Re: cvs commit: src/sys/kern kern_sig.c

["Simon 'corecode' Schubert" <corecode@xxxxxxxxxxxx>]

Matthew Dillon wrote:
> :corecode    2007/02/22 07:48:55 PST
> :
> :DragonFly src repository
> :
> :  Modified files:
> :    sys/kern             kern_sig.c 
> :  Log:
> :  Fix stopping of the current process, which was broken by the last commit.
> :  
> :  If the current process is still far inside the kernel and is going to
> :  tsleep(), CURSIG()/issignal() in tsleep() have to see the pending signal;
> :..
>
>     You can't actually stop a process in tsleep, ever.  The kernel might
>     be holding resources (such as lockmgr locks) that would never be 
>     released.
>
>     A process can only be stopped in userret() (platform/pc32/i386/trap.c).
>
>     Please make sure that those semantics are retained.

That always happened.  CURSIG/issignal check for a SA_STOP signal and calls tstop, which in turn calls tsleep() in a loop until SSTOP (previously P_STOPPED) is cleared.

I'm also not fond of this, but seems the kernel code assumes this behaviour.  I noticed this in ttread():  it calls ttysleep()/tsleep(), but really expects the thread to be stopped *inside*.

I'd rather have the threads stop only in userret() as well (and possibly when entering the kernel), but we'd need very careful audit of CURSIG/issignal() users.

cheers
 simon

--
Serve - BSD     +++  RENT this banner advert  +++    ASCII Ribbon   /"\
Work - Mac      +++  space for low €€€ NOW!1  +++      Campaign     \ /
Party Enjoy Relax   |   http://dragonflybsd.org      Against  HTML   \
Dude 2c 2 the max   !   http://golden-apple.biz       Mail + News   / \

Attachment: signature.asc
Description: OpenPGP digital signature