Add initial support for i386 builds #2347

jonathanmetzman · 2019-04-25T17:19:16Z

Add initial support for i386 (ie: 32 bit/x86) builds.
Build + install i386 versions of clang libraries.
Build + install an i386 version of libc++.
Create a new build option ARCHITECTURE (that defaults to x86_64).
Use the correct CFLAGS and CXXFLAGS (to compile for i386 and use the right libc++) when ARCHITECTURE is i386.

jonathanmetzman · 2019-04-25T17:23:37Z

I've only tested this with skcms so far I need to do more.
This is currently blocked on my LLVM patch adding i386 support to libFuzzer.

Once that patch lands it is unclear if we will simply roll clang forward early or if we will need to wait for Chromium's clang to be updated.
We could use libFuzzingEngine.a in $LIB_FUZZING_ENGINE as a temporary workaround that unblocks this patch.

jonathanmetzman · 2019-04-25T18:17:43Z

The runner needs a 32-bit compatible version of ld-linux.so and other basic libraries (including libgcc for some strange reason?) for this to work. That can be done by installing libc6-dev-i386 on the runner.

jonathanmetzman · 2019-04-25T20:06:33Z

The runner needs a 32-bit compatible version of ld-linux.so and other basic libraries (including libgcc for some strange reason?) for this to work. That can be done by installing libc6-dev-i386 on the runner.

I do this now by installing libc6-dev-i386 to the runner (as well as the runner). With dependencies it adds about 50 MB to the image size. It pulls in a lot that is potentially not needed on the runner so reducing that might be worth exploring.

evverx · 2019-04-26T06:11:59Z

@jonathanmetzman I think it's a great idea but having opened google/sanitizers#778 (which was closed as a duplicate of google/sanitizers#628) I was under the impression that i386 wasn't the most popular platform among the developers of sanitizers. I'm wondering if this has changed.

evverx · 2019-04-26T06:30:29Z

What I was trying to say is that it would be great if there was a way to turn it off completely when (if?) it's merged because after I peppered the code with kludges to get around MSan false positives nobody is willing to fix I'd wait for someone else to test how stable sanitizers on i386 are.

jonathanmetzman · 2019-04-29T16:06:00Z

I was under the impression that i386 wasn't the most popular platform among the developers of sanitizers. I'm wondering if this has changed.

It isn't the most popular but we've had Chrome's libFuzzer targets running on i386 for months now without any issues. I don't know enough to say it will work but I think it will. CC @kcc @morehouse
if they have any comments.

What I was trying to say is that it would be great if there was a way to turn it off completely when (if?) it's merged

There will be an easy way to turn it off. Initially it will be off by default, projects will need to opt-in to it by adding something to project.yaml.

I peppered the code with kludges to get around MSan false positives nobody is willing to fix I'd wait for someone else to test how stable sanitizers on i386 are.

MSAN is totally unsupported on i386. We only plan on using ASAN for now.

morehouse · 2019-04-29T16:30:38Z

I was under the impression that i386 wasn't the most popular platform among the developers of sanitizers. I'm wondering if this has changed.

We're not opposed to sanitizers on 32-bit, we just don't have the resources to support them ourselves. We have similar stories for sanitizers on Mac, Windows, etc., but many of those platforms have other interested parties that support them (e.g., Apple).

evverx · 2019-04-30T11:00:00Z

I thought you were going to roll it out globally. Looks like it doesn't seem to be the case and I could wait for more issues found by ASan on i386 to be opened to the public (or as a last resort to experiment with it locally) before turning it on. Sorry for the noise.

I also went through issues at https://bugs.chromium.org/p/chromium/issues/list?can=1&q=description%3Ax86_libfuzzer_chrome_asan%2Cx86_libfuzzer_chrome_asan_debug and I'm wondering what is going to happen when ASan finds a memory leak for example on i386 and x86_64 with AFL and libFuzzer. Is my understanding correct that it will open 3 different issues?

jonathanmetzman · 2019-04-30T13:38:04Z

I thought you were going to roll it out globally. Looks like it doesn't seem to be the case and I could wait for more issues found by ASan on i386 to be opened to the public (or as a last resort to experiment with it locally) before turning it on. Sorry for the noise.

No problem. This is a fine approach to take.

I also went through issues at https://bugs.chromium.org/p/chromium/issues/list?can=1&q=description%3Ax86_libfuzzer_chrome_asan%2Cx86_libfuzzer_chrome_asan_debug

Good sleuthing!

and I'm wondering what is going to happen when ASan finds a memory leak for example on i386 and x86_64 with AFL and libFuzzer. Is my understanding correct that it will open 3 different issues?

No. It will be deduplicated and only one issue should be filed (this will make it harder to determine which bugs are x86 only but iirc there are definitely some clear cases of this). Also we will only use libFuzzer on x86 (AFL doesn't use LSAN btw).

jonathanmetzman · 2019-04-30T20:59:07Z

Need to roll clang past r359802 to support this.

jonathanmetzman · 2019-05-13T20:08:42Z

Chrome rolled to 360094 so this can land even if my OSS-Fuzz clang roll doesn't stick.

jonathanmetzman · 2019-05-13T20:22:15Z

This PR won't contain any documentation changes as I want to land this change before I fully support its use by OSS-Fuzz users.

jonathanmetzman · 2019-05-13T21:06:31Z

@oliverchang @inferno-chromium PTAL

oliverchang

LGTM. this will need changes to CF images as well for 32-bit libs right?

infra/travis/travis_build.py

jonathanmetzman · 2019-05-13T21:18:08Z

LGTM. this will need changes to CF images as well for 32-bit libs right?

Yep. Will change that with my CF-repo changes.

infra/base-images/base-builder/Dockerfile

inferno-chromium · 2019-05-13T21:12:28Z

infra/base-images/base-builder/compile

@@ -22,6 +22,10 @@ if [ -z "${SANITIZER_FLAGS-}" ]; then
  export SANITIZER_FLAGS=${!FLAGS_VAR-}
 fi

+if [[ $ARCHITECTURE == "i386" ]]; then
+    export CFLAGS="-m32 $CFLAGS"


can you make this a CFLAGS_EXTRA and then add support for that similar to
infra/base-images/base-clang/Dockerfile:ENV CXXFLAGS "$CFLAGS $CXXFLAGS_EXTRA"

Why? as far as I can tell CXXFLAGS_EXTRA is to have "extra" flags to append to CFLAGS to get CXXFLAGS instead of having to repeat the flags. I don't think I understand the benefits having CFLAGS_EXTRA provides.

ah my bad, this is flags itself, so appending to CFLAGS directly is fine.

infra/helper.py

infra/travis/travis_build.py

infra/helper.py

inferno-chromium · 2019-05-13T21:45:47Z

--architecture is fine.

jonathanmetzman · 2019-05-13T21:48:31Z

--architecture is fine.

Thanks. I'm not opposed to arch, we can discuss before we make this feature supported.

jonathanmetzman changed the title ~~DONT MERGE: Initial support for (i386) builds~~ DONT MERGE: Initial support for i386 builds Apr 25, 2019

jonathanmetzman force-pushed the x86-2 branch from cbb1feb to 93ac13f Compare April 25, 2019 20:04

jonathanmetzman mentioned this pull request May 13, 2019

[infra] Roll clang to r360398 #2405

Merged

jonathanmetzman closed this May 13, 2019

jonathanmetzman reopened this May 13, 2019

jonathanmetzman added 6 commits May 13, 2019 13:18

initial

1b296a8

install g++-multilib

fb5c4d8

add todo

22a54ff

working

60ed58b

remove unneeded newline

f539508

add arch arg to command line, fix problems building and running

677fa96

jonathanmetzman force-pushed the x86-2 branch from 93ac13f to 677fa96 Compare May 13, 2019 20:19

jonathanmetzman changed the title ~~DONT MERGE: Initial support for i386 builds~~ Add initial support for i386 builds May 13, 2019

jonathanmetzman added 5 commits May 13, 2019 13:28

Make travis_build.py recognize different architectures.

e9322c8

fix

047fb20

lint

0d88b3f

make i386 a choice

d4a9d1e

improve travis script

0f8bee5

jonathanmetzman requested review from oliverchang and inferno-chromium May 13, 2019 21:06

oliverchang approved these changes May 13, 2019

View reviewed changes

infra/travis/travis_build.py Outdated Show resolved Hide resolved

inferno-chromium reviewed May 13, 2019

View reviewed changes

add architecture to check_build and fix nits

68b35d5

inferno-chromium approved these changes May 13, 2019

View reviewed changes

jonathanmetzman merged commit c60fcd7 into master May 13, 2019

jonathanmetzman deleted the x86-2 branch May 13, 2019 22:01

jonathanmetzman mentioned this pull request May 13, 2019

Running fuzzers on non-x64 systems #2295

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add initial support for i386 builds #2347

Add initial support for i386 builds #2347

jonathanmetzman commented Apr 25, 2019

jonathanmetzman commented Apr 25, 2019 •

edited

Loading

jonathanmetzman commented Apr 25, 2019

jonathanmetzman commented Apr 25, 2019

evverx commented Apr 26, 2019

evverx commented Apr 26, 2019

jonathanmetzman commented Apr 29, 2019

morehouse commented Apr 29, 2019

evverx commented Apr 30, 2019

jonathanmetzman commented Apr 30, 2019

jonathanmetzman commented Apr 30, 2019 •

edited

Loading

jonathanmetzman commented May 13, 2019 •

edited

Loading

jonathanmetzman commented May 13, 2019

jonathanmetzman commented May 13, 2019

oliverchang left a comment

jonathanmetzman commented May 13, 2019

inferno-chromium May 13, 2019

jonathanmetzman May 13, 2019 •

edited

Loading

inferno-chromium May 13, 2019

inferno-chromium commented May 13, 2019

jonathanmetzman commented May 13, 2019

Add initial support for i386 builds #2347

Add initial support for i386 builds #2347

Conversation

jonathanmetzman commented Apr 25, 2019

jonathanmetzman commented Apr 25, 2019 • edited Loading

jonathanmetzman commented Apr 25, 2019

jonathanmetzman commented Apr 25, 2019

evverx commented Apr 26, 2019

evverx commented Apr 26, 2019

jonathanmetzman commented Apr 29, 2019

morehouse commented Apr 29, 2019

evverx commented Apr 30, 2019

jonathanmetzman commented Apr 30, 2019

jonathanmetzman commented Apr 30, 2019 • edited Loading

jonathanmetzman commented May 13, 2019 • edited Loading

jonathanmetzman commented May 13, 2019

jonathanmetzman commented May 13, 2019

oliverchang left a comment

Choose a reason for hiding this comment

jonathanmetzman commented May 13, 2019

inferno-chromium May 13, 2019

Choose a reason for hiding this comment

jonathanmetzman May 13, 2019 • edited Loading

Choose a reason for hiding this comment

inferno-chromium May 13, 2019

Choose a reason for hiding this comment

inferno-chromium commented May 13, 2019

jonathanmetzman commented May 13, 2019

jonathanmetzman commented Apr 25, 2019 •

edited

Loading

jonathanmetzman commented Apr 30, 2019 •

edited

Loading

jonathanmetzman commented May 13, 2019 •

edited

Loading

jonathanmetzman May 13, 2019 •

edited

Loading