Skip to contentSkip to conversation

vmcoreinfo: export SECTION_HAS_MEM_MAP

PATCH Applied David Hildenbrand (Arm) 6 participants 11 messages Sep 28 – Sep 29

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vmcoreinfo: export SECTION_HAS_MEM_MAP

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +1 -0
index 5a417f8a922a..7833a36064a8 100644
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

---

base-commit: 57e4ac91fc62d75d84b5a03827a19ceb4094ecd9

change-id: 20260928-section_has_mem_map-6205dc90a09f

--

Cheers,

David

#1 lore raw
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

It would be good to add one section in Documentation/admin-guide/kdump/vmcoreinfo.rst, e.g

SECTION_HAS_MEM_MAP
-------------------

Indicates a memory section has struct page array (mem_map). User-space tools should read this value but not hardcode it since the bit position could change between kernel versions.

The code change looks good to me.

Quoted 13 lines

---

base-commit: 57e4ac91fc62d75d84b5a03827a19ceb4094ecd9

change-id: 20260928-section_has_mem_map-6205dc90a09f

--

Cheers,

David

#2 lore raw
Quoted 37 lines · Baoquan He
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

It would be good to add one section in Documentation/admin-guide/kdump/vmcoreinfo.rst, e.g

Quoted 6 lines

SECTION_HAS_MEM_MAP
-------------------

Indicates a memory section has struct page array (mem_map). User-space tools should read this value but not hardcode it since the bit position could change between kernel versions.

I'm not particularly happy about having core-kernel defines documented in kdump/vmcoreinfo.rst, though?

If it would be something vmcoreinfo special, sure. But not for basic defines.

"See the kernel source code" is really what anybody using this should be doing?

Signature · 3 lines
Cheers,

David
Acked-by
#3 lore raw
Quoted 45 lines · David Hildenbrand (Arm)
Quoted 37 lines · Baoquan He
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

It would be good to add one section in Documentation/admin-guide/kdump/vmcoreinfo.rst, e.g

Quoted 6 lines

SECTION_HAS_MEM_MAP
-------------------

Indicates a memory section has struct page array (mem_map). User-space tools should read this value but not hardcode it since the bit position could change between kernel versions.

Quoted 6 lines

I'm not particularly happy about having core-kernel defines documented in kdump/vmcoreinfo.rst, though?

If it would be something vmcoreinfo special, sure. But not for basic defines.

"See the kernel source code" is really what anybody using this should be doing?

I can't remember who suggested adding this document. The benefit is users can read document to understand how it works, no need to look at the code. I don't have strong preference on this.

So I am fine with both. Or we can add it if anyone really complains.

For the patch itself,

Thanks
Baoquan

#4 lore raw
In reply to David Hildenbrand (Arm)
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

Current crash code looks below:

#define SECTION_MARKED_PRESENT          (1UL<<0)
#define SECTION_HAS_MEM_MAP             (1UL<<1)
#define SECTION_IS_ONLINE               (1UL<<2)
#define SECTION_IS_EARLY                (1UL<<3)
#define SECTION_TAINT_ZONE_DEVICE       (1UL<<4)
#define SECTION_MAP_LAST_BIT            (1UL<<5)
#define SECTION_MAP_MASK                (~(SECTION_MAP_LAST_BIT-1))

And it does not match the kernel for below chunk with ifdef #ifdef CONFIG_ZONE_DEVICE
SECTION_TAINT_ZONE_DEVICE_BIT,
#endif
#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
SECTION_IS_VMEMMAP_PREINIT_BIT,
#endif

The last bit could be wrong in crash. So all numbers should be exported in vmcoreinfo.

Thoughts?

