A recent bug report led to a discussion about potentially dropping support for pre-4.0 versions of GCC. Adrian Bunk noted, "currently we support 6 different stable gcc release series, and it might be the right time to consider dropping support for the older ones. Are there any architectures still requiring a gcc < 4.0 ?" Russell King noted that on some architectures GCC 3.x is still preferable to the newer 4.x branch, "I want to keep support for gcc 3.4.3 for ARM for the foreseeable future. From my point of view, gcc 4 compilers have been something of a development thing as far as the ARM architecture goes. Also, gcc 3.4.3 is faster and significantly less noisy than gcc 4."

When it was asked how many kernel developers use older version of GCC, Linus Torvalds explained that it really doesn't matter, "it's NOT about 'kernel developers'. It's about random people testing kernels. If we make it harder for people to test kernels, we're going to lose. So no, I vote for *not* cutting off old gcc versions unless it's absolutely fatal."

From:	Adrian Bunk [email blocked]
Subject: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 19:35:50 +0200

On Tue, Aug 21, 2007 at 09:31:03AM -0700, Randy Dunlap wrote:
> On Tue, 21 Aug 2007 15:20:38 +0200 Jarek Poplawski wrote:
> 
> > Here are some more of, probably well-known, warnings with attached
> > testing-only .config.
> >...
> > drivers/pci/msi.c:686: warning: weak declaration of `arch_msi_check_device' after first use results in unspecified behavior
> > drivers/pci/msi.c:698: warning: weak declaration of `arch_setup_msi_irqs' after first use results in unspecified behavior
> > drivers/pci/msi.c:718: warning: weak declaration of `arch_teardown_msi_irqs' after first use results in unspecified behavior
> >...
> 
> What gcc version?  I don't get the arch_ warnings in drivers/pci/msi.c.

Obviously a gcc <= 3.4 [1], and therefore no unit-at-a-time.

You can reproduce it with a more recent gcc when adding 
-fno-unit-at-a-time to the CFLAGS.

And it's becoming a real maintainance problem that not only this problem 
but also other problems like some section mismatches [2] are only 
present without unit-at-a-time.

Currently we support 6 different stable gcc release series, and it might 
be the right time to consider dropping support for the older ones.

Are there any architectures still requiring a gcc < 4.0 ?

> ~Randy

cu
Adrian

[1] unit-at-a-time was added in gcc 3.4, but on gcc 3.4 we disable it on
    i386 due to stack usage problems
[2] example: static __init function with exactly one caller, and this
             caller is non-__init

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed




From: Russell King [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 18:54:53 +0100

On Tue, Aug 21, 2007 at 07:35:50PM +0200, Adrian Bunk wrote:
> On Tue, Aug 21, 2007 at 09:31:03AM -0700, Randy Dunlap wrote:
> > On Tue, 21 Aug 2007 15:20:38 +0200 Jarek Poplawski wrote:
> > 
> > > Here are some more of, probably well-known, warnings with attached
> > > testing-only .config.
> > >...
> > > drivers/pci/msi.c:686: warning: weak declaration of `arch_msi_check_device' after first use results in unspecified behavior
> > > drivers/pci/msi.c:698: warning: weak declaration of `arch_setup_msi_irqs' after first use results in unspecified behavior
> > > drivers/pci/msi.c:718: warning: weak declaration of `arch_teardown_msi_irqs' after first use results in unspecified behavior
> > >...
> > 
> > What gcc version?  I don't get the arch_ warnings in drivers/pci/msi.c.
> 
> Obviously a gcc <= 3.4 [1], and therefore no unit-at-a-time.
> 
> You can reproduce it with a more recent gcc when adding 
> -fno-unit-at-a-time to the CFLAGS.
> 
> And it's becoming a real maintainance problem that not only this problem 
> but also other problems like some section mismatches [2] are only 
> present without unit-at-a-time.
> 
> Currently we support 6 different stable gcc release series, and it might 
> be the right time to consider dropping support for the older ones.
> 
> Are there any architectures still requiring a gcc < 4.0 ?

I want to keep support for gcc 3.4.3 for ARM for the forseeable future.
From my point of view, gcc 4 compilers have been something of a development
thing as far as the ARM architecture goes.  Also, gcc 3.4.3 is faster and
significantly less noisy than gcc 4.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:



From:	Kyle McMartin [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 14:14:20 -0400

On Tue, Aug 21, 2007 at 06:54:53PM +0100, Russell King wrote:
> I want to keep support for gcc 3.4.3 for ARM for the forseeable future.
> From my point of view, gcc 4 compilers have been something of a development
> thing as far as the ARM architecture goes.  Also, gcc 3.4.3 is faster and
> significantly less noisy than gcc 4.
> 

+1, similar reasons, for parisc. Not that gcc4 doesn't work, but there's
some cases where gcc 3.4 still generates better code...

Regards,
	Kyle M.



