Re: [PATCH 0/3] map multiple blocks in get_block() and mpage_readpages()

[Nathan Scott <nathans@sgi.com>]

On Tue, Feb 21, 2006 at 08:03:54AM -0800, Badari Pulavarty wrote:
> ...
> Thinking more about it, currently (without my patches) only DIO
> code can request for large chunks of mapping through get_blocks().
> Since b_size is only u32, you can only map 4GB max. Isn't it ?
> With my patches, now mpage_readpages() also can request large
> chunks. (through readahead). So, my patches are not adding any
> extra limitation. Its carrying the same existing limitation. 

The DIO get_blocks is super-freaky at the moment, in that it passes
in an unsigned long max_blocks (units of blocks) but it expects the
filesystem to fill in a u32 b_size (in bytes).  This does at least
give the filesystem the hint that this was a monster write and it
gets the chance to allocate contiguously, but that was probably
unintentional, who knows?  I guess that subtlety would be dropped
with this change if b_size isn't made a 64 bit entity.

> In order to handle larger chunks of disk mapping, changing b_size 
> to u64 is required and we should request for it irrespective of my
> patches.

*shrug*... its all very closely inter-related, I'd keep it together.
I'd also thought that size_t was the right type here, not u64?  32
bit platforms aren't capable of submitting IOs this large, so they
needn't be doing any 64 bit manipulation here, no?

cheers.

-- 
Nathan