[RFC v2 PATCH] reserve_mem: add support for static memory

[RFC v2 PATCH] reserve_mem: add support for static memory

[Shyam Saini]

reserve_mem relies on dynamic memory allocation, this limits the
usecase where memory is required to be preserved across the boots.
Eg: ramoops memory reservation on ACPI platforms

So add support to pass a pre-determined static address and reserve
memory at a specified location. This enables use case like ramoops
on ACPI platforms to reliably access ramoops region with previous
boot logs.

Also skip the parsing of <align> when static address is passed.

Example syntax for static address
 reserve_mem=4M@0x1E0000000:oops

Signed-off-by: Shyam Saini <shyamsaini@linux.microsoft.com>
---
v1: https://lore.kernel.org/lkml/0eaf3be2-5121-48b7-aeed-196405c0a480@infradead.org/
v2: Fix code logic and incorporate Randy's suggestion
---
 .../admin-guide/kernel-parameters.txt         | 15 ++++++
 mm/memblock.c                                 | 47 +++++++++++++------
 2 files changed, 47 insertions(+), 15 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index b5493a7f8f228..7e0baca564b97 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -6563,6 +6563,21 @@ Kernel parameters
 
 			reserve_mem=12M:4096:oops ramoops.mem_name=oops
 
+	reserve_mem=	[RAM]
+			Format: nn[KMG]:<@offset>:<label>
+			Reserve physical memory at predetermined location and label it with
+			a name that other subsystems can use to access it. This is typically
+			used for systems that do not wipe the RAM, and this command
+			line will try to reserve the same physical memory on
+			soft reboots. Note, it is guaranteed to be the same
+			location unless some other early allocation, e.g.: crashkernel=256M
+                        (without static address) is reserved or overlaps this region.
+
+			The format is size:offset:label for example, to request
+			4 megabytes for ramoops at 0x1E0000000:
+
+			reserve_mem=4M@0x1E0000000:oops ramoops.mem_name=oops
+
 	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 6349c48154f4b..c76cefa0a8a83 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -2721,6 +2721,7 @@ static int __init reserve_mem(char *p)
 	char *name;
 	char *oldp;
 	int len;
+	bool addr_is_static = false;
 
 	if (!p)
 		goto err_param;
@@ -2736,19 +2737,27 @@ static int __init reserve_mem(char *p)
 	if (!size || p == oldp)
 		goto err_param;
 
-	if (*p != ':')
-		goto err_param;
+	/* parse the static memory address */
+	if (*p == '@') {
+		start = memparse(p+1, &p);
+		addr_is_static = true;
+	}
 
-	align = memparse(p+1, &p);
 	if (*p != ':')
 		goto err_param;
 
-	/*
-	 * 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 (!addr_is_static) {
+		align = memparse(p+1, &p);
+		if (*p != ':')
+			goto err_param;
+
+		/*
+		 * 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;
+	}
 
 	name = p + 1;
 	len = strlen(name);
@@ -2772,14 +2781,22 @@ static int __init reserve_mem(char *p)
 	}
 
 	/* Pick previous allocations up from KHO if available */
-	if (reserve_mem_kho_revive(name, size, align))
+	if (!addr_is_static && reserve_mem_kho_revive(name, size, align))
 		return 1;
 
-	/* TODO: Allocation must be outside of scratch region */
-	start = memblock_phys_alloc(size, align);
-	if (!start) {
-		pr_err("reserve_mem: memblock allocation failed\n");
-		return -ENOMEM;
+	if (addr_is_static) {
+		if (memblock_reserve(start, size)) {
+			pr_err("reserve_mem: memblock reservation failed\n");
+			return -ENOMEM;
+		}
+
+	} else {
+		/* TODO: Allocation must be outside of scratch region */
+		start = memblock_phys_alloc(size, align);
+		if (!start) {
+			pr_err("reserve_mem: memblock allocation failed\n");
+			return -ENOMEM;
+		}
 	}
 
 	reserved_mem_add(start, size, name);
-- 
2.43.0

Re: [RFC v2 PATCH] reserve_mem: add support for static memory

[Randy Dunlap]

