[PATCH V2 0/2] Small improvements to reserve_mem, take 2

[PATCH V2 0/2] Small improvements to reserve_mem, take 2

[Guilherme G. Piccoli]

Hi folks, this is the 2nd version of the patchset, V1 is
here: https://lore.kernel.org/r/20260217195816.861684-1-gpiccoli@igalia.com/

So, quick summary: we are adding
(a) Error messages if "reserve_mem=" fails;

(b) A debugfs file in case "reserve_mem=" succeeds, showing name and
size of the allocation.

Use case of (a) is kinda obvious - we want to know if there are failures.
Use case for (b) are users or tools like kdumpst[0] that might want to
check if their reserve_mem allocation(s) succeeded in order to load
ramoops with proper parameters, using the reserved areas.

Thanks a lot Mike for the great review so far! Questions or comments
are welcome,

Guilherme G. Piccoli (2):
  mm/memblock: Print out errors on reserve_mem parser
  mm/memblock: Add reserve_mem debugfs info

 mm/memblock.c | 65 ++++++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 54 insertions(+), 11 deletions(-)


[0] https://aur.archlinux.org/packages/kdumpst

-- 
2.50.1

[PATCH V2 1/2] mm/memblock: Print out errors on reserve_mem parser

[Guilherme G. Piccoli]

The parsing of kernel parameter "reserve_mem=" is subject to
multiple failures, like duplicate naming, malformed expression
or even lack of available memory. Right now, all of these fail
silently. Let's add some messages so the kernel log can provide
useful information in case of failures.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Guilherme G. Piccoli <gpiccoli@igalia.com>
---


V2: no changes.

 mm/memblock.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)


diff --git a/mm/memblock.c b/mm/memblock.c
index b3ddfdec7a80..2d2646f7a120 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -2642,23 +2642,25 @@ static int __init reserve_mem(char *p)
 	int len;
 
 	if (!p)
-		return -EINVAL;
+		goto err_param;
 
 	/* Check if there's room for more reserved memory */
-	if (reserved_mem_count >= RESERVE_MEM_MAX_ENTRIES)
+	if (reserved_mem_count >= RESERVE_MEM_MAX_ENTRIES) {
+		pr_err("reserve_mem: no more room for reserved memory\n");
 		return -EBUSY;
+	}
 
 	oldp = p;
 	size = memparse(p, &p);
 	if (!size || p == oldp)
-		return -EINVAL;
+		goto err_param;
 
 	if (*p != ':')
-		return -EINVAL;
+		goto err_param;
 
 	align = memparse(p+1, &p);
 	if (*p != ':')
-		return -EINVAL;
+		goto err_param;
 
 	/*
 	 * memblock_phys_alloc() doesn't like a zero size align,
@@ -2672,7 +2674,7 @@ static int __init reserve_mem(char *p)
 
 	/* name needs to have length but not too big */
 	if (!len || len >= RESERVE_MEM_NAME_SIZE)
-		return -EINVAL;
+		goto err_param;
 
 	/* Make sure that name has text */
 	for (p = name; *p; p++) {
@@ -2680,11 +2682,13 @@ static int __init reserve_mem(char *p)
 			break;
 	}
 	if (!*p)
-		return -EINVAL;
+		goto err_param;
 
 	/* Make sure the name is not already used */
-	if (reserve_mem_find_by_name(name, &start, &tmp))
+	if (reserve_mem_find_by_name(name, &start, &tmp)) {
+		pr_err("reserve_mem: name \"%s\" was already used\n", name);
 		return -EBUSY;
+	}
 
 	/* Pick previous allocations up from KHO if available */
 	if (reserve_mem_kho_revive(name, size, align))
@@ -2692,12 +2696,18 @@ static int __init reserve_mem(char *p)
 
 	/* TODO: Allocation must be outside of scratch region */
 	start = memblock_phys_alloc(size, align);
-	if (!start)
+	if (!start) {
+		pr_err("reserve_mem: memblock allocation failed\n");
 		return -ENOMEM;
+	}
 
 	reserved_mem_add(start, size, name);
 
 	return 1;
+err_param:
+	pr_err("reserve_mem: empty or malformed parameter\n");
+	return -EINVAL;
+
 }
 __setup("reserve_mem=", reserve_mem);
 
-- 
2.50.1

[PATCH V2 2/2] mm/memblock: Add reserve_mem debugfs info

[Guilherme G. Piccoli]

When using the "reserve_mem" parameter, users aim at having an
area that (hopefully) persists across boots, so pstore infrastructure
(like ramoops module) can make use of that to save oops/ftrace logs,
for example.

