Re: [PATCH] kho: fix restore order for NUMA-split preservations

[Jose A. Perez de Azpillaga]

On 11/5/26 2:21 PM, Mukesh Pilaniya wrote:
> kho_preserve_pages() uses __kho_preserve_pages_order() to reduce the
> order of a preservation so that it does not span multiple NUMA nodes.
> kho_restore_pages() does not use that helper; it recomputes the order
> from the pfn alignment and the remaining length only.
> 
> When a preservation was split at a NUMA boundary, restore picks a larger
> order than preserve did and jumps over the next key. Those pages never
> get restored: they stay memblock-reserved and keep the KHOP magic in
> page->private, while kho_restore_pages() returns success for the pages
> that were restored. The consumer has no way to notice that part of its
> buffer was not restored.
> 
> Fix this by reading the order from the key being restored. The order is
> already stored in page->private by kho_preserved_memory_reserve() and is
> validated by kho_restore_page(). Advance by that order instead of
> recomputing it. This also keeps restore independent of the incoming
> kernel's NUMA topology.
> 
> Example: QEMU is started with two NUMA nodes:
> 
>   node0: [0x40000000, 0xC0400000)   2052 MiB
>   node1: [0xC0400000, 0x1C0000000)  4092 MiB
> 
> The module preserves [0xC0000000, 0xC0800000) (8 MiB, 2048 pages). The
> range is 8 MiB aligned and the node boundary sits exactly 4 MiB into it, so
> preserve emits two order-10 keys, the first at 0xC0000000 and the second at
> 0xC0400000. Restore computes order 11 for the first page, restores the
> first key and jumps straight to the end of the range, skipping the second
> key.
> 
> Fixes: 91e74fa8b1bc ("kho: make sure preservations do not span multiple NUMA nodes")
> Signed-off-by: Mukesh Pilaniya <mpilaniy@xxxxxxxxxx>
> ---
>  kernel/liveupdate/kexec_handover.c | 16 ++++++++++------
>  1 file changed, 10 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/liveupdate/kexec_handover.c b/kernel/liveupdate/kexec_handover.c
> index 6f1f55fe3181..7d30ad9cc512 100644
> --- a/kernel/liveupdate/kexec_handover.c
> +++ b/kernel/liveupdate/kexec_handover.c
> @@ -511,7 +511,8 @@ static void kho_init_folio(struct page *page, unsigned int order)
>  		prep_compound_page(page, order);
>  }
>  
> -static struct page *kho_restore_page(phys_addr_t phys, bool is_folio)
> +static struct page *kho_restore_page(phys_addr_t phys, bool is_folio,
> +				     unsigned int *order)
>  {
>  	struct page *page = pfn_to_online_page(PHYS_PFN(phys));
>  	unsigned long nr_pages;
> @@ -530,6 +531,9 @@ static struct page *kho_restore_page(phys_addr_t phys, bool is_folio)
>  		return NULL;
>  	nr_pages = (1 << info.order);
>  
> +	if (order)
> +		*order = info.order;
> +
>  	/* Clear private to make sure later restores on this page error out. */
>  	page->private = 0;
>  
> @@ -550,7 +554,7 @@ static struct page *kho_restore_page(phys_addr_t phys, bool is_folio)
>   */
>  struct folio *kho_restore_folio(phys_addr_t phys)
>  {
> -	struct page *page = kho_restore_page(phys, true);
> +	struct page *page = kho_restore_page(phys, true, NULL);
>  
>  	return page ? page_folio(page) : NULL;
>  }
> @@ -573,13 +577,13 @@ struct page *kho_restore_pages(phys_addr_t phys, unsigned long nr_pages)
>  	unsigned long pfn = start_pfn;
>  
>  	while (pfn < end_pfn) {
> -		const unsigned int order =
> -			min(count_trailing_zeros(pfn), ilog2(end_pfn - pfn));
> -		struct page *page = kho_restore_page(PFN_PHYS(pfn), false);
> +		unsigned int order;
> +		struct page *page;
>  
> +		page = kho_restore_page(PFN_PHYS(pfn), false, &order);
>  		if (!page)
>  			return NULL;
> -		pfn += 1 << order;
> +		pfn += 1UL << order;
>  	}
>  
>  	return pfn_to_page(start_pfn);
> 

LGTM:

Reviewed-by: Jose A. Perez de Azpillaga <azpijr@xxxxxxxxx>

-- 
cheers,
jose a. p-a