From:	Andi Kleen [email blocked]
To:	Adrian Bunk [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 21:19:59 +0200

On Tue, Aug 21, 2007 at 07:35:50PM +0200, Adrian Bunk wrote:
> Obviously a gcc <= 3.4 [1], and therefore no unit-at-a-time.

Actually there are widely used 3.3 variants that support unit-at-a-time
(e.g. 3.3-hammer which was shipped by several distributions for some time)

There are still a lot of systems around which use gcc 3.3 (less so with
3.4). Unless there's a major bug that is hard to work around I would
prefer to keep it supported.

Bogus warnings should be relatively harmless.

> And it's becoming a real maintainance problem that not only this problem 
> but also other problems like some section mismatches [2] are only 
> present without unit-at-a-time.

The unit-at-a-time output order is not defined, so even if it works
with the current compiler a compiler change might still trigger
that problem. So it would be better to fix those anyways.

-Andi



From:	Adrian Bunk [email blocked]
To:	Andi Kleen [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 21:54:33 +0200

On Tue, Aug 21, 2007 at 09:19:59PM +0200, Andi Kleen wrote:
> On Tue, Aug 21, 2007 at 07:35:50PM +0200, Adrian Bunk wrote:
> > Obviously a gcc <= 3.4 [1], and therefore no unit-at-a-time.
> 
> Actually there are widely used 3.3 variants that support unit-at-a-time
> (e.g. 3.3-hammer which was shipped by several distributions for some time)
> 
> There are still a lot of systems around which use gcc 3.3 (less so with
> 3.4). Unless there's a major bug that is hard to work around I would
> prefer to keep it supported.
> 
> Bogus warnings should be relatively harmless.

How many kernel developers use such old gcc versions?

And how many people notice the valid modpost warnings that can indicate
a runtime Oops?

> > And it's becoming a real maintainance problem that not only this problem 
> > but also other problems like some section mismatches [2] are only 
> > present without unit-at-a-time.
> 
> The unit-at-a-time output order is not defined, so even if it works
> with the current compiler a compiler change might still trigger
> that problem. So it would be better to fix those anyways.

The example [2] from my email is guaranteed to not be a problem with 
unit-at-a-time (as long as unit-at-a-time implies 
inline-functions-called-once - and that's although theoretically 
possible quite unlikely to change in practice).

This example is for a bug that should be fixed, but my point is the
maintainability, IOW: issues with older compilers might not be 
discovered and fixed before they go into a stable kernel.

We currently support 6 different stable gcc release series plus heavily 
modified vendor branches like 3.3-hammer. We can discuss whether it is 
now already the right time, and where to make the cut, but medium-term 
we must reduce the number of supported compilers.

> -Andi

cu
Adrian

[2] example: static __init function with exactly one caller, and this
             caller is non-__init

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed



From:	Matthew Wilcox [email blocked]
To:	Adrian Bunk [email blocked]
Subject: Re: [RFC][PATCH] introduce TASK_SIZE_OF() for all arches
Date:	Tue, 21 Aug 2007 14:07:11 -0600

On Tue, Aug 21, 2007 at 09:54:33PM +0200, Adrian Bunk wrote:
> We currently support 6 different stable gcc release series plus heavily 
> modified vendor branches like 3.3-hammer. We can discuss whether it is 
> now already the right time, and where to make the cut, but medium-term 
> we must reduce the number of supported compilers.

I don't think that's feasible.  We're dealing with a situation where:

 - GCC aim to release a new compiler series every 6 months.
   Fortunately, they don't achieve this goal, but they do release every
   12 months or thereabouts [1].
 - Enterprise distros are supported for seven years
 - We still care about people being able to compile kernels on
   enterprise distros that are still supported by their vendor.

Yes, it causes us some pain to support all these different compilers,
but it's not *that* big a pain.

[1] Release dates, according to the GCC website

2007-05-17 4.2.0	(14 months)
2006-02-28 4.1.0	(10 months)
2005-04-20 4.0.0	(12 months)
2004-04-20 3.4.0	(11 months)
2003-05-20 3.3		(12 months, ignoring gcc 3.2 which was really 3.1.2)
2002-05-15 3.1		(11 months)
2001-06-18 3.0

-- 
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."



From:	Linus Torvalds [email blocked]
To:	Adrian Bunk [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 13:08:25 -0700 (PDT)



On Tue, 21 Aug 2007, Adrian Bunk wrote:
> > 
> > Bogus warnings should be relatively harmless.
> 
> How many kernel developers use such old gcc versions?

It's NOT about "kernel developers".

It's about random people testing kernels.

If we make it harder for people to test kernels, we're going to lose. So 
no, I vote for *not* cutting off old gcc versions unless it's absolutely 
fatal.

		Linus



From:	Adrian Bunk [email blocked]
To:	Linus Torvalds [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 22:21:13 +0200

On Tue, Aug 21, 2007 at 01:08:25PM -0700, Linus Torvalds wrote:
> 
> 
> On Tue, 21 Aug 2007, Adrian Bunk wrote:
> > > 
> > > Bogus warnings should be relatively harmless.
> > 
> > How many kernel developers use such old gcc versions?
> 
> It's NOT about "kernel developers".
> 
> It's about random people testing kernels.
> 
> If we make it harder for people to test kernels, we're going to lose. So 
> no, I vote for *not* cutting off old gcc versions unless it's absolutely 
> fatal.

