Re: [PATCH 2/3] percpu: add __GFP_NORETRY semantics to the percpu balancing path

[Dennis Zhou <dennisszhou@gmail.com>]

Hi Tejun,

On Thu, Feb 15, 2018 at 01:39:09PM -0800, Tejun Heo wrote:
> On Thu, Feb 15, 2018 at 10:08:15AM -0600, Dennis Zhou wrote:
> > -static struct pcpu_chunk *pcpu_create_chunk(void)
> > +static struct pcpu_chunk *pcpu_create_chunk(gfp_t gfp)
> >  {
> >  	const int nr_pages = pcpu_group_sizes[0] >> PAGE_SHIFT;
> >  	struct pcpu_chunk *chunk;
> >  	struct page *pages;
> >  	int i;
> >  
> > -	chunk = pcpu_alloc_chunk();
> > +	chunk = pcpu_alloc_chunk(gfp);
> >  	if (!chunk)
> >  		return NULL;
> >  
> > -	pages = alloc_pages(GFP_KERNEL, order_base_2(nr_pages));
> > +	pages = alloc_pages(gfp | GFP_KERNEL, order_base_2(nr_pages));
> 
> Is there a reason to set GFP_KERNEL in this function?  I'd prefer
> pushing this to the callers.
> 

Not particularly. As I wasn't sure of the original decision to use
GFP_KERNEL for all percpu underlying allocations, I didn't want to
add the gfp passthrough and remove functionality.

> > diff --git a/mm/percpu-vm.c b/mm/percpu-vm.c
> > index 9158e5a..ea9906a 100644
> > --- a/mm/percpu-vm.c
> > +++ b/mm/percpu-vm.c
> > @@ -37,7 +37,7 @@ static struct page **pcpu_get_pages(void)
> >  	lockdep_assert_held(&pcpu_alloc_mutex);
> >  
> >  	if (!pages)
> > -		pages = pcpu_mem_zalloc(pages_size);
> > +		pages = pcpu_mem_zalloc(pages_size, 0);
>                                                   ^^^^
> 						  because this is confusing

Yeah.. The next patch removes this as the additional gfp flags is weird.

> >  static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
> > -			    struct page **pages, int page_start, int page_end)
> > +			    struct page **pages, int page_start, int page_end,
> > +			    gfp_t gfp)
> >  {
> > -	const gfp_t gfp = GFP_KERNEL | __GFP_HIGHMEM;
> >  	unsigned int cpu, tcpu;
> >  	int i;
> >  
> > +	gfp |=  GFP_KERNEL | __GFP_HIGHMEM;
>               ^^
> 	      double space
> 

I'll fix this with any other updates.

> So, setting __GFP_HIGHMEM unconditionally here makes sense because
> it's indicating the types of pages we can use (we also accept high
> pages); however, I'm not sure GFP_KERNEL makes sense.  That's about
> "how to allocate" and looks like it should be left to the caller.
> 

That makes sense, I can remove the forced GFP_KERNEL use in the next
patch as that patch moves the flags to the caller.

I'd rather be explicit though and whitelist GFP_KERNEL as I don't have a
full grasp of all the flags. Our use case is a little different because
we ultimately become the owner of the pages until the chunk is freed. So
there are certain flags such as __GFP_HARDWALL (poor example), the
difference between GFP_KERNEL and GFP_USER, which don't make sense here.

Regarding high pages, I think you're referring to GFP_ATOMIC
allocations? We actually never allocate on this path as allocations must
be served out of parts of chunks that are already backed.

Thanks,
Dennis