Hi,

On 6/18/26 11:23 PM, Shyam Saini wrote:
> reserve_mem relies on dynamic memory allocation, this limits the
> usecase where memory is required to be preserved across the boots.
> Eg: ramoops memory reservation on ACPI platforms
> 
> So add support to pass a pre-determined static address and reserve
> memory at a specified location. This enables use case like ramoops
> on ACPI platforms to reliably access ramoops region with previous
> boot logs.
> 
> Also skip the parsing of <align> when static address is passed.
> 
> Example syntax for static address
>  reserve_mem=4M@0x1E0000000:oops
> 
> Signed-off-by: Shyam Saini <shyamsaini@linux.microsoft.com>
> ---
> v1: https://lore.kernel.org/lkml/0eaf3be2-5121-48b7-aeed-196405c0a480@infradead.org/
> v2: Fix code logic and incorporate Randy's suggestion

OK, you fixed a few typos.
There are some bigger things that you seem to have ignored.

> ---
>  .../admin-guide/kernel-parameters.txt         | 15 ++++++
>  mm/memblock.c                                 | 47 +++++++++++++------
>  2 files changed, 47 insertions(+), 15 deletions(-)
> 
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index b5493a7f8f228..7e0baca564b97 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -6563,6 +6563,21 @@ Kernel parameters
>  
>  			reserve_mem=12M:4096:oops ramoops.mem_name=oops
>  
> +	reserve_mem=	[RAM]

[RAM] means "RAM disk support is enabled."
Is that the case here?  Is "reserve_mem=" only for use in case
RAM disk support is enabled?

ISTM that you need a new designator instead of RAM...
or overload the use of RAM by adding more info near the top of
Documentation/admin-guide/kernel-parameters.txt.


> +			Format: nn[KMG]:<@offset>:<label>
> +			Reserve physical memory at predetermined location and label it with
> +			a name that other subsystems can use to access it. This is typically
> +			used for systems that do not wipe the RAM, and this command
> +			line will try to reserve the same physical memory on
> +			soft reboots. Note, it is guaranteed to be the same
> +			location unless some other early allocation, e.g.: crashkernel=256M
> +                        (without static address) is reserved or overlaps this region.
> +
> +			The format is size:offset:label for example, to request
> +			4 megabytes for ramoops at 0x1E0000000:
> +
> +			reserve_mem=4M@0x1E0000000:oops ramoops.mem_name=oops
> +
>  	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 6349c48154f4b..c76cefa0a8a83 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -2721,6 +2721,7 @@ static int __init reserve_mem(char *p)
>  	char *name;
>  	char *oldp;
>  	int len;
> +	bool addr_is_static = false;
>  
>  	if (!p)
>  		goto err_param;
> @@ -2736,19 +2737,27 @@ static int __init reserve_mem(char *p)
>  	if (!size || p == oldp)
>  		goto err_param;
>  
> -	if (*p != ':')
> -		goto err_param;
> +	/* parse the static memory address */
> +	if (*p == '@') {
> +		start = memparse(p+1, &p);
> +		addr_is_static = true;
> +	}
>  
> -	align = memparse(p+1, &p);
>  	if (*p != ':')
>  		goto err_param;
>  
> -	/*
> -	 * 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 (!addr_is_static) {
> +		align = memparse(p+1, &p);
> +		if (*p != ':')
> +			goto err_param;
> +
> +		/*
> +		 * 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;
> +	}
>  
>  	name = p + 1;
>  	len = strlen(name);
> @@ -2772,14 +2781,22 @@ static int __init reserve_mem(char *p)
>  	}
>  
>  	/* Pick previous allocations up from KHO if available */
> -	if (reserve_mem_kho_revive(name, size, align))
> +	if (!addr_is_static && reserve_mem_kho_revive(name, size, align))
>  		return 1;
>  
> -	/* TODO: Allocation must be outside of scratch region */
> -	start = memblock_phys_alloc(size, align);
> -	if (!start) {
> -		pr_err("reserve_mem: memblock allocation failed\n");
> -		return -ENOMEM;

		return 1;

