/* * Copyright (C) 2017-2019 Alyssa Rosenzweig * Copyright (C) 2017-2019 Connor Abbott * Copyright (C) 2019 Collabora, Ltd. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include #include #include #include #include #include #include #include "decode.h" #include "util/macros.h" #include "util/u_math.h" #include "midgard/disassemble.h" #include "bifrost/disassemble.h" #include "pan_encoder.h" #define MEMORY_PROP(obj, p) {\ if (obj->p) { \ char *a = pointer_as_memory_reference(obj->p); \ pandecode_prop("%s = %s", #p, a); \ free(a); \ } \ } #define MEMORY_PROP_DIR(obj, p) {\ if (obj.p) { \ char *a = pointer_as_memory_reference(obj.p); \ pandecode_prop("%s = %s", #p, a); \ free(a); \ } \ } #define DUMP_UNPACKED(T, var, ...) { \ pandecode_log(__VA_ARGS__); \ pan_print(pandecode_dump_stream, T, var, (pandecode_indent + 1) * 2); \ } #define DUMP_CL(T, cl, ...) {\ pan_unpack(cl, T, temp); \ DUMP_UNPACKED(T, temp, __VA_ARGS__); \ } #define DUMP_SECTION(A, S, cl, ...) { \ pan_section_unpack(cl, A, S, temp); \ pandecode_log(__VA_ARGS__); \ pan_section_print(pandecode_dump_stream, A, S, temp, (pandecode_indent + 1) * 2); \ } #define MAP_ADDR(T, addr, cl) \ const uint8_t *cl = 0; \ { \ struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(addr); \ cl = pandecode_fetch_gpu_mem(mapped_mem, addr, MALI_ ## T ## _LENGTH); \ } #define DUMP_ADDR(T, addr, ...) {\ MAP_ADDR(T, addr, cl) \ DUMP_CL(T, cl, __VA_ARGS__); \ } FILE *pandecode_dump_stream; /* Semantic logging type. * * Raw: for raw messages to be printed as is. * Message: for helpful information to be commented out in replays. * Property: for properties of a struct * * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar. */ enum pandecode_log_type { PANDECODE_RAW, PANDECODE_MESSAGE, PANDECODE_PROPERTY }; #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__) #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__) #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__) unsigned pandecode_indent = 0; static void pandecode_make_indent(void) { for (unsigned i = 0; i < pandecode_indent; ++i) fprintf(pandecode_dump_stream, " "); } static void PRINTFLIKE(2, 3) pandecode_log_typed(enum pandecode_log_type type, const char *format, ...) { va_list ap; pandecode_make_indent(); if (type == PANDECODE_MESSAGE) fprintf(pandecode_dump_stream, "// "); else if (type == PANDECODE_PROPERTY) fprintf(pandecode_dump_stream, "."); va_start(ap, format); vfprintf(pandecode_dump_stream, format, ap); va_end(ap); if (type == PANDECODE_PROPERTY) fprintf(pandecode_dump_stream, ",\n"); } static void pandecode_log_cont(const char *format, ...) { va_list ap; va_start(ap, format); vfprintf(pandecode_dump_stream, format, ap); va_end(ap); } /* To check for memory safety issues, validates that the given pointer in GPU * memory is valid, containing at least sz bytes. The goal is to eliminate * GPU-side memory bugs (NULL pointer dereferences, buffer overflows, or buffer * overruns) by statically validating pointers. */ static void pandecode_validate_buffer(mali_ptr addr, size_t sz) { if (!addr) { pandecode_msg("XXX: null pointer deref"); return; } /* Find a BO */ struct pandecode_mapped_memory *bo = pandecode_find_mapped_gpu_mem_containing(addr); if (!bo) { pandecode_msg("XXX: invalid memory dereference\n"); return; } /* Bounds check */ unsigned offset = addr - bo->gpu_va; unsigned total = offset + sz; if (total > bo->length) { pandecode_msg("XXX: buffer overrun. " "Chunk of size %zu at offset %d in buffer of size %zu. " "Overrun by %zu bytes. \n", sz, offset, bo->length, total - bo->length); return; } } /* Midgard's tiler descriptor is embedded within the * larger FBD */ static void pandecode_midgard_tiler_descriptor( const struct mali_midgard_tiler_packed *tp, const struct mali_midgard_tiler_weights_packed *wp, unsigned width, unsigned height, bool is_fragment, bool has_hierarchy) { pan_unpack(tp, MIDGARD_TILER, t); DUMP_UNPACKED(MIDGARD_TILER, t, "Tiler:\n"); MEMORY_PROP_DIR(t, polygon_list); /* The body is offset from the base of the polygon list */ //assert(t->polygon_list_body > t->polygon_list); unsigned body_offset = t.polygon_list_body - t.polygon_list; /* It needs to fit inside the reported size */ //assert(t->polygon_list_size >= body_offset); /* Now that we've sanity checked, we'll try to calculate the sizes * ourselves for comparison */ unsigned ref_header = panfrost_tiler_header_size(width, height, t.hierarchy_mask, has_hierarchy); unsigned ref_size = panfrost_tiler_full_size(width, height, t.hierarchy_mask, has_hierarchy); if (!