase
923 for (i = 0; i < objbufptr; i++)
924 dump_object(device, i, snapshot);
927 * Incase snapshot static blob is running out of memory, Add Active IB1
928 * and IB2 entries to obj_list so that active ib's can be dumped to
929 * snapshot dynamic blob.
931 if (!snapshot->ib1dumped || !snapshot->ib2dumped)
932 kgsl_snapshot_add_active_ib_obj_list(device, snapshot);
935 KGSL_CORE_ERR("Max objects found in IB\n");
936 if (snapshot_frozen_objsize)
937 KGSL_CORE_ERR("GPU snapshot froze %zdKb of GPU buffers\n",
938 snapshot_frozen_objsize / 1024);
943 * adreno_snapshot_cp_roq - Dump CP merciu data in snapshot
944 * @device: Device being snapshotted
945 * @remain: Bytes remaining in snapshot memory
946 * @priv: Size of merciu data in Dwords
948 size_t adreno_snapshot_cp_merciu(struct kgsl_device *device, u8 *buf,
949 size_t remain, void *priv)
951 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
952 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
953 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
954 int i, size = *((int *)priv);
956 /* The MERCIU data is two dwords per entry */
959 if (remain < DEBUG_SECTION_SZ(size)) {
960 SNAPSHOT_ERR_NOMEM(device, "CP MERCIU DEBUG");
964 header->type = SNAPSHOT_DEBUG_CP_MERCIU;
967 adreno_writereg(adreno_dev, ADRENO_REG_CP_MERCIU_ADDR, 0x0);
969 for (i = 0; i < size; i++) {
970 adreno_readreg(adreno_dev, ADRENO_REG_CP_MERCIU_DATA,
972 adreno_readreg(adreno_dev, ADRENO_REG_CP_MERCIU_DATA2,
976 return DEBUG_SECTION_SZ(size);
980 * adreno_snapshot_cp_roq - Dump ROQ data in snapshot
981 * @device: Device being snapshotted
982 * @remain: Bytes remaining in snapshot memory
983 * @priv: Size of ROQ data in Dwords
985 size_t adreno_snapshot_cp_roq(struct kgsl_device *device, u8 *buf,
986 size_t remain, void *priv)
988 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
989 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
990 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
991 int i, size = *((int *)priv);
993 if (remain < DEBUG_SECTION_SZ(size)) {
994 SNAPSHOT_ERR_NOMEM(device, "CP ROQ DEBUG");
998 header->type = SNAPSHOT_DEBUG_CP_ROQ;
1001 adreno_writereg(adreno_dev, ADRENO_REG_CP_ROQ_ADDR, 0x0);
1002 for (i = 0; i < size; i++)
1003 adreno_readreg(adreno_dev, ADRENO_REG_CP_ROQ_DATA, &data[i]);
1005 return DEBUG_SECTION_SZ(size);
1009 * adreno_snapshot_cp_pm4_ram() - Dump PM4 data in snapshot
1010 * @device: Device being snapshotted
1011 * @buf: Snapshot memory
1012 * @remain: Number of bytes left in snapshot memory
1015 size_t adreno_snapshot_cp_pm4_ram(struct kgsl_device *device, u8 *buf,
1016 size_t remain, void *priv)
1018 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
1019 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
1020 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
1022 size_t size = adreno_dev->pm4_fw_size - 1;
1024 if (remain < DEBUG_SECTION_SZ(size)) {
1025 SNAPSHOT_ERR_NOMEM(device, "CP PM4 RAM DEBUG");
1029 header->type = SNAPSHOT_DEBUG_CP_PM4_RAM;
1030 header->size = size;
1033 * Read the firmware from the GPU rather than use our cache in order to
1034 * try to catch mis-programming or corruption in the hardware. We do
1035 * use the cached version of the size, however, instead of trying to
1036 * maintain always changing hardcoded constants
1039 adreno_writereg(adreno_dev, ADRENO_REG_CP_ME_RAM_RADDR, 0x0);
1040 for (i = 0; i < size; i++)
1041 adreno_readreg(adreno_dev, ADRENO_REG_CP_ME_RAM_DATA, &data[i]);
1043 return DEBUG_SECTION_SZ(size);
1047 * adreno_snapshot_cp_pfp_ram() - Dump the PFP data on snapshot
1048 * @device: Device being snapshotted
1049 * @buf: Snapshot memory
1050 * @remain: Amount of butes left in snapshot memory
1053 size_t adreno_snapshot_cp_pfp_ram(struct kgsl_device *device, u8 *buf,
1054 size_t remain, void *priv)
1056 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
1057 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
1058 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
1059 int i, size = adreno_dev->pfp_fw_size - 1;
1061 if (remain < DEBUG_SECTION_SZ(size)) {
