Re: [PATCH] bpf: Try harder when allocating memory for maps

[Martynas Pumputis <m@lambda.lt>]

On 3/8/19 9:44 AM, Michal Hocko wrote:
> On Fri 08-03-19 09:08:57, Martynas Pumputis wrote:
>> It has been observed that sometimes memory allocation for BPF maps
>> fails when there is no obvious memory pressure in a system.
>>
>> E.g. the map (BPF_MAP_TYPE_LRU_HASH, key=38, value=56, max_elems=524288)
>> could not be created due to due to vmalloc unable to allocate 75497472B,
>> when the system's memory consumption (in MB) was the following:
>>
>>      Total: 3942 Used: 837 (21.24%) Free: 138 Buffers: 239 Cached: 2727
> 
> Hmm 75MB is quite large and much larger than the slab/page allocator
> cann provide so this is not really a fragmentation issue. Vmalloc does
> respect noretry but considering that there shouldn't be a large memory
> pressure I wonder how NORETRY managed to fail the allocation. Do you
> happen to have the allocation failure report?

I got /proc/{meminfo,vmstat,vmallocinfo} just after the allocation has 
failed:
https://gist.github.com/brb/62092c1d83daa6527271b88f0352e32d

Let me know if more info is required, I can reproduce the failure. Thanks.

> 
> Btw. is there any real reason to opencode and duplicate kvmalloc logic
> here? In other words why not simply make bpf_map_area_alloc use
> kvmalloc_node with GFP_KERNEL?
> 
>> Considering dcda9b0471 ("mm, tree wide: replace __GFP_REPEAT by
>> __GFP_RETRY_MAYFAIL with more useful semantic") we can replace
>> __GFP_NORETRY with __GFP_RETRY_MAYFAIL, as it won't invoke OOM killer
>> and will try harder to fulfil allocation requests.
>>
>> The change has been tested with the workloads mentioned above and by
>> observing oom_kill value from /proc/vmstat.
>>
>> Signed-off-by: Martynas Pumputis <m@lambda.lt>
>> ---
>>   kernel/bpf/syscall.c | 8 ++++----
>>   1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
>> index 62f6bced3a3c..eb5cefe44af3 100644
>> --- a/kernel/bpf/syscall.c
>> +++ b/kernel/bpf/syscall.c
>> @@ -136,11 +136,11 @@ static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
>>   
>>   void *bpf_map_area_alloc(size_t size, int numa_node)
>>   {
>> -	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
>> -	 * trigger under memory pressure as we really just want to
>> -	 * fail instead.
>> +	/* We definitely need __GFP_NORETRY or __GFP_RETRY_MAYFAIL, so
>> +	 * OOM killer doesn't trigger under memory pressure as we really
>> +	 * just want to fail instead.
>>   	 */
>> -	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
>> +	const gfp_t flags = __GFP_NOWARN | __GFP_RETRY_MAYFAIL | __GFP_ZERO;
>>   	void *area;
>>   
>>   	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
>> -- 
>> 2.21.0
>>
>