2013/4/22 Tomasz Figa t.figa@samsung.com
On Sunday 21 of April 2013 22:36:08 Inki Dae wrote:
2013/4/21 Tomasz Figa tomasz.figa@gmail.com
Hi,
On Monday 08 of April 2013 16:41:54 Viresh Kumar wrote:
On 8 April 2013 16:37, Vikas Sajjan vikas.sajjan@linaro.org
wrote:
> While migrating to common clock framework (CCF), I found that
the
> FIMD > clocks were pulled down by the CCF. > If CCF finds any clock(s) which has NOT been claimed by any of
the
> drivers, then such clock(s) are PULLed low by CCF. > > Calling clk_prepare() for FIMD clocks fixes the issue. > > This patch also replaces clk_disable() with clk_unprepare()
during
> exit, since clk_prepare() is called in fimd_probe().
I asked you about fixing your commit log too.. It still looks incorrect to me.
This patch doesn't have anything to do with CCF pulling clocks
down,
but calling clk_prepare() before clk_enable() is must now..
that's
it.. nothing more.
I fully agree.
The message should be something like:
Common Clock Framework introduced the need to prepare clocks before enabling them, otherwise clk_enable() fails. This patch adds clk_prepare calls to the driver.
and that's all.
What you are observing as "CCF pulling clocks down" is the fact
that
clk_enable() fails if the clock is not prepared and so the clock is not enabled in result.
Another thing is that CCF is not pulling anything down. GPIO pins
can
be pulled down (or up or not pulled), but clocks can be masked,
gated
or simply disabled - this does not imply their signal level.
> Signed-off-by: Vikas Sajjan vikas.sajjan@linaro.org > --- > > Changes since v3: > - added clk_prepare() in fimd_probe() and
clk_unprepare()
> in > fimd_remove()> > > as suggested by Viresh Kumar <viresh.kumar@linaro.org
> > Changes since v2: > - moved clk_prepare_enable() and
clk_disable_unprepare()
> from > fimd_probe() to fimd_clock() as suggested by Inki Dae > inki.dae@samsung.com> > > Changes since v1: > - added error checking for clk_prepare_enable() and
also
> replaced > clk_disable() with clk_disable_unprepare() during exit. > > --- > > drivers/gpu/drm/exynos/exynos_drm_fimd.c | 14 ++++++++++++-- > 1 file changed, 12 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/exynos/exynos_drm_fimd.c > b/drivers/gpu/drm/exynos/exynos_drm_fimd.c index
9537761..aa22370
> 100644 > --- a/drivers/gpu/drm/exynos/exynos_drm_fimd.c > +++ b/drivers/gpu/drm/exynos/exynos_drm_fimd.c > @@ -934,6 +934,16 @@ static int fimd_probe(struct
platform_device
> *pdev)> > > return ret; > > } > > + ret = clk_prepare(ctx->bus_clk); > + if (ret < 0) > + return ret; > + > + ret = clk_prepare(ctx->lcd_clk); > + if (ret < 0) { > + clk_unprepare(ctx->bus_clk); > + return ret; > + } > +
Why not just simply use clk_prepare_enable() instead of all calls
to
clk_enable() in the driver?
Same goes for s/clk_disable/clk_disable_unprepare/ .
I agree with you. Using clk_prepare_enable() is more clear. Actually
I
had already commented on this. Please see the patch v2. But this way also looks good to me.
Well, both versions are technically correct and will have the same
effect
for Exynos SoC clocks, since only enable/disable ops change hardware state.
However if we look at general meaning of those generic ops, the clock
will
remain prepared for all the time the driver is loaded, even if the
device
Right, so I said previous one is more clear. I gonna revert current one
and then merge previous one(v3)
is runtime suspended. Again on Exynos SoCs this won't have any effect,
but
I think we should respect general Common Clock Framework semantics
anyway.
> ctx->vidcon0 = pdata->vidcon0; > ctx->vidcon1 = pdata->vidcon1; > ctx->default_win = pdata->default_win; > > @@ -981,8 +991,8 @@ static int fimd_remove(struct
platform_device
> *pdev)> > > if (ctx->suspended) > > goto out; > > - clk_disable(ctx->lcd_clk); > - clk_disable(ctx->bus_clk); > + clk_unprepare(ctx->lcd_clk); > + clk_unprepare(ctx->bus_clk);
This looks wrong again.. You still need to call clk_disable() to make clk enabled count zero...
