svn commit: r184558 - head/sys/dev/acpica/Osd
jkim at FreeBSD.org
Mon Nov 3 09:27:02 PST 2008
On Monday 03 November 2008 11:08 am, Alexander Motin wrote:
> Jung-uk Kim wrote:
> > On Sunday 02 November 2008 07:50 am, Alexander Motin wrote:
> >> Author: mav
> >> Date: Sun Nov 2 12:50:16 2008
> >> New Revision: 184558
> >> URL: http://svn.freebsd.org/changeset/base/184558
> >> Log:
> >> As soon as we have several threads per process now, it is not
> >> correct to use process ID as ACPI thread ID. Concurrent requests
> >> with equal thread IDs broke ACPI mutexes operation causing
> >> unpredictable errors including AE_AML_MUTEX_NOT_ACQUIRED that I
> >> have seen.
> >> Use kernel thread ID instead of process ID for ACPI thread.
> > Sorry but this patch is incorrect, i.e., td_tid is not unique.
> > You have to use curthread or (p_pid, td_tid) pair.
> > Unfortunately, even if you correct this problem, you also have to
> > correct ACPI_THREAD_ID definition, which is in the vendor code.
> > That's why it wasn't done yet and it is more complicated than you
> > think, i.e., ACPI-CA assumes sizeof(ACPI_THREAD_ID) ==
> > sizeof(int), etc. Please see the related ACPI-CA bugs:
> I'm also sorry, but that is what I see:
> typedef __int32_t __lwpid_t; /* Thread ID (a.k.a. LWP)
> */ ...
> td->td_tid = alloc_unr(tid_unrhdr);
> tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
> So what have I missed, where is the problem? Why td_tid is not
> unique and where is the size problem?
On top of that:
/* Returning 0 is not allowed. */
return (curthread->td_tid + 1);
may actually return 0 because it can be INT_MAX. :-)
More information about the svn-src-all