Skip to contentSkip to conversation

Documentation/ABI: Document the non-ABI status of Kconfig and symbols

PATCH In review Josh Triplett 4 participants 6 messages Oct 24, 2013 – Nov 10, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Documentation/ABI: Document the non-ABI status of Kconfig and symbols

Discussion at Kernel Summit made it clear that the presence or absence of specific Kconfig symbols are not considered ABI, and that no userspace (or bootloader, etc) should rely on them.

In addition, kernel-internal symbols are well established as non-ABI, per Documentation/stable_api_nonsense.txt.

Document both of these in Documentation/ABI/README, in a new section for notable bits of non-ABI.

---
 Documentation/ABI/README | 13 +++++++++++++
 1 file changed, 13 insertions(+)
Documentation/ABI/README +13 -0
index 1006982..1fafc4b 100644
@@ -72,3 +72,16 @@ kernel tree without going through the obsolete state first.
 
 It's up to the developer to place their interfaces in the category they
 wish for it to start out in.
+
+
+Notable bits of non-ABI, which should not under any circumstances be considered
+stable:
+
+- Kconfig.  Userspace should not rely on the presence or absence of any
+  particular Kconfig symbol, in /proc/config.gz, in the copy of .config
+  commonly installed to /boot, or in any invocation of the kernel build
+  process.
+
+- Kernel-internal symbols.  Do not rely on the presence, absence, location, or
+  type of any kernel symbol, either in System.map files or the kernel binary
+  itself.  See Documentation/stable_api_nonsense.txt.
Signature · 1 line
1.8.4.rc3
Acked-by on v1 1/1
#1 lore raw
Quoted 11 lines · Josh Triplett

Discussion at Kernel Summit made it clear that the presence or absence of specific Kconfig symbols are not considered ABI, and that no userspace (or bootloader, etc) should rely on them.

In addition, kernel-internal symbols are well established as non-ABI, per Documentation/stable_api_nonsense.txt.

Document both of these in Documentation/ABI/README, in a new section for notable bits of non-ABI.

Good initiative.

#2 lore raw
In reply to Josh Triplett
Quoted 26 lines · Josh Triplett

Discussion at Kernel Summit made it clear that the presence or absence of specific Kconfig symbols are not considered ABI, and that no userspace (or bootloader, etc) should rely on them.

In addition, kernel-internal symbols are well established as non-ABI, per Documentation/stable_api_nonsense.txt.

Document both of these in Documentation/ABI/README, in a new section for notable bits of non-ABI.

---
 Documentation/ABI/README | 13 +++++++++++++
 1 file changed, 13 insertions(+)
Documentation/ABI/README +3 -0
index 1006982..1fafc4b 100644
@@ -72,3 +72,16 @@ kernel tree without going through the obsolete state first.
 
 It's up to the developer to place their interfaces in the category they
 wish for it to start out in.
+
+
+Notable bits of non-ABI, which should not under any circumstances be considered

Documentation/ABI/README kernel tree without going through the obsolete state first. · line 87

Quoted 10 lines
Documentation/ABI/README +10 -0
+stable:
+
+- Kconfig.  Userspace should not rely on the presence or absence of any
+  particular Kconfig symbol, in /proc/config.gz, in the copy of .config
+  commonly installed to /boot, or in any invocation of the kernel build
+  process.
+
+- Kernel-internal symbols.  Do not rely on the presence, absence, location, or
+  type of any kernel symbol, either in System.map files or the kernel binary
+  itself.  See Documentation/stable_api_nonsense.txt.

And what about the kernel make "interface", is it considered also as non-ABI?

E.g.
Before ffee0de (x86: Default to ARCH=x86 to avoid overriding CONFIG_64BIT) "make defconfig ARCH=x86" produced a i386 defconfig. Now it produces a x86_64 defconfig.
I'm sure some build scripted failed badly.

Signature · 2 lines
Thanks,
//richard
#3 lore raw
Quoted 40 lines · Richard Weinberger
Quoted 26 lines · Josh Triplett

Discussion at Kernel Summit made it clear that the presence or absence of specific Kconfig symbols are not considered ABI, and that no userspace (or bootloader, etc) should rely on them.

In addition, kernel-internal symbols are well established as non-ABI, per Documentation/stable_api_nonsense.txt.

Document both of these in Documentation/ABI/README, in a new section for notable bits of non-ABI.

---
 Documentation/ABI/README | 13 +++++++++++++
 1 file changed, 13 insertions(+)
Documentation/ABI/README +3 -0
index 1006982..1fafc4b 100644
@@ -72,3 +72,16 @@ kernel tree without going through the obsolete state first.
 
 It's up to the developer to place their interfaces in the category they
 wish for it to start out in.
+
+
+Notable bits of non-ABI, which should not under any circumstances be considered

Documentation/ABI/README kernel tree without going through the obsolete state first. · line 87

Quoted 10 lines
Documentation/ABI/README +10 -0
+stable:
+
+- Kconfig.  Userspace should not rely on the presence or absence of any
+  particular Kconfig symbol, in /proc/config.gz, in the copy of .config
+  commonly installed to /boot, or in any invocation of the kernel build
+  process.
+
+- Kernel-internal symbols.  Do not rely on the presence, absence, location, or
+  type of any kernel symbol, either in System.map files or the kernel binary
+  itself.  See Documentation/stable_api_nonsense.txt.

And what about the kernel make "interface", is it considered also as non-ABI?

Quoted 5 lines

E.g.
Before ffee0de (x86: Default to ARCH=x86 to avoid overriding CONFIG_64BIT) "make defconfig ARCH=x86" produced a i386 defconfig. Now it produces a x86_64 defconfig.
I'm sure some build scripted failed badly.

I seriously considered including that, but I could imagine that some parts of that interface might be considered more stable than others (the names of targets, for instance), and I wanted to stick to completely uncontroversial items. If there's consensus that part or all of the invocation of "make" to build the kernel is non-ABI, we can easily enough add that to the non-ABI list, but let's start with Kconfig and symbols.

- Josh Triplett

#4 lore raw
Quoted 8 lines · Josh Triplett
Quoted 1 line · Richard Weinberger

On Thu, Oct 24, 2013 at 10:41 AM, Josh Triplett

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line

Discussion at Kernel Summit made it clear that the presence or

absence

Quoted 3 lines
Quoted 3 lines

of specific Kconfig symbols are not considered ABI, and that no userspace (or bootloader, etc) should rely on them.

Quoted 12 lines
Quoted 1 line
Quoted 1 line

In addition, kernel-internal symbols are well established as

non-ABI,

Quoted 3 lines
Quoted 3 lines

per Documentation/stable_api_nonsense.txt.

Document both of these in Documentation/ABI/README, in a new

section for

Quoted 6 lines
Quoted 6 lines

notable bits of non-ABI.

---
 Documentation/ABI/README | 13 +++++++++++++
 1 file changed, 13 insertions(+)

Should I grab this, or is it already going in through somebody else's tree?

Rob

#5 lore raw
In reply to Richard Weinberger
Quoted 10 lines · Richard Weinberger

And what about the kernel make "interface", is it considered also as non-ABI?

E.g.
Before ffee0de (x86: Default to ARCH=x86 to avoid overriding CONFIG_64BIT) "make defconfig ARCH=x86" produced a i386 defconfig. Now it produces a x86_64 defconfig.
I'm sure some build scripted failed badly.

Definitely not ABI, although obviously not a completely "break blindly" either.

-hpa