[llvm-size] Add --exclude-pagezero option for Mach-O to exclude __PAGEZERO size. #159574

rjmansfield · 2025-09-18T14:04:05Z

Fixes #86644

Fixes llvm#86644

llvmbot · 2025-09-18T14:04:42Z

@llvm/pr-subscribers-llvm-binary-utilities

Author: Ryan Mansfield (rjmansfield)

Changes

Fixes #86644

Full diff: https://github.com/llvm/llvm-project/pull/159574.diff

3 Files Affected:

(added) llvm/test/tools/llvm-size/macho-pagezero.test (+60)
(modified) llvm/tools/llvm-size/Opts.td (+2)
(modified) llvm/tools/llvm-size/llvm-size.cpp (+8-2)

diff --git a/llvm/test/tools/llvm-size/macho-pagezero.test b/llvm/test/tools/llvm-size/macho-pagezero.test
new file mode 100644
index 0000000000000..d53067504c3ad
--- /dev/null
+++ b/llvm/test/tools/llvm-size/macho-pagezero.test
@@ -0,0 +1,60 @@
+# Test the -z option to skip __PAGEZERO segment in Mach-O files
+
+# RUN: yaml2obj %s --docnum=1 -o %t-pagezero.o
+# RUN: llvm-size %t-pagezero.o | \
+# RUN:   FileCheck %s --check-prefix=NORMAL --match-full-lines \
+# RUN:                --strict-whitespace --implicit-check-not={{.}}
+# RUN: llvm-size -z %t-pagezero.o | \
+# RUN:   FileCheck %s --check-prefix=SKIP --match-full-lines \
+# RUN:                --strict-whitespace --implicit-check-not={{.}}
+
+# NORMAL:__TEXT	__DATA	__OBJC	others	dec	hex
+# NORMAL-NEXT:20	100	0	4096	4216	1078	
+
+# SKIP:__TEXT	__DATA	__OBJC	others	dec	hex
+# SKIP-NEXT:20	100	0	0	120	78	
+
+--- !mach-o
+FileHeader:
+  magic:           0xFEEDFACF
+  cputype:         0x100000C
+  cpusubtype:      0x0
+  filetype:        0x2
+  ncmds:           3
+  sizeofcmds:      216
+  flags:           0x2000
+  reserved:        0x0
+LoadCommands:
+  - cmd:             LC_SEGMENT_64
+    cmdsize:         72
+    segname:         __PAGEZERO
+    vmaddr:          0x0
+    vmsize:          4096
+    fileoff:         0
+    filesize:        0
+    maxprot:         0
+    initprot:        0
+    nsects:          0
+    flags:           0
+  - cmd:             LC_SEGMENT_64
+    cmdsize:         72
+    segname:         __TEXT
+    vmaddr:          0x100000000
+    vmsize:          20
+    fileoff:         248
+    filesize:        20
+    maxprot:         7
+    initprot:        5
+    nsects:          0
+    flags:           0
+  - cmd:             LC_SEGMENT_64
+    cmdsize:         72
+    segname:         __DATA
+    vmaddr:          0x100001000
+    vmsize:          100
+    fileoff:         268
+    filesize:        100
+    maxprot:         7
+    initprot:        3
+    nsects:          0
+    flags:           0
diff --git a/llvm/tools/llvm-size/Opts.td b/llvm/tools/llvm-size/Opts.td
index edae43f1abd24..65478730c2801 100644
--- a/llvm/tools/llvm-size/Opts.td
+++ b/llvm/tools/llvm-size/Opts.td
@@ -21,6 +21,8 @@ def grp_mach_o : OptionGroup<"kind">, HelpText<"OPTIONS (Mach-O specific)">;
 def arch_EQ : Joined<["--"], "arch=">, HelpText<"architecture(s) from a Mach-O file to dump">, Group<grp_mach_o>;
 def : Separate<["--", "-"], "arch">, Alias<arch_EQ>;
 def l : F<"l", "When format is darwin, use long format to include addresses and offsets">, Group<grp_mach_o>;
+def z : F<"z", "Do not include __PAGEZERO segment in totals">,
+        Group<grp_mach_o>;
 
 def : F<"A", "Alias for --format">, Alias<format_EQ>, AliasArgs<["sysv"]>;
 def : F<"B", "Alias for --format">, Alias<format_EQ>, AliasArgs<["berkeley"]>;
diff --git a/llvm/tools/llvm-size/llvm-size.cpp b/llvm/tools/llvm-size/llvm-size.cpp
index acc7843ffac8b..805f8ed1e6dcd 100644
--- a/llvm/tools/llvm-size/llvm-size.cpp
+++ b/llvm/tools/llvm-size/llvm-size.cpp
@@ -79,6 +79,7 @@ static bool DarwinLongFormat;
 static RadixTy Radix = RadixTy::decimal;
 static bool TotalSizes;
 static bool HasMachOFiles = false;
+static bool SkipPageZero = false;
 
 static std::vector<std::string> InputFilenames;
 
