[llvm-readobj] Dump callgraph section info for ELF #157499

Prabhuk · 2025-09-08T15:58:03Z

Introduce a new flag --call-graph-info which outputs callgraph ELF
section information to the console as a text output or as JSON output.

github-actions · 2025-09-08T16:00:45Z

✅ With the latest revision this PR passed the C/C++ code formatter.

Introduce a new flag `--call-graph-info` which outputs callgraph ELF section information to the console as a text output or as JSON output.

…dth size.

…toms out to llvm_unreachable.

llvm/tools/llvm-readobj/ELFDumper.cpp

llvm/test/tools/llvm-readobj/ELF/call-graph-info.test

llvm/tools/llvm-readobj/ELFDumper.cpp

llvm/test/tools/llvm-readobj/ELF/call-graph-info.test

llvm/test/tools/llvm-readobj/ELF/call-graph-info-err-missing-indirect.s

jh7370

Please make sure to update the llvm-readobj and llvm-readelf CommandGuide documents under llvm/docs.

I'm keen to review this, as I've seen several things when skimming I'd like addressed, but don't have time to review it properly this week, due to other reviews and some planned time off.

jh7370 · 2025-11-03T09:34:03Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+#include "llvm/Support/DataExtractor.h"
+#include "llvm/Support/Debug.h"
 #include "llvm/Support/Endian.h"
+#include "llvm/Support/Error.h"


I'm surprised some/most/all of these new headers need to be added? Is your IDE adding additional headers unnecessarily?

I think the IDE is adding the corresponding header whenever I initialize a type. Cleaned up the header includes. PTAL.

jh7370 · 2025-11-03T09:41:46Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+  }
+}
+
+template <class ELFT> void GNUELFDumper<ELFT>::printCallGraphInfo() {


Do we need a GNU-style dumper? In other words, does GNU readelf support the section? If it doesn't, this function shouldn't be implemented, or at the very most, should do the same as the LLVM dumper. See various other modes for examples.

For context: llvm-readobj has three ELF output modes: JSON (which produces legal JSON output), LLVM (which is a human-readable dictionary-like structure), and GNU. GNU output is supposed to mirror GNU's readelf output, as close as possible. This is because people expect it to be a drop-in replacement in this mode. If there isn't a GNU readelf option to dump it, we shouldn't try to make up one.

Dropped the GNU dumper implementation as suggested.

ilovepi

I think this is basically LGTM from me. Lets make sure we wait until @jh7370 weighs in w/ a full review to land though.

Prabhuk · 2025-11-04T23:04:42Z

I think this is basically LGTM from me. Lets make sure we wait until @jh7370 weighs in w/ a full review to land though.

Thank you. I will be adding a few more tests and will wait for @jh7370's review.

llvm/test/tools/llvm-readobj/ELF/call-graph-info-relocatable-obj.test

ilovepi · 2025-11-05T17:10:40Z

llvm/test/tools/llvm-readobj/ELF/call-graph-info-callgraph-section.s

+# JSON-NEXT:          "NumDirectCallees": 1,
+# JSON-NEXT:          "DirectCallees": [
+# JSON-NEXT:            {
+# JSON-NEXT:              "Offset": 19


What is Offset here? is it the address? offset from the start of the section?

Offset is pointless actually. It's the offset of the symbol for which we couldn't find corresponding relocation. I removed "offset" and turned it into a warning for this scenario.

jh7370

Sorry for the delay in looking. The CommandGuide docs still haven't been updated. Please update them with the new option.

I've made another pass, but have run out of time to look at everything. Hopefully there's enough here to keep going on.

jh7370 · 2025-11-19T09:05:58Z

llvm/test/tools/llvm-readobj/ELF/call-graph-info-callgraph-section.s

@@ -0,0 +1,150 @@
+## Tests --call-graph-info prints information from call graph section.
+
+# REQUIRES: x86-registered-target


Could we convert the test input to use yaml2obj, instead of requiring x86? It looks like the file format is trivial enough that something like a YAML ELF input using ContentArray would be just as descriptive, whilst also allowing more precise control of everything.

jh7370 · 2025-11-19T09:06:41Z

llvm/test/tools/llvm-readobj/ELF/call-graph-info-callgraph-section.s

+# RUN: llvm-readelf --elf-output-style=LLVM --call-graph-info %t 2>&1 | FileCheck %s -DFILE=%t --check-prefix=LLVM
+# RUN: llvm-readelf --elf-output-style=JSON --pretty-print --call-graph-info %t 2>&1 | FileCheck %s -DFILE=%t --check-prefix=JSON
+
+


Nit: why the double blank line?

jh7370 · 2025-11-19T09:10:02Z

llvm/test/tools/llvm-readobj/ELF/call-graph-info-callgraph-section.s

+
+
+## We do not support GNU format console output for --call-graph-info as it is an LLVM only info.
+# CHECK-NOT: .