How many people e.g. test -rc kernels compiled with gcc 3.2?

> 		Linus

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed



From:	Linus Torvalds [email blocked]
To:	Adrian Bunk [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 13:32:50 -0700 (PDT)



On Tue, 21 Aug 2007, Adrian Bunk wrote:
> 
> How many people e.g. test -rc kernels compiled with gcc 3.2?

How could anybody know?

		Linus



From:	Segher Boessenkool [email blocked]
To:	Adrian Bunk [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 22:49:49 +0200

> How many people e.g. test -rc kernels compiled with gcc 3.2?

Why would that matter?  It either works or not.  If it doesn't
work, it can either be fixed, or support for that old compiler
version can be removed.

The only other policy than "only remove support if things are
badly broken" would be "only support what the GCC team supports",
which would be >= 4.1 now; and there are very good arguments for
supporting more than that with the Linux kernel.


Segher



From:	Adrian Bunk [email blocked]
To:	Segher Boessenkool [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 23:21:29 +0200

On Tue, Aug 21, 2007 at 10:49:49PM +0200, Segher Boessenkool wrote:
>> How many people e.g. test -rc kernels compiled with gcc 3.2?
>
> Why would that matter?  It either works or not.  If it doesn't
> work, it can either be fixed, or support for that old compiler
> version can be removed.

One bug report "kernel doesn't work / crash / ... when compiled with
gcc 3.2, but works when compiled with gcc 4.2" will most likely be lost 
in the big pile of unhandled bugs, not cause the removal of gcc 3.2 
support...

> The only other policy than "only remove support if things are
> badly broken" would be "only support what the GCC team supports",
> which would be >= 4.1 now; and there are very good arguments for
> supporting more than that with the Linux kernel.

No, it's not about bugs in gcc, it's about kernel+gcc combinations that 
are mostly untested but officially supported.

E.g. how many kernel developers use kernels compiled without 
unit-at-a-time? And unit-at-a-time does paper over some bugs,
e.g. at about half a dozen section mismatch bugs I've fixed
recently are not present with it.

But as the discussions have shown gcc 4.0 is currently too high for 
making a cut, and it is not yet the right time for raising the minimum 
required gcc version.

> Segher

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed



From:	James Bottomley [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Tue, 21 Aug 2007 16:49:38 -0500

On Tue, 2007-08-21 at 23:21 +0200, Adrian Bunk wrote:
> On Tue, Aug 21, 2007 at 10:49:49PM +0200, Segher Boessenkool wrote:
> >> How many people e.g. test -rc kernels compiled with gcc 3.2?
> >
> > Why would that matter?  It either works or not.  If it doesn't
> > work, it can either be fixed, or support for that old compiler
> > version can be removed.
> 
> One bug report "kernel doesn't work / crash / ... when compiled with
> gcc 3.2, but works when compiled with gcc 4.2" will most likely be lost 
> in the big pile of unhandled bugs, not cause the removal of gcc 3.2 
> support...

What's the bugzilla or pointer to this report please?  Those of us who
use gcc-3 as the default kernel compiler will take it seriously (if it
looks to have an impact to our kernel builds) otherwise we can tell you
it's unreproducible/not a problem etc.

James



From:	Adrian Bunk [email blocked]
Subject: Re: RFC: drop support for gcc < 4.0
Date:	Wed, 22 Aug 2007 00:09:52 +0200

On Tue, Aug 21, 2007 at 04:49:38PM -0500, James Bottomley wrote:
> On Tue, 2007-08-21 at 23:21 +0200, Adrian Bunk wrote:
> > On Tue, Aug 21, 2007 at 10:49:49PM +0200, Segher Boessenkool wrote:
> > >> How many people e.g. test -rc kernels compiled with gcc 3.2?
> > >
> > > Why would that matter?  It either works or not.  If it doesn't
> > > work, it can either be fixed, or support for that old compiler
> > > version can be removed.
> > 
> > One bug report "kernel doesn't work / crash / ... when compiled with
> > gcc 3.2, but works when compiled with gcc 4.2" will most likely be lost 
> > in the big pile of unhandled bugs, not cause the removal of gcc 3.2 
> > support...
> 
> What's the bugzilla or pointer to this report please?  Those of us who
> use gcc-3 as the default kernel compiler will take it seriously (if it
> looks to have an impact to our kernel builds) otherwise we can tell you
> it's unreproducible/not a problem etc.

This was an example in response to Segher's point we would remove 
support for a gcc version in such a case.

I remember we had such issues, but I don't find any pointer to a 
specific one at the moment.

I'll keep you informed when bug reports come in that only occur with 
older gcc versions and that aren't easily fixable.

> James

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

• Archive of above thread
• KernelTrap interview with Russell King