[FATAL src/ReplaySession.cc:1181:check_ticks_consistency()]

[tysmith]

[FATAL src/ReplaySession.cc:1181:check_ticks_consistency()]

[khuey]

What CPU did this happen on?

If an AMD system, does it reproduce on Intel hardware?

[tysmith]

I am using an Intel i7-8650U.

[khuey]

Can you get a recording with the --intel-pt option that reproduces the replay failure?

https://github.com/rr-debugger/rr/wiki/Debugging-divergence#intel-pt

[tysmith]

Running with --intel-pt leads to:

[FATAL src/PerfCounters.cc:768:open() errno: EPERM] Can't allocate memory for PT AUX area

How do I "Make sure that the "max locked memory" limit is very high..."? Could that be the issue with --intel-pt?

[khuey]

You need to set ulimit -l really high (e.g. ulimit -l unlimited, though that may consume all your RAM).

[tysmith]

I can't figure out how to do what is needed on Ubuntu.

ulimit -l unlimited returns
bash: ulimit: max locked memory: cannot modify limit: Operation not permitted

I tried modifying /etc/security/limits.conf and rebooting and that doesn't work either.

[tysmith]

Got it. Typo in my /etc/security/limits.conf entry.

[tysmith]

Same result: [FATAL src/PerfCounters.cc:768:open() errno: EPERM] Can't allocate memory for PT AUX area but it uses a lot more memory first.

[rocallahan]

What happens if you run rr as root?

…

On Thu, 14 Mar 2024, 2:44 pm Tyson Smith, ***@***.***> wrote: Same result: [FATAL src/PerfCounters.cc:768:open() errno: EPERM] Can't allocate memory for PT AUX area but it uses a lot more memory first. — Reply to this email directly, view it on GitHub <#3711 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACIJR6GCHIKWJUAGVH4YE3YYD6ILAVCNFSM6AAAAABEU672VSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOJWGI2DAOJVGU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[tysmith]

Looks like Firefox does not support it.
Running Nightly as root in a regular user's session is not supported

I also tried --headless to hopefully same some memory but it still fails to launch,

[rocallahan]

What if it's something simple like 'ls'?

…

On Thu, 14 Mar 2024, 3:40 pm Tyson Smith, ***@***.***> wrote: Looks like Firefox does not support it. Running Nightly as root in a regular user's session is not supported I also tried --headless to hopefully same some memory but it still fails to launch, — Reply to this email directly, view it on GitHub <#3711 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACIJR4EK2CGUHJHNDYPKP3YYEEZXAVCNFSM6AAAAABEU672VSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOJWGI4DSMJRGU> . You are receiving this because you commented.Message ID: ***@***.***>

[tysmith]

I disabled a few extra browser processes and managed to get it to launch. After using up most of the available memory and hanging for a long period I got this:

[FATAL src/PerfCounters.cc:811:flush()] Unexpected AUX packet flags 1

[tysmith]

And running rr record --intel-pt ls works quickly as expected.

[khuey]

that's PERF_AUX_FLAG_TRUNCATED

[khuey]

Can you share the steps to reproduce (privately, if necessary)?

[tysmith]

I am unable to reproduce the issue independent of the tools we use in automation.

STR:

• download build https://firefox-ci-tc.services.mozilla.com/tasks/index/gecko.v2.mozilla-central.revision.4f7a5399c1ccd3d580631d65356e681ad7e28d74.firefox/linux64-fuzzing-noopt-debug
• run python3 -m pip install site-scout
• run python3 -m site_scout <firefox> -u http://genral.fund --pernosco -o . --time-limit 300

It can take a few minutes for the crash. The rr command is actually created in ffpuppet[1] so if that needs to change you can either modify the version installed by pip or pull down a copy from GitHub, modify that and do a local install of ffpuppet. Let me know if that works.

[1] https://github.com/MozillaSecurity/ffpuppet/blob/master/src/ffpuppet/core.py#L479

[khuey]

Is that supposed to be http://general.fund instead of http://genral.fund?

[khuey]

Assuming it's supposed to be http://general.fund, the bad news is that I can't reproduce the rr issue. The good news is that I can reproduce the Firefox issue. Do you want me to upload it to Pernosco?

[tysmith]

Yep sorry that was a typo. I was able to capture it without the tools this morning and submit to Pernosco already, thanks.

[khuey]

What rr rev are you recording with?

[tysmith]

2bb38b9

[khuey]

Ok, that's what I tried. I can't reproduce an rr divergence here unfortunately.

How much RAM does your machine have?

[tysmith]

16GB

[khuey]

You could try reducing PT_PERF_AUX_SIZE in src/PerfCounters.cc

[tysmith]

OK I set PT_PERF_AUX_SIZE to 32*1024*1024 and now I get the following error: [FATAL src/PerfCounters.cc:811:flush()] Unexpected AUX packet flags 1

[khuey]

Ok that's the same issue as before unfortunately.

[tysmith]

I just rebuilt with clang 14 and can't reproduce.

[khuey]

rebuilt rr?

[tysmith]

Yes

[khuey]

Alright, if even you can't reproduce it there's nothing actionable left here.