Re: [PATCH V4 10/15] xen/page_alloc: vmap heap nodes when they are outside the direct map

["Alejandro Vallejo" <alejandro.vallejo@cloud.com>]

Hi,

I'm seeing crashes on NUMA machines, which can be attributed to a bug below:

On Mon Nov 11, 2024 at 1:11 PM GMT, Elias El Yandouzi wrote:
> From: Hongyan Xia <hongyxia@amazon.com>
>
> When we do not have a direct map, archs_mfn_in_direct_map() will always
> return false, thus init_node_heap() will allocate xenheap pages from an
> existing node for the metadata of a new node. This means that the
> metadata of a new node is in a different node, slowing down heap
> allocation.
>
> Since we now have early vmap, vmap the metadata locally in the new node.
>
> Signed-off-by: Hongyan Xia <hongyxia@amazon.com>
> Signed-off-by: Julien Grall <jgrall@amazon.com>
> Signed-off-by: Elias El Yandouzi <eliasely@amazon.com>
>
> ----
>
>     Changes in v4:
>         * Change type of the parameters to paddr_t
>         * Use clear_domain_page() instead of open-coding it
>
>     Changes in v2:
>         * vmap_contig_pages() was renamed to vmap_contig()
>         * Fix indentation and coding style
>
>     Changes from Hongyan's version:
>         * arch_mfn_in_direct_map() was renamed to
>           arch_mfns_in_direct_map()
>         * Use vmap_contig_pages() rather than __vmap(...).
>         * Add missing include (xen/vmap.h) so it compiles on Arm
>
> diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
> index 2cef521ad85a..62cdeb5013a3 100644
> --- a/xen/common/page_alloc.c
> +++ b/xen/common/page_alloc.c
> @@ -137,6 +137,7 @@
>  #include <xen/sections.h>
>  #include <xen/softirq.h>
>  #include <xen/spinlock.h>
> +#include <xen/vmap.h>
>  
>  #include <asm/flushtlb.h>
>  #include <asm/page.h>
> @@ -606,22 +607,32 @@ static unsigned long init_node_heap(int node, unsigned long mfn,
>          needed = 0;
>      }
>      else if ( *use_tail && nr >= needed &&
> -              arch_mfns_in_directmap(mfn + nr - needed, needed) &&
>                (!xenheap_bits ||
>                 !((mfn + nr - 1) >> (xenheap_bits - PAGE_SHIFT))) )
>      {
> -        _heap[node] = mfn_to_virt(mfn + nr - needed);
> -        avail[node] = mfn_to_virt(mfn + nr - 1) +
> -                      PAGE_SIZE - sizeof(**avail) * NR_ZONES;
> +        if ( arch_mfns_in_directmap(mfn + nr - needed, needed) )
> +            _heap[node] = mfn_to_virt(mfn + nr - needed);
> +        else
> +            _heap[node] = vmap_contig(_mfn(mfn + nr - needed), needed);
> +
> +        BUG_ON(!_heap[node]);
> +        avail[node] = (void *)(_heap[node]) + (needed << PAGE_SHIFT) -
> +                        sizeof(**avail) * NR_ZONES;
> +
>      }
>      else if ( nr >= needed &&
> -              arch_mfns_in_directmap(mfn, needed) &&
>                (!xenheap_bits ||
> -               !((mfn + needed - 1) >> (xenheap_bits - PAGE_SHIFT))) )
> +              !((mfn + needed - 1) >> (xenheap_bits - PAGE_SHIFT))) )
>      {
> -        _heap[node] = mfn_to_virt(mfn);
> -        avail[node] = mfn_to_virt(mfn + needed - 1) +
> -                      PAGE_SIZE - sizeof(**avail) * NR_ZONES;
> +        if ( arch_mfns_in_directmap(mfn + nr - needed, needed) )
> +            _heap[node] = mfn_to_virt(mfn + nr - needed);
> +        else
> +            _heap[node] = vmap_contig(_mfn(mfn + nr - needed), needed);

This isn't quite the same thing, I think it regressed in v4 when acting on
Roger's feedback. It should be:

   if ( arch_mfns_in_directmap(mfn, needed) )
       _heap[node] = mfn_to_virt(mfn);
   else
       _heap[node] = vmap_contig(_mfn(mfn), needed);

Otherwise `use_tail` serves is unconditionally considered as set. With this
change in place, I can boot on NUMA machines.

> +
> +        BUG_ON(!_heap[node]);
> +        avail[node] = (void *)(_heap[node]) + (needed << PAGE_SHIFT) -
> +                        sizeof(**avail) * NR_ZONES;
> +
>          *use_tail = false;
>      }
>      else if ( get_order_from_bytes(sizeof(**_heap)) ==

Cheers,
Alejandro