Re: [PATCH] mm/alloc: fallback to first node if the wanted node offline

[Pingfan Liu]

On Mon, Dec 10, 2018 at 8:37 PM Michal Hocko <mhocko@xxxxxxxxxx> wrote:
>
> On Fri 07-12-18 16:56:27, Michal Hocko wrote:
> > On Fri 07-12-18 22:27:13, Pingfan Liu wrote:
> > [...]
> > > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> > > index 1308f54..4dc497d 100644
> > > --- a/arch/x86/mm/numa.c
> > > +++ b/arch/x86/mm/numa.c
> > > @@ -754,18 +754,23 @@ void __init init_cpu_to_node(void)
> > >  {
> > >         int cpu;
> > >         u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
> > > +       int node, nr;
> > >
> > >         BUG_ON(cpu_to_apicid == NULL);
> > > +       nr = cpumask_weight(cpu_possible_mask);
> > > +
> > > +       /* bring up all possible node, since dev->numa_node */
> > > +       //should check acpi works for node possible,
> > > +       for_each_node(node)
> > > +               if (!node_online(node))
> > > +                       init_memory_less_node(node);
> >
> > I suspect there is no change if you replace for_each_node by
> >       for_each_node_mask(nid, node_possible_map)
> >
> > here. If that is the case then we are probably calling
> > free_area_init_node too early. I do not see it yet though.
>
> OK, so it is not about calling it late or soon. It is just that
> node_possible_map is a misnomer and it has a different semantic than
> I've expected. numa_nodemask_from_meminfo simply considers only nodes
> with some memory. So my patch didn't really make any difference and the
> node stayed uninialized.
>
> In other words. Does the following work? I am sorry to wildguess this
> way but I am not able to recreate your setups to play with this myself.
>
No problem. Yeah, in order to debug the patch, you need a numa machine
with a memory-less node. And unlucky, the patch can not work either by
grub bootup or kexec -l boot. There is nothing, just silent.  I will
dig into numa_register_memblks() to figure out the problem.

Thanks,
Pingfan
> diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> index 1308f5408bf7..d51643e10d00 100644
> --- a/arch/x86/mm/numa.c
> +++ b/arch/x86/mm/numa.c
> @@ -216,8 +216,6 @@ static void __init alloc_node_data(int nid)
>
>         node_data[nid] = nd;
>         memset(NODE_DATA(nid), 0, sizeof(pg_data_t));
> -
> -       node_set_online(nid);
>  }
>
>  /**
> @@ -527,6 +525,19 @@ static void __init numa_clear_kernel_node_hotplug(void)
>         }
>  }
>
> +static void __init init_memory_less_node(int nid)
> +{
> +       unsigned long zones_size[MAX_NR_ZONES] = {0};
> +       unsigned long zholes_size[MAX_NR_ZONES] = {0};
> +
> +       free_area_init_node(nid, zones_size, 0, zholes_size);
> +
> +       /*
> +        * All zonelists will be built later in start_kernel() after per cpu
> +        * areas are initialized.
> +        */
> +}
> +
>  static int __init numa_register_memblks(struct numa_meminfo *mi)
>  {
>         unsigned long uninitialized_var(pfn_align);
> @@ -570,7 +581,7 @@ static int __init numa_register_memblks(struct numa_meminfo *mi)
>                 return -EINVAL;
>
>         /* Finally register nodes. */
> -       for_each_node_mask(nid, node_possible_map) {
> +       for_each_node(nid) {
>                 u64 start = PFN_PHYS(max_pfn);
>                 u64 end = 0;
>
> @@ -592,6 +603,10 @@ static int __init numa_register_memblks(struct numa_meminfo *mi)
>                         continue;
>
>                 alloc_node_data(nid);
> +               if (!end)
> +                       init_memory_less_node(nid);
> +               else
> +                       node_set_online(nid);
>         }
>
>         /* Dump memblock with node info and return. */
> @@ -721,21 +736,6 @@ void __init x86_numa_init(void)
>         numa_init(dummy_numa_init);
>  }
>
> -static void __init init_memory_less_node(int nid)
> -{
> -       unsigned long zones_size[MAX_NR_ZONES] = {0};
> -       unsigned long zholes_size[MAX_NR_ZONES] = {0};
> -
> -       /* Allocate and initialize node data. Memory-less node is now online.*/
> -       alloc_node_data(nid);
> -       free_area_init_node(nid, zones_size, 0, zholes_size);
> -
> -       /*
> -        * All zonelists will be built later in start_kernel() after per cpu
> -        * areas are initialized.
> -        */
> -}
> -
>  /*
>   * Setup early cpu_to_node.
>   *
> @@ -763,9 +763,6 @@ void __init init_cpu_to_node(void)
>                 if (node == NUMA_NO_NODE)
>                         continue;
>
> -               if (!node_online(node))
> -                       init_memory_less_node(node);
> -
>                 numa_set_node(cpu, node);
>         }
>  }
> --
> Michal Hocko
> SUSE Labs