Re: [ros-diffs] [hbirr] 16714: - Removed R_EXTRA_STACK because R_EXTRA_STACK and R_STACK must have the same size.

[Royce Mitchell III]

[e-mail] wrote:
> Initialized an allocated block completely before release the lock.
>
> @@ -719,9 +725,9 @@
>
> {
>    	if ( (NewBlock = RQueRemove ( &pool->Que[que][align] )) )
>    	{
> 	R_RELEASE_MUTEX(pool);
>
>
> RiUsedInit ( NewBlock, Tag );
> RiUsedInitRedZone ( NewBlock, NumberOfBytes );
>
>
> 	R_RELEASE_MUTEX(pool);
>
>
> return RHdrToBody(NewBlock);
>
> }
> queBytes = 16 << que;
>
> @@ -865,11 +871,10 @@
>
> NewBlock = (PR_USED)BestBlock;
>    RiUsedInit ( NewBlock, Tag );
>
> R_RELEASE_MUTEX(pool);
>
> /*  RtlZeroMemory(RHdrToBody(NewBlock), NumberOfBytes);*/
>
> RiUsedInitRedZone ( NewBlock, NumberOfBytes );
>
> R_RELEASE_MUTEX(pool);
>
> return RHdrToBody(NewBlock);
>
> }

Hartmut,

If you don't mind, I'm curious how this was causing a problem, the stuff 
that was being initialized outside of the lock shouldn't (in my feeble 
mind) have been "visible" upon reentry until this block is freed again.

Thanks,

Royce

[Hartmut Birr]

Royce Mitchell III wrote:
> [e-mail] wrote:
> > Initialized an allocated block completely before release the lock.
> >
> > @@ -719,9 +725,9 @@
> > {
> >     if ( (NewBlock = RQueRemove ( &pool->Que[que][align] )) )
> >     {
> >
> >
> >        R_RELEASE_MUTEX(pool);
> >
> >
> >       RiUsedInit ( NewBlock, Tag );
> >       RiUsedInitRedZone ( NewBlock, NumberOfBytes );
> >
> >
> >        R_RELEASE_MUTEX(pool);
> >
> >
> >       return RHdrToBody(NewBlock);
> >   }
> >   queBytes = 16 << que;
> >
> > @@ -865,11 +871,10 @@
> > NewBlock = (PR_USED)BestBlock;
> > RiUsedInit ( NewBlock, Tag );
> >
> >
> >
> > R_RELEASE_MUTEX(pool);
> >
> > /*  RtlZeroMemory(RHdrToBody(NewBlock), NumberOfBytes);*/
> >
> > RiUsedInitRedZone ( NewBlock, NumberOfBytes );
> >
> > R_RELEASE_MUTEX(pool);
> >
> > return RHdrToBody(NewBlock);
> >
> > }
>
> Hartmut,
>
> If you don't mind, I'm curious how this was causing a problem, the
> stuff that was being initialized outside of the lock shouldn't (in my
> feeble mind) have been "visible" upon reentry until this block is
> freed again.

I've add a 'paged pool check thread' which is activated each 10 minutes.
The thread allocates a zero size block, which triggers the checking. If
the thread does check a new allocated block between releasing the lock
and intializing the red zone, I get a wrong hit.

- Hartmut