why the early_initcall(au1x00_setup) do not work?

why the early_initcall(au1x00_setup) do not work?

[zhuzhenhua]

i download a standard 2.6.14 kernel, and compile it for dbau1100, and
i find the early_initcall(au1x00_setup) was not compiled into the
vmlinux.
and i find at the linux-mips cvs, it used plat_setup instead of early_initcall.
does it means my toolchain is not correct to compile the
early_initcall, or in the standard 2.6.14 kernel, the early_initcall
do not work well?

Best regards!

zhuzhenhua

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Mon, Jan 09, 2006 at 01:59:34PM +0800, zhuzhenhua wrote:

> i download a standard 2.6.14 kernel, and compile it for dbau1100, and
> i find the early_initcall(au1x00_setup) was not compiled into the
> vmlinux.
> and i find at the linux-mips cvs, it used plat_setup instead of early_initcall.
> does it means my toolchain is not correct to compile the
> early_initcall, or in the standard 2.6.14 kernel, the early_initcall
> do not work well?

You've downloaded a kernel.org kernel it would seem - doesn't fly for MIPS.
Instead get a kernel from linux-mips.org.

The early_initcall() construct has been removed.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[P. Christeas]

On Monday 09 January 2006 4:56 pm, Ralf Baechle wrote:

> You've downloaded a kernel.org kernel it would seem - doesn't fly for MIPS.
> Instead get a kernel from linux-mips.org.
>
> The early_initcall() construct has been removed.
>
>   Ralf

What's the difference between the trees?
Aren't the MIPS patches supposed to be merged to Linus' tree?

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Mon, Jan 09, 2006 at 05:20:01PM +0200, P. Christeas wrote:

> > You've downloaded a kernel.org kernel it would seem - doesn't fly for MIPS.
> > Instead get a kernel from linux-mips.org.
> >
> > The early_initcall() construct has been removed.
> >
> >   Ralf
> 
> What's the difference between the trees?

All the MIPS work is happening in the linux-mips.org tree.

> Aren't the MIPS patches supposed to be merged to Linus' tree?

In 2.6.15 things were alomst fully merged but several megabytes of
patches are between the linux-mips.org and kernel.org versions of 2.6.14.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[zhuzhenhua]

On 1/9/06, Ralf Baechle <ralf@linux-mips.org> wrote:
> On Mon, Jan 09, 2006 at 05:20:01PM +0200, P. Christeas wrote:
>
> > > You've downloaded a kernel.org kernel it would seem - doesn't fly for MIPS.
> > > Instead get a kernel from linux-mips.org.
> > >
> > > The early_initcall() construct has been removed.
> > >
> > >   Ralf
> >
> > What's the difference between the trees?
>
> All the MIPS work is happening in the linux-mips.org tree.
>
> > Aren't the MIPS patches supposed to be merged to Linus' tree?
>
> In 2.6.15 things were alomst fully merged but several megabytes of
> patches are between the linux-mips.org and kernel.org versions of 2.6.14.
>
>   Ralf
>

in linux-mips, where to download the patches for standard kernel?

zhuzhenhua

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:

> > In 2.6.15 things were alomst fully merged but several megabytes of
> > patches are between the linux-mips.org and kernel.org versions of 2.6.14.
> >
> >   Ralf
> >
> 
> in linux-mips, where to download the patches for standard kernel?

I don't publish such patches - but they're eassy to generate.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/10, Ralf Baechle <ralf@linux-mips.org>:
> On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:
>
> > > In 2.6.15 things were alomst fully merged but several megabytes of
> > > patches are between the linux-mips.org and kernel.org versions of 2.6.14.
> > >
> > >   Ralf
> > >
> >
> > in linux-mips, where to download the patches for standard kernel?
>
> I don't publish such patches - but they're eassy to generate.
>
Could you be a bit more explicit about that "easy" way you generate them ?

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[P. Christeas]

On Tuesday 10 January 2006 5:40 pm, Ivan Korzakow wrote:
> 2006/1/10, Ralf Baechle <ralf@linux-mips.org>:
> > On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:
> > > > In 2.6.15 things were alomst fully merged but several megabytes of
> > > > patches are between the linux-mips.org and kernel.org versions of
> > > > 2.6.14.
> > > >
> > > >   Ralf
> > >
> > > in linux-mips, where to download the patches for standard kernel?
> >
> > I don't publish such patches - but they're eassy to generate.
>
> Could you be a bit more explicit about that "easy" way you generate them ?
>
> Ivan
You make sure you have the two trees and diff them.
They 're both in git and typically you could do that only using git.

