Re: [PATCH][RELEASE] megaraid 2.10.2 Driver

[Matthew Wilcox <willy@debian.org>]

On Mon, Mar 22, 2004 at 07:02:39PM -0500, Bagalkote, Sreenivas wrote:
> >>For upstream, this should just be CONFIG_COMPAT I presume.
> 
> For 2.6 kernels, this would be just CONFIG_COMPAT. Or have I
> misunderstood your comment?
> 
> >>I don't see how this construct will work in all cases.  Hence my 
> >>CONFIG_COMPAT command above.
> 
> We saw the need for ioctl compatibility in __x86_64__ cases so far.
> What other cases will this not work in?

I don't think you understand how CONFIG_COMPAT works.  x86-64 defines it
when it wants it:

config COMPAT
        bool
        depends on IA32_EMULATION
        default y

just like every other architecture.  Just use
#ifdef CONFIG_COMPAT
	... 32bit compat code ...
#endif

and everything will be fine.  Please don't introduce this stupid
unnecessary LSI_CONFIG_COMPAT.  That just makes people say "what the
fuck are they doing?".

> >>Bug -- always set dma mask.  Do not conditionally _not_ call 
> >>pci_set_dma_mask(), for the 64-bit case.
> 
> The code does not __not__ call pci_set_dma_mask() conditionally.
> It is always calling with either 64-bit or 32-bit mask.
> 
> >>ummmm what???    uxferaddr is u32.  why are you casting it to a pointer?
> 
> Both copy_to_user and copy_from_user take pointers, don't they?

So you can only copy to the bottom 4GB of user address space?  That
seems like a recipe for disaster.  Particularly on ia64.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain