mirror/zig
1
0
Fork 0
mirror of https://github.com/ziglang/zig.git synced 2025-12-16 19:23:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

more stuff

Browse Source
This commit is contained in:
mlugg 2025-09-02 12:12:54 +01:00
parent b750e7cf9e
commit ed6ed62c42
No known key found for this signature in database
GPG Key ID: 3F5B7DCCBF4AF02E
3 changed files with 557 additions and 611 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
lib/std/debug/Dwarf.zig
View File

@ -1449,6 +1449,7 @@ fn getStringGeneric(opt_str: ?[]const u8, offset: u64) ![:0]const u8 {
return str[casted_offset..last :0]; return str[casted_offset..last :0];
} }
// MLUGG TODO: i am dubious of this whole thing being here atp. look closely and see if it depends on being the self process
pub const ElfModule = struct { pub const ElfModule = struct {
unwind: Dwarf.Unwind, unwind: Dwarf.Unwind,
dwarf: Dwarf, dwarf: Dwarf,
@ -1456,10 +1457,7 @@ pub const ElfModule = struct {
external_mapped_memory: ?[]align(std.heap.page_size_min) const u8, external_mapped_memory: ?[]align(std.heap.page_size_min) const u8,
pub const init: ElfModule = .{ pub const init: ElfModule = .{
.unwind = .{ .unwind = .init,
.debug_frame = null,
.eh_frame = null,
},
.dwarf = .{}, .dwarf = .{},
.mapped_memory = null, .mapped_memory = null,
.external_mapped_memory = null, .external_mapped_memory = null,
@ -1508,6 +1506,8 @@ pub const ElfModule = struct {
/// If the required sections aren't present but a reference to external debug /// If the required sections aren't present but a reference to external debug
/// info is, then this this function will recurse to attempt to load the debug /// info is, then this this function will recurse to attempt to load the debug
/// sections from an external file. /// sections from an external file.
///
/// MLUGG TODO: this should *return* a thing
pub fn load( pub fn load(
em: *ElfModule, em: *ElfModule,
gpa: Allocator, gpa: Allocator,
@ -1518,6 +1518,8 @@ pub const ElfModule = struct {
parent_mapped_mem: ?[]align(std.heap.page_size_min) const u8, parent_mapped_mem: ?[]align(std.heap.page_size_min) const u8,
elf_filename: ?[]const u8, elf_filename: ?[]const u8,
) LoadError!void { ) LoadError!void {
assert(em.mapped_memory == null);
if (expected_crc) |crc| if (crc != std.hash.crc.Crc32.hash(mapped_mem)) return error.InvalidDebugInfo; if (expected_crc) |crc| if (crc != std.hash.crc.Crc32.hash(mapped_mem)) return error.InvalidDebugInfo;
const hdr: *const elf.Ehdr = @ptrCast(&mapped_mem[0]); const hdr: *const elf.Ehdr = @ptrCast(&mapped_mem[0]);
@ -1709,8 +1711,9 @@ pub const ElfModule = struct {
separate_debug_crc, separate_debug_crc,
&sections, &sections,
mapped_mem, mapped_mem,
)) |debug_info| { )) |v| {
return debug_info; v;
return;
} else |_| {} } else |_| {}
// <exe_dir>/.debug/<gnu_debuglink> // <exe_dir>/.debug/<gnu_debuglink>

357
lib/std/debug/Dwarf/Unwind.zig
View File

@ -1,28 +1,35 @@
//! MLUGG TODO DOCUMENT THIS
pub const VirtualMachine = @import("Unwind/VirtualMachine.zig"); pub const VirtualMachine = @import("Unwind/VirtualMachine.zig");
/// The contents of the `.debug_frame` section as specified by DWARF. This might be a more reliable frame_section: ?struct {
/// stack unwind mechanism in some cases, or it may be present when `.eh_frame` is not, but fetching id: Section,
/// the data requires loading the binary, so it is not a viable approach for fast stack trace /// The virtual address of the start of the section. "Virtual address" refers to the address in
/// capturing within a process. /// the binary (e.g. `sh_addr` in an ELF file); the equivalent runtime address may be relocated
debug_frame: ?struct { /// in position-independent binaries.
data: []const u8, vaddr: u64,
/// Offsets into `data` of FDEs, sorted by ascending `pc_begin`. /// The full contents of the section. May have imprecise bounds depending on `section`.
sorted_fdes: []SortedFdeEntry, ///
/// For `.debug_frame`, the slice length is exactly equal to the section length. This is needed
/// to know the number of CIEs and FDEs.
///
/// For `.eh_frame`, the slice length may exceed the section length, i.e. the slice may refer to
/// more bytes than are in the second. This restriction exists because `.eh_frame_hdr` only
/// includes the address of the loaded `.eh_frame` data, not its length. It is not a problem
/// because unlike `.debug_frame`, the end of the CIE/FDE list is signaled through a sentinel
/// value. If this slice does have bounds, they will still be checked, preventing crashes when
/// reading potentially-invalid `.eh_frame` data from files.
bytes: []const u8,
}, },
/// Data associated with the `.eh_frame` and `.eh_frame_hdr` sections as defined by LSB Core. The lookup: ?union(enum) {
/// format of `.eh_frame` is an extension of that of DWARF's `.debug_frame` -- in fact it is almost eh_frame_hdr: struct {
/// identical, though subtly different in a few places. /// Virtual address of the `.eh_frame_hdr` section.
eh_frame: ?struct { vaddr: u64,
header: EhFrameHeader, table: EhFrameHeader.SearchTable,
/// Though this is a slice, it may be longer than the `.eh_frame` section. When unwinding },
/// through the runtime-loaded `.eh_frame_hdr` data, we are not told the size of the `.eh_frame` /// Offsets into `frame_section` of FDEs, sorted by ascending `pc_begin`.
/// section, so construct a slice referring to all of the rest of memory. The end of the section sorted_fdes: []SortedFdeEntry,
/// must be detected through `EntryHeader.terminator`.
eh_frame_data: []const u8,
/// Offsets into `eh_frame_data` of FDEs, sorted by ascending `pc_begin`.
/// Populated only if `header` does not already contain a lookup table.
sorted_fdes: ?[]SortedFdeEntry,
}, },
const SortedFdeEntry = struct { const SortedFdeEntry = struct {
@ -34,17 +41,61 @@ const SortedFdeEntry = struct {
const Section = enum { debug_frame, eh_frame }; const Section = enum { debug_frame, eh_frame };
// MLUGG TODO deinit?
pub const init: Unwind = .{
.frame_section = null,
.lookup = null,
};
/// This represents the decoded .eh_frame_hdr header /// This represents the decoded .eh_frame_hdr header
pub const EhFrameHeader = struct { pub const EhFrameHeader = struct {
vaddr: u64,
eh_frame_vaddr: u64, eh_frame_vaddr: u64,
search_table: ?struct { search_table: ?SearchTable,
pub const SearchTable = struct {
/// The byte offset of the search table into the `.eh_frame_hdr` section. /// The byte offset of the search table into the `.eh_frame_hdr` section.
offset: u8, offset: u8,
encoding: EH.PE, encoding: EH.PE,
fde_count: usize, fde_count: usize,
entries: []const u8, entries: []const u8,
},
/// Returns the vaddr of the FDE for `pc`, or `null` if no matching FDE was found.
