| Message ID | 20240425065011.105915-3-ming.qian@nxp.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2,1/3] media: v4l2-ctrls: Add average qp control | expand |
Hi Ming Qian, W dniu 25.04.2024 o 08:50, Ming Qian pisze: > There is no need to add lock in s_ctrl callback, it has been > synchronized by the ctrl_handler's lock, otherwise it may led to > deadlock if driver call v4l2_ctrl_s_ctrl(). > > Signed-off-by: Ming Qian <ming.qian@nxp.com> > --- > drivers/media/platform/amphion/vdec.c | 2 -- > drivers/media/platform/amphion/venc.c | 2 -- > 2 files changed, 4 deletions(-) > > diff --git a/drivers/media/platform/amphion/vdec.c b/drivers/media/platform/amphion/vdec.c > index a57f9f4f3b87..6a38a0fa0e2d 100644 > --- a/drivers/media/platform/amphion/vdec.c > +++ b/drivers/media/platform/amphion/vdec.c > @@ -195,7 +195,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) > struct vdec_t *vdec = inst->priv; > int ret = 0; > > - vpu_inst_lock(inst); I assume that PATCH v2 2/3 might cause the said deadlock to happen? If so, maybe it would make more sense to make the current patch preceed PATCH v2 2/3? Otherwise the kernel at PATCH v2 2/3 introduces a potential deadlock. Regards, Andrzej > switch (ctrl->id) { > case V4L2_CID_MPEG_VIDEO_DEC_DISPLAY_DELAY_ENABLE: > vdec->params.display_delay_enable = ctrl->val; > @@ -207,7 +206,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) > ret = -EINVAL; > break; > } > - vpu_inst_unlock(inst); > > return ret; > } > diff --git a/drivers/media/platform/amphion/venc.c b/drivers/media/platform/amphion/venc.c > index cdfaba9d107b..351b4edc8742 100644 > --- a/drivers/media/platform/amphion/venc.c > +++ b/drivers/media/platform/amphion/venc.c > @@ -518,7 +518,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > struct venc_t *venc = inst->priv; > int ret = 0; > > - vpu_inst_lock(inst); > switch (ctrl->id) { > case V4L2_CID_MPEG_VIDEO_H264_PROFILE: > venc->params.profile = ctrl->val; > @@ -579,7 +578,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) > ret = -EINVAL; > break; > } > - vpu_inst_unlock(inst); > > return ret; > }
Hi Andrzej, > Hi Ming Qian, > > W dniu 25.04.2024 o 08:50, Ming Qian pisze: >> There is no need to add lock in s_ctrl callback, it has been >> synchronized by the ctrl_handler's lock, otherwise it may led to >> deadlock if driver call v4l2_ctrl_s_ctrl(). >> >> Signed-off-by: Ming Qian <ming.qian@nxp.com> >> --- >> drivers/media/platform/amphion/vdec.c | 2 -- >> drivers/media/platform/amphion/venc.c | 2 -- >> 2 files changed, 4 deletions(-) >> >> diff --git a/drivers/media/platform/amphion/vdec.c >> b/drivers/media/platform/amphion/vdec.c >> index a57f9f4f3b87..6a38a0fa0e2d 100644 >> --- a/drivers/media/platform/amphion/vdec.c >> +++ b/drivers/media/platform/amphion/vdec.c >> @@ -195,7 +195,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >> struct vdec_t *vdec = inst->priv; >> int ret = 0; >> - vpu_inst_lock(inst); > > I assume that PATCH v2 2/3 might cause the said deadlock to happen? > If so, maybe it would make more sense to make the current patch preceed > PATCH v2 2/3? Otherwise the kernel at PATCH v2 2/3 introduces a potential > deadlock. > > Regards, > > Andrzej > I actually discovered this problem when I was preparing the v2 2/3 patch. But in the v2 2/3 patch, it tried to add a read-only ctrl, then I just unset the s_ctrl callback for the new added ctrl, the potential deadlock is caused by call the s_ctrl back in a locked environment, so after unset the s_ctrl callback, the 2/3 patch won't trigger the deadlock even if this patch is missing. In order to avoid encountering similar problems in the future, that driver may set or get some ctrl, I added this patch. Best regards, Ming >> switch (ctrl->id) { >> case V4L2_CID_MPEG_VIDEO_DEC_DISPLAY_DELAY_ENABLE: >> vdec->params.display_delay_enable = ctrl->val; >> @@ -207,7 +206,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >> ret = -EINVAL; >> break; >> } >> - vpu_inst_unlock(inst); >> return ret; >> } >> diff --git a/drivers/media/platform/amphion/venc.c >> b/drivers/media/platform/amphion/venc.c >> index cdfaba9d107b..351b4edc8742 100644 >> --- a/drivers/media/platform/amphion/venc.c >> +++ b/drivers/media/platform/amphion/venc.c >> @@ -518,7 +518,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> struct venc_t *venc = inst->priv; >> int ret = 0; >> - vpu_inst_lock(inst); >> switch (ctrl->id) { >> case V4L2_CID_MPEG_VIDEO_H264_PROFILE: >> venc->params.profile = ctrl->val; >> @@ -579,7 +578,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >> ret = -EINVAL; >> break; >> } >> - vpu_inst_unlock(inst); >> return ret; >> } >