> +	if (addr_is_static) {
> +		if (memblock_reserve(start, size)) {
> +			pr_err("reserve_mem: memblock reservation failed\n");
> +			return -ENOMEM;

			return 1;

> +		}
> +
> +	} else {
> +		/* TODO: Allocation must be outside of scratch region */
> +		start = memblock_phys_alloc(size, align);
> +		if (!start) {
> +			pr_err("reserve_mem: memblock allocation failed\n");
> +			return -ENOMEM;

			return 1;

> +		}
>  	}
>  
>  	reserved_mem_add(start, size, name);


__setup() functions return 1 for "yes, I recognized this string/option
and attempted to handle it" or 0 for "This string/option is meaningless."
There is no "return -Eerror".
If you need that, you could consider using early_param() [see
<linux/init.h>].

-- 
~Randy

Re: [RFC v2 PATCH] reserve_mem: add support for static memory

[Mike Rapoport]

On Thu, Jun 18, 2026 at 11:23:31PM -0700, Shyam Saini wrote:
> reserve_mem relies on dynamic memory allocation, this limits the
> usecase where memory is required to be preserved across the boots.
> Eg: ramoops memory reservation on ACPI platforms
>
> So add support to pass a pre-determined static address and reserve
> memory at a specified location. This enables use case like ramoops
> on ACPI platforms to reliably access ramoops region with previous
> boot logs.
> 
> Also skip the parsing of <align> when static address is passed.
> 
> Example syntax for static address
>  reserve_mem=4M@0x1E0000000:oops

reserve_mem is best effort by design because such hacks as well as memmap=
cannot guarantee this memory is actually free.

If you want to preserve ramoops reliably, use KHO with reserve_mem.
The first kernel will allocate memory, this memory will be preserved by KHO
and could be picked up by the second kernel.
 
> Signed-off-by: Shyam Saini <shyamsaini@linux.microsoft.com>
> ---
> v1: https://lore.kernel.org/lkml/0eaf3be2-5121-48b7-aeed-196405c0a480@infradead.org/
> v2: Fix code logic and incorporate Randy's suggestion
> ---
>  .../admin-guide/kernel-parameters.txt         | 15 ++++++
>  mm/memblock.c                                 | 47 +++++++++++++------
>  2 files changed, 47 insertions(+), 15 deletions(-)

-- 
Sincerely yours,
Mike.

Re: [RFC v2 PATCH] reserve_mem: add support for static memory

[Shyam Saini]

Hi,


On 19 Jun 2026 11:35, Randy Dunlap wrote:
> Hi,
> 
> On 6/18/26 11:23 PM, Shyam Saini wrote:
> > reserve_mem relies on dynamic memory allocation, this limits the
> > usecase where memory is required to be preserved across the boots.
> > Eg: ramoops memory reservation on ACPI platforms
> > 
> > So add support to pass a pre-determined static address and reserve
> > memory at a specified location. This enables use case like ramoops
> > on ACPI platforms to reliably access ramoops region with previous
> > boot logs.
> > 
> > Also skip the parsing of <align> when static address is passed.
> > 
> > Example syntax for static address
> >  reserve_mem=4M@0x1E0000000:oops
> > 
> > Signed-off-by: Shyam Saini <shyamsaini@linux.microsoft.com>
> > ---
> > v1: https://lore.kernel.org/lkml/0eaf3be2-5121-48b7-aeed-196405c0a480@infradead.org/
> > v2: Fix code logic and incorporate Randy's suggestion
> 
> OK, you fixed a few typos.
> There are some bigger things that you seem to have ignored.

Thanks for calling this out. You are right that I did not address all
comments in v2.
My goal for v2 was to quickly fix the core logic issue and keep
discussion focused on the reserve_mem static address direction in this
RFC cycle. I should have stated that clearly.
 