fn findEntry(
table: *const SearchTable,
eh_frame_hdr_vaddr: u64,
pc: u64,
addr_size_bytes: u8,
endian: Endian,
) !?u64 {
const table_vaddr = eh_frame_hdr_vaddr + table.offset;
const entry_size = try EhFrameHeader.entrySize(table.encoding, addr_size_bytes);
var left: usize = 0;
var len: usize = table.fde_count;
while (len > 1) {
const mid = left + len / 2;
var entry_reader: Reader = .fixed(table.entries[mid * entry_size ..][0..entry_size]);
const pc_begin = try readEhPointer(&entry_reader, table.encoding, addr_size_bytes, .{
.pc_rel_base = table_vaddr + left * entry_size,
.data_rel_base = eh_frame_hdr_vaddr,
}, endian);
if (pc < pc_begin) {
len /= 2;
} else {
left = mid;
len -= len / 2;
}
}
if (len == 0) return null;
var entry_reader: Reader = .fixed(table.entries[left * entry_size ..][0..entry_size]);
// Skip past `pc_begin`; we're now interested in the fde offset
_ = try readEhPointerAbs(&entry_reader, table.encoding.type, addr_size_bytes, endian);
const fde_ptr = try readEhPointer(&entry_reader, table.encoding, addr_size_bytes, .{
.pc_rel_base = table_vaddr + left * entry_size,
.data_rel_base = eh_frame_hdr_vaddr,
}, endian);
return fde_ptr;
}
};
pub fn entrySize(table_enc: EH.PE, addr_size_bytes: u8) !u8 { pub fn entrySize(table_enc: EH.PE, addr_size_bytes: u8) !u8 {
return switch (table_enc.type) { return switch (table_enc.type) {
@ -76,10 +127,7 @@ pub const EhFrameHeader = struct {
.pc_rel_base = eh_frame_hdr_vaddr + r.seek, .pc_rel_base = eh_frame_hdr_vaddr + r.seek,
}, endian); }, endian);
return .{ const table: ?SearchTable = table: {
.vaddr = eh_frame_hdr_vaddr,
.eh_frame_vaddr = eh_frame_ptr,
.search_table = table: {
if (fde_count_enc == EH.PE.omit) break :table null; if (fde_count_enc == EH.PE.omit) break :table null;
if (table_enc == EH.PE.omit) break :table null; if (table_enc == EH.PE.omit) break :table null;
const fde_count = try readEhPointer(&r, fde_count_enc, addr_size_bytes, .{ const fde_count = try readEhPointer(&r, fde_count_enc, addr_size_bytes, .{
@ -95,45 +143,12 @@ pub const EhFrameHeader = struct {
.entries = bytes, .entries = bytes,
.offset = @intCast(bytes_offset), .offset = @intCast(bytes_offset),
}; };
},
}; };
}
/// Asserts that `eh_frame_hdr.search_table != null`. return .{
fn findEntry( .eh_frame_vaddr = eh_frame_ptr,
eh_frame_hdr: *const EhFrameHeader, .search_table = table,
pc: u64, };
addr_size_bytes: u8,
endian: Endian,
) !?u64 {
const table = &eh_frame_hdr.search_table.?;
const table_vaddr = eh_frame_hdr.vaddr + table.offset;
const entry_size = try EhFrameHeader.entrySize(table.encoding, addr_size_bytes);
var left: usize = 0;
var len: usize = table.fde_count;
while (len > 1) {
const mid = left + len / 2;
var entry_reader: Reader = .fixed(table.entries[mid * entry_size ..][0..entry_size]);
const pc_begin = try readEhPointer(&entry_reader, table.encoding, addr_size_bytes, .{
.pc_rel_base = table_vaddr + left * entry_size,
.data_rel_base = eh_frame_hdr.vaddr,
}, endian);
if (pc < pc_begin) {
len /= 2;
} else {
left = mid;
len -= len / 2;
}
}
if (len == 0) return null;
var entry_reader: Reader = .fixed(table.entries[left * entry_size ..][0..entry_size]);
// Skip past `pc_begin`; we're now interested in the fde offset
_ = try readEhPointerAbs(&entry_reader, table.encoding.type, addr_size_bytes, endian);
const fde_ptr = try readEhPointer(&entry_reader, table.encoding, addr_size_bytes, .{
.pc_rel_base = table_vaddr + left * entry_size,
.data_rel_base = eh_frame_hdr.vaddr,
}, endian);
return std.math.sub(u64, fde_ptr, eh_frame_hdr.eh_frame_vaddr) catch bad(); // offset into .eh_frame
} }
}; };
@ -356,133 +371,84 @@ pub const FrameDescriptionEntry = struct {
} }
}; };
pub fn scanDebugFrame( /// Load unwind information from the contents of an `.eh_frame` or `.debug_frame` section.
unwind: *Unwind, ///
gpa: Allocator, /// If the `.eh_frame_hdr` section is available, consider instead using `loadFromEhFrameHdr`. This
section_vaddr: u64, /// allows the implementation to use a search table embedded in that section if it is available.
section_bytes: []const u8, pub fn loadFromSection(unwind: *Unwind, section: Section, section_vaddr: u64, section_bytes: []const u8) void {
addr_size_bytes: u8, assert(unwind.frame_section == null);
endian: Endian, assert(unwind.lookup == null);
) void { unwind.frame_section = .{
assert(unwind.debug_frame == null); .id = section,
.bytes = section_bytes,
var fbr: Reader = .fixed(section_bytes); .vaddr = section_vaddr,
var fde_list: std.ArrayList(SortedFdeEntry) = .empty;
defer fde_list.deinit(gpa);
while (fbr.seek < fbr.buffer.len) {
const entry_offset = fbr.seek;
switch (try EntryHeader.read(&fbr, fbr.seek, .debug_frame, endian)) {
// Ignore CIEs; we only need them to parse the FDEs!
.cie => |info| {
try fbr.discardAll(info.bytes_len);
continue;
},
.fde => |info| {
const cie: CommonInformationEntry = cie: {
var cie_reader: Reader = .fixed(section_bytes[info.cie_offset..]);
const cie_info = switch (try EntryHeader.read(&cie_reader, info.cie_offset, .debug_frame, endian)) {
.cie => |cie_info| cie_info,
.fde, .terminator => return bad(), // This is meant to be a CIE
}; };
break :cie try .parse(try cie_reader.take(cie_info.bytes_len), .debug_frame, addr_size_bytes);
};
const fde: FrameDescriptionEntry = try .parse(
section_vaddr + fbr.seek,
try fbr.take(info.bytes_len),
cie,
endian,
);
try fde_list.append(.{
.pc_begin = fde.pc_begin,
.fde_offset = entry_offset, // *not* `fde_offset`, because we need to include the entry header
});
},
.terminator => return bad(), // DWARF `.debug_frame` isn't meant to have terminators
}
}
const fde_slice = try fde_list.toOwnedSlice(gpa);
errdefer comptime unreachable;
std.mem.sortUnstable(SortedFdeEntry, fde_slice, {}, struct {
fn lessThan(ctx: void, a: SortedFdeEntry, b: SortedFdeEntry) bool {
ctx;
return a.pc_begin < b.pc_begin;
}
}.lessThan);
unwind.debug_frame = .{ .data = section_bytes, .sorted_fdes = fde_slice };
} }
pub fn scanEhFrame( /// Load unwind information from a header loaded from an `.eh_frame_hdr` section, and a pointer to
/// the contents of the `.eh_frame` section.
///
/// This differs from `loadFromSection` because `.eh_frame_hdr` may embed a binary search table, and
/// if it does, this function will use that for address lookups instead of constructing our own
/// search table.
pub fn loadFromEhFrameHdr(
unwind: *Unwind, unwind: *Unwind,
gpa: Allocator,
header: EhFrameHeader, header: EhFrameHeader,
section_vaddr: u64,
section_bytes_ptr: [*]const u8, section_bytes_ptr: [*]const u8,
/// This is separate from `section_bytes_ptr` because it is unknown when `.eh_frame` is accessed
/// through the pointer in the `.eh_frame_hdr` section. If this is non-`null`, we avoid reading
/// past this number of bytes, but if `null`, we must assume that the `.eh_frame` data has a
/// valid terminator.
