link/macho: apply fixes to deduping logic

* test non-ObjC literal deduping logic
This commit is contained in:
Jakub Konka 2024-05-22 07:47:08 +02:00
parent 434e69482e
commit 8fc0c7dce1
6 changed files with 316 additions and 42 deletions

View File

@ -1500,14 +1500,25 @@ pub fn dedupLiterals(self: *MachO) !void {
const gpa = self.base.comp.gpa;
var lp: LiteralPool = .{};
defer lp.deinit(gpa);
if (self.getZigObject()) |zo| {
try zo.dedupLiterals(&lp, self);
try zo.resolveLiterals(&lp, self);
}
for (self.objects.items) |index| {
try self.getFile(index).?.object.dedupLiterals(&lp, self);
try self.getFile(index).?.object.resolveLiterals(&lp, self);
}
if (self.getInternalObject()) |object| {
try object.dedupLiterals(&lp, self);
try object.resolveLiterals(&lp, self);
}
if (self.getZigObject()) |zo| {
zo.dedupLiterals(lp, self);
}
for (self.objects.items) |index| {
self.getFile(index).?.object.dedupLiterals(lp, self);
}
if (self.getInternalObject()) |object| {
object.dedupLiterals(lp, self);
}
}
@ -4415,8 +4426,14 @@ pub const LiteralPool = struct {
lp.data.deinit(allocator);
}
pub fn getAtom(lp: LiteralPool, index: Index, macho_file: *MachO) *Atom {
assert(index < lp.values.items.len);
return macho_file.getAtom(lp.values.items[index]).?;
}
const InsertResult = struct {
found_existing: bool,
index: Index,
atom: *Atom.Index,
};
@ -4434,6 +4451,7 @@ pub const LiteralPool = struct {
}
return .{
.found_existing = gop.found_existing,
.index = @intCast(gop.index),
.atom = &lp.values.items[gop.index],
};
}
@ -4472,6 +4490,8 @@ pub const LiteralPool = struct {
return key.hash(ctx.lp);
}
};
pub const Index = u32;
};
const HotUpdateState = struct {

View File

@ -113,12 +113,18 @@ pub fn getThunk(self: Atom, macho_file: *MachO) *Thunk {
return macho_file.getThunk(extra.thunk);
}
pub fn getLiteralPoolIndex(self: Atom, macho_file: *MachO) ?MachO.LiteralPool.Index {
if (!self.flags.literal_pool) return null;
return self.getExtra(macho_file).?.literal_index;
}
const AddExtraOpts = struct {
thunk: ?u32 = null,
rel_index: ?u32 = null,
rel_count: ?u32 = null,
unwind_index: ?u32 = null,
unwind_count: ?u32 = null,
literal_index: ?u32 = null,
};
pub fn addExtra(atom: *Atom, opts: AddExtraOpts, macho_file: *MachO) !void {
@ -1177,7 +1183,7 @@ pub const Flags = packed struct {
/// Specifies whether this atom is alive or has been garbage collected.
alive: bool = true,
/// Specifies if the atom has been visited during garbage collection.
/// Specifies if this atom has been visited during garbage collection.
visited: bool = false,
/// Whether this atom has a range extension thunk.
@ -1188,6 +1194,9 @@ pub const Flags = packed struct {
/// Whether this atom has any unwind records.
unwind: bool = false,
/// Whether this atom has LiteralPool entry.
literal_pool: bool = false,
};
pub const Extra = struct {
@ -1205,6 +1214,9 @@ pub const Extra = struct {
/// Count of relocations belonging to this atom.
unwind_count: u32 = 0,
/// Index into LiteralPool entry for this atom.
literal_index: u32 = 0,
};
pub const Alignment = @import("../../InternPool.zig").Alignment;