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/10, P. Christeas <p_christ@hol.gr>:
> On Tuesday 10 January 2006 5:40 pm, Ivan Korzakow wrote:
> > 2006/1/10, Ralf Baechle <ralf@linux-mips.org>:
> > > On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:
> > > > > In 2.6.15 things were alomst fully merged but several megabytes of
> > > > > patches are between the linux-mips.org and kernel.org versions of
> > > > > 2.6.14.
> > > > >
> > > > >   Ralf
> > > >
> > > > in linux-mips, where to download the patches for standard kernel?
> > >
> > > I don't publish such patches - but they're eassy to generate.
> >
> > Could you be a bit more explicit about that "easy" way you generate them ?
> >
> > Ivan
> You make sure you have the two trees and diff them.
> They 're both in git and typically you could do that only using git.
>

Have you ever tried what you're talking about or is it a guess ?
For example, let's say that there's a bug introduced when merging
Linus tree with mips branch. How do you easily "bisect" in order to do
a binary seek of this bug ?

BTW, I noticed that Linux mips repository have several branches. Each
branch is tracking each kernel minor version. I think this is
completly useless :
1 - it makes the repository huge by keeping history since 2.2 (It's
pointless for most people)
2 - The way branches are made is broken : you can not fetch separate
branches without doing some "grafts" things

Why not simply ask Linus to pull from the mips tree ?

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[P. Christeas]

On Tuesday 10 January 2006 6:38 pm, Ivan Korzakow wrote:
> 2006/1/10, P. Christeas <p_christ@hol.gr>:

> > You make sure you have the two trees and diff them.
> > They 're both in git and typically you could do that only using git.
>
> Have you ever tried what you're talking about or is it a guess ?
> For example, let's say that there's a bug introduced when merging
> Linus tree with mips branch. How do you easily "bisect" in order to do
> a binary seek of this bug ?

 I was about to mention the git incompatibility as I gave you the first 
answer. It's true what you say and I also find it annoying. That's why I use 
Linus tree.
I cannot, though, complaint to Ralph about this practice. It seems that he has 
done his best to preserve the CVS history of his tree. We can only thank him 
for that. It is just a lot of work to trim the tree and make it 
Linus-parallel.
There is also things that I don't know about git, so I won't jump to 
conclusions yet (as to whether it is feasible to merge with Linus).

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/10, P. Christeas <p_christ@hol.gr>:
> On Tuesday 10 January 2006 6:38 pm, Ivan Korzakow wrote:
> > 2006/1/10, P. Christeas <p_christ@hol.gr>:
>
> > > You make sure you have the two trees and diff them.
> > > They 're both in git and typically you could do that only using git.
> >
> > Have you ever tried what you're talking about or is it a guess ?
> > For example, let's say that there's a bug introduced when merging
> > Linus tree with mips branch. How do you easily "bisect" in order to do
> > a binary seek of this bug ?
>
>  I was about to mention the git incompatibility as I gave you the first
> answer. It's true what you say and I also find it annoying. That's why I use
> Linus tree.
> I cannot, though, complaint to Ralph about this practice. It seems that he has
> done his best to preserve the CVS history of his tree. We can only thank him
> for that. It is just a lot of work to trim the tree and make it
> Linus-parallel.
Why not simply keep CVS repository available for 1% of people willing
to browse the history ? And make life easier for 99% of people willing
to work on 2.6 ... (2.4 work may continue to use CVS too).

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[Jordan Crouse]

> Have you ever tried what you're talking about or is it a guess ?

I just did that.  I have Linus's tree and tags, and Ralf's tree:

git checkout mips
git diff v2.6.14..HEAD

the nice thing about git is that it was written to do exactly the sort 
of things we need it to do, like this.

> For example, let's say that there's a bug introduced when merging
> Linus tree with mips branch. How do you easily "bisect" in order to do
> a binary seek of this bug ?

Linus wrote a HOWTO on that very subject:

http://www.kernel.org/pub/software/scm/git/docs/howto/isolate-bugs-with-bisect.txt

> 2 - The way branches are made is broken : you can not fetch separate
> branches without doing some "grafts" things

I'm no git expert, but can't you just

git-pull rsync://ftp.linux-mips.org/git/linux.git remote:local

That works for me.

