[ANNOUNCE:RFC] convert OMAP to use clkdev

[ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

All,

In a bid to clean up the clk API implementations, and more importantly
ensure that implementations are using the clk API correctly, I've
converted OMAP over to the generic clock lookup (clkdev) implementation.

In doing so, I've shaken out a number of OMAP quirks and OMAP1 specific
clock code from the drivers.

The set of patches (minus two[*]) are available in my 'omap-clks2' branch.
This is built on top of the omap-clks1 branch, which some people here
already know about.

I do need to make some points:

1. the omap-clks1 branch has been updated with various silly fixes I've
   found along the way, so this won't be the same as the omap-clks1 branch
   from a month ago.

2. omap-clks2 branch follows on from the omap-clks1 branch - in other words,
   arm:omap-clks2.diff is a superset of omap-clks1.diff.

3. clkdev's current matching algorithm leads to false positives on OMAP,
   so you need the clkdev fix which is in the master branch as well.

With this, I've been able to boot my LDP board (with the HS MMC driver)
through to a shell prompt, and kernel messages look good.  I've also a
bodgerated QEMU H2-based-on-OMAP310 environment which I've been checking
OMAP1 on, and that looks reasonable.  Obviously, it's no subsitute for
real hardware.

Reviews and comments obviously welcome, testing more preferred.  Probably
the easiest way to apply these changes for testing is to download the
diffs from the FTP site: arm.diff and arm:omap-clks2.diff, and apply
them in that order to a fresh mainline tree.

There's still a bit of work to be done here - I've left the original
'name' and 'id' members in the struct clk solely for the debugfs code.
Obviously this is wasteful if you don't have debugfs enabled.

Finally, the diffstat for the clkdev conversion for OMAP (omap-clks1
to omap-clks2) shows a saving of 500 lines of code - definitely
something not to be sniffed at:

 arch/arm/mach-omap1/clock.c              |  152 ++++++++--
 arch/arm/mach-omap1/clock.h              |  171 ++--------
 arch/arm/mach-omap1/mcbsp.c              |  133 ++------
 arch/arm/mach-omap2/clock24xx.c          |  197 +++++++++++-
 arch/arm/mach-omap2/clock24xx.h          |  353 ++-------------------
 arch/arm/mach-omap2/clock34xx.c          |  268 ++++++++++++++--
 arch/arm/mach-omap2/clock34xx.h          |  513 +++++-------------------------
 arch/arm/mach-omap2/devices.c            |    1 +
 arch/arm/mach-omap2/mcbsp.c              |  121 -------
 arch/arm/plat-omap/Kconfig               |    3 +
 arch/arm/plat-omap/clock.c               |   48 +---
 arch/arm/plat-omap/include/mach/clkdev.h |   13 +
 arch/arm/plat-omap/include/mach/clock.h  |   11 +-
 arch/arm/plat-omap/include/mach/mcbsp.h  |    4 +-
 arch/arm/plat-omap/mcbsp.c               |   66 +++--
 drivers/i2c/busses/i2c-omap.c            |   47 ++--
 drivers/mmc/host/omap.c                  |   24 +-
 drivers/spi/omap2_mcspi.c                |    4 +-
 drivers/spi/omap_uwire.c                 |    9 +-
 drivers/w1/masters/omap_hdq.c            |    4 +-
 drivers/watchdog/omap_wdt.c              |   94 ++-----
 21 files changed, 868 insertions(+), 1368 deletions(-)

[*] - these are for the RNG and OHCI drivers, and have been left out
for the time being because they conflict with my master branch, and
would make the omap-clks2 patch unnecessarily difficult for people to
try.

Note: please don't think about polluting mach/clkdev.h, thanks.

RE: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Woodruff, Richard]

> From: linux-omap-owner@vger.kernel.org [mailto:linux-omap-
> owner@vger.kernel.org] On Behalf Of Russell King - ARM Linux

> Reviews and comments obviously welcome, testing more preferred.  Probably
> the easiest way to apply these changes for testing is to download the
> diffs from the FTP site: arm.diff and arm:omap-clks2.diff, and apply
> them in that order to a fresh mainline tree.

