Re: [PATCH bpf-next 1/6] mm/percpu.c: introduce alloc_size_percpu()

[Hou Tao <houtao@huaweicloud.com>]

Hi,

On 10/9/2023 6:32 AM, Dennis Zhou wrote:
> Hello,
>
> On Sat, Oct 07, 2023 at 09:51:01PM +0800, Hou Tao wrote:
>> From: Hou Tao <houtao1@huawei.com>
>>
>> Introduce alloc_size_percpu() to get the size of the dynamic per-cpu
>> area. It will be used by bpf memory allocator in the following patches.
>> BPF memory allocator maintains multiple per-cpu area caches for multiple
>> area sizes and it needs the size of dynamic per-cpu area to select the
>> corresponding cache when bpf program frees the dynamic per-cpu area.
>>
>> Signed-off-by: Hou Tao <houtao1@huawei.com>
>> ---
>>  include/linux/percpu.h |  1 +
>>  mm/percpu.c            | 29 +++++++++++++++++++++++++++++
>>  2 files changed, 30 insertions(+)
>>
>> diff --git a/include/linux/percpu.h b/include/linux/percpu.h
>> index 68fac2e7cbe6..d140d9d79567 100644
>> --- a/include/linux/percpu.h
>> +++ b/include/linux/percpu.h
>> @@ -132,6 +132,7 @@ extern void __init setup_per_cpu_areas(void);
>>  extern void __percpu *__alloc_percpu_gfp(size_t size, size_t align, gfp_t gfp) __alloc_size(1);
>>  extern void __percpu *__alloc_percpu(size_t size, size_t align) __alloc_size(1);
>>  extern void free_percpu(void __percpu *__pdata);
>> +extern size_t alloc_size_percpu(void __percpu *__pdata);
>>  
>>  DEFINE_FREE(free_percpu, void __percpu *, free_percpu(_T))
>>  
>> diff --git a/mm/percpu.c b/mm/percpu.c
>> index 7b40b3963f10..f541cfc3cb2d 100644
>> --- a/mm/percpu.c
>> +++ b/mm/percpu.c
>> @@ -2244,6 +2244,35 @@ static void pcpu_balance_workfn(struct work_struct *work)
>>  	mutex_unlock(&pcpu_alloc_mutex);
>>  }
>>  
>> +/**
>> + * alloc_size_percpu - the size of the dynamic percpu area
> Can we name this pcpu_alloc_size(). A few other functions are
> exposed under pcpu_* so it's a bit easier to keep track of.

Will do in v2.
>
>> + * @ptr: pointer to the dynamic percpu area
>> + *
>> + * Return the size of the dynamic percpu area @ptr.
>> + *
>> + * RETURNS:
>> + * The size of the dynamic percpu area.
>> + *
>> + * CONTEXT:
>> + * Can be called from atomic context.
>> + */
>> +size_t alloc_size_percpu(void __percpu *ptr)
>> +{
>> +	struct pcpu_chunk *chunk;
>> +	int bit_off, end;
>> +	void *addr;
>> +
>> +	if (!ptr)
>> +		return 0;
>> +
>> +	addr = __pcpu_ptr_to_addr(ptr);
>> +	/* No pcpu_lock here: ptr has not been freed, so chunk is still alive */
> Now that percpu variables are floating around more commonly, I think we
> or I need to add more validation guards so it's easier to
> debug bogus/stale pointers. Potentially like a static_key or Kconfig so
> that we take the lock and `test_bit()`.

Before the patch, it seems that free_percpu() is the only user of
__pcpu_ptr_to_addr(ptr). I can move the invocation of both
__pcpu_ptr_to_addr() and pcpu_chunk_addr_search() into a common helper
if you are OK with it. And I think it will ease the work of adding
validation guard on the per-cpu pointer.
>
>> +	chunk = pcpu_chunk_addr_search(addr);
>> +	bit_off = (addr - chunk->base_addr) / PCPU_MIN_ALLOC_SIZE;
>> +	end = find_next_bit(chunk->bound_map, pcpu_chunk_map_bits(chunk), bit_off + 1);
> Nit: can you please reflow `bit_off + 1` to the next line. I know we
> dropped the line requirement, but percpu.c almost completely still
> follows it.

Will do in v2.
>
>> +	return (end - bit_off) * PCPU_MIN_ALLOC_SIZE;
>> +}
>> +
>>  /**
>>   * free_percpu - free percpu area
>>   * @ptr: pointer to area to free
>> -- 
>> 2.29.2
>>
> Thanks,
> Dennis