Smalltalk › Squeak › Squeak VM

Problems involving cogvm and Unix OSProcessPlugin OSProcess PipeableOSProcess

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

CdAB63

Problems involving cogvm and Unix OSProcessPlugin OSProcess PipeableOSProcess

Here's the log:

Untrustedir /home/casimiro/Desenvolvimento/cog.work/secure
UnixOSPrReceiversessionIdentvaria DependentsArray(a UnixProcess with pid
3490)
fier: #[202 58grimReaper: siona Process in [] in [] in
UnixOSProcessAccessor>>
grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
in Uni
UnixOSPrReceiversessionIdentifier: epend#[202 58 13 UnixProcess with pid
3490)
6] grimReaper: sio a Process in [] in [] in UnixOSProcessAccessor>>
grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
in Uni
UnixOSPrReceivercanObtainSessionIdenpend#[202 58 13 86]xProcess with pid
3490)
ifierFromPlugin:grimReaper: a Process in [] in [] in
UnixOSProcessAccessor>>
grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
in Uni
UnixOSPrReceiversessionIdentifier: epend#[2sArray(a UnixProcess with pid
3490)
2 58 13 86] grimReaper: siona Process in [] in [] in
UnixOSProcessAccessor>>
grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
in Uni
UnixProcess>>redirect:to:- - - - - - - - - - - - - - - - - - - - -
[] in PipeableOSProcess
class>>forkSqueakAndDoThenQuit:errorPipelineStream:headl
Compiler>>evaluate:in:to:notifyiningTo:toPrecision:maxEpochs:reportingEach:ss:
PluggableTextMorphPlus(PluggableTextMorph)>>handleEdit:
MorphicEventh(Morph)>>handleEvent:nt:ndTo:
[] in HandMhandleFocusEvent:nt:using::with:h:
...etc...>>handleEvent:nt::clear:r:ing:

BTW, XDisplayControlPlugin explodes cogvm at my face... but seems to be
required by TranscriptStream (trying to figure out what's going on
there... then I'll report the problem properly).

signature.asc (269 bytes) Download Attachment

David T. Lewis

Re: Problems involving cogvm and Unix OSProcessPlugin OSProcess PipeableOSProcess

I think that you have confirmed that #forkSqueak does not work on a Cog VM.
The likely reason for this is that #forkSqueak involves temporarily disabling
timer signals from the operating system, then reenabling them in both the
child and parent after the fork(). The Cog VM is using a different underlying
timer mechanism (pthreads etc) so the current #forkSqueak mechanism would
probably need to be redone for Cog. I have not looked in to this at all,
but I did expect that this might be a problem and I think you have confirmed
it in the log below.

Dave

On Fri, Sep 03, 2010 at 09:52:29AM -0300, Casimiro de Almeida Barreto wrote:

>
> Here's the log:
>
> Untrustedir /home/casimiro/Desenvolvimento/cog.work/secure
> UnixOSPrReceiversessionIdentvaria DependentsArray(a UnixProcess with pid
> 3490)
> fier: #[202 58grimReaper: siona Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceiversessionIdentifier: epend#[202 58 13 UnixProcess with pid
> 3490)
> 6] grimReaper: sio a Process in [] in [] in UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceivercanObtainSessionIdenpend#[202 58 13 86]xProcess with pid
> 3490)
> ifierFromPlugin:grimReaper: a Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceiversessionIdentifier: epend#[2sArray(a UnixProcess with pid
> 3490)
> 2 58 13 86] grimReaper: siona Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixProcess>>redirect:to:- - - - - - - - - - - - - - - - - - - - -
> [] in PipeableOSProcess
> class>>forkSqueakAndDoThenQuit:errorPipelineStream:headl
> Compiler>>evaluate:in:to:notifyiningTo:toPrecision:maxEpochs:reportingEach:ss:
> PluggableTextMorphPlus(PluggableTextMorph)>>handleEdit:
> MorphicEventh(Morph)>>handleEvent:nt:ndTo:
> [] in HandMhandleFocusEvent:nt:using::with:h:
> ...etc...>>handleEvent:nt::clear:r:ing:
>
>
> BTW, XDisplayControlPlugin explodes cogvm at my face... but seems to be
> required by TranscriptStream (trying to figure out what's going on
> there... then I'll report the problem properly).
>

