From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Chen, Kenneth W" Date: Fri, 13 Oct 2006 01:28:46 +0000 Subject: RE: [RFC] Variable Kernel Page size support Message-Id: <000701c6ee66$edb8e470$db34030a@amr.corp.intel.com> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org Christoph Lameter wrote on Thursday, October 12, 2006 5:33 PM > If set then a lookup is performed using the region7_pgdir table. > That table is segmented into 8 section for the varying page > sizes supported. > > 0 = _PAGE_SIZE_64K > 1 = _PAGE_SIZE_256K > 2 = _PAGE_SIZE_1M > 3 = _PAGE_SIZE_4M > 4 = _PAGE_SIZE_16M > 5 = _PAGE_SIZE_64M > 6 = _PAGE_SIZE_256M and _PAGE_SIZE_4K and _PAGE_SIZE_8K > 7 = _PAGE_SIZE_1G and PAGE_SIZE_16k Another thing cross my mind is that this is in region 7, and effectively vhpt is disabled for those sections of virtual address. I suppose the larger page size will win back the loss from not using vhpt. And this trade off also depends on data access pattern. Some of the small page size will take double hit on (1) being very small and (2) not having vhpt to mitigate tlb miss latency. You might want to just drop those small sizes.