1/*
2 * Copyright (C) 2019 Alyssa Rosenzweig
3 * Copyright (C) 2017-2018 Lyude Paul
4 * Copyright (C) 2019 Collabora, Ltd.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 */
25
26#include <stdio.h>
27#include <stdlib.h>
28#include <assert.h>
29#include <stdint.h>
30#include <string.h>
31#include <sys/mman.h>
32
33#include "decode.h"
34#include "util/macros.h"
35#include "util/u_debug.h"
36#include "util/u_dynarray.h"
37#include "util/hash_table.h"
38
39FILE *pandecode_dump_stream;
40
41/* Memory handling */
42
43static struct hash_table_u64 *mmap_table;
44
45static struct util_dynarray ro_mappings;
46
47static struct pandecode_mapped_memory *
48pandecode_find_mapped_gpu_mem_containing_rw(uint64_t addr)
49{
50        return _mesa_hash_table_u64_search(mmap_table, addr & ~(4096 - 1));
51}
52
53struct pandecode_mapped_memory *
54pandecode_find_mapped_gpu_mem_containing(uint64_t addr)
55{
56        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing_rw(addr);
57
58        if (mem && mem->addr && !mem->ro) {
59                mprotect(mem->addr, mem->length, PROT_READ);
60                mem->ro = true;
61                util_dynarray_append(&ro_mappings, struct pandecode_mapped_memory *, mem);
62        }
63
64        return mem;
65}
66
67void
68pandecode_map_read_write(void)
69{
70        util_dynarray_foreach(&ro_mappings, struct pandecode_mapped_memory *, mem) {
71                (*mem)->ro = false;
72                mprotect((*mem)->addr, (*mem)->length, PROT_READ | PROT_WRITE);
73        }
74        util_dynarray_clear(&ro_mappings);
75}
76
77static void
78pandecode_add_name(struct pandecode_mapped_memory *mem, uint64_t gpu_va, const char *name)
79{
80        if (!name) {
81                /* If we don't have a name, assign one */
82
83                snprintf(mem->name, sizeof(mem->name) - 1,
84                         "memory_%" PRIx64, gpu_va);
85        } else {
86                assert((strlen(name) + 1) < sizeof(mem->name));
87                memcpy(mem->name, name, strlen(name) + 1);
88        }
89}
90
91void
92pandecode_inject_mmap(uint64_t gpu_va, void *cpu, unsigned sz, const char *name)
93{
94        /* First, search if we already mapped this and are just updating an address */
95
96        struct pandecode_mapped_memory *existing =
97                pandecode_find_mapped_gpu_mem_containing_rw(gpu_va);
98
99        if (existing && existing->gpu_va == gpu_va) {
100                existing->length = sz;
101                existing->addr = cpu;
102                pandecode_add_name(existing, gpu_va, name);
103                return;
104        }
105
106        /* Otherwise, add a fresh mapping */
107        struct pandecode_mapped_memory *mapped_mem = NULL;
108
109        mapped_mem = calloc(1, sizeof(*mapped_mem));
110        mapped_mem->gpu_va = gpu_va;
111        mapped_mem->length = sz;
112        mapped_mem->addr = cpu;
113        pandecode_add_name(mapped_mem, gpu_va, name);
114
115        /* Add it to the table */
116        assert((gpu_va & 4095) == 0);
117
118        for (unsigned i = 0; i < sz; i += 4096)
119                _mesa_hash_table_u64_insert(mmap_table, gpu_va + i, mapped_mem);
120}
121
122void
123pandecode_inject_free(uint64_t gpu_va, unsigned sz)
124{
125        struct pandecode_mapped_memory *mem =
126                pandecode_find_mapped_gpu_mem_containing_rw(gpu_va);
127
128        if (!mem)
129                return;
130
131        assert(mem->gpu_va == gpu_va);
132        assert(mem->length == sz);
133
134        free(mem);
135
136        for (unsigned i = 0; i < sz; i += 4096)
137                _mesa_hash_table_u64_remove(mmap_table, gpu_va + i);
138}
139
140char *
141pointer_as_memory_reference(uint64_t ptr)
142{
143        struct pandecode_mapped_memory *mapped;
144        char *out = malloc(128);
145
146        /* Try to find the corresponding mapped zone */
147
148        mapped = pandecode_find_mapped_gpu_mem_containing_rw(ptr);
149
150        if (mapped) {
151                snprintf(out, 128, "%s + %d", mapped->name, (int) (ptr - mapped->gpu_va));
152                return out;
153        }
154
155        /* Just use the raw address if other options are exhausted */
156
157        snprintf(out, 128, "0x%" PRIx64, ptr);
158        return out;
159
160}
161
162static int pandecode_dump_frame_count = 0;
163
164static bool force_stderr = false;
165
166void
167pandecode_dump_file_open(void)
168{
169        if (pandecode_dump_stream)
170                return;
171
172        /* This does a getenv every frame, so it is possible to use
173         * setenv to change the base at runtime.