Eliot Miranda-2

Re: Problems involving cogvm and Unix OSProcessPlugin OSProcess PipeableOSProcess

Hi David,

what's the reason for shutting down interval timers around fork? What was the issue that necessitated this in the first place? We're likely getting bit by this and I want to try and keep the solution as simple as possible.

best

Eliot

On Sat, Sep 4, 2010 at 7:20 AM, David T. Lewis <[hidden email]> wrote:

I think that you have confirmed that #forkSqueak does not work on a Cog VM.
The likely reason for this is that #forkSqueak involves temporarily disabling
timer signals from the operating system, then reenabling them in both the
child and parent after the fork(). The Cog VM is using a different underlying
timer mechanism (pthreads etc) so the current #forkSqueak mechanism would
probably need to be redone for Cog. I have not looked in to this at all,
but I did expect that this might be a problem and I think you have confirmed
it in the log below.

Dave

On Fri, Sep 03, 2010 at 09:52:29AM -0300, Casimiro de Almeida Barreto wrote:
>
> Here's the log:
>
> Untrustedir /home/casimiro/Desenvolvimento/cog.work/secure
> UnixOSPrReceiversessionIdentvaria DependentsArray(a UnixProcess with pid
> 3490)
> fier: #[202 58grimReaper: siona Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceiversessionIdentifier: epend#[202 58 13 UnixProcess with pid
> 3490)
> 6] grimReaper: sio a Process in [] in [] in UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceivercanObtainSessionIdenpend#[202 58 13 86]xProcess with pid
> 3490)
> ifierFromPlugin:grimReaper: a Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixOSPrReceiversessionIdentifier: epend#[2sArray(a UnixProcess with pid
> 3490)
> 2 58 13 86] grimReaper: siona Process in [] in [] in
> UnixOSProcessAccessor>>
> grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> in Uni
> UnixProcess>>redirect:to:- - - - - - - - - - - - - - - - - - - - -
> [] in PipeableOSProcess
> class>>forkSqueakAndDoThenQuit:errorPipelineStream:headl
> Compiler>>evaluate:in:to:notifyiningTo:toPrecision:maxEpochs:reportingEach:ss:
> PluggableTextMorphPlus(PluggableTextMorph)>>handleEdit:
> MorphicEventh(Morph)>>handleEvent:nt:ndTo:
> [] in HandMhandleFocusEvent:nt:using::with:h:
> ...etc...>>handleEvent:nt::clear:r:ing:
>
>
> BTW, XDisplayControlPlugin explodes cogvm at my face... but seems to be
> required by TranscriptStream (trying to figure out what's going on
> there... then I'll report the problem properly).
>

David T. Lewis

Re: Problems involving cogvm and Unix OSProcessPlugin OSProcess PipeableOSProcess

Hi Eliot,

The child process needed to establish signal handlers for the timer
signals, which IIRC did not get passed to the child process on a
fork() call. The solution was to turn off these signals before the
fork, and turn them back on afterwards (in both parent and child).
There are some comments in the UnixOSProcessPlugin>>forkSqueak:
method that give some further explanation.

I have not yet had a chance to look into this with Cog but I suspect
that the use of pthreads for timer events may be a problem for
#forkSqueak:. I don't think that pthreads are reestablished in a
child process after a fork(), so I am guessing that the missing piece
would be a way for the child Squeak process to restart the timer
threads. If that's the issue, it would probably be easy to resolve
if we provide a restartTimerThreads() function, which could be a
no-op for the traditional VM. I cannot explain why the parent process
crashes when calling #forkSqueak: from Cog, possibly there is some
interaction between the signal handling and pthreads (signal delivery
to pthreads on Linux is a bit awkward).

