Re: [patch 2/2] backport of sba sg list management to ccio-dma

On Sun, Oct 28, 2007 at 02:45:24PM +0000, Joel Soete wrote:
...
> e.g.
> # more GetOrder-poc.c
...
> give me well same results:
> # ./GetOrder-poc
> get_order(1) = 0 (0x0);
> get_order_poc(1) = 0 (0x0).
...

Good!

...
> But I need to check more on a detail:
> here is the prototype of fls():
> static inline int fls(int x)
>
> so I need to be sure that the value of ((size - 1) >> (PAGE_SHIFT)) would 
> always stand in a int (not unsigned but sign) for 32bit (iirc 
> size_of(unsigned long)==32) and 64bit (size_of(unsigned long)==64) 
> implementations?

Huh?
Why do you care how (unsigned long) is handled when the function prototype
defines the API to pass in an "int"?

FWIW, __fls could be extended to handle unsigned long the same way
that __ffs has been...but I don't think we needed to for anything so far.

...
>>> 	return (size ? fls((size - 1) >> (PAGE_SHIFT)): 0);
>> fls is inlined already and does the same test for size.
>
> Yes but here the error (well imo it's an error) came from the unsign long 
> of (-1 >> PAGE_SHIFT) <> 0

Yeah, we'll have to make sure signed int values are handled correctly too.
But something specific to the parisc version of get_order() can deal
with sign bits and zero values then call __fls(unsigned long) instead
of fls(signed int).

grant