Re: PATCH

[Pete Popov <ppopov@embeddedalley.com>]

Pete Popov wrote:
> Atsushi Nemoto wrote:
> 
>>>>>>> On Sun, 10 Oct 2004 17:47:20 -0700, Pete Popov 
>>>>>>> <ppopov@embeddedalley.com> said:
>>
>>
>>
>> ppopov> Clearly a buglet, carried over from 2.4. That section of the
>> ppopov> code wouldn't even be compiled, since CONFIG_MIPS64 is not
>> ppopov> defined. I'll remove that and send a new patch. Anything else
>> ppopov> you see that's suspicious :)?
>>
>> Hi.  I wonder why following change is needed.
>>
>>
>>> --- include/asm-mips/page.h    20 Aug 2004 12:02:18 -0000    1.44
>>> +++ include/asm-mips/page.h    19 Sep 2004 22:51:29 -0000
>>> @@ -32,7 +32,7 @@
>>> #ifdef CONFIG_PAGE_SIZE_64KB
>>> #define PAGE_SHIFT    16
>>> #endif
>>> -#define PAGE_SIZE    (1UL << PAGE_SHIFT)
>>> +#define PAGE_SIZE    (1L << PAGE_SHIFT)
>>> #define PAGE_MASK    (~(PAGE_SIZE-1))
>>>
>>> #ifdef __KERNEL__
> 
> 
> It was related to some compiler problem I mentioned to Ralf sometime 
> ago. Perhaps it's not needed anymore, I'll take a look.

Turns out the problem is with the #define PAGE_MASK, when (1UL << 
PAGE_SHIFT) is used for PAGE_SIZE. The correct fix is already in the PPC 
tree and it doesn't use PAGE_SIZE to define PAGE_MASK. From the PPC tree:

/*
  * Subtle: this is an int (not an unsigned long) and so it
  * gets extended to 64 bits the way [we] want (i.e. with 1s).  -- paulus
  */
#define PAGE_MASK       (~((1 << PAGE_SHIFT) - 1))

I updated the patch and put a new version in my directory (v1.3).  The 
diff is:

-#define PAGE_MASK      (~(PAGE_SIZE-1))
+#define PAGE_MASK       (~((1 << PAGE_SHIFT) - 1))

I'll work on the remap_pfn_range after those bits get to kernel.org. 
That will clean up the patch nicely.

Pete