On Sat, Sep 11, 2010 at 02:36:26PM +0100, Rui Paulo wrote: > On 11 Sep 2010, at 14:23, Kostik Belousov wrote: > > > On Sat, Sep 11, 2010 at 12:58:31PM +0000, Rui Paulo wrote: > >> Author: rpaulo > >> Date: Sat Sep 11 12:58:31 2010 > >> New Revision: 212465 > >> URL: http://svn.freebsd.org/changeset/base/212465 > >> > >> Log: > >> Avoid a LOR (sleepable after non-sleepable) in > >> fasttrap_tracepoint_enable(). > >> > >> Sponsored by: The FreeBSD Foundation > >> > >> Modified: > >> head/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c > >> > >> Modified: head/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c > >> ============================================================================== > >> --- head/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c Sat Sep > >> 11 12:51:01 2010 (r212464) > >> +++ head/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c Sat Sep > >> 11 12:58:31 2010 (r212465) > >> @@ -591,7 +591,9 @@ fasttrap_tracepoint_enable(proc_t *p, fa > >> * Before we make any modifications, make sure we've imposed a barrier > >> * on the generation in which this probe was last modified. > >> */ > >> + PROC_UNLOCK(p); > >> fasttrap_mod_barrier(probe->ftp_gen); > >> + PROC_LOCK(p); > >> > >> bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; > >> > > I suspect that you should hold the process around unlocked region. > > The unlocked region in the diff ? You're saying I shouldn't unlock/lock here? No, I am saying that you should do _PHOLD before dropping process lock.
> > > fasttrap_pid_enable() also does unlock, so it might be better to hold > > the process immediately after pfind(). > > You mean use _PHOLD after pfind()? Yes.
pgp0bMJL9Wh9S.pgp
Description: PGP signature