Hi Ming, W dniu 30.04.2024 o 04:32, ming qian pisze: > Hi Andrzej, > >> Hi Ming Qian, >> >> W dniu 25.04.2024 o 08:50, Ming Qian pisze: >>> There is no need to add lock in s_ctrl callback, it has been >>> synchronized by the ctrl_handler's lock, otherwise it may led to >>> deadlock if driver call v4l2_ctrl_s_ctrl(). >>> >>> Signed-off-by: Ming Qian <ming.qian@nxp.com> >>> --- >>> drivers/media/platform/amphion/vdec.c | 2 -- >>> drivers/media/platform/amphion/venc.c | 2 -- >>> 2 files changed, 4 deletions(-) >>> >>> diff --git a/drivers/media/platform/amphion/vdec.c >>> b/drivers/media/platform/amphion/vdec.c >>> index a57f9f4f3b87..6a38a0fa0e2d 100644 >>> --- a/drivers/media/platform/amphion/vdec.c >>> +++ b/drivers/media/platform/amphion/vdec.c >>> @@ -195,7 +195,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >>> struct vdec_t *vdec = inst->priv; >>> int ret = 0; >>> - vpu_inst_lock(inst); >> >> I assume that PATCH v2 2/3 might cause the said deadlock to happen? >> If so, maybe it would make more sense to make the current patch preceed >> PATCH v2 2/3? Otherwise the kernel at PATCH v2 2/3 introduces a potential >> deadlock. >> >> Regards, >> >> Andrzej >> > > I actually discovered this problem when I was preparing the v2 2/3 patch. > > But in the v2 2/3 patch, it tried to add a read-only ctrl, then I just > unset the s_ctrl callback for the new added ctrl, the potential deadlock > is caused by call the s_ctrl back in a locked environment, so after unset > the s_ctrl callback, the 2/3 patch won't trigger the deadlock even if > this patch is missing. > > In order to avoid encountering similar problems in the future, that > driver may set or get some ctrl, I added this patch. Do I understand you correctly that patch 2/3 is written in such a way that it does not introduce a deadlock, and you add patch 3/3 only to prevent future problems? If so, it seems to me that patch 3/3 could/should be separate from this series, as it does not quite match "Add average qp control". Regards, Andrzej > > Best regards, > Ming > >>> switch (ctrl->id) { >>> case V4L2_CID_MPEG_VIDEO_DEC_DISPLAY_DELAY_ENABLE: >>> vdec->params.display_delay_enable = ctrl->val; >>> @@ -207,7 +206,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >>> ret = -EINVAL; >>> break; >>> } >>> - vpu_inst_unlock(inst); >>> return ret; >>> } >>> diff --git a/drivers/media/platform/amphion/venc.c >>> b/drivers/media/platform/amphion/venc.c >>> index cdfaba9d107b..351b4edc8742 100644 >>> --- a/drivers/media/platform/amphion/venc.c >>> +++ b/drivers/media/platform/amphion/venc.c >>> @@ -518,7 +518,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>> struct venc_t *venc = inst->priv; >>> int ret = 0; >>> - vpu_inst_lock(inst); >>> switch (ctrl->id) { >>> case V4L2_CID_MPEG_VIDEO_H264_PROFILE: >>> venc->params.profile = ctrl->val; >>> @@ -579,7 +578,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>> ret = -EINVAL; >>> break; >>> } >>> - vpu_inst_unlock(inst); >>> return ret; >>> } >>
Hi Andrzej, > Hi Ming, > > W dniu 30.04.2024 o 04:32, ming qian pisze: >> Hi Andrzej, >> >>> Hi Ming Qian, >>> >>> W dniu 25.04.2024 o 08:50, Ming Qian pisze: >>>> There is no need to add lock in s_ctrl callback, it has been >>>> synchronized by the ctrl_handler's lock, otherwise it may led to >>>> deadlock if driver call v4l2_ctrl_s_ctrl(). >>>> >>>> Signed-off-by: Ming Qian <ming.qian@nxp.com> >>>> --- >>>> drivers/media/platform/amphion/vdec.c | 2 -- >>>> drivers/media/platform/amphion/venc.c | 2 -- >>>> 2 files changed, 4 deletions(-) >>>> >>>> diff --git a/drivers/media/platform/amphion/vdec.c >>>> b/drivers/media/platform/amphion/vdec.c >>>> index a57f9f4f3b87..6a38a0fa0e2d 100644 >>>> --- a/drivers/media/platform/amphion/vdec.c >>>> +++ b/drivers/media/platform/amphion/vdec.c >>>> @@ -195,7 +195,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>> struct vdec_t *vdec = inst->priv; >>>> int ret = 0; >>>> - vpu_inst_lock(inst); >>> >>> I assume that PATCH v2 2/3 might cause the said deadlock to happen? >>> If so, maybe it would make more sense to make the current patch preceed >>> PATCH v2 2/3? Otherwise the kernel at PATCH v2 2/3 introduces a >>> potential >>> deadlock. >>> >>> Regards, >>> >>> Andrzej >>> >> >> I actually discovered this problem when I was preparing the v2 2/3 patch. >> >> But in the v2 2/3 patch, it tried to add a read-only ctrl, then I just >> unset the s_ctrl callback for the new added ctrl, the potential deadlock >> is caused by call the s_ctrl back in a locked environment, so after unset >> the s_ctrl callback, the 2/3 patch won't trigger the deadlock even if >> this patch is missing. >> >> In order to avoid encountering similar problems in the future, that >> driver may set or get some ctrl, I added this patch. > > Do I understand you correctly that patch 2/3 is written in such a way that > it does not introduce a deadlock, and you add patch 3/3 only to prevent > future > problems? If so, it seems to me that patch 3/3 could/should be separate > from > this series, as it does not quite match "Add average qp control". > > Regards, > > Andrzej > You're right, when I prepare the 2/3 patch, I add the new ctrl just like the previous ctrl, but I got a deadlock. then I tried to fix a deadlock in 2/3 patch first, and make this patch again to avoid similar errors. I put this patch to this series, as I think this better explains why I made this change. Regards, Ming >> >> Best regards, >> Ming >> >>>> switch (ctrl->id) { >>>> case V4L2_CID_MPEG_VIDEO_DEC_DISPLAY_DELAY_ENABLE: >>>> vdec->params.display_delay_enable = ctrl->val; >>>> @@ -207,7 +206,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>> ret = -EINVAL; >>>> break; >>>> } >>>> - vpu_inst_unlock(inst); >>>> return ret; >>>> } >>>> diff --git a/drivers/media/platform/amphion/venc.c >>>> b/drivers/media/platform/amphion/venc.c >>>> index cdfaba9d107b..351b4edc8742 100644 >>>> --- a/drivers/media/platform/amphion/venc.c >>>> +++ b/drivers/media/platform/amphion/venc.c >>>> @@ -518,7 +518,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>> struct venc_t *venc = inst->priv; >>>> int ret = 0; >>>> - vpu_inst_lock(inst); >>>> switch (ctrl->id) { >>>> case V4L2_CID_MPEG_VIDEO_H264_PROFILE: >>>> venc->params.profile = ctrl->val; >>>> @@ -579,7 +578,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) >>>> ret = -EINVAL; >>>> break; >>>> } >>>> - vpu_inst_unlock(inst); >>>> return ret; >>>> } >>> >
diff --git a/drivers/media/platform/amphion/vdec.c b/drivers/media/platform/amphion/vdec.c index a57f9f4f3b87..6a38a0fa0e2d 100644 --- a/drivers/media/platform/amphion/vdec.c +++ b/drivers/media/platform/amphion/vdec.c @@ -195,7 +195,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) struct vdec_t *vdec = inst->priv; int ret = 0; - vpu_inst_lock(inst); switch (ctrl->id) { case V4L2_CID_MPEG_VIDEO_DEC_DISPLAY_DELAY_ENABLE: vdec->params.display_delay_enable = ctrl->val; @@ -207,7 +206,6 @@ static int vdec_op_s_ctrl(struct v4l2_ctrl *ctrl) ret = -EINVAL; break; } - vpu_inst_unlock(inst); return ret; } diff --git a/drivers/media/platform/amphion/venc.c b/drivers/media/platform/amphion/venc.c index cdfaba9d107b..351b4edc8742 100644 --- a/drivers/media/platform/amphion/venc.c +++ b/drivers/media/platform/amphion/venc.c @@ -518,7 +518,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) struct venc_t *venc = inst->priv; int ret = 0; - vpu_inst_lock(inst); switch (ctrl->id) { case V4L2_CID_MPEG_VIDEO_H264_PROFILE: venc->params.profile = ctrl->val; @@ -579,7 +578,6 @@ static int venc_op_s_ctrl(struct v4l2_ctrl *ctrl) ret = -EINVAL; break; } - vpu_inst_unlock(inst); return ret; }
There is no need to add lock in s_ctrl callback, it has been synchronized by the ctrl_handler's lock, otherwise it may led to deadlock if driver call v4l2_ctrl_s_ctrl(). Signed-off-by: Ming Qian <ming.qian@nxp.com> --- drivers/media/platform/amphion/vdec.c | 2 -- drivers/media/platform/amphion/venc.c | 2 -- 2 files changed, 4 deletions(-)