There is no easy way to determine if this kernel parameter is properly
set though; the kernel doesn't show information about this memory in
memblock debugfs, neither in /proc/iomem nor dmesg. This is a relevant
information for tools like kdumpst[0], to determine if it's reliable
to use the reserved area as ramoops persistent storage; checking only
/proc/cmdline is not sufficient as it doesn't tell if the reservation
effectively succeeded or not.

Add here a new file under memblock debugfs showing properly set memory
reservations, with name and size as passed to "reserve_mem". Notice that
if no "reserve_mem=" is passed on command-line or if the reservation
attempts fail, the file is not created.

[0] https://aur.archlinux.org/packages/kdumpst

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Guilherme G. Piccoli <gpiccoli@igalia.com>
---


Thanks a lot for the suggestions Mike! I'm not sure if you would
prefer a Co-Developed-by or Suggested-by instead of CC, you helped
a lot improving the code. Lemme know and either I can re-submit
(with potential other changes) or even, you can change while merging.

V2: (all suggestions by Mike Rapoport)

- Commit message (showing use case);
- Drop ifdef on include "string_helpers.h";
- Don't show the address of reserve_mem, only name and size;
- Fixed flag names inside ARCH_KEEP_MEMBLOCK ifdef;
- Make use of its own show_attribute instead of refactoring
  the memblock one;
- Use sizeof() instead of magical numbers for the size;
- Don't show memblock directory if no reserve_mem succeeded
  and ARCH_KEEP_MEMBLOCK isn't defined (keeping current behavior).


 mm/memblock.c | 37 +++++++++++++++++++++++++++++++++++--
 1 file changed, 35 insertions(+), 2 deletions(-)

diff --git a/mm/memblock.c b/mm/memblock.c
index 2d2646f7a120..d816796ab919 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -17,6 +17,7 @@
 #include <linux/seq_file.h>
 #include <linux/memblock.h>
 #include <linux/mutex.h>
+#include <linux/string_helpers.h>
 
 #ifdef CONFIG_KEXEC_HANDOVER
 #include <linux/libfdt.h>
@@ -2711,7 +2712,8 @@ static int __init reserve_mem(char *p)
 }
 __setup("reserve_mem=", reserve_mem);
 
-#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
+#ifdef CONFIG_DEBUG_FS
+#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
 static const char * const flagname[] = {
 	[ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
 	[ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
@@ -2758,10 +2760,40 @@ static int memblock_debug_show(struct seq_file *m, void *private)
 }
 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
 
+#endif /* CONFIG_ARCH_KEEP_MEMBLOCK */
+
+static int memblock_reserve_mem_show(struct seq_file *m, void *private)
+{
+	struct reserve_mem_table *map;
+	char txtsz[16];
+
+	for (int i = 0; i < reserved_mem_count; i++) {
+		map = &reserved_mem_table[i];
+		if (!map->size)
+			continue;
+
+		memset(txtsz, 0, sizeof(txtsz));
+		string_get_size(map->size, 1, STRING_UNITS_2, txtsz, sizeof(txtsz));
+		seq_printf(m, "%s\t\t(%s)\n", map->name, txtsz);
+	}
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(memblock_reserve_mem);
+
 static int __init memblock_init_debugfs(void)
 {
-	struct dentry *root = debugfs_create_dir("memblock", NULL);
+	struct dentry *root;
 
+	if (!(IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK) || reserved_mem_count))
+		return 0;
+
+	root = debugfs_create_dir("memblock", NULL);
+
+	if (reserved_mem_count)
+		debugfs_create_file("reserve_mem_param", 0444, root, NULL,
+				    &memblock_reserve_mem_fops);
+#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
 	debugfs_create_file("memory", 0444, root,
 			    &memblock.memory, &memblock_debug_fops);
 	debugfs_create_file("reserved", 0444, root,
@@ -2771,6 +2803,7 @@ static int __init memblock_init_debugfs(void)
 			    &memblock_debug_fops);
 #endif
 
+#endif /* CONFIG_ARCH_KEEP_MEMBLOCK */
 	return 0;
 }
 __initcall(memblock_init_debugfs);
-- 
2.50.1

Re: [PATCH V2 1/2] mm/memblock: Print out errors on reserve_mem parser

[SeongJae Park]

On Wed,  4 Mar 2026 17:14:10 -0300 "Guilherme G. Piccoli" <gpiccoli@igalia.com> wrote:

> The parsing of kernel parameter "reserve_mem=" is subject to
> multiple failures, like duplicate naming, malformed expression
> or even lack of available memory. Right now, all of these fail
> silently. Let's add some messages so the kernel log can provide
> useful information in case of failures.

