Re: [PATCH v3 2/3] mm: disable LRU pagevec during the migration temporarily

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 10 Mar 2021 08:14:28 -0800 Minchan Kim <minchan@kernel.org> wrote:

> LRU pagevec holds refcount of pages until the pagevec are drained.
> It could prevent migration since the refcount of the page is greater
> than the expection in migration logic. To mitigate the issue,
> callers of migrate_pages drains LRU pagevec via migrate_prep or
> lru_add_drain_all before migrate_pages call.
> 
> However, it's not enough because pages coming into pagevec after the
> draining call still could stay at the pagevec so it could keep
> preventing page migration. Since some callers of migrate_pages have
> retrial logic with LRU draining, the page would migrate at next trail
> but it is still fragile in that it doesn't close the fundamental race
> between upcoming LRU pages into pagvec and migration so the migration
> failure could cause contiguous memory allocation failure in the end.
> 
> To close the race, this patch disables lru caches(i.e, pagevec)
> during ongoing migration until migrate is done.
> 
> Since it's really hard to reproduce, I measured how many times
> migrate_pages retried with force mode(it is about a fallback to a
> sync migration) with below debug code.
> 
> int migrate_pages(struct list_head *from, new_page_t get_new_page,
> 			..
> 			..
> 
> if (rc && reason == MR_CONTIG_RANGE && pass > 2) {
>        printk(KERN_ERR, "pfn 0x%lx reason %d\n", page_to_pfn(page), rc);
>        dump_page(page, "fail to migrate");
> }
> 
> The test was repeating android apps launching with cma allocation
> in background every five seconds. Total cma allocation count was
> about 500 during the testing. With this patch, the dump_page count
> was reduced from 400 to 30.
> 
> The new interface is also useful for memory hotplug which currently
> drains lru pcp caches after each migration failure. This is rather
> suboptimal as it has to disrupt others running during the operation.
> With the new interface the operation happens only once. This is also in
> line with pcp allocator cache which are disabled for the offlining as
> well.
> 

This is really a rather ugly thing, particularly from a maintainability
point of view.  Are you sure you found all the sites which need the
enable/disable?  How do we prevent new ones from creeping in which need
the same treatment?  Is there some way of adding a runtime check which
will trip if a conversion was missed?

> ...
>
> +bool lru_cache_disabled(void)
> +{
> +	return atomic_read(&lru_disable_count);
> +}
> +
> +void lru_cache_enable(void)
> +{
> +	atomic_dec(&lru_disable_count);
> +}
> +
> +/*
> + * lru_cache_disable() needs to be called before we start compiling
> + * a list of pages to be migrated using isolate_lru_page().
> + * It drains pages on LRU cache and then disable on all cpus until
> + * lru_cache_enable is called.
> + *
> + * Must be paired with a call to lru_cache_enable().
> + */
> +void lru_cache_disable(void)
> +{
> +	atomic_inc(&lru_disable_count);
> +#ifdef CONFIG_SMP
> +	/*
> +	 * lru_add_drain_all in the force mode will schedule draining on
> +	 * all online CPUs so any calls of lru_cache_disabled wrapped by
> +	 * local_lock or preemption disabled would be ordered by that.
> +	 * The atomic operation doesn't need to have stronger ordering
> +	 * requirements because that is enforeced by the scheduling
> +	 * guarantees.
> +	 */
> +	__lru_add_drain_all(true);
> +#else
> +	lru_add_drain();
> +#endif
> +}

I guess at least the first two of these functions should be inlined.