"xlock already held" panics on 5.3-BETA(3,4,5)

On Thu, Sep 23, 2004 at 10:29:47AM -0400, John Baldwin wrote:

> > > > panic: _sx_xlock (user map): xlock already held _at_
> > > > /var/data/5/src/sys/vm/vm_map.c:2997 cpuid = 0
> > > > KDB: stack backtrace:
> > > > kdb_backtrace(100,c0f794e0,c0e983c8,1,bb5) at kdb_backtrace+0x29
> > > > panic(c05eaffa,c05d320c,c05fd313,c05fd393,bb5) at panic+0x114
> > > > _sx_xlock(c0e983c8,c05fd393,bb5) at _sx_xlock+0x44
> > > > _vm_map_lock_read(c0e98384,c05fd393,bb5,1f49a0c,c113692c) at  _vm_map_lock_read+0x37
> > > > vm_map_lookup(c9f49a64,bfca2000,1,c9f49a68,c9f49a58) at vm_map_lookup+0x28
> > > > vm_fault(c0e98384,bfca2000,1,0,c0f794e0) at vm_fault+0x66
> > > > trap_pfault(c9f49b2c,0,bfca253c) at trap_pfault+0xd2
> > > > trap(18,10,10,c0e98444,c0c95c48) at trap+0x30d
> > > > calltrap() at calltrap+0x5
> > > > --- trap 0xc, eip = 0xc05b541c, esp = 0xc9f49b6c, ebp = 0xc9f49b78 ---
> > > > pmap_enter_quick(c0e98444,2894f000,c0d7c2e8,0,0) at pmap_enter_quick+0xc8
> > >
> > > This is the real bug, can you do 'l *0xc05b541c' from kgdb?
> >
> > (kgdb) l *0xc05b541c
> > 0xc05b541c is in pmap_enter_quick
> > (/var/data/5/src/sys/i386/i386/pmap.c:2130).
[]
> > 2129            pte = vtopte(va);
> > 2130            if (*pte) {
> > 2131                    if (mpte != NULL) {
> > 2132                            pmap_unwire_pte_hold(pmap, mpte);
> > 2133                            mpte = NULL;
> > 2134                    }

  I've digged a bit...
  Seems that `if (*pte)' causes a Page Fault. So, is it legal for page to
be non-resident here?

-- 
Pavel Gubin
TUSC&R / Industrial Electronics dept / System Administrator
2:5005/14_at_fidonet / Phone +7-3822-423067 / ICQ 28835566