Skip to contentSkip to conversation

kbuild: give SUBLEVEL more room in KERNEL_VERSION

PATCH Superseded Sasha Levin 4 participants 11 messages Jan 18, 2021 Kbuild

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kbuild: give SUBLEVEL more room in KERNEL_VERSION

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +2 -2
index 9e73f82e0d863..dc2bad7a440d8 100644
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'
 endef
 
 $(version_h): FORCE
Signature · 1 line
2.27.0
#1 lore raw
Quoted 27 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644

Makefile endef · line 1256

Quoted 7 lines
Makefile +2 -2
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

thanks,

greg k-h

#2 lore raw
Quoted 42 lines · Greg KH
Quoted 27 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644

Makefile endef · line 1256

Quoted 7 lines
Makefile +2 -2
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

Quoted 3 lines

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

Or, I can just stop doing stable releases at .255 and then abuse the EXTRAVERSION field to put in sub-revision values.

Or, we can just not worry about it as anyone using these really old kernels, userspace will work just fine (the number going backwards for these fields isn't going to break anything), it's only any crazy out-of-tree code that will get confused if they are trying to do different build options based on SUBLEVEL :)

thanks,

greg k-h

#3 lore raw
Quoted 47 lines · Greg KH
Quoted 42 lines · Greg KH
Quoted 27 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644

Makefile endef · line 1256

Quoted 7 lines
Makefile +2 -2
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

Quoted 3 lines

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

Quoted 8 lines

Or, I can just stop doing stable releases at .255 and then abuse the EXTRAVERSION field to put in sub-revision values.

Or, we can just not worry about it as anyone using these really old kernels, userspace will work just fine (the number going backwards for these fields isn't going to break anything), it's only any crazy out-of-tree code that will get confused if they are trying to do different build options based on SUBLEVEL :)

I think it would also affect code that doesn't do things based on SBULEVEL. Consider something like:

	if (LINUX_VERSION_CODE < KERNEL_VERSION(4,5,0))

Which will cause 4.4.256 to now change the result of that comparison.

Signature · 2 lines
Thanks,
Sasha
#4 lore raw
Quoted 57 lines · Sasha Levin
Quoted 47 lines · Greg KH
Quoted 42 lines · Greg KH
Quoted 27 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644

Makefile endef · line 1256

Quoted 7 lines
Makefile +2 -2
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

Quoted 3 lines

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

Quoted 8 lines

Or, I can just stop doing stable releases at .255 and then abuse the EXTRAVERSION field to put in sub-revision values.

Or, we can just not worry about it as anyone using these really old kernels, userspace will work just fine (the number going backwards for these fields isn't going to break anything), it's only any crazy out-of-tree code that will get confused if they are trying to do different build options based on SUBLEVEL :)

Quoted 6 lines

I think it would also affect code that doesn't do things based on SBULEVEL. Consider something like:

	if (LINUX_VERSION_CODE < KERNEL_VERSION(4,5,0))

Which will cause 4.4.256 to now change the result of that comparison.

Sure, but there are no in-kernel users like this, so my sympathy is quite low :)

#5 lore raw
Quoted 10 lines · Greg KH
Quoted 6 lines · Sasha Levin

I think it would also affect code that doesn't do things based on SBULEVEL. Consider something like:

	if (LINUX_VERSION_CODE < KERNEL_VERSION(4,5,0))

Which will cause 4.4.256 to now change the result of that comparison.

Sure, but there are no in-kernel users like this, so my sympathy is quite low :)

Wouldn't it be an issue for the hacky in-kernel users too? For example, right now the USB code does:

	#define KERNEL_REL      bin2bcd(((LINUX_VERSION_CODE >> 16) & 0x0ff))
	#define KERNEL_VER      bin2bcd(((LINUX_VERSION_CODE >> 8) & 0x0ff))

After 4.4.256, KERNEL_VER will be (5) rather than (4), indicating a version of 4.5.

Signature · 2 lines
Thanks,
Sasha
#6 lore raw
Quoted 12 lines · Sasha Levin
Quoted 10 lines · Greg KH
Quoted 6 lines · Sasha Levin

I think it would also affect code that doesn't do things based on SBULEVEL. Consider something like:

	if (LINUX_VERSION_CODE < KERNEL_VERSION(4,5,0))

Which will cause 4.4.256 to now change the result of that comparison.

Sure, but there are no in-kernel users like this, so my sympathy is quite low :)

Quoted 8 lines

Wouldn't it be an issue for the hacky in-kernel users too? For example, right now the USB code does:

	#define KERNEL_REL      bin2bcd(((LINUX_VERSION_CODE >> 16) & 0x0ff))
	#define KERNEL_VER      bin2bcd(((LINUX_VERSION_CODE >> 8) & 0x0ff))

After 4.4.256, KERNEL_VER will be (5) rather than (4), indicating a version of 4.5.