> > ---
> >  .../admin-guide/kernel-parameters.txt         | 15 ++++++
> >  mm/memblock.c                                 | 47 +++++++++++++------
> >  2 files changed, 47 insertions(+), 15 deletions(-)
> > 
> > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > index b5493a7f8f228..7e0baca564b97 100644
> > --- a/Documentation/admin-guide/kernel-parameters.txt
> > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > @@ -6563,6 +6563,21 @@ Kernel parameters
> >  
> >  			reserve_mem=12M:4096:oops ramoops.mem_name=oops
> >  
> > +	reserve_mem=	[RAM]
> 
> [RAM] means "RAM disk support is enabled."
> Is that the case here?  Is "reserve_mem=" only for use in case
> RAM disk support is enabled?
> 
> ISTM that you need a new designator instead of RAM...
> or overload the use of RAM by adding more info near the top of
> Documentation/admin-guide/kernel-parameters.txt.

will address them in future iterations
> 
> > +			Format: nn[KMG]:<@offset>:<label>
> > +			Reserve physical memory at predetermined location and label it with
> > +			a name that other subsystems can use to access it. This is typically
> > +			used for systems that do not wipe the RAM, and this command
> > +			line will try to reserve the same physical memory on
> > +			soft reboots. Note, it is guaranteed to be the same
> > +			location unless some other early allocation, e.g.: crashkernel=256M
> > +                        (without static address) is reserved or overlaps this region.
> > +
> > +			The format is size:offset:label for example, to request
> > +			4 megabytes for ramoops at 0x1E0000000:
> > +
> > +			reserve_mem=4M@0x1E0000000:oops ramoops.mem_name=oops
> > +
> >  	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 6349c48154f4b..c76cefa0a8a83 100644
> > --- a/mm/memblock.c
> > +++ b/mm/memblock.c
> > @@ -2721,6 +2721,7 @@ static int __init reserve_mem(char *p)
> >  	char *name;
> >  	char *oldp;
> >  	int len;
> > +	bool addr_is_static = false;
> >  
> >  	if (!p)
> >  		goto err_param;
> > @@ -2736,19 +2737,27 @@ static int __init reserve_mem(char *p)
> >  	if (!size || p == oldp)
> >  		goto err_param;
> >  
> > -	if (*p != ':')
> > -		goto err_param;
> > +	/* parse the static memory address */
> > +	if (*p == '@') {
> > +		start = memparse(p+1, &p);
> > +		addr_is_static = true;
> > +	}
> >  
> > -	align = memparse(p+1, &p);
> >  	if (*p != ':')
> >  		goto err_param;
> >  
> > -	/*
> > -	 * 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 (!addr_is_static) {
> > +		align = memparse(p+1, &p);
> > +		if (*p != ':')
> > +			goto err_param;
> > +
> > +		/*
> > +		 * 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;
> > +	}
> >  
> >  	name = p + 1;
> >  	len = strlen(name);
> > @@ -2772,14 +2781,22 @@ static int __init reserve_mem(char *p)
> >  	}
> >  
> >  	/* Pick previous allocations up from KHO if available */
> > -	if (reserve_mem_kho_revive(name, size, align))
> > +	if (!addr_is_static && reserve_mem_kho_revive(name, size, align))
> >  		return 1;
> >  
> > -	/* TODO: Allocation must be outside of scratch region */
> > -	start = memblock_phys_alloc(size, align);
> > -	if (!start) {
> > -		pr_err("reserve_mem: memblock allocation failed\n");
> > -		return -ENOMEM;
> 
> 		return 1;
> 
> > +	if (addr_is_static) {
> > +		if (memblock_reserve(start, size)) {
> > +			pr_err("reserve_mem: memblock reservation failed\n");
> > +			return -ENOMEM;
> 
> 			return 1;
> 
> > +		}
> > +
> > +	} else {
> > +		/* TODO: Allocation must be outside of scratch region */
> > +		start = memblock_phys_alloc(size, align);
> > +		if (!start) {
> > +			pr_err("reserve_mem: memblock allocation failed\n");
> > +			return -ENOMEM;
> 
> 			return 1;
> 
> > +		}
> >  	}
> >  
> >  	reserved_mem_add(start, size, name);
> 
> 
> __setup() functions return 1 for "yes, I recognized this string/option
> and attempted to handle it" or 0 for "This string/option is meaningless."
> There is no "return -Eerror".
> If you need that, you could consider using early_param() [see
> <linux/init.h>].
> 
same for this concern, will address them in next iteration.

