On 10/09/2013 10:40 PM, Seth Jennings wrote: > On Tue, Oct 08, 2013 at 01:08:53PM -0700, Andrew Morton wrote: >> On Tue, 08 Oct 2013 10:13:20 +0200 Krzysztof Kozlowski <k.kozlowski@xxxxxxxxxxx> wrote: >> >>> On pon, 2013-10-07 at 15:03 -0700, Andrew Morton wrote: >>>> On Mon, 07 Oct 2013 17:25:41 +0200 Krzysztof Kozlowski <k.kozlowski@xxxxxxxxxxx> wrote: >>>> >>>>> During swapoff the frontswap_map was NULL-ified before calling >>>>> frontswap_invalidate_area(). However the frontswap_invalidate_area() >>>>> exits early if frontswap_map is NULL. Invalidate was never called during >>>>> swapoff. >>>>> >>>>> This patch moves frontswap_map_set() in swapoff just after calling >>>>> frontswap_invalidate_area() so outside of locks >>>>> (swap_lock and swap_info_struct->lock). This shouldn't be a problem as >>>>> during swapon the frontswap_map_set() is called also outside of any >>>>> locks. >>>>> >>>> >>>> Ahem. So there's a bunch of code in __frontswap_invalidate_area() >>>> which hasn't ever been executed and nobody noticed it. So perhaps that >>>> code isn't actually needed? >>>> >>>> More seriously, this patch looks like it enables code which hasn't been >>>> used or tested before. How well tested was this? >>>> >>>> Are there any runtime-visible effects from this change? >>> >>> I tested zswap on x86 and x86-64 and there was no difference. This is >>> good as there shouldn't be visible anything because swapoff is unusing >>> all pages anyway: >>> try_to_unuse(type, false, 0); /* force all pages to be unused */ >>> >>> I haven't tested other frontswap users. >> >> So is that code in __frontswap_invalidate_area() unneeded? > > Yes, to expand on what Bob said, __frontswap_invalidate_area() is still > needed to let any frontswap backend free per-swaptype resources. > > __frontswap_invalidate_area() is _not_ for freeing structures associated > with individual swapped out pages since all of the pages should be > brought back into memory by try_to_unuse() before > __frontswap_invalidate_area() is called. > > The reason we never noticed this for zswap is that zswap has no > dynamically allocated per-type resources. In the expected case, > where all of the pages have been drained from zswap, > zswap_frontswap_invalidate_area() is a no-op. > Not exactly, see the bug fix "mm/zswap: bugfix: memory leak when re-swapon" from Weijie. Zswap needs invalidate_area() also. Thanks, -Bob -- 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>