ext3 oops on shutdown

ext3 oops on shutdown

[David Miller]

I just started to see the following ext3 umount() oops on
shutdown, it happens every shutdown, has anyone else seen
it?

I'll start bisecting soon... this was on sparc64.

free_block()+0x50/0x190
cache_flusharray()+0x74/0xa0
kmem_cache_free()
journal_destroy_revoke()
journal_destroy()
ext3_put_super()
generic_shutdown_super()
kill_block_super()
deactivate_super()
sys_umount()

Re: ext3 oops on shutdown

[David Miller]

From: David Miller <davem@davemloft.net>
Date: Tue, 24 Oct 2006 17:57:51 -0700 (PDT)

> I just started to see the following ext3 umount() oops on
> shutdown, it happens every shutdown, has anyone else seen
> it?
> 
> I'll start bisecting soon... this was on sparc64.
> 
> free_block()+0x50/0x190
> cache_flusharray()+0x74/0xa0
> kmem_cache_free()
> journal_destroy_revoke()
> journal_destroy()
> ext3_put_super()
> generic_shutdown_super()
> kill_block_super()
> deactivate_super()
> sys_umount()

Just as an FYI, I bisected it down to this commit:

commit 6b4b78fed47e7380dfe9280b154e8b9bfcd4c86c
Author: Matt Domsch <Matt_Domsch@dell.com>
Date:   Fri Sep 29 15:23:23 2006 -0500

    PCI: optionally sort device lists breadth-first

which makes no sense.  Effectively all this changeset does is add some
__init section code to the PCI layer which never gets executed on
non-i386.  And even on i386 it only runs if you give a special kernel
command line option.

So the only side effect of this code for sparc64, by checking the
System.map before/after this change, is to move the __init_end ahead
by 1 PAGE_SIZE.

It's probably some sparc64 issue, maybe even some error wrt.  freeing
up init memory.  I'll try to track it down.  I'm actually quite
curious what this bug is :-)

Re: ext3 oops on shutdown

[David Miller]

From: David Miller <davem@davemloft.net>
Date: Wed, 25 Oct 2006 00:51:34 -0700 (PDT)

> It's probably some sparc64 issue, maybe even some error wrt.  freeing
> up init memory.  I'll try to track it down.  I'm actually quite
> curious what this bug is :-)

Yep, a sparc64 bug, I'll push this to Linus shortly...

commit 2506be0657c7bd1befdb616fb0e86d87c6a288cd
Author: David S. Miller <davem@sunset.davemloft.net>
Date:   Wed Oct 25 22:33:07 2006 -0700

    [SPARC64]: Fix memory corruption in pci_4u_free_consistent().
    
    The second argument to free_npages() was being incorrectly
    calculated, which would thus access far past the end of the
    arena->map[] bitmap.
    
    Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/arch/sparc64/kernel/pci_iommu.c b/arch/sparc64/kernel/pci_iommu.c
index 82e5455..2e7f142 100644
--- a/arch/sparc64/kernel/pci_iommu.c
+++ b/arch/sparc64/kernel/pci_iommu.c
@@ -281,7 +281,7 @@ static void pci_4u_free_consistent(struc
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	free_npages(iommu, dvma, npages);
+	free_npages(iommu, dvma - iommu->page_table_map_base, npages);
 
 	spin_unlock_irqrestore(&iommu->lock, flags);