[PATCH v2 0/2] media: camss: CSID power-on unwind and VFE stream handling

[Qianhui Xu]

This is version 2 of the three camss fixes I sent on 29 October.  Two
things changed.

The DMA-contiguity check and its prepare_buffer hook are gone.  Bryan
measured that plane0 of a normal buffer has five physical breaks and no
IOVA breaks, and the pipeline still works; that matches how dma-iommu
merges the segments when the table is mapped.  The check tested
sg_dma_address() (IOVA) while the commit message claimed physical
contiguity, so the commit was wrong even before the measurement, and the
prepare_buffer hook had no implementer or caller anywhere in the tree.
I dropped the whole patch rather than rewording it.

The streaming fix is reshaped after Loic's review: the start and stop
walks now share camss_video_subdevs_set_stream(), the source pad is taken
from media_entity_remote_source_pad_unique() instead of pads[0], and the
subdevices are toggled with v4l2_subdev_enable_streams() and
v4l2_subdev_disable_streams().  The camss subdevices only implement
.s_stream(), so the helpers fall back to it; that is the only reason this
still works without changing the subdevice drivers.

The scope of this series is fixes only.  The SM8750 camss enablement is
not part of it; that work is with Hangxiang.  Both patches apply on
v7.3-rc7 and the camss module builds clean.

Changes since the 29 October posting:
 - drop the DMA-contiguity patch and its prepare hook (rejected on plane0
   evidence; it also never had a user)
 - factor the stream walk into one helper, use the remote source pad and
   the v4l2_subdev_*_streams() helpers
 - keep the csid power-on unwind unchanged


Qianhui Xu (2):
  media: camss: csid: unwind the power-on path through a single error
    label
  media: camss: video: factor pipeline stream on/off into a helper

 .../media/platform/qcom/camss/camss-csid.c    |  45 ++++----
 .../media/platform/qcom/camss/camss-video.c   | 109 +++++++++++-------
 2 files changed, 85 insertions(+), 69 deletions(-)

-- 
2.53.0