Which, really, is just fine. This is an informational string that shows up in 'lsusb' for root hubs. Same for V4L devices, they just want to send some string to userspace. Yes, it might look odd, but nothing is going to break, it's just a string :)

thanks,

greg k-h

#7 lore raw
In reply to Greg KH
Quoted 30 lines · Greg KH
Quoted 28 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644
@@ -1252,8 +1252,8 @@ endef
 
Quoted 12 lines

Makefile endef · line 1256

Quoted 6 lines
Makefile +2 -2
 define filechk_version.h
      echo \#define LINUX_VERSION_CODE $(shell                         \
-     expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-     echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+     expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+     echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

I can see a checkpatch warning about LINUX_VERSION_CODE.

See line 4528 of scripts/checkpatch.pl

WARN("LINUX_VERSION_CODE",
"LINUX_VERSION_CODE should be avoided, code should be for the
version to which it is merged\n" . $herecurr);

It helps external modules to be compiled for multiple kernel versions.

#if KERNEL_VERSION_CODE < KERNEL_VERSION(5, 4, 0)
code for the kernel versions older than 5.4.0
#endif

The upstream code does not do this.
But, LINUX_VERSION_CODE is actually used in many places...

Quoted 7 lines

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

thanks,

greg k-h

Signature · 2 lines
Best Regards
Masahiro Yamada
#8 lore raw
Quoted 58 lines · Masahiro Yamada
Quoted 30 lines · Greg KH
Quoted 28 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644
@@ -1252,8 +1252,8 @@ endef
 
Quoted 12 lines

Makefile endef · line 1256

Quoted 6 lines
Makefile +2 -2
 define filechk_version.h
      echo \#define LINUX_VERSION_CODE $(shell                         \
-     expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-     echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+     expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+     echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

I can see a checkpatch warning about LINUX_VERSION_CODE.

See line 4528 of scripts/checkpatch.pl

WARN("LINUX_VERSION_CODE",
"LINUX_VERSION_CODE should be avoided, code should be for the
version to which it is merged\n" . $herecurr);

It helps external modules to be compiled for multiple kernel versions.

Quoted 7 lines

#if KERNEL_VERSION_CODE < KERNEL_VERSION(5, 4, 0)
code for the kernel versions older than 5.4.0
#endif

The upstream code does not do this.
But, LINUX_VERSION_CODE is actually used in many places...

Yes, it is used in a number of user/kernel apis for various reasons.

And the above patch will break them :(

thanks,

greg k-h

#9 lore raw
In reply to Masahiro Yamada

From: Masahiro Yamada

Quoted 1 line

Sent: 18 January 2021 10:28

...

Quoted 5 lines

It helps external modules to be compiled for multiple kernel versions.

#if KERNEL_VERSION_CODE < KERNEL_VERSION(5, 4, 0)
code for the kernel versions older than 5.4.0
#endif

I've just done a scan through some drivers.
The only checks with a non-zero sub-rev are for 2.6.nnn. So provided KERNEL_VERSION_CODE is changed to match nothing of ours breaks.

I've only found tests for the following:
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 28)
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 34)
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 10, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 17, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 13, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 17, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 18, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 1, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 7, 0)
#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 8, 0)

Although some seem to be against VERSION_CODE()
rather than KERNEL_VERSION().

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#10 lore raw
In reply to Greg KH
Quoted 42 lines · Greg KH
Quoted 27 lines · Sasha Levin

SUBLEVEL only has 8 bits of space, which means that we'll overflow it once it reaches 256.

Few of the stable branches will imminently overflow SUBLEVEL while there's no risk of overflowing VERSION.

Thus, give SUBLEVEL 8 more bits which will be stolen from VERSION, this should create a better balance between the different version numbers we use.

The downside here is that Linus will have 8 bits less to play with, but given our current release cadence (~10 weeks), the number of Linus's fingers & toes (20), and the current VERSION (5) we can calculate that VERSION will overflow in just over 1,000 years, so I'm kicking this can down the road.

---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Makefile +0 -0
index 9e73f82e0d863..dc2bad7a440d8 100644

Makefile endef · line 1256

Quoted 7 lines
Makefile +2 -2
@@ -1252,8 +1252,8 @@ endef
 
 define filechk_version.h
 	echo \#define LINUX_VERSION_CODE $(shell                         \
-	expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
-	echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))'
+	expr $(VERSION) \* 16777216 + 0$(PATCHLEVEL) \* 65536 + 0$(SUBLEVEL)); \
+	echo '#define KERNEL_VERSION(a,b,c) (((a) << 24) + ((b) << 16) + (c))'

As much as I agree, this will break in-tree users of LINUX_VERSION_CODE that try to suck out the version/patchlevel number of the kernel release into their own fields. Things like USB host controller strings, v4l ioctl reports, scsi driver ioctls, and other places do fun bit-movements to try to unreverse this bit packing.

Quoted 3 lines

So how about we just provide a "real" version/subversion/revision #define as well, and clean up all in-kernel users, so we can get this to work, and we can change it in the future more easily.

Uh, yes, I see what you mean. I'll fix those up and resend.

Signature · 2 lines
Thanks,
Sasha