Re: [PATCH v9 2/8] media: Add registration helpers for V4L2 flash sub-devices

[Jacek Anaszewski <j.anaszewski@samsung.com>]

Hi Sakari,

On 06/03/2015 10:59 PM, Sakari Ailus wrote:
> Hi Jacek,
>
> On Wed, Jun 03, 2015 at 09:56:39AM +0200, Jacek Anaszewski wrote:
>> Hi Sakari,
>>
>> On 06/02/2015 05:32 PM, Sakari Ailus wrote:
>>> Hi, Jacek!
>>>
>>> On Tue, Jun 02, 2015 at 11:13:54AM +0200, Jacek Anaszewski wrote:
>>>> Hi Sakari,
>>>>
>>>> On 06/01/2015 10:59 PM, Sakari Ailus wrote:
>>>>> Hi Jacek,
>>>>>
>>>>> On Mon, May 25, 2015 at 05:13:57PM +0200, Jacek Anaszewski wrote:
>>>>>> This patch adds helper functions for registering/unregistering
>>>>>> LED Flash class devices as V4L2 sub-devices. The functions should
>>>>>> be called from the LED subsystem device driver. In case the
>>>>>> support for V4L2 Flash sub-devices is disabled in the kernel
>>>>>> config the functions' empty versions will be used.
>>>>>>
>>>>>> Signed-off-by: Jacek Anaszewski <j.anaszewski@samsung.com>
>>>>>> Acked-by: Kyungmin Park <kyungmin.park@samsung.com>
>>>>>> Cc: Sakari Ailus <sakari.ailus@iki.fi>
>>>>>> Cc: Hans Verkuil <hans.verkuil@cisco.com>
>>>>>
>>>>> Thanks for adding indicator support!
>>>>>
>>>>> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
>>>>>
>>>>
>>>> I missed one thing - sysfs interface of the indicator LED class
>>>> also has to be disabled/enabled of v4l2_flash_open/close.
>>>
>>> Good catch.
>>>
>>>>
>>>> I am planning to reimplement the functions as follows,
>>>> please let me know if you see any issues here:
>>>>
>>>> static int v4l2_flash_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh
>>>> *fh)
>>>> {
>>>>          struct v4l2_flash *v4l2_flash = v4l2_subdev_to_v4l2_flash(sd);
>>>>
>>>>          struct led_classdev_flash *fled_cdev = v4l2_flash->fled_cdev;
>>>>
>>>>          struct led_classdev *led_cdev = &fled_cdev->led_cdev;
>>>>          struct led_classdev_flash *iled_cdev = v4l2_flash->iled_cdev;
>>>>
>>>>          struct led_classdev *led_cdev_ind;
>>>>
>>>>          int ret = 0;
>>>
>>> I think you could spare some newlines above (and below as well).
>>
>> There must have been some issue on thunderbird side,
>> originally there were no newlines above.
>
> Ok. I've used Seamonkey, I believe it uses the same editor. Some bugs were
> unintroduced a few years ago, and since that I've mostly used a different
> editor (and MUA) when replying to patches.
>
>>>>
>>>>
>>>>          mutex_lock(&led_cdev->led_access);
>>>>
>>>>
>>>>          if (!v4l2_fh_is_singular(&fh->vfh))
>>>>
>>>>                  goto unlock;
>>>>
>>>>
>>>>          led_sysfs_disable(led_cdev);
>>>>          led_trigger_remove(led_cdev);
>>>>
>>>>
>>>>          if (iled_cdev) {
>>>>                  led_cdev_ind = &iled_cdev->led_cdev;
>>>
>>> You can also declare led_cdev_ind here as you don't need it outside the
>>> basic block.
>>
>> With new approach it will be required also in error path.
>
> True.
>
>>>>
>>>>
>>>>                  mutex_lock(&led_cdev_ind->led_access);
>>>>
>>>>
>>>>                  led_sysfs_disable(led_cdev_ind);
>>>>                  led_trigger_remove(led_cdev_ind);
>>>>
>>>>
>>>>                  mutex_unlock(&led_cdev_ind->led_access);
>>>
>>> Please don't acquire the indicator mutex while holding the flash mutex. I
>>> don't think there's a need to do so, thus creating a dependency between the
>>> two.Just remember to check for v4l2_fh_is_singular() in both cases.
>>
>> I thought that the code would be simpler this way. Nevertheless I
>> produced a new version by following your advise.
>>
>>>
>>>>
>>>>          }
>>>>
>>>>
>>>>          ret = __sync_device_with_v4l2_controls(v4l2_flash);
>>>
>>> If ret is < 0 here, you end up disabling the sysfs interface while open()
>>> fails (and v4l2_flash_close() will not be run). Shouldn't you handle that?
>>
>> Good point.
>>
>> Please find the new version of v4l2_flash{open|close} functions below:
>>
>> static int v4l2_flash_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh
>> *fh)
>> {
>>          struct v4l2_flash *v4l2_flash = v4l2_subdev_to_v4l2_flash(sd);
>>          struct led_classdev_flash *fled_cdev = v4l2_flash->fled_cdev;
>>          struct led_classdev *led_cdev = &fled_cdev->led_cdev;
>>          struct led_classdev_flash *iled_cdev = v4l2_flash->iled_cdev;
>>          struct led_classdev *led_cdev_ind = NULL;
>>          int ret = 0;
>>
>>          mutex_lock(&led_cdev->led_access);
>>
>>          if (!v4l2_fh_is_singular(&fh->vfh)) {
>
> Hmm. I just realised that this might be a bit broken ---
> v4l2_fh_is_singular() should return the same value independently of where in
> subdev's open() handler it is called.
>
> But adding file handles to the list and checking how many there are of them
> is not properly serialised, i.e. another process could change the list in
> between. I'll try to submit a patch for that. No need to wait for that
> though.
>
> So please ignore my earlier request to check for v4l2_fh_is_singular() for
> multiple times, it won't help. Once is enough, albeit more than that won't
> hurt.

