On Mon, Jun 08, 2026 at 04:38:00AM -0400, Michael S. Tsirkin wrote: > Replace user_alloc_needs_zeroing() with the direct aliasing checks > (cpu_dcache_is_aliasing() || cpu_icache_is_aliasing()) in the > post_alloc_hook aliasing guard. > > user_alloc_needs_zeroing() includes a !init_on_alloc term that > means "allocator didn't zero this page." But in this guard's > context (!zeroed && !init && __GFP_ZERO), we already know the page > is zero; init incorporates init_on_alloc via want_init_on_alloc(). > The only question left is whether the cache architecture needs > the data re-zeroed through a congruent mapping, which is purely > cpu_dcache_is_aliasing() || cpu_icache_is_aliasing(). > > On non-aliasing architectures with init_on_free=true and > init_on_alloc=false, this avoids a redundant re-zero of an > already-zero page. > > Note on PowerPC: PowerPC overrides clear_user_page to call > flush_dcache_page after clear_page, but on freshly allocated > pages PG_dcache_clean is already clear (cleared by > __free_pages_prepare), so flush_dcache_page is a no-op. > Skipping this here thus has no effect. > > Signed-off-by: Michael S. Tsirkin <[email protected]> > Assisted-by: Claude:claude-opus-4-6
This seems like an odd ordering of patches, can we group like changes together? > --- > mm/page_alloc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 45e824b1ec75..edfc83571985 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -1880,7 +1880,7 @@ inline void post_alloc_hook(struct page *page, unsigned > int order, > */ > if (!zeroed && !init && (gfp_flags & __GFP_ZERO) && > user_addr != USER_ADDR_NONE && > - user_alloc_needs_zeroing()) > + (cpu_dcache_is_aliasing() || cpu_icache_is_aliasing())) Let's try and simplify things rather than adding endlessly huge if conditionals? It's now incredibly hard to track exactly what's going on here, and that is bug-bait. > init = true; > /* > * If memory is still not initialized, initialize it now. > -- > MST > Thanks, Lorenzo
