Re: [PATCH v2 7/7] docs/core-api: mm-api: add section about GFP flags

From: Michal Hocko
Date: Thu Jul 26 2018 - 12:41:57 EST


On Thu 26-07-18 18:18:53, Mike Rapoport wrote:
> On Thu, Jul 26, 2018 at 04:20:39PM +0200, Michal Hocko wrote:
> > On Thu 26-07-18 06:01:06, Matthew Wilcox wrote:
> > > On Thu, Jul 26, 2018 at 03:22:02PM +0300, Mike Rapoport wrote:
> > > > +Memory Allocation Controls
> > > > +==========================
> > >
> > > Perhaps call this section "Memory Allocation Flags" instead?
> > >
> > > > +Linux provides a variety of APIs for memory allocation from direct
> > > > +calls to page allocator through slab caches and vmalloc to allocators
> > > > +of compressed memory. Although these allocators have different
> > > > +semantics and are used in different circumstances, they all share the
> > > > +GFP (get free page) flags that control behavior of each allocation
> > > > +request.
> > >
> > > While this isn't /wrong/, I think it might not be the most useful way
> > > of explaining what the GFP flags are to someone who's just come across
> > > them in some remote part of the kernel. How about this paragraph instead?
> > >
> > > Functions which need to allocate memory often use GFP flags to express
> > > how that memory should be allocated. The GFP acronym stands for "get
> > > free pages", the underlying memory allocation function.
> >
> > OK.
> >
> > > Not every GFP
> > > flag is allowed to every function which may allocate memory. Most
> > > users will want to use a plain ``GFP_KERNEL`` or ``GFP_ATOMIC``.
> >
> > Or rather than mentioning the two just use "Useful GFP flag
> > combinations" comment segment from gfp.h
>
> The comment there includes GFP_DMA, GFP_NOIO etc so I'd prefer Matthew's
> version and maybe even omit GFP_ATOMIC from it.
>
> Some grepping shows that roughly 80% of allocations are GFP_KERNEL, 12% are
> GFP_ATOMIC and ... I didn't count the usage of other flags ;-)

Well, I will certainly not insist... I don't know who is the expected
audience of this documentation. That section was meant for kernel
developers to know which of the high level flags to use.
--
Michal Hocko
SUSE Labs