Re: Panic on dom0 boot

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Fri, 21 Mar 2025, Manuel Bouyer wrote:
> On Fri, Mar 21, 2025 at 04:41:22PM +0000, Stephen Borrill wrote:
> > I have just installed a 10.1_STABLE machine with xenkernel418.
> >
> > After making the boot.cfg changes to boot xen and the the XEN3_DOMO kernel,
> > I get a panic during autoconfiguration.
> >
> > Initially it appeared to be related to ixgbe_attach(), so I tried without
> > ixgbe and instead it panicked during mfii_attach(), so it seems a general
> > PCI attach or MSI-X problem.
> >
> > Traceback without details:
> >
> > panic()
> > bind_pirq_to_evtch()
> > intr_establish_xname()
> > x86_pci_msix_establish()
> > mfii_attach() ot ixgbe_attach()
> > config_attached_internal)
> > config_found()
> > pci_probe_device1()
> >
> > If it will help, I can try to connect a serial console after the weekend.
>
> Yes, it can help to get the complete panic message, and also to boot with
> the xen-debug.gz kernel (which is usefull only with a serial console in
> this case)

I finally managed to get a serial console up (and will fix the how-to, hi 
gdt@!). The problem was reasonably obvious:

[   1.0000030] mfii0 at pci4 dev 0 function 0: "ServeRAID M5110", firmware 23.34.0-0023, 512MB cache
[   1.0000030] panic: pirq 4991 out of bound, increase NR_PIRQS
[   1.0000030] cpu0: Begin traceback...
[   1.0000030] vpanic() at netbsd:vpanic+0x177
[   1.0000030] panic() at netbsd:panic+0x3c
[   1.0000030] bind_pirq_to_evtch() at netbsd:bind_pirq_to_evtch+0x97

From sys/arch/xen/xen/evtchn.c:

#define NR_PIRQS        NR_EVENT_CHANNELS

This equates to 4096 and, indeed, 4991 is greater than 4096.

I applied the following brute-force patch and it boots fine:

--- sys/arch/xen/xen/evtchn.c   7 Sep 2022 00:40:19 -0000       1.100
+++ sys/arch/xen/xen/evtchn.c   24 Mar 2025 14:18:26 -0000
@@ -89,7 +89,7 @@
 #define NBSD_XEN_MAX_VCPUS HVM_MAX_VCPUS
 #endif

-#define        NR_PIRQS        NR_EVENT_CHANNELS
+#define        NR_PIRQS        (NR_EVENT_CHANNELS*2)

 /*
  * This lock protects updates to the following mapping and 
reference-count
@@ -656,7 +656,7 @@
                return -1;

        if (pirq >= NR_PIRQS) {
-               panic("pirq %d out of bound, increase NR_PIRQS", pirq);
+               panic("pirq %d out of bound, increase NR_PIRQS (currently %lu)", pirq, NR_PIRQS);
        }
        mutex_spin_enter(&evtchn_lock);

@@ -674,7 +674,7 @@
        int evtchn;

        if (pirq >= NR_PIRQS) {
-               panic("pirq %d out of bound, increase NR_PIRQS", pirq);
+               panic("pirq %d out of bound, increase NR_PIRQS (currently %lu)", pirq, NR_PIRQS);
        }

        mutex_spin_enter(&evtchn_lock);



--
Stephen