((ref_header == body_offset) && (ref_size == t.polygon_list_size))) { pandecode_msg("XXX: bad polygon list size (expected %d / 0x%x)\n", ref_header, ref_size); pandecode_prop("polygon_list_size = 0x%x", t.polygon_list_size); pandecode_msg("body offset %d\n", body_offset); } /* The tiler heap has a start and end specified -- it should be * identical to what we have in the BO. The exception is if tiling is * disabled. */ MEMORY_PROP_DIR(t, heap_start); assert(t.heap_end >= t.heap_start); unsigned heap_size = t.heap_end - t.heap_start; /* Tiling is enabled with a special flag */ unsigned hierarchy_mask = t.hierarchy_mask & MALI_MIDGARD_TILER_HIERARCHY_MASK; unsigned tiler_flags = t.hierarchy_mask ^ hierarchy_mask; bool tiling_enabled = hierarchy_mask; if (tiling_enabled) { /* We should also have no other flags */ if (tiler_flags) pandecode_msg("XXX: unexpected tiler %X\n", tiler_flags); } else { /* When tiling is disabled, we should have that flag and no others */ if (tiler_flags != MALI_MIDGARD_TILER_DISABLED) { pandecode_msg("XXX: unexpected tiler flag %X, expected MALI_MIDGARD_TILER_DISABLED\n", tiler_flags); } /* We should also have an empty heap */ if (heap_size) { pandecode_msg("XXX: tiler heap size %d given, expected empty\n", heap_size); } /* Disabled tiling is used only for clear-only jobs, which are * purely FRAGMENT, so we should never see this for * non-FRAGMENT descriptors. */ if (!is_fragment) pandecode_msg("XXX: tiler disabled for non-FRAGMENT job\n"); } /* We've never seen weights used in practice, but we know from the * kernel these fields is there */ pan_unpack(wp, MIDGARD_TILER_WEIGHTS, w); bool nonzero_weights = false; nonzero_weights |= w.weight0 != 0x0; nonzero_weights |= w.weight1 != 0x0; nonzero_weights |= w.weight2 != 0x0; nonzero_weights |= w.weight3 != 0x0; nonzero_weights |= w.weight4 != 0x0; nonzero_weights |= w.weight5 != 0x0; nonzero_weights |= w.weight6 != 0x0; nonzero_weights |= w.weight7 != 0x0; if (nonzero_weights) DUMP_UNPACKED(MIDGARD_TILER_WEIGHTS, w, "Tiler Weights:\n"); } /* Information about the framebuffer passed back for * additional analysis */ struct pandecode_fbd { unsigned width; unsigned height; unsigned rt_count; bool has_extra; }; static struct pandecode_fbd pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); const void *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va); struct pandecode_fbd info = { .has_extra = false, .rt_count = 1 }; pandecode_log("Single-Target Framebuffer:\n"); pandecode_indent++; DUMP_SECTION(SINGLE_TARGET_FRAMEBUFFER, LOCAL_STORAGE, s, "Local Storage:\n"); pan_section_unpack(s, SINGLE_TARGET_FRAMEBUFFER, PARAMETERS, p); DUMP_UNPACKED(SINGLE_TARGET_FRAMEBUFFER_PARAMETERS, p, "Parameters:\n"); const void *t = pan_section_ptr(s, SINGLE_TARGET_FRAMEBUFFER, TILER); const void *w = pan_section_ptr(s, SINGLE_TARGET_FRAMEBUFFER, TILER_WEIGHTS); bool has_hierarchy = !(gpu_id == 0x0720 || gpu_id == 0x0820 || gpu_id == 0x0830); pandecode_midgard_tiler_descriptor(t, w, p.bound_max_x + 1, p.bound_max_y + 1, is_fragment, has_hierarchy); pandecode_indent--; /* Dummy unpack of the padding section to make sure all words are 0. * No need to call print here since the section is supposed to be empty. */ pan_section_unpack(s, SINGLE_TARGET_FRAMEBUFFER, PADDING_1, padding1); pan_section_unpack(s, SINGLE_TARGET_FRAMEBUFFER, PADDING_2, padding2); pandecode_log("\n"); return info; } static void pandecode_compute_fbd(uint64_t gpu_va, int job_no) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); const struct mali_local_storage_packed *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va); DUMP_CL(LOCAL_STORAGE, s, "Local Storage:\n"); } static void pandecode_render_target(uint64_t gpu_va, unsigned job_no, bool is_bifrost, unsigned gpu_id, const struct MALI_MULTI_TARGET_FRAMEBUFFER_PARAMETERS *fb) { pandecode_log("Color Render Targets:\n"); pandecode_indent++; for (int i = 0; i < (fb->render_target_count); i++) { mali_ptr rt_va = gpu_va + i * MALI_RENDER_TARGET_LENGTH; struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(rt_va); const struct mali_render_target_packed *PANDECODE_PTR_VAR(rtp, mem, (mali_ptr) rt_va); DUMP_CL(RENDER_TARGET, rtp, "Color Render Target %d:\n", i); } pandecode_indent--; pandecode_log("\n"); } static void pandecode_mfbd_bifrost_deps(const void *fb, int job_no) { pan_section_unpack(fb, MULTI_TARGET_FRAMEBUFFER, BIFROST_PARAMETERS, params); /* The blob stores all possible sample locations in a single buffer * allocated on startup, and just switches the pointer when switching * MSAA state. For now, we just put the data into the cmdstream, but we * should do something like what the blob does with a real driver. * * There seem to be 32 slots for sample locations, followed by another * 16. The second 16 is just the center location followed by 15 zeros * in all the cases I've identified (maybe shader