Which/where on ftp site?

ftp://ftp.arm.linux.org.uk/

Regards,
Richard W.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

On Sat, Jan 24, 2009 at 10:41:26AM -0600, Woodruff, Richard wrote:
> 
> > From: linux-omap-owner@vger.kernel.org [mailto:linux-omap-
> > owner@vger.kernel.org] On Behalf Of Russell King - ARM Linux
> 
> > Reviews and comments obviously welcome, testing more preferred.  Probably
> > the easiest way to apply these changes for testing is to download the
> > diffs from the FTP site: arm.diff and arm:omap-clks2.diff, and apply
> > them in that order to a fresh mainline tree.
> 
> Which/where on ftp site?

Unfortunately, I can never remember the URL.  There's a link off of
http://www.arm.linux.org.uk/developer/ which takes you directly to
where I keep kernel patches.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Tony Lindgren]

[-- Attachment #1: Type: text/plain, Size: 4532 bytes --]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [090124 04:03]:
> All,
> 
> In a bid to clean up the clk API implementations, and more importantly
> ensure that implementations are using the clk API correctly, I've
> converted OMAP over to the generic clock lookup (clkdev) implementation.
> 
> In doing so, I've shaken out a number of OMAP quirks and OMAP1 specific
> clock code from the drivers.

To summarize, these patches solve the problem of matching clocks
with the driver where omap1 usually has one clock, and omap2 & 3 devices
have two clocks.

> The set of patches (minus two[*]) are available in my 'omap-clks2' branch.
> This is built on top of the omap-clks1 branch, which some people here
> already know about.

Russell, can you please post your omap-clks[12] patches here too for review
so people can look at them? It's ok to post a large series of patches to
this list.

> I do need to make some points:
> 
> 1. the omap-clks1 branch has been updated with various silly fixes I've
>    found along the way, so this won't be the same as the omap-clks1 branch
>    from a month ago.
> 
> 2. omap-clks2 branch follows on from the omap-clks1 branch - in other words,
>    arm:omap-clks2.diff is a superset of omap-clks1.diff.
> 
> 3. clkdev's current matching algorithm leads to false positives on OMAP,
>    so you need the clkdev fix which is in the master branch as well.
> 
> With this, I've been able to boot my LDP board (with the HS MMC driver)
> through to a shell prompt, and kernel messages look good.  I've also a
> bodgerated QEMU H2-based-on-OMAP310 environment which I've been checking
> OMAP1 on, and that looks reasonable.  Obviously, it's no subsitute for
> real hardware.

Seems to boot on overo after applying the attached hsmmc init fix.

> Reviews and comments obviously welcome, testing more preferred.  Probably
> the easiest way to apply these changes for testing is to download the
> diffs from the FTP site: arm.diff and arm:omap-clks2.diff, and apply
> them in that order to a fresh mainline tree.

Or even easier for the omap folks is to use the new clks-testing branch
I've just added to the linux-omap tree on kernel.org.

That's v2.6.26-rc2 + rmk/master + rmk/omap-clks1 + rmk/omap-clks2 +
one fix for hsmmc init (attached also).

> There's still a bit of work to be done here - I've left the original
> 'name' and 'id' members in the struct clk solely for the debugfs code.
> Obviously this is wasteful if you don't have debugfs enabled.

Is this series missing a patch to just clk_get ick and fck for
omap_hsmmc.c?

