x86/mm/mtrr: Fix alignment determination in range_to_mtrr()

With the variable operated on being of "unsigned long" type,
neither ffs() nor fls() are suitable to use on them, as those
truncate their arguments to 32 bits. Using __ffs() and __fls()
respectively at once eliminates the need to subtract 1 from their
results.

Additionally, with the alignment value subsequently used as a
shift count, it must be enforced to be less than BITS_PER_LONG
(and on 64-bit there's no need for it to be any smaller).

Signed-off-by: Jan Beulich <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Yinghai Lu <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

arch/x86/kernel/cpu/mtrr/cleanup.c

@@ -258,11 +258,11 @@ range_to_mtrr(unsigned int reg, unsigned long range_startk,
 
 		/* Compute the maximum size with which we can make a range: */
 		if (range_startk)
-			max_align = ffs(range_startk) - 1;
+			max_align = __ffs(range_startk);
 		else
-			max_align = 32;
+			max_align = BITS_PER_LONG - 1;
 
-		align = fls(range_sizek) - 1;
+		align = __fls(range_sizek);
 		if (align > max_align)
 			align = max_align;