vs. depth/color * samples?). */ struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(params.sample_locations); const u16 *PANDECODE_PTR_VAR(samples, smem, params.sample_locations); pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no); pandecode_indent++; for (int i = 0; i < 32 + 16; i++) { pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]); } pandecode_indent--; pandecode_log("};\n"); } static struct pandecode_fbd pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, bool is_compute, bool is_bifrost, unsigned gpu_id) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); const void *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va); pan_section_unpack(fb, MULTI_TARGET_FRAMEBUFFER, PARAMETERS, params); struct pandecode_fbd info; if (is_bifrost) pandecode_mfbd_bifrost_deps(fb, job_no); pandecode_log("Multi-Target Framebuffer:\n"); pandecode_indent++; if (is_bifrost) { DUMP_SECTION(MULTI_TARGET_FRAMEBUFFER, BIFROST_PARAMETERS, fb, "Bifrost Params:\n"); } else { DUMP_SECTION(MULTI_TARGET_FRAMEBUFFER, LOCAL_STORAGE, fb, "Local Storage:\n"); } info.width = params.width; info.height = params.height; info.rt_count = params.render_target_count; DUMP_UNPACKED(MULTI_TARGET_FRAMEBUFFER_PARAMETERS, params, "Parameters:\n"); if (!is_compute) { if (is_bifrost) { DUMP_SECTION(MULTI_TARGET_FRAMEBUFFER, BIFROST_TILER_POINTER, fb, "Tiler Pointer"); } else { const void *t = pan_section_ptr(fb, MULTI_TARGET_FRAMEBUFFER, TILER); const void *w = pan_section_ptr(fb, MULTI_TARGET_FRAMEBUFFER, TILER_WEIGHTS); pandecode_midgard_tiler_descriptor(t, w, params.width, params.height, is_fragment, true); } } else { pandecode_msg("XXX: skipping compute MFBD, fixme\n"); } if (is_bifrost) { pan_section_unpack(fb, MULTI_TARGET_FRAMEBUFFER, BIFROST_PADDING, padding); } pandecode_indent--; pandecode_log("\n"); gpu_va += MALI_MULTI_TARGET_FRAMEBUFFER_LENGTH; info.has_extra = params.has_zs_crc_extension; if (info.has_extra) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); const struct mali_zs_crc_extension_packed *PANDECODE_PTR_VAR(zs_crc, mem, (mali_ptr)gpu_va); DUMP_CL(ZS_CRC_EXTENSION, zs_crc, "ZS CRC Extension:\n"); pandecode_log("\n"); gpu_va += MALI_ZS_CRC_EXTENSION_LENGTH; } if (is_fragment) pandecode_render_target(gpu_va, job_no, is_bifrost, gpu_id, ¶ms); return info; } static void pandecode_attributes(const struct pandecode_mapped_memory *mem, mali_ptr addr, int job_no, char *suffix, int count, bool varying, enum mali_job_type job_type) { char *prefix = varying ? "Varying" : "Attribute"; assert(addr); if (!count) { pandecode_msg("warn: No %s records\n", prefix); return; } MAP_ADDR(ATTRIBUTE_BUFFER, addr, cl); for (int i = 0; i < count; ++i) { pan_unpack(cl + i * MALI_ATTRIBUTE_BUFFER_LENGTH, ATTRIBUTE_BUFFER, temp); DUMP_UNPACKED(ATTRIBUTE_BUFFER, temp, "%s:\n", prefix); if (temp.type != MALI_ATTRIBUTE_TYPE_1D_NPOT_DIVISOR) continue; pan_unpack(cl + (i + 1) * MALI_ATTRIBUTE_BUFFER_LENGTH, ATTRIBUTE_BUFFER_CONTINUATION_NPOT, temp2); pan_print(pandecode_dump_stream, ATTRIBUTE_BUFFER_CONTINUATION_NPOT, temp2, (pandecode_indent + 1) * 2); } pandecode_log("\n"); } static mali_ptr pandecode_shader_address(const char *name, mali_ptr ptr) { /* TODO: Decode flags */ mali_ptr shader_ptr = ptr & ~15; char *a = pointer_as_memory_reference(shader_ptr); pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15)); free(a); return shader_ptr; } /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */ static mali_ptr pandecode_bifrost_blend(void *descs, int job_no, int rt_no, mali_ptr frag_shader) { pan_unpack(descs + (rt_no * MALI_BLEND_LENGTH), BLEND, b); DUMP_UNPACKED(BLEND, b, "Blend RT %d:\n", rt_no); if (b.bifrost.internal.mode != MALI_BIFROST_BLEND_MODE_SHADER) return 0; return (frag_shader & 0xFFFFFFFF00000000ULL) | b.bifrost.internal.shader.pc; } static mali_ptr pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no) { pan_unpack(descs + (rt_no * MALI_BLEND_LENGTH), BLEND, b); DUMP_UNPACKED(BLEND, b, "Blend RT %d:\n", rt_no); return b.midgard.blend_shader ? (b.midgard.shader_pc & ~0xf) : 0; } /* Attributes and varyings have descriptor records, which contain information * about their format and ordering with the attribute/varying buffers. We'll * want to validate that the combinations specified are self-consistent. */ static int pandecode_attribute_meta(int count, mali_ptr attribute, bool varying, char *suffix) { for (int i = 0; i < count; ++i, attribute += MALI_ATTRIBUTE_LENGTH) DUMP_ADDR(ATTRIBUTE, attribute, "%s:\n", varying ? "Varying" : "Attribute"); pandecode_log("\n"); return count; } /* return bits [lo, hi) of word */ static u32 bits(u32 word, u32 lo, u32 hi) { if (hi - lo >= 32) return word; // avoid undefined behavior with