Re: [PATCH 1/8] mm: cma: introduce cma_release_nowait()

[Oscar Salvador]

On Thu, Mar 25, 2021 at 10:45:18AM +0100, Michal Hocko wrote:
> On Wed 24-03-21 17:28:28, Mike Kravetz wrote:
> [...]
> >  phys_addr_t cma_get_base(const struct cma *cma)
> >  {
> >  	return PFN_PHYS(cma->base_pfn);
> > @@ -146,6 +155,10 @@ static int __init cma_init_reserved_areas(void)
> >  	for (i = 0; i < cma_area_count; i++)
> >  		cma_activate_area(&cma_areas[i]);
> >  
> > +	cma_release_wq = create_workqueue("cma_release");
> 
> Considering the workqueue is used to free up memory it should likely be
> WQ_MEM_RECLAIM to prevent from long stalls when WQs are overloaded.

I might be missing something but the worqueue->func() seems to not be anything
for memory related purposes.

The worqueue calls cma_clear_bitmap_fn(), and the only think that does
is freeing one page, and clearing the bitmap. 
I might be reading the CMA code wrongly, but I cannot see any freeing up
of memory in cma_clear_bitmap()->__bitmap_clear() (or its siblings).

The actual freeing of memory seems to happen in cma_release_no_wait()
with: 

 if (count > 1)
       free_contig_range(pfn + 1, count - 1);

-- 
Oscar Salvador
SUSE L3