View File

@ -109,12 +109,9 @@ fn addObjcSelrefsSection(self: *InternalObject, methname_atom_index: Atom.Index,
return atom_index;
}
pub fn dedupLiterals(self: InternalObject, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
pub fn resolveLiterals(self: InternalObject, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
const gpa = macho_file.base.comp.gpa;
var killed_atoms = std.AutoHashMap(Atom.Index, Atom.Index).init(gpa);
defer killed_atoms.deinit();
var buffer = std.ArrayList(u8).init(gpa);
defer buffer.deinit();
@ -126,10 +123,9 @@ pub fn dedupLiterals(self: InternalObject, lp: *MachO.LiteralPool, macho_file: *
const res = try lp.insert(gpa, header.type(), data);
if (!res.found_existing) {
res.atom.* = atom_index;
continue;
}
atom.flags.alive = false;
try killed_atoms.putNoClobber(atom_index, res.atom.*);
atom.flags.literal_pool = true;
try atom.addExtra(.{ .literal_index = res.index }, macho_file);
} else if (Object.isPtrLiteral(header)) {
const atom = macho_file.getAtom(atom_index).?;
const relocs = atom.getRelocs(macho_file);
@ -145,32 +141,45 @@ pub fn dedupLiterals(self: InternalObject, lp: *MachO.LiteralPool, macho_file: *
buffer.clearRetainingCapacity();
if (!res.found_existing) {
res.atom.* = atom_index;
continue;
}
atom.flags.alive = false;
try killed_atoms.putNoClobber(atom_index, res.atom.*);
atom.flags.literal_pool = true;
try atom.addExtra(.{ .literal_index = res.index }, macho_file);
}
}
}
pub fn dedupLiterals(self: InternalObject, lp: MachO.LiteralPool, macho_file: *MachO) void {
for (self.atoms.items) |atom_index| {
if (killed_atoms.get(atom_index)) |_| continue;
const atom = macho_file.getAtom(atom_index) orelse continue;
if (!atom.flags.alive) continue;
if (!atom.flags.relocs) continue;
const relocs = blk: {
const extra = atom.getExtra(macho_file).?;
const relocs = slice.items(.relocs)[atom.n_sect].items;
const relocs = self.sections.items(.relocs)[atom.n_sect].items;
break :blk relocs[extra.rel_index..][0..extra.rel_count];
};
for (relocs) |*rel| switch (rel.tag) {
.local => if (killed_atoms.get(rel.target)) |new_target| {
rel.target = new_target;
.local => {
const target = macho_file.getAtom(rel.target).?;
if (target.getLiteralPoolIndex(macho_file)) |lp_index| {
const lp_atom = lp.getAtom(lp_index, macho_file);
if (target.atom_index != lp_atom.atom_index) {
target.flags.alive = false;
rel.target = lp_atom.atom_index;
}
}
},
.@"extern" => {
const target = rel.getTargetSymbol(macho_file);
if (killed_atoms.get(target.atom)) |new_atom| {
target.atom = new_atom;
const target_sym = rel.getTargetSymbol(macho_file);
if (target_sym.getAtom(macho_file)) |target_atom| {
if (target_atom.getLiteralPoolIndex(macho_file)) |lp_index| {
const lp_atom = lp.getAtom(lp_index, macho_file);
if (target_atom.atom_index != lp_atom.atom_index) {
target_atom.flags.alive = false;
target_sym.atom = lp_atom.atom_index;
}
}
}
},
};
@ -179,9 +188,15 @@ pub fn dedupLiterals(self: InternalObject, lp: *MachO.LiteralPool, macho_file: *
for (self.symbols.items) |sym_index| {
const sym = macho_file.getSymbol(sym_index);
if (!sym.flags.objc_stubs) continue;
const extra = sym.getExtra(macho_file).?;
if (killed_atoms.get(extra.objc_selrefs)) |new_atom| {
try sym.addExtra(.{ .objc_selrefs = new_atom }, macho_file);
var extra = sym.getExtra(macho_file).?;
const atom = macho_file.getAtom(extra.objc_selrefs).?;
if (atom.getLiteralPoolIndex(macho_file)) |lp_index| {
const lp_atom = lp.getAtom(lp_index, macho_file);
if (atom.atom_index != lp_atom.atom_index) {
atom.flags.alive = false;
extra.objc_selrefs = lp_atom.atom_index;
sym.setExtra(extra, macho_file);
}
}
}
}