This isn't going to do what you think it will. This checks simply that the character "." doesn't appear anywhere in the output. What you probably wanted was {{.}} which is a regex pattern of any character. However, if the output is truly empty, the better method is to use count 0. You should find other examples elsewhere in the testsuite.

Neat. Thank you. Done.

jh7370 · 2025-11-19T09:14:20Z

llvm/tools/llvm-readobj/Opts.td

 def arch_specific : FF<"arch-specific", "Display architecture-specific information">;
 def bb_addr_map : FF<"bb-addr-map", "Display the BB address map section">;
 def pretty_pgo_analysis_map : FF<"pretty-pgo-analysis-map", "Display PGO analysis values with formatting rather than raw numbers">;
+def call_graph_info : FF<"call-graph-info", "Display call graph information">;


The option for the call graph profile abbreviates it to cg-profile. Should this also be cg-info? I don't know much about the functionality, so I might be mistaken.

These two things are unrelated. I'd like to keep the name to call_graph_info which is consistent with the flags in Clang/LLVM. Same for other similar suggestions made towards abbreviating call_graph to cg.

jh7370 · 2025-11-19T09:15:08Z

llvm/tools/llvm-readobj/ObjDumper.h

  virtual void printGroupSections() {}
  virtual void printHashHistograms() {}
  virtual void printCGProfile() {}
+  virtual void printCallGraphInfo() {}


Perhaps printCGInfo() to mirror printCGProfile?

jh7370 · 2025-11-19T10:12:50Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+      }
+    }
+    W.printNumber("NumIndirectTargetTypeIDs", CGInfo.IndirectTypeIDs.size());
+    auto IndirectTypeIdsList = SmallVector<uint64_t, 4>(


Suggested change

auto IndirectTypeIdsList = SmallVector<uint64_t, 4>(

SmallVector<uint64_t, 4> IndirectTypeIdsList(

This is more standard syntax in LLVM, I feel?

jh7370 · 2025-11-19T10:14:51Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+    reportError(createError("No .llvm.callgraph section found."),
+                "Missing section");


https://llvm.org/docs/CodingStandards.html#error-and-warning-messages

Same applies to other error/warning messages.

That being said, we don't tend to emit an error for other options that do similar things, if the respective section doesn't exist.

When a diagnostic message is appropriate, we usually prefer warnings, rather than errors, so that we can continue dumping other requested items, or we emit a warning then try to dump the information we can for the requested section and use placeholder values (e.g. "<?>").

Done. PTAL.

jh7370 · 2025-11-19T10:15:58Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+
+  Expected<ArrayRef<uint8_t>> SectionBytesOrErr =
+      Obj.getSectionContents(*CGSection);
+  if (!SectionBytesOrErr) {


https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

Applies in several other places too.

I think I have addressed all the cases. PTAL.

jh7370 · 2025-11-19T10:22:54Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+                    .str());
+  };
+
+  DataExtractor Data(SectionBytesOrErr.get(), Obj.isLE(),


Have you looked into using the DataExtractor::Cursor class? I think that would simplify some of your parsing logic below.

jh7370 · 2025-11-19T10:25:08Z

llvm/tools/llvm-readobj/ELFDumper.cpp

+      if (Name == CallGraphSectionName)
+        return Sec;
+    } else
+      consumeError(NameOrErr.takeError());


consumeError is generally a code smell.

Can an actual error be emitted here? If so, it should either be propagated or reported as a warning. If it can't ever fail here, cantFail is more appropriate.

This function was not used and should've been removed in the first place. Thank you.

…f the output which is not useful.

github-actions · 2025-11-21T03:11:01Z

🐧 Linux x64 Test Results

186436 tests passed
4868 tests skipped

Prabhuk mentioned this pull request Sep 9, 2025

[llvm-objdump] Handle .callgraph section #151009

Closed

Prabhuk added 13 commits September 17, 2025 18:59

[llvm-readobj] Dump callgraph section info for ELF

ac4b7bc

Introduce a new flag `--call-graph-info` which outputs callgraph ELF section information to the console as a text output or as JSON output.

Format code

31c4d2b

Implement JSON output for printCallGraph()

4d1a7c7

ELFDumper cleanups

f0810b2

Add readelf callgraph info tests.

