Re: [PATCH v3 2/4] mm/sparsemem: Defer the ms->section_mem_map clearing

[Baoquan He]

On 04/09/18 at 09:02am, Dave Hansen wrote:
> On 04/07/2018 11:50 PM, Baoquan He wrote:
> >> Should the " = 0" instead be clearing SECTION_MARKED_PRESENT or
> >> something?  That would make it easier to match the code up with the code
> >> that it is effectively undoing.
> > 
> > Not sure if I understand your question correctly. From memory_present(),
> > information encoded into ms->section_mem_map including numa node,
> > SECTION_IS_ONLINE and SECTION_MARKED_PRESENT. Not sure if it's OK to only
> > clear SECTION_MARKED_PRESENT.  People may wrongly check SECTION_IS_ONLINE
> > and do something on this memory section?
> 
> What is mean is that, instead of:

I mean that in memory_present() all present sections are marked with
below information.

	ms->section_mem_map = (nid << SECTION_NID_SHIFT) |
			      SECTION_MARKED_PRESENT |
			      SECTION_IS_ONLINE;

Later in sparse_init(), if we failed to allocate mem map, the
corresponding section need clear its ->section_mem_map. The existing
code does the clearing with: 

	ms->section_mem_map = 0;

If with 'ms->section_mem_map &= ~SECTION_MARKED_PRESENT', the nid and
SECTION_IS_ONLINE are still left in ms->section_mem_map. Someone may
probably mistakenly check if this section is online and do something, or
still get nid from this section. Just worried.

> 
> 	
> 	ms->section_mem_map = 0;
> 
> we could literally do:
> 
> 	ms->section_mem_map &= ~SECTION_MARKED_PRESENT;
> 
> That does the same thing in practice, but makes the _intent_ much more
> clear.