@@ -307,7 +308,9 @@ static void printDarwinSegmentSizes(MachOObjectFile *MachO) {
         }
       } else {
         StringRef SegmentName = StringRef(Seg.segname);
-        if (SegmentName == "__TEXT")
+        if (SkipPageZero && SegmentName == "__PAGEZERO")
+          ; // Skip __PAGEZERO segment
+        else if (SegmentName == "__TEXT")
           total_text += Seg.vmsize;
         else if (SegmentName == "__DATA")
           total_data += Seg.vmsize;
@@ -333,7 +336,9 @@ static void printDarwinSegmentSizes(MachOObjectFile *MachO) {
         }
       } else {
         StringRef SegmentName = StringRef(Seg.segname);
-        if (SegmentName == "__TEXT")
+        if (SkipPageZero && SegmentName == "__PAGEZERO")
+          ; // Skip __PAGEZERO segment
+        else if (SegmentName == "__TEXT")
           total_text += Seg.vmsize;
         else if (SegmentName == "__DATA")
           total_data += Seg.vmsize;
@@ -914,6 +919,7 @@ int llvm_size_main(int argc, char **argv, const llvm::ToolContext &) {
 
   ELFCommons = Args.hasArg(OPT_common);
   DarwinLongFormat = Args.hasArg(OPT_l);
+  SkipPageZero = Args.hasArg(OPT_z);
   TotalSizes = Args.hasArg(OPT_totals);
   StringRef V = Args.getLastArgValue(OPT_format_EQ, "berkeley");
   if (V == "berkeley")

jh7370

Please make sure to update the llvm-size command guide (llvm/docs/CommandGuide/llvm-size.rst) with the new option.

This will need a Mach-O developer to review, since I don't know the intricacies of the segment layout for these files. @drodriguez / @danzimm / @alexander-shaposhnikov, can any of you help?

llvm/test/tools/llvm-size/macho-pagezero.test

jh7370 · 2025-09-23T08:51:55Z

llvm/test/tools/llvm-size/macho-pagezero.test

+# RUN:   FileCheck %s --check-prefix=NORMAL --match-full-lines \
+# RUN:                --strict-whitespace --implicit-check-not={{.}}
+# RUN: llvm-size -z %t-pagezero.o | \
+# RUN:   FileCheck %s --check-prefix=SKIP --match-full-lines \
+# RUN:                --strict-whitespace --implicit-check-not={{.}}


As this test isn't about the format of the output, you can get away with omitting the --strict-whitespace and --implicit-check-not options. You might want to keep --match-full-lines, so that you don't accidentally miss a leading/trailing digit in the first/last number.

llvm/tools/llvm-size/Opts.td

llvm/tools/llvm-size/llvm-size.cpp

drodriguez · 2025-09-23T17:00:10Z

If we use Xcode's size instead of llvm-size, I get the same display.

$ build/bin/yaml2obj llvm/test/tools/llvm-size/macho-pagezero.test --docnum=1 -o macho-pagezero.o
$ build/bin/llvm-size macho-pagezero.o
__TEXT	__DATA	__OBJC	others	dec	hex
20	100	0	4096	4216	1078
$ size macho-pagezero.o
__TEXT	__DATA	__OBJC	others	dec	hex
20	100	0	4096	4216	1078

The man page for size does not have a -z option (it has -m, -l, -x and -arch. Except -arch, the rest seems to be supported). Like @jh7370 says, maybe -z is not the best spelling for an option here, and a long one that makes clear is not for compatibility makes more sense.

But in the Xcode man page one can also find the following:

Size (without the -m option) prints the (decimal) number of bytes required by the __TEXT, __DATA and __OBJC segments. All other segments are totaled and that size is listed in the `others' column.

