Random Kernel Panic on Dreamplug (FS related)
Warner Losh
imp at bsdimp.com
Sun May 17 23:21:54 UTC 2015
> On May 17, 2015, at 1:26 PM, Ronald Klop <ronald-lists at klop.ws> wrote:
>
> On Sun, 17 May 2015 00:16:23 +0200, Ian Lepore <ian at freebsd.org> wrote:
>
>> On Tue, 2014-09-30 at 16:29 +0200, Mattia Rossi wrote:
>>> Am 30.09.2014 16:19, schrieb Ian Lepore:
>>> > On Tue, 2014-09-30 at 16:05 +0200, Mattia Rossi wrote:
>>> >> Am 30.09.2014 14:30, schrieb John-Mark Gurney:
>>> >>> Mattia Rossi wrote this message on Tue, Sep 30, 2014 at 14:14 +0200:
>>> >>>> Am 30.09.2014 13:29, schrieb John-Mark Gurney:
>>> >>>>> Mattia Rossi wrote this message on Mon, Sep 29, 2014 at 10:42 +0200:
>>> >>>>>> Am 29.09.2014 06:01, schrieb John-Mark Gurney:
>>> >>>>>>> Mattia Rossi wrote this message on Fri, Sep 26, 2014 at 14:19 +0200:
>>> >>>>>>>> This might be part of the weird FFS issues the Dreamplug has and no-one
>>> >>>>>>>> knows why they're happening.
>>> >>>>>>> Are you running w/ FFS journaling? If so, try turning it off, but
>>> >>>>>>> keeping softupdates on..
>>> >>>>>> No journaling, no softupdates. I'll try enabling softupdates next time.
>>> >>>>>> don't know if it will panic though
>>> >>>>>>>> data_abort_handler() at data_abort_handler+0x5c0
>>> >>>>>>>> pc = 0xc0de7a28 lr = 0xc0dd711c (exception_exit)
>>> >>>>>>>> sp = 0xde019898 fp = 0xde019a20
>>> >>>>>>>> r4 = 0xffffffff r5 = 0xffff1004
>>> >>>>>>>> r6 = 0xc3f3f6c0 r7 = 0x00001000
>>> >>>>>>>> r8 = 0xc443e880 r9 = 0x00000000
>>> >>>>>>>> r10 = 0xc3d69000
>>> >>>>>>>> exception_exit() at exception_exit
>>> >>>>>>>> pc = 0xc0dd711c lr = 0xc0d53828 (ffs_truncate+0xaa8)
>>> >>>>>>>> sp = 0xde0198e8 fp = 0xde019a20
>>> >>>>>>>> r0 = 0xd0238120 r1 = 0x00000e60
>>> >>>>>>>> r2 = 0x00000000 r3 = 0x00000000
>>> >>>>>>>> r4 = 0x00000120 r5 = 0x00000000
>>> >>>>>>>> r6 = 0xc3f3f6c0 r7 = 0x00001000
>>> >>>>>>>> r8 = 0xc443e880 r9 = 0x00000000
>>> >>>>>>>> r10 = 0xc3d69000 r12 = 0xd0238120
>>> >>>>>>>> memset() at memset+0x48
>>> >>>>>>>> pc = 0xc0de521c lr = 0xc0d53828 (ffs_truncate+0xaa8)
>>> >>>>>>>> sp = 0xde0198e8 fp = 0xde019a20
>>> >>>>>>>> Unwind failure (no registers changed)
>>> >>>>>>> No more beyond this? If you could run addr2line on 0xc0d53828 so
>>> >>>>>>> that we know where in ffs_truncate it's failing, that'd be very
>>> >>>>>>> nice...
>>> >>>>>> So I was trying to save the coredump in order to reboot and run
>>> >>>>>> addr2line, but that failed:
>>> >>>>>>
>>> >>>>>> Physical memory: 504 MB
>>> >>>>>> Dumping 67 MB:(da0:umass-sim0:0:0:0): WRITE(10). CDB: 2a 00 01 d5 1f20
>>> >>>>>> 00 00 01 00 <sip:2000000100>
>>> >>>>>> (da0:umass-sim0:0:0:0): CAM status: Resource Unavailable
>>> >>>>>> (da0:umass-sim0:0:0:0): Error 5, Retries exhausted
>>> >>>>>> Aborting dump due to I/O error.
