lkml.org 
[lkml]   [2020]   [Feb]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC PATCH v3 4/6] media: tegra: Add Tegra210 Video input driver
From
Date

On 2/20/20 11:11 AM, Sowjanya Komatineni wrote:
>
> On 2/20/20 5:33 AM, Hans Verkuil wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> (Replying to myself so I can explain this a bit more)
>>
>> On 2/20/20 1:44 PM, Hans Verkuil wrote:
>>>> +
>>>> +static int tegra_csi_tpg_channels_alloc(struct tegra_csi *csi)
>>>> +{
>>>> +    struct device_node *node = csi->dev->of_node;
>>>> +    unsigned int port_num;
>>>> +    int ret;
>>>> +    struct tegra_csi_channel *item;
>>>> +    unsigned int tpg_channels = csi->soc->csi_max_channels;
>>>> +
>>>> +    /* allocate CSI channel for each CSI x2 ports */
>>>> +    for (port_num = 0; port_num < tpg_channels; port_num++) {
>>>> +            item = devm_kzalloc(csi->dev, sizeof(*item), GFP_KERNEL);
>>> Using devm_*alloc can be dangerous. If someone unbinds the driver, then
>>> all memory allocated with devm_ is immediately freed. But if an
>>> application
>>> still has a filehandle open, then when it closes it it might still
>>> reference
>>> this already-freed memory.
>>>
>>> I recommend that you avoid using devm_*alloc for media drivers.
>> A good test is to unbind & bind the driver:
>>
>> cd /sys/devices/platform/50000000.host1x/54080000.vi/driver
>> echo -n 54080000.vi >unbind
>> echo -n 54080000.vi >bind
>>
>> First just do this without the driver being used. That already
>> gives me 'list_del corruption' kernel messages (list debugging
>> is turned on in my kernel).

Will fix in v4 to use kzalloc and also proper release v4l2 to make sure
unbind/bind works properly.

BTW, tegra vi and csi are registered as clients to host1x video driver.

So, unbind and bind should be done with host1x video driver "tegra-video"

cd /sys/devices/platform/50000000.host1x/tegra-video/driver
echo -n tegra-video > unbind
echo -n tegra-video > bind

>>
>> Note that this first test is basically identical to a rmmod/modprobe
>> of the driver. But when I compiled the driver as a module it didn't
>> create any video device nodes! Nor did I see any errors in the kernel
>> log. I didn't pursue this, and perhaps I did something wrong, but it's
>> worth taking a look at.
>>
>> The next step would be to have a video node open with:
>>
>> v4l2-ctl --sleep 10
>>
>> then while it is sleeping unbind the driver and see what happens
>> when v4l2-ctl exits.
>>
>> Worst case is when you are streaming:
>>
>> v4l2-ctl --stream-mmap
>>
>> and then unbind.
>>
>> In general, the best way to get this to work correctly is:
>>
>> 1) don't use devm_*alloc
>> 2) set the release callback of struct v4l2_device and do all freeing
>> there.
>> 3) in the platform remove() callback you call media_device_unregister()
>>     and video_unregister_device().
> Reg 3, in current patch, media_device_unregister is called in
> host1x_video_remove
> video_unregister_device happens during host1x_video_remove ->
> host1x_device_exit -> tegra_vi_exit -> tegra_vi_channels_cleanup
>
>> It's worth getting this right in this early stage, rather than fixing it
>> in the future.
>>
>> Regards,
>>
>>          Hans

\
 
 \ /
  Last update: 2020-02-24 05:46    [W:0.234 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site