[PATCH 03/10] Use percpu stats

[Nitin Gupta]

On 08/10/2010 10:04 AM, Andrew Morton wrote:
> On Mon,  9 Aug 2010 22:56:49 +0530 Nitin Gupta <ngupta at vflare.org> wrote:
> 
>> +/*
>> + * Individual percpu values can go negative but the sum across all CPUs
>> + * must always be positive (we store various counts). So, return sum as
>> + * unsigned value.
>> + */
>> +static u64 zram_get_stat(struct zram *zram, enum zram_stats_index idx)
>>  {
>> -	u64 val;
>> -
>> -	spin_lock(&zram->stat64_lock);
>> -	val = *v;
>> -	spin_unlock(&zram->stat64_lock);
>> +	int cpu;
>> +	s64 val = 0;
>> +
>> +	for_each_possible_cpu(cpu) {
>> +		s64 temp;
>> +		unsigned int start;
>> +		struct zram_stats_cpu *stats;
>> +
>> +		stats = per_cpu_ptr(zram->stats, cpu);
>> +		do {
>> +			start = u64_stats_fetch_begin(&stats->syncp);
>> +			temp = stats->count[idx];
>> +		} while (u64_stats_fetch_retry(&stats->syncp, start));
>> +		val += temp;
>> +	}
>>  
>> +	WARN_ON(val < 0);
>>  	return val;
>>  }
> 
> That reimplements include/linux/percpu_counter.h, poorly.
> 
> Please see the June discussion "[PATCH v2 1/2] tmpfs: Quick token
> library to allow scalable retrieval of tokens from token jar" for some
> discussion.
> 
> 

I read the discussion you pointed out but still fail to see how percpu_counters,
with all their overhead, are better than simple pcpu variable used in current
version. What is the advantage?

Thanks,
Nitin