the shift return (word >> lo) & ((1 << (hi - lo)) - 1); } static void pandecode_invocation(const void *i, bool graphics) { /* Decode invocation_count. See the comment before the definition of * invocation_count for an explanation. */ pan_unpack(i, INVOCATION, invocation); unsigned size_x = bits(invocation.invocations, 0, invocation.size_y_shift) + 1; unsigned size_y = bits(invocation.invocations, invocation.size_y_shift, invocation.size_z_shift) + 1; unsigned size_z = bits(invocation.invocations, invocation.size_z_shift, invocation.workgroups_x_shift) + 1; unsigned groups_x = bits(invocation.invocations, invocation.workgroups_x_shift, invocation.workgroups_y_shift) + 1; unsigned groups_y = bits(invocation.invocations, invocation.workgroups_y_shift, invocation.workgroups_z_shift) + 1; unsigned groups_z = bits(invocation.invocations, invocation.workgroups_z_shift, 32) + 1; /* Even though we have this decoded, we want to ensure that the * representation is "unique" so we don't lose anything by printing only * the final result. More specifically, we need to check that we were * passed something in canonical form, since the definition per the * hardware is inherently not unique. How? Well, take the resulting * decode and pack it ourselves! If it is bit exact with what we * decoded, we're good to go. */ struct mali_invocation_packed ref; panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, graphics); if (memcmp(&ref, i, sizeof(ref))) { pandecode_msg("XXX: non-canonical workgroups packing\n"); DUMP_UNPACKED(INVOCATION, invocation, "Invocation:\n") } /* Regardless, print the decode */ pandecode_log("Invocation (%d, %d, %d) x (%d, %d, %d)\n", size_x, size_y, size_z, groups_x, groups_y, groups_z); } static void pandecode_primitive(const void *p) { pan_unpack(p, PRIMITIVE, primitive); DUMP_UNPACKED(PRIMITIVE, primitive, "Primitive:\n"); /* Validate an index buffer is present if we need one. TODO: verify * relationship between invocation_count and index_count */ if (primitive.indices) { /* Grab the size */ unsigned size = (primitive.index_type == MALI_INDEX_TYPE_UINT32) ? sizeof(uint32_t) : primitive.index_type; /* Ensure we got a size, and if so, validate the index buffer * is large enough to hold a full set of indices of the given * size */ if (!size) pandecode_msg("XXX: index size missing\n"); else pandecode_validate_buffer(primitive.indices, primitive.index_count * size); } else if (primitive.index_type) pandecode_msg("XXX: unexpected index size\n"); } static void pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no) { struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs); uint64_t *PANDECODE_PTR_VAR(ubufs, umem, pubufs); for (int i = 0; i < ubufs_count; i++) { unsigned size = (ubufs[i] & ((1 << 10) - 1)) * 16; mali_ptr addr = (ubufs[i] >> 10) << 2; pandecode_validate_buffer(addr, size); char *ptr = pointer_as_memory_reference(addr); pandecode_log("ubuf_%d[%u] = %s;\n", i, size, ptr); free(ptr); } pandecode_log("\n"); } static void pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count) { pandecode_validate_buffer(uniforms, uniform_count * 16); char *ptr = pointer_as_memory_reference(uniforms); pandecode_log("vec4 uniforms[%u] = %s;\n", uniform_count, ptr); free(ptr); pandecode_log("\n"); } static const char * shader_type_for_job(unsigned type) { switch (type) { case MALI_JOB_TYPE_VERTEX: return "VERTEX"; case MALI_JOB_TYPE_TILER: return "FRAGMENT"; case MALI_JOB_TYPE_COMPUTE: return "COMPUTE"; default: return "UNKNOWN"; } } static unsigned shader_id = 0; static struct midgard_disasm_stats pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type, bool is_bifrost, unsigned gpu_id) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr); uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr); /* Compute maximum possible size */ size_t sz = mem->length - (shader_ptr - mem->gpu_va); /* Print some boilerplate to clearly denote the assembly (which doesn't * obey indentation rules), and actually do the disassembly! */ pandecode_log_cont("\n\n"); struct midgard_disasm_stats stats; if (is_bifrost) { disassemble_bifrost(pandecode_dump_stream, code, sz, true); /* TODO: Extend stats to Bifrost */ stats.texture_count = -128; stats.sampler_count = -128; stats.attribute_count = -128; stats.varying_count = -128; stats.uniform_count = -128; stats.uniform_buffer_count = -128; stats.work_count = -128; stats.instruction_count = 0; stats.bundle_count = 0; stats.quadword_count = 0; stats.helper_invocations = false; } else { stats = disassemble_midgard(pandecode_dump_stream, code, sz, gpu_id, type == MALI_JOB_TYPE_TILER ? MESA_SHADER_FRAGMENT : MESA_SHADER_VERTEX); } unsigned nr_threads = (stats.work_count <= 4) ? 4 : (stats.work_count <= 8) ? 2 : 1; pandecode_log_cont("shader%d - MESA_SHADER_%s shader: " "%u inst, %u bundles, %u quadwords, " "%u registers, %u threads, 0 loops, 0:0 spills:fills\n\n\n", shader_id++, shader_type_for_job(type), stats.instruction_count, stats.bundle_count, stats.quadword_count, stats.work_count, nr_threads); return stats; } static void pandecode_texture_payload(mali_ptr payload, enum mali_texture_dimension dim, enum mali_texture_layout layout, bool manual_stride, uint8_t levels, uint16_t depth, uint16_t array_size, struct pandecode_mapped_memory *tmem) { pandecode_log(".payload = {\n"); pandecode_indent++; /* A bunch of bitmap pointers follow. * We work out the correct number, * based on the mipmap/cubemap * properties, but dump extra * possibilities to futureproof */ int bitmap_count = levels + 1; /* Miptree for each face */ if (dim == MALI_TEXTURE_DIMENSION_CUBE) bitmap_count *= 6; /* Array of layers */ bitmap_count *= depth; /* Array of textures */ bitmap_count *= array_size; /* Stride for each element */ if (manual_stride) bitmap_count *= 2; mali_ptr *pointers_and_strides = pandecode_fetch_gpu_mem(tmem, payload, sizeof(mali_ptr) * bitmap_count); for (int i = 0; i < bitmap_count; ++i) { /* How we dump depends if this is a stride or a pointer */ if (manual_stride && (i & 1)) { /* signed 32-bit snuck in as a 64-bit pointer */ uint64_t stride_set = pointers_and_strides[i]; uint32_t clamped_stride = stride_set; int32_t stride = clamped_stride; assert(stride_set == clamped_stride); pandecode_log("(mali_ptr) %d /* stride */, \n", stride); } else { char *a = pointer_as_memory_reference(pointers_and_strides[i]); pandecode_log("%s, \n", a); free(a); } } pandecode_indent--; pandecode_log("},\n"); } static void pandecode_texture(mali_ptr u, struct pandecode_mapped_memory *tmem, unsigned job_no, unsigned tex) { struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(u); const uint8_t *cl = pandecode_fetch_gpu_mem(mapped_mem, u, MALI_MIDGARD_TEXTURE_LENGTH); pan_unpack(cl, MIDGARD_TEXTURE, temp); DUMP_UNPACKED(MIDGARD_TEXTURE, temp, "Texture:\n") pandecode_indent++; pandecode_texture_payload(u + MALI_MIDGARD_TEXTURE_LENGTH, temp.dimension, temp.texel_ordering, temp.manual_stride, temp.levels, temp.depth, temp.array_size, mapped_mem); pandecode_indent--; } static void pandecode_bifrost_texture( const void *cl, unsigned job_no, unsigned tex) { pan_unpack(cl, BIFROST_TEXTURE, temp); DUMP_UNPACKED(BIFROST_TEXTURE, temp, "Texture:\n") struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(temp.surfaces); pandecode_indent++; pandecode_texture_payload(temp.surfaces, temp.dimension, temp.texel_ordering, true, temp.levels, 1, 1, tmem); pandecode_indent--; } /* For shader properties like texture_count, we have a claimed property in the shader_meta, and the actual Truth from static analysis (this may just be an upper limit). We validate accordingly */ static void pandecode_shader_prop(const char *name, unsigned claim, signed truth, bool fuzzy) { /* Nothing to do */ if (claim == truth) return; if (fuzzy && (truth < 0)) pandecode_msg("XXX: fuzzy %s, claimed %d, expected %d\n", name, claim, truth); if ((truth >= 0) && !fuzzy) { pandecode_msg("%s: expected %s = %d, claimed %u\n", (truth < claim) ? "warn" : "XXX", name, truth, claim); } else if ((claim > -truth) && !fuzzy) { pandecode_msg("XXX: expected %s <= %u, claimed %u\n", name, -truth, claim); } else if (fuzzy && (claim < truth)) pandecode_msg("XXX: expected %s >= %u, claimed %u\n", name, truth, claim); pandecode_log(".%s = %" PRId16, name, claim); if (fuzzy) pandecode_log_cont(" /* %u used */", truth); pandecode_log_cont(",\n"); } static void pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type, bool is_bifrost, unsigned gpu_id) { struct midgard_disasm_stats stats = pandecode_shader_disassemble(shader, job_no, job_type, is_bifrost, gpu_id); bool has_texture = (stats.texture_count > 0); bool has_sampler = (stats.sampler_count > 0); bool has_attribute = (stats.attribute_count > 0); bool has_varying = (stats.varying_count > 0); bool has_uniform = (stats.uniform_count > 0); bool has_ubo = (stats.uniform_buffer_count > 0); if (has_texture || has_sampler) pandecode_msg("XXX: blend shader accessing textures\n"); if (has_attribute || has_varying) pandecode_msg("XXX: blend shader accessing interstage\n"); if (has_uniform || has_ubo) pandecode_msg("XXX: blend shader accessing uniforms\n"); } static void pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no, bool is_bifrost) { struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(textures); if (!mmem) return; pandecode_log("Textures %"PRIx64"_%d:\n", textures, job_no); pandecode_indent++; if (is_bifrost) { const void *cl = pandecode_fetch_gpu_mem(mmem, textures, MALI_BIFROST_TEXTURE_LENGTH * texture_count); for (unsigned tex = 0; tex < texture_count; ++tex) { pandecode_bifrost_texture(cl + MALI_BIFROST_TEXTURE_LENGTH * tex, job_no, tex); } } else { mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures); for (int tex = 0; tex < texture_count; ++tex) { mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr)); char *a = pointer_as_memory_reference(*u); pandecode_log("%s,\n", a); free(a); } /* Now, finally, descend down into the texture descriptor */ for (unsigned tex = 0; tex < texture_count; ++tex) { mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr)); struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u); if (tmem) pandecode_texture(*u, tmem, job_no, tex); } } pandecode_indent--; pandecode_log("\n"); } static void pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no, bool is_bifrost) { pandecode_log("Samplers %"PRIx64"_%d:\n", samplers, job_no); pandecode_indent++; for (int i = 0; i < sampler_count; ++i) { if (is_bifrost) { DUMP_ADDR(BIFROST_SAMPLER, samplers + (MALI_BIFROST_SAMPLER_LENGTH * i), "Sampler %d:\n", i); } else { DUMP_ADDR(MIDGARD_SAMPLER, samplers + (MALI_MIDGARD_SAMPLER_LENGTH * i), "Sampler %d:\n", i); } } pandecode_indent--; pandecode_log("\n"); } static void pandecode_vertex_tiler_postfix_pre( const struct MALI_DRAW *p, int job_no, enum mali_job_type job_type, char *suffix, bool is_bifrost, unsigned gpu_id) { struct pandecode_mapped_memory *attr_mem; struct pandecode_fbd fbd_info = { /* Default for Bifrost */ .rt_count = 1 }; if (is_bifrost) pandecode_compute_fbd(p->fbd & ~1, job_no); else if (p->fbd & MALI_FBD_TAG_IS_MFBD) fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->fbd) & ~MALI_FBD_TAG_MASK, job_no, false, job_type == MALI_JOB_TYPE_COMPUTE, is_bifrost, gpu_id); else if (job_type == MALI_JOB_TYPE_COMPUTE) pandecode_compute_fbd((u64) (uintptr_t) p->fbd, job_no); else fbd_info = pandecode_sfbd((u64) (uintptr_t) p->fbd, job_no, false, gpu_id); int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0; int texture_count = 0, sampler_count = 0; if (p->state) { struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->state); uint32_t *cl = pandecode_fetch_gpu_mem(smem, p->state, MALI_RENDERER_STATE_LENGTH); /* Disassemble ahead-of-time to get stats. Initialize with * stats for the missing-shader case so we get validation * there, too */ struct midgard_disasm_stats info = { .texture_count = 0, .sampler_count = 0, .attribute_count = 0, .varying_count = 0, .work_count = 1, .uniform_count = -128, .uniform_buffer_count = 0 }; pan_unpack(cl, RENDERER_STATE, state); if (state.shader.shader & ~0xF) info = pandecode_shader_disassemble(state.shader.shader & ~0xF, job_no, job_type, is_bifrost, gpu_id); DUMP_UNPACKED(RENDERER_STATE, state, "State:\n"); pandecode_indent++; /* Save for dumps */ attribute_count = state.shader.attribute_count; varying_count = state.shader.varying_count; texture_count = state.shader.texture_count; sampler_count = state.shader.sampler_count; uniform_buffer_count = state.properties.uniform_buffer_count; if (is_bifrost) uniform_count = state.preload.uniform_count; else uniform_count = state.properties.midgard.uniform_count; pandecode_shader_prop("texture_count", texture_count, info.texture_count, false); pandecode_shader_prop("sampler_count", sampler_count, info.sampler_count, false); pandecode_shader_prop("attribute_count", attribute_count, info.attribute_count, false); pandecode_shader_prop("varying_count", varying_count, info.varying_count, false); if (is_bifrost) DUMP_UNPACKED(PRELOAD, state.preload, "Preload:\n"); if (!is_bifrost) { /* TODO: Blend shaders routing/disasm */ pandecode_log("SFBD Blend:\n"); pandecode_indent++; if (state.multisample_misc.sfbd_blend_shader) { pandecode_shader_address("Shader", state.sfbd_blend_shader); } else { DUMP_UNPACKED(BLEND_EQUATION, state.sfbd_blend_equation, "Equation:\n"); pandecode_prop("Constant = %f", state.sfbd_blend_constant); } pandecode_indent--; pandecode_log("\n"); mali_ptr shader = state.sfbd_blend_shader & ~0xF; if (state.multisample_misc.sfbd_blend_shader && shader) pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id); } pandecode_indent--; pandecode_log("\n"); /* MRT blend fields are used whenever MFBD is used, with * per-RT descriptors */ if (job_type == MALI_JOB_TYPE_TILER && (is_bifrost || p->fbd & MALI_FBD_TAG_IS_MFBD)) { void* blend_base = ((void *) cl) + MALI_RENDERER_STATE_LENGTH; for (unsigned i = 0; i < fbd_info.rt_count; i++) { mali_ptr shader = 0; if (is_bifrost) shader = pandecode_bifrost_blend(blend_base, job_no, i, state.shader.shader); else shader = pandecode_midgard_blend_mrt(blend_base, job_no, i); if (shader & ~0xF) pandecode_blend_shader_disassemble(shader, job_no, job_type, is_bifrost, gpu_id); } } } else pandecode_msg("XXX: missing shader descriptor\n"); if (p->viewport) { DUMP_ADDR(VIEWPORT, p->viewport, "Viewport:\n"); pandecode_log("\n"); } unsigned max_attr_index = 0; if (p->attributes) max_attr_index = pandecode_attribute_meta(attribute_count, p->attributes, false, suffix); if (p->attribute_buffers) { attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attribute_buffers); pandecode_attributes(attr_mem, p->attribute_buffers, job_no, suffix, max_attr_index, false, job_type); } if (p->varyings) { varying_count = pandecode_attribute_meta(varying_count, p->varyings, true, suffix); } if (p->varying_buffers) { attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varying_buffers); pandecode_attributes(attr_mem, p->varying_buffers, job_no, suffix, varying_count, true, job_type); } if (p->uniform_buffers) { if (uniform_buffer_count) pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no); else pandecode_msg("warn: UBOs specified but not referenced\n"); } else if (uniform_buffer_count) pandecode_msg("XXX: UBOs referenced but not specified\n"); /* We don't want to actually dump uniforms, but we do need to validate * that the counts we were given are sane */ if (p->push_uniforms) { if (uniform_count) pandecode_uniforms(p->push_uniforms, uniform_count); else pandecode_msg("warn: Uniforms specified but not referenced\n"); } else if (uniform_count) pandecode_msg("XXX: Uniforms referenced but not specified\n"); if (p->textures) pandecode_textures(p->textures, texture_count, job_no, is_bifrost); if (p->samplers) pandecode_samplers(p->samplers, sampler_count, job_no, is_bifrost); } static void pandecode_bifrost_tiler_heap(mali_ptr gpu_va, int job_no) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); pan_unpack(PANDECODE_PTR(mem, gpu_va, void), BIFROST_TILER_HEAP, h); DUMP_UNPACKED(BIFROST_TILER_HEAP, h, "Bifrost Tiler Heap:\n"); } static void pandecode_bifrost_tiler(mali_ptr gpu_va, int job_no) { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); pan_unpack(PANDECODE_PTR(mem, gpu_va, void), BIFROST_TILER, t); pandecode_bifrost_tiler_heap(t.heap, job_no); DUMP_UNPACKED(BIFROST_TILER, t, "Bifrost Tiler:\n"); pandecode_indent++; if (t.hierarchy_mask != 0xa && t.hierarchy_mask != 0x14 && t.hierarchy_mask != 0x28 && t.hierarchy_mask != 0x50 && t.hierarchy_mask != 0xa0) pandecode_prop("XXX: Unexpected hierarchy_mask (not 0xa, 0x14, 0x28, 0x50 or 0xa0)!"); pandecode_indent--; } static void pandecode_primitive_size(const void *s, bool constant) { pan_unpack(s, PRIMITIVE_SIZE, ps); if (ps.size_array == 0x0) return; DUMP_UNPACKED(PRIMITIVE_SIZE, ps, "Primitive Size:\n") } static void pandecode_vertex_compute_geometry_job(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, bool is_bifrost, unsigned gpu_id) { struct mali_compute_job_packed *PANDECODE_PTR_VAR(p, mem, job); pan_section_unpack(p, COMPUTE_JOB, DRAW, draw); pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", is_bifrost, gpu_id); pandecode_log("Vertex Job Payload:\n"); pandecode_indent++; pandecode_invocation(pan_section_ptr(p, COMPUTE_JOB, INVOCATION), h->type != MALI_JOB_TYPE_COMPUTE); DUMP_SECTION(COMPUTE_JOB, PARAMETERS, p, "Vertex Job Parameters:\n"); DUMP_UNPACKED(DRAW, draw, "Draw:\n"); pandecode_indent--; pandecode_log("\n"); } static void pandecode_tiler_job_bfr(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id) { struct mali_bifrost_tiler_job_packed *PANDECODE_PTR_VAR(p, mem, job); pan_section_unpack(p, BIFROST_TILER_JOB, DRAW, draw); pan_section_unpack(p, BIFROST_TILER_JOB, TILER, tiler_ptr); pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", true, gpu_id); pandecode_log("Tiler Job Payload:\n"); pandecode_indent++; pandecode_bifrost_tiler(tiler_ptr.address, job_no); pandecode_invocation(pan_section_ptr(p, BIFROST_TILER_JOB, INVOCATION), true); pandecode_primitive(pan_section_ptr(p, BIFROST_TILER_JOB, PRIMITIVE)); /* TODO: gl_PointSize on Bifrost */ pandecode_primitive_size(pan_section_ptr(p, BIFROST_TILER_JOB, PRIMITIVE_SIZE), true); pan_section_unpack(p, BIFROST_TILER_JOB, PADDING, padding); DUMP_UNPACKED(DRAW, draw, "Draw:\n"); pandecode_indent--; pandecode_log("\n"); } static void pandecode_tiler_job_mdg(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id) { struct mali_midgard_tiler_job_packed *PANDECODE_PTR_VAR(p, mem, job); pan_section_unpack(p, MIDGARD_TILER_JOB, DRAW, draw); pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", false, gpu_id); pandecode_log("Tiler Job