> Finally, the diffstat for the clkdev conversion for OMAP (omap-clks1
> to omap-clks2) shows a saving of 500 lines of code - definitely
> something not to be sniffed at:
> 
>  arch/arm/mach-omap1/clock.c              |  152 ++++++++--
>  arch/arm/mach-omap1/clock.h              |  171 ++--------
>  arch/arm/mach-omap1/mcbsp.c              |  133 ++------
>  arch/arm/mach-omap2/clock24xx.c          |  197 +++++++++++-
>  arch/arm/mach-omap2/clock24xx.h          |  353 ++-------------------
>  arch/arm/mach-omap2/clock34xx.c          |  268 ++++++++++++++--
>  arch/arm/mach-omap2/clock34xx.h          |  513 +++++-------------------------
>  arch/arm/mach-omap2/devices.c            |    1 +
>  arch/arm/mach-omap2/mcbsp.c              |  121 -------
>  arch/arm/plat-omap/Kconfig               |    3 +
>  arch/arm/plat-omap/clock.c               |   48 +---
>  arch/arm/plat-omap/include/mach/clkdev.h |   13 +
>  arch/arm/plat-omap/include/mach/clock.h  |   11 +-
>  arch/arm/plat-omap/include/mach/mcbsp.h  |    4 +-
>  arch/arm/plat-omap/mcbsp.c               |   66 +++--
>  drivers/i2c/busses/i2c-omap.c            |   47 ++--
>  drivers/mmc/host/omap.c                  |   24 +-
>  drivers/spi/omap2_mcspi.c                |    4 +-
>  drivers/spi/omap_uwire.c                 |    9 +-
>  drivers/w1/masters/omap_hdq.c            |    4 +-
>  drivers/watchdog/omap_wdt.c              |   94 ++-----
>  21 files changed, 868 insertions(+), 1368 deletions(-)

Shrinking the clock*-h is a good thing.

However my main concern is that this series is going to cause a massive
manual merge pain for us to get in sync with the linux-omap tree.

Regards,

Tony


> [*] - these are for the RNG and OHCI drivers, and have been left out
> for the time being because they conflict with my master branch, and
> would make the omap-clks2 patch unnecessarily difficult for people to
> try.
> 
> Note: please don't think about polluting mach/clkdev.h, thanks.

Huh?

[-- Attachment #2: hsmmc-init-fix.patch --]
[-- Type: text/x-diff, Size: 926 bytes --]

commit 418c92c9fa4e7e56cacd14f0ed24cf7cfdc3f477
Author: Tony Lindgren <tony@atomide.com>
Date:   Sat Jan 24 08:34:37 2009 -0800

    Fix hsmmc init
    
    It accidentally broke while changing the name for the driver
    to not to conflict with the other mmc driver.
    
    Signed-off-by: Tony Lindgren <tony@atomide.com>

diff --git a/arch/arm/plat-omap/devices.c b/arch/arm/plat-omap/devices.c
index ac15c23..d22529c 100644
--- a/arch/arm/plat-omap/devices.c
+++ b/arch/arm/plat-omap/devices.c
@@ -205,9 +205,15 @@ int __init omap_mmc_add(int id, unsigned long base, unsigned long size,
 {
 	struct platform_device *pdev;
 	struct resource res[OMAP_MMC_NR_RES];
+	char *name;
 	int ret;
 
-	pdev = platform_device_alloc("mmci-omap", id);
+	if (cpu_class_is_omap1() || cpu_is_omap242x())
+		name = "mmci-omap";
+	else
+		name = "mmci-omap-hs";
+
+	pdev = platform_device_alloc(name, id);
 	if (!pdev)
 		return -ENOMEM;
 

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

On Sat, Jan 24, 2009 at 09:04:08AM -0800, Tony Lindgren wrote:
> Russell, can you please post your omap-clks[12] patches here too for review
> so people can look at them? It's ok to post a large series of patches to
> this list.

TBH, I haven't worked out how to get git to sanely do that yet, and
I suspect that git-send-email will overload my mail server at this
end due to lack of grace periods between messages.  However, I'll
try to sort something out over the next week or so.

> > There's still a bit of work to be done here - I've left the original
> > 'name' and 'id' members in the struct clk solely for the debugfs code.
> > Obviously this is wasteful if you don't have debugfs enabled.
> 
> Is this series missing a patch to just clk_get ick and fck for
> omap_hsmmc.c?