Regards,
Jordan

-- 
Jordan Crouse
Senior Linux Engineer
AMD - Personal Connectivity Solutions Group
<www.amd.com/embeddedprocessors>

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Tue, Jan 10, 2006 at 06:57:24PM +0200, P. Christeas wrote:

> > > You make sure you have the two trees and diff them.
> > > They 're both in git and typically you could do that only using git.
> >
> > Have you ever tried what you're talking about or is it a guess ?
> > For example, let's say that there's a bug introduced when merging
> > Linus tree with mips branch. How do you easily "bisect" in order to do
> > a binary seek of this bug ?
> 
>  I was about to mention the git incompatibility as I gave you the first 
> answer. It's true what you say and I also find it annoying. That's why I use 
> Linus tree.
> I cannot, though, complaint to Ralph about this practice. It seems that he has 
> done his best to preserve the CVS history of his tree. We can only thank him 
> for that. It is just a lot of work to trim the tree and make it 
> Linus-parallel.
> There is also things that I don't know about git, so I won't jump to 
> conclusions yet (as to whether it is feasible to merge with Linus).

It is possible to keep both Linus's and the lmo tree in the same
repository with a _little_ care.  I do that all the time.  When compressed
this will result in a bloat of just about 10-20MB.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Tue, Jan 10, 2006 at 06:15:18PM +0100, Ivan Korzakow wrote:

> Why not simply keep CVS repository available for 1% of people willing
> to browse the history ? And make life easier for 99% of people willing
> to work on 2.6 ... (2.4 work may continue to use CVS too).

The CVS repository is still available - and will stay for a long time so
people have a chance to do diffs against their existing checked out trees.
But no more changes.

Note, this does not affect the other projects in the CVS server.  Each
of them is maintained independantly and for each the decission to use CVS,
git or some flying saucer technology to use is made by whoever is the
caretaker of that project.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[zhuzhenhua]

On 1/10/06, Ralf Baechle <ralf@linux-mips.org> wrote:
> On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:
>
> > > In 2.6.15 things were alomst fully merged but several megabytes of
> > > patches are between the linux-mips.org and kernel.org versions of 2.6.14.
> > >
> > >   Ralf
> > >
> >
> > in linux-mips, where to download the patches for standard kernel?
>
> I don't publish such patches - but they're eassy to generate.
i find on http://www.linux-mips.org/pub/linux/mips/kernel/v2.6/ can
download the kernel tarball, i compire the 2.6.14 with the stand
kernel, and found they are not same.

is the tarball in
http://www.linux-mips.org/pub/linux/mips/kernel/v2.6/ is tared with
the cvs tree code in linux-mips?



>
>   Ralf
>

Best regards

zhuzhenhua

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/10, Ralf Baechle <ralf@linux-mips.org>:
> On Tue, Jan 10, 2006 at 06:57:24PM +0200, P. Christeas wrote:
>
> > > > You make sure you have the two trees and diff them.
> > > > They 're both in git and typically you could do that only using git.
> > >
> > > Have you ever tried what you're talking about or is it a guess ?
> > > For example, let's say that there's a bug introduced when merging
> > > Linus tree with mips branch. How do you easily "bisect" in order to do
> > > a binary seek of this bug ?
> >
> >  I was about to mention the git incompatibility as I gave you the first
> > answer. It's true what you say and I also find it annoying. That's why I use
> > Linus tree.
> > I cannot, though, complaint to Ralph about this practice. It seems that he has
> > done his best to preserve the CVS history of his tree. We can only thank him
> > for that. It is just a lot of work to trim the tree and make it
> > Linus-parallel.
> > There is also things that I don't know about git, so I won't jump to
> > conclusions yet (as to whether it is feasible to merge with Linus).
>
> It is possible to keep both Linus's and the lmo tree in the same
> repository with a _little_ care.  I do that all the time.  When compressed
> this will result in a bloat of just about 10-20MB.
>

It would be great to be a little bit more explicit by giving some
_little_ examples ! Why not enlighting us directly instead of being so
vague.

Thanks

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Wed, Jan 11, 2006 at 12:10:05PM +0100, Ivan Korzakow wrote:

> It would be great to be a little bit more explicit by giving some
> _little_ examples ! Why not enlighting us directly instead of being so
> vague.

