On Wed, Feb 29, 2012 at 11:28:30AM +0000, bill4carson wrote:

On 2012?02?29? 18:32, Catalin Marinas wrote:

quoted

On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:

quoted

+static int __init add_huge_page_size(unsigned long long size)
+{
+	int shift = __ffs(size);
+	u32 mmfr3 = 0;
+
+	/* Check that it is a page size supported by the hardware and
+	 * that it fits within pagetable and slice limits. */
+	if (!is_power_of_2(size) || (shift != HPAGE_SHIFT))
+		return -EINVAL;

You could use get_order() instead of __ffs(), the latter just finds the
first bit set.

With all due respect, I'm afraid I can't agree with you on this.
here, we should use __ffs to return this "shift" not the order.

For "hugepagesz=2M", hpage_shift/HPAGE_SHIFT should be set to 21,
*not* the order 9(21-12), that's what HUGETLB_PAGE_ORDER for.

I agree (I got confused by get_order() and log2()).

-- 
Catalin