[Xen-devel] Re: xen/stable 2.6.32.9 32bit dom0 kernel crashes early on boot

[Jeremy Fitzhardinge <jeremy@xxxxxxxx>]

On 03/04/2010 11:53 AM, Pasi Kärkkäinen wrote:
> Hello,
>
> 32bit PAE 2.6.32.9 xen/stable dom0 kernel crashes early on my (old) testbox..
>
> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-xen-stable-2.6.32.9-32b-crash-log01.txt
>
> ACPI: LAPIC_NMI (acpi_id[0x00] high edge lint[0x1])
> ACPI: LAPIC_NMI (acpi_id[0x01] high edge lint[0x1])
> ACPI: LAPIC_NMI (acpi_id[0x02] high edge lint[0x1])
> ACPI: LAPIC_NMI (acpi_id[0x03] high edge lint[0x1])
> ACPI: IOAPIC (id[0x04] address[0xfec00000] gsi_base[0])
> (XEN) mm.c:1746:d0 Bad L1 flags 80000100
> (XEN) mm.c:778:d0 Bad L1 flags 80000100
> (XEN) mm.c:4637:d0 ptwr_emulate: could not get_page_from_l1e()
> (XEN) d0:v0: unhandled page fault (ec=0003)
> (XEN) Pagetable walk from c0259fd0:
> (XEN)  L3[0x003] = 000000003c912001 00000912
> (XEN)  L2[0x001] = 000000003d3b6067 000013b6
> (XEN)  L1[0x059] = 000000003c259061 00000259
> (XEN) domain_crash_sync called from entry.S (ff1cb962)
> (XEN) Domain 0 (vcpu#0) crashed on cpu#0:
> (XEN) ----[ Xen-4.0.0-rc5  x86_32p  debug=y  Not tainted ]----
> (XEN) CPU:    0
> (XEN) EIP:    e019:[<c0405dbe>]
> (XEN) EFLAGS: 00000296   EM: 1   CONTEXT: pv guest
> (XEN) eax: 00000000   ebx: c0259fd0   ecx: 80000000   edx: 3c913163
> (XEN) esi: 80000000   edi: 00000000   ebp: c0848e54   esp: c0848e34
> (XEN) cr0: 8005003b   cr4: 000006f0   cr3: 00bc6c80   cr2: c0259fd0
> (XEN) ds: e021   es: e021   fs: 00d8   gs: 0000   ss: e021   cs: e019
> (XEN) Guest stack trace from esp=c0848e34:
> (XEN)    00000003 c0405dbe 0001e019 00010096 3c913163 deadbeef deadbeef f57fa000
> (XEN)    c0848e74 c0405f1c c0259fd0 80000000 3c913163 80000000 c0259fd0 f57fa000
> (XEN)    c0848e98 c04281f9 3c913163 80000000 80000000 00913163 00005000 f57ff000
> (XEN)    0000017b c0848ea8 c042747f 00000005 80000000 c0848ec4 c0404f98 c07c3370
> (XEN)    c0848ec4 00000000 00000000 00000004 c0848ee0 c089aeec 0000017b 80000000
> (XEN)    f55002cc f55002cc c07c3370 c0848eec c0895f7a 00000001 c0848f10 c08ad384
> (XEN)    f5500310 00000001 00000090 f5500280 00000000 00000000 c12aea00 c0848f20
> (XEN)    c08ad3fa c0895f4d 00000040 c0848f30 c089650f 168ff000 2d1fe000 c0848fac
> (XEN)    c088f56a 00000000 00000000 012aea00 00000000 00000000 00ad5000 00000000
> (XEN)    c0848f60 00000092 00000000 2d1fe000 00000000 40000000 00000000 c0945be4
> (XEN)    00000000 00000000 00000000 00000000 c085ab8c c0848f9c c04063b6 00000000
> (XEN)    00000000 00000000 00000000 00000000 00000000 c085ab8c c0848fc4 c088a5bd
> (XEN)    c07b9c5c c06e8010 c08bfb44 00ad5000 c0848fd4 c088a0a8 00ad5000 c08be930
> (XEN)    c0848ffc c088d113 1fc89375 80000401 00010800 00000f34 00000001 00000000
> (XEN)    c13af000 00000000 00000000 3d3b1001 00000000 3d3b2001 00000000 3d3b3001
> (XEN)    00000000 3c912001 00000000 00000000 00000000 00000000 00000000 00000000
> (XEN)    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> (XEN)    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> (XEN)    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> (XEN)    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> (XEN) Domain 0 crashed: rebooting machine in 5 seconds.
>
>
>
> (gdb) x/i 0xc0405dbe
> 0xc0405dbe<xen_set_pte+137>:   mov    %edx,(%ebx)
>
> (gdb) list *0xc0405dbe
> 0xc0405dbe is in xen_set_pte (arch/x86/xen/mmu.c:703).
> 698             ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == 
> PARAVIRT_LAZY_MMU);
> 699
> 700     #ifdef CONFIG_X86_PAE
> 701             ptep->pte_high = pte.pte_high;
> 702             smp_wmb();
> 703             ptep->pte_low = pte.pte_low;
> 704     #else
> 705             *ptep = pte;
> 706     #endif
> 707     }
>    
Xen is 32 bit, right?

Could you disassemble this whole function, so I can see what register 
the other half is in? (I'm guessing its ecx or esi; nothing else makes 
much sense.)
Could you also try to work out at least some of the call stack to see 
what is trying to set a pte?  The presence of _PAGE_UNUSED1 suggests its 
the CPA test.  If so, you could try disabling that (CONFIG_CPA_DEBUG).
Keir, 32-bit Xen isn't reserving bit 9 in the PTE is it?  It just uses 
the high user bits?
Thanks,
    J

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel