Re: [PATCH] change b_size to size_t

[Badari Pulavarty <pbadari@gmail.com>]

On Thu, 2006-02-23 at 09:28 -0800, Badari Pulavarty wrote:
> On Thu, 2006-02-23 at 11:32 -0500, Benjamin LaHaise wrote:
> > On Thu, Feb 23, 2006 at 08:28:12AM -0800, Badari Pulavarty wrote:
> > > Here is the updated version of the patch, which changes
> > > buffer_head.b_size to size_t to support mapping large
> > > amount of disk blocks (for large IOs).
> > 
> > Your patch doesn't seem to be inline, so I can't quote it.  Several 
> > problems: on 64 bit platforms you introduced 4 bytes of padding into 
> > buffer_head.  atomic_t only takes up 4 byte, while size_t is 8 byte 
> > aligned. 
> 
> 
> Ignore my previous mail.
> 
> How about doing this ? Change b_state to u32 and change b_size
> to "size_t". This way, we don't increase the overall size of
> the structure on 64-bit machines. Isn't it ?

I hate to correct myself again. But this won't work either.
If we do this, we can use bit_spin_lock() helpers any more
to do bh_state manipulation :(

Yes. Bottom line is, we would increase the size of the structure
by 8-bytes on 64-bit machines. I don't see any way out of it.
But this would provide ability to let the filesystems know
that the we are dealing with large (> 4GB) of IOs (may be they
can allocated as much as possible contiguously), even if
we don't really do that big IOs.

Thanks,
Badari