| Message ID | 20250502-qcom-iris-hevc-vp9-v3-13-552158a10a7d@quicinc.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v3,01/23] media: iris: Skip destroying internal buffer if not dequeued | expand |
On 01/05/2025 20:13, Dikshita Agarwal wrote: > Firmware sends buffers with 0 filled length which needs to be dropped, > to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. > Also make sure: > - These 0 length buffers are not returned as result of flush. > - Its not a buffer with LAST flag enabled which will also have 0 filled > length. > > Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> > Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> > --- > drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > index 4488540d1d41..3bb326843a7b 100644 > --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, > > buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); > > + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && > + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && > + !(inst->sub_state & IRIS_INST_SUB_DRC)) { > + buf->flags |= V4L2_BUF_FLAG_ERROR; > + } > + > return 0; > } > > This is a pretty complex conjunctive clause. Is it possible for say 1/3 of these logical criteria to be false ? i.e. if you get to: 1. buf->data_size && inst->state == IRIS_INST_STREAMING 2. !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) is it possible to get to 3. !(inst->sub_state & IRIS_INST_SUB_DRC) ? This also feels like a bugfix ? --- bod
On 5/2/2025 6:19 PM, Bryan O'Donoghue wrote: > On 01/05/2025 20:13, Dikshita Agarwal wrote: >> Firmware sends buffers with 0 filled length which needs to be dropped, >> to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. >> Also make sure: >> - These 0 length buffers are not returned as result of flush. >> - Its not a buffer with LAST flag enabled which will also have 0 filled >> length. >> >> Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> >> Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> >> --- >> drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ >> 1 file changed, 6 insertions(+) >> >> diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> index 4488540d1d41..3bb326843a7b 100644 >> --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct >> iris_inst *inst, >> buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, >> hfi_buffer->flags); >> + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && >> + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && >> + !(inst->sub_state & IRIS_INST_SUB_DRC)) { >> + buf->flags |= V4L2_BUF_FLAG_ERROR; >> + } >> + >> return 0; >> } >> > > This is a pretty complex conjunctive clause. > > Is it possible for say 1/3 of these logical criteria to be false ? > > i.e. if you get to: > > 1. buf->data_size && inst->state == IRIS_INST_STREAMING This makes sure that the 0 length buffers are not returned as response to flush. > 2. !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) This makes sure that we don't mark the buffer with last flag as error since this will be also be a zero length buffer but a valid one. > > is it possible to get to > > 3. !(inst->sub_state & IRIS_INST_SUB_DRC) I can remove this condition as first one captures the essence of this one. Thanks, Dikshita > > ? > > This also feels like a bugfix ? > > --- > bod
Hi Dikshita, Le vendredi 02 mai 2025 à 00:43 +0530, Dikshita Agarwal a écrit : > Firmware sends buffers with 0 filled length which needs to be dropped, > to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. > Also make sure: > - These 0 length buffers are not returned as result of flush. > - Its not a buffer with LAST flag enabled which will also have 0 filled > length. This message is quite vague, is this about capture or output buffers ? If its output buffers that don't produce capture, I don't see why they have to be flagged as errors, or why the payload size matter. Then, if its about assigned capture buffers that did not get used in the end, you should put them back in the queue instead of returning them to user space. Returning a capture buffers to userspace should only be used if a frame could not be produced. That imply copying the cookie timestamp from the src buffers into the capture buffer. Please make sure you don't endup returning fake erorrs to userspace, which may lead to some frame metadata being dropped erroneously. Nicolas > > Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> > Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> > --- > drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > index 4488540d1d41..3bb326843a7b 100644 > --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, > > buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); > > + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && > + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && > + !(inst->sub_state & IRIS_INST_SUB_DRC)) { > + buf->flags |= V4L2_BUF_FLAG_ERROR; > + } > + > return 0; > } >
On 5/3/2025 9:39 PM, Nicolas Dufresne wrote: > Hi Dikshita, > > Le vendredi 02 mai 2025 à 00:43 +0530, Dikshita Agarwal a écrit : >> Firmware sends buffers with 0 filled length which needs to be dropped, >> to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. >> Also make sure: >> - These 0 length buffers are not returned as result of flush. >> - Its not a buffer with LAST flag enabled which will also have 0 filled >> length. > > This message is quite vague, is this about capture or output buffers ? > If its output buffers that don't produce capture, I don't see why they > have to be flagged as errors, or why the payload size matter. Then, if > its about assigned capture buffers that did not get used in the end, you > should put them back in the queue instead of returning them to user > space. > > Returning a capture buffers to userspace should only be used if a frame > could not be produced. That imply copying the cookie timestamp from the > src buffers into the capture buffer. Please make sure you don't endup > returning fake erorrs to userspace, which may lead to some frame > metadata being dropped erroneously. > The capture buffers which I am trying to handle here are of 0 byteused which means they don't have any valid data and they have the timestamp copied from src buffers. How these buffers will be handled by client? if we don't associate error flag to such buffers? Thanks, Dikshita > Nicolas > >> >> Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> >> Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> >> --- >> drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ >> 1 file changed, 6 insertions(+) >> >> diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> index 4488540d1d41..3bb326843a7b 100644 >> --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >> @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, >> >> buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); >> >> + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && >> + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && >> + !(inst->sub_state & IRIS_INST_SUB_DRC)) { >> + buf->flags |= V4L2_BUF_FLAG_ERROR; >> + } >> + >> return 0; >> } >>
