Re: [PATCH] mm/hotplug: Adjust shrink_zone_span() to keep the old logic

[Baoquan He]

On 02/06/20 at 11:05am, David Hildenbrand wrote:
> On 06.02.20 11:02, David Hildenbrand wrote:
> > On 06.02.20 11:00, Baoquan He wrote:
> >> On 02/06/20 at 10:48am, David Hildenbrand wrote:
> >>> On 06.02.20 10:35, Baoquan He wrote:
> >>>> On 02/06/20 at 09:50am, David Hildenbrand wrote:
> >>>>> On 06.02.20 06:39, Baoquan He wrote:
> >>>>>> In commit 950b68d9178b ("mm/memory_hotplug: don't check for "all holes"
> >>>>>> in shrink_zone_span()"), the zone->zone_start_pfn/->spanned_pages
> >>>>>> resetting is moved into the if()/else if() branches, if the zone becomes
> >>>>>> empty. However the 2nd resetting code block may cause misunderstanding.
> >>>>>>
> >>>>>> So take the resetting codes out of the conditional checking and handling
> >>>>>> branches just as the old code does, the find_smallest_section_pfn()and
> >>>>>> find_biggest_section_pfn() searching have done the the same thing as
> >>>>>> the old for loop did, the logic is kept the same as the old code. This
> >>>>>> can remove the possible confusion.
> >>>>>>
> >>>>>> Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
> >>>>>> ---
> >>>>>>  mm/memory_hotplug.c | 14 ++++++--------
> >>>>>>  1 file changed, 6 insertions(+), 8 deletions(-)
> >>>>>>
> >>>>>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> >>>>>> index 089b6c826a9e..475d0d68a32c 100644
> >>>>>> --- a/mm/memory_hotplug.c
> >>>>>> +++ b/mm/memory_hotplug.c
> >>>>>> @@ -398,7 +398,7 @@ static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
> >>>>>>  static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
> >>>>>>  			     unsigned long end_pfn)
> >>>>>>  {
> >>>>>> -	unsigned long pfn;
> >>>>>> +	unsigned long pfn = zone->zone_start_pfn;
> >>>>>>  	int nid = zone_to_nid(zone);
> >>>>>>  
> >>>>>>  	zone_span_writelock(zone);
> >>>>>> @@ -414,9 +414,6 @@ static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
> >>>>>>  		if (pfn) {
> >>>>>>  			zone->spanned_pages = zone_end_pfn(zone) - pfn;
> >>>>>>  			zone->zone_start_pfn = pfn;
> >>>>>> -		} else {
> >>>>>> -			zone->zone_start_pfn = 0;
> >>>>>> -			zone->spanned_pages = 0;
> >>>>>>  		}
> >>>>>>  	} else if (zone_end_pfn(zone) == end_pfn) {
> >>>>>>  		/*
> >>>>>> @@ -429,10 +426,11 @@ static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
> >>>>>>  					       start_pfn);
> >>>>>>  		if (pfn)
> >>>>>>  			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
> >>>>>> -		else {
> >>>>>> -			zone->zone_start_pfn = 0;
> >>>>>> -			zone->spanned_pages = 0;
> >>>>>> -		}
> >>>>>> +	}
> >>>>>> +
> >>>>>> +	if (!pfn) {
> >>>>>> +		zone->zone_start_pfn = 0;
> >>>>>> +		zone->spanned_pages = 0;
> >>>>>>  	}
> >>>>>>  	zone_span_writeunlock(zone);
> >>>>>>  }
> >>>>>>
> >>>>>
> >>>>> So, what if your zone starts at pfn 0? Unlikely that we can actually
> >>>>> offline that, but still it is more confusing than the old code IMHO.
> >>>>> Then I prefer to drop the second else case as discussed instead.
> >>>>
> >>>> Hmm, pfn is initialized as zone->zone_start_pfn, does it matter?
> >>>> The impossible empty zone won't go wrong if it really happen.
> >>>>
> >>>
> >>> If you offline any memory block that belongs to the lowest zone
> >>> (zone->zone_start_pfn == 0) but does not fall on a boundary (so that you
> >>> can actually shrink), you would mark the whole zone offline. That's
> >>> broken unless I am missing something.
> >>
> >> AFAIK, the page 0 is reserved. No valid zone can start at 0, only empty
> >> zone is. Please correct me if I am wrong.
> > 
> > At least on x86 it indeed is :) So if this holds true for all archs
> > 
> > Acked-by: David Hildenbrand <david@xxxxxxxxxx>
> > 
> > Thanks!
> > 
> > 
> 
> Correction
> 
> Nacked-by: David Hildenbrand <david@xxxxxxxxxx>
> 
> s390x:
> [linux1@rhkvm01 ~]$ cat /proc/zoneinfo
> Node 0, zone      DMA
>   per-node stats
> [...]
>   node_unreclaimable:  0
>   start_pfn:           0

OK, it's very interesting, and good to know. This should be discarded.