Yes and no - I can't do that until we sort out where/when omap_hsmmc is
going to get merged.

> Shrinking the clock*-h is a good thing.
> 
> However my main concern is that this series is going to cause a massive
> manual merge pain for us to get in sync with the linux-omap tree.

That's something I can't help.

I requested that the clock changes be merged upstream ASAP several months
ago so I could work on the OMAP clock API without causing major merge
pain.  Unfortunately, this hasn't been forthcoming, and after waiting a
reasonable amount of time, I've given up waiting.

I had given up all hope of ever seeing the omap-clks1 branch in mainline.
Really.  Until I saw a comment from you about apparantly wanting to see it
upstream.  However, I still have a very big question whether any of this
work will ever reach mainline, especially since my experimental omap-clks1
branch seems to have been broken up with random bits applied to your tree
with zero discussion with me.

> > [*] - these are for the RNG and OHCI drivers, and have been left out
> > for the time being because they conflict with my master branch, and
> > would make the omap-clks2 patch unnecessarily difficult for people to
> > try.
> > 
> > Note: please don't think about polluting mach/clkdev.h, thanks.
> 
> Huh?

Don't put anything else in mach/clkdev.h - it's full.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Tony Lindgren]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [090124 09:38]:
> On Sat, Jan 24, 2009 at 09:04:08AM -0800, Tony Lindgren wrote:
> > Russell, can you please post your omap-clks[12] patches here too for review
> > so people can look at them? It's ok to post a large series of patches to
> > this list.
> 
> TBH, I haven't worked out how to get git to sanely do that yet, and
> I suspect that git-send-email will overload my mail server at this
> end due to lack of grace periods between messages.  However, I'll
> try to sort something out over the next week or so.

I've found that stg-mail works pretty good. At least I like it more
than git-send-email. And it has the --sleep option between mails.

You can also output the messages into a mbox first, then formail to
send them. I have not tried that, but been meaning to.

> > > There's still a bit of work to be done here - I've left the original
> > > 'name' and 'id' members in the struct clk solely for the debugfs code.
> > > Obviously this is wasteful if you don't have debugfs enabled.
> > 
> > Is this series missing a patch to just clk_get ick and fck for
> > omap_hsmmc.c?
> 
> Yes and no - I can't do that until we sort out where/when omap_hsmmc is
> going to get merged.

OK. Still seems to work with old clock names too.

Also, your patches boot fine on osk5912.

> > Shrinking the clock*-h is a good thing.
> > 
> > However my main concern is that this series is going to cause a massive
> > manual merge pain for us to get in sync with the linux-omap tree.
> 
> That's something I can't help.
> 
> I requested that the clock changes be merged upstream ASAP several months
> ago so I could work on the OMAP clock API without causing major merge
> pain.  Unfortunately, this hasn't been forthcoming, and after waiting a
> reasonable amount of time, I've given up waiting.

Hmm, didn't Paul offer several times to send you clock update patches?

> I had given up all hope of ever seeing the omap-clks1 branch in mainline.
> Really.  Until I saw a comment from you about apparantly wanting to see it
> upstream.  However, I still have a very big question whether any of this
> work will ever reach mainline, especially since my experimental omap-clks1
> branch seems to have been broken up with random bits applied to your tree
> with zero discussion with me.

Paul has spent some time to merge some of your earlier omap-clks1
stuff into linux-omap tree. Manually merging them, there's really no
other way to merge these changes. And Paul posted some of your earlier
omap-clks1 patches to this list along with some other patches before
they got merged.

Anyways, too late to bitch. Merging this stuff is going to be a pain.
And in the long run we need both Russell's and Paul's patches. So let's
just figure out a way how we can get it all merged and into the mainline
tree.

