On 01/15/2013 11:46 AM, H. Peter Anvin wrote:
> I object to this switch statement. If we are going to create new
> primitives, let's create a primitive that embody this and put it in
> pgtypes_types.h, especially since it is simply an algorithmic operation:
Yeah, that's a good point. I did at least copy part of the switch from
elsewhere in the file, so there's certainly room for consolidating some
things.
> static inline unsigned long page_level_size(int level)
> {
> return (PAGE_SIZE/PGDIR_SIZE) << (PGDIR_SHIFT*level);
> }
> static inline unsigned long page_level_shift(int level)
> {
> return (PAGE_SHIFT-PGDIR_SHIFT) + (PGDIR_SHIFT*level);
> }
(PAGE_SHIFT-PGDIR_SHIFT) == -27, so this can't possibly work, right?
How about something like this?
/*
* Note: this only holds true for pagetable levels where PTEs can be
* present. It would break if you used it on the PGD level where PAE
* is in use. It basically assumes that the shift between _all_
* adjacent levels of the pagetables are the same as the lowest-level
* shift.
*/
#define PG_SHIFT_PER_LEVEL (PMD_SHIFT-PAGE_SHIFT)
static inline unsigned long page_level_shift(int level)
{
return PAGE_SHIFT + (level - PG_LEVEL_4K) * PG_SHIFT_PER_LEVEL;
}
static inline unsigned long page_level_size(int level)
{
return 1 << page_level_shift(level);
}
The generated code for page_level_size() looks pretty good, despite it
depending on page_level_shift(), so we might as well leave it defined
this way for simplicity:
0000000000400610 <plsize>:
400610: 8d 7c bf fb lea -0x5(%rdi,%rdi,4),%edi
400614: b8 01 00 00 00 mov $0x1,%eax
400619: 8d 4c 3f 0c lea 0xc(%rdi,%rdi,1),%ecx
40061d: d3 e0 shl %cl,%eax
40061f: c3 retq
I'll send out another series doing this.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>