[PATCH v2] memblock: reject non-page-aligned reserve_mem regions

[PATCH v2] memblock: reject non-page-aligned reserve_mem regions

[Sang-Heon Jeon]

When map->start or map->size is not page aligned,
reserved_mem_preserve() does not preserve the last one or two pages of
the region from map->start to map->start + map->size.

After kexec the new kernel reserves only the preserved pages, so those
pages are released to the buddy allocator. But
reserve_mem_find_by_name() still returns the whole region including
them, so they can be allocated while the subsystem is still using them.

So reject a reserve_mem= size or alignment that is not a multiple of
PAGE_SIZE to keep both map->start and map->size page aligned.

Fixes: 8375b76517cb ("kho: replace kho_preserve_phys() with kho_preserve_pages()")
Signed-off-by: Sang-Heon Jeon <ekffu200098@gmail.com>
---
Changes from v1 [1]
- reject non-page-aligned size and align instead of fix page count when
  preserving reserve_mem regions

[1] https://lore.kernel.org/all/20260901165237.1025973-1-ekffu200098@gmail.com/
---
 Documentation/admin-guide/kernel-parameters.txt | 3 +++
 mm/memblock.c                                   | 9 +++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 68647ff4bdd2..f969085fffb7 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -6648,6 +6648,9 @@ Kernel parameters
 
 			reserve_mem=12M:4096:oops ramoops.mem_name=oops
 
+			Both the size and the alignment must be multiples of
+			the page size.
+
 	reservetop=	[X86-32,EARLY]
 			Format: nn[KMG]
 			Reserves a hole at the top of the kernel virtual
diff --git a/mm/memblock.c b/mm/memblock.c
index 4302fb4ab85c..5e93b6da53e3 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -2816,12 +2816,17 @@ static int __init reserve_mem(char *p)
 	if (*p != ':')
 		goto err_param;
 
+	if (!IS_ALIGNED(size, PAGE_SIZE) || !IS_ALIGNED(align, PAGE_SIZE)) {
+		pr_err("reserve_mem: size and align must be multiples of the page size\n");
+		return -EINVAL;
+	}
+
 	/*
 	 * memblock_phys_alloc() doesn't like a zero size align,
 	 * but it is OK for this command to have it.
 	 */
-	if (align < SMP_CACHE_BYTES)
-		align = SMP_CACHE_BYTES;
+	if (!align)
+		align = PAGE_SIZE;
 
 	name = p + 1;
 	len = strlen(name);

base-commit: dbffc67777cd752ef3f77818109e1d2ef13e2949
-- 
2.43.0

Re: [PATCH v2] memblock: reject non-page-aligned reserve_mem regions

[Mike Rapoport]

(adding tracing and pstore folks)

On Sat, Sep 05, 2026 at 11:15:24PM +0900, Sang-Heon Jeon wrote:
> When map->start or map->size is not page aligned,
> reserved_mem_preserve() does not preserve the last one or two pages of
> the region from map->start to map->start + map->size.
> 
> After kexec the new kernel reserves only the preserved pages, so those
> pages are released to the buddy allocator. But
> reserve_mem_find_by_name() still returns the whole region including
> them, so they can be allocated while the subsystem is still using them.
> 
> So reject a reserve_mem= size or alignment that is not a multiple of
> PAGE_SIZE to keep both map->start and map->size page aligned.
> 
> Fixes: 8375b76517cb ("kho: replace kho_preserve_phys() with kho_preserve_pages()")
> Signed-off-by: Sang-Heon Jeon <ekffu200098@gmail.com>
> ---
> Changes from v1 [1]
> - reject non-page-aligned size and align instead of fix page count when
>   preserving reserve_mem regions
> 
> [1] https://lore.kernel.org/all/20260901165237.1025973-1-ekffu200098@gmail.com/
> ---
>  Documentation/admin-guide/kernel-parameters.txt | 3 +++
>  mm/memblock.c                                   | 9 +++++++--
>  2 files changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 68647ff4bdd2..f969085fffb7 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -6648,6 +6648,9 @@ Kernel parameters
>  
>  			reserve_mem=12M:4096:oops ramoops.mem_name=oops
>  
> +			Both the size and the alignment must be multiples of
> +			the page size.
> +
>  	reservetop=	[X86-32,EARLY]
>  			Format: nn[KMG]
>  			Reserves a hole at the top of the kernel virtual
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 4302fb4ab85c..5e93b6da53e3 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -2816,12 +2816,17 @@ static int __init reserve_mem(char *p)
>  	if (*p != ':')
>  		goto err_param;
>  
> +	if (!IS_ALIGNED(size, PAGE_SIZE) || !IS_ALIGNED(align, PAGE_SIZE)) {
> +		pr_err("reserve_mem: size and align must be multiples of the page size\n");
> +		return -EINVAL;
> +	}
> +
>  	/*
>  	 * memblock_phys_alloc() doesn't like a zero size align,
>  	 * but it is OK for this command to have it.
>  	 */
> -	if (align < SMP_CACHE_BYTES)
> -		align = SMP_CACHE_BYTES;
> +	if (!align)
> +		align = PAGE_SIZE;
>  
>  	name = p + 1;
>  	len = strlen(name);
> 
> base-commit: dbffc67777cd752ef3f77818109e1d2ef13e2949
> -- 
> 2.43.0
> 

-- 
Sincerely yours,
Mike.