Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Linus Torvalds]

On Tue, 13 Mar 2007, Steven Rostedt wrote:
> 
> What we have currently is a bunch of hacks.  Seems that people can't make
> up their mind to what to do.

I don't mind the patches, but I'd be a lot happier if it also was a stated 
intention to actually make it be buildable as "x86", the same way that the 
separate 32-bit and 64-bit POWER architectures were merged into just one 
architecture that could be built either way.

For the POWER merge, we had (and probably still have) legacy platforms 
that could only be built the old way (ie if you needed to build for 
certain legacy 32-bit targets, you still needed to use the "ppc" 
architecture, so I'm not saying that it has to all be converted, but I 
think that we should at least *aim* for unifying 32-bit and 64-bit x86.

The "32-bit code has legcay issues" thing that Andi complained about (eg 
there's no guarantee of a HPET on 32-bit x86) doesn't really change the 
fact that yes, we have to support those legacy issues *anyway*, and 64-bit 
x86 certainly has its set of issues already too.

We've started to notice that the i386 build gets broken now that most 
developers tend to have newer CPU's and run mostly on x86-64 (and yes, 
that's me too), and while I don't think unifying things will guarantee 
that doesn't happen in the future, it will hopefully at least help make it 
not get much *worse*.

As it is, the build environment has to know to pass in "-m32/-m64" 
anyway..

		Linus

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Steven Rostedt]

On Tue, 2007-03-13 at 14:39 -0700, Linus Torvalds wrote:
> 
> On Tue, 13 Mar 2007, Steven Rostedt wrote:
> > 
> > What we have currently is a bunch of hacks.  Seems that people can't make
> > up their mind to what to do.
> 
> I don't mind the patches, but I'd be a lot happier if it also was a stated 
> intention to actually make it be buildable as "x86", the same way that the 
> separate 32-bit and 64-bit POWER architectures were merged into just one 
> architecture that could be built either way.

That's actually a larger goal, but for the immediate future, I figure
this would be a good first step.  Start out by stating what's similar,
and then build off of this for something bigger.  But in the mean time,
we can have a staging ground for work that's for both i386 and x86_64
archs. And for those that know these systems in a more intimate way
(Andi :) they can work off of this to make that monster.

-- Steve

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Andi Kleen]

On Tue, Mar 13, 2007 at 02:39:33PM -0700, Linus Torvalds wrote:
> 
> 
> On Tue, 13 Mar 2007, Steven Rostedt wrote:
> > 
> > What we have currently is a bunch of hacks.  Seems that people can't make
> > up their mind to what to do.

I think they work fine. I don't like such large scale renaming -- they
are a pain for anybody with external patches and make regression hunting
later painful. And the gain is about zero as far as I can see.
It won't fix a single bug, just cause pain.

> 
> I don't mind the patches, but I'd be a lot happier if it also was a stated 
> intention to actually make it be buildable as "x86", the same way that the 
> separate 32-bit and 64-bit POWER architectures were merged into just one 
> architecture that could be built either way.

I've thought about it, but it would be a lot of work and regression
test on old hardware would be a nightmare. 

Besides still x86-64 is a lot cleaner than i386 and imho
easier to hack and with all the 32bit quirks readded it would probably 
become worse than current i386. The only good option would be 
a "modern 32bit only" but even that gets complicated quickly when
you consider all the corner cases. And a clean 32bit port wouldn't
cover enough hardware to be usable by distributions.

And I also don't have really time to work on that.

> The "32-bit code has legcay issues" thing that Andi complained about (eg 
> there's no guarantee of a HPET on 32-bit x86) doesn't really change the 

Most 64bit doesn't neither.

> fact that yes, we have to support those legacy issues *anyway*, and 64-bit 
> x86 certainly has its set of issues already too.

Yes :/  The more supported systems, the more junk.

The recent nmi watchdog issues are a good example. All just because
a few vendors write crappy AML/SMM code.

