Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <8a60b6a0-9289-0d89-c543-02386a721bca@baylibre.com>
Date:   Fri, 23 Sep 2022 14:32:26 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.2
Subject: Re: [PATCH v8 3/4] counter: ti-ecap-capture: capture driver support
 for ECAP
Content-Language: en-US
From:   Julien Panis <jpanis@baylibre.com>
To:     William Breathitt Gray <william.gray@linaro.org>
Cc:     robh+dt@kernel.org, krzysztof.kozlowski+dt@linaro.org,
        linux-iio@vger.kernel.org, linux-kernel@vger.kernel.org,
        devicetree@vger.kernel.org, mranostay@ti.com
References: <20220922170402.403683-1-jpanis@baylibre.com>
 <20220922170402.403683-4-jpanis@baylibre.com> <Yy0G9a5S3OzwyEwW@fedora>
 <2f3e5036-caab-f892-a4ad-b852f72db331@baylibre.com> <Yy2aAMv5PRjsJ4s2@fedora>
 <923f5876-d692-7e0e-f351-f0e05869ccd8@baylibre.com>
In-Reply-To: <923f5876-d692-7e0e-f351-f0e05869ccd8@baylibre.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 23/09/2022 14:17, Julien Panis wrote:
>
>
> On 23/09/2022 13:35, William Breathitt Gray wrote:
>> On Fri, Sep 23, 2022 at 09:23:26AM +0200, Julien Panis wrote:
>>>
>>> On 23/09/2022 03:08, William Breathitt Gray wrote:
>>>> On Thu, Sep 22, 2022 at 07:04:01PM +0200, Julien Panis wrote:
>>>>> ECAP hardware on TI AM62x SoC supports capture feature. It can be 
>>>>> used
>>>>> to timestamp events (falling/rising edges) detected on input signal.
>>>>>
>>>>> This commit adds capture driver support for ECAP hardware on AM62x 
>>>>> SoC.
>>>>>
>>>>> In the ECAP hardware, capture pin can also be configured to be in
>>>>> PWM mode. Current implementation only supports capture operating 
>>>>> mode.
>>>>> Hardware also supports timebase sync between multiple instances, but
>>>>> this driver supports simple independent capture functionality.
>>>>>
>>>>> Signed-off-by: Julien Panis <jpanis@baylibre.com>
>>>> Hello Julien,
>>>>
>>>> Comments follow inline below.
>>>>
>>>>> +/**
>>>>> + * struct ecap_cnt_dev - device private data structure
>>>>> + * @enabled: device state
>>>>> + * @clk:     device clock
>>>>> + * @regmap:  device register map
>>>>> + * @nb_ovf:  number of overflows since capture start
>>>>> + * @pm_ctx:  device context for PM operations
>>>>> + */
>>>>> +struct ecap_cnt_dev {
>>>>> +    bool enabled;
>>>>> +    struct clk *clk;
>>>>> +    struct regmap *regmap;
>>>>> +    atomic_t nb_ovf;
>>>>> +    struct {
>>>>> +        u8 ev_mode;
>>>>> +        u32 time_cntr;
>>>>> +    } pm_ctx;
>>>>> +};
>>>> Provide documentation for the ev_mode and time_cntr members. You
>>>> probably need a lock as well to protect access to this structure or
>>>> you'll end up with race problems.
>>> Hi William,
>>>
>>> How can I end up with race problems ? pm_ctx members are only 
>>> accessed at
>>> suspend (after capture/IRQ are disabled) and resume (before 
>>> capture/IRQ are
>>> re-enabled).
>>> Is there any risk I did not identify ?
>>>
>>> Julien
>> I was thinking of the ecap_cnt_dev enabled member. The Counter callbacks
>> may execute in concurrent threads, so races can appear when you access
>> members of the ecap_cnt_dev structure in these callbacks.
>>
>> Take for example this section of ecap_cnt_enable_write():
>>
>>          if (enable == ecap_dev->enabled)
>>                  return 0;
>>          if (enable)
>>                  ecap_cnt_capture_enable(counter);
>>          else
>>                  ecap_cnt_capture_disable(counter);
>>          ecap_dev->enabled = enable
>>
>> Suppose two threads try to enable the count capture. A race condition is
>> present where the two threads could see ecap_dev->enabled as false and
>> both proceed to call ecap_cnt_capture_enable(). This results in
>> pm_runtime_get_sync() bumping the usage count twice and we're left with
>> a mismatch the next time ecap_cnt_capture_disable() is called.
>>
>> William Breathitt Gray
>
> OK, If I understand well there's the same problem with IO access with 
> regmap ?
> Julien

[ERRATUM] It seems that some spinlock is already used by regmap API.
So, only the 'enabled' member needs a lock.