Viresh is right again here.
Ok, you two guys say together this looks wrong so I'd like to take
more
checking. I thought that clk->clk_enable is 1 at here and it would
be 0
by pm_runtimg_put_sync(). Is there any my missing point?
You're reasoning is correct, but only assuming that runtime PM is
enabled.
When it is disabled, pm_runtime_put_sync() is a no-op.
Well, after digging into the exynos_drm_fimd driver a bit more, it
seems
like its power management code needs a serious rework, because I was
able
to find more problems:
- fimd_activate() does not get called at all if CONFIG_PM_RUNTIME is
not
enabled (except in system-wide suspend callbacks, but this is
irrelevant
to this point) - this means that the hardware is not properly
initialized
without CONFIG_PM_RUNTIME - at least clocks does not get enabled.
- pm_runtime_set_suspended() can be used only when runtime PM is
disabled
for the device (i.e. by calling pm_runtime_disable() or not calling pm_runtime_enable() at all) - when runtime PM is enabled it is
basically a
no-op returning -EAGAIN error.
So here's my proposed solution:
- call fimd_activate() and pm_runtime_set_active() explicitly in
fimd_probe(), before calling pm_runtime_enable():
mutex_init(&ctx->lock); platform_set_drvdata(pdev, ctx);
fimd_activate(ctx, true);pm_runtime_set_active(dev); pm_runtime_enable(dev); pm_runtime_get_sync(dev);This would power up the device even if CONFIG_PM_RUNTIME is not
enabled.
Note that pm_runtime_get_sync() after marking the device as active with pm_runtime_set_active() won't result in calling fimd_runtime_resume(), because the device is considered already resumed.
- in fimd_remove():
pm_runtime_disable(dev);if (ctx->suspended)
goto out;
return 0;
clk_disable(ctx->lcd_clk);clk_disable(ctx->bus_clk);
fimd_activate(ctx, false);pm_runtime_put_noidle(dev); pm_runtime_set_suspended(dev);
pm_runtime_put_sync(dev);-out:
pm_runtime_disable(dev);First, pm_runtime_disable() will prevent any further runtime PM
operations
that could change ctx->suspended state. Then, if ctx->suspended is
true,
there is no need to suspend anything and we can leave. Otherwise, we
power
down the hardware manually - which will work with both
CONFIG_PM_RUNTIME
enabled and disabled, and then mark the hardware as suspended and free remaining reference in runtime PM core. Note that pm_runtime_put_noidle just decreases the reference counter and nothing else.
- after those two changes, all that remains is to fix compliance with
Common Clock Framework, in other words:
s/clk_enable/clk_prepare_enable/
and
s/clk_disable/clk_disable_unprepare/
Also looks good to me. But what if power domain was disabled without pm runtime? In this case, you must enable the power domain at machine code
or
bootloader somewhere. This way would not only need some hard codes to
turn
the power domain on but also not manage power management fully. This is
same
as only the use of pm runtime interface(needing some hard codes without
pm
runtime) so I don't prefer to add clk_enable/disable to fimd probe(). I
quite
tend to force only the use of pm runtime as possible. So please add the
hard
codes to machine code or bootloader like you did for power domain if you want to use drm fimd without pm runtime.
That's not how the runtime PM, clock subsystems work:
- When CONFIG_PM_RUNTIME is disabled, all the used hardware must be kept
powered on all the time.
- Common Clock Framework will always gate all clocks that have zero
enable_count. Note that CCF support for Exynos is already merged for 3.10 and it will be the only available clock support method for Exynos.
AFAIK, drivers must work correctly in both cases, with CONFIG_PM_RUNTIME enabled and disabled.
Then is the driver worked correctly if the power domain to this device was disabled at bootloader without CONFIG_PM_RUNTIME and with clk_enable()? I think, in this case, the device wouldn't be worked correctly because the power of the device remains off. So you must enable the power domain somewhere. What is the difference between these two cases?
Best regards, Tomasz