View File

@ -527,12 +527,9 @@ fn initPointerLiterals(self: *Object, macho_file: *MachO) !void {
}
}
pub fn dedupLiterals(self: Object, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
pub fn resolveLiterals(self: Object, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
const gpa = macho_file.base.comp.gpa;
var killed_atoms = std.AutoHashMap(Atom.Index, Atom.Index).init(gpa);
defer killed_atoms.deinit();
var buffer = std.ArrayList(u8).init(gpa);
defer buffer.deinit();
@ -548,10 +545,9 @@ pub fn dedupLiterals(self: Object, lp: *MachO.LiteralPool, macho_file: *MachO) !
const res = try lp.insert(gpa, header.type(), atom_data);
if (!res.found_existing) {
res.atom.* = sub.atom;
continue;
}
atom.flags.alive = false;
try killed_atoms.putNoClobber(sub.atom, res.atom.*);
atom.flags.literal_pool = true;
try atom.addExtra(.{ .literal_index = res.index }, macho_file);
}
} else if (isPtrLiteral(header)) {
for (subs.items) |sub| {
@ -572,33 +568,46 @@ pub fn dedupLiterals(self: Object, lp: *MachO.LiteralPool, macho_file: *MachO) !
buffer.clearRetainingCapacity();
if (!res.found_existing) {
res.atom.* = sub.atom;
continue;
}
atom.flags.alive = false;
try killed_atoms.putNoClobber(sub.atom, res.atom.*);
atom.flags.literal_pool = true;
try atom.addExtra(.{ .literal_index = res.index }, macho_file);
}
}
}
}
pub fn dedupLiterals(self: Object, lp: MachO.LiteralPool, macho_file: *MachO) void {
for (self.atoms.items) |atom_index| {
if (killed_atoms.get(atom_index)) |_| continue;
const atom = macho_file.getAtom(atom_index) orelse continue;
if (!atom.flags.alive) continue;
if (!atom.flags.relocs) continue;
const relocs = blk: {
const extra = atom.getExtra(macho_file).?;
const relocs = slice.items(.relocs)[atom.n_sect].items;
const relocs = self.sections.items(.relocs)[atom.n_sect].items;
break :blk relocs[extra.rel_index..][0..extra.rel_count];
};
for (relocs) |*rel| switch (rel.tag) {
.local => if (killed_atoms.get(rel.target)) |new_target| {
rel.target = new_target;
.local => {
const target = macho_file.getAtom(rel.target).?;
if (target.getLiteralPoolIndex(macho_file)) |lp_index| {
const lp_atom = lp.getAtom(lp_index, macho_file);
if (target.atom_index != lp_atom.atom_index) {
target.flags.alive = false;
rel.target = lp_atom.atom_index;
}
}
},
.@"extern" => {
const target = rel.getTargetSymbol(macho_file);
if (killed_atoms.get(target.atom)) |new_atom| {
target.atom = new_atom;
const target_sym = rel.getTargetSymbol(macho_file);
if (target_sym.getAtom(macho_file)) |target_atom| {
if (target_atom.getLiteralPoolIndex(macho_file)) |lp_index| {
const lp_atom = lp.getAtom(lp_index, macho_file);
if (target_atom.atom_index != lp_atom.atom_index) {
target_atom.flags.alive = false;
target_sym.atom = lp_atom.atom_index;
}
}
}
},
};