> We've started to notice that the i386 build gets broken now that most 
> developers tend to have newer CPU's and run mostly on x86-64 (and yes, 
> that's me too), and while I don't think unifying things will guarantee 
> that doesn't happen in the future, it will hopefully at least help make it 
> not get much *worse*.

Build test would be needed anyways, doesn't make much difference I guess.

> As it is, the build environment has to know to pass in "-m32/-m64" 
> anyway..

It already does that.

-Andi

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Rusty Russell]

On Wed, 2007-03-14 at 11:12 +0100, Andi Kleen wrote:
> On Tue, Mar 13, 2007 at 02:39:33PM -0700, Linus Torvalds wrote:
> > 
> > 
> > On Tue, 13 Mar 2007, Steven Rostedt wrote:
> > > 
> > > What we have currently is a bunch of hacks.  Seems that people can't make
> > > up their mind to what to do.
> 
> I think they work fine.

My first instinct was to agree with Andi, or maybe have a "shared"
directory a-la x86/.  But I had a fruitful discussion with the PowerPC
people, who went through (and are still completing) the transition to
one directory.

Their verdict, surprisingly, was definitely positive.  They started with
a single powerpc dir which simply included ppc64/ or ppc/, then pulled
parts into that directory until ppc64/ was gone.  The things left in
ppc/ are being slowly merged, and some may be dropped entirely.

They can obviously speak for themselves, but the cleanup caused by the
movement was seen as a net win.

Rusty.

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Ingo Molnar]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> We've started to notice that the i386 build gets broken now that most 
> developers tend to have newer CPU's and run mostly on x86-64 (and yes, 
> that's me too), and while I don't think unifying things will guarantee 
> that doesn't happen in the future, it will hopefully at least help 
> make it not get much *worse*.

actually, even for 'cutting edge' testers, the number of 32-bit x86 
systems outnumber the number of 64-bit test-systems by a factor of 5. 
For distributions with real users, the ratio is closer to 1:10...

so i typically get bugs reported on 32-bit first, and for 64-bit it is 
usually only reported if the bug is 64-bit /only/. That basic property 
is true for both upstream, -rt and rawhide kernels.

Andi's x86 merge tree /does/ break quite often on 32-bit, but that's not 
really due to testers, that's due to Andi being primarily the 64-bit 
maintainer ;)

	Ingo

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Chris Wright]

* Steven Rostedt (rostedt@goodmis.org) wrote:
> Recently I've been doing some work that will affect both the i386 and x86_64
> architectures.  So there will be common code for both, as well as code
> that will be unique for the specific arch.  So I was looking into a way
> to do this cleanly, and found that there is no clean way to share code
> between x86_64 and i386.

Thanks for taking this on.  I'm sure Andi has a bunch of ideas on this
topic, but it would be nice to see the consolidation.

> So the move patches are a simple move of one file, with the slight exception 
> of files that hold the speedstep-lib.h file.  This file was moved from the
> arch/i386/kerne/cpu/cpufreq directory and put into the include/asm-i386
> directory.  This was due to the fact that some of the moved files included
> it, and some files that were not moved also included it. Instead of using
> the #include "../../x86/" hack again, I just simply moved it to the global
> i386 include directory.  Only the arch/x86 will use the include/asm-i386

what about asm-x86/ dir?  the asm/ symlink would still point to relevant
arch, but the file there could be simply #include <asm-x86/file.h> ?

thanks,
-chris

Re: [RFC/PATCH 00/59] Make common x86 arch area for i386 and x86_64

[Steven Rostedt]

On Tue, 2007-03-13 at 14:45 -0700, Chris Wright wrote:

> what about asm-x86/ dir?  the asm/ symlink would still point to relevant
> arch, but the file there could be simply #include <asm-x86/file.h> ?

Would it be acceptable to have an include/asm-x86/ dir with one file?
Of course it will open the door to merge current code and share it there
too.

-- Steve