ThanksDave

#5 lore raw
Quoted 52 lines · Dave Young
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

Current crash code looks below:

#define SECTION_MARKED_PRESENT          (1UL<<0)
#define SECTION_HAS_MEM_MAP             (1UL<<1)
#define SECTION_IS_ONLINE               (1UL<<2)
#define SECTION_IS_EARLY                (1UL<<3)
#define SECTION_TAINT_ZONE_DEVICE       (1UL<<4)
#define SECTION_MAP_LAST_BIT            (1UL<<5)
#define SECTION_MAP_MASK                (~(SECTION_MAP_LAST_BIT-1))

And it does not match the kernel for below chunk with ifdef #ifdef CONFIG_ZONE_DEVICE
SECTION_TAINT_ZONE_DEVICE_BIT,
#endif
#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
SECTION_IS_VMEMMAP_PREINIT_BIT,
#endif

Quoted 3 lines

The last bit could be wrong in crash. So all numbers should be exported in vmcoreinfo.

Thoughts?

You know the crash code best, so likely ... it's time one of the crash people takes over this patch?

Signature · 3 lines
Cheers,

David
#6 lore raw
Quoted 53 lines · David Hildenbrand (Arm)
Quoted 52 lines · Dave Young
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);

Current crash code looks below:

#define SECTION_MARKED_PRESENT          (1UL<<0)
#define SECTION_HAS_MEM_MAP             (1UL<<1)
#define SECTION_IS_ONLINE               (1UL<<2)
#define SECTION_IS_EARLY                (1UL<<3)
#define SECTION_TAINT_ZONE_DEVICE       (1UL<<4)
#define SECTION_MAP_LAST_BIT            (1UL<<5)
#define SECTION_MAP_MASK                (~(SECTION_MAP_LAST_BIT-1))

And it does not match the kernel for below chunk with ifdef #ifdef CONFIG_ZONE_DEVICE
SECTION_TAINT_ZONE_DEVICE_BIT,
#endif
#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
SECTION_IS_VMEMMAP_PREINIT_BIT,
#endif

Quoted 6 lines
Quoted 3 lines

The last bit could be wrong in crash. So all numbers should be exported in vmcoreinfo.

Thoughts?

You know the crash code best, so likely ... it's time one of the crash people takes over this patch?

Rethinking about it I'm hesitating to add many more to vmcoreinfo although it is reasonable, about the #ifdef things, there are other places other than mm, crash can not depend on the /proc/config.gz which is not reliable in the old kernel memory, maybe it will be helpful to export another elf note for old kernel's kconfig. So I'd like to support to use your patch for the time being as the first bit is always needed and leave the remaining concerns a follow up item.

Thanks
Dave

#7 lore raw
Quoted 54 lines · Dave Young
Quoted 53 lines · David Hildenbrand (Arm)
Quoted 52 lines · Dave Young
Quoted 25 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
  kernel/vmcore_info.c | 1 +
  1 file changed, 1 insertion(+)
kernel/vmcore_info.c +0 -0
index 5a417f8a922a..7833a36064a8 100644

kernel/vmcore_info.c static int __init crash_save_vmcoreinfo_init(void) · line 188

Quoted 7 lines
kernel/vmcore_info.c +1 -0
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
  	VMCOREINFO_STRUCT_SIZE(mem_section);
  	VMCOREINFO_OFFSET(mem_section, section_mem_map);
  	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
  	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
  #endif
  	VMCOREINFO_STRUCT_SIZE(page);

Current crash code looks below:

#define SECTION_MARKED_PRESENT          (1UL<<0)
#define SECTION_HAS_MEM_MAP             (1UL<<1)
#define SECTION_IS_ONLINE               (1UL<<2)
#define SECTION_IS_EARLY                (1UL<<3)
#define SECTION_TAINT_ZONE_DEVICE       (1UL<<4)
#define SECTION_MAP_LAST_BIT            (1UL<<5)
#define SECTION_MAP_MASK                (~(SECTION_MAP_LAST_BIT-1))

And it does not match the kernel for below chunk with ifdef #ifdef CONFIG_ZONE_DEVICE
SECTION_TAINT_ZONE_DEVICE_BIT,
#endif
#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
SECTION_IS_VMEMMAP_PREINIT_BIT,
#endif

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

The last bit could be wrong in crash. So all numbers should be exported in vmcoreinfo.

