Re: Segfault fix in scsi_id-0.6

[Andrew Patterson <andrew.patterson@hp.com>]

[-- Attachment #1: Type: text/plain, Size: 1397 bytes --]

On Wed, 2004-09-29 at 15:03 -0400, Dave Dodge wrote:
> On Tue, Sep 28, 2004 at 05:45:39PM -0700, Patrick Mansfield wrote:
> > On Tue, Sep 28, 2004 at 03:33:59PM -0600, Andrew Patterson wrote:
> 
> > > +       serial = (char*) (((unaligned_buf - (char *)0) + (ALIGN - 1)) &
> > > ~(ALIGN - 1));
> 
> Quick comment: the C compiler will treat the (char*)0 as a null
> pointer constant.  C makes no guarantee that the resulting value will
> have anything to do with address 0.  I believe there are architectures
> where a null pointer has all-bits-on or some other non-zero
> representation, though I don't know if Linux has ever been made to run
> on any of them.
> 
> > But how about just casting to unsigned long instead? Like:
> 
> > +	serial = (char*) (((unsigned long) unaligned_buf + (ALIGN - 1))
> > +			  & ~(ALIGN - 1));
> 

This works on i386 and ia64.  As Dave points out, it may not work on
architectures where a pointer size != unsigned long.

Andrew Patterson

> None of this pointer manipulation is technically defined to work, but
> from a practical standpoint this will probably be safe.  There's still
> the possibility that an unsigned long can't hold all pointer values.
> If you're willing to rely on a C99 construct, you could include
> <stdint.h> and use uintptr_t instead.
> 
>                                                   -Dave Dodge
> 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]