diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 9c0244f26..06d3d8897 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -278,7 +278,7 @@ Findings that matter beyond one subsystem go into `device//NOTES.md`, - `make CHECK_DTBS=y W=1 exynos/google/-.dtb` (or `dt-validate`), with no new warnings, also with your bring-up overlays applied. The plans record these results (for example [HAPTICS-TORCH-PLAN.md](device/cheetah/HAPTICS-TORCH-PLAN.md)). `tools/kernel-pr-check --build OUT --merge BASE` in this repository runs the sign-off, prefix, checkpatch and identifier checks on your branch's commits and the `W=1`, `dtbs_check` and `dt_binding_check` builds on your branch merged onto the current base, comparing device-tree warnings against that base; CI and the review run it the same way. If it says "rebase needed", rebase onto `omamobile`. -- **Shared drivers leave the Pixel 7 Pro as it is.** Most drivers serve several phones: the display engine and DSI, the Synaptics touch controller, Broadcom Wi-Fi, the NFC core, UFS, ACPM, the PMU. The reference phone runs them every day, so put new behaviour behind your SoC's or board's match data (a compatible string, a per-SoC table entry, a quirk flag), and show in the pull request that cheetah's objects build unchanged. A change that alters the Pixel 7 Pro on purpose needs a reason that holds there too (a bug it also has, with the evidence) and merges only after it has run on that phone. +- **Shared drivers leave the Pixel 7 Pro as it is.** Most drivers serve several phones: the display engine and DSI, the Synaptics touch controller, Broadcom Wi-Fi, the NFC core, UFS, ACPM, the PMU. The reference phone runs them every day, so put new behaviour behind your SoC's or board's match data (a compatible string, a per-SoC table entry, a quirk flag), and show in the pull request that cheetah's code is unchanged: identical objects, or, where a flag is tested inside a shared function, a per-function diff showing that only your SoC's path changes. Remember that per-SoC data also reaches every board that binds it through a fallback compatible. A change that alters the Pixel 7 Pro on purpose needs a reason that holds there too (a bug it also has, with the evidence) and merges only after it has run on that phone. - **New hardware is `disabled` by default.** A separate, later commit turns it on once it has passed on the phone. One exception: a board node whose pins, flags and driver are already proven on another phone with the same SoC (the Pixel Fold's volume keys, wired as the Pixel 7 Pro's) may be added enabled in one commit, once it has passed on your phone; say which phone's node it matches. - **Code imported from Google's GPL tree** keeps its copyright and is marked as such. Port it to upstream APIs and style before proposing it for `omamobile`. - **Sending upstream:** the published `omamobile` branch sits on one squashed commit that holds the tree of the upstream tag (see [`tools/publish-kernel`](tools/publish-kernel)). Rebase your commits onto the real tag (`git rebase --onto v7.3-rc6 `), run `scripts/get_maintainer.pl`, and follow `Documentation/process/submitting-patches.rst`. diff --git a/device/husky/UFS-PLAN.md b/device/husky/UFS-PLAN.md index 951d03eb4..d4bdfcf3b 100644 --- a/device/husky/UFS-PLAN.md +++ b/device/husky/UFS-PLAN.md @@ -132,7 +132,7 @@ Let the UFS device runtime-suspend, with the link in hibern8. zuma's CAL has hib - **a guard flag changed:** `reboot-to bootloader` at once (Power + Volume Up if it won't go down, then Power + Volume Down), and run no more cycles. The initramfs guard marks each disk once and never looks at that name again, and this boot never reaches the check before `switch_root`, so a LUN that a host reset dropped and re-added would come back writable. The per-cycle check is the only thing that would see it. - **an error handler run, a reset, or no answer:** if ssh still answers, put `power/control` back to `on`; otherwise use the buttons. - Then add the CAL's hibern8 sequences before trying again. For context: on the Pixel 7 Pro, the first hibern8 exit at `spm_lvl` 3 needed a host reset (AGENTS.md, [SUSPEND-PLAN](../cheetah/SUSPEND-PLAN.md) S1). There, gs101's vendor `suspend` hook asserts the device reset (`HCI_GPIO_OUT`). Tensor G4's host data has no such hook, so that cause doesn't carry over; its `drv_init` sets `spm_lvl` 5, and `UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8` is why `auto_hibern8` is empty (L/`drivers/ufs/host/ufs-exynos.c`) [V]. + Then add the CAL's hibern8 sequences before trying again. For context: on the Pixel 7 Pro, the first hibern8 exit at `spm_lvl` 3 needed a host reset (AGENTS.md, [SUSPEND-PLAN](../cheetah/SUSPEND-PLAN.md) S1). There, gs101's vendor `suspend` hook asserts the device reset (`HCI_GPIO_OUT`). Tensor G4's host data (which husky binds through its fallback compatible) had no such hook until kernel #83 (2026-10-27) added a `suspend` hook that holds the device in reset, as Google's driver does on zuma and zumapro, so that cause now carries over; its `drv_init` sets `spm_lvl` 5, and `UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8` is why `auto_hibern8` is empty (L/`drivers/ufs/host/ufs-exynos.c`) [V]. - **Rollback:** `on` again, or a reboot: nothing persists. - **Risk:** medium. A failed hibern8 exit can leave the link down until the host resets it. From the initramfs, nothing writes to UFS.