Tony


> > > [*] - these are for the RNG and OHCI drivers, and have been left out
> > > for the time being because they conflict with my master branch, and
> > > would make the omap-clks2 patch unnecessarily difficult for people to
> > > try.
> > > 
> > > Note: please don't think about polluting mach/clkdev.h, thanks.
> > 
> > Huh?
> 
> Don't put anything else in mach/clkdev.h - it's full.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

On Sat, Jan 24, 2009 at 10:32:29AM -0800, Tony Lindgren wrote:
> * Russell King - ARM Linux <linux@arm.linux.org.uk> [090124 09:38]:
> > I had given up all hope of ever seeing the omap-clks1 branch in mainline.
> > Really.  Until I saw a comment from you about apparantly wanting to see it
> > upstream.  However, I still have a very big question whether any of this
> > work will ever reach mainline, especially since my experimental omap-clks1
> > branch seems to have been broken up with random bits applied to your tree
> > with zero discussion with me.
> 
> Paul has spent some time to merge some of your earlier omap-clks1
> stuff into linux-omap tree. Manually merging them, there's really no
> other way to merge these changes. And Paul posted some of your earlier
> omap-clks1 patches to this list along with some other patches before
> they got merged.
> 
> Anyways, too late to bitch. Merging this stuff is going to be a pain.
> And in the long run we need both Russell's and Paul's patches. So let's
> just figure out a way how we can get it all merged and into the mainline
> tree.

If Paul sends me his changes, I'm entirely willing to do the dovetailing
work to get them merged with my changes.  I already do that - what you
see as the changes between clks1 and clks2 were resorted this morning
before sending that email out.  There's interdependencies between each
commit which make such operations non-trivial with traditional git usage.
I have my own scripts which make it possible to move, delete, and merge
a set of commits and patches together.

Therefore, it probably makes sense for me to do it.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Tony Lindgren]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [090124 10:42]:
> On Sat, Jan 24, 2009 at 10:32:29AM -0800, Tony Lindgren wrote:
> > * Russell King - ARM Linux <linux@arm.linux.org.uk> [090124 09:38]:
> > > I had given up all hope of ever seeing the omap-clks1 branch in mainline.
> > > Really.  Until I saw a comment from you about apparantly wanting to see it
> > > upstream.  However, I still have a very big question whether any of this
> > > work will ever reach mainline, especially since my experimental omap-clks1
> > > branch seems to have been broken up with random bits applied to your tree
> > > with zero discussion with me.
> > 
> > Paul has spent some time to merge some of your earlier omap-clks1
> > stuff into linux-omap tree. Manually merging them, there's really no
> > other way to merge these changes. And Paul posted some of your earlier
> > omap-clks1 patches to this list along with some other patches before
> > they got merged.
> > 
> > Anyways, too late to bitch. Merging this stuff is going to be a pain.
> > And in the long run we need both Russell's and Paul's patches. So let's
> > just figure out a way how we can get it all merged and into the mainline
> > tree.
> 
> If Paul sends me his changes, I'm entirely willing to do the dovetailing
> work to get them merged with my changes.  I already do that - what you
> see as the changes between clks1 and clks2 were resorted this morning
> before sending that email out.  There's interdependencies between each
> commit which make such operations non-trivial with traditional git usage.
> I have my own scripts which make it possible to move, delete, and merge
> a set of commits and patches together.
> 
> Therefore, it probably makes sense for me to do it.

Wow, that's the best thing I've ever seen during all the merge hells!

I'll keep the clks-testing branch updated so we all can test it easily,
and then we'll just flip linux-omap to use that branch for the clocks
once everything is merged.

Let me know if I can help with anything else!

Tony

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Paul Walmsley]

On Sat, 24 Jan 2009, Russell King - ARM Linux wrote:

> If Paul sends me his changes, I'm entirely willing to do the dovetailing
> work to get them merged with my changes.

That's great.  I'll send a series to the lists ASAP.


