harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pavel Pervov" <pmcfi...@gmail.com>
Subject Re: [harmony][Thread]
Date Tue, 18 Dec 2007 14:55:15 GMT
Chengrong,

I think it is vice versa. Wherever you have a safe point, stack must be
enumerable.

Fast path does not contain safe points, so no need for m2n there. Slow path
contains some - and we have to create m2n to allow stack enumeration if GC
happens.

Pavel.

P.S. BTW, not creating m2n makes fast path even faster.

On 12/18/07, 史成荣 <icyrong@gmail.com> wrote:

> I am not familiar with the relationship between setting the m2nframe and
> GC.
> I guess if there is no potential garbage collection, it's better to take
> the
> fast path,because the fast path does not contain safepoint for GC to
> start,
> and if GC really happens when a thread is requiring or releasing a lock,
> then it should transfer from the fast path to the slow one so that GC
> can start, because the slow path contains safepoint for GC to start. And
> in order to unwind the stack correctly, the m2nframe should be setted
> before
> the slow path.
> Am I right or I missed some thing else?
>
> Thanks,
> Chengrong
>
>
> 2007/12/18, Xiao-Feng Li <xiaofeng.li@gmail.com>:
> >
> > On Dec 18, 2007 8:43 PM, 史成荣 <icyrong@gmail.com> wrote:
> > > why does it need setting the m2nframe before the slow path while it
> > needn't
> > > when taking the fast path?
> >
> > As I said below, that's for potential garbage collection or exception
> > throwing, when you need unwind the stack correctly.
> >
> > Thanks,
> > xiaofeng
> >
> > > 在07-12-18,Xiao-Feng Li <xiaofeng.li@gmail.com> 写道:
> > >
> > > >
> > > > Chengrong, locking is expensive, which should be optimized or
> avoided.
> > > > The fast path acquiring the lock without Java-native border crossing
> > > > can be much fast. It's expected the fast path be executed in most
> > > > cases. This is possible if the code sequence doesn't trigger garbage
> > > > collection or exception, hence no need to bookkeeping the frames.
> > > >
> > > > Thanks,
> > > > xiaofeng
> > > >
> > > >
> > >
> >
> >
> >
> > --
> > http://xiao-feng.blogspot.com
> >
>



-- 
Pavel Pervov,
Intel Enterprise Solutions Software Division
Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message