View File

@ -314,7 +314,14 @@ pub fn checkDuplicates(self: *ZigObject, dupes: anytype, macho_file: *MachO) !vo
}
}
pub fn dedupLiterals(self: *ZigObject, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
pub fn resolveLiterals(self: *ZigObject, lp: *MachO.LiteralPool, macho_file: *MachO) !void {
_ = self;
_ = lp;
_ = macho_file;
// TODO
}
pub fn dedupLiterals(self: *ZigObject, lp: MachO.LiteralPool, macho_file: *MachO) void {
_ = self;
_ = lp;
_ = macho_file;

View File

@ -38,6 +38,8 @@ pub fn testAll(b: *Build, build_opts: BuildOptions) *Step {
macho_step.dependOn(testLayout(b, .{ .target = default_target }));
macho_step.dependOn(testLinkingStaticLib(b, .{ .target = default_target }));
macho_step.dependOn(testLinksection(b, .{ .target = default_target }));
macho_step.dependOn(testMergeLiterals(b, .{ .target = aarch64_target }));
macho_step.dependOn(testMergeLiterals2(b, .{ .target = aarch64_target }));
macho_step.dependOn(testMhExecuteHeader(b, .{ .target = default_target }));
macho_step.dependOn(testNoDeadStrip(b, .{ .target = default_target }));
macho_step.dependOn(testNoExportsDylib(b, .{ .target = default_target }));
@ -914,6 +916,215 @@ fn testLinksection(b: *Build, opts: Options) *Step {
return test_step;
}
fn testMergeLiterals(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "merge-literals", opts);
const a_o = addObject(b, opts, .{ .name = "a", .asm_source_bytes =
\\.globl _q1
\\.globl _s1
\\
\\.align 4
\\_q1:
\\ adrp x8, L._q1@PAGE
\\ ldr d0, [x8, L._q1@PAGEOFF]
\\ ret
\\
\\.section __TEXT,__cstring,cstring_literals
\\l._s1:
\\ .asciz "hello"
\\
\\.section __TEXT,__literal8,8byte_literals
\\.align 8
\\L._q1:
\\ .double 1.2345
\\
\\.section __DATA,__data
\\.align 8
\\_s1:
\\ .quad l._s1
});
const b_o = addObject(b, opts, .{ .name = "b", .asm_source_bytes =
\\.globl _q2
\\.globl _s2
\\.globl _s3
\\
\\.align 4
\\_q2:
\\ adrp x8, L._q2@PAGE
\\ ldr d0, [x8, L._q2@PAGEOFF]
\\ ret
\\
\\.section __TEXT,__cstring,cstring_literals
\\l._s2:
\\ .asciz "hello"
\\l._s3:
\\ .asciz "world"
\\
\\.section __TEXT,__literal8,8byte_literals
\\.align 8
\\L._q2:
\\ .double 1.2345
\\
\\.section __DATA,__data
\\.align 8
\\_s2:
\\ .quad l._s2
\\_s3:
\\ .quad l._s3
});
const main_o = addObject(b, opts, .{ .name = "main", .c_source_bytes =
\\#include <stdio.h>
\\extern double q1();
\\extern double q2();
\\extern const char* s1;
\\extern const char* s2;
\\extern const char* s3;
\\int main() {
\\ printf("%s, %s, %s, %f, %f", s1, s2, s3, q1(), q2());
\\ return 0;
\\}
});
{
const exe = addExecutable(b, opts, .{ .name = "main1" });
exe.addObject(a_o);
exe.addObject(b_o);
exe.addObject(main_o);
const run = addRunArtifact(exe);
run.expectStdOutEqual("hello, hello, world, 1.234500, 1.234500");
test_step.dependOn(&run.step);
const check = exe.checkObject();
check.dumpSection("__TEXT,__const");
check.checkContains("\x8d\x97n\x12\x83\xc0\xf3?");
check.dumpSection("__TEXT,__cstring");
check.checkContains("hello\x00world\x00%s, %s, %s, %f, %f\x00");
test_step.dependOn(&check.step);
}
{
const exe = addExecutable(b, opts, .{ .name = "main2" });
