Hi Akinobu, let me see if I got this right... On Mon, Apr 30, 2018 at 02:13:11AM +0900, Akinobu Mita wrote: > The set_fmt() in subdev pad ops, the s_ctrl() for subdev control handler, > and the s_frame_interval() in subdev video ops could be called when the > device is under power saving mode. These callbacks for ov772x driver > cause updating H/W registers that will fail under power saving mode. > > This avoids it by not apply any changes to H/W if the device is not powered > up. Instead the changes will be restored right after power-up. > > Cc: Jacopo Mondi > Cc: Laurent Pinchart > Cc: Hans Verkuil > Cc: Sakari Ailus > Cc: Mauro Carvalho Chehab > Signed-off-by: Akinobu Mita > --- > * v4 > - No changes > > drivers/media/i2c/ov772x.c | 79 +++++++++++++++++++++++++++++++++++++--------- > 1 file changed, 64 insertions(+), 15 deletions(-) > > diff --git a/drivers/media/i2c/ov772x.c b/drivers/media/i2c/ov772x.c > index 3e6ca98..bd37169 100644 > --- a/drivers/media/i2c/ov772x.c > +++ b/drivers/media/i2c/ov772x.c > @@ -741,19 +741,30 @@ static int ov772x_s_frame_interval(struct v4l2_subdev *sd, > struct ov772x_priv *priv = to_ov772x(sd); > struct v4l2_fract *tpf = &ival->interval; > unsigned int fps; > - int ret; > + int ret = 0; > > fps = ov772x_select_fps(priv, tpf); > > - ret = ov772x_set_frame_rate(priv, fps, priv->cfmt, priv->win); > - if (ret) > - return ret; > + mutex_lock(&priv->lock); > + /* > + * If the device is not powered up by the host driver do > + * not apply any changes to H/W at this time. Instead > + * the frame rate will be restored right after power-up. > + */ > + if (priv->power_count > 0) { If I'm not wrong this is not protected when the device is streaming. Since Hans' series frame control is called by g/s_parm (until a new ioctl is not introduced) and I've looked at the call stack and it seems to me it does not check for active streaming[1]. I -think- this is even worse when this is called on the subdev, as there is no shared notion of active streaming. Am I wrong? If you have to check for active streaming here (I see some other drivers doing that, eg ov5640), then I see two ways, or you just return -EBUSY, or you save the desired FPS for later, but then it gets messy as you won't be able to just restore paramters at power_on() time, but you would need to do that also at start streaming time. My opinion is that you should check for streaming active (as you're doing for the set_fmt() function in [13/14], do you agree? [1] This calls for a device wise 'streaming' state handler. Maybe it's there but I failed to find checks for that. > + ret = ov772x_set_frame_rate(priv, fps, priv->cfmt, priv->win); > + if (ret) > + goto error; > + } > > tpf->numerator = 1; > tpf->denominator = fps; > priv->fps = fps; > > - return 0; > +error: > + mutex_unlock(&priv->lock); > + > + return ret; > } > > static int ov772x_s_ctrl(struct v4l2_ctrl *ctrl) > @@ -765,6 +776,16 @@ static int ov772x_s_ctrl(struct v4l2_ctrl *ctrl) > int ret = 0; > u8 val; > > + /* v4l2_ctrl_lock() locks our own mutex */ > + > + /* > + * If the device is not powered up by the host driver do > + * not apply any controls to H/W at this time. Instead > + * the controls will be restored right after power-up. > + */ > + if (priv->power_count == 0) > + return 0; > + > switch (ctrl->id) { > case V4L2_CID_VFLIP: > val = ctrl->val ? VFLIP_IMG : 0x00; > @@ -885,6 +906,10 @@ static int ov772x_power_off(struct ov772x_priv *priv) > return 0; > } > > +static int ov772x_set_params(struct ov772x_priv *priv, > + const struct ov772x_color_format *cfmt, > + const struct ov772x_win_size *win); > + > static int ov772x_s_power(struct v4l2_subdev *sd, int on) > { > struct ov772x_priv *priv = to_ov772x(sd); > @@ -895,8 +920,20 @@ static int ov772x_s_power(struct v4l2_subdev *sd, int on) > /* If the power count is modified from 0 to != 0 or from != 0 to 0, > * update the power state. > */ > - if (priv->power_count == !on) > - ret = on ? ov772x_power_on(priv) : ov772x_power_off(priv); > + if (priv->power_count == !on) { > + if (on) { > + ret = ov772x_power_on(priv); > + /* > + * Restore the format, the frame rate, and > + * the controls > + */ > + if (!ret) > + ret = ov772x_set_params(priv, priv->cfmt, > + priv->win); > + } else { > + ret = ov772x_power_off(priv); > + } > + } > > if (!ret) { > /* Update the power count. */ > @@ -1163,7 +1200,7 @@ static int ov772x_set_fmt(struct v4l2_subdev *sd, > struct v4l2_mbus_framefmt *mf = &format->format; > const struct ov772x_color_format *cfmt; > const struct ov772x_win_size *win; > - int ret; > + int ret = 0; > > if (format->pad) > return -EINVAL; > @@ -1184,14 +1221,24 @@ static int ov772x_set_fmt(struct v4l2_subdev *sd, > return 0; > } > > - ret = ov772x_set_params(priv, cfmt, win); > - if (ret < 0) > - return ret; > - > + mutex_lock(&priv->lock); > + /* > + * If the device is not powered up by the host driver do > + * not apply any changes to H/W at this time. Instead > + * the format will be restored right after power-up. > + */ > + if (priv->power_count > 0) { > + ret = ov772x_set_params(priv, cfmt, win); > + if (ret < 0) > + goto error; > + } > priv->win = win; > priv->cfmt = cfmt; > > - return 0; > +error: > + mutex_unlock(&priv->lock); > + > + return ret; > } > > static int ov772x_video_probe(struct ov772x_priv *priv) > @@ -1201,7 +1248,7 @@ static int ov772x_video_probe(struct ov772x_priv *priv) > const char *devname; > int ret; > > - ret = ov772x_s_power(&priv->subdev, 1); > + ret = ov772x_power_on(priv); > if (ret < 0) > return ret; > > @@ -1241,7 +1288,7 @@ static int ov772x_video_probe(struct ov772x_priv *priv) > ret = v4l2_ctrl_handler_setup(&priv->hdl); > > done: > - ov772x_s_power(&priv->subdev, 0); > + ov772x_power_off(priv); > > return ret; > } > @@ -1340,6 +1387,8 @@ static int ov772x_probe(struct i2c_client *client, > > v4l2_i2c_subdev_init(&priv->subdev, client, &ov772x_subdev_ops); > v4l2_ctrl_handler_init(&priv->hdl, 3); > + /* Use our mutex for the controls */ > + priv->hdl.lock = &priv->lock; This is unrelated and should probably go in the previous patch (which I would move before the one where you split s_frame_interval up, for clearness) Thanks j > priv->vflip_ctrl = v4l2_ctrl_new_std(&priv->hdl, &ov772x_ctrl_ops, > V4L2_CID_VFLIP, 0, 1, 1, 0); > priv->hflip_ctrl = v4l2_ctrl_new_std(&priv->hdl, &ov772x_ctrl_ops, > -- > 2.7.4 >