Thanks,
Shyam

Re: [RFC v2 PATCH] reserve_mem: add support for static memory

[Shyam Saini]

Hi Mike,

On 21 Jun 2026 13:36, Mike Rapoport wrote:
> On Thu, Jun 18, 2026 at 11:23:31PM -0700, Shyam Saini wrote:
> > reserve_mem relies on dynamic memory allocation, this limits the
> > usecase where memory is required to be preserved across the boots.
> > Eg: ramoops memory reservation on ACPI platforms
> >
> > So add support to pass a pre-determined static address and reserve
> > memory at a specified location. This enables use case like ramoops
> > on ACPI platforms to reliably access ramoops region with previous
> > boot logs.
> > 
> > Also skip the parsing of <align> when static address is passed.
> > 
> > Example syntax for static address
> >  reserve_mem=4M@0x1E0000000:oops
> 
> reserve_mem is best effort by design because such hacks as well as memmap=
> cannot guarantee this memory is actually free.
> 
> If you want to preserve ramoops reliably, use KHO with reserve_mem.
> The first kernel will allocate memory, this memory will be preserved by KHO
> and could be picked up by the second kernel.

ok, On ARM64 DTS systems, we can reserve ramoops memory in the device tree during
the warm reboot.
For an equivalent ARM64 ACPI platform, what is the recommended way to reserve
and preserve that memory across the boots? 

> > Signed-off-by: Shyam Saini <shyamsaini@linux.microsoft.com>
> > ---
> > v1: https://lore.kernel.org/lkml/0eaf3be2-5121-48b7-aeed-196405c0a480@infradead.org/
> > v2: Fix code logic and incorporate Randy's suggestion
> > ---
> >  .../admin-guide/kernel-parameters.txt         | 15 ++++++
> >  mm/memblock.c                                 | 47 +++++++++++++------
> >  2 files changed, 47 insertions(+), 15 deletions(-)
> 
> -- 
> Sincerely yours,
> Mike.

Thanks,
Shyam

Re: [RFC v2 PATCH] reserve_mem: add support for static memory

[Mike Rapoport]

Hi Shyam,

On Wed, Jun 24, 2026 at 06:22:33PM -0700, Shyam Saini wrote:
> On 21 Jun 2026 13:36, Mike Rapoport wrote:
> > On Thu, Jun 18, 2026 at 11:23:31PM -0700, Shyam Saini wrote:
> > > reserve_mem relies on dynamic memory allocation, this limits the
> > > usecase where memory is required to be preserved across the boots.
> > > Eg: ramoops memory reservation on ACPI platforms
> > >
> > > So add support to pass a pre-determined static address and reserve
> > > memory at a specified location. This enables use case like ramoops
> > > on ACPI platforms to reliably access ramoops region with previous
> > > boot logs.
> > > 
> > > Also skip the parsing of <align> when static address is passed.
> > > 
> > > Example syntax for static address
> > >  reserve_mem=4M@0x1E0000000:oops
> > 
> > reserve_mem is best effort by design because such hacks as well as memmap=
> > cannot guarantee this memory is actually free.
> > 
> > If you want to preserve ramoops reliably, use KHO with reserve_mem.
> > The first kernel will allocate memory, this memory will be preserved by KHO
> > and could be picked up by the second kernel.
> 
> ok, On ARM64 DTS systems, we can reserve ramoops memory in the device tree during
> the warm reboot.

The cc list actually implied x86 ;-)
Added arm64 folks now.

> For an equivalent ARM64 ACPI platform, what is the recommended way to reserve
> and preserve that memory across the boots? 

I don't think it exists, but a command line option (be it memmap= or
reserve_mem=) does not seem the right way to me.

Most of the arguments that were made against adding memmap= to arm64 [1]
apply here.

If kexec is an option, KHO provides a reliable way to preserve memory
across boots.

If kexec is not an option, we should look for a generic way to specify
something like DT's reserved_mem for ACPI/EFI systems.

[1] https://lkml.kernel.org/lkml/20201118063314.22940-1-song.bao.hua@hisilicon.com/T/

> Thanks,
> Shyam

-- 
Sincerely yours,
Mike.