lkml.org 
[lkml]   [2012]   [Jan]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 3/6] clk: introduce the common clock framework
On 01/13/2012 08:39 PM, Turquette, Mike wrote:
> On Fri, Jan 13, 2012 at 8:18 PM, Saravana Kannan<skannan@codeaurora.org> wrote:
>> On 12/17/2011 03:04 AM, Russell King - ARM Linux wrote:
>>>
>>> On Fri, Dec 16, 2011 at 04:45:48PM -0800, Turquette, Mike wrote:
>>>>
>>>> On Wed, Dec 14, 2011 at 5:18 AM, Thomas Gleixner<tglx@linutronix.de>
>>>> wrote:
>>>>>
>>>>> On Tue, 13 Dec 2011, Mike Turquette wrote:
>>>>>>
>>>>>> +void __clk_unprepare(struct clk *clk)
>>>>>> +{
>>>>>> + if (!clk)
>>>>>> + return;
>>>>>> +
>>>>>> + if (WARN_ON(clk->prepare_count == 0))
>>>>>> + return;
>>>>>> +
>>>>>> + if (--clk->prepare_count> 0)
>>>>>> + return;
>>>>>> +
>>>>>> + WARN_ON(clk->enable_count> 0);
>>>>>
>>>>>
>>>>> So this leaves the clock enable count set. I'm a bit wary about
>>>>> that. Shouldn't it either return (including bumping the prepare_count
>>>>> again) or call clk_disable() ?
>>>
>>>
>>> No it should not.
>>>
>>>> I've hit this in my port of OMAP. It comes from this simple situation:
>>>>
>>>> driver 1 (adapted for clk_prepare/clk_unprepare):
>>>> clk_prepare(clk);
>>>> clk_enable(clk);
>>>>
>>>> ...
>>>>
>>>> driver2 (not adapted for clk_prepare/clk_unprepare):
>>>> clk_enable(clk);
>>>
>>>
>>> So this is basically buggy. Look, it's quite simple. Convert _all_
>>> your drivers to clk_prepare/clk_unprepare _before_ you start switching
>>> your platform to use these new functions. You can do that _today_
>>> without exception.
>>>
>>> We must refuse to merge _any_ user which does this the old way - and
>>> we should have been doing this since my commit was merged into mainline
>>> to allow drivers to be converted.
>>>
>>> And stop trying to think of ways around this inside clk_prepare/
>>> clk_unprepare/clk_enable/clk_disable. You can't do it. Just fix _all_
>>> the drivers. Now. Before you start implementing
>>> clk_prepare/clk_unprepare.
>>
>>
>> I agree with Russell's suggestion. This is what I'm trying to do with the
>> MSM platform. Not sure if I'm too optimistic, but as of today, I'm still
>> optimistic I can push the MSM driver devs to get this done before we enable
>> real prepare/unprepare support.
>
> Just to reach closure on this topic: I don't plan to change
> __clk_unprepare in the next version of the patches. The warnings are
> doing a fine job of catching code which has yet to be properly
> converted to use clk_(un)prepare.

To be fair, I also have to improve the stub clk_prepare/unprepare to
maintain ref counts and do refcount checking before I plan to cut off to
the real prepare/unprepare implementations. So, I'm guessing Mike is
just trying to partly add that support in this patch series.

My goal is to have MSM converted fully before switching to this. So,
this code that we are debating about won't directly impact MSM. For that
reason, I won't be trying to hold off the more important common clock
framework due to unconventional error handling.

Thanks,
Saravana

--
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.


\
 
 \ /
  Last update: 2012-01-14 05:55    [W:0.019 / U:0.112 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site