1062 SNAPSHOT_ERR_NOMEM(device, "CP PFP RAM DEBUG");
1066 header->type = SNAPSHOT_DEBUG_CP_PFP_RAM;
1067 header->size = size;
1070 * Read the firmware from the GPU rather than use our cache in order to
1071 * try to catch mis-programming or corruption in the hardware. We do
1072 * use the cached version of the size, however, instead of trying to
1073 * maintain always changing hardcoded constants
1075 adreno_writereg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_ADDR, 0x0);
1076 for (i = 0; i < size; i++)
1077 adreno_readreg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_DATA,
1080 return DEBUG_SECTION_SZ(size);
1084 * adreno_snapshot_vpc_memory() - Save VPC data in snapshot
1085 * @device: Device being snapshotted
1086 * @buf: Snapshot memory
1087 * @remain: Number of bytes left in snapshot memory
1088 * @priv: Private data for VPC if any
1090 size_t adreno_snapshot_vpc_memory(struct kgsl_device *device, u8 *buf,
1091 size_t remain, void *priv)
1093 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
1094 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
1095 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
1096 int vpc_mem_size = *((int *)priv);
1097 size_t size = VPC_MEMORY_BANKS * vpc_mem_size;
1098 int bank, addr, i = 0;
1100 if (remain < DEBUG_SECTION_SZ(size)) {
1101 SNAPSHOT_ERR_NOMEM(device, "VPC MEMORY");
1105 header->type = SNAPSHOT_DEBUG_VPC_MEMORY;
1106 header->size = size;
1108 for (bank = 0; bank < VPC_MEMORY_BANKS; bank++) {
1109 for (addr = 0; addr < vpc_mem_size; addr++) {
1110 unsigned int val = bank | (addr << 4);
1111 adreno_writereg(adreno_dev,
1112 ADRENO_REG_VPC_DEBUG_RAM_SEL, val);
1113 adreno_readreg(adreno_dev,
1114 ADRENO_REG_VPC_DEBUG_RAM_READ, &data[i++]);
1118 return DEBUG_SECTION_SZ(size);
1122 * adreno_snapshot_cp_meq() - Save CP MEQ data in snapshot
1123 * @device: Device being snapshotted
1124 * @buf: Snapshot memory
1125 * @remain: Number of bytes left in snapshot memory
1126 * @priv: Contains the size of MEQ data
1128 size_t adreno_snapshot_cp_meq(struct kgsl_device *device, u8 *buf,
1129 size_t remain, void *priv)
1131 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
1132 struct kgsl_snapshot_debug *header = (struct kgsl_snapshot_debug *)buf;
1133 unsigned int *data = (unsigned int *)(buf + sizeof(*header));
1135 int cp_meq_sz = *((int *)priv);
1137 if (remain < DEBUG_SECTION_SZ(cp_meq_sz)) {
1138 SNAPSHOT_ERR_NOMEM(device, "CP MEQ DEBUG");
1142 header->type = SNAPSHOT_DEBUG_CP_MEQ;
1143 header->size = cp_meq_sz;
1145 adreno_writereg(adreno_dev, ADRENO_REG_CP_MEQ_ADDR, 0x0);
1146 for (i = 0; i < cp_meq_sz; i++)
1147 adreno_readreg(adreno_dev, ADRENO_REG_CP_MEQ_DATA, &data[i]);
1149 return DEBUG_SECTION_SZ(cp_meq_sz);
1152 static const struct adreno_vbif_snapshot_registers *vbif_registers(
1153 struct adreno_device *adreno_dev,
1154 const struct adreno_vbif_snapshot_registers *list,
1157 unsigned int version;
1160 adreno_readreg(adreno_dev, ADRENO_REG_VBIF_VERSION, &version);
1162 for (i = 0; i < count; i++) {
1163 if ((list[i].version & list[i].mask) ==
1164 (version & list[i].mask))
1169 "snapshot: Registers for VBIF version %X register were not dumped\n",
1175 void adreno_snapshot_registers(struct kgsl_device *device,
1176 struct kgsl_snapshot *snapshot,
1177 const unsigned int *regs, unsigned int count)
1179 struct kgsl_snapshot_registers r;
1184 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_REGS, snapshot,
1185 kgsl_snapshot_dump_registers, &r);
1188 void adreno_snapshot_vbif_registers(struct kgsl_device *device,
1189 struct kgsl_snapshot *snapshot,
1190 const struct adreno_vbif_snapshot_registers *list,
1193 struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
1194 struct kgsl_snapshot_registers regs;
1195 const struct adreno_vbif_snapshot_registers *vbif;
1197 vbif = vbif_registers(adreno_dev, list, count);
1200 regs.regs = vbif->registers;
1201 regs.count = vbif->count;
1203 kgsl_snapshot_add_section(device, KGSL_SNAPSHOT_SECTION_REGS,
1204 snapshot, kgsl_snapshot_dump_registers, ®s);