Re: xhci panic

[Nick Hudson <skrll%netbsd.org@localhost>]

On 01/20/17 14:35, Patrick Welche wrote:
> On Fri, Jan 20, 2017 at 02:10:24PM +0000, Nick Hudson wrote:
> > On 01/20/17 11:50, Patrick Welche wrote:
> > > On Fri, Jan 20, 2017 at 11:08:14AM +0000, Nick Hudson wrote:
> > > > On 01/20/17 10:56, Patrick Welche wrote:
> > > > > Just tried a this morning's kernel on amd64 (was looking forward to the
> > > > > new xhci.c :-) ), and (laptop so no serial console):
> > > > >
> > > > > xhci0 at pci4 dev 0 function 0: NEC USB3 Houst Controller (rev. 0x04)
> > > > > xhci0: interrupting at msi2 vec 0
> > > > > uvm_fault(0x..., 0x0, 1) -> e
> > > > >
> > > > > Stopped in pid 0.1 (system) at netbsd:vmem_alloc
> > > > > vmem_alloc
> > > > > uvm_km_kmem_alloc
> > > > > kmem_intr_alloc
> > > > > kmem_intr_zalloc
> > > > > xhci_init
> > > > > xhci_pci_attach
> > > > > ...
> > > > > mp_pci_scan
> > > > Can you boot with -x and send the output please
> > > OK - going off-list!
> > >
> > > P
> > can you show the full backtrace,e.g.
> >
> > Stopped in pid 0.1 (system) at netbsd:vmem_alloc+FOO
> >
> > need to know the source line of vmem_alloc+FOO is
> (gdb) info line *(vmem_alloc+0x3e)
> Line 1056 of "../../../../kern/subr_vmem.c"
>     starts at address 0xffffffff8052efa2 <vmem_alloc+60>
>     and ends at 0xffffffff8052efac <vmem_alloc+70>.
>
>       1  /*      $NetBSD: subr_vmem.c,v 1.95 2016/07/07 06:55:43 msaitoh Exp $   */
> ...
>    1050  #if defined(QCACHE)
>    1051          if (size <= vm->vm_qcache_max) {
>    1052                  void *p;
>    1053                  int qidx = (size + vm->vm_quantum_mask) >> vm->vm_quantum_shift;
>    1054                  qcache_t *qc = vm->vm_qcache[qidx - 1];
>    1055
>    1056                  p = pool_cache_get(qc->qc_cache, vmf_to_prf(flags));
>    1057                  if (addrp != NULL)
>    1058                          *addrp = (vmem_addr_t)p;
>    1059                  return (p == NULL) ? ENOMEM : 0;
>    1060          }
>    1061  #endif /* defined(QCACHE) */
>
> For more, it'll be easier to do later when I'm near a second computer...
> (I can boot this laptop with boot -c, disable xhci, using an envelope as
> memory...)
>
> Cheers,
>
> Patrick

Does this help?

Nick

Index: sys/dev/usb/xhci.c
===================================================================
RCS file: /cvsroot/src/sys/dev/usb/xhci.c,v
retrieving revision 1.69
diff -u -p -r1.69 xhci.c
--- sys/dev/usb/xhci.c	19 Jan 2017 20:35:44 -0000	1.69
+++ sys/dev/usb/xhci.c	20 Jan 2017 14:52:33 -0000
@@ -958,12 +958,12 @@ xhci_init(struct xhci_softc *sc)
 	aprint_debug_dev(sc->sc_dev, "xECP %x\n", XHCI_HCC_XECP(hcc) * 4);
 
 	/* default all ports to bus 0, i.e. usb 3 */
-	sc->sc_ctlrportbus = kmem_zalloc(sc->sc_maxports * sizeof(uint8_t) / NBBY, KM_SLEEP);
-	sc->sc_ctlrportmap = kmem_zalloc(sc->sc_maxports * sizeof(int), KM_SLEEP);
+	sc->sc_ctlrportbus = kmem_zalloc(max(sc->sc_maxports * sizeof(uint8_t) / NBBY, 32), KM_SLEEP);
+	sc->sc_ctlrportmap = kmem_zalloc(max(sc->sc_maxports * sizeof(int), 32), KM_SLEEP);
 
 	/* controller port to bus roothub port map */
 	for (size_t j = 0; j < __arraycount(sc->sc_rhportmap); j++) {
-		sc->sc_rhportmap[j] = kmem_zalloc(sc->sc_maxports * sizeof(int), KM_SLEEP);
+		sc->sc_rhportmap[j] = kmem_zalloc(max(sc->sc_maxports * sizeof(int), 32), KM_SLEEP);
 	}
 
 	/*