Re: Bug in master branch: IbmPrep40pMachine.test_factory_firmware_and_netbsd broken

[Philippe Mathieu-Daudé]

On 2/3/23 10:48, Cédric Le Goater wrote:
> On 3/2/23 08:47, Thomas Huth wrote:
> >   Hi all,
> >
> > seems like we've got another bug that sneaked in during the CI minutes 
> > blackout: The avocado test 
> > IbmPrep40pMachine.test_factory_firmware_and_netbsd is now broken 
> > reliably, see:
> >
> >   https://gitlab.com/qemu-project/qemu/-/jobs/3858833617#L300
> >   https://gitlab.com/thuth/qemu/-/jobs/3858727901#L300
> >   https://gitlab.com/thuth/qemu/-/jobs/3857804517#L300
> >
> > Is anybody already looking into this?
> >
> >   Thomas
>
> I tried manually and QEMU reached :
>        >> (builds@wb29, Tue Oct  7 23:59:44 PDT 2008)
>      Boot:
>      Loading in()
>      3558080+232036 [197696+188133]=0x3fbbf0
>      start=0x100000
>      Copyright (c) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 
> 2005,
>          2006, 2007
>          The NetBSD Foundation, Inc.  All rights reserved.
>      Copyright (c) 1982, 1986, 1989, 1991, 1993
>          The Regents of the University of California.  All rights reserved.
>      NetBSD 4.0.1 (GENERIC) #0: Wed Oct  8 00:46:56 PDT 2008
>          
> builds@wb29:/home/builds/ab/netbsd-4-0-1-RELEASE/prep/200810080053Z-obj/home/builds/ab/netbsd-4-0-1-RELEASE/src/sys/arch/prep/compile/GENERIC
>      Model: IBM PPS Model 6015
>      total memory = 128 MB
>      avail memory = 119 MB
>      trap: kernel read DSI trap @ 0x7c3042a6 by 0x3e4cf8 (DSISR 
> 0x40000000, err=14), lr 0x377eac
>      panic: trap
>      Stopped in pid 0.1 () at        netbsd:cpu_Debugger+0x10:       
> lwz     r0, r1, 0
>      x14

Bisecting using:

     make print-V \
  && ninja qemu-system-ppc \
  && AVOCADO_ALLOW_UNTRUSTED_CODE=1 \
     avocado --show=app,console \
       run --test-runner=nrunner -t slowness:high \
         tests/avocado/ppc_prep_40p.py

commit cef2e7148e32d61338de0220619d308bf42af770
Author: Philippe Mathieu-Daudé <f4bug@amsat.org>
Date:   Fri Feb 10 16:05:34 2023 +0100

    hw/isa/i82378: Remove intermediate IRQ forwarder

    When the i82378 model was added in commit a04ff940974 ("prep:
    Add i82378 PCI-to-ISA bridge emulation") the i8259 model was
    not yet QOM'ified. This happened later in commit 747c70af78f
    ("i8259: Convert to qdev").

    Directly dispatch ISA IRQs to 'cpu_intr' output IRQ
    by removing the intermediate i82378_request_out0_irq()
    handler. Rename ISA IRQs array as 'isa_irqs_in' to
    emphasize these are input IRQs.

Sorry :/ I'll look at this.