diff --git a/device/tegu/NOTES.md b/device/tegu/NOTES.md index 2d9ae83cd..a762361a1 100644 --- a/device/tegu/NOTES.md +++ b/device/tegu/NOTES.md @@ -13,6 +13,56 @@ hash/build ID; this dirty release is not a published kernel package. `[V]` marks verified observations; `[I]` marks an inference still to test. Raw captures, recordings, backups and vendor files remain off repository. +## 2026-11-08: swap on zram + +Kernel `7.3.0-rc6-omamobile-push1-ge8d722982099`, a local build, with the +zram options of `kernel/configs/omamobile.config` (set in the device's +`kernel.config` for the test). They build only modules, so `zram.ko` and +`zsmalloc.ko` went onto the running release with `bin/omamobile +modules-push`, without a new boot image. Omarchy 4.0.6 on the phone. + +- [V] **Before:** no swap (`CONFIG_ZRAM` not set; Omarchy 4.0.6 ships no + zram-generator on aarch64). A `cargo build` of squeekboard on the phone, + next to Brave and the shell, left 177 MB available: memory PSI `full` + about 50 %, load 13, the session stuttering and SSH timing out in the + banner exchange. The OOM killer ended `rustc` (1.2 GB anonymous), and + systemd stopped the app scope it ran in, the terminal with it. +- [V] **Setup**, as a newer Omarchy does it on aarch64 (Omarchy v4.5.0): + `zram-generator` 1.2.1 from Arch, Omarchy's `90-omarchy.conf` (zstd, + `swap-priority = 100`) and the `vm.*` lines of its + `99-omarchy-sysctl.conf` (`swappiness=150`, `vfs_cache_pressure=50`, + `page-cluster=0`, the watermark and dirty settings). zram-generator + loads the module itself. +- [V] **`zram-size = ram / 2` (3.6 GB), 5 GB more allocated** on top of + about 4 GB in use: no OOM kill, PSI `full` peaked at 32 % and fell below + 10 %, and SSH answered throughout. zram held 3.2 GB in 1.7 GB; the test + data compresses about 2:1, ordinary app memory closer to 3:1. +- [V] **`zram-size = ram` (Omarchy's default, 7.3 GB), 6 GB more + allocated:** the phone stopped answering, ping included, and came back + after a restart about six minutes later. The journal has no oops or + panic and no OOM kill for this run; its last lines are the CP's PCIe + link going down and brcmfmac failing to allocate RX buffers. [I] RAM ran + out before zram filled, and reclaim spun instead of killing. Not a + like-for-like comparison (5 GB against 6 GB), so this says only that a + RAM-sized zram is not a safe default here yet. +- [V] **Stock Android on the 9a** (BP3A.250905.014 vendor image): + `vendor.zram.size=50p` selects `fstab.zram.50p` (`zramsize=50%`, plus a + 1 GB `zram_backingdev_size` writeback file for idle pages), started at + `sys.boot_completed` by `init.zumapro.board.rc` with `page-cluster 0`; + no swap partition. +- [V] **After the hang:** `tools/nv-backup -d tegu --check`: every NV + partition the same. `omamobile-tegu-modem-boot` refused to start (its + latch for a boot that didn't shut down cleanly), as designed; the modem + came up ONLINE once the logs were read and the latch removed. +- **By hand, until the image ships it:** build with the options, install + `zram-generator`, put `[zram0]` with `zram-size = ram / 2`, + `compression-algorithm = zstd` and `swap-priority = 100` in + `/etc/systemd/zram-generator.conf`, then `systemctl daemon-reload` and + `systemctl start systemd-zram-setup@zram0.service`. `zramctl` and + `swapon --show` show it. Resize only while it is empty: stopping the unit + removes `/dev/zram0`, and `cat /sys/class/zram-control/hot_add` brings + it back. + ## 2026-11-03: the radio backup check with tools/nv-backup - [V] **NV partitions unchanged (`7.3.0-rc6-omamobile-tegu-s2mpg15-m1-g991042e2eff3-dirty`):** diff --git a/kernel/configs/omamobile.config b/kernel/configs/omamobile.config index f6adc6055..5e9dfd58e 100644 --- a/kernel/configs/omamobile.config +++ b/kernel/configs/omamobile.config @@ -422,3 +422,16 @@ CONFIG_NETFILTER_XT_TARGET_TPROXY=m # Traffic accounting through BPF classifiers (netd's bandwidth controller). CONFIG_NET_CLS_BPF=m CONFIG_NET_ACT_BPF=m + +# Swap on zram. Omarchy's settings set it up on every aarch64 machine +# (zram-generator, default/systemd/zram-generator.conf.d/90-omarchy.conf and +# the vm.* tuning in etc/sysctl.d/99-omarchy-sysctl.conf), and stock Android +# runs one too (half the RAM on the Pixel 9a), but without the driver a phone +# has no swap at all: a Rust build on the 9a (7.4 GB) left 177 MB, stalled +# the session and ended at the OOM killer (device/tegu/NOTES.md, 2026-11-08). +# Modules, loaded by systemd-zram-setup@zram0 when a config asks for a device; +# ZRAM selects ZSMALLOC. zstd is Omarchy's algorithm, lz4 the faster choice. +CONFIG_ZRAM=m +CONFIG_ZRAM_BACKEND_LZ4=y +CONFIG_ZRAM_BACKEND_ZSTD=y +CONFIG_ZRAM_DEF_COMP_ZSTD=y