On Tue, Aug 18, 2015 at 05:18:16PM +0200, Oleg Nesterov wrote:
> On 08/18, Oleg Nesterov wrote:
> >
> > When I tried to run all tests, I
> > got the new reports from lockdep.
>
> Just in case... when I run all tests I see misc failures (with or without
> the changes above) which I didn't try to interpret. In particular xfs/073
> just hangs, "shutdown -r" doesn't work, the serial console continues to
> print
>
> [10784.605139] XFS (loop2): metadata I/O error: block 0x7d4948
> ("xfs_buf_iodone_callbacks") error 5 numblks 8
EIO from the loop device.
> [10784.605207] loop: Write error at byte offset 26843578368, length
> 4096.
> [10784.605222] loop: Write error at byte offset 26843578368, length
> 4096.
> [10784.605235] loop: Write error at byte offset 26843578368, length
> 4096.
> [10784.605248] loop: Write error at byte offset 26843578368, length
> 4096.
> [10784.605261] loop: Write error at byte offset 26843578368, length
> 4096.
That makes me think the underlying filesystem (i.e. the fs the loop
image file is hosted on) ran out of space. Everything will go to
shit if that happens.
> again and again.
>
> Plus some tests need a lot of time (for example generic/127 more than 70
> minutes!),
Yup, that one only does about 50,000 synchronous writes via fsx.
If you want to run fast, use a virtual machine with 10GB RAM and use
a pair of 4GB ramdisks as the storage.
Or expunge the long running tests (look up the -X option) so they
are skipped.
> Also, tests/generic/078 on 4.2.0-rc6 (without other changes) triggered
>
> [ 2100.404545] BUG: looking up invalid subclass: 8
> [ 2100.409600] turning off the locking correctness validator.
That's another lockdep annotation limitation I've had to work
around. The RENAME_WHITEOUT changes caused that by introducing the
need to lock 5 inodes at the same time (src/dst dir, src/dst file,
whiteout inode) instead of only 4.
> perhaps this was fixed by Dave's ILOCK patches.
It greatly complicated those patches, you mean?
But, yes, those patches should have fixed it - I had to shoehorn
about 15 different XFS inode locking subclasses into the 8
subclasses that lockdep supports. And I used all 8 subclasses for
the ILOCK annotations, so if theres some other annotation we need to
add (e.g more than one level of ILOCK_PARENT lock nesting) then,
well, we will need at least 16 subclasses to annotate everything.
I hate lockdep.
Cheers,
Dave.
--
Dave Chinner
da...@fromorbit.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
