further progress on disassembly cache; instead of caching the text via the disassembly cache, simply submit into the hash store; this allows the disassembly cache to be composed with other caches, like the text analysis cache. + plug this system into the disassembly view rule, so now we can disassemble arbitrary pointers in the watch window

This commit is contained in:
Ryan Fleury
2024-03-28 11:36:57 -07:00
parent c91cd69536
commit d3570e8147
4 changed files with 231 additions and 88 deletions
+94 -81
View File
@@ -1,6 +1,18 @@
// Copyright (c) 2024 Epic Games Tools
// Licensed under the MIT license (https://opensource.org/license/mit/)
////////////////////////////////
//~ rjf: Third Party Includes
#include "third_party/udis86/config.h"
#include "third_party/udis86/udis86.h"
#include "third_party/udis86/libudis86/decode.c"
#include "third_party/udis86/libudis86/itab.c"
#include "third_party/udis86/libudis86/syn-att.c"
#include "third_party/udis86/libudis86/syn-intel.c"
#include "third_party/udis86/libudis86/syn.c"
#include "third_party/udis86/libudis86/udis86.c"
////////////////////////////////
//~ rjf: Instruction Type Functions
@@ -62,84 +74,6 @@ dasm_inst_array_code_off_from_idx(DASM_InstArray *array, U64 idx)
return off;
}
////////////////////////////////
//~ rjf: Disassembly Decoding Function
#include "third_party/udis86/config.h"
#include "third_party/udis86/udis86.h"
#include "third_party/udis86/libudis86/decode.c"
#include "third_party/udis86/libudis86/itab.c"
#include "third_party/udis86/libudis86/syn-att.c"
#include "third_party/udis86/libudis86/syn-intel.c"
#include "third_party/udis86/libudis86/syn.c"
#include "third_party/udis86/libudis86/udis86.c"
internal DASM_Info
dasm_info_from_arch_addr_data(Arena *arena, U64 *bytes_processed_counter, Architecture arch, U64 addr, String8 data)
{
Temp scratch = scratch_begin(&arena, 1);
//- rjf: decode
DASM_InstChunkList inst_list = {0};
String8List inst_strings = {0};
switch(arch)
{
default:{}break;
//- rjf: x86/x64 decoding
case Architecture_x64:
case Architecture_x86:
{
// rjf: grab context
struct ud udc;
ud_init(&udc);
ud_set_mode(&udc, bit_size_from_arch(arch));
ud_set_pc(&udc, addr);
ud_set_input_buffer(&udc, data.str, data.size);
ud_set_vendor(&udc, UD_VENDOR_ANY);
ud_set_syntax(&udc, UD_SYN_INTEL);
// rjf: disassemble
U64 byte_process_start_off = 0;
for(U64 off = 0; off < data.size;)
{
// rjf: disassemble one instruction
U64 size = ud_disassemble(&udc);
if(size == 0)
{
break;
}
// rjf: analyze
struct ud_operand *first_op = (struct ud_operand *)ud_insn_opr(&udc, 0);
U64 rel_voff = (first_op != 0 && first_op->type == UD_OP_JIMM) ? ud_syn_rel_target(&udc, first_op) : 0;
// rjf: push
String8 string = push_str8f(scratch.arena, "%s\n", udc.asm_buf);
DASM_Inst inst = {off, rel_voff, r1u64(inst_strings.total_size, inst_strings.total_size+string.size-1)};
dasm_inst_chunk_list_push(scratch.arena, &inst_list, 1024, &inst);
str8_list_push(scratch.arena, &inst_strings, string);
// rjf: increment
off += size;
if(bytes_processed_counter != 0 && (off-byte_process_start_off >= 1000))
{
ins_atomic_u64_add_eval(bytes_processed_counter, (off-byte_process_start_off));
byte_process_start_off = off;
}
}
}break;
}
//- rjf: fill result
DASM_Info info = {0};
info.text = str8_list_join(arena, &inst_strings, 0);
info.insts = dasm_inst_array_from_chunk_list(arena, &inst_list);
scratch_end(scratch);
return info;
}
////////////////////////////////
//~ rjf: Main Layer Initialization
@@ -406,6 +340,8 @@ dasm_parse_thread__entry_point(void *p)
ThreadNameF("[dasm] parse thread #%I64u", (U64)p);
for(;;)
{
Temp scratch = scratch_begin(0, 0);
//- rjf: get next request
U128 hash = {0};
U64 addr = 0;
