Re: [PATCH 3/5] qlge: bugfix: Fix endian issue regarding shadow registers.

[Harvey Harrison <harvey.harrison@gmail.com>]

On Wed, 2008-12-24 at 16:15 -0500, Christoph Hellwig wrote:
> On Wed, Dec 24, 2008 at 06:35:31PM +0000, Ben Hutchings wrote:
> > > -static inline unsigned int ql_read_sh_reg(const volatile void  *addr)
> > > +static inline u32 ql_read_sh_reg(const volatile void  *addr)
> > >  {
> > > -	return *(volatile unsigned int __force *)addr;
> > > +	return le32_to_cpu(*(volatile unsigned int __force *)addr);
> > 
> > I think that should be:
> > 	???return le32_to_cpu(*(const volatile __le32 *)addr);
> 
> No.  It needs more love than that.
> 
> ql_read_sh_reg is only used to access .prod_idx_sh_reg in the rx_ring
> structure, which is an offset from the .rx_ring_shadow_reg_area member.
> So instead of all this werid casting .prod_idx_sh_reg should simply
> be a __le32 pointer, whithout volatile and the other mess, and then
> you can use le32_to_cpu directly and kill the helper.
> 
> And please verify the whole driver for endianess issues using sparse
> while you're at it...

I've already deleted the patches from my inbox, but if you send them
again off-list I'll have a look as well.

Harvey