Skip to contentSkip to conversation

mm: use zonelist_zone() to get zone

PATCH NAK Wei Yang 4 participants 13 messages Jul 2, 2024 – Jul 7, 2024

Subsystems (5)
lore
Share & tools
lore

Wei Yang

author
#0 lore raw

PATCH 1/3 mm: use zonelist_zone() to get zone

Instead of accessing zoneref->zone directly, use zonelist_zone() like other places for consistency.

No functional change.

CC Mike Rapoport (IBM) <[email protected]>
---
 include/linux/mmzone.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/mmzone.h +1 -1
index cb7f265c2b96..a34a74f5b113 100644
@@ -1690,7 +1690,7 @@ static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
 			zone = zonelist_zone(z))
 
 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
-	for (zone = z->zone;	\
+	for (zone = zonelist_zone(z);	\
 		zone;							\
 		z = next_zones_zonelist(++z, highidx, nodemask),	\
 			zone = zonelist_zone(z))
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 2/3 modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
scripts/mod/modpost.c +10 -0
index f48d72d22dc2..81134403d4d7 100644
@@ -22,6 +22,7 @@
 #include <errno.h>
 #include "modpost.h"
 #include "../../include/linux/license.h"
+#include "../../include/generated/autoconf.h"
 
 static bool module_enabled;
 /* Are we using CONFIG_MODVERSIONS? */
@@ -775,9 +776,14 @@ static void check_section(const char *modname, struct elf_info *elf,
 
 
 
+#if defined(CONFIG_MEMORY_HOTPLUG)
+#define ALL_INIT_DATA_SECTIONS \
+	".init.setup", ".init.rodata", ".init.data"
+#else
 #define ALL_INIT_DATA_SECTIONS \
 	".init.setup", ".init.rodata", ".meminit.rodata", \
 	".init.data", ".meminit.data"
+#endif
 
 #define ALL_PCI_INIT_SECTIONS	\
 	".pci_fixup_early", ".pci_fixup_header", ".pci_fixup_final", \
@@ -786,7 +792,11 @@ static void check_section(const char *modname, struct elf_info *elf,
 
 #define ALL_XXXINIT_SECTIONS ".meminit.*"
 
+#if defined(CONFIG_MEMORY_HOTPLUG)
+#define ALL_INIT_SECTIONS INIT_SECTIONS
+#else
 #define ALL_INIT_SECTIONS INIT_SECTIONS, ALL_XXXINIT_SECTIONS
+#endif
 #define ALL_EXIT_SECTIONS ".exit.*"
 
 #define DATA_SECTIONS ".data", ".data.rel"
Signature · 1 line
2.34.1
maintainer
#2 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 3 lines · On Tue, 2 Jul 2024 23:40:07 +0000 Wei Yang <[email protected]> wrote:

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

Please describe how this changes modpost behaviour.

Something like: "we're currently not checking for references into meminit and meminitdata when CONFIG_HOTPLUG=y, which may cause us to fail to notice incorrect references.". But I don't think that's correct. So what *is* wrong with the current code?

#3 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 7 lines · Andrew Morton
Quoted 3 lines · On Tue, 2 Jul 2024 23:40:07 +0000 Wei Yang <[email protected]> wrote:

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

Quoted 7 lines

Please describe how this changes modpost behaviour.

Something like: "we're currently not checking for references into meminit and meminitdata when CONFIG_HOTPLUG=y, which may cause us to fail to notice incorrect references.". But I don't think that's correct. So what *is* wrong with the current code?

Sigh.

If you do not understand, you should not apply it.

I am surprised that there exists a person who
attempted to apply this.