@@ -440,13 +376,89 @@ dasm_parse_thread__entry_point(void *p)
data = hs_data_from_hash(hs_scope, hash);
}
//- rjf: data -> disassembly info
//- rjf: data * arch * addr -> decode artifacts
DASM_InstChunkList inst_list = {0};
String8List inst_strings = {0};
if(got_task)
{
switch(arch)
{
default:{}break;
//- rjf: x86/x64 decoding
case Architecture_x64:
case Architecture_x86:
{
// rjf: grab context
struct ud udc;
ud_init(&udc);
ud_set_mode(&udc, bit_size_from_arch(arch));
ud_set_pc(&udc, addr);
ud_set_input_buffer(&udc, data.str, data.size);
ud_set_vendor(&udc, UD_VENDOR_ANY);
ud_set_syntax(&udc, UD_SYN_INTEL);
// rjf: disassemble
U64 byte_process_start_off = 0;
for(U64 off = 0; off < data.size;)
{
// rjf: disassemble one instruction
U64 size = ud_disassemble(&udc);
if(size == 0)
{
break;
}
// rjf: analyze
struct ud_operand *first_op = (struct ud_operand *)ud_insn_opr(&udc, 0);
U64 rel_voff = (first_op != 0 && first_op->type == UD_OP_JIMM) ? ud_syn_rel_target(&udc, first_op) : 0;
// rjf: push
String8 string = push_str8f(scratch.arena, "%s", udc.asm_buf);
DASM_Inst inst = {off, rel_voff, r1u64(inst_strings.total_size + inst_strings.node_count,
inst_strings.total_size + inst_strings.node_count + string.size)};
dasm_inst_chunk_list_push(scratch.arena, &inst_list, 1024, &inst);
str8_list_push(scratch.arena, &inst_strings, string);
// rjf: increment
off += size;
}
}break;
}
}
//- rjf: artifacts -> value bundle
Arena *info_arena = 0;
DASM_Info info = {0};
if(got_task && data.size != 0)
if(got_task)
{
//- rjf: produce joined text
Arena *text_arena = arena_alloc();
StringJoin text_join = {0};
text_join.sep = str8_lit("\n");
String8 text = str8_list_join(text_arena, &inst_strings, &text_join);
//- rjf: produce unique key for this disassembly's text
U128 text_key = {0};
{
U64 hash_data[] =
{
hash.u64[0],
hash.u64[1],
addr,
(U64)arch,
0x4d534144,
};
text_key = hs_hash_from_data(str8((U8 *)hash_data, sizeof(hash_data)));
}
//- rjf: submit text data to hash store
U128 text_hash = hs_submit_data(text_key, &text_arena, text);
//- rjf: produce value bundle
info_arena = arena_alloc();
info = dasm_info_from_arch_addr_data(info_arena, 0, arch, addr, data);
info.text_key = text_key;
info.insts = dasm_inst_array_from_chunk_list(info_arena, &inst_list);
}
//- rjf: commit results to cache
@@ -466,6 +478,7 @@ dasm_parse_thread__entry_point(void *p)
}
hs_scope_close(hs_scope);
scratch_end(scratch);
}
}
+2 -6
View File
@@ -46,7 +46,7 @@ struct DASM_InstArray
typedef struct DASM_Info DASM_Info;
struct DASM_Info
{
String8 text;
U128 text_key;
DASM_InstArray insts;
};
@@ -170,11 +170,6 @@ internal DASM_InstArray dasm_inst_array_from_chunk_list(Arena *arena, DASM_InstC
internal U64 dasm_inst_array_idx_from_code_off__linear_scan(DASM_InstArray *array, U64 off);
internal U64 dasm_inst_array_code_off_from_idx(DASM_InstArray *array, U64 idx);
////////////////////////////////
//~ rjf: Disassembly Decoding Function
internal DASM_Info dasm_info_from_arch_addr_data(Arena *arena, U64 *bytes_processed_counter, Architecture arch, U64 addr, String8 data);
////////////////////////////////
//~ rjf: Main Layer Initialization
@@ -198,6 +193,7 @@ internal void dasm_scope_touch_node__stripe_r_guarded(DASM_Scope *scope, DASM_No
internal DASM_Info dasm_info_from_hash_addr_arch(DASM_Scope *scope, U128 hash, U64 addr, Architecture arch);
internal DASM_Info dasm_info_from_key_addr_arch(DASM_Scope *scope, U128 key, U64 addr, Architecture arch, U128 *hash_out);
////////////////////////////////
//~ rjf: Parse Threads
+127 -1
View File
@@ -314,6 +314,36 @@ df_view_rule_hooks__txt_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ct
return result;
}
internal DF_DisasmTopologyInfo
