On Wed, May 20, 2015 at 1:45 PM, PaX Team <pageexec@xxxxxxxxxxx> wrote: > >> Moreover, why is the resume code path the only one where freed pages need to >> be sanitized? > > ... i had a bug report before (http://marc.info/?l=linux-pm&m=132871433416256) > which is why i asked Anisse to figure this out before upstreaming the feature. > i've also asked him already to explain why his approach is the proper fix for > the problem (which should include the description of the root cause as a start) > but he hasn't answered that yet. > > anyway, the big question is how there can be free memory pages after resume > which are not sanitized. now i have no idea about the hibernation logic but > i assume that it doesn't save/restore free pages so the question is how the > kernel gets to learn about these free pages during resume and whether there's > a path where __free_page() or some other wrapper around free_pages_prepare() > doesn't get called at all. In my opinion the free pages left are those used by the loading kernel. If I understand correctly, a suspend (hibernate) image contains *all* the memory necessary for the OS to work; so when you restore it, you restore it all, page tables, and kernel code section included. So when the kernel does a hibernate restoration, it loads it all the pages into memory, then architecture-specific code will jump into the new "resumed" kernel by restoring page table entries and CPU context. When it does that, it leaves the "loader" kernel memory hanging; this memory is seen as free pages by the resumed kernel, but it isn't cleared. Rafael, am I getting something wrong on the hibernation resume process ? What do you think of this analysis ? Regards, Anisse -- 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>