section_bytes_len: ?usize,
addr_size_bytes: u8,
endian: Endian,
) !void { ) !void {
assert(unwind.eh_frame == null); assert(unwind.frame_section == null);
assert(unwind.lookup == null);
const section_bytes: []const u8 = bytes: { unwind.frame_section = .{
// If the length is unknown, let the slice span from `section_bytes_ptr` to the end of memory. .id = .eh_frame,
const len = section_bytes_len orelse (std.math.maxInt(usize) - @intFromPtr(section_bytes_ptr)); .bytes = maxSlice(section_bytes_ptr),
break :bytes section_bytes_ptr[0..len]; .vaddr = header.eh_frame_vaddr,
}; };
if (header.search_table) |table| {
if (header.search_table != null) { unwind.lookup = .{ .eh_frame_hdr = .{
// No need to populate `sorted_fdes`, the header contains a search table. .vaddr = section_vaddr,
unwind.eh_frame = .{ .table = table,
.header = header, } };
.eh_frame_data = section_bytes,
.sorted_fdes = null,
};
return;
} }
}
// We aren't told the length of this section. Luckily, we don't need it, because there will be pub fn prepareLookup(unwind: *Unwind, gpa: Allocator, addr_size_bytes: u8, endian: Endian) !void {
// an `EntryHeader.terminator` after the last CIE/FDE. Just make a `Reader` which will give us const section = unwind.frame_section.?;
// alllll of the bytes! if (unwind.lookup != null) return;
var fbr: Reader = .fixed(section_bytes);
var r: Reader = .fixed(section.bytes);
var fde_list: std.ArrayList(SortedFdeEntry) = .empty; var fde_list: std.ArrayList(SortedFdeEntry) = .empty;
defer fde_list.deinit(gpa); defer fde_list.deinit(gpa);
while (true) { const saw_terminator = while (r.seek < r.buffer.len) {
const entry_offset = fbr.seek; const entry_offset = r.seek;
switch (try EntryHeader.read(&fbr, fbr.seek, .eh_frame, endian)) { switch (try EntryHeader.read(&r, entry_offset, section.id, endian)) {
// Ignore CIEs; we only need them to parse the FDEs! .cie => |cie_info| {
.cie => |info| { // Ignore CIEs for now; we'll parse them when we read a corresponding FDE
try fbr.discardAll(info.bytes_len); try r.discardAll(cie_info.bytes_len);
continue; continue;
}, },
.fde => |info| { .fde => |fde_info| {
const cie: CommonInformationEntry = cie: { var cie_r: Reader = .fixed(section.bytes[fde_info.cie_offset..]);
var cie_reader: Reader = .fixed(section_bytes[info.cie_offset..]); const cie_info = switch (try EntryHeader.read(&cie_r, fde_info.cie_offset, section.id, endian)) {
const cie_info = switch (try EntryHeader.read(&cie_reader, info.cie_offset, .eh_frame, endian)) {
.cie => |cie_info| cie_info, .cie => |cie_info| cie_info,
.fde, .terminator => return bad(), // This is meant to be a CIE .fde, .terminator => return bad(), // this is meant to be a CIE
}; };
break :cie try .parse(try cie_reader.take(cie_info.bytes_len), .eh_frame, addr_size_bytes); const cie: CommonInformationEntry = try .parse(try cie_r.take(cie_info.bytes_len), section.id, addr_size_bytes);
}; const fde: FrameDescriptionEntry = try .parse(section.vaddr + r.seek, try r.take(fde_info.bytes_len), cie, endian);
const fde: FrameDescriptionEntry = try .parse(
header.eh_frame_vaddr + fbr.seek,
try fbr.take(info.bytes_len),
cie,
endian,
);
try fde_list.append(gpa, .{ try fde_list.append(gpa, .{
.pc_begin = fde.pc_begin, .pc_begin = fde.pc_begin,
.fde_offset = entry_offset, // *not* `fde_offset`, because we need to include the entry header .fde_offset = entry_offset,
}); });
}, },
// Unlike `.debug_frame`, the `.eh_frame` section does have a terminator CIE -- this is .terminator => break true,
// necessary because `header` doesn't include the length of the `.eh_frame` section
.terminator => break,
} }
} else false;
switch (section.id) {
.eh_frame => if (!saw_terminator) return bad(), // `.eh_frame` indicates the end of the CIE/FDE list with a sentinel entry
.debug_frame => if (saw_terminator) return bad(), // `.debug_frame` uses the section bounds and does not specify a sentinel entry
} }
const fde_slice = try fde_list.toOwnedSlice(gpa); const fde_slice = try fde_list.toOwnedSlice(gpa);
errdefer comptime unreachable; errdefer comptime unreachable;
std.mem.sortUnstable(SortedFdeEntry, fde_slice, {}, struct { std.mem.sortUnstable(SortedFdeEntry, fde_slice, {}, struct {
@ -491,26 +457,29 @@ pub fn scanEhFrame(
return a.pc_begin < b.pc_begin; return a.pc_begin < b.pc_begin;
} }
}.lessThan); }.lessThan);
unwind.eh_frame = .{ unwind.lookup = .{ .sorted_fdes = fde_slice };
.header = header,
.eh_frame_data = section_bytes,
.sorted_fdes = fde_slice,
};
} }
/// Given a program counter value, returns the offset of the corresponding FDE, or `null` if no
/// matching FDE was found. The returned offset can be passed to `getFde` to load the data
/// associated with the FDE.
///
/// Before calling this function, `prepareLookup` must return successfully.
///
/// The return value may be a false positive. After loading the FDE with `loadFde`, the caller must /// The return value may be a false positive. After loading the FDE with `loadFde`, the caller must
/// validate that `pc` is indeed in its range -- if it is not, then no FDE matches `pc`. /// validate that `pc` is indeed in its range -- if it is not, then no FDE matches `pc`.
pub fn findFdeOffset(unwind: *const Unwind, pc: u64, addr_size_bytes: u8, endian: Endian) !?u64 { pub fn lookupPc(unwind: *const Unwind, pc: u64, addr_size_bytes: u8, endian: Endian) !?u64 {
// We'll break from this block only if we have a manually-constructed search table. const sorted_fdes: []const SortedFdeEntry = switch (unwind.lookup.?) {
const sorted_fdes: []const SortedFdeEntry = fdes: { .eh_frame_hdr => |eh_frame_hdr| {
if (unwind.debug_frame) |df| break :fdes df.sorted_fdes; const fde_vaddr = try eh_frame_hdr.table.findEntry(
if (unwind.eh_frame) |eh_frame| { eh_frame_hdr.vaddr,
if (eh_frame.sorted_fdes) |fdes| break :fdes fdes; pc,
// Use the search table from the `.eh_frame_hdr` section rather than one of our own addr_size_bytes,
return eh_frame.header.findEntry(pc, addr_size_bytes, endian); endian,
} ) orelse return null;
// We have no available unwind info return std.math.sub(u64, fde_vaddr, unwind.frame_section.?.vaddr) catch bad(); // convert vaddr to offset
return null; },
.sorted_fdes => |sorted_fdes| sorted_fdes,
}; };
const first_bad_idx = std.sort.partitionPoint(SortedFdeEntry, sorted_fdes, pc, struct { const first_bad_idx = std.sort.partitionPoint(SortedFdeEntry, sorted_fdes, pc, struct {
fn canIncludePc(target_pc: u64, entry: SortedFdeEntry) bool { fn canIncludePc(target_pc: u64, entry: SortedFdeEntry) bool {
@ -523,33 +492,29 @@ pub fn findFdeOffset(unwind: *const Unwind, pc: u64, addr_size_bytes: u8, endian
return sorted_fdes[first_bad_idx - 1].fde_offset; return sorted_fdes[first_bad_idx - 1].fde_offset;
} }
pub fn loadFde(unwind: *const Unwind, fde_offset: u64, addr_size_bytes: u8, endian: Endian) !struct { Format, CommonInformationEntry, FrameDescriptionEntry } { pub fn getFde(unwind: *const Unwind, fde_offset: u64, addr_size_bytes: u8, endian: Endian) !struct { Format, CommonInformationEntry, FrameDescriptionEntry } {
const section_bytes: []const u8, const section_vaddr: u64, const section: Section = s: { const section = unwind.frame_section.?;
if (unwind.debug_frame) |df| break :s .{ df.data, if (true) @panic("MLUGG TODO"), .debug_frame };
if (unwind.eh_frame) |ef| break :s .{ ef.eh_frame_data, ef.header.eh_frame_vaddr, .eh_frame };
unreachable; // how did you get `fde_offset`?!