Which explains the "others" category including __PAGEZERO.

If we want to provide this special behaviour because of a feature request, it should be obvious that it is not a compatibility flag or similar.

rjmansfield · 2025-09-23T17:47:21Z

Something like --skip-pagezero or --exclude-pagezero maybe? From the issue, the counting of the 4G __PAGEZERO segment is making the totals appear skewed and causing some confusion. The bug report calls the default output "useless" because of it. While the default output is correct, a new option that excludes would help people see the numbers are expecting to see.

jh7370 · 2025-09-24T07:21:18Z

I'd go with --exclude-pagezero, personally. Or you could go with a more general-purpose option of --exclude-segment=<segment-name>, but that's probably overkill.

Address code review comments.

jh7370

The documentation still needs updating, as stated in my earlier comment.

Also, the test only covers the LC_SEGMENT_64 case, but you've modified both that and the LC_SEGMENT path.

Add test for 32 bit __PAGEZERO segment.

jh7370 · 2025-09-26T07:01:42Z

llvm/docs/CommandGuide/llvm-size.rst

+.. option:: --exclude-pagezero
+
+ Do not include the ``__PAGEZERO`` segment when calculating size information
+ for Mach-O files. ``__PAGEZERO`` segment is a virtual memory region used


Suggested change

for Mach-O files. ``__PAGEZERO`` segment is a virtual memory region used

for Mach-O files. The ``__PAGEZERO`` segment is a virtual memory region used

llvm/test/tools/llvm-size/macho-pagezero.test

jh7370

LGTM, thanks. Let me know if you need me to merge this for you again.

rjmansfield · 2025-09-29T11:30:02Z

@jh7370 If you could merge it, I'd appreciate it. Thanks.

…EZERO size. (llvm#159574) Do not include the ``__PAGEZERO`` segment when calculating size information for Mach-O files when `--exclude-pagezero` is used. The ``__PAGEZERO`` segment is a virtual memory region used for memory protection that does not contribute to actual size, and excluding can provide a better representation of actual size. Fixes llvm#86644

[llvm-size] Add -z option for Mach-O to exclude __PAGEZERO size.

91ba62d

Fixes llvm#86644

llvmbot added the llvm:binary-utilities label Sep 18, 2025

jh7370 reviewed Sep 23, 2025

View reviewed changes

drodriguez mentioned this pull request Sep 23, 2025

Default llvm-size output format berkeley (but also darwin) is useless on macOS #86644

Closed

Update new option to --exclude-pagezero.

7e7b187

Address code review comments.

drodriguez approved these changes Sep 24, 2025

View reviewed changes

rjmansfield changed the title ~~[llvm-size] Add -z option for Mach-O to exclude __PAGEZERO size.~~ [llvm-size] Add --exclude-pagezero option for Mach-O to exclude __PAGEZERO size. Sep 24, 2025

jh7370 requested changes Sep 25, 2025

View reviewed changes

Add entry in the command guide for --exclude-pagezero option.

ed93efb

Add test for 32 bit __PAGEZERO segment.

jh7370 reviewed Sep 26, 2025

View reviewed changes

Use same check values for 32bit test.

711273c

jh7370 approved these changes Sep 29, 2025

View reviewed changes

drodriguez merged commit 30b0215 into llvm:main Sep 29, 2025
10 checks passed

rjmansfield deleted the llvmsize_exclude_pagezero branch September 29, 2025 23:29

	for Mach-O files. ``__PAGEZERO`` segment is a virtual memory region used
	for Mach-O files. The ``__PAGEZERO`` segment is a virtual memory region used

[llvm-size] Add --exclude-pagezero option for Mach-O to exclude __PAGEZERO size. #159574

[llvm-size] Add --exclude-pagezero option for Mach-O to exclude __PAGEZERO size. #159574

Uh oh!

Conversation

rjmansfield commented Sep 18, 2025

Uh oh!

llvmbot commented Sep 18, 2025

Uh oh!

jh7370 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jh7370 Sep 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

drodriguez commented Sep 23, 2025

Uh oh!

rjmansfield commented Sep 23, 2025

Uh oh!

jh7370 commented Sep 24, 2025

Uh oh!

jh7370 left a comment

Choose a reason for hiding this comment

Uh oh!

jh7370 Sep 26, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jh7370 left a comment

Choose a reason for hiding this comment

Uh oh!

rjmansfield commented Sep 29, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants