Re: poison_element vs highmem, was Re: [linux-next:master] [block] ec7f31b2a2: BUG:unable_to_handle_page_fault_for_address

[Vlastimil Babka <vbabka@suse.cz>]

On 11/13/25 08:44, Oliver Sang wrote:
> hi, Vlastimil Babka,
> 
> On Wed, Nov 12, 2025 at 10:33:32AM +0100, Vlastimil Babka wrote:
>> On 11/11/25 08:48, Christoph Hellwig wrote:
>> > Looks like this is due to the code in poison_element, which tries
>> > to memset more than PAGE_SIZE for a single page.  This probably
>> > implies we are the first users of the mempool page helpers for order > 0,
>> > or at least the first one tested by anyone on 32-bit with highmem :)
>> > 
>> > That code seems to come from
>> > 
>> > commit bdfedb76f4f5aa5e37380e3b71adee4a39f30fc6
>> > Author: David Rientjes <rientjes@google.com>
>> > Date:   Wed Apr 15 16:14:17 2015 -0700
>> > 
>> >     mm, mempool: poison elements backed by slab allocator
>> > 
>> > originally.  The easiest fix would be to just skip poisoning for this
>> > case, although that would reduce the usefulness of the poisoning.
>> 
>> #syz test
> 
> we applied below patch upon ec7f31b2a2 directly, and confirmed the issue we
> reported gone now with the patch.
> 
> Tested-by: kernel test robot <oliver.sang@intel.com>

Thanks!

> BTW, we are kernel test robot, not the syzbot :) thanks

Yeah I realized only after sending...

I'll make this a full patch then. How urgent is it, Christoph? I suppose
this is related to the bulk mempool changes, and we discussed the users will
target 6.20 (7.0?) merge window? So landing this fix in 6.19 is enough?

>> ----8<----
>> From 4d97b55c208c611cb01062e0fbf9dbda9f5617d5 Mon Sep 17 00:00:00 2001
>> From: Vlastimil Babka <vbabka@suse.cz>
>> Date: Wed, 12 Nov 2025 10:29:52 +0100
>> Subject: [PATCH] mm/mempool: fix poisoning order>0 pages with HIGHMEM
>> 
>> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
>> ---
>>  mm/mempool.c | 28 ++++++++++++++++++++++------
>>  1 file changed, 22 insertions(+), 6 deletions(-)
>> 
>> diff --git a/mm/mempool.c b/mm/mempool.c
>> index 1c38e873e546..75fea9441b93 100644
>> --- a/mm/mempool.c
>> +++ b/mm/mempool.c
>> @@ -68,10 +68,18 @@ static void check_element(mempool_t *pool, void *element)
>>  	} else if (pool->free == mempool_free_pages) {
>>  		/* Mempools backed by page allocator */
>>  		int order = (int)(long)pool->pool_data;
>> -		void *addr = kmap_local_page((struct page *)element);
>> +#ifdef CONFIG_HIGHMEM
>> +		for (int i = 0; i < (1 << order); i++) {
>> +			struct page *page = (struct page *)element;
>> +			void *addr = kmap_local_page(page + i);
>> 
>> -		__check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
>> -		kunmap_local(addr);
>> +			__check_element(pool, addr, PAGE_SIZE);
>> +			kunmap_local(addr);
>> +		}
>> +#else
>> +		void *addr = page_address((struct page *)element);
>> +		__check_element(pool, addr, PAGE_SIZE << order);
>> +#endif
>>  	}
>>  }
>> 
>> @@ -97,10 +105,18 @@ static void poison_element(mempool_t *pool, void *element)
>>  	} else if (pool->alloc == mempool_alloc_pages) {
>>  		/* Mempools backed by page allocator */
>>  		int order = (int)(long)pool->pool_data;
>> -		void *addr = kmap_local_page((struct page *)element);
>> +#ifdef CONFIG_HIGHMEM
>> +		for (int i = 0; i < (1 << order); i++) {
>> +			struct page *page = (struct page *)element;
>> +			void *addr = kmap_local_page(page + i);
>> 
>> -		__poison_element(addr, 1UL << (PAGE_SHIFT + order));
>> -		kunmap_local(addr);
>> +			__poison_element(addr, PAGE_SIZE);
>> +			kunmap_local(addr);
>> +		}
>> +#else
>> +		void *addr = page_address((struct page *)element);
>> +		__poison_element(addr, PAGE_SIZE << order);
>> +#endif
>>  	}
>>  }
>>  #else /* CONFIG_SLUB_DEBUG_ON */
>> -- 
>> 2.51.1
>> 
>>