Payload:\n"); pandecode_indent++; pandecode_invocation(pan_section_ptr(p, MIDGARD_TILER_JOB, INVOCATION), true); pandecode_primitive(pan_section_ptr(p, MIDGARD_TILER_JOB, PRIMITIVE)); DUMP_UNPACKED(DRAW, draw, "Draw:\n"); pan_section_unpack(p, MIDGARD_TILER_JOB, PRIMITIVE, primitive); pandecode_primitive_size(pan_section_ptr(p, MIDGARD_TILER_JOB, PRIMITIVE_SIZE), primitive.point_size_array_format == MALI_POINT_SIZE_ARRAY_FORMAT_NONE); pandecode_indent--; pandecode_log("\n"); } static void pandecode_fragment_job(const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, bool is_bifrost, unsigned gpu_id) { struct mali_fragment_job_packed *PANDECODE_PTR_VAR(p, mem, job); pan_section_unpack(p, FRAGMENT_JOB, PAYLOAD, s); bool is_mfbd = s.framebuffer & MALI_FBD_TAG_IS_MFBD; if (!is_mfbd && is_bifrost) pandecode_msg("XXX: Bifrost fragment must use MFBD\n"); struct pandecode_fbd info; if (is_mfbd) info = pandecode_mfbd_bfr(s.framebuffer & ~MALI_FBD_TAG_MASK, job_no, true, false, is_bifrost, gpu_id); else info = pandecode_sfbd(s.framebuffer & ~MALI_FBD_TAG_MASK, job_no, true, gpu_id); /* Compute the tag for the tagged pointer. This contains the type of * FBD (MFBD/SFBD), and in the case of an MFBD, information about which * additional structures follow the MFBD header (an extra payload or * not, as well as a count of render targets) */ unsigned expected_tag = is_mfbd ? MALI_FBD_TAG_IS_MFBD : 0; if (is_mfbd) { if (info.has_extra) expected_tag |= MALI_FBD_TAG_HAS_ZS_RT; expected_tag |= (MALI_POSITIVE(info.rt_count) << 2); } /* Extract tile coordinates */ unsigned min_x = s.bound_min_x << MALI_TILE_SHIFT; unsigned min_y = s.bound_min_y << MALI_TILE_SHIFT; unsigned max_x = s.bound_max_x << MALI_TILE_SHIFT; unsigned max_y = s.bound_max_y << MALI_TILE_SHIFT; /* Validate the coordinates are well-ordered */ if (min_x > max_x) pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x); if (min_y > max_y) pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x); /* Validate the coordinates fit inside the framebuffer. We use floor, * rather than ceil, for the max coordinates, since the tile * coordinates for something like an 800x600 framebuffer will actually * resolve to 800x608, which would otherwise trigger a Y-overflow */ if (max_x + 1 > info.width) pandecode_msg("XXX: tile coordinates overflow in X direction\n"); if (max_y + 1 > info.height) pandecode_msg("XXX: tile coordinates overflow in Y direction\n"); /* After validation, we print */ DUMP_UNPACKED(FRAGMENT_JOB_PAYLOAD, s, "Fragment Job Payload:\n"); /* The FBD is a tagged pointer */ unsigned tag = (s.framebuffer & MALI_FBD_TAG_MASK); if (tag != expected_tag) pandecode_msg("XXX: expected FBD tag %X but got %X\n", expected_tag, tag); pandecode_log("\n"); } static void pandecode_write_value_job(const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no) { struct mali_write_value_job_packed *PANDECODE_PTR_VAR(p, mem, job); pan_section_unpack(p, WRITE_VALUE_JOB, PAYLOAD, u); DUMP_SECTION(WRITE_VALUE_JOB, PAYLOAD, p, "Write Value Payload:\n"); pandecode_log("\n"); } /* Entrypoint to start tracing. jc_gpu_va is the GPU address for the first job * in the chain; later jobs are found by walking the chain. Bifrost is, well, * if it's bifrost or not. GPU ID is the more finegrained ID (at some point, we * might wish to combine this with the bifrost parameter) because some details * are model-specific even within a particular architecture. Minimal traces * *only* examine the job descriptors, skipping printing entirely if there is * no faults, and only descends into the payload if there are faults. This is * useful for looking for faults without the overhead of invasive traces. */ void pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal) { pandecode_dump_file_open(); unsigned job_descriptor_number = 0; mali_ptr next_job = 0; do { struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(jc_gpu_va); pan_unpack(PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_header_packed), JOB_HEADER, h); next_job = h.next; int job_no = job_descriptor_number++; /* If the job is good to go, skip it in minimal mode */ if (minimal && (h.exception_status == 0x0 || h.exception_status == 0x1)) continue; DUMP_UNPACKED(JOB_HEADER, h, "Job Header:\n"); pandecode_log("\n"); switch (h.type) { case MALI_JOB_TYPE_WRITE_VALUE: pandecode_write_value_job(mem, jc_gpu_va, job_no); break; case MALI_JOB_TYPE_TILER: if (bifrost) pandecode_tiler_job_bfr(&h, mem, jc_gpu_va, job_no, gpu_id); else pandecode_tiler_job_mdg(&h, mem, jc_gpu_va, job_no, gpu_id); break; case MALI_JOB_TYPE_VERTEX: case MALI_JOB_TYPE_COMPUTE: pandecode_vertex_compute_geometry_job(&h, mem, jc_gpu_va, job_no, bifrost, gpu_id); break; case MALI_JOB_TYPE_FRAGMENT: pandecode_fragment_job(mem, jc_gpu_va, job_no, bifrost, gpu_id); break; default: break; } } while ((jc_gpu_va = next_job)); pandecode_map_read_write(); }