174         */
175        const char *dump_file_base = debug_get_option("PANDECODE_DUMP_FILE", "pandecode.dump");
176        if (force_stderr || !strcmp(dump_file_base, "stderr"))
177                pandecode_dump_stream = stderr;
178        else {
179                char buffer[1024];
180                snprintf(buffer, sizeof(buffer), "%s.%04d", dump_file_base, pandecode_dump_frame_count);
181                printf("pandecode: dump command stream to file %s\n", buffer);
182                pandecode_dump_stream = fopen(buffer, "w");
183                if (!pandecode_dump_stream)
184                        fprintf(stderr,
185                                "pandecode: failed to open command stream log file %s\n",
186                                buffer);
187        }
188}
189
190static void
191pandecode_dump_file_close(void)
192{
193        if (pandecode_dump_stream && pandecode_dump_stream != stderr) {
194                if (fclose(pandecode_dump_stream))
195                        perror("pandecode: dump file");
196
197                pandecode_dump_stream = NULL;
198        }
199}
200
201void
202pandecode_initialize(bool to_stderr)
203{
204        force_stderr = to_stderr;
205        mmap_table = _mesa_hash_table_u64_create(NULL);
206        util_dynarray_init(&ro_mappings, NULL);
207}
208
209void
210pandecode_next_frame(void)
211{
212        pandecode_dump_file_close();
213        pandecode_dump_frame_count++;
214}
215
216void
217pandecode_close(void)
218{
219        _mesa_hash_table_u64_destroy(mmap_table);
220        util_dynarray_fini(&ro_mappings);
221        pandecode_dump_file_close();
222}
223
224void pandecode_abort_on_fault_v4(mali_ptr jc_gpu_va);
225void pandecode_abort_on_fault_v5(mali_ptr jc_gpu_va);
226void pandecode_abort_on_fault_v6(mali_ptr jc_gpu_va);
227void pandecode_abort_on_fault_v7(mali_ptr jc_gpu_va);
228
229void
230pandecode_abort_on_fault(mali_ptr jc_gpu_va, unsigned gpu_id)
231{
232        switch (pan_arch(gpu_id)) {
233        case 4: pandecode_abort_on_fault_v4(jc_gpu_va); return;
234        case 5: pandecode_abort_on_fault_v5(jc_gpu_va); return;
235        case 6: pandecode_abort_on_fault_v6(jc_gpu_va); return;
236        case 7: pandecode_abort_on_fault_v7(jc_gpu_va); return;
237        default: unreachable("Unsupported architecture");
238        }
239}
240
241void pandecode_jc_v4(mali_ptr jc_gpu_va, unsigned gpu_id);
242void pandecode_jc_v5(mali_ptr jc_gpu_va, unsigned gpu_id);
243void pandecode_jc_v6(mali_ptr jc_gpu_va, unsigned gpu_id);
244void pandecode_jc_v7(mali_ptr jc_gpu_va, unsigned gpu_id);
245
246void
247pandecode_jc(mali_ptr jc_gpu_va, unsigned gpu_id)
248{
249        switch (pan_arch(gpu_id)) {
250        case 4: pandecode_jc_v4(jc_gpu_va, gpu_id); return;
251        case 5: pandecode_jc_v5(jc_gpu_va, gpu_id); return;
252        case 6: pandecode_jc_v6(jc_gpu_va, gpu_id); return;
253        case 7: pandecode_jc_v7(jc_gpu_va, gpu_id); return;
254        default: unreachable("Unsupported architecture");
255        }
256}
257