};
var fde_reader: Reader = .fixed(section_bytes[fde_offset..]); var fde_reader: Reader = .fixed(section.bytes[fde_offset..]);
const fde_info = switch (try EntryHeader.read(&fde_reader, fde_offset, section, endian)) { const fde_info = switch (try EntryHeader.read(&fde_reader, fde_offset, section.id, endian)) {
.fde => |info| info, .fde => |info| info,
.cie, .terminator => return bad(), // This is meant to be an FDE .cie, .terminator => return bad(), // This is meant to be an FDE
}; };
const cie_offset = fde_info.cie_offset; const cie_offset = fde_info.cie_offset;
var cie_reader: Reader = .fixed(section_bytes[cie_offset..]); var cie_reader: Reader = .fixed(section.bytes[cie_offset..]);
const cie_info = switch (try EntryHeader.read(&cie_reader, cie_offset, section, endian)) { const cie_info = switch (try EntryHeader.read(&cie_reader, cie_offset, section.id, endian)) {
.cie => |info| info, .cie => |info| info,
.fde, .terminator => return bad(), // This is meant to be a CIE .fde, .terminator => return bad(), // This is meant to be a CIE
}; };
const cie: CommonInformationEntry = try .parse( const cie: CommonInformationEntry = try .parse(
try cie_reader.take(cie_info.bytes_len), try cie_reader.take(cie_info.bytes_len),
section, section.id,
addr_size_bytes, addr_size_bytes,
); );
const fde: FrameDescriptionEntry = try .parse( const fde: FrameDescriptionEntry = try .parse(
section_vaddr + fde_offset + fde_reader.seek, section.vaddr + fde_offset + fde_reader.seek,
try fde_reader.take(fde_info.bytes_len), try fde_reader.take(fde_info.bytes_len),
cie, cie,
endian, endian,

676
lib/std/debug/SelfInfo.zig
View File

@ -26,10 +26,13 @@ const regValueNative = Dwarf.abi.regValueNative;
const SelfInfo = @This(); const SelfInfo = @This();
/// MLUGG TODO: what if this field had a less stupid name... modules: std.AutoHashMapUnmanaged(usize, struct {
address_map: std.AutoHashMapUnmanaged(usize, Module.DebugInfo), di: Module.DebugInfo,
loaded_debug: bool,
module_cache: if (native_os == .windows) std.ArrayListUnmanaged(windows.MODULEENTRY32) else void, loaded_unwind: bool,
const init: @This() = .{ .di = .init, .loaded_debug = false, .loaded_unwind = false };
}),
lookup_cache: Module.LookupCache,
pub const target_supported: bool = switch (native_os) { pub const target_supported: bool = switch (native_os) {
.linux, .linux,
@ -46,19 +49,19 @@ pub const target_supported: bool = switch (native_os) {
}; };
pub const init: SelfInfo = .{ pub const init: SelfInfo = .{
.address_map = .empty, .modules = .empty,
.module_cache = if (native_os == .windows) .empty, .lookup_cache = if (Module.LookupCache != void) .init,
}; };
pub fn deinit(self: *SelfInfo) void { pub fn deinit(self: *SelfInfo) void {
// MLUGG TODO: that's amusing, this function is straight-up unused. i... wonder if it even should be used anywhere? perhaps not... so perhaps it should not even exist...???? // MLUGG TODO: that's amusing, this function is straight-up unused. i... wonder if it even should be used anywhere? perhaps not... so perhaps it should not even exist...????
var it = self.address_map.iterator(); var it = self.modules.iterator();
while (it.next()) |entry| { while (it.next()) |entry| {
const mdi = entry.value_ptr.*; const mdi = entry.value_ptr.*;
mdi.deinit(self.allocator); mdi.deinit(self.allocator);
self.allocator.destroy(mdi); self.allocator.destroy(mdi);
} }
self.address_map.deinit(self.allocator); self.modules.deinit(self.allocator);
if (native_os == .windows) { if (native_os == .windows) {
for (self.modules.items) |module| { for (self.modules.items) |module| {
self.allocator.free(module.name); self.allocator.free(module.name);
@ -68,94 +71,26 @@ pub fn deinit(self: *SelfInfo) void {
} }
} }
fn lookupModuleForAddress(self: *SelfInfo, gpa: Allocator, address: usize) !Module {
if (builtin.target.os.tag.isDarwin()) {
return self.lookupModuleDyld(address);
} else if (native_os == .windows) {
return self.lookupModuleWin32(gpa, address);
} else if (native_os == .haiku) {
@panic("TODO implement lookup module for Haiku");
} else if (builtin.target.cpu.arch.isWasm()) {
@panic("TODO implement lookup module for Wasm");
} else {
return self.lookupModuleDl(address);
}
}
fn loadModuleDebugInfo(gpa: Allocator, module: *const Module, di: *Module.DebugInfo) !void {
// MLUGG TODO: this should totally just go into the `Module` impl or something, right? lol
if (builtin.target.os.tag.isDarwin()) {
try loadMachODebugInfo(gpa, module, di);
} else if (native_os == .windows) {
// MLUGG TODO: deal with 'already loaded' properly
try readCoffDebugInfo(gpa, module, di);
} else if (native_os == .haiku) {
unreachable;
} else if (builtin.target.cpu.arch.isWasm()) {
unreachable;
} else {
if (di.mapped_memory != null) return; // already loaded
const filename: ?[]const u8 = if (module.name.len > 0) module.name else null;
const mapped_mem = mapFileOrSelfExe(filename) catch |err| switch (err) {
error.FileNotFound => return error.MissingDebugInfo,
error.FileTooBig => return error.InvalidDebugInfo,
else => |e| return e,
};
errdefer posix.munmap(mapped_mem);
try di.load(gpa, mapped_mem, module.build_id, null, null, null, filename);
assert(di.mapped_memory != null);
}
}
fn loadModuleUnwindInfo(gpa: Allocator, module: *const Module, di: *Module.DebugInfo) !void {
if (builtin.target.os.tag.isDarwin()) {
// MLUGG TODO HACKHACK
try loadMachODebugInfo(gpa, module, di);
} else if (native_os == .windows) {
comptime unreachable; // not supported
} else if (native_os == .haiku) {
comptime unreachable; // not supported
} else if (builtin.target.cpu.arch.isWasm()) {
comptime unreachable; // not supported
} else {
eh_frame: {
if (di.unwind.eh_frame != null) break :eh_frame; // already loaded
const eh_frame_hdr_bytes = module.gnu_eh_frame orelse break :eh_frame;
const eh_frame_hdr: Dwarf.Unwind.EhFrameHeader = try .parse(
@intFromPtr(eh_frame_hdr_bytes.ptr) - module.load_offset,
eh_frame_hdr_bytes,
@sizeOf(usize),
native_endian,
);
const eh_frame_addr = module.load_offset + @as(usize, @intCast(eh_frame_hdr.eh_frame_vaddr));
try di.unwind.scanEhFrame(
gpa,
eh_frame_hdr,
@ptrFromInt(eh_frame_addr),
null,
@sizeOf(usize),
native_endian,
);
}
}
}
pub fn unwindFrame(self: *SelfInfo, gpa: Allocator, context: *UnwindContext) !usize { pub fn unwindFrame(self: *SelfInfo, gpa: Allocator, context: *UnwindContext) !usize {
comptime assert(target_supported); comptime assert(target_supported);
const module = try self.lookupModuleForAddress(gpa, context.pc); const module: Module = try .lookup(&self.lookup_cache, gpa, context.pc); // MLUGG TODO: don't take gpa
const gop = try self.address_map.getOrPut(gpa, module.load_offset); const gop = try self.modules.getOrPut(gpa, module.load_offset);
if (!gop.found_existing) gop.value_ptr.* = .init; if (!gop.found_existing) gop.value_ptr.* = .init;
try loadModuleUnwindInfo(gpa, &module, gop.value_ptr); if (!gop.value_ptr.loaded_unwind) {
try module.loadUnwindInfo(gpa, &gop.value_ptr.di);
gop.value_ptr.loaded_unwind = true;
}
// MLUGG TODO: the stuff below is impl!
if (native_os.isDarwin()) { if (native_os.isDarwin()) {
// __unwind_info is a requirement for unwinding on Darwin. It may fall back to DWARF, but unwinding // __unwind_info is a requirement for unwinding on Darwin. It may fall back to DWARF, but unwinding
// via DWARF before attempting to use the compact unwind info will produce incorrect results. // via DWARF before attempting to use the compact unwind info will produce incorrect results.
