Re: [PATCH v4 1/4] mm: cma: introduce gfp flag in cma_alloc instead of no_warn

[Minchan Kim]

On Tue, Jan 26, 2021 at 11:12:14AM -0800, Minchan Kim wrote:
> On Tue, Jan 26, 2021 at 08:38:08AM +0100, Michal Hocko wrote:
> > On Mon 25-01-21 11:42:34, Minchan Kim wrote:
> > > On Mon, Jan 25, 2021 at 02:07:01PM +0100, Michal Hocko wrote:
> > > > On Thu 21-01-21 09:54:59, Minchan Kim wrote:
> > > > > The upcoming patch will introduce __GFP_NORETRY semantic
> > > > > in alloc_contig_range which is a failfast mode of the API.
> > > > > Instead of adding a additional parameter for gfp, replace
> > > > > no_warn with gfp flag.
> > > > > 
> > > > > To keep old behaviors, it follows the rule below.
> > > > > 
> > > > >   no_warn 			gfp_flags
> > > > > 
> > > > >   false         		GFP_KERNEL
> > > > >   true          		GFP_KERNEL|__GFP_NOWARN
> > > > >   gfp & __GFP_NOWARN		GFP_KERNEL | (gfp & __GFP_NOWARN)
> > > > > 
> > > > > Reviewed-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> > > > > Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
> > > > [...]
> > > > > diff --git a/mm/cma.c b/mm/cma.c
> > > > > index 0ba69cd16aeb..d50627686fec 100644
> > > > > --- a/mm/cma.c
> > > > > +++ b/mm/cma.c
> > > > > @@ -419,13 +419,13 @@ static inline void cma_debug_show_areas(struct cma *cma) { }
> > > > >   * @cma:   Contiguous memory region for which the allocation is performed.
> > > > >   * @count: Requested number of pages.
> > > > >   * @align: Requested alignment of pages (in PAGE_SIZE order).
> > > > > - * @no_warn: Avoid printing message about failed allocation
> > > > > + * @gfp_mask: GFP mask to use during the cma allocation.
> > > > 
> > > > Call out supported gfp flags explicitly. Have a look at kvmalloc_node
> > > > for a guidance.
> > > 
> > > How about this?
> > > 
> > > 
> > > diff --git a/mm/cma.c b/mm/cma.c
> > > index d50627686fec..b94727b694d6 100644
> > > --- a/mm/cma.c
> > > +++ b/mm/cma.c
> > > @@ -423,6 +423,10 @@ static inline void cma_debug_show_areas(struct cma *cma) { }
> > >   *
> > >   * This function allocates part of contiguous memory on specific
> > >   * contiguous memory area.
> > > + *
> > > + * For gfp_mask, GFP_KERNEL and __GFP_NORETRY are supported. __GFP_NORETRY
> > > + * will avoid costly functions(e.g., waiting on page_writeback and locking)
> > > + * at current implementaion during the page migration.
> > 
> > rather than explicitly mentioning what the flag implies I think it would
> > be more useful to state the intended usecase. See how kvmalloc_node says
> > "__GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
> > preferable to the vmalloc fallback, due to visible performance
> > drawbacks.
> > __GFP_NOWARN is also supported to suppress allocation failure messages."
> > 
> > This would help people not familiar with internals to see whether this
> > flag is a good fit for them.
> > 
> > In this case I woul go with
> > "
> > @flags: gfp mask. Must be compatible (superset) with GFP_KERNEL.
> > [...]
> > Reclaim modifiers (__GFP_RETRY_MAYFAIL, __GFP_NOFAIL) are not supported.
> > __GFP_NORETRY is supported, and it should be used for opportunistic
> > allocation attempts that should rather fail quickly when the caller has
> > a fallback strategy.
> > "
> > 
> > Obviously for this patch you will go with a simple statement that
> > Reclaim modifiers are not supported at all.
> 
> After more discussion for gfp_flags in thread of next patch, let me
> changes a bit more based on it.
> 
> Thanks for the suggestion, Michal.

Based on the discussion in other thread, I want to go with __GFP_NORETRY
to indicate "opportunistic-easy-to-fail attempt" so suggestion words
Michal is valid so I will carry it on next version.

Thank you.