- Paul

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Paul Walmsley]

On Sat, 24 Jan 2009, Russell King - ARM Linux wrote:

> In a bid to clean up the clk API implementations, and more importantly
> ensure that implementations are using the clk API correctly, I've
> converted OMAP over to the generic clock lookup (clkdev) implementation.

I have some comments on this proposal; I will post them after the clock 
update series have been sent to the lists.


- Paul

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

On Sat, Jan 24, 2009 at 11:01:36AM -0800, Tony Lindgren wrote:
> I'll keep the clks-testing branch updated so we all can test it easily,
> and then we'll just flip linux-omap to use that branch for the clocks
> once everything is merged.

Note that I'm going to rebase the two omap-clks branches to be on top of
Linus' tree now that my 'master' was merged.

I would also like to have Paul's patches by the weekend - I'm not asking
for them to already be manipulated to apply on top of my patches, so it
shouldn't take very much time to sort out which patches need to be sent
to me.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Tony Lindgren]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [090128 02:17]:
> On Sat, Jan 24, 2009 at 11:01:36AM -0800, Tony Lindgren wrote:
> > I'll keep the clks-testing branch updated so we all can test it easily,
> > and then we'll just flip linux-omap to use that branch for the clocks
> > once everything is merged.
> 
> Note that I'm going to rebase the two omap-clks branches to be on top of
> Linus' tree now that my 'master' was merged.

OK, I'll refresh the clks-testing branch in linux-omap tree so people
can test it easily.

> I would also like to have Paul's patches by the weekend - I'm not asking
> for them to already be manipulated to apply on top of my patches, so it
> shouldn't take very much time to sort out which patches need to be sent
> to me.

Sounds doable, I believe Paul is pretty close to posting his series.

Regards,

Tony

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Russell King - ARM Linux]

On Wed, Jan 28, 2009 at 07:59:02AM -0800, Tony Lindgren wrote:
> * Russell King - ARM Linux <linux@arm.linux.org.uk> [090128 02:17]:
> > On Sat, Jan 24, 2009 at 11:01:36AM -0800, Tony Lindgren wrote:
> > > I'll keep the clks-testing branch updated so we all can test it easily,
> > > and then we'll just flip linux-omap to use that branch for the clocks
> > > once everything is merged.
> > 
> > Note that I'm going to rebase the two omap-clks branches to be on top of
> > Linus' tree now that my 'master' was merged.
> 
> OK, I'll refresh the clks-testing branch in linux-omap tree so people
> can test it easily.

Note that you only need my omap-clks2 branch.  As I said in the original
message, omap-clks2 is a superset of omap-clks1 - so merging omap-clks2
gets you everything already in omap-clks1.

Re: [ANNOUNCE:RFC] convert OMAP to use clkdev

[Tony Lindgren]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [090128 08:10]:
> On Wed, Jan 28, 2009 at 07:59:02AM -0800, Tony Lindgren wrote:
> > * Russell King - ARM Linux <linux@arm.linux.org.uk> [090128 02:17]:
> > > On Sat, Jan 24, 2009 at 11:01:36AM -0800, Tony Lindgren wrote:
> > > > I'll keep the clks-testing branch updated so we all can test it easily,
> > > > and then we'll just flip linux-omap to use that branch for the clocks
> > > > once everything is merged.
> > > 
> > > Note that I'm going to rebase the two omap-clks branches to be on top of
> > > Linus' tree now that my 'master' was merged.
> > 
> > OK, I'll refresh the clks-testing branch in linux-omap tree so people
> > can test it easily.
> 
> Note that you only need my omap-clks2 branch.  As I said in the original
> message, omap-clks2 is a superset of omap-clks1 - so merging omap-clks2
> gets you everything already in omap-clks1.

OK, the clks-testing in linux-omap tree is updated now. Everybody,
please note that this is a temporary branch that will get rebased often.

So if pulling it fails the next time, just do a hard reset on your local
clks-testing branch with git-reset --hard origin/clks-testing after
doing a git-fetch.

Tony