if (gop.value_ptr.unwind_info) |unwind_info| { if (gop.value_ptr.di.unwind_info) |unwind_info| {
if (unwindFrameMachO( if (unwindFrameMachO(
module.text_base, module.text_base,
module.load_offset, module.load_offset,
context, context,
unwind_info, unwind_info,
gop.value_ptr.eh_frame, gop.value_ptr.di.eh_frame,
)) |return_address| { )) |return_address| {
return return_address; return return_address;
} else |err| { } else |err| {
@ -164,7 +99,7 @@ pub fn unwindFrame(self: *SelfInfo, gpa: Allocator, context: *UnwindContext) !us
} }
return error.MissingUnwindInfo; return error.MissingUnwindInfo;
} }
if (try gop.value_ptr.getDwarfUnwindForAddress(gpa, context.pc)) |unwind| { if (try gop.value_ptr.di.getDwarfUnwindForAddress(gpa, context.pc)) |unwind| {
return unwindFrameDwarf(unwind, module.load_offset, context, null); return unwindFrameDwarf(unwind, module.load_offset, context, null);
} }
return error.MissingDebugInfo; return error.MissingDebugInfo;
@ -172,11 +107,15 @@ pub fn unwindFrame(self: *SelfInfo, gpa: Allocator, context: *UnwindContext) !us
pub fn getSymbolAtAddress(self: *SelfInfo, gpa: Allocator, address: usize) !std.debug.Symbol { pub fn getSymbolAtAddress(self: *SelfInfo, gpa: Allocator, address: usize) !std.debug.Symbol {
comptime assert(target_supported); comptime assert(target_supported);
const module = try self.lookupModuleForAddress(gpa, address); const module: Module = try .lookup(&self.lookup_cache, gpa, address); // MLUGG TODO: don't take gpa
const gop = try self.address_map.getOrPut(gpa, module.key()); const gop = try self.modules.getOrPut(gpa, module.key());
if (!gop.found_existing) gop.value_ptr.* = .init; if (!gop.found_existing) gop.value_ptr.* = .init;
try loadModuleDebugInfo(gpa, &module, gop.value_ptr); if (!gop.value_ptr.loaded_debug) {
return module.getSymbolAtAddress(gpa, gop.value_ptr, address); // MLUGG TODO: this overloads the name 'debug info' with including vs excluding unwind info
// figure out a better name for one or the other (i think the inner one is maybe 'symbol info' or something idk)
try module.loadDebugInfo(gpa, &gop.value_ptr.di);
}
return module.getSymbolAtAddress(gpa, &gop.value_ptr.di, address);
} }
/// Returns the module name for a given address. /// Returns the module name for a given address.
@ -184,82 +123,25 @@ pub fn getSymbolAtAddress(self: *SelfInfo, gpa: Allocator, address: usize) !std.
/// a path that doesn't rely on any side-effects of a prior successful module lookup. /// a path that doesn't rely on any side-effects of a prior successful module lookup.
pub fn getModuleNameForAddress(self: *SelfInfo, gpa: Allocator, address: usize) error{ Unexpected, OutOfMemory, MissingDebugInfo }![]const u8 { pub fn getModuleNameForAddress(self: *SelfInfo, gpa: Allocator, address: usize) error{ Unexpected, OutOfMemory, MissingDebugInfo }![]const u8 {
comptime assert(target_supported); comptime assert(target_supported);
const module = try self.lookupModuleForAddress(gpa, address); const module: Module = try .lookup(&self.lookup_cache, gpa, address); // MLUGG TODO: don't take gpa
return module.name; return module.name;
} }
fn lookupModuleDl(self: *SelfInfo, address: usize) !Module { const Module = switch (native_os) {
_ = self; // MLUGG else => {}, // Dwarf, // TODO MLUGG: it's this on master but that's definitely broken atm...
const DlIterContext = struct { .macos, .ios, .watchos, .tvos, .visionos => struct {
/// input /// The runtime address where __TEXT is loaded.
address: usize, text_base: usize,
/// output load_offset: usize,
module: Module, name: []const u8,
unwind_info: ?[]const u8,
fn callback(info: *posix.dl_phdr_info, size: usize, context: *@This()) !void { eh_frame: ?[]const u8,
_ = size; fn key(m: *const Module) usize {
// The base address is too high return m.text_base;
if (context.address < info.addr)
return;
const phdrs = info.phdr[0..info.phnum];
for (phdrs) |*phdr| {
if (phdr.p_type != elf.PT_LOAD) continue;
// Overflowing addition is used to handle the case of VSDOs having a p_vaddr = 0xffffffffff700000
const seg_start = info.addr +% phdr.p_vaddr;
const seg_end = seg_start + phdr.p_memsz;
if (context.address >= seg_start and context.address < seg_end) {
context.module = .{
.load_offset = info.addr,
// Android libc uses NULL instead of "" to mark the main program
.name = mem.sliceTo(info.name, 0) orelse "",
.build_id = null,
.gnu_eh_frame = null,
};
break;
} }
} else return; fn lookup(cache: *LookupCache, gpa: Allocator, address: usize) !Module {
_ = cache;
for (info.phdr[0..info.phnum]) |phdr| { _ = gpa;
switch (phdr.p_type) {
elf.PT_NOTE => {
// Look for .note.gnu.build-id
const segment_ptr: [*]const u8 = @ptrFromInt(info.addr + phdr.p_vaddr);
var r: std.Io.Reader = .fixed(segment_ptr[0..phdr.p_memsz]);
const name_size = r.takeInt(u32, native_endian) catch continue;
const desc_size = r.takeInt(u32, native_endian) catch continue;
const note_type = r.takeInt(u32, native_endian) catch continue;
const name = r.take(name_size) catch continue;
if (note_type != elf.NT_GNU_BUILD_ID) continue;
if (!mem.eql(u8, name, "GNU\x00")) continue;
const desc = r.take(desc_size) catch continue;
context.module.build_id = desc;
},
elf.PT_GNU_EH_FRAME => {
const segment_ptr: [*]const u8 = @ptrFromInt(info.addr + phdr.p_vaddr);
context.module.gnu_eh_frame = segment_ptr[0..phdr.p_memsz];
},
else => {},
}
}
// Stop the iteration
return error.Found;
}
};
var ctx: DlIterContext = .{
.address = address,
.module = undefined,
};
posix.dl_iterate_phdr(&ctx, error{Found}, DlIterContext.callback) catch |err| switch (err) {
error.Found => return ctx.module,
};
return error.MissingDebugInfo;
}
fn lookupModuleDyld(self: *SelfInfo, address: usize) !Module {
_ = self; // MLUGG
const image_count = std.c._dyld_image_count(); const image_count = std.c._dyld_image_count();
for (0..image_count) |image_idx| { for (0..image_count) |image_idx| {
const header = std.c._dyld_get_image_header(@intCast(image_idx)) orelse continue; const header = std.c._dyld_get_image_header(@intCast(image_idx)) orelse continue;
@ -306,165 +188,13 @@ fn lookupModuleDyld(self: *SelfInfo, address: usize) !Module {
return result; return result;
} }
return error.MissingDebugInfo; return error.MissingDebugInfo;
}
fn lookupModuleWin32(self: *SelfInfo, gpa: Allocator, address: usize) !Module {
if (self.lookupModuleWin32Cache(address)) |m| return m;
{
// Check a new module hasn't been loaded
self.module_cache.clearRetainingCapacity();
const handle = windows.kernel32.CreateToolhelp32Snapshot(windows.TH32CS_SNAPMODULE | windows.TH32CS_SNAPMODULE32, 0);
if (handle == windows.INVALID_HANDLE_VALUE) {
return windows.unexpectedError(windows.GetLastError());
} }
defer windows.CloseHandle(handle); fn loadDebugInfo(module: *const Module, gpa: Allocator, di: *Module.DebugInfo) !void {
return loadMachODebugInfo(gpa, module, di);
var entry: windows.MODULEENTRY32 = undefined;
entry.dwSize = @sizeOf(windows.MODULEENTRY32);
if (windows.kernel32.Module32First(handle, &entry) != 0) {
try self.module_cache.append(gpa, entry);
while (windows.kernel32.Module32Next(handle, &entry) != 0) {
try self.module_cache.append(gpa, entry);
} }
} fn loadUnwindInfo(module: *const Module, gpa: Allocator, di: *Module.DebugInfo) !void {
} // MLUGG TODO HACKHACK
try loadMachODebugInfo(gpa, module, di);
if (self.lookupModuleWin32Cache(address)) |m| return m;
return error.MissingDebugInfo;
}
fn lookupModuleWin32Cache(self: *SelfInfo, address: usize) ?Module {
for (self.module_cache.items) |*entry| {
const base_address = @intFromPtr(entry.modBaseAddr);
if (address >= base_address and address < base_address + entry.modBaseSize) {
return .{
.base_address = base_address,
.size = entry.modBaseSize,
.name = std.mem.sliceTo(&entry.szModule, 0),
.handle = entry.hModule,
};
}
}
return null;
}
fn readCoffDebugInfo(gpa: Allocator, module: *const Module, di: *Module.DebugInfo) !void {
const mapped_ptr: [*]const u8 = @ptrFromInt(module.base_address);
const mapped = mapped_ptr[0..module.size];
var coff_obj = coff.Coff.init(mapped, true) catch return error.InvalidDebugInfo;
// The string table is not mapped into memory by the loader, so if a section name is in the
// string table then we have to map the full image file from disk. This can happen when
// a binary is produced with -gdwarf, since the section names are longer than 8 bytes.
