ed/raddebugger
1
0 Fork 0
mirror of https://github.com/Ed94/raddebugger.git synced 2026-11-15 10:40:12 -08:00
Code Issues Packages Projects Releases Wiki Activity

dbg_engine: heuristically clear memory map & grab new stack page when advancing to new regions of stack for large unwinds

Browse Source
This commit is contained in:
Ryan Fleury committed 2026-09-30 07:55:55 -07:00
1 parent db645e5c90
commit 00d4a1a56f
1 file changed
+32 -20
Download Patch File Download Diff File Expand all files Collapse all files
src/dbg_engine/dbg_engine_ctrl.c
+32 -20
View File
@@ -1905,24 +1905,9 @@ d_unwind_from_thread(Arena *arena, D_Handle thread, U64 endt_us)
void *regs_block_restore = push_array(scratch.arena, U8, arch_reg_block_size);
//////////////////////////////
//- rjf: grab initial memory
//
// NOTE(rjf): we can pre-fill memory that we can expect to load here -
// otherwise we will lazily evaluate it via the unwinding backends below
//- rjf: set up empty memory map (will be lazily filled out via unwinding)
//
MemoryMap memory_map = {0};
if(regs_block_good)
{
U64 sp = arch_sp_from_reg_block(arch_info, regs_block);
U64 sp_rounded_down = AlignDownPow2(sp, KB(4));
Rng1U64 top_of_stack_vaddr_range = r1u64(sp_rounded_down, sp_rounded_down+KB(4));
D_ProcessMemorySlice slice = d_process_memory_slice_from_vaddr_range(scratch.arena, process_entity->handle, top_of_stack_vaddr_range, 1, endt_us);
String8 data = slice.data;
if(!slice.stale)
{
memory_map_push(scratch.arena, &memory_map, top_of_stack_vaddr_range, data.str);
}
}
//////////////////////////////
//- rjf: push one frame for register block we already have
@@ -1970,6 +1955,33 @@ d_unwind_from_thread(Arena *arena, D_Handle thread, U64 endt_us)
U64 tls_vaddr = 0;
d_thread_get_module_tls_vaddr(thread_entity->handle, module_entity->handle, &tls_vaddr);
//- rjf: clear memory map every so many frames, heuristically. preemptively pull in next 4k of stack.
if(frame_node_count%10 == 0)
{
// rjf: compute the next 4k of stack memory addresses
U64 sp_rounded_down = AlignDownPow2(start_sp, KB(4));
Rng1U64 top_of_stack_vaddr_range = r1u64(sp_rounded_down, sp_rounded_down+KB(4));
// rjf: determine if we already have this stack
B32 already_have_this_portion_of_stack = 0;
for(MemoryMapRangeNode *n = memory_map.first_range; n != 0; n = n->next)
{
if(n->v.vaddr_range.min == top_of_stack_vaddr_range.min && n->v.vaddr_range.max == top_of_stack_vaddr_range.max)
{
already_have_this_portion_of_stack = 1;
break;
}
}
// rjf: if we've moved to a new 4k region of stack, clear memory map, read the new portion of stack, & push
if(!already_have_this_portion_of_stack)
{
MemoryZeroStruct(&memory_map);
String8 data = d_data_from_process_vaddr_range(scratch.arena, process_entity->handle, top_of_stack_vaddr_range, 0);
memory_map_push(scratch.arena, &memory_map, top_of_stack_vaddr_range, data.str);
}
}
//- rjf: do one unwind step
B32 step_is_good = 0;
for(;!unwind.flags && !step_is_good;)
@@ -2026,9 +2038,9 @@ d_unwind_from_thread(Arena *arena, D_Handle thread, U64 endt_us)
// (b) advance ip to 0
// (c) do not modify either ip *or* sp (just one is fine)
//
B32 step_made_process = (arch_ip_from_reg_block(arch_info, regs_block) != start_ip ||
arch_sp_from_reg_block(arch_info, regs_block) != start_sp);
if(step_is_good && step_made_process && arch_ip_from_reg_block(arch_info, regs_block) != 0)
B32 step_made_progress = (arch_ip_from_reg_block(arch_info, regs_block) != start_ip ||
arch_sp_from_reg_block(arch_info, regs_block) != start_sp);
if(step_is_good && step_made_progress && arch_ip_from_reg_block(arch_info, regs_block) != 0)
{
D_UnwindFrameNode *frame_node = push_array(scratch.arena, D_UnwindFrameNode, 1);
D_UnwindFrame *f = &frame_node->v;
@@ -2042,7 +2054,7 @@ d_unwind_from_thread(Arena *arena, D_Handle thread, U64 endt_us)
access_close(access);
//- rjf: exit if we made no progress on the unwind
if(!step_made_process)
if(!step_made_progress)
{
break;
}