exe.addObject(b_o);
exe.addObject(a_o);
exe.addObject(main_o);
const run = addRunArtifact(exe);
run.expectStdOutEqual("hello, hello, world, 1.234500, 1.234500");
test_step.dependOn(&run.step);
const check = exe.checkObject();
check.dumpSection("__TEXT,__const");
check.checkContains("\x8d\x97n\x12\x83\xc0\xf3?");
check.dumpSection("__TEXT,__cstring");
check.checkContains("hello\x00world\x00%s, %s, %s, %f, %f\x00");
test_step.dependOn(&check.step);
}
{
const c_o = addObject(b, opts, .{ .name = "c" });
c_o.addObject(a_o);
c_o.addObject(b_o);
c_o.addObject(main_o);
const exe = addExecutable(b, opts, .{ .name = "main3" });
exe.addObject(c_o);
const run = addRunArtifact(exe);
run.expectStdOutEqual("hello, hello, world, 1.234500, 1.234500");
test_step.dependOn(&run.step);
const check = exe.checkObject();
check.dumpSection("__TEXT,__const");
check.checkContains("\x8d\x97n\x12\x83\xc0\xf3?");
check.dumpSection("__TEXT,__cstring");
check.checkContains("hello\x00world\x00%s, %s, %s, %f, %f\x00");
test_step.dependOn(&check.step);
}
return test_step;
}
/// This particular test case will generate invalid machine code that will segfault at runtime.
/// However, this is by design as we want to test that the linker does not panic when linking it
/// which is also the case for the system linker and lld - linking succeeds, runtime segfaults.
/// It should also be mentioned that runtime segfault is not due to the linker but faulty input asm.
fn testMergeLiterals2(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "merge-literals-2", opts);
const a_o = addObject(b, opts, .{ .name = "a", .asm_source_bytes =
\\.globl _q1
\\.globl _s1
\\
\\.align 4
\\_q1:
\\ adrp x0, L._q1@PAGE
\\ ldr x0, [x0, L._q1@PAGEOFF]
\\ ret
\\
\\.section __TEXT,__cstring,cstring_literals
\\_s1:
\\ .asciz "hello"
\\
\\.section __TEXT,__literal8,8byte_literals
\\.align 8
\\L._q1:
\\ .double 1.2345
});
const b_o = addObject(b, opts, .{ .name = "b", .asm_source_bytes =
\\.globl _q2
\\.globl _s2
\\.globl _s3
\\
\\.align 4
\\_q2:
\\ adrp x0, L._q2@PAGE
\\ ldr x0, [x0, L._q2@PAGEOFF]
\\ ret
\\
\\.section __TEXT,__cstring,cstring_literals
\\_s2:
\\ .asciz "hello"
\\_s3:
\\ .asciz "world"
\\
\\.section __TEXT,__literal8,8byte_literals
\\.align 8
\\L._q2:
\\ .double 1.2345
});
const main_o = addObject(b, opts, .{ .name = "main", .c_source_bytes =
\\#include <stdio.h>
\\extern double q1();
\\extern double q2();
\\extern const char* s1;
\\extern const char* s2;
\\extern const char* s3;
\\int main() {
\\ printf("%s, %s, %s, %f, %f", s1, s2, s3, q1(), q2());
\\ return 0;
\\}
});
const exe = addExecutable(b, opts, .{ .name = "main1" });
exe.addObject(a_o);
exe.addObject(b_o);
exe.addObject(main_o);
const check = exe.checkObject();
check.dumpSection("__TEXT,__const");
check.checkContains("\x8d\x97n\x12\x83\xc0\xf3?");
check.dumpSection("__TEXT,__cstring");
check.checkContains("hello\x00world\x00%s, %s, %s, %f, %f\x00");
test_step.dependOn(&check.step);
return test_step;
}
fn testMhExecuteHeader(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "mh-execute-header", opts);