On Wed, Feb 11, 2015 at 02:09:15PM -0800, Andrew Morton wrote: > On Wed, 14 Jan 2015 01:08:40 +0000 Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> wrote: > > > On Tue, Jan 13, 2015 at 05:57:04PM +0530, Shiraz Hashim wrote: > > > pagemap_read scans through the virtual address space of a > > > task till it prepares 'count' pagemaps or it reaches end > > > of task. > > > > > > This presents a problem when the page walk doesn't happen > > > for vma with VM_PFNMAP set. In which case walk is silently > > > skipped and no pagemap is prepare, in turn making > > > pagemap_read to scan through task end, even crossing beyond > > > 'count', landing into a different vma region. This leads to > > > wrong presentation of mappings for that vma. > > > > > > Fix this by limiting end_vaddr to the end of the virtual > > > address region being scanned. > > > > > > Signed-off-by: Shiraz Hashim <shashim@xxxxxxxxxxxxxx> > > > > This patch works in some case, but there still seems a problem in another case. > > > > Consider that we have two vmas within some narrow (PAGEMAP_WALK_SIZE) region. > > One vma in lower address is VM_PFNMAP, and the other vma in higher address is not. > > Then a single call of walk_page_range() skips the first vma and scans the > > second vma, but the pagemap record of the second vma will be stored on the > > wrong offset in the buffer, because we just skip vma(VM_PFNMAP) without calling > > any callbacks (within which add_to_pagemap() increments pm.pos). > > > > So calling pte_hole() for vma(VM_PFNMAP) looks a better fix to me. > > > > Can we get this finished off? ASAP, please. Yes, I think so. The patch "mm: pagewalk: call pte_hole() for VM_PFNMAP during walk_page_range" (replacing this patch) is already in mainline. Thanks, Naoya Horiguchi -- 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