df_view_rule_hooks__disasm_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ctrl_ctx, EVAL_ParseCtx *parse_ctx, EVAL_String2ExprMap *macro_map, DF_CfgNode *cfg)
{
Temp scratch = scratch_begin(0, 0);
DF_DisasmTopologyInfo result = zero_struct;
{
StringJoin join = {0};
join.sep = str8_lit(" ");
DF_CfgNode *size_cfg = df_cfg_node_child_from_string(cfg, str8_lit("size"), 0);
DF_CfgNode *arch_cfg = df_cfg_node_child_from_string(cfg, str8_lit("arch"), 0);
String8List size_expr_strs = {0};
String8 arch_string = {0};
for(DF_CfgNode *child = size_cfg->first; child != &df_g_nil_cfg_node; child = child->next)
{
str8_list_push(scratch.arena, &size_expr_strs, child->string);
}
arch_string = arch_cfg->first->string;
String8 size_expr = str8_list_join(scratch.arena, &size_expr_strs, &join);
DF_Eval size_eval = df_eval_from_string(scratch.arena, scope, ctrl_ctx, parse_ctx, macro_map, size_expr);
DF_Eval size_val_eval = df_value_mode_eval_from_eval(parse_ctx->type_graph, parse_ctx->rdi, ctrl_ctx, size_eval);
if(str8_match(arch_string, str8_lit("x64"), StringMatchFlag_CaseInsensitive))
{
result.arch = Architecture_x64;
}
result.size_cap = size_val_eval.imm_u64;
}
scratch_end(scratch);
return result;
}
////////////////////////////////
//~ rjf: "array"
@@ -736,6 +766,17 @@ DF_GFX_VIEW_RULE_BLOCK_UI_FUNCTION_DEF(text)
////////////////////////////////
//~ rjf: "disasm"
typedef struct DF_ViewRuleHooks_DisasmState DF_ViewRuleHooks_DisasmState;
struct DF_ViewRuleHooks_DisasmState
{
B32 initialized;
TxtPt cursor;
TxtPt mark;
S64 preferred_column;
U64 last_open_frame_idx;
F32 loaded_t;
};
DF_CORE_VIEW_RULE_VIZ_BLOCK_PROD_FUNCTION_DEF(disasm)
{
DF_EvalVizBlock *vb = df_eval_viz_block_begin(arena, DF_EvalVizBlockKind_Canvas, key, df_expand_key_make(df_hash_from_expand_key(key), 1), depth);
@@ -748,7 +789,92 @@ DF_CORE_VIEW_RULE_VIZ_BLOCK_PROD_FUNCTION_DEF(disasm)
DF_GFX_VIEW_RULE_BLOCK_UI_FUNCTION_DEF(disasm)
{
Temp scratch = scratch_begin(0, 0);
HS_Scope *hs_scope = hs_scope_open();
TXT_Scope *txt_scope = txt_scope_open();
DASM_Scope *dasm_scope = dasm_scope_open();
DF_ViewRuleHooks_DisasmState *state = df_view_rule_block_user_state(key, DF_ViewRuleHooks_DisasmState);
if(!state->initialized)
{
state->initialized = 1;
state->cursor = state->mark = txt_pt(1, 1);
}
if(state->last_open_frame_idx+1 < df_frame_index())
{
state->loaded_t = 0;
}
state->last_open_frame_idx = df_frame_index();
{
//- rjf: unpack params
DF_DisasmTopologyInfo top = df_view_rule_hooks__disasm_topology_info_from_cfg(dbgi_scope, ctrl_ctx, parse_ctx, macro_map, cfg);
//- rjf: resolve to address value & range
DF_Eval value_eval = df_value_mode_eval_from_eval(parse_ctx->type_graph, parse_ctx->rdi, ctrl_ctx, eval);
U64 base_vaddr = value_eval.imm_u64 ? value_eval.imm_u64 : value_eval.offset;
Rng1U64 vaddr_range = r1u64(base_vaddr, base_vaddr + (top.size_cap ? top.size_cap : 2048));
//- rjf: unpack thread/process of eval
DF_Entity *thread = df_entity_from_handle(ctrl_ctx->thread);
DF_Entity *process = df_entity_ancestor_from_kind(thread, DF_EntityKind_Process);
//- rjf: unpack key for this region in memory
U128 dasm_key = ctrl_hash_store_key_from_process_vaddr_range(process->ctrl_machine_id, process->ctrl_handle, vaddr_range, 0);
//- rjf: key -> parsed text info
U128 data_hash = {0};
DASM_Info dasm_info = dasm_info_from_key_addr_arch(dasm_scope, dasm_key, vaddr_range.min, top.arch, &data_hash);
String8 dasm_text_data = {0};
TXT_TextInfo dasm_text_info = {0};
for(U64 rewind_idx = 0; rewind_idx < 2; rewind_idx += 1)
{
U128 dasm_text_hash = hs_hash_from_key(dasm_info.text_key, rewind_idx);
dasm_text_data = hs_data_from_hash(hs_scope, dasm_text_hash);
dasm_text_info = txt_text_info_from_hash_lang(txt_scope, dasm_text_hash, TXT_LangKind_Null);
