[PATCH] clk: Fix race conditions between clk_set_parent() and clk_enable()

[skannan@codeaurora.org (Saravana Kannan)]

On 05/23/2012 02:16 AM, Peter De Schrijver wrote:
> On Tue, May 22, 2012 at 08:06:45PM +0200, Turquette, Mike wrote:
>> On Tue, May 22, 2012 at 6:58 AM, Peter De Schrijver
>> <pdeschrijver@nvidia.com>  wrote:
>>> On Tue, May 15, 2012 at 08:20:44PM +0200, Saravana Kannan wrote:
>>>> On 05/11/2012 09:59 PM, Saravana Kannan wrote:
>>>>> Without this patch, the following race conditions are possible.
>>>>>
>>>>> Race condition 1:
>>>>> * clk-A has two parents - clk-X and clk-Y.
>>>>> * All three are disabled and clk-X is current parent.
>>>>> * Thread A: clk_set_parent(clk-A, clk-Y).
>>>>> * Thread A:<snip execution flow>
>>>>> * Thread A: Grabs enable lock.
>>>>> * Thread A: Sees enable count of clk-A is 0, so doesn't enable clk-Y.
>>>>> * Thread A: Releases enable lock.
>>>>> * Thread B: Calls clk_enable(clk-A), which in turn enables clk-X.
>>>>> * Thread A: Switches clk-A's parent to clk-Y in hardware.
>>>>>
>>>>> clk-A is now enabled in software, but not clocking in hardware.
>>>>>
>>>>> Race condition 2:
>>>>> * clk-A has two parents - clk-X and clk-Y.
>>>>> * All three are disabled and clk-X is current parent.
>>>>> * Thread A: clk_set_parent(clk-A, clk-Y).
>>>>> * Thread A:<snip execution flow>
>>>>> * Thread A: Switches parent in hardware to clk-Y.
>>>>> * Thread A: Grabs enable lock.
>>>>> * Thread A: Sees enable count of clk-A is 0, so doesn't disable clk-X.
>>>>> * Thread A: Releases enable lock.
>>>>> * Thread B: Calls clk_enable(clk-A)
>>>>> * Thread B: Software state still says parent is clk-X.
>>>>> * Thread B: So, enables clk-X and then itself.
>>>>> * Thread A: Updates parent in software state to clk-Y.
>>>>>
>>>
>>> This looks correct to me. Is there any usecase where enabling/disabling a
>>> clock would require sleeping but changing the parent would not?
>>>
>>
>> clk_enable&  clk_disable must never sleep.  clk_prepare and
>> clk_unprepare may sleep.
>>
>
> In that case the clock is actually enabled in clk_prepare and disabled in
> clk_unprepare I guess (and clk_enable/clk_disable are dummy functions)?
>
> What I'm trying to say is that I don't think there are clocks which can be
> enabled/disabled using non blocking operations, but where a parent change
> would require a blocking operation.
>
> Cheers,
>
> Peter.

Mark, Shawn, Russell,

Can you guys please respond? I'm surprised that no one seem to care 
about fixing race conditions between clk_set_parent/clk_set_rate() and 
clk_enable() that will result in incorrect enable count propagation and 
have the SW get out of sync with HW.

If we absolutely need to support clocks that where the ops->set_parent() 
is not atomic and can't go with the CLK_SET_PARENT_GATE option, then 
maybe we can add a "I promise the consumers of this clock won't call 
clk_set_parent() and clk_enable() in a racy way" clock flag 
(CLK_IGNORE_PARENT_ENABLE_RACE). Yes, it would be a hack for such 
clocks, but that's still better than leaving a gaping hole for all the 
clocks.

-Saravana

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