if (coff_obj.strtabRequired()) {
var name_buffer: [windows.PATH_MAX_WIDE + 4:0]u16 = undefined;
name_buffer[0..4].* = .{ '\\', '?', '?', '\\' }; // openFileAbsoluteW requires the prefix to be present
const process_handle = windows.GetCurrentProcess();
const len = windows.kernel32.GetModuleFileNameExW(
process_handle,
module.handle,
name_buffer[4..],
windows.PATH_MAX_WIDE,
);
if (len == 0) return error.MissingDebugInfo;
const coff_file = fs.openFileAbsoluteW(name_buffer[0 .. len + 4 :0], .{}) catch |err| switch (err) {
error.FileNotFound => return error.MissingDebugInfo,
else => |e| return e,
};
errdefer coff_file.close();
var section_handle: windows.HANDLE = undefined;
const create_section_rc = windows.ntdll.NtCreateSection(
&section_handle,
windows.STANDARD_RIGHTS_REQUIRED | windows.SECTION_QUERY | windows.SECTION_MAP_READ,
null,
null,
windows.PAGE_READONLY,
// The documentation states that if no AllocationAttribute is specified, then SEC_COMMIT is the default.
// In practice, this isn't the case and specifying 0 will result in INVALID_PARAMETER_6.
windows.SEC_COMMIT,
coff_file.handle,
);
if (create_section_rc != .SUCCESS) return error.MissingDebugInfo;
errdefer windows.CloseHandle(section_handle);
var coff_len: usize = 0;
var section_view_ptr: [*]const u8 = undefined;
const map_section_rc = windows.ntdll.NtMapViewOfSection(
section_handle,
process_handle,
@ptrCast(&section_view_ptr),
null,
0,
null,
&coff_len,
.ViewUnmap,
0,
windows.PAGE_READONLY,
);
if (map_section_rc != .SUCCESS) return error.MissingDebugInfo;
errdefer assert(windows.ntdll.NtUnmapViewOfSection(process_handle, @constCast(section_view_ptr)) == .SUCCESS);
const section_view = section_view_ptr[0..coff_len];
coff_obj = coff.Coff.init(section_view, false) catch return error.InvalidDebugInfo;
di.mapped_file = .{
.file = coff_file,
.section_handle = section_handle,
.section_view = section_view,
};
}
di.coff_image_base = coff_obj.getImageBase();
if (coff_obj.getSectionByName(".debug_info")) |_| {
di.dwarf = .{};
inline for (@typeInfo(Dwarf.Section.Id).@"enum".fields, 0..) |section, i| {
di.dwarf.?.sections[i] = if (coff_obj.getSectionByName("." ++ section.name)) |section_header| blk: {
break :blk .{
.data = try coff_obj.getSectionDataAlloc(section_header, gpa),
.virtual_address = section_header.virtual_address,
.owned = true,
};
} else null;
}
try di.dwarf.?.open(gpa, native_endian);
}
if (try coff_obj.getPdbPath()) |raw_path| pdb: {
const path = blk: {
if (fs.path.isAbsolute(raw_path)) {
break :blk raw_path;
} else {
const self_dir = try fs.selfExeDirPathAlloc(gpa);
defer gpa.free(self_dir);
break :blk try fs.path.join(gpa, &.{ self_dir, raw_path });
}
};
defer if (path.ptr != raw_path.ptr) gpa.free(path);
di.pdb = Pdb.init(gpa, path) catch |err| switch (err) {
error.FileNotFound, error.IsDir => break :pdb,
else => return err,
};
try di.pdb.?.parseInfoStream();
try di.pdb.?.parseDbiStream();
if (!mem.eql(u8, &coff_obj.guid, &di.pdb.?.guid) or coff_obj.age != di.pdb.?.age)
return error.InvalidDebugInfo;
di.coff_section_headers = try coff_obj.getSectionHeadersAlloc(gpa);
}
}
const Module = switch (native_os) {
else => "MLUGG TODO", // Dwarf, // TODO MLUGG: it's this on master but that's definitely broken atm...
.macos, .ios, .watchos, .tvos, .visionos => struct {
/// The runtime address where __TEXT is loaded.
text_base: usize,
load_offset: usize,
name: []const u8,
unwind_info: ?[]const u8,
eh_frame: ?[]const u8,
fn key(m: *const Module) usize {
return m.text_base;
} }
fn getSymbolAtAddress(module: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol { fn getSymbolAtAddress(module: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol {
const vaddr = address - module.load_offset; const vaddr = address - module.load_offset;
@ -524,8 +254,8 @@ const Module = switch (native_os) {
}, },
}; };
} }
const LookupCache = void;
const DebugInfo = struct { const DebugInfo = struct {
// MLUGG TODO: these are duplicated state. i actually reckon they should be removed from Module, and loadMachODebugInfo should be the one discovering them!
mapped_memory: []align(std.heap.page_size_min) const u8, mapped_memory: []align(std.heap.page_size_min) const u8,
symbols: []const MachoSymbol, symbols: []const MachoSymbol,
strings: [:0]const u8, strings: [:0]const u8,
@ -533,6 +263,7 @@ const Module = switch (native_os) {
ofiles: std.StringArrayHashMapUnmanaged(OFile), ofiles: std.StringArrayHashMapUnmanaged(OFile),
// Backed by the in-memory sections mapped by the loader // Backed by the in-memory sections mapped by the loader
// MLUGG TODO: these are duplicated state. i actually reckon they should be removed from Module, and loadMachODebugInfo should be the one discovering them!