Makes sense to me.

> 
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Mike Rapoport <rppt@kernel.org>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Signed-off-by: Guilherme G. Piccoli <gpiccoli@igalia.com>

Reviewed-by: SeongJae Park <sj@kernel.org>

> ---
> 
> 
> V2: no changes.
> 
>  mm/memblock.c | 28 +++++++++++++++++++---------
>  1 file changed, 19 insertions(+), 9 deletions(-)
> 
> 
> diff --git a/mm/memblock.c b/mm/memblock.c
> index b3ddfdec7a80..2d2646f7a120 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -2642,23 +2642,25 @@ static int __init reserve_mem(char *p)
>  	int len;
>  
>  	if (!p)
> -		return -EINVAL;
> +		goto err_param;
>  
>  	/* Check if there's room for more reserved memory */
> -	if (reserved_mem_count >= RESERVE_MEM_MAX_ENTRIES)
> +	if (reserved_mem_count >= RESERVE_MEM_MAX_ENTRIES) {
> +		pr_err("reserve_mem: no more room for reserved memory\n");
>  		return -EBUSY;
> +	}
>  
>  	oldp = p;
>  	size = memparse(p, &p);
>  	if (!size || p == oldp)
> -		return -EINVAL;
> +		goto err_param;
>  
>  	if (*p != ':')
> -		return -EINVAL;
> +		goto err_param;
>  
>  	align = memparse(p+1, &p);
>  	if (*p != ':')
> -		return -EINVAL;
> +		goto err_param;
>  
>  	/*
>  	 * memblock_phys_alloc() doesn't like a zero size align,
> @@ -2672,7 +2674,7 @@ static int __init reserve_mem(char *p)
>  
>  	/* name needs to have length but not too big */
>  	if (!len || len >= RESERVE_MEM_NAME_SIZE)
> -		return -EINVAL;
> +		goto err_param;
>  
>  	/* Make sure that name has text */
>  	for (p = name; *p; p++) {
> @@ -2680,11 +2682,13 @@ static int __init reserve_mem(char *p)
>  			break;
>  	}
>  	if (!*p)
> -		return -EINVAL;
> +		goto err_param;
>  
>  	/* Make sure the name is not already used */
> -	if (reserve_mem_find_by_name(name, &start, &tmp))
> +	if (reserve_mem_find_by_name(name, &start, &tmp)) {
> +		pr_err("reserve_mem: name \"%s\" was already used\n", name);
>  		return -EBUSY;
> +	}
>  
>  	/* Pick previous allocations up from KHO if available */
>  	if (reserve_mem_kho_revive(name, size, align))
> @@ -2692,12 +2696,18 @@ static int __init reserve_mem(char *p)
>  
>  	/* TODO: Allocation must be outside of scratch region */
>  	start = memblock_phys_alloc(size, align);
> -	if (!start)
> +	if (!start) {
> +		pr_err("reserve_mem: memblock allocation failed\n");
>  		return -ENOMEM;
> +	}
>  
>  	reserved_mem_add(start, size, name);
>  
>  	return 1;
> +err_param:
> +	pr_err("reserve_mem: empty or malformed parameter\n");
> +	return -EINVAL;
> +

Nit.  Above blank line seems not needed.

>  }
>  __setup("reserve_mem=", reserve_mem);
>  
> -- 
> 2.50.1


Thanks,
SJ

Re: [PATCH V2 2/2] mm/memblock: Add reserve_mem debugfs info

[SeongJae Park]

On Wed,  4 Mar 2026 17:14:11 -0300 "Guilherme G. Piccoli" <gpiccoli@igalia.com> wrote:

> When using the "reserve_mem" parameter, users aim at having an
> area that (hopefully) persists across boots, so pstore infrastructure
> (like ramoops module) can make use of that to save oops/ftrace logs,
> for example.
> 
> There is no easy way to determine if this kernel parameter is properly
> set though; the kernel doesn't show information about this memory in
> memblock debugfs, neither in /proc/iomem nor dmesg. This is a relevant
> information for tools like kdumpst[0], to determine if it's reliable
> to use the reserved area as ramoops persistent storage; checking only
> /proc/cmdline is not sufficient as it doesn't tell if the reservation
> effectively succeeded or not.

Asking out of curiosity.  The previous patch has added the error log for
failure cases.  Could checking the kernel log to see if it was failed be an
option?

I think this debugfs approach is easier to check for the user space, though.
If that is the reason of this patch, adding the clarity would be nice for a
theoretical case that debugfs cannot be mounted.

> 
> Add here a new file under memblock debugfs showing properly set memory
> reservations, with name and size as passed to "reserve_mem". Notice that
> if no "reserve_mem=" is passed on command-line or if the reservation
> attempts fail, the file is not created.
> 
> [0] https://aur.archlinux.org/packages/kdumpst
> 
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Mike Rapoport <rppt@kernel.org>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Signed-off-by: Guilherme G. Piccoli <gpiccoli@igalia.com>
> ---
> 
> 
> Thanks a lot for the suggestions Mike! I'm not sure if you would
> prefer a Co-Developed-by or Suggested-by instead of CC, you helped
> a lot improving the code. Lemme know and either I can re-submit
> (with potential other changes) or even, you can change while merging.
> 
> V2: (all suggestions by Mike Rapoport)
> 
> - Commit message (showing use case);
> - Drop ifdef on include "string_helpers.h";
> - Don't show the address of reserve_mem, only name and size;
> - Fixed flag names inside ARCH_KEEP_MEMBLOCK ifdef;
> - Make use of its own show_attribute instead of refactoring
>   the memblock one;
> - Use sizeof() instead of magical numbers for the size;
> - Don't show memblock directory if no reserve_mem succeeded
>   and ARCH_KEEP_MEMBLOCK isn't defined (keeping current behavior).
> 
> 
>  mm/memblock.c | 37 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 35 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 2d2646f7a120..d816796ab919 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -17,6 +17,7 @@
>  #include <linux/seq_file.h>
>  #include <linux/memblock.h>
>  #include <linux/mutex.h>
> +#include <linux/string_helpers.h>
>  
>  #ifdef CONFIG_KEXEC_HANDOVER
>  #include <linux/libfdt.h>
> @@ -2711,7 +2712,8 @@ static int __init reserve_mem(char *p)
>  }
>  __setup("reserve_mem=", reserve_mem);
>  
> -#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
> +#ifdef CONFIG_DEBUG_FS
> +#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
>  static const char * const flagname[] = {
>  	[ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
>  	[ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
> @@ -2758,10 +2760,40 @@ static int memblock_debug_show(struct seq_file *m, void *private)
>  }
>  DEFINE_SHOW_ATTRIBUTE(memblock_debug);
>  
> +#endif /* CONFIG_ARCH_KEEP_MEMBLOCK */
> +
> +static int memblock_reserve_mem_show(struct seq_file *m, void *private)
> +{
> +	struct reserve_mem_table *map;
> +	char txtsz[16];
> +
> +	for (int i = 0; i < reserved_mem_count; i++) {
> +		map = &reserved_mem_table[i];
> +		if (!map->size)
> +			continue;
> +
> +		memset(txtsz, 0, sizeof(txtsz));
> +		string_get_size(map->size, 1, STRING_UNITS_2, txtsz, sizeof(txtsz));
> +		seq_printf(m, "%s\t\t(%s)\n", map->name, txtsz);
> +	}
> +
> +	return 0;
> +}
> +DEFINE_SHOW_ATTRIBUTE(memblock_reserve_mem);
> +
>  static int __init memblock_init_debugfs(void)
>  {
> -	struct dentry *root = debugfs_create_dir("memblock", NULL);
> +	struct dentry *root;
>  
> +	if (!(IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK) || reserved_mem_count))
> +		return 0;

One trivial comment.  I'd slightly prefer having one less parentheses level as
a tradeoff of having one more exclamation mark, e.g.,

    if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK) && !reserved_mem_count)

> +
> +	root = debugfs_create_dir("memblock", NULL);
> +
> +	if (reserved_mem_count)
> +		debugfs_create_file("reserve_mem_param", 0444, root, NULL,
> +				    &memblock_reserve_mem_fops);
> +#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
>  	debugfs_create_file("memory", 0444, root,
>  			    &memblock.memory, &memblock_debug_fops);
>  	debugfs_create_file("reserved", 0444, root,
> @@ -2771,6 +2803,7 @@ static int __init memblock_init_debugfs(void)
>  			    &memblock_debug_fops);
>  #endif
>  
> +#endif /* CONFIG_ARCH_KEEP_MEMBLOCK */

So, we get two level of nested ifdef...  I'm wondering if returning earlier
when CONFIG_ARCH_KEEP_MEMBLOCK is undefined is easier to read.  E.g.,

	#ifndef CONFIG_ARCH_KEEP_MEMBLOCK
		return 0;
	#endif

		debugfs_create_file("memory", 0444, root,
	[...]

>  	return 0;

Very trivial comment.  Why don't you keep the original blank line above the
return statemtnt?

>  }
>  __initcall(memblock_init_debugfs);
> -- 
> 2.50.1


Thanks,
SJ