Signature · 2 lines
Best Regards
Masahiro Yamada
#4 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

In reply to Wei Yang
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

Signature · 2 lines
Best Regards
Masahiro Yamada
author
#5 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 32 lines · Masahiro Yamada
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100
Quoted 8 lines

Use separate sections for __dev/__cpu/__mem code/data

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

You mean something like this?

include/linux/init.h +2 -0
index 58cef4c2e59a..388f0a4c34e9 100644
@@ -85,10 +85,12 @@
 #define __exit          __section(".exit.text") __exitused __cold notrace
 
 /* Used for MEMORY_HOTPLUG */
+#ifndef CONFIG_MEMORY_HOTPLUG
 #define __meminit        __section(".meminit.text") __cold notrace \
 						  __latent_entropy
 #define __meminitdata    __section(".meminit.data")
 #define __meminitconst   __section(".meminit.rodata")
+#endif
 
 /* For assembly routines */
 #define __HEAD		.section	".head.text","ax"
Quoted 3 lines

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

I guess you mean this is not a good practice.

Then I am confused how we do the mismatch check unconditionally?

After commit

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

Sections .meminit.* will be put into INIT_SECTION conditionally, but we always do the mismatch check unconditionally. It will report mismatch when .meminit.* is not in INIT_SECTION. It looks not correct to me.

Maybe I am not fully understand your message. Would you mind explaining more on what is the correct way to do?

Signature · 2 lines
Wei Yang
Help you, Help me
author
#6 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

In reply to Masahiro Yamada
Quoted 37 lines · Masahiro Yamada
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

Quoted 6 lines

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

Masahiro

If you would give me some suggestions, I'd appreciate it a lot.

The original thing I want to do is to put function __free_pages_core() in __meminit section, since this function is only used by __init functions and in memory_hotplug.c. This means it is save to release it if CONFIG_MEMORY_HOTPLUG is set.

Then I add __meminit to function __free_pages_core() and face the warning from modpost.

WARNING: modpost: vmlinux: section mismatch in reference: generic_online_page+0xa (section: .text) -> __free_pages_core (section: .meminit.text)

A .text function calls init code is not proper. Then I add __meminit to generic_online_page too. Then I face this warning from modpost.

WARNING: modpost: vmlinux: generic_online_page: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.

Function generic_online_page() is exported and be used in some modules. And is not allowed to use init tag.

When looking into the code, this warning is printed by this code:

   #define ALL_INIT_SECTIONS INIT_SECTIONS, ALL_XXXINIT_SECTIONS

if (match(secname, PATTERNS(ALL_INIT_SECTIONS)))
warn("%s: %s: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.\n",
mod->name, name);

If my understanding is correct, the code means we can't use a function in init section, since the code will be released after bootup.

But for this case, when CONFIG_MEMORY_HOTPLUG is set, the section .meminit.* is not put into the real init sections. So the functions are not released and could be used by modules. This behavior is introduced in commit eb8f689046b8(the one you mentioned above).

So the check here is not proper to me. We should exclude .meminit.* from ALL_INIT_SECTIONS.

Looking forward your suggestion and a proper way to handle this.

Signature · 2 lines
Wei Yang
Help you, Help me
author
#7 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 57 lines · Wei Yang
Quoted 37 lines · Masahiro Yamada
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

Quoted 6 lines

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

Masahiro

If you would give me some suggestions, I'd appreciate it a lot.

The original thing I want to do is to put function __free_pages_core() in __meminit section, since this function is only used by __init functions and in memory_hotplug.c. This means it is save to release it if CONFIG_MEMORY_HOTPLUG is set.

Then I add __meminit to function __free_pages_core() and face the warning from modpost.

Quoted 7 lines

WARNING: modpost: vmlinux: section mismatch in reference: generic_online_page+0xa (section: .text) -> __free_pages_core (section: .meminit.text)

A .text function calls init code is not proper. Then I add __meminit to generic_online_page too. Then I face this warning from modpost.

WARNING: modpost: vmlinux: generic_online_page: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.

I guess I found the correct way.

Add __ref to generic_online_page to not issue a warning.

Quoted 27 lines

Function generic_online_page() is exported and be used in some modules. And is not allowed to use init tag.

