Re: [PATCH] dma-debug: Fix the overlap() function to be correct and readable

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> 
> 
> On Fri, 10 Jul 2009, Ingo Molnar wrote:
> > 
> > How about the patch below? Lightly tested.
> >  
> >  	if (!PageHighMem(page)) {
> > -		void *addr = ((char *)page_address(page)) + offset;
> > +		void *addr = (void *)page_address(page) + offset;
> > +
> 
> Why is that 'void *' cast there? page_address() is already a void *.
> 
> Other than that it obviously looks good to me. But I never see my 
> own bugs.

hm, indeed. I distinctly remember page_address() having been 
unsigned long ten years ago.

And yes, i still have a "highmem-2.2.21-A0" patch proving it:

--- linux/include/linux/pagemap.h.orig  Thu Oct  7 14:54:24 1999
+++ linux/include/linux/pagemap.h       Wed Oct 13 02:44:03 1999
@@ -11,12 +11,24 @@
 
 #include <linux/mm.h>
 #include <linux/fs.h>
+#include <linux/list.h>
 
-static inline unsigned long page_address(struct page * page)
+extern inline unsigned long FIXME_page_address(struct page * page)
 {
+       if (PageHIGHMEM(page))
+               BUG();
        return PAGE_OFFSET + ((page - mem_map) << PAGE_SHIFT);

Which, beyond being a rather embarrasing hunk (whose author i wont 
name voluntarily - i'll rather take the 5th), also shows that 
page_address() started out as an unsigned long.

Which got cleaned up for good once we added page->address, instead 
of the above direct calculation.

( Note to self: consider checking the types of core MM facilities 
  somewhat more frequently than every 10 years. )

	Ingo