Thoughts?

You know the crash code best, so likely ... it's time one of the crash people takes over this patch?

Rethinking about it I'm hesitating to add many more to vmcoreinfo although it is reasonable, about the #ifdef things, there are other places other than mm, crash can not depend on the /proc/config.gz which is not reliable in the old kernel memory, maybe it will be helpful to export another elf note for old kernel's kconfig. So I'd like to support to use your patch for the time being as the first bit is always needed and leave the remaining concerns a follow up item.

I agree, from makedumpfile's view point.

(to be exact, makedumpfile needs this but maybe crash can read enum values from vmlinux.)

Thanks,
Kazu

Quoted 3 lines

Thanks
Dave

#8 lore raw
Quoted 5 lines · HAGIO KAZUHITO(萩尾 一仁)
Quoted 2 lines · Dave Young
Quoted 1 line · David Hildenbrand (Arm)
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

You know the crash code best, so likely ... it's time one of the crash people takes over this patch?

Rethinking about it I'm hesitating to add many more to vmcoreinfo although it is reasonable, about the #ifdef things, there are other places other than mm, crash can not depend on the /proc/config.gz which is not reliable in the old kernel memory, maybe it will be helpful to export another elf note for old kernel's kconfig. So I'd like to support to use your patch for the time being as the first bit is always needed and leave the remaining concerns a follow up item.

I agree, from makedumpfile's view point.

(to be exact, makedumpfile needs this but maybe crash can read enum values from vmlinux.)

That's a good point, maybe that indeed works for crash.

Signature · 3 lines
Cheers,

David
Reviewed-by on v1 1/1
#9 lore raw
In reply to David Hildenbrand (Arm)
Quoted 6 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

I guess that works

Quoted 32 lines

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1]
https://lore.kernel.org/r/[email protected]
---
kernel/vmcore_info.c | 1 +

1 file changed, 1 insertion(+)
kernel/vmcore_info.c +1 -1
index 5a417f8a922a..7833a36064a8 100644
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);
 
---

base-commit: 57e4ac91fc62d75d84b5a03827a19ceb4094ecd9

Quoted 10 lines

change-id: 20260928-section_has_mem_map-6205dc90a09f

--

Cheers,

David

--- Thanks!
"I'm not a very positive person" - Linus torvalds

Reviewed-by on v1 1/1
#10 lore raw
In reply to David Hildenbrand (Arm)
Quoted 8 lines · David Hildenbrand (Arm)

The crash tool currently hardcodes SECTION_HAS_MEM_MAP, and makedumpfile needs similar information (although still relying on SECTION_MARKED_PRESENT, it should switch to SECTION_HAS_MEM_MAP).

Let's just export the value instead, so tools that work on vmcoreinfo will not have to guess.

Quoted 37 lines

---
Result of the discussion in reply to "[PATCH v2 00/13] mm/sparse: remove SECTION_MARKED_PRESENT and further cleanups" [1]

This patch can go in independently. It would be preferable if both go into the same kernel release ;)

[1] https://lore.kernel.org/r/[email protected]

---
 kernel/vmcore_info.c | 1 +
 1 file changed, 1 insertion(+)
kernel/vmcore_info.c +1 -1
index 5a417f8a922a..7833a36064a8 100644
@@ -182,6 +182,7 @@ static int __init crash_save_vmcoreinfo_init(void)
 	VMCOREINFO_STRUCT_SIZE(mem_section);
 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
+	VMCOREINFO_NUMBER(SECTION_HAS_MEM_MAP);
 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
 #endif
 	VMCOREINFO_STRUCT_SIZE(page);
 
---

base-commit: 57e4ac91fc62d75d84b5a03827a19ceb4094ecd9

change-id: 20260928-section_has_mem_map-6205dc90a09f

--

Cheers,

David