Let's agree on the new shape of the functions then:

tatic int v4l2_flash_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh)
{
         struct v4l2_flash *v4l2_flash = v4l2_subdev_to_v4l2_flash(sd);
         struct led_classdev_flash *fled_cdev = v4l2_flash->fled_cdev;
         struct led_classdev *led_cdev = &fled_cdev->led_cdev;
         struct led_classdev_flash *iled_cdev = v4l2_flash->iled_cdev;
         struct led_classdev *led_cdev_ind = NULL;
         int ret = 0;

         if (!v4l2_fh_is_singular(&fh->vfh))
                 return 0;

         mutex_lock(&led_cdev->led_access);

         led_sysfs_disable(led_cdev);
         led_trigger_remove(led_cdev);

         mutex_unlock(&led_cdev->led_access);

         if (iled_cdev) {
                 led_cdev_ind = &iled_cdev->led_cdev;

                 mutex_lock(&led_cdev_ind->led_access);

                 led_sysfs_disable(led_cdev_ind);
                 led_trigger_remove(led_cdev_ind);

                 mutex_unlock(&led_cdev_ind->led_access);
         }

         ret = __sync_device_with_v4l2_controls(v4l2_flash);
         if (ret < 0)
                 goto out_sync_device;

         return 0;
out_sync_device:
         mutex_lock(&led_cdev->led_access);
         led_sysfs_enable(led_cdev);
         mutex_unlock(&led_cdev->led_access);

         if (led_cdev_ind) {
                 mutex_lock(&led_cdev_ind->led_access);
                 led_sysfs_enable(led_cdev_ind);
                 mutex_unlock(&led_cdev_ind->led_access);
         }

         return ret;
}

static int v4l2_flash_close(struct v4l2_subdev *sd, struct 
v4l2_subdev_fh *fh)
{
         struct v4l2_flash *v4l2_flash = v4l2_subdev_to_v4l2_flash(sd);
         struct led_classdev_flash *fled_cdev = v4l2_flash->fled_cdev;
         struct led_classdev *led_cdev = &fled_cdev->led_cdev;
         struct led_classdev_flash *iled_cdev = v4l2_flash->iled_cdev;
         int ret = 0;

         if (!v4l2_fh_is_singular(&fh->vfh))
                 return 0;

         mutex_lock(&led_cdev->led_access);

         if (v4l2_flash->ctrls[STROBE_SOURCE])
                 ret = v4l2_ctrl_s_ctrl(v4l2_flash->ctrls[STROBE_SOURCE],
                                 V4L2_FLASH_STROBE_SOURCE_SOFTWARE);
         led_sysfs_enable(led_cdev);

         mutex_unlock(&led_cdev->led_access);

         if (iled_cdev) {
                 struct led_classdev *led_cdev_ind = &iled_cdev->led_cdev;

                 mutex_lock(&led_cdev_ind->led_access);
                 led_sysfs_enable(led_cdev_ind);
                 mutex_unlock(&led_cdev_ind->led_access);
         }

         return ret;
}


-- 
Best Regards,
Jacek Anaszewski