>>> >>>>>>
>>> >>>>>> ** DUMP FAILED (ERROR 5) **
>>> >>>>>>
>>> >>>>>> So I guess this error is related to the CAM errors I'm getting from time
>>> >>>>>> to time. I was hoping that those errors were related to the INVARIANTS
>>> >>>>>> option that slowed down the system and thus might have triggered CAM
>>> >>>>>> errors, but obviously the SD Card seems to be the real issue here.
>>> >>>>>> So no crashdump for further analysis.
>>> >>>>> That's fine.. w/ the addr2line we have some lines to explore...
>>> >>>>>
>>> >>>>>> Interestingly the CAM errors didn't show up on the terminal as other
>>> >>>>>> times, the kernel just panicked straight away.
>>> >>>>> Hmm.. that is odd.. someone who knows the SD card layer should look
>>> >>>>> at this part... It could be that the SD card driver doesn't handle
>>> >>>>> dumping (there is this global flag that gets set) properly and the driver
>>> >>>>> needs to behave differently when it's set...
>>> >>>> I also need to grab a new SD card, just to make sure it's really not the
>>> >>>> card.
>>> >>>>
>>> >>>>>> But I've got the addr2line output, even though I'm not sure it makes any
>>> >>>>>> difference:
>>> >>>>>>
>>> >>>>>> addr2line -f -e /mnt/kernel.debug 0xc0d53828
>>> >>>>>>
>>> >>>>>> ffs_truncate
>>> >>>>>> /usr/devel/dreamplug/sys/ufs/ffs/ffs_inode.c:321
>>> >>>>> can you give me the contents of the line? and a few lines of context
>>> >>>>> around it? In HEAD's source, this is DOINGASYNC, and there is no call
>>> >>>>> to memset, nor a variable assignment that would result in memset being
>>> >>>>> called...
>>> >>>> Same here.. The file hasn't been changed in a while (Fri, 31 May 2013):
>>> >>>>
>>> >>>> ip->i_size = length;
>>> >>>> DIP_SET(ip, i_size, length);
>>> >>>> if (bp->b_bufsize == fs->fs_bsize)
>>> >>>> bp->b_flags |= B_CLUSTEROK;
>>> >>>> if (flags & IO_SYNC)
>>> >>>> bwrite(bp);
>>> >>>> 321: else if (DOINGASYNC(vp))
>>> >>>> bdwrite(bp);
>>> >>>> else
>>> >>>> bawrite(bp);
>>> >>>> ip->i_flag |= IN_CHANGE | IN_UPDATE;
>>> >>>> return (ffs_update(vp, !DOINGASYNC(vp)));
>>> >>>>
>>> >>>> No idea what's going on.
>>> >>> ok, could you send me the output of objdump -dSl, but you only need
>>> >>> to include the part from XXXXX <ffs_truncate>: to the next XXX<func>:
>>> >>> line... probably off list as it'll be quite long...
>>> >> I'm sorry, but given that I just broke all my working worlds using fsck,
>>> >> I'm not going to be able to do that until I'm back from holidays....
>>> >> currently working on the stuff remotely and after today's work day, I'm
>>> >> not going to be able to get my hands on the dreamplug.
>>> >>
>>> >>
>>> > BTW, for anyone playing with this problem, step one is to edit
>>> > your /etc/fstab and set the fsck pass number to 0 for all filesystems.
>>> > There's a risk of filesystem corruption after a crash, but it's smaller
>>> > than the 100% corruption rate of letting fsck run. :)
>>> >
>>> Of course! Great idea :-) Sometimes just can't think of the right tweak
>>> to save a lot of pain...
>>>
>>> Anyhow, I just found out, that I was rebooting the dreamplug from the sd
>>> card instead of the usb stick the whole time, and the usb stick hasn't
>>> been damaged enough by fsck, so it actually booted :-) I'll send the
>>> objdump soon.
>>
>> A (very) late update on this.... It looks like we may have tracked the
>> change that started all this down to the introduction of unmapped IO,
>> almost 2 years ago now. I still can't find the root cause, but I think
>> disabling unmapped IO on armv4/5 is a viable workaround, which Warner
>> committed this morning as r283014.
>>
>> --Ian
>
>
> This sounds promising for the use of my Sheevaplugs.
> I will try this soon. Thanks.
I plan on MFCing this change to 10 on Monday or Tuesday after fixing some
typos.
Warner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 842 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/freebsd-arm/attachments/20150517/727df2b0/attachment.sig>
More information about the freebsd-arm
mailing list