defaults.md

Shared execution defaults

中文: defaults_ZH.md

These are the execution-tuning defaults every kernel shares. The resolver loads them from execution-tuning.conf and execution-<route>.conf as presets, merges the built-in profile, imported offsets, and user overrides, then hands the result to native. A profile that omits a value gets the value here.

Everything in the table is advisory. Only kernel geometry is required; a profile may omit or override any row. Field-name suffixes encode units: _ms for milliseconds, _us for microseconds. The Impact and Rationale columns explain the trade-offs; change execution only from real-device measurements.

Parent	Child	Default	Purpose	Impact	Rationale
Recommended CPUs	recommended_cpus.main	0	Suggested main race CPU.	Changes scheduling, heat, and race timing; explicit UI selection takes precedence.	Keeps the legacy fallback without topology assumptions.
	recommended_cpus.consumer	1	Suggested consumer CPU.	Same-core placement increases contention; nonexistent cores fail or fall back.	Keeps the legacy separation.
Heap preparation	heap.prepare_max_attempts	4	Full heap-preparation retry cap.	More retries may recover transient misses but add time and heat.	Matches the former four-pass bound.
	heap.prepare_timeout_ms	240000	Overall heap preparation timeout.	Too short misjudges slow devices; too long delays failure feedback.	Four minutes keeps the old bounded worst case.
	heap.kernelsnitch_timeout_ms	60000	KernelSnitch scan timeout.	Affects scan completeness versus stall time.	Matches the prior 60-second scan window.
Shared race timing	race.route_wait_ms	1000	Shared pre-route wait.	Changes readiness ordering.	Keeps the conservative one-second window.
	race.route_done_timeout_ms	300000	Maximum wait for the route worker to publish its terminal status.	Expiry fail-stops the native process because PI ownership cannot be safely unwound.	Five minutes bounds a stuck route without shortening existing TCP attempt windows.
	race.setup_settle_us	50000	Post-setup settle delay.	Too short can trigger before setup completes.	Legacy 50 ms balance.
	race.state_poll_interval_us	1000	Shared state polling interval.	Trades CPU load for reaction latency.	Keeps the prior polling cadence.
W1/W2/W3 stages	stages.w1_attempts	15	W1 write-attempt cap.	More attempts add heat and corruption exposure.	Copied from the original loop bound.
	stages.w1_settle_us	100000	W1 post-attempt settle.	Controls stabilization before verification.	Keeps 100 ms.
	stages.w1_scratch_repair_attempts	3	W1 scratch-repair cap.	More repairs may recover or compound damage.	Matches the former three repairs.
	stages.w2_attempts	15	W2 credential-write cap.	Affects elevation probability, time, and risk.	Matches legacy W2.
	stages.w2_settle_us	100000	W2 settle delay.	Too short may verify before visibility.	Keeps 100 ms.
	stages.w3_chain_rounds	3	Whole W3-chain rounds.	More rounds increase coverage and exposure.	Legacy cap is three.
	stages.w3_attempts	6	W3 attempts per round.	Changes bypass opportunity and contention.	Keeps six attempts.
	stages.w3_settle_us	50000	W3 settle delay.	Controls verification timing and speed.	Keeps 50 ms.
TCP zerocopy route	routes.tcp_zerocopy.attempts	2000	TCP zerocopy probe cap.	Major heat/time contributor; too low reduces hit rate.	Matches the original cap.
	routes.tcp_zerocopy.arm_sequence	16	Arm-sequence length.	Changes spray/trigger ordering; wrong values miss the window.	Keeps the validated 16-step sequence.
	routes.tcp_zerocopy.post_receive_hold_iterations	20000	Post-receive hold iterations.	Longer hold widens window but adds heat.	Copies the former busy-loop constant.
Select-stack route	routes.select_stack.enter_delay_us	50000	Delay before the main select trigger after consumer entry.	Directly shifts the select-stack race window.	Keeps the validated 50 ms timing.
	routes.select_stack.timeout_us	200000	Per-select timeout.	Balances premature exit and recovery time.	Keeps 200 ms.
	routes.select_stack.consumer_max_calls	1	Consumer call cap.	Multiple calls alter the proven layout.	The stable path is single-shot.
	routes.select_stack.consumer_burst_calls	1	Calls per consumer burst.	Changes scheduling and stack lifetime.	Keeps one-call bursts.
Privilege handoff	handoff.pre_dispatch_settle_ms	2000	Pre-handoff settle.	Too short may dispatch before privilege/SELinux state settles.	Keeps two seconds.
	handoff.module_poll_attempts	30	Module-state poll count.	Together with the interval, forms the 3-second total window.	30×100 ms keeps three seconds.
	handoff.module_poll_interval_ms	100	Module poll interval.	Trades wakeups for latency.	Legacy cadence.
	handoff.enforce_poll_attempts	200	Enforcing-state poll count.	Together with the interval, defines the 20-second recovery window.	200×100 ms keeps twenty seconds.
	handoff.enforce_poll_interval_ms	100	Enforcing poll interval.	Affects detection latency and wakeups.	Legacy cadence.