In books that kind of stuff is usually marked as "left as an exercise to
the reader" ;-)

  git-clone rsync://ftp.linux-mips.org/pub/scm/linux.git repository
  cd repository
  git-fetch git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6 master:linus
  git-repack -a -d

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Wed, Jan 11, 2006 at 10:04:50AM +0800, zhuzhenhua wrote:

> On 1/10/06, Ralf Baechle <ralf@linux-mips.org> wrote:
> > On Tue, Jan 10, 2006 at 09:02:43AM +0800, zhuzhenhua wrote:
> >
> > > > In 2.6.15 things were alomst fully merged but several megabytes of
> > > > patches are between the linux-mips.org and kernel.org versions of 2.6.14.
> > > >
> > > >   Ralf
> > > >
> > >
> > > in linux-mips, where to download the patches for standard kernel?
> >
> > I don't publish such patches - but they're eassy to generate.
> i find on http://www.linux-mips.org/pub/linux/mips/kernel/v2.6/ can
> download the kernel tarball, i compire the 2.6.14 with the stand
> kernel, and found they are not same.
> 
> is the tarball in
> http://www.linux-mips.org/pub/linux/mips/kernel/v2.6/ is tared with
> the cvs tree code in linux-mips?

Some tarballs were rather old, were created before the switch to git.
The 2.4 and 2.6 tarballs were created from the git tags.

In case of doubt you could always try git-get-tar-commit-id on the tarball.

  Ralf

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/11, Ralf Baechle <ralf@linux-mips.org>:
> On Wed, Jan 11, 2006 at 12:10:05PM +0100, Ivan Korzakow wrote:
>
> > It would be great to be a little bit more explicit by giving some
> > _little_ examples ! Why not enlighting us directly instead of being so
> > vague.
>
> In books that kind of stuff is usually marked as "left as an exercise to
> the reader" ;-)

yeah, but we are chating on a forum ;0)

>
>   git-clone rsync://ftp.linux-mips.org/pub/scm/linux.git repository
>   cd repository
>   git-fetch git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6 master:linus
>   git-repack -a -d
>

I know that but you miss my point. GIT is a tool to ease work on linux
kernel, but the way you use it makes harder life of users of your
tree. For example your tree contains more than 350 000 objects ! That
makes a lot of git commands running slow...

Let's say I'm developing a net drivers on ARM platform. I'm actually
do not care about ARM development, but I do care about net tree. To do
that, I just need to clone net tree because I know that ARM should be
OK with this tree. What about MIPS ?

I'm just wondering why not asking to Linus to pull from your tree like
every others maintainers do ?

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/10, Ralf Baechle <ralf@linux-mips.org>:
> It is possible to keep both Linus's and the lmo tree in the same
> repository with a _little_ care.  I do that all the time.  When compressed
> this will result in a bloat of just about 10-20MB.
>

well, in a MIPS repo:

$ git repack -a -d
$ git prune-packed
$ git prune
$ du .git
266M    .git
$ tree .git/objects/
.git/objects/
|-- info
|   `-- packs
`-- pack
    |-- pack-77ee75692f2944708c9dd65d6ba9100f6647b414.idx
    `-- pack-77ee75692f2944708c9dd65d6ba9100f6647b414.pack

2 directories, 3 files

What kind of magical git commands are you using ?

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[Ivan Korzakow]

2006/1/11, Ralf Baechle <ralf@linux-mips.org>:
> On Tue, Jan 10, 2006 at 06:15:18PM +0100, Ivan Korzakow wrote:
>
> > Why not simply keep CVS repository available for 1% of people willing
> > to browse the history ? And make life easier for 99% of people willing
> > to work on 2.6 ... (2.4 work may continue to use CVS too).
>
> The CVS repository is still available - and will stay for a long time so
> people have a chance to do diffs against their existing checked out trees.
> But no more changes.
>

again why not letting 2.2 2.4 stay in CVS ? I don't think there are
any active development, isn't it ?

Ivan

Re: why the early_initcall(au1x00_setup) do not work?

[Ralf Baechle]

On Wed, Jan 11, 2006 at 04:07:25PM +0100, Ivan Korzakow wrote:

> I know that but you miss my point. GIT is a tool to ease work on linux
> kernel, but the way you use it makes harder life of users of your
> tree. For example your tree contains more than 350 000 objects ! That
> makes a lot of git commands running slow...

Linus promised too look into that.  He has too - it's what his tree would
be facing in the not too distant future, otherwise.

> Let's say I'm developing a net drivers on ARM platform. I'm actually
> do not care about ARM development, but I do care about net tree. To do
> that, I just need to clone net tree because I know that ARM should be
> OK with this tree. What about MIPS ?
> 
> I'm just wondering why not asking to Linus to pull from your tree like
> every others maintainers do ?

Like most other developers I create throw-away trees for that purpose, see
the upstream-linus tree.  Due to the way the Linux release process is
working anything else is unrealistic.

  Ralf