Message-ID: <0162E372-2B68-494D-8DFA-F3AF0EC6F5E8@nvidia.com>
Date: Mon, 09 Dec 2024 16:42:21 -0500
From: Zi Yan <ziy@...dia.com>
To: David Hildenbrand <david@...hat.com>, Vlastimil Babka <vbabka@...e.cz>
Cc: linux-kernel@...r.kernel.org, Johannes Weiner <hannes@...xchg.org>,
 linux-mm@...ck.org, Andrew Morton <akpm@...ux-foundation.org>,
 Yu Zhao <yuzhao@...gle.com>
Subject: Re: [PATCH v1 1/2] mm/page_alloc: conditionally split >
 pageblock_order pages in free_one_page() and move_freepages_block_isolate()

On 9 Dec 2024, at 16:35, David Hildenbrand wrote:

> On 09.12.24 20:23, Zi Yan wrote:
>> On 9 Dec 2024, at 14:01, Vlastimil Babka wrote:
>>
>>> On 12/6/24 10:59, David Hildenbrand wrote:
>>>> Let's special-case for the common scenarios that:
>>>>
>>>> (a) We are freeing pages <= pageblock_order
>>>> (b) We are freeing a page <= MAX_PAGE_ORDER and all pageblocks match
>>>>      (especially, no mixture of isolated and non-isolated pageblocks)
>>>
>>> Well in many of those cases we could also just adjust the pageblocks... But
>>> perhaps they indeed shouldn't differ in the first place, unless there's an
>>> isolation attempt.
>>>
>>>> When we encounter a > MAX_PAGE_ORDER page, it can only come from
>>>> alloc_contig_range(), and we can process MAX_PAGE_ORDER chunks.
>>>>
>>>> When we encounter a >pageblock_order <= MAX_PAGE_ORDER page,
>>>> check whether all pageblocks match, and if so (common case), don't
>>>> split them up just for the buddy to merge them back.
>>>>
>>>> This makes sure that when we free MAX_PAGE_ORDER chunks to the buddy,
>>>> for example during system startups, memory onlining, or when isolating
>>>> consecutive pageblocks via alloc_contig_range()/memory offlining, that
>>>> we don't unnecessarily split up what we'll immediately merge again,
>>>> because the migratetypes match.
>>>>
>>>> Rename split_large_buddy() to __free_one_page_maybe_split(), to make it
>>>> clearer what's happening, and handle in it only natural buddy orders,
>>>> not the alloc_contig_range(__GFP_COMP) special case: handle that in
>>>> free_one_page() only.
>>>>
>>>> Signed-off-by: David Hildenbrand <david@...hat.com>
>>>
>>> Acked-by: Vlastimil Babka <vbabka@...e.cz
>>>
>>> Hm but noticed something:
>>>
>>>> +static void __free_one_page_maybe_split(struct zone *zone, struct page *page,
>>>> +		unsigned long pfn, int order, fpi_t fpi_flags)
>>>> +{
>>>> +	const unsigned long end_pfn = pfn + (1 << order);
>>>> +	int mt = get_pfnblock_migratetype(page, pfn);
>>>> +
>>>> +	VM_WARN_ON_ONCE(order > MAX_PAGE_ORDER);
>>>>   	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn, 1 << order));
>>>>   	/* Caller removed page from freelist, buddy info cleared! */
>>>>   	VM_WARN_ON_ONCE(PageBuddy(page));
>>>>
>>>> -	if (order > pageblock_order)
>>>> -		order = pageblock_order;
>>>> -
>>>> -	while (pfn != end) {
>>>> -		int mt = get_pfnblock_migratetype(page, pfn);
>>>> +	/*
>>>> +	 * With CONFIG_MEMORY_ISOLATION, we might be freeing MAX_ORDER_NR_PAGES
>>>> +	 * pages that cover pageblocks with different migratetypes; for example
>>>> +	 * only some migratetypes might be MIGRATE_ISOLATE. In that (unlikely)
>>>> +	 * case, fallback to freeing individual pageblocks so they get put
>>>> +	 * onto the right lists.
>>>> +	 */
>>>> +	if (!IS_ENABLED(CONFIG_MEMORY_ISOLATION) ||
>>>> +	    likely(order <= pageblock_order) ||
>>>> +	    pfnblock_migratetype_equal(pfn + pageblock_nr_pages, end_pfn, mt)) {
>>>> +		__free_one_page(page, pfn, zone, order, mt, fpi_flags);
>>>> +		return;
>>>> +	}
>>>>
>>>> -		__free_one_page(page, pfn, zone, order, mt, fpi);
>>>> -		pfn += 1 << order;
>>>> +	while (pfn != end_pfn) {
>>>> +		mt = get_pfnblock_migratetype(page, pfn);
>>>> +		__free_one_page(page, pfn, zone, pageblock_order, mt, fpi_flags);
>>>> +		pfn += pageblock_nr_pages;
>>>>   		page = pfn_to_page(pfn);
>>>
>>> This predates your patch, but seems potentially dangerous to attempt
>>> pfn_to_page(end_pfn) with SPARSEMEM and no vmemmap and the end_pfn perhaps
>>> being just outside of the valid range? Should we change that?
>>>
>>> But seems this code was initially introduced as part of Johannes'
>>> migratetype hygiene series.
>>
>> It starts as split_free_page() from commit b2c9e2fbba32 ("mm: make
>> alloc_contig_range work at pageblock granularity”), but harmless since
>> it is only used to split a buddy page. Then commit fd919a85cd55 ("mm:
>> page_isolation: prepare for hygienic freelists") refactored it, which
>> should be fine, since it is still used for the same purpose in page
>> isolation. Then commit e98337d11bbd ("mm/contig_alloc: support __GFP_COMP")
>> used it for gigantic hugetlb.
>>
>> For SPARSEMEM && !SPARSEMEM_VMEMMAP, PFNs are contiguous, vmemmap might not
>> be. The code above using pfn in the loop might be fine. And since order
>> is provided, unless the caller is providing a falsely large order, pfn
>> should be valid. Or am I missing anything?
>
> I think the question is, what happens when we call pfn_to_page() on a PFN that falls into a memory section that is either offline, doesn't have a memmap, or does not exist.
>
> With CONFIG_SPARSEMEM, we do a
>
> struct mem_section *__sec = __pfn_to_section(__pfn)
> __section_mem_map_addr(__sec) + __pfn;
>
> __pfn_to_section() can return NULL, in which case __section_mem_map_addr() would dereference NULL.
>
> I assume it ould happen in corner cases, if we'd exceed NR_SECTION_ROOTS. (IOW, large memory, and we free a page that is at the very end of physical memory).
>
> Likely, we should do the pfn_to_page() before the __free_one_page() call.

Got it. Both you and Vlastimil gave the same corner case issue.
I agree that doing pfn_to_page() before the __free_one_page() could get rid of
the concern.

Thank you both.


Best Regards,
Yan, Zi

Powered by blists - more mailing lists