• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2017-2019 Lyude Paul
3  * Copyright (C) 2017-2019 Alyssa Rosenzweig
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  *
24  */
25 
26 #ifndef __PAN_DECODE_H__
27 #define __PAN_DECODE_H__
28 
29 #include "genxml/gen_macros.h"
30 #include "util/rb_tree.h"
31 #include "util/simple_mtx.h"
32 #include "util/u_dynarray.h"
33 
34 #include "wrap.h"
35 
36 struct pandecode_context {
37    int id; /* only used for the filename */
38    FILE *dump_stream;
39    unsigned indent;
40    struct rb_tree mmap_tree;
41    struct util_dynarray ro_mappings;
42    int dump_frame_count;
43    simple_mtx_t lock;
44 
45    /* On CSF context, set to true if the root CS ring buffer
46     * is managed in userspace. The blob does that, and mesa might use
47     * usermode queues too at some point.
48     */
49    bool usermode_queue;
50 };
51 
52 void pandecode_dump_file_open(struct pandecode_context *ctx);
53 
54 struct pandecode_mapped_memory {
55    struct rb_node node;
56    size_t length;
57    void *addr;
58    uint64_t gpu_va;
59    bool ro;
60    char name[32];
61 };
62 
63 char *pointer_as_memory_reference(struct pandecode_context *ctx, uint64_t ptr);
64 
65 struct pandecode_mapped_memory *
66 pandecode_find_mapped_gpu_mem_containing(struct pandecode_context *ctx,
67                                          uint64_t addr);
68 
69 void pandecode_map_read_write(struct pandecode_context *ctx);
70 
71 void pandecode_dump_mappings(struct pandecode_context *ctx);
72 
73 static inline void *
__pandecode_fetch_gpu_mem(struct pandecode_context * ctx,uint64_t gpu_va,size_t size,int line,const char * filename)74 __pandecode_fetch_gpu_mem(struct pandecode_context *ctx, uint64_t gpu_va,
75                           size_t size, int line, const char *filename)
76 {
77    const struct pandecode_mapped_memory *mem =
78       pandecode_find_mapped_gpu_mem_containing(ctx, gpu_va);
79 
80    if (!mem) {
81       fprintf(stderr, "Access to unknown memory %" PRIx64 " in %s:%d\n", gpu_va,
82               filename, line);
83       fflush(ctx->dump_stream);
84       assert(0);
85    }
86 
87    assert(size + (gpu_va - mem->gpu_va) <= mem->length);
88 
89    return mem->addr + gpu_va - mem->gpu_va;
90 }
91 
92 #define pandecode_fetch_gpu_mem(ctx, gpu_va, size)                             \
93    __pandecode_fetch_gpu_mem(ctx, gpu_va, size, __LINE__, __FILE__)
94 
95 /* Returns a validated pointer to mapped GPU memory with the given pointer type,
96  * size automatically determined from the pointer type
97  */
98 #define PANDECODE_PTR(ctx, gpu_va, type)                                       \
99    ((type *)(__pandecode_fetch_gpu_mem(ctx, gpu_va, sizeof(type), __LINE__,    \
100                                        __FILE__)))
101 
102 /* Usage: <variable type> PANDECODE_PTR_VAR(name, gpu_va) */
103 #define PANDECODE_PTR_VAR(ctx, name, gpu_va)                                   \
104    name = __pandecode_fetch_gpu_mem(ctx, gpu_va, sizeof(*name), __LINE__,      \
105                                     __FILE__)
106 
107 void pandecode_validate_buffer(struct pandecode_context *ctx, uint64_t addr,
108                                size_t sz);
109 
110 /* Forward declare for all supported gens to permit thunking */
111 void pandecode_jc_v4(struct pandecode_context *ctx, uint64_t jc_gpu_va,
112                      unsigned gpu_id);
113 void pandecode_jc_v5(struct pandecode_context *ctx, uint64_t jc_gpu_va,
114                      unsigned gpu_id);
115 void pandecode_jc_v6(struct pandecode_context *ctx, uint64_t jc_gpu_va,
116                      unsigned gpu_id);
117 void pandecode_jc_v7(struct pandecode_context *ctx, uint64_t jc_gpu_va,
118                      unsigned gpu_id);
119 void pandecode_jc_v9(struct pandecode_context *ctx, uint64_t jc_gpu_va,
120                      unsigned gpu_id);
121 
122 void pandecode_abort_on_fault_v4(struct pandecode_context *ctx,
123                                  uint64_t jc_gpu_va);
124 void pandecode_abort_on_fault_v5(struct pandecode_context *ctx,
125                                  uint64_t jc_gpu_va);
126 void pandecode_abort_on_fault_v6(struct pandecode_context *ctx,
127                                  uint64_t jc_gpu_va);
128 void pandecode_abort_on_fault_v7(struct pandecode_context *ctx,
129                                  uint64_t jc_gpu_va);
130 void pandecode_abort_on_fault_v9(struct pandecode_context *ctx,
131                                  uint64_t jc_gpu_va);
132 
133 void pandecode_interpret_cs_v10(struct pandecode_context *ctx, uint64_t queue,
134                                 uint32_t size, unsigned gpu_id, uint32_t *regs);
135 void pandecode_cs_binary_v10(struct pandecode_context *ctx, uint64_t bin,
136                              uint32_t bin_size, unsigned gpu_id);
137 void pandecode_cs_trace_v10(struct pandecode_context *ctx, uint64_t trace,
138                             uint32_t trace_size, unsigned gpu_id);
139 
140 /* Logging infrastructure */
141 static void
pandecode_make_indent(struct pandecode_context * ctx)142 pandecode_make_indent(struct pandecode_context *ctx)
143 {
144    for (unsigned i = 0; i < ctx->indent; ++i)
145       fprintf(ctx->dump_stream, "  ");
146 }
147 
148 static inline void PRINTFLIKE(2, 3)
pandecode_log(struct pandecode_context * ctx,const char * format,...)149    pandecode_log(struct pandecode_context *ctx, const char *format, ...)
150 {
151    va_list ap;
152 
153    pandecode_make_indent(ctx);
154    va_start(ap, format);
155    vfprintf(ctx->dump_stream, format, ap);
156    va_end(ap);
157 }
158 
159 static inline void PRINTFLIKE(2, 3)
pandecode_user_msg(struct pandecode_context * ctx,const char * format,...)160    pandecode_user_msg(struct pandecode_context *ctx, const char *format, ...)
161 {
162    va_list ap;
163 
164    simple_mtx_lock(&ctx->lock);
165    pandecode_dump_file_open(ctx);
166    pandecode_make_indent(ctx);
167    va_start(ap, format);
168    vfprintf(ctx->dump_stream, format, ap);
169    va_end(ap);
170    simple_mtx_unlock(&ctx->lock);
171 }
172 
173 static inline void
pandecode_log_cont(struct pandecode_context * ctx,const char * format,...)174 pandecode_log_cont(struct pandecode_context *ctx, const char *format, ...)
175 {
176    va_list ap;
177 
178    va_start(ap, format);
179    vfprintf(ctx->dump_stream, format, ap);
180    va_end(ap);
181 }
182 
183 /* Convenience methods */
184 #define DUMP_UNPACKED(ctx, T, var, ...)                                        \
185    {                                                                           \
186       pandecode_log(ctx, __VA_ARGS__);                                         \
187       pan_print(ctx->dump_stream, T, var, (ctx->indent + 1) * 2);              \
188    }
189 
190 #define DUMP_CL(ctx, T, cl, ...)                                               \
191    {                                                                           \
192       pan_unpack((MALI_##T##_PACKED_T *)(cl), T, temp);                        \
193       DUMP_UNPACKED(ctx, T, temp, __VA_ARGS__);                                \
194    }
195 
196 #define DUMP_SECTION(ctx, A, S, cl, ...)                                       \
197    {                                                                           \
198       pan_section_unpack(cl, A, S, temp);                                      \
199       pandecode_log(ctx, __VA_ARGS__);                                         \
200       pan_section_print(ctx->dump_stream, A, S, temp, (ctx->indent + 1) * 2);  \
201    }
202 
203 #define MAP_ADDR(ctx, T, addr, cl)                                             \
204    const MALI_##T##_PACKED_T *cl =                                             \
205       pandecode_fetch_gpu_mem(ctx, addr, pan_size(T));
206 
207 #define DUMP_ADDR(ctx, T, addr, ...)                                           \
208    {                                                                           \
209       MAP_ADDR(ctx, T, addr, cl)                                               \
210       DUMP_CL(ctx, T, cl, __VA_ARGS__);                                        \
211    }
212 
213 void pandecode_shader_disassemble(struct pandecode_context *ctx,
214                                   uint64_t shader_ptr, unsigned gpu_id);
215 
216 #ifdef PAN_ARCH
217 
218 /* Information about the framebuffer passed back for additional analysis */
219 struct pandecode_fbd {
220    unsigned rt_count;
221    bool has_extra;
222 };
223 
224 struct pandecode_fbd GENX(pandecode_fbd)(struct pandecode_context *ctx,
225                                          uint64_t gpu_va, bool is_fragment,
226                                          unsigned gpu_id);
227 
228 #if PAN_ARCH >= 9
229 void GENX(pandecode_dcd)(struct pandecode_context *ctx,
230                          const struct MALI_DRAW *p, unsigned unused,
231                          unsigned gpu_id);
232 #else
233 void GENX(pandecode_dcd)(struct pandecode_context *ctx,
234                          const struct MALI_DRAW *p, enum mali_job_type job_type,
235                          unsigned gpu_id);
236 #endif
237 
238 #if PAN_ARCH <= 5
239 void GENX(pandecode_texture)(struct pandecode_context *ctx, uint64_t u,
240                              unsigned tex);
241 #else
242 void GENX(pandecode_texture)(struct pandecode_context *ctx,
243                              const struct mali_texture_packed *cl,
244                              unsigned tex);
245 #endif
246 
247 #if PAN_ARCH >= 5
248 uint64_t GENX(pandecode_blend)(struct pandecode_context *ctx,
249                                struct mali_blend_packed *descs, int rt_no,
250                                uint64_t frag_shader);
251 #endif
252 
253 #if PAN_ARCH >= 6
254 void GENX(pandecode_tiler)(struct pandecode_context *ctx, uint64_t gpu_va,
255                            unsigned gpu_id);
256 #endif
257 
258 #if PAN_ARCH >= 9
259 void GENX(pandecode_shader_environment)(struct pandecode_context *ctx,
260                                         const struct MALI_SHADER_ENVIRONMENT *p,
261                                         unsigned gpu_id);
262 
263 void GENX(pandecode_resource_tables)(struct pandecode_context *ctx,
264                                      uint64_t addr, const char *label);
265 
266 void GENX(pandecode_fau)(struct pandecode_context *ctx, uint64_t addr,
267                          unsigned count, const char *name);
268 
269 uint64_t GENX(pandecode_shader)(struct pandecode_context *ctx, uint64_t addr,
270                                 const char *label, unsigned gpu_id);
271 
272 void GENX(pandecode_blend_descs)(struct pandecode_context *ctx, uint64_t blend,
273                                  unsigned count, uint64_t frag_shader,
274                                  unsigned gpu_id);
275 
276 void GENX(pandecode_depth_stencil)(struct pandecode_context *ctx,
277                                    uint64_t addr);
278 #endif
279 
280 #endif
281 
282 #endif /* __MMAP_TRACE_H__ */
283