5f20b7b

more tests

d991194

Extend tests to json

c2d8d22

Expand all tests to include JSON and LLVM formats.

2874918

Improve tests. Match full lines.

9803c62

Make call graph processing agnostic of target architecture pointer wi…

7285210

…dth size.

Remove unnecessary function returns after reportError calls which bot…

8236798

…toms out to llvm_unreachable.

Formatting fixes.

abe189e

Remove callsites from direct call data.

0b9c46d

Prabhuk force-pushed the readelf_callgraph_info branch from 3499c2d to 0b9c46d Compare September 18, 2025 03:09

Prabhuk added 12 commits September 17, 2025 20:10

Format

5b504fc

Simplify readelf output.

5b6bb00

format

937ff61

Clean up callgraph output

1b86857

Format

7280ff3

Show symbol names.

bdb8683

Format

09d3a87

Make llvm output better.

b851b8d

format source.

fd50ff0

Minor improvements to LLVM output.

ee3a081

Fix tests to match the new output.

3c94a44

Add REQUIRES: x86-registered-target to tests.

026b184

Prabhuk requested review from frobtech and ilovepi September 19, 2025 00:12

Prabhuk requested a review from jh7370 October 23, 2025 18:05

Update error message.

c429dd8

ilovepi reviewed Oct 24, 2025

View reviewed changes

Prabhuk added 3 commits October 31, 2025 13:31

Address review comments.

d04fbd0

Function name printing to match stack size printing behavior.

75bffeb

Refactor getCalGraphSection function as suggested in review.

77acb10

ilovepi reviewed Oct 31, 2025

View reviewed changes

jh7370 reviewed Nov 3, 2025

View reviewed changes

Prabhuk added 4 commits November 4, 2025 11:36

Declutter ouput format. Remove GNU output.

246275d

Remove unnecessary header includes.

402c90f

More header include cleanups.

87d3dd8

Rename test files to .S

20ea016

ilovepi approved these changes Nov 4, 2025

View reviewed changes

Prabhuk added 8 commits November 4, 2025 16:21

Fix relocation printing handling.

c5823aa

Remove addend, type and offset info from relocations.

8d3a27c

Fix tests.

aaa2cd4

Fix tests.

afc13ee

Rename files to .s

6e22254

Make relocatable test to be built from YAML.

5b3bff0

Relocatable file tests fixed.

c4ea41d

Nits. Alignment within test files.

53ffcaa

ilovepi reviewed Nov 5, 2025

View reviewed changes

Prabhuk added 2 commits November 5, 2025 11:00

Merge remote-tracking branch 'upstream/main' into readelf_callgraph_info

edc5eaa

Merge branch 'main' into readelf_callgraph_info

14b5cf0

jh7370 reviewed Nov 19, 2025

View reviewed changes

Prabhuk added 3 commits November 20, 2025 17:19

Warn for unknown relocations instead of printing the offset as part o…

b56a17c

…f the output which is not useful.

Address review comments.

ae501fa

Fix messages. Remove unnecessary errors and unused functions.

f454957

		@@ -0,0 +1,150 @@
		## Tests --call-graph-info prints information from call graph section.

		# REQUIRES: x86-registered-target

		# RUN: llvm-readelf --elf-output-style=LLVM --call-graph-info %t 2>&1 \| FileCheck %s -DFILE=%t --check-prefix=LLVM
		# RUN: llvm-readelf --elf-output-style=JSON --pretty-print --call-graph-info %t 2>&1 \| FileCheck %s -DFILE=%t --check-prefix=JSON



		## We do not support GNU format console output for --call-graph-info as it is an LLVM only info.
		# CHECK-NOT: .

	auto IndirectTypeIdsList = SmallVector<uint64_t, 4>(
	SmallVector<uint64_t, 4> IndirectTypeIdsList(

		reportError(createError("No .llvm.callgraph section found."),
		"Missing section");

[llvm-readobj] Dump callgraph section info for ELF #157499

Are you sure you want to change the base?

[llvm-readobj] Dump callgraph section info for ELF #157499

Conversation

Prabhuk commented Sep 8, 2025

Uh oh!

github-actions bot commented Sep 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jh7370 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ilovepi left a comment

Choose a reason for hiding this comment

Uh oh!

Prabhuk commented Nov 4, 2025

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jh7370 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Nov 21, 2025

🐧 Linux x64 Test Results

Uh oh!

github-actions bot commented Sep 8, 2025 •

edited

Loading