I have not yet had a chance to look into this so treat the above as
pure speculation.

If we don't have a solution to this, I don't think it's unreasonable
to just fail the #forkSqueak primitives when running on a VM that does
not support them. I could put a check for this based on image format
version in the plugin as a workaround to prevent VM crashes. For most
people, the only practical impact would be that many of the OSProcess
unit tests would not run.

Dave

On Mon, Sep 20, 2010 at 02:49:35PM -0700, Eliot Miranda wrote:

> Hi David,
>
> what's the reason for shutting down interval timers around fork? What
> was the issue that necessitated this in the first place? We're likely
> getting bit by this and I want to try and keep the solution as simple as
> possible.
>
> best
> Eliot
>
> On Sat, Sep 4, 2010 at 7:20 AM, David T. Lewis <[hidden email]> wrote:
>
> >
> > I think that you have confirmed that #forkSqueak does not work on a Cog VM.
> > The likely reason for this is that #forkSqueak involves temporarily
> > disabling
> > timer signals from the operating system, then reenabling them in both the
> > child and parent after the fork(). The Cog VM is using a different
> > underlying
> > timer mechanism (pthreads etc) so the current #forkSqueak mechanism would
> > probably need to be redone for Cog. I have not looked in to this at all,
> > but I did expect that this might be a problem and I think you have
> > confirmed
> > it in the log below.
> >
> > Dave
> >
> > On Fri, Sep 03, 2010 at 09:52:29AM -0300, Casimiro de Almeida Barreto
> > wrote:
> > >
> > > Here's the log:
> > >
> > > Untrustedir /home/casimiro/Desenvolvimento/cog.work/secure
> > > UnixOSPrReceiversessionIdentvaria DependentsArray(a UnixProcess with pid
> > > 3490)
> > > fier: #[202 58grimReaper: siona Process in [] in [] in
> > > UnixOSProcessAccessor>>
> > > grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> > > in Uni
> > > UnixOSPrReceiversessionIdentifier: epend#[202 58 13 UnixProcess with pid
> > > 3490)
> > > 6] grimReaper: sio a Process in [] in [] in
> > UnixOSProcessAccessor>>
> > > grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> > > in Uni
> > > UnixOSPrReceivercanObtainSessionIdenpend#[202 58 13 86]xProcess with pid
> > > 3490)
> > > ifierFromPlugin:grimReaper: a Process in [] in [] in
> > > UnixOSProcessAccessor>>
> > > grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> > > in Uni
> > > UnixOSPrReceiversessionIdentifier: epend#[2sArray(a UnixProcess with pid
> > > 3490)
> > > 2 58 13 86] grimReaper: siona Process in [] in [] in
> > > UnixOSProcessAccessor>>
> > > grimReaperProcessigChldSemaphore: a Semaphore(a Process in [] in []
> > > in Uni
> > > UnixProcess>>redirect:to:- - - - - - - - - - - - - - - - - - - - -
> > > [] in PipeableOSProcess
> > > class>>forkSqueakAndDoThenQuit:errorPipelineStream:headl
> > >
> > Compiler>>evaluate:in:to:notifyiningTo:toPrecision:maxEpochs:reportingEach:ss:
> > > PluggableTextMorphPlus(PluggableTextMorph)>>handleEdit:
> > > MorphicEventh(Morph)>>handleEvent:nt:ndTo:
> > > [] in HandMhandleFocusEvent:nt:using::with:h:
> > > ...etc...>>handleEvent:nt::clear:r:ing:
> > >
> > >
> > > BTW, XDisplayControlPlugin explodes cogvm at my face... but seems to be
> > > required by TranscriptStream (trying to figure out what's going on
> > > there... then I'll report the problem properly).
> > >
> >
> >
> >
> >