if(dasm_text_info.lines_count != 0)
{
break;
}
}
TXT_LineTokensSlice line_tokens_slice = txt_line_tokens_slice_from_info_data_line_range(scratch.arena, &dasm_text_info, dasm_text_data, r1s64(1, dasm_info.insts.count));
//- rjf: info -> code slice info
DF_CodeSliceParams code_slice_params = {0};
{
code_slice_params.flags = DF_CodeSliceFlag_LineNums;
code_slice_params.line_num_range = r1s64(1, dasm_text_info.lines_count);
code_slice_params.line_text = push_array(scratch.arena, String8, dasm_text_info.lines_count);
code_slice_params.line_ranges = push_array(scratch.arena, Rng1U64, dasm_text_info.lines_count);
code_slice_params.line_tokens = push_array(scratch.arena, TXT_TokenArray, dasm_text_info.lines_count);
code_slice_params.line_bps = push_array(scratch.arena, DF_EntityList, dasm_text_info.lines_count);
code_slice_params.line_ips = push_array(scratch.arena, DF_EntityList, dasm_text_info.lines_count);
code_slice_params.line_pins = push_array(scratch.arena, DF_EntityList, dasm_text_info.lines_count);
code_slice_params.line_dasm2src = push_array(scratch.arena, DF_TextLineDasm2SrcInfoList, dasm_text_info.lines_count);
code_slice_params.line_src2dasm = push_array(scratch.arena, DF_TextLineSrc2DasmInfoList, dasm_text_info.lines_count);
for(U64 line_idx = 0; line_idx < dasm_text_info.lines_count; line_idx += 1)
{
code_slice_params.line_text[line_idx] = str8_substr(dasm_text_data, dasm_info.insts.v[line_idx].text_range);
code_slice_params.line_ranges[line_idx] = dasm_info.insts.v[line_idx].text_range;
code_slice_params.line_tokens[line_idx] = line_tokens_slice.line_tokens[line_idx];
}
code_slice_params.font = df_font_from_slot(DF_FontSlot_Code);
code_slice_params.font_size = ui_top_font_size();
code_slice_params.line_height_px = ui_top_font_size()*1.5f;
code_slice_params.margin_width_px = 0;
code_slice_params.line_num_width_px = ui_top_font_size()*5.f;
code_slice_params.line_text_max_width_px = ui_top_font_size()*2.f*dasm_text_info.lines_max_size;
}
//- rjf: build code slice
if(dasm_info.insts.count != 0 && dasm_text_info.lines_count != 0)
UI_Padding(ui_pct(1, 0)) UI_PrefWidth(ui_px(dasm_text_info.lines_max_size*ui_top_font_size()*1.2f, 1.f)) UI_Column UI_Padding(ui_pct(1, 0))
{
DF_CodeSliceSignal sig = df_code_slice(ws, ctrl_ctx, parse_ctx, &code_slice_params, &state->cursor, &state->mark, &state->preferred_column, str8_lit("###code_slice"));
}
}
dasm_scope_close(dasm_scope);
txt_scope_close(txt_scope);
hs_scope_close(hs_scope);
scratch_end(scratch);
}
////////////////////////////////
+8
View File
@@ -29,6 +29,13 @@ struct DF_TxtTopologyInfo
U64 size_cap;
};
typedef struct DF_DisasmTopologyInfo DF_DisasmTopologyInfo;
struct DF_DisasmTopologyInfo
{
Architecture arch;
U64 size_cap;
};
////////////////////////////////
//~ rjf: Helpers
@@ -37,5 +44,6 @@ internal void df_view_rule_hooks__eval_commit_rgba(DF_Eval eval, TG_Graph *graph
internal DF_BitmapTopologyInfo df_view_rule_hooks__bitmap_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ctrl_ctx, EVAL_ParseCtx *parse_ctx, EVAL_String2ExprMap *macro_map, DF_CfgNode *cfg);
internal DF_GeoTopologyInfo df_view_rule_hooks__geo_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ctrl_ctx, EVAL_ParseCtx *parse_ctx, EVAL_String2ExprMap *macro_map, DF_CfgNode *cfg);
internal DF_TxtTopologyInfo df_view_rule_hooks__txt_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ctrl_ctx, EVAL_ParseCtx *parse_ctx, EVAL_String2ExprMap *macro_map, DF_CfgNode *cfg);
internal DF_DisasmTopologyInfo df_view_rule_hooks__disasm_topology_info_from_cfg(DBGI_Scope *scope, DF_CtrlCtx *ctrl_ctx, EVAL_ParseCtx *parse_ctx, EVAL_String2ExprMap *macro_map, DF_CfgNode *cfg);
#endif //DF_VIEW_RULE_HOOKS_H