Re: [PATCH v2] mm/memory: drop PROCESS_PAGES_NON_PREEMPT_BATCH

[Pedro Falcato <pfalcato@suse.de>]

On Mon, Nov 09, 2026 at 06:51:07AM +0000, Shivank Garg wrote:
> clear_contig_highpages() clears the whole range at once under
> preemptible models. Otherwise it clears PROCESS_PAGES_NON_PREEMPT_BATCH
> pages at a time, with a cond_resched() in between. The batch is 32MB if
> the architecture defines clear_pages(), and one page otherwise.
> 
> Since commit 7dadeaa6e851 ("sched: Further restrict the preemption
> modes"), architectures that support lazy preemption only offer the
> full, lazy and RT preemption models, also with PREEMPT_DYNAMIC. x86 is
> one of them, and it is the only architecture that defines
> clear_pages(). So the 32MB batch is never used: whenever
> clear_contig_highpages() runs under a cooperative model, the batch is
> one page.
> 
> Make that explicit. Clear the whole range when CONFIG_PREEMPTION is
> set, and one page at a time with cond_resched() in between otherwise.
> Drop PROCESS_PAGES_NON_PREEMPT_BATCH, which has no other users.
> 
> PREEMPT_DYNAMIC kernels also have CONFIG_PREEMPTION set, and
> cond_resched() does nothing on any of these kernels. So a build-time
> CONFIG_PREEMPTION check is enough, and preempt_model_preemptible() is
> not needed.
> 
> No functional change intended.
> 
> Suggested-by: David Hildenbrand <david@kernel.org>
> Signed-off-by: Shivank Garg <shivankg@amd.com>

Reviewed-by: Pedro Falcato <pfalcato@suse.de>

with a question below.

> ---
> This came up in review of the copy_mc_highpages() series [1]. Pedro
> pointed out that most architecture run with full or lazy preempt, so mm
> does not need to check the preemption model at runtime or batch the copy
> for cond_resched(), and a simple per-page fallback is enough for
> non-preemptible models.
> David suggested doing the same for the zeroing path, and this patch
> does that.
> 
> [1] https://lore.kernel.org/all/20261019-batpagecopy-v2-1-d75e7d0dfaae@amd.com 
> ---
> Changes in v2:
> - Rebased on mm-next/for-next, and drop RFC tag.
> - Link to v1: https://lore.kernel.org/r/20261022-clear-pages-batch-v1-1-13d3bd0b876e@amd.com
> ---
>  include/linux/mm.h | 28 +---------------------------
>  mm/memory.c        | 29 ++++++++++++++++-------------
>  2 files changed, 17 insertions(+), 40 deletions(-)
> 
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index d89e6f823c9f..59bc3685aac5 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -5242,7 +5242,7 @@ static inline void clear_page_guard(struct zone *zone, struct page *page,
>   *
>   * When running under preemptible models this is not a problem. Under
>   * cooperatively scheduled models, however, the caller is expected to
> - * limit @npages to no more than PROCESS_PAGES_NON_PREEMPT_BATCH.
> + * clear one page at a time and call cond_resched() in between.
>   */
>  static inline void clear_pages(void *addr, unsigned int npages)
>  {
> @@ -5253,32 +5253,6 @@ static inline void clear_pages(void *addr, unsigned int npages)
>  }
>  #endif
>  
> -#ifndef PROCESS_PAGES_NON_PREEMPT_BATCH
> -#ifdef clear_pages
> -/*
> - * The architecture defines clear_pages(), and we assume that it is
> - * generally "fast". So choose a batch size large enough to allow the processor
> - * headroom for optimizing the operation and yet small enough that we see
> - * reasonable preemption latency for when this optimization is not possible
> - * (ex. slow microarchitectures, memory bandwidth saturation.)
> - *
> - * With a value of 32MB and assuming a memory bandwidth of ~10GBps, this should
> - * result in worst case preemption latency of around 3ms when clearing pages.
> - *
> - * (See comment above clear_pages() for why preemption latency is a concern
> - * here.)
> - */
> -#define PROCESS_PAGES_NON_PREEMPT_BATCH		(SZ_32M >> PAGE_SHIFT)
> -#else /* !clear_pages */
> -/*
> - * The architecture does not provide a clear_pages() implementation. Assume
> - * that clear_page() -- which clear_pages() will fallback to -- is relatively
> - * slow and choose a small value for PROCESS_PAGES_NON_PREEMPT_BATCH.
> - */
> -#define PROCESS_PAGES_NON_PREEMPT_BATCH		1
> -#endif
> -#endif
> -
>  #ifdef __HAVE_ARCH_GATE_AREA
>  extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm);
>  extern int in_gate_area_no_mm(unsigned long addr);
> diff --git a/mm/memory.c b/mm/memory.c
> index c46482e35418..9bfd49ea990f 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -7612,27 +7612,30 @@ static inline int process_huge_page(
>  }
>  
>  static void clear_contig_highpages(struct page *page, unsigned long addr,
> -				   unsigned int nr_pages)
> +		unsigned int nr_pages)
>  {
> -	unsigned int i, count;
> +	unsigned int i;
> +
> +	might_sleep();
> +
>  	/*
>  	 * When clearing we want to operate on the largest extent possible to
>  	 * allow for architecture specific extent based optimizations.
>  	 *
> -	 * However, since clear_user_highpages() (and primitives clear_user_pages(),
> -	 * clear_pages()), do not call cond_resched(), limit the unit size when
> -	 * running under non-preemptible scheduling models.
> +	 * clear_user_highpages() (and primitives clear_user_pages(),
> +	 * clear_pages()) do not call cond_resched(). That is fine on
> +	 * preemptible kernels, where the scheduler can preempt the clearing
> +	 * at any point, so clear the whole range at once. Otherwise, clear
> +	 * one page at a time and reschedule in between.
>  	 */
> -	const unsigned int unit = preempt_model_preemptible() ?
> -				   nr_pages : PROCESS_PAGES_NON_PREEMPT_BATCH;
> -
> -	might_sleep();
> +	if (IS_ENABLED(CONFIG_PREEMPTION)) {
> +		clear_user_highpages(page, addr, nr_pages);
> +		return;
> +	}
>  
> -	for (i = 0; i < nr_pages; i += count) {
> +	for (i = 0; i < nr_pages; i++) {
>  		cond_resched();
> -
> -		count = min(unit, nr_pages - i);
> -		clear_user_highpages(page + i, addr + i * PAGE_SIZE, count);
> +		clear_user_highpages(page + i, addr + i * PAGE_SIZE, 1);

I think we can use clear_user_highpage() here. I don't know if that's
necessarily desirable, but it would make this a little simpler.

-- 
Pedro