[Vm-dev] Segfaulting cog vm on FreeBSD

[Petr Fischer]

Next findings. My first test crashed in function "scavengeReferentsOf" on this line:
...
if (((longAt(referent)) & ((classIndexMask()) - (isForwardedObjectClassIndexPun()))) == 0) {
...
http://pastebin.com/yK7YJig2 (line 110)

My second test is crashing (segfault) in function "eliminateAndFreeForwardersForPigCompact" on this line:
...
if (((f & (tagMask())) == 0)  
     && (((longAt(f)) & (classIndexMask())) == (isForwardedObjectClassIndexPun()))) {
...
http://pastebin.com/CaYzLPkZ (line  243)

The core C line/condition is very similar (there is even more similar lines/conditions in src/vm/gcc3x-cointerp.c source).

Is it more helpful?                                                                                                                                                                                                 

pf


>  
> Crashing on the same line as before: http://pastebin.com/yK7YJig2 line 110.
> 
> Backtrace:
> http://pastebin.com/r5bgn6f4
> 
> printAllStacks() - this is different than before, no UUID generator in stacktrace, crashed somewhere in Fuel serializer
> http://pastebin.com/VDJnFtFp
> 
> Now what? How to debug this? Add some printf outputs to Core C sources?
> 
> 
> >  
> > Look for the method that calls #primitiveMakeUUID. Probably that will be
> > UUID>>primMakeUUID. Then comment out this line:
> > 
> > 	"<primitive: 'primitiveMakeUUID' module: 'UUIDPlugin'>"
> > 
> > The will prevent calls to the plugin, effectively disabling it.
> > 
> > When you are done testing, revert your #primMakeUUID method back to the
> > way it was before.
> > 
> > Dave
> > 
> > 
> > >
> > > OK - how to  "temporarily remove UUIDPlugin"? Thanks, pf
> > >
> > >
> > >>
> > >> That's all fine. What I was trying to hint that maybe the UUID plugin is
> > >> broken (it is definitely not generating UUIDs, since the fallback code
> > >> is executed) and it messes up the object memory, which results in a
> > >> crash in the GC. That's why I asked if it crashes with the plugin
> > >> removed.
> > >>
> > >> Levente
> > >>
> > >> On Mon, 18 Jul 2016, Petr Fischer wrote:
> > >>
> > >> >
> > >> > I am generating/instantiating thousand of objects (my simple data
> > >> classes), and then persisting this medium sized object graph with Fuel
> > >> to file.
> > >> > While I am preparing object graph, UUID generator is used for
> > >> generating random IDs and is called thousands times.
> > >> > Final crash is in "scavengeReferentsOf" (via gdb) - probably Garbage
> > >> Collector.
> > >> >
> > >> > pf
> > >> >
> > >> >>
> > >> >> The image was generating an UUID when the crash happened. But
> > >> primMakeUUID
> > >> >> seem to had failed and the crash happened in the fallback code.
> > >> >> Does the VM crash if you temporarily remove UUIDPlugin?
> > >> >>
> > >> >> Levente
> > >> >>
> > >> >> On Sun, 17 Jul 2016, Petr Fischer wrote:
> > >> >>
> > >> >>>
> > >> >>> Hello! Moving discussion about %subj% from pharo-dev list.
> > >> >>> My problem on FreeBSD:
> > >> >>>
> > >> >>> There is a linux compatibility layer on FreeBSD (even 64bits on
> > >> 11-CURRENT), so there is no theoretical problem running pharo linux
> > >> vms.
> > >> >>> But: complete linux libraries comming from centos 6.7, where is
> > >> glibc older than 2.15 - so, no luck with actual default pharo vm
> > >> from pharo.org.
> > >> >>> So I compiled my own pharo-vm on CentOS 6.7 with old glibc,
> > >> everything is OK on Linux CentOS box, but on the FreeBSD side, vm is
> > >> segfaulting (my own compiled "stack" vm is more stable, but
> > >> segfaulting too).
> > >> >>> I compiled from pharo-vm git:
> > >> >>> https://github.com/pharo-project/pharo-vm
> > >> >>>
> > >> >>> I am building "PharoVMSpur32Builder buildUnix32" (there is some old
> > >> "PharoVMBuilder buildFreeBSD", but probably unmaintained).
> > >> >>>
> > >> >>> Why I can't compile directly on the FreeBSD without Linux
> > >> compatibility layer? Because there is a step, when I need to Load
> > >> VMMaker into functional stable image - when I use my compiled vm, of
> > >> course, segfaulting while loading VMMaker... Therefore I am going
> > >> via Linux comp. layer. Native port of vm for FreeBSD will be the
> > >> next step.
> > >> >>>
> > >> >>> Ok, ok, next steps... I compiled my own "debug vm" (cog, spur, again
> > >> on CentOS 6.7) and here is some outputs from my gdb session (I am
> > >> definitely not Core C hacker):
> > >> >>>
> > >> >>> Crashing code snippet - function is "scavengeReferentsOf" and
> > >> crashing line is 110:
> > >> >>> http://pastebin.com/yK7YJig2
> > >> >>>
> > >> >>> Backtrace:
> > >> >>> http://pastebin.com/jMgzE9G0
> > >> >>>
> > >> >>> printAllStacks:
> > >> >>> http://pastebin.com/VP1nwrsC
> > >> >>>
> > >> >>> Now I am at end with my knowledge (vm internals) - but I can add
> > >> some printf/log outputs into generated C files, recompile and
> > >> retest.
> > >> >>>
> > >> >>> Any ideas? Thanks very much. pf
> > >> >>>
> > >> >
> > >
> > 
> >