When looking into the code, this warning is printed by this code:

  #define ALL_INIT_SECTIONS INIT_SECTIONS, ALL_XXXINIT_SECTIONS

if (match(secname, PATTERNS(ALL_INIT_SECTIONS)))
warn("%s: %s: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.\n",
mod->name, name);

If my understanding is correct, the code means we can't use a function in init section, since the code will be released after bootup.

But for this case, when CONFIG_MEMORY_HOTPLUG is set, the section .meminit.* is not put into the real init sections. So the functions are not released and could be used by modules. This behavior is introduced in commit eb8f689046b8(the one you mentioned above).

So the check here is not proper to me. We should exclude .meminit.* from ALL_INIT_SECTIONS.

Looking forward your suggestion and a proper way to handle this.

Signature · 2 lines
Wei Yang
Help you, Help me
Signature · 2 lines
Wei Yang
Help you, Help me
#8 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 64 lines · Wei Yang
Quoted 62 lines · Wei Yang
Quoted 37 lines · Masahiro Yamada
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

Quoted 6 lines

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

Masahiro

If you would give me some suggestions, I'd appreciate it a lot.

The original thing I want to do is to put function __free_pages_core() in __meminit section, since this function is only used by __init functions and in memory_hotplug.c. This means it is save to release it if CONFIG_MEMORY_HOTPLUG is set.

Then I add __meminit to function __free_pages_core() and face the warning from modpost.

WARNING: modpost: vmlinux: section mismatch in reference: generic_online_page+0xa (section: .text) -> __free_pages_core (section: .meminit.text)

A .text function calls init code is not proper. Then I add __meminit to generic_online_page too. Then I face this warning from modpost.

Quoted 6 lines
Quoted 2 lines

WARNING: modpost: vmlinux: generic_online_page: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.

I guess I found the correct way.

Add __ref to generic_online_page to not issue a warning.

Yes, __ref is used to bypass the section mismatch check.

Some functions in mm/memory_hotplug.c are annotated as __ref to reference __meminit functions.

Adding __ref is the easy solution.

Having said that, I started to think eb8f689046b857874e964463619f09df06d59fad was the wrong decision. I will revert it.

Signature · 2 lines
Best Regards
Masahiro Yamada
author
#9 lore raw

Re: [PATCH 2/3] modpost: .meminit.* is not in init section when CONFIG_MEMORY_HOTPLUG set

Quoted 68 lines · Masahiro Yamada
Quoted 67 lines · Wei Yang
Quoted 64 lines · Wei Yang
Quoted 37 lines · Masahiro Yamada
Quoted 10 lines · Wei Yang

.meminit.* is not put into init section when CONFIG_MEMORY_HOTPLUG is set, since we define MEM_KEEP()/MEM_DISCARD() according to CONFIG_MEMORY_HOTPLUG.

CC Mike Rapoport (IBM) <[email protected]>
---
 scripts/mod/modpost.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

NACK.

The section mismatch is performed _unconditionally_.

In the old days, we did this depending on relevant CONFIG options. It was more than 15 years ago that we stopped doing that.

See this:

commit eb8f689046b857874e964463619f09df06d59fad

Author: Sam Ravnborg <[email protected]>
Date:   Sun Jan 20 20:07:28 2008 +0100

Use separate sections for __dev/__cpu/__mem code/data

Quoted 6 lines

So, if you wanted to check this only when CONFIG_MEMORY_HOTPLUG=n, you would need to add #ifdef CONFIG_MEMORY_HOTPLUG to include/linux/init.h

That is what we did in the Linux 2.6.* era, which had much worse section mismatch coverage.

Masahiro

If you would give me some suggestions, I'd appreciate it a lot.

The original thing I want to do is to put function __free_pages_core() in __meminit section, since this function is only used by __init functions and in memory_hotplug.c. This means it is save to release it if CONFIG_MEMORY_HOTPLUG is set.

Then I add __meminit to function __free_pages_core() and face the warning from modpost.

