| A venus firmware assert wedges the video GDSC, and the driver’s recovery then retries every 10 ms forever |
finding |
soc:msm8998 |
proven |
| GStreamer’s V4L2 pool copies every venus frame when the sink offers no video meta: 12 fps at 4K, 82 fps otherwise |
finding |
device:google-taimen |
proven |
| Hardware decode does work in Epiphany – the residual ceiling is WebKit’s ~7 processes per tab, not the decoder and not GEM runaway |
finding |
device:google-taimen |
proven |
| There is no hardware video decode on taimen – venus is described in DT but not built |
finding |
device:google-taimen |
proven |
| The “SIGKILL wedges venus until reboot” was VP9 bandwidth starvation misread – venus survives SIGKILL on both codecs |
finding |
device:google-taimen |
proven |
| The msm8998 venus wedge was wrapper clock auto-gating, and one write closes it |
finding |
device:google-taimen |
proven |
| Hardware video decode works on taimen – three more root causes, and no power collapse |
finding |
device:google-taimen |
proven |
| Enabling venus on msm8998 kills the SoC instantly, and the missing bus vote is not why |
finding |
device:google-taimen |
proven |
| msm8998 TZ refuses venus resume with -EINVAL, mainline swallows it, and the whole block stays dark |
finding |
device:google-taimen |
proven |
| A second environment.d file setting the same variable deletes the first one’s value |
trap |
generic |
proven |
| The taimen-v7.2 tree was ten venus patches behind its own aport series, and a venus_core built from it wedges the SoC |
trap |
device:google-taimen |
proven |
