cvs commit: src/sys/alpha/alpha mp_machdep.c

[Peter Wemm]

John Baldwin wrote:
> 
> On 01-Dec-2003 Jeff Roberson wrote:
> > 
> > On Mon, 1 Dec 2003, John Baldwin wrote:
> > 
> >>
> >> On 01-Dec-2003 Jeff Roberson wrote:
> >> > On Mon, 1 Dec 2003, John Baldwin wrote:
> >> >
> >> >>
> >> >> On 30-Nov-2003 Jeff Roberson wrote:
> >> >> > jeff        2003/11/30 14:08:24 PST
> >> >> >
> >> >> >   FreeBSD src repository
> >> >> >
> >> >> >   Modified files:
> >> >> >     sys/alpha/alpha      mp_machdep.c
> >> >> >   Log:
> >> >> >    - Set mp_maxid in a way that is consistent with every other arch. 
     It is
> >> >> >      one more than the last valid 'cpuid'.
> >> >> >
> >> >> >   Approved by:    re (rwatson)
> >> >>
> >> >> Eh?  No it's not.  All that needs to be true is that for every processo
    r,
> >> >> PCPU_GET(cpuid) <= mp_maxid.
> >> >>
> >> >> It can be equal to the highest ID and should be so for most cases.
> >> >
> >> > From x86's mp_machdep:
> >> >         mp_maxid = MAXCPU;
> >>
> >> That's a bug.
> >>
> >> > ia64
> >> > mp_maxid = min(mp_ncpus, MAXCPU) - 1;
> >>
> >> This is correct.  Note - 1.
> >>
> >> > sparc64
> >> >         cpus = 0;
> >> >         root = OF_peer(0);
> >> >         for (child = OF_child(root); child != 0; child = OF_peer(child))
     {
> >> >                 if (OF_getprop(child, "device_type", buf, sizeof(buf)) >
     0
> >> > &&
> >> >                     strcmp(buf, "cpu") == 0)
> >> >                         cpus++;
> >> >         }
> >> >         mp_maxid = cpus;
> >>
> >> This should be cpus - 1.
> >>
> >> > And I just changed amd64 and alpha to mean maxid + 1.  Damn. it looks li
    ke
> >> > just sparc64 and x86 were wrong.  Maybe I should change those to mean th
    e
> >> > right thing, and backout the changes to uma.  yack.
> >>
> >> Heh.  I am already waiting for approval from re@ to fix x86.  That was
> >> definitely my bug.
> > 
> > Well, since ULE and UMA are the only two subsystems which make use of
> > mp_maxid, can I request that it be one over the maximum id so that I can
> > use it in loops like:
> > for (i = 0; i < mp_maxid; i++) {
> > 
> > Otherwise I guess I can do <=.  If you fix x86, can you fix alpha and
> > amd64, and uma and ule? :-)
> 
> I'll fix all of them and work up a patch tomorrow, sure. :)

Well, whatever happens, please do something promptly because the amd64 GENERIC
kernel is broken by all this and panics on a UP machine.  I think the problem
is currently bandaid-able by doing this:

+++ mp_machdep.c        2 Dec 2003 01:41:02 -0000
@@ -201,3 +201,3 @@
        else if (mp_ncpus == 1)
-               mp_maxid = 0;
+               mp_maxid = 1;
        else
@@ -234,3 +234,3 @@
                 */
-               mp_maxid = 0;
+               mp_maxid = 1;
                return (0);

In other words, the code sets it to 1 and then changes it back to 0 again.

Cheers,
-Peter
--
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com
"All of this is for nothing if we don't go to the stars" - JMS/B5