On Sat, 6 Mar 2021 00:54:32 +0100 Andrey Konovalov <andreyknvl@xxxxxxxxxx> wrote: > On Sat, Mar 6, 2021 at 12:50 AM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > On Sat, 6 Mar 2021 00:36:33 +0100 Andrey Konovalov <andreyknvl@xxxxxxxxxx> wrote: > > > > > Currently, kasan_free_nondeferred_pages()->kasan_free_pages() is called > > > after debug_pagealloc_unmap_pages(). This causes a crash when > > > debug_pagealloc is enabled, as HW_TAGS KASAN can't set tags on an > > > unmapped page. > > > > > > This patch puts kasan_free_nondeferred_pages() before > > > debug_pagealloc_unmap_pages() and arch_free_page(), which can also make > > > the page unavailable. > > > > > > ... > > > > > > --- a/mm/page_alloc.c > > > +++ b/mm/page_alloc.c > > > @@ -1304,6 +1304,12 @@ static __always_inline bool free_pages_prepare(struct page *page, > > > > > > kernel_poison_pages(page, 1 << order); > > > > > > + /* > > > + * With hardware tag-based KASAN, memory tags must be set before the > > > + * page becomes unavailable via debug_pagealloc or arch_free_page. > > > + */ > > > + kasan_free_nondeferred_pages(page, order, fpi_flags); > > > + > > > /* > > > * arch_free_page() can make the page's contents inaccessible. s390 > > > * does this. So nothing which can access the page's contents should > > > @@ -1313,8 +1319,6 @@ static __always_inline bool free_pages_prepare(struct page *page, > > > > > > debug_pagealloc_unmap_pages(page, 1 << order); > > > > > > - kasan_free_nondeferred_pages(page, order, fpi_flags); > > > - > > > return true; > > > } > > > > kasan_free_nondeferred_pages() has only two args in current mainline. > > Ah, yes, forgot to mention: this goes on top of: > > kasan: initialize shadow to TAG_INVALID for SW_TAGS > mm, kasan: don't poison boot memory with tag-based modes This patch (kasan, mm: fix crash with HW_TAGS and DEBUG_PAGEALLOC) is cc:stable, so it should come head of the other two lower priority patches. > > > > I fixed that in the obvious manner... > > Thanks! > > If you changed this patch, you'll also need to change the other one though. Yup.
