[fedora-arm] F17 rootfs on Trim Slice = kernel problems

[David A. Marlin]

Richard W.M. Jones wrote:
> I tried the rootfs posted here:
>
> http://fedora.roving-it.com/rootfs-f17-hfp-alpha1.tar.bz2
>
> on a Trim Slice.  However the included kernel
> (3.3.0-0.rc4.git3.1.fc17.armv7hl.tegra) gives lots of errors like
> this:
>
> [  572.198943] BUG: sleeping function called from invalid context at include/linux/freezer.h:46
> [  572.207448] in_atomic(): 0, irqs_disabled(): 128, pid: 538, name: bash
> [  572.214012] INFO: lockdep is turned off.
> [  572.217969] irq event stamp: 0
> [  572.221047] hardirqs last  enabled at (0): [<  (null)>]   (null)
> [  572.227115] hardirqs last disabled at (0): [<c0027054>] copy_process.part.31+0x54c/0x12cc
> [  572.235380] softirqs last  enabled at (0): [<c0027054>] copy_process.part.31+0x54c/0x12cc
> [  572.243628] softirqs last disabled at (0): [<  (null)>]   (null)
> [  572.249789] [<c001613c>] (unwind_backtrace+0x0/0x124) from [<c00110f4>] (do_signal+0x94/0x5ac)
> [  572.258482] [<c00110f4>] (do_signal+0x94/0x5ac) from [<c0011b80>] (do_notify_resume+0x20/0x5c)
> [  572.267168] [<c0011b80>] (do_notify_resume+0x20/0x5c) from [<c000e284>] (work_pending+0x24/0x28)
>
> This breaks every process (not just bash), so not much works.  I've no
> idea why it's trying to suspend.
>
> There's apparently a patch that fixes this:
>
> http://lists.infradead.org/pipermail/linux-arm-kernel/2012-January/079842.html
>
> However I cannot at the moment test it because I don't have a
> cross-compiler setup.  Surely I can't be the only one seeing these
> failures?
>   
Others have seen it as well,

    http://www.spinics.net/lists/linux-omap/msg64732.html

but while it is a known issue affecting ARM (not just OMAP or Tegra), it 
does not appear to have any acceptable resolution as of 3.3-rc4, at 
least nothing that has been accepted upstream.  The thread ends with:

-----
I had found this patch also when I searched the list. That's what I

meant by "one patch proposal by Russell", but it was back in August and
the patch doesn't seem to have been ever applied.  The discussion also
seemed to end abruptly, so my assumption was that no conclusion has been
really reached.

Instead of applying that, I have simply removed the might_sleep() call
in the try_to_freeze() function.  This definitely doesn't solve the
problem, but at least my console doesn't get spammed and it's better
than disabling CONFIG_DEBUG_ATOMIC_SLEEP, because at least I still catch
other sleeps while atomic. ;)
-----



d.marlin
========