Smalltalk › Squeak › Squeak VM

Segfaults from #copyStack

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

Max Leske

Segfaults from #copyStack

Hi,

I'm seeing many segfaults (around 20 to 30% of the time) when running Fuel builds on Linux. Everytime I see it it's caused by #copyStack (e.g. https://travis-ci.org/theseion/Fuel/jobs/345741310). AFAICT it happens more often (exclusively?) on 32 bit with Pharo 7 (latest VM + latest image).

Any ideas?

Cheers,

Max

Eliot Miranda-2

Re: Segfaults from #copyStack

Hi Max,

On Sat, Feb 24, 2018 at 1:07 PM, Max Leske <[hidden email]> wrote:

Hi,

I'm seeing many segfaults (around 20 to 30% of the time) when running Fuel builds on Linux. Everytime I see it it's caused by #copyStack (e.g. https://travis-ci.org/theseion/Fuel/jobs/345741310). AFAICT it happens more often (exclusively?) on 32 bit with Pharo 7 (latest VM + latest image).

Try and construct a test case that crashes without user interaction. i.e. something that starts up with a known image, supplying a script to load via Fuel. Once this crashes reliably debugging is easy. Otherwise its very hard. As soon as you have the crashing test case let me know and I'll take a look. Ideally it would run on Mac OS X.

_,,,^..^,,,_

best, Eliot

Max Leske

Re: Segfaults from #copyStack

In reply to this post by Max Leske

Thanks Eliot.

On 24 February 2018 at 22:15:21, [hidden email] ([hidden email]) wrote:

Hi Max,

<a href="http://airmail.calendar/2018-02-24 13:07:00 CET" style="font-family:helvetica;font-size:13px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">On Sat, Feb 24, 2018 at 1:07 PM, Max Leske <[hidden email]> wrote:

>
> Hi,
>
> I'm seeing many segfaults (around 20 to 30% of the time) when running Fuel
> builds on Linux. Everytime I see it it's caused by #copyStack (e.g.
> https://travis-ci.org/theseion/Fuel/jobs/345741310). AFAICT it happens
> more often (exclusively?) on 32 bit with Pharo 7 (latest VM + latest image).
>

Try and construct a test case that crashes without user interaction. i.e.
something that starts up with a known image, supplying a script to load via
Fuel. Once this crashes reliably debugging is easy. Otherwise its very
hard. As soon as you have the crashing test case let me know and I'll take
a look. Ideally it would run on Mac OS X.

I've tried for the last two hours to come up with a reliable way of triggering the segfault but haven't had any luck. I can only reproduce it on Linux, not macOS and I've only been able to force the segfault with using smalltalkCI:

git clone https://github.com/hpi-swa/smalltalkCI.git && git clone https://github.com/theseion/Fuel.git && smalltalkCI/run.sh -s Pharo-alpha Fuel/.default.ston

I used Ubuntu 17.04 (32-bit), TravisCI uses 14.04.5 (64-bit).

Maybe there's a way I can extract some information you can use? And then I'll just play mole whacking until I trigger it... :)

Cheers,

Max

Eliot Miranda-2

Re: Segfaults from #copyStack

Hi Max,

On Sat, Feb 24, 2018 at 2:57 PM, Max Leske <[hidden email]> wrote:

Thanks Eliot.

On 24 February 2018 at 22:15:21, [hidden email] ([hidden email]) wrote:

Hi Max,

On Sat, Feb 24, 2018 at 1:07 PM, Max Leske <[hidden email]> wrote:

>
> Hi,
>
> I'm seeing many segfaults (around 20 to 30% of the time) when running Fuel
> builds on Linux. Everytime I see it it's caused by #copyStack (e.g.
> https://travis-ci.org/theseion/Fuel/jobs/345741310). AFAICT it happens
> more often (exclusively?) on 32 bit with Pharo 7 (latest VM + latest image).
>

Try and construct a test case that crashes without user interaction. i.e.
something that starts up with a known image, supplying a script to load via
Fuel. Once this crashes reliably debugging is easy. Otherwise its very
hard. As soon as you have the crashing test case let me know and I'll take
a look. Ideally it would run on Mac OS X.

I've tried for the last two hours to come up with a reliable way of triggering the segfault but haven't had any luck. I can only reproduce it on Linux, not macOS and I've only been able to force the segfault with using smalltalkCI:

git clone https://github.com/hpi-swa/smalltalkCI.git && git clone https://github.com/theseion/Fuel.git && smalltalkCI/run.sh -s Pharo-alpha Fuel/.default.ston

I used Ubuntu 17.04 (32-bit), TravisCI uses 14.04.5 (64-bit).

Maybe there's a way I can extract some information you can use? And then I'll just play mole whacking until I trigger it... :)

There should at least be a crash.dmp file that should contain a reasonably informative stack trace and register set. What does that show you? When you combine the crash.mp output with a gdb session you can find the offending instruction and relate that to the register set in crash.mp to find out the instruction that caused the crash. Tracing back form that isn't necessarily trivial but at least one has started :-)

Cheers,
Max

_,,,^..^,,,_

best, Eliot