Re: 2.6.14-rc4-rt1 - enable IRQ-off tracing causes kernel to fault at boot

From: Mark Knecht (markknecht_at_gmail.com)
Date: 10/14/05

  • Next message: Stephan Brodkorb: "[PATCH 1/1] n_r3964 fix - char" 
    Date:	Fri, 14 Oct 2005 07:56:05 -0700
To: Ingo Molnar <mingo@elte.hu>

    On 10/13/05, Ingo Molnar <mingo@elte.hu> wrote:
    >
    > * Mark Knecht <markknecht@gmail.com> wrote:
    >
    > > Ingo & Steve,
    > > Thank you for your great instructions that even a guitar player
    > > could basically follow. After about an hour of messing around I did
    > > manage to capture the crash. The console file is attached.
    > >
    > > NOTE: The first time I booted the kernel it got to the crash point and
    > > the machine rebooted. The second time it booted I got the trace. Both
    > > boots are in the capture file.
    >
    > thanks, this log is much more informative. No smoking gun though, but it
    > seems something fundamental (probably lowlevel x64 code) has been broken
    > by -rt1.
    >
    > Do the crashes go away if you take the -rc3-rt13 version of
    > arch/x86_64/kernel/entry.S and copy it over into the -rc4-rt1 tree?
    > [this undoes a particular set of CONFIG_CRITICAL_IRQSOFF_TIMING fixes
    > from the x64 code, which i did during -rc3-rt13 => -rc4-rt1]

    Indeed it is fixed by doing this. Options are on but the modified
    kernel does boot:

    *****************************************************************************
    * *
    * REMINDER, the following debugging options are turned on in your .config: *
    * *
    * CONFIG_DEBUG_PREEMPT *
    * CONFIG_CRITICAL_PREEMPT_TIMING *
    * CONFIG_CRITICAL_IRQSOFF_TIMING *
    * *
    * they may increase runtime overhead and latencies. *
    * *
    *****************************************************************************

    mark@lightning ~ $ uname -a
    Linux lightning 2.6.14-rc4-rt1 #8 PREEMPT Fri Oct 14 07:46:29 PDT 2005
    x86_64 AMD Athlon(tm) 64 Processor 3000+ AuthenticAMD GNU/Linux
    mark@lightning ~ $

    >
    > (Note that doing this will re-introduce tracing bugs, which can result
    > in false-positive latency readings - but it should fix any related
    > lowlevel bug in the assembly code.)
    >
    > if this indeed solves the crash then i'd suggest to restore the -rt1
    > version of entry.S, and i'd suggest to disable CRITICAL_IRQSOFF_TIMING
    > until i fix it. You should be able to get pretty good latency tracing
    > info even without CRITICAL_IRQSOFF_TIMING.
    >
    > Ingo
    >

    Will got back to the original entry.S file with the IRQoff option
    turned off. Let me know when you have a fix to test, or if you need
    more data.

    Thanks,
    Mark
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Stephan Brodkorb: "[PATCH 1/1] n_r3964 fix - char"

    Relevant Pages

    • Re: 2.6.0-test2-mm5
      ... 4/4 split now boots again, but behaves rather oddly. ... > Nick's AS fix, plus the 4/4 fix Andrew sent me last night, which I presume ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: [PATCH] Fix for vmalloc problem was Re: 2.6.10-rc1-mm3
      ... > Boots fine now thanks. ... It broke x86-64 unfortunately. ... Fix vmalloc overflow with 4levels. ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: 2.6.13-rt3
      ... i've released -rt4 with this fix included. ... The 8-way box boots fine ... send the line "unsubscribe linux-kernel" in ... Please read the FAQ at http://www.tux.org/lkml/ ...
      (Linux-Kernel)
    • Re: Power Management Update
      ... I encountered this problem by having an IDE CD-ROM, ... He mentioned producing a cleaner patch, but this should at least fix the ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: [PATCH] fs/fcntl.c : dont test unsigned value for less than zero
      ... I think the real problem here is that 'arg' ... architecture's ptrace code could easily make use of the latter, ... But be careful not to "fix up" ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)