From acacf3d7ebd26261fbe69db5283440c068262ce2 Mon Sep 17 00:00:00 2001 From: Markus Fritsche Date: Thu, 14 May 2026 15:41:05 +0000 Subject: [PATCH] =?UTF-8?q?iter32=20close:=20kernel=20substrate=20cleanup?= =?UTF-8?q?=20landed=20=E2=86=92=207.0-11=20SHIPPING?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit All iter17/20-31 diagnostic pr_info printks removed from v4l2-ctrls-request.c + rkvdec-hevc.c. 6 base patches retained. 3-codec anchor regression on 7.0-11 post-reboot: - H.264 10F: PASS (sha dd4f5f2d552c) - HEVC 10F: PASS (sha 108f925bb6cb) - VP9 10F: PASS (sha cf35908ae0f9) Hashes identical to 7.0-10 — confirms no kernel-side regression from removing printks. dmesg clean of iter* entries. Memory entry reference_fresnel_kernel_substrate.md updated: substrate now at 7.0-11; Bug 4/Bug 5 marked RESOLVED as libva-side fixes (NOT kernel-side as originally hypothesised). --- phase8_iteration32_close.md | 60 +++++++++++++++++++++++++++++++++++++ 1 file changed, 60 insertions(+) create mode 100644 phase8_iteration32_close.md diff --git a/phase8_iteration32_close.md b/phase8_iteration32_close.md new file mode 100644 index 0000000..60ef774 --- /dev/null +++ b/phase8_iteration32_close.md @@ -0,0 +1,60 @@ +## Iteration 32 — Phase 8 (close): Kernel substrate cleanup → 7.0-11 SHIPPING + +Closes 2026-05-14, third campaign-day milestone after iter31 α-29 fix. Mechanical cleanup — no behavior change. + +### Goal + +Strip the diagnostic `pr_info` printks added during iter17–iter31 root-cause investigation, ship a clean kernel for ongoing fresnel-fourier work. + +### Printks removed + +In `drivers/media/v4l2-core/v4l2-ctrls-request.c`: +- iter21_setup, iter21_setup_ref +- iter22_clone_* +- iter23_loop +- iter24_req_to_new, iter24_try_or_set, iter24_loop_break, iter24_loop_done + +In `drivers/media/platform/rockchip/rkvdec/rkvdec-hevc.c`: +- iter17 (`rkvdec_hevc_run:` printk) +- iter20 (ctrl_hdl + sps/dp byte dump) +- iter27_dpb01, iter31_dpb23 (dpb dumps) +- iter27_sl0, iter31_sl1 (slice_params dumps) + +### Substrate state + +- `linux-fresnel-fourier 7.0-11` built on boltzmann, installed on fresnel. +- 6 load-bearing base patches retained: + - `0001-arm64-dts-rk3399-pinebook-pro-add-OC-OPP-tables-1704-2184.patch` + - `0002-arm64-dts-rk3399-pinebook-pro-enable-hdmi-sound.patch` + - `0003-arm64-dts-rk3399-pinebook-pro-spi1-max-freq-10MHz.patch` + - `0004-media-videobuf2-add-opt-in-dma_resv-producer-fence-h.patch` (RFC v2 #1) + - `0005-media-hantro-attach-dma_resv-release-fence-at-device.patch` (RFC v2 #2) + - `0006-media-rockchip-rga-attach-dma_resv-release-fence-at-.patch` (RFC v2 #3) + +### Verification (post-reboot on 7.0-11) + +3-codec anchor regression: + +| Codec | Status | SHA-12 | +|---|---|---| +| H.264 10F | PASS | dd4f5f2d552c | +| HEVC 10F | PASS | 108f925bb6cb | +| VP9 10F | PASS | cf35908ae0f9 | + +Hashes IDENTICAL to those captured on diagnostic 7.0-10 build — confirms no kernel-side regression from removing the printks (they were pure logging, no behavior touch). + +dmesg `grep -E "rkvdec_iter|iter2[0-7]|iter3[01]|rkvdec_hevc_run:"` post-decode: **empty** (as expected). + +### Substrate-memory updated + +`reference_fresnel_kernel_substrate.md` bumped to 7.0-11, with a new "RESOLVED 2026-05-14" footnote noting that Bug 4 + Bug 5 turned out to be libva-side (not kernel-side) — the libva-vaapi cached-mmap readback path actually works correctly on this kernel; α-25 + α-29 fixes are in libva backend, not kernel. + +### Remaining campaign open items (post iter32) + +1. **MPEG-2 / VP8 multi-device probe** — libva backend's `find_codec_device` picks ONE device for the entire session. Forcing hantro disables H264/HEVC/VP9. Architectural change required; design decision belongs to user. +2. **Backend env-gated diagnostics** (iter29 `LIBVA_HEVC_DUMP_SLICE_TAIL`, iter30 `LIBVA_TS_SCALE`) — no behavior change without env set, retained for future regression debugging. Low priority cleanup. +3. **α-26 cosmetic revert** — `decode_params->short_term_ref_pic_set_size = picture->st_rps_bits` was mis-routed (right value to wrong V4L2 field; rkvdec doesn't use that field). Could revert to 0. Cosmetic. + +### Campaign final state + +After today's arc: Bug 4 fixed (iter25 α-25), Bug 5 fully fixed (iter25 α-25 + iter31 α-29), kernel substrate shipping (iter32 7.0-11). The 5-codec target reduced to 3/5 strict-PASS + 2/5 untestable through libva's single-device probe — pre-existing limitation orthogonal to Bug 4/5.