langsdorff@3.5

Performance & I/O

I/O and virtual-memory tuning is now compiled directly into langsdorffkernel
as build-time defaults, instead of being written at boot through init.s5e3830.rc.
This means the tuning is applied earlier in the boot sequence, consistently, and
without the per-boot sysfs/procfs injection.

Highlights

• Smoother app launches and file loading on the eMMC storage bottleneck.
• Reduced I/O-wait stalls, especially for SQLite-heavy apps (messaging, browser, etc.).
• Less wasted CPU on pointless ZRAM read-ahead and I/O accounting.

Technical details

Storage (mmcblk0)

Setting	Value	Purpose
scheduler	mq-deadline	Low-latency, lightweight scheduler suited to eMMC
read_ahead_kb	2048	Larger sequential read-ahead to hide eMMC latency
nr_requests	128	Deeper queue for better parallelism under load
iostats	0	Disable per-request I/O accounting overhead

ZRAM (zram0)

• read_ahead_kb = 0 — read-ahead is meaningless for in-RAM compressed swap; disabling it avoids wasted decompression/CPU.

FSync

• fsync_enabled = 0 — Kernel FSync disabled to eliminate I/O-wait on fsync() calls (mainly SQLite writes). Trade-off: last few seconds of unflushed data may be lost on a sudden power loss / panic.
• Bugfix: the old injection wrote N to a non-existent path (a silent no-op). The kernel now exposes /sys/module/sync/parameters/fsync_enabled as an int; corrected to 0.

VM (keep dirty pages & caches in RAM longer, avoid eMMC write stalls)

Setting	Value	Purpose
vm.swappiness	30	Keep app pages in RAM, reduce swap I/O
vm.vfs_cache_pressure	30	Retain inode/dentry caches longer
vm.dirty_ratio	80	Allow more dirty pages before forced writeback
vm.dirty_background_ratio	40	Defer background flush, batch writes