Alignment of bitmaps for ext2_set_bit et al.

Alignment of bitmaps for ext2_set_bit et al.

[Paul Mackerras]

What can we assume about the alignment of the bitmap pointer passed to
the ext2_{set,clear}_bit_atomic functions?  Can we assume that they
will be aligned to an long boundary (8 bytes on 64-bit)?

Olof has made a patch that uses atomics for these on ppc64 rather than
locking and unlocking a lock, but it will only work correctly if the
bitmap is always 8-byte aligned.

Paul.

Re: Alignment of bitmaps for ext2_set_bit et al.

[Andrew Morton]

Paul Mackerras <paulus@samba.org> wrote:
>
> What can we assume about the alignment of the bitmap pointer passed to
> the ext2_{set,clear}_bit_atomic functions?  Can we assume that they
> will be aligned to an long boundary (8 bytes on 64-bit)?

For ext2 and ext3 it's safe to assume that they are 1024-byte aligned.

But it's possible that other filesystems are using these (awfully named)
functions.

Well.  I doubt if anyone is using the _atomic versions yet.

> Olof has made a patch that uses atomics for these on ppc64 rather than
> locking and unlocking a lock, but it will only work correctly if the
> bitmap is always 8-byte aligned.

Sounds sane, as long as you get firmly notified when a poorly-aligned
address is fed in.

Re: Alignment of bitmaps for ext2_set_bit et al.

[David S. Miller]

On Thu, 19 Aug 2004 04:22:34 -0700
Andrew Morton <akpm@osdl.org> wrote:

> Paul Mackerras <paulus@samba.org> wrote:
> >
> > What can we assume about the alignment of the bitmap pointer passed to
> > the ext2_{set,clear}_bit_atomic functions?  Can we assume that they
> > will be aligned to an long boundary (8 bytes on 64-bit)?
> 
> For ext2 and ext3 it's safe to assume that they are 1024-byte aligned.
> 
> But it's possible that other filesystems are using these (awfully named)
> functions.

FWIW, I've been using 4-byte word atomics for those routines
on sparc64 and have never gotten an unaligned trap for it yet.

Re: Alignment of bitmaps for ext2_set_bit et al.

[Olof Johansson]

Andrew Morton wrote:

>Paul Mackerras <paulus@samba.org> wrote:
>  
>
>>Olof has made a patch that uses atomics for these on ppc64 rather than
>>locking and unlocking a lock, but it will only work correctly if the
>>bitmap is always 8-byte aligned.
>>    
>>
>
>Sounds sane, as long as you get firmly notified when a poorly-aligned
>address is fed in.
>  
>
The notification is in the form of an unalignment trap resulting in a 
panic, is that firm enough? :-)


-Olof