unwind_info: ?[]const u8, unwind_info: ?[]const u8,
eh_frame: ?[]const u8, eh_frame: ?[]const u8,
@ -642,28 +373,137 @@ const Module = switch (native_os) {
}; };
}, },
.wasi, .emscripten => struct { .wasi, .emscripten => struct {
const LookupCache = void;
const DebugInfo = struct { const DebugInfo = struct {
const init: DebugInfo = .{}; const init: DebugInfo = .{};
fn getSymbolAtAddress(di: *DebugInfo, gpa: Allocator, base_address: usize, address: usize) !std.debug.Symbol { };
_ = di; fn lookup(cache: *LookupCache, gpa: Allocator, address: usize) !Module {
_ = cache;
_ = gpa; _ = gpa;
_ = base_address; _ = address;
@panic("TODO implement lookup module for Wasm");
}
fn getSymbolAtAddress(module: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol {
_ = module;
_ = gpa;
_ = di;
_ = address; _ = address;
unreachable; unreachable;
} }
}; fn loadDebugInfo(module: *const Module, gpa: Allocator, di: *DebugInfo) !void {
_ = module;
_ = gpa;
_ = di;
unreachable;
}
fn loadUnwindInfo(module: *const Module, gpa: Allocator, di: *DebugInfo) !void {
_ = module;
_ = gpa;
_ = di;
unreachable;
}
}, },
.linux, .netbsd, .freebsd, .dragonfly, .openbsd, .haiku, .solaris, .illumos => struct { .linux, .netbsd, .freebsd, .dragonfly, .openbsd, .haiku, .solaris, .illumos => struct {
load_offset: usize, load_offset: usize,
name: []const u8, name: []const u8,
build_id: ?[]const u8, build_id: ?[]const u8,
gnu_eh_frame: ?[]const u8, gnu_eh_frame: ?[]const u8,
const LookupCache = void;
const DebugInfo = Dwarf.ElfModule;
fn key(m: Module) usize { fn key(m: Module) usize {
return m.load_offset; // MLUGG TODO: is this technically valid? idk return m.load_offset; // MLUGG TODO: is this technically valid? idk
} }
const DebugInfo = Dwarf.ElfModule; fn lookup(cache: *LookupCache, gpa: Allocator, address: usize) !Module {
fn getSymbolAtAddress(mod: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol { _ = cache;
return di.getSymbolAtAddress(gpa, native_endian, mod.load_offset, address); _ = gpa;
if (native_os == .haiku) @panic("TODO implement lookup module for Haiku");
const DlIterContext = struct {
/// input
address: usize,
/// output
module: Module,
fn callback(info: *posix.dl_phdr_info, size: usize, context: *@This()) !void {
_ = size;
// The base address is too high
if (context.address < info.addr)
return;
const phdrs = info.phdr[0..info.phnum];
for (phdrs) |*phdr| {
if (phdr.p_type != elf.PT_LOAD) continue;
// Overflowing addition is used to handle the case of VSDOs having a p_vaddr = 0xffffffffff700000
const seg_start = info.addr +% phdr.p_vaddr;
const seg_end = seg_start + phdr.p_memsz;
if (context.address >= seg_start and context.address < seg_end) {
context.module = .{
.load_offset = info.addr,
// Android libc uses NULL instead of "" to mark the main program
.name = mem.sliceTo(info.name, 0) orelse "",
.build_id = null,
.gnu_eh_frame = null,
};
break;
}
} else return;
for (info.phdr[0..info.phnum]) |phdr| {
switch (phdr.p_type) {
elf.PT_NOTE => {
// Look for .note.gnu.build-id
const segment_ptr: [*]const u8 = @ptrFromInt(info.addr + phdr.p_vaddr);
var r: std.Io.Reader = .fixed(segment_ptr[0..phdr.p_memsz]);
const name_size = r.takeInt(u32, native_endian) catch continue;
const desc_size = r.takeInt(u32, native_endian) catch continue;
const note_type = r.takeInt(u32, native_endian) catch continue;
const name = r.take(name_size) catch continue;
if (note_type != elf.NT_GNU_BUILD_ID) continue;
if (!mem.eql(u8, name, "GNU\x00")) continue;
const desc = r.take(desc_size) catch continue;
context.module.build_id = desc;
},
elf.PT_GNU_EH_FRAME => {
const segment_ptr: [*]const u8 = @ptrFromInt(info.addr + phdr.p_vaddr);
context.module.gnu_eh_frame = segment_ptr[0..phdr.p_memsz];
},
else => {},
}
}
// Stop the iteration
return error.Found;
}
};
var ctx: DlIterContext = .{
.address = address,
.module = undefined,
};
posix.dl_iterate_phdr(&ctx, error{Found}, DlIterContext.callback) catch |err| switch (err) {
error.Found => return ctx.module,
};
return error.MissingDebugInfo;
}
fn loadDebugInfo(module: *const Module, gpa: Allocator, di: *Module.DebugInfo) !void {
const filename: ?[]const u8 = if (module.name.len > 0) module.name else null;
const mapped_mem = mapFileOrSelfExe(filename) catch |err| switch (err) {
error.FileNotFound => return error.MissingDebugInfo,
error.FileTooBig => return error.InvalidDebugInfo,
else => |e| return e,
};
errdefer posix.munmap(mapped_mem);
try di.load(gpa, mapped_mem, module.build_id, null, null, null, filename);
assert(di.mapped_memory != null);
}
fn loadUnwindInfo(module: *const Module, gpa: Allocator, di: *Module.DebugInfo) !void {
const section_bytes = module.gnu_eh_frame orelse return error.MissingUnwindInfo; // MLUGG TODO: load from file
const section_vaddr: u64 = @intFromPtr(section_bytes.ptr) - module.load_offset;
const header: Dwarf.Unwind.EhFrameHeader = try .parse(section_vaddr, section_bytes, @sizeOf(usize), native_endian);
try di.unwind.loadFromEhFrameHdr(header, section_vaddr, @ptrFromInt(module.load_offset + header.eh_frame_vaddr));
try di.unwind.prepareLookup(gpa, @sizeOf(usize), native_endian);
}
fn getSymbolAtAddress(module: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol {
return di.getSymbolAtAddress(gpa, native_endian, module.load_offset, address);
} }
}, },
.uefi, .windows => struct { .uefi, .windows => struct {
@ -674,6 +514,152 @@ const Module = switch (native_os) {
fn key(m: Module) usize { fn key(m: Module) usize {
return m.base_address; return m.base_address;
} }
fn lookup(cache: *LookupCache, gpa: Allocator, address: usize) !Module {
if (lookupInCache(cache, address)) |m| return m;
{
// Check a new module hasn't been loaded
cache.modules.clearRetainingCapacity();
const handle = windows.kernel32.CreateToolhelp32Snapshot(windows.TH32CS_SNAPMODULE | windows.TH32CS_SNAPMODULE32, 0);
if (handle == windows.INVALID_HANDLE_VALUE) {
return windows.unexpectedError(windows.GetLastError());
}
defer windows.CloseHandle(handle);
var entry: windows.MODULEENTRY32 = undefined;
entry.dwSize = @sizeOf(windows.MODULEENTRY32);
if (windows.kernel32.Module32First(handle, &entry) != 0) {
try cache.modules.append(gpa, entry);
while (windows.kernel32.Module32Next(handle, &entry) != 0) {
try cache.modules.append(gpa, entry);
}
}
}
if (lookupInCache(cache, address)) |m| return m;
return error.MissingDebugInfo;
}
fn lookupInCache(cache: *const LookupCache, address: usize) ?Module {
for (cache.modules.items) |*entry| {
const base_address = @intFromPtr(entry.modBaseAddr);
if (address >= base_address and address < base_address + entry.modBaseSize) {
return .{
.base_address = base_address,
.size = entry.modBaseSize,
.name = std.mem.sliceTo(&entry.szModule, 0),
.handle = entry.hModule,
};
}
}
return null;
}
fn loadDebugInfo(module: *const Module, gpa: Allocator, di: *DebugInfo) !void {
const mapped_ptr: [*]const u8 = @ptrFromInt(module.base_address);
const mapped = mapped_ptr[0..module.size];
var coff_obj = coff.Coff.init(mapped, true) catch return error.InvalidDebugInfo;
// The string table is not mapped into memory by the loader, so if a section name is in the
// string table then we have to map the full image file from disk. This can happen when
// a binary is produced with -gdwarf, since the section names are longer than 8 bytes.
if (coff_obj.strtabRequired()) {
var name_buffer: [windows.PATH_MAX_WIDE + 4:0]u16 = undefined;
name_buffer[0..4].* = .{ '\\', '?', '?', '\\' }; // openFileAbsoluteW requires the prefix to be present
const process_handle = windows.GetCurrentProcess();
const len = windows.kernel32.GetModuleFileNameExW(
process_handle,
module.handle,
name_buffer[4..],
windows.PATH_MAX_WIDE,
);
if (len == 0) return error.MissingDebugInfo;
const coff_file = fs.openFileAbsoluteW(name_buffer[0 .. len + 4 :0], .{}) catch |err| switch (err) {
error.FileNotFound => return error.MissingDebugInfo,
else => |e| return e,
};
errdefer coff_file.close();
var section_handle: windows.HANDLE = undefined;
const create_section_rc = windows.ntdll.NtCreateSection(
&section_handle,
windows.STANDARD_RIGHTS_REQUIRED | windows.SECTION_QUERY | windows.SECTION_MAP_READ,
null,
null,
windows.PAGE_READONLY,
// The documentation states that if no AllocationAttribute is specified, then SEC_COMMIT is the default.