WARNING: modpost: vmlinux: section mismatch in reference: generic_online_page+0xa (section: .text) -> __free_pages_core (section: .meminit.text)

A .text function calls init code is not proper. Then I add __meminit to generic_online_page too. Then I face this warning from modpost.

WARNING: modpost: vmlinux: generic_online_page: EXPORT_SYMBOL used for init symbol. Remove __init or EXPORT_SYMBOL.

Quoted 8 lines
Quoted 3 lines

I guess I found the correct way.

Add __ref to generic_online_page to not issue a warning.

Yes, __ref is used to bypass the section mismatch check.

I am think whether __ref is providing a gate to escape the check of modpost?

Quoted 11 lines

Some functions in mm/memory_hotplug.c are annotated as __ref to reference __meminit functions.

Adding __ref is the easy solution.

Having said that, I started to think eb8f689046b857874e964463619f09df06d59fad was the wrong decision. I will revert it.

Oh, I finally understand it... didn't think that was a wrong decision :-(

Quoted 13 lines
Signature · 3 lines
-- 
Best Regards
Masahiro Yamada
Signature · 2 lines
Wei Yang
Help you, Help me
author
#10 lore raw

PATCH 3/3 mm/page_alloc: put __free_pages_core() in __meminit section

In reply to Wei Yang

Function __free_pages_core() is only used in bootmem init and hot-add memory init path. Let's put it in __meminit section.

CC Mike Rapoport (IBM) <[email protected]>
---
 mm/page_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index 480e4416131f..c46aedfc9a12 100644
@@ -1219,7 +1219,7 @@ static void __free_pages_ok(struct page *page, unsigned int order,
 	__count_vm_events(PGFREE, 1 << order);
 }
 
-void __free_pages_core(struct page *page, unsigned int order,
+void __meminit __free_pages_core(struct page *page, unsigned int order,
 		enum meminit_context context)
 {
 	unsigned int nr_pages = 1 << order;
Signature · 1 line
2.34.1
#11 lore raw
In reply to Wei Yang
Quoted 16 lines · Wei Yang

Instead of accessing zoneref->zone directly, use zonelist_zone() like other places for consistency.

No functional change.

CC Mike Rapoport (IBM) <[email protected]>
---
  include/linux/mmzone.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/mmzone.h +0 -0
index cb7f265c2b96..a34a74f5b113 100644

include/linux/mmzone.h static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, · line 1696

Quoted 8 lines
include/linux/mmzone.h +1 -1
@@ -1690,7 +1690,7 @@ static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
  			zone = zonelist_zone(z))
 
  #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
-	for (zone = z->zone;	\
+	for (zone = zonelist_zone(z);	\
  		zone;							\
  		z = next_zones_zonelist(++z, highidx, nodemask),	\
  			zone = zonelist_zone(z))

Should we do the same in movable_only_nodes as well to be consistent in that file?

Signature · 3 lines
Cheers,

David / dhildenb
author
#12 lore raw
Quoted 12 lines · David Hildenbrand
Quoted 11 lines · Wei Yang

Instead of accessing zoneref->zone directly, use zonelist_zone() like other places for consistency.

No functional change.

CC Mike Rapoport (IBM) <[email protected]>
---
  include/linux/mmzone.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 16 lines

include/linux/mmzone.h static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, · line 1695

Quoted 12 lines
include/linux/mmzone.h +1 -1
index cb7f265c2b96..a34a74f5b113 100644
@@ -1690,7 +1690,7 @@ static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
  			zone = zonelist_zone(z))
  #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
-	for (zone = z->zone;	\
+	for (zone = zonelist_zone(z);	\
  		zone;							\
  		z = next_zones_zonelist(++z, highidx, nodemask),	\
  			zone = zonelist_zone(z))

Should we do the same in movable_only_nodes as well to be consistent in that file?

Agree, thanks

Quoted 4 lines
Signature · 3 lines
Cheers,

David / dhildenb
Signature · 2 lines
Wei Yang
Help you, Help me