Hi Dikshita, Le dimanche 04 mai 2025 à 20:53 +0530, Dikshita Agarwal a écrit : > > > On 5/3/2025 9:39 PM, Nicolas Dufresne wrote: > > Hi Dikshita, > > > > Le vendredi 02 mai 2025 à 00:43 +0530, Dikshita Agarwal a écrit : > > > Firmware sends buffers with 0 filled length which needs to be dropped, > > > to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. > > > Also make sure: > > > - These 0 length buffers are not returned as result of flush. > > > - Its not a buffer with LAST flag enabled which will also have 0 filled > > > length. > > > > This message is quite vague, is this about capture or output buffers ? > > If its output buffers that don't produce capture, I don't see why they > > have to be flagged as errors, or why the payload size matter. Then, if > > its about assigned capture buffers that did not get used in the end, you > > should put them back in the queue instead of returning them to user > > space. > > > > Returning a capture buffers to userspace should only be used if a frame > > could not be produced. That imply copying the cookie timestamp from the > > src buffers into the capture buffer. Please make sure you don't endup > > returning fake erorrs to userspace, which may lead to some frame > > metadata being dropped erroneously. > > > The capture buffers which I am trying to handle here are of 0 byteused > which means they don't have any valid data and they have the timestamp > copied from src buffers. > How these buffers will be handled by client? if we don't associate error > flag to such buffers? Please share a link to the stream and specify which frames are handled this way by your firmware. The answer to your question is entirely dependent on the stream you are decoding. Nicolas > > Thanks, > Dikshita > > Nicolas > > > > > > > > Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> > > > Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> > > > --- > > > drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ > > > 1 file changed, 6 insertions(+) > > > > > > diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > > > b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > > > index 4488540d1d41..3bb326843a7b 100644 > > > --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > > > +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c > > > @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, > > > > > > buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); > > > > > > + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && > > > + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && > > > + !(inst->sub_state & IRIS_INST_SUB_DRC)) { > > > + buf->flags |= V4L2_BUF_FLAG_ERROR; > > > + } > > > + > > > return 0; > > > } > > >
On 5/5/2025 6:07 PM, Nicolas Dufresne wrote: > Hi Dikshita, > > Le dimanche 04 mai 2025 à 20:53 +0530, Dikshita Agarwal a écrit : >> >> >> On 5/3/2025 9:39 PM, Nicolas Dufresne wrote: >>> Hi Dikshita, >>> >>> Le vendredi 02 mai 2025 à 00:43 +0530, Dikshita Agarwal a écrit : >>>> Firmware sends buffers with 0 filled length which needs to be dropped, >>>> to achieve the same, add V4L2_BUF_FLAG_ERROR to such buffers. >>>> Also make sure: >>>> - These 0 length buffers are not returned as result of flush. >>>> - Its not a buffer with LAST flag enabled which will also have 0 filled >>>> length. >>> >>> This message is quite vague, is this about capture or output buffers ? >>> If its output buffers that don't produce capture, I don't see why they >>> have to be flagged as errors, or why the payload size matter. Then, if >>> its about assigned capture buffers that did not get used in the end, you >>> should put them back in the queue instead of returning them to user >>> space. >>> >>> Returning a capture buffers to userspace should only be used if a frame >>> could not be produced. That imply copying the cookie timestamp from the >>> src buffers into the capture buffer. Please make sure you don't endup >>> returning fake erorrs to userspace, which may lead to some frame >>> metadata being dropped erroneously. >>> >> The capture buffers which I am trying to handle here are of 0 byteused >> which means they don't have any valid data and they have the timestamp >> copied from src buffers. >> How these buffers will be handled by client? if we don't associate error >> flag to such buffers? > > Please share a link to the stream and specify which frames are handled > this way by your firmware. The answer to your question is entirely > dependent on the stream you are decoding. > one such stream is JCT-VC-HEVC_V1 -tv NoOutPrior_A_Qualcomm_1 (fluster stream) for this stream, POC 17, 22 and 24 are not supposed to be displayed (output is false), the corresponding capture buffers are returned with 0 filled length by firmware to driver and expected to be dropped by the client. Thanks, Dikshita > Nicolas > >> >> Thanks, >> Dikshita >>> Nicolas >>> >>>> >>>> Acked-by: Vikash Garodia <quic_vgarodia@quicinc.com> >>>> Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com> >>>> --- >>>> drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c | 6 ++++++ >>>> 1 file changed, 6 insertions(+) >>>> >>>> diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >>>> b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >>>> index 4488540d1d41..3bb326843a7b 100644 >>>> --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >>>> +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c >>>> @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, >>>> >>>> buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); >>>> >>>> + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && >>>> + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && >>>> + !(inst->sub_state & IRIS_INST_SUB_DRC)) { >>>> + buf->flags |= V4L2_BUF_FLAG_ERROR; >>>> + } >>>> + >>>> return 0; >>>> } >>>>
diff --git a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c index 4488540d1d41..3bb326843a7b 100644 --- a/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c +++ b/drivers/media/platform/qcom/iris/iris_hfi_gen2_response.c @@ -378,6 +378,12 @@ static int iris_hfi_gen2_handle_output_buffer(struct iris_inst *inst, buf->flags = iris_hfi_gen2_get_driver_buffer_flags(inst, hfi_buffer->flags); + if (!buf->data_size && inst->state == IRIS_INST_STREAMING && + !(hfi_buffer->flags & HFI_BUF_FW_FLAG_LAST) && + !(inst->sub_state & IRIS_INST_SUB_DRC)) { + buf->flags |= V4L2_BUF_FLAG_ERROR; + } + return 0; }