Re: [PATCH v2 04/14] mm: handle hashdist initialization in mm/mm_init.c

[Mike Rapoport]

On Wed, Mar 22, 2023 at 03:49:24PM +0100, Vlastimil Babka wrote:
> On 3/21/23 18:05, Mike Rapoport wrote:
> > From: "Mike Rapoport (IBM)" <rppt@xxxxxxxxxx>
> > 
> > The hashdist variable must be initialized before the first call to
> > alloc_large_system_hash() and free_area_init() looks like a better place
> > for it than page_alloc_init().
> > 
> > Move hashdist handling to mm/mm_init.c
> > 
> > Signed-off-by: Mike Rapoport (IBM) <rppt@xxxxxxxxxx>
> > Acked-by: David Hildenbrand <david@xxxxxxxxxx>
> 
> Reviewed-by: Vlastimil Babka <vbabka@xxxxxxx>
> 
> Looks like this will move the fixup_hashdist() call earlier, but can't
> result in seeing less N_MEMORY nodes than before, right?

hashdist must be set before the first call to alloc_large_system_hash() and
after the nodes present at boot time are initialized, so setting it in the
end of free_area_init() is Ok.

> I wonder if the whole thing lacks hotplug support anyway, what if system
> boots with one node and more are added later? Hmm.

alloc_large_system_hash() is called really early even for !HASH_EARLY
cases. Not sure it's feasible to redistribute the hashes allocated with it
when new node is added.
 
> > ---
> >  mm/mm_init.c    | 22 ++++++++++++++++++++++
> >  mm/page_alloc.c | 18 ------------------
> >  2 files changed, 22 insertions(+), 18 deletions(-)
> > 
> > diff --git a/mm/mm_init.c b/mm/mm_init.c
> > index 68d0187c7886..2e60c7186132 100644
> > --- a/mm/mm_init.c
> > +++ b/mm/mm_init.c
> > @@ -607,6 +607,25 @@ int __meminit early_pfn_to_nid(unsigned long pfn)
> >  
> >  	return nid;
> >  }
> > +
> > +int hashdist = HASHDIST_DEFAULT;
> > +
> > +static int __init set_hashdist(char *str)
> > +{
> > +	if (!str)
> > +		return 0;
> > +	hashdist = simple_strtoul(str, &str, 0);
> > +	return 1;
> > +}
> > +__setup("hashdist=", set_hashdist);
> > +
> > +static inline void fixup_hashdist(void)
> > +{
> > +	if (num_node_state(N_MEMORY) == 1)
> > +		hashdist = 0;
> > +}
> > +#else
> > +static inline void fixup_hashdist(void) {}
> >  #endif /* CONFIG_NUMA */
> >  
> >  #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
> > @@ -1855,6 +1874,9 @@ void __init free_area_init(unsigned long *max_zone_pfn)
> >  	}
> >  
> >  	memmap_init();
> > +
> > +	/* disable hash distribution for systems with a single node */
> > +	fixup_hashdist();
> >  }
> >  
> >  /**
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index c56c147bdf27..ff6a2fff2880 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -6383,28 +6383,10 @@ static int page_alloc_cpu_online(unsigned int cpu)
> >  	return 0;
> >  }
> >  
> > -#ifdef CONFIG_NUMA
> > -int hashdist = HASHDIST_DEFAULT;
> > -
> > -static int __init set_hashdist(char *str)
> > -{
> > -	if (!str)
> > -		return 0;
> > -	hashdist = simple_strtoul(str, &str, 0);
> > -	return 1;
> > -}
> > -__setup("hashdist=", set_hashdist);
> > -#endif
> > -
> >  void __init page_alloc_init(void)
> >  {
> >  	int ret;
> >  
> > -#ifdef CONFIG_NUMA
> > -	if (num_node_state(N_MEMORY) == 1)
> > -		hashdist = 0;
> > -#endif
> > -
> >  	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC,
> >  					"mm/page_alloc:pcp",
> >  					page_alloc_cpu_online,
> 

-- 
Sincerely yours,
Mike.