[PATCH RFC 3/4] mm: percpu: Add generic pcpu_fc_alloc/free funciton

[Kefeng Wang]

On 2021/11/30 6:45, Dennis Zhou wrote:
> On Sun, Nov 21, 2021 at 05:35:56PM +0800, Kefeng Wang wrote:
>> With previous patch, we could add a generic pcpu first chunk
>> allocation and free function to cleanup the duplicated definations
>> on each architecture.
>>
>> Signed-off-by: Kefeng Wang <wangkefeng.wang at huawei.com>
>> ---
>>   arch/mips/mm/init.c            | 16 +--------
>>   arch/powerpc/kernel/setup_64.c | 51 ++------------------------
>>   arch/sparc/kernel/smp_64.c     | 50 +-------------------------
>>   arch/x86/kernel/setup_percpu.c | 59 +-----------------------------
>>   drivers/base/arch_numa.c       | 19 +---------
>>   include/linux/percpu.h         |  9 +----
>>   mm/percpu.c                    | 66 ++++++++++++++++++----------------
>>   7 files changed, 42 insertions(+), 228 deletions(-)
...
>> diff --git a/include/linux/percpu.h b/include/linux/percpu.h
>> index 41bb54715b0c..d73c97ef4ff4 100644
>> --- a/include/linux/percpu.h
>> +++ b/include/linux/percpu.h
>> @@ -95,9 +95,6 @@ extern const char * const pcpu_fc_names[PCPU_FC_NR];
>>   extern enum pcpu_fc pcpu_chosen_fc;
>>   
>>   typedef int (pcpu_fc_cpu_to_node_fn_t)(int cpu);
>> -typedef void * (*pcpu_fc_alloc_fn_t)(unsigned int cpu, size_t size, size_t align,
>> -				     pcpu_fc_cpu_to_node_fn_t cpu_to_nd_fn);
>> -typedef void (*pcpu_fc_free_fn_t)(void *ptr, size_t size);
>>   typedef void (*pcpu_fc_populate_pte_fn_t)(unsigned long addr);
>>   typedef int (pcpu_fc_cpu_distance_fn_t)(unsigned int from, unsigned int to);
>>   
>> @@ -112,15 +109,11 @@ extern void __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
>>   extern int __init pcpu_embed_first_chunk(size_t reserved_size, size_t dyn_size,
>>   				size_t atom_size,
>>   				pcpu_fc_cpu_distance_fn_t cpu_distance_fn,
>> -				pcpu_fc_cpu_to_node_fn_t cpu_to_nd_fn,
>> -				pcpu_fc_alloc_fn_t alloc_fn,
>> -				pcpu_fc_free_fn_t free_fn);
>> +				pcpu_fc_cpu_to_node_fn_t cpu_to_nd_fn);
>>   #endif
>>   
>>   #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
>>   extern int __init pcpu_page_first_chunk(size_t reserved_size,
>> -				pcpu_fc_alloc_fn_t alloc_fn,
>> -				pcpu_fc_free_fn_t free_fn,
>>   				pcpu_fc_cpu_to_node_fn_t cpu_to_nd_fn,
>>   				pcpu_fc_populate_pte_fn_t populate_pte_fn);
>>   #endif
>> diff --git a/mm/percpu.c b/mm/percpu.c
>> index 3f6cf1ff0be2..efaa1cbaf73d 100644
>> --- a/mm/percpu.c
>> +++ b/mm/percpu.c
>> @@ -2992,6 +2992,30 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
>>   
>>   	return ai;
>>   }
>> +
>> +static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align,
>> +				   pcpu_fc_cpu_to_node_fn_t cpu_to_nd_fn)
>> +{
>> +	const unsigned long goal = __pa(MAX_DMA_ADDRESS);
>> +	int node = NUMA_NO_NODE;
>> +	void *ptr;
>> +
>> +	if (cpu_to_nd_fn)
>> +		node = cpu_to_nd_fn(cpu);
>> +
>> +	if (node == NUMA_NO_NODE || !node_online(node) || !NODE_DATA(node)) {
>> +		ptr = memblock_alloc_from(size, align, goal);
>> +	} else {
>> +		ptr = memblock_alloc_try_nid(size, align, goal,
>> +					     MEMBLOCK_ALLOC_ACCESSIBLE, node);
>> +	}
>> +	return ptr;
>> +}
> My preference here would be to keep this identical to the x86
> implementation where we #ifdef CONFIG_NUMA.

I will add back '#ifdef CONFIG_NUMA',  also will add back the 
pr_debug/pr_info part as x86 does,

> Overall this makes sense.
>
> Thanks,
> Dennis
> .