// In practice, this isn't the case and specifying 0 will result in INVALID_PARAMETER_6.
windows.SEC_COMMIT,
coff_file.handle,
);
if (create_section_rc != .SUCCESS) return error.MissingDebugInfo;
errdefer windows.CloseHandle(section_handle);
var coff_len: usize = 0;
var section_view_ptr: [*]const u8 = undefined;
const map_section_rc = windows.ntdll.NtMapViewOfSection(
section_handle,
process_handle,
@ptrCast(&section_view_ptr),
null,
0,
null,
&coff_len,
.ViewUnmap,
0,
windows.PAGE_READONLY,
);
if (map_section_rc != .SUCCESS) return error.MissingDebugInfo;
errdefer assert(windows.ntdll.NtUnmapViewOfSection(process_handle, @constCast(section_view_ptr)) == .SUCCESS);
const section_view = section_view_ptr[0..coff_len];
coff_obj = coff.Coff.init(section_view, false) catch return error.InvalidDebugInfo;
di.mapped_file = .{
.file = coff_file,
.section_handle = section_handle,
.section_view = section_view,
};
}
di.coff_image_base = coff_obj.getImageBase();
if (coff_obj.getSectionByName(".debug_info")) |_| {
di.dwarf = .{};
inline for (@typeInfo(Dwarf.Section.Id).@"enum".fields, 0..) |section, i| {
di.dwarf.?.sections[i] = if (coff_obj.getSectionByName("." ++ section.name)) |section_header| blk: {
break :blk .{
.data = try coff_obj.getSectionDataAlloc(section_header, gpa),
.virtual_address = section_header.virtual_address,
.owned = true,
};
} else null;
}
try di.dwarf.?.open(gpa, native_endian);
}
if (try coff_obj.getPdbPath()) |raw_path| pdb: {
const path = blk: {
if (fs.path.isAbsolute(raw_path)) {
break :blk raw_path;
} else {
const self_dir = try fs.selfExeDirPathAlloc(gpa);
defer gpa.free(self_dir);
break :blk try fs.path.join(gpa, &.{ self_dir, raw_path });
}
};
defer if (path.ptr != raw_path.ptr) gpa.free(path);
di.pdb = Pdb.init(gpa, path) catch |err| switch (err) {
error.FileNotFound, error.IsDir => break :pdb,
else => return err,
};
try di.pdb.?.parseInfoStream();
try di.pdb.?.parseDbiStream();
if (!mem.eql(u8, &coff_obj.guid, &di.pdb.?.guid) or coff_obj.age != di.pdb.?.age)
return error.InvalidDebugInfo;
di.coff_section_headers = try coff_obj.getSectionHeadersAlloc(gpa);
}
}
const LookupCache = struct {
modules: std.ArrayListUnmanaged(windows.MODULEENTRY32),
const init: LookupCache = .{ .modules = .empty };
};
const DebugInfo = struct { const DebugInfo = struct {
coff_image_base: u64, coff_image_base: u64,
mapped_file: ?struct { mapped_file: ?struct {
@ -747,9 +733,9 @@ const Module = switch (native_os) {
} }
}; };
fn getSymbolAtAddress(mod: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol { fn getSymbolAtAddress(module: *const Module, gpa: Allocator, di: *DebugInfo, address: usize) !std.debug.Symbol {
// Translate the runtime address into a virtual address into the module // Translate the runtime address into a virtual address into the module
const vaddr = address - mod.base_address; const vaddr = address - module.base_address;
if (di.pdb != null) { if (di.pdb != null) {
if (try di.getSymbolFromPdb(vaddr)) |symbol| return symbol; if (try di.getSymbolFromPdb(vaddr)) |symbol| return symbol;
@ -1091,12 +1077,12 @@ fn unwindFrameDwarf(
const pc_vaddr = context.pc - load_offset; const pc_vaddr = context.pc - load_offset;
const fde_offset = explicit_fde_offset orelse try unwind.findFdeOffset( const fde_offset = explicit_fde_offset orelse try unwind.lookupPc(
pc_vaddr, pc_vaddr,
@sizeOf(usize), @sizeOf(usize),
native_endian, native_endian,
) orelse return error.MissingDebugInfo; ) orelse return error.MissingDebugInfo;
const format, const cie, const fde = try unwind.loadFde(fde_offset, @sizeOf(usize), native_endian); const format, const cie, const fde = try unwind.getFde(fde_offset, @sizeOf(usize), native_endian);
// Check if this FDE *actually* includes the address. // Check if this FDE *actually* includes the address.
if (pc_vaddr < fde.pc_begin or pc_vaddr >= fde.pc_begin + fde.pc_range) return error.MissingDebugInfo; if (pc_vaddr < fde.pc_begin or pc_vaddr >= fde.pc_begin + fde.pc_range) return error.MissingDebugInfo;
@ -1294,7 +1280,7 @@ fn unwindFrameMachO(
load_offset: usize, load_offset: usize,
context: *UnwindContext, context: *UnwindContext,
unwind_info: []const u8, unwind_info: []const u8,
eh_frame: ?[]const u8, opt_eh_frame: ?[]const u8,
) !usize { ) !usize {
if (unwind_info.len < @sizeOf(macho.unwind_info_section_header)) return error.InvalidUnwindInfo; if (unwind_info.len < @sizeOf(macho.unwind_info_section_header)) return error.InvalidUnwindInfo;
const header: *align(1) const macho.unwind_info_section_header = @ptrCast(unwind_info); const header: *align(1) const macho.unwind_info_section_header = @ptrCast(unwind_info);
@ -1304,20 +1290,6 @@ fn unwindFrameMachO(
const indices: []align(1) const macho.unwind_info_section_header_index_entry = @ptrCast(unwind_info[header.indexSectionOffset..][0..index_byte_count]); const indices: []align(1) const macho.unwind_info_section_header_index_entry = @ptrCast(unwind_info[header.indexSectionOffset..][0..index_byte_count]);
if (indices.len == 0) return error.MissingUnwindInfo; if (indices.len == 0) return error.MissingUnwindInfo;
// MLUGG TODO HACKHACK -- Unwind needs a slight refactor to make this work well
const opt_dwarf_unwind: ?Dwarf.Unwind = if (eh_frame) |eh_frame_data| .{
.debug_frame = null,
.eh_frame = .{
.header = .{
.vaddr = undefined,
.eh_frame_vaddr = @intFromPtr(eh_frame_data.ptr) - load_offset,
.search_table = null,
},
.eh_frame_data = eh_frame_data,
.sorted_fdes = null,
},
} else null;
// offset of the PC into the `__TEXT` segment // offset of the PC into the `__TEXT` segment
const pc_text_offset = context.pc - text_base; const pc_text_offset = context.pc - text_base;
@ -1533,8 +1505,11 @@ fn unwindFrameMachO(
break :ip new_ip; break :ip new_ip;
}, },
.DWARF => { .DWARF => {
const dwarf_unwind = &(opt_dwarf_unwind orelse return error.MissingEhFrame); const eh_frame = opt_eh_frame orelse return error.MissingEhFrame;
return unwindFrameDwarf(dwarf_unwind, load_offset, context, @intCast(encoding.value.x86_64.dwarf)); const eh_frame_vaddr = @intFromPtr(eh_frame.ptr) - load_offset;
var dwarf_unwind: Dwarf.Unwind = .init;
dwarf_unwind.loadFromSection(.eh_frame, eh_frame_vaddr, eh_frame);
return unwindFrameDwarf(&dwarf_unwind, load_offset, context, @intCast(encoding.value.x86_64.dwarf));
}, },
}, },
.aarch64, .aarch64_be => switch (encoding.mode.arm64) { .aarch64, .aarch64_be => switch (encoding.mode.arm64) {
@ -1547,7 +1522,10 @@ fn unwindFrameMachO(
break :ip new_ip; break :ip new_ip;
}, },
.DWARF => { .DWARF => {
const dwarf_unwind = &(opt_dwarf_unwind orelse return error.MissingEhFrame); const eh_frame = opt_eh_frame orelse return error.MissingEhFrame;
const eh_frame_vaddr = @intFromPtr(eh_frame.ptr) - load_offset;
var dwarf_unwind: Dwarf.Unwind = .init;
dwarf_unwind.loadFromSection(.eh_frame, eh_frame_vaddr, eh_frame);
return unwindFrameDwarf(dwarf_unwind, load_offset, context, @intCast(encoding.value.arm64.dwarf)); return unwindFrameDwarf(dwarf_unwind, load_offset, context, @intCast(encoding.value.arm64.dwarf));
}, },
.FRAME => ip: { .FRAME => ip: {