Re: [PATCH v3 2/2] mm/sparse: start using sparse_init_nid(), and remove old code

[Pavel Tatashin]

On 07/02/2018 03:47 PM, Dave Hansen wrote:
> On 07/01/2018 07:04 PM, Pavel Tatashin wrote:
>> +	for_each_present_section_nr(pnum_begin + 1, pnum_end) {
>> +		int nid = sparse_early_nid(__nr_to_section(pnum_end));
>>  
>> +		if (nid == nid_begin) {
>> +			map_count++;
>>  			continue;
>>  		}
> 
>> +		sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
>> +		nid_begin = nid;
>> +		pnum_begin = pnum_end;
>> +		map_count = 1;
>>  	}
> 
> Ugh, this is really hard to read.  Especially because the pnum "counter"
> is called "pnum_end".

I called it pnum_end, because that is what is passed to sparse_init_nid(), but I see your point, and I can rename pnum_end to simply pnum if that will make things look better.

> 
> So, this is basically a loop that collects all of the adjacent sections
> in a given single nid and then calls sparse_init_nid().  pnum_end in
> this case is non-inclusive, so the sparse_init_nid() call is actually
> for the *previous* nid that pnum_end is pointing _past_.
> 
> This *really* needs commenting.

There is a comment before sparse_init_nid() about inclusiveness:

434 /*
435  * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
436  * And number of present sections in this node is map_count.
437  */
438 static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
439                                    unsigned long pnum_end,
440                                    unsigned long map_count)


Thank you,
Pavel