Apple's open-source XNU kernel on Android phones: its own loader and drivers, Metal on the Adreno. Runs from RAM; a reboot gives the phone back to Android.
C
docs/air: how much room the log has left, measured rather than assumed
vcvkkc3dd423e2325
diff --git a/docs/air.md b/docs/air.mdindex fae3547..07726e3 100644--- a/docs/air.md+++ b/docs/air.md@@ -679,9 +679,26 @@ Three references, each independent of the translator:line. With `XL_SECS=120` and `REST=1`, iOS wrote more than a megabyte of `os_log` after thesuite had finished, so the suite's own output was in the dropped middle — 27 modules survived,at the front. `XL_FOLLOW`'s intermediate log does not help, because the drop has already- happened by the time it is fetched. The cure is a short window, not a filter: a filter removes- lines that are already gone. `XL_SECS=30` leaves the whole run at 1.19 MB with all 225 modules- in it.+ happened by the time it is fetched, and a filter removes lines that are already gone. Two levers+ keep a run under the cap and neither is sufficient alone — `QUIET=1`, which drops `launchd`'s own+ thousands of lines, and a short `XL_SECS` — because the volume comes from the number of services+ and from the length of the session both. What is actually diagnostic is the size of the file:+ 1 466 58x bytes *is* the cap, so a log that size is truncated however many drop markers it+ carries (the intact run here has two and lost nothing; a truncated one had one). `README.md`+ carries this beside `./xl log`, with the three runs tabulated, because it catches anything+ verbose and not just this suite.++ **How much room is left is worth knowing before the suite grows.** In the run that worked, at+ `XL_SECS=30`, `air_run`'s own output was 93 459 bytes for 225 modules — 415 bytes each — and the+ other 1 092 137 bytes, 92% of the log, was iOS booting *with* `QUIET=1` already on. So the+ headroom is about 280 KB, or roughly 650 more modules if they cost nothing in time, and fewer+ because each one adds 22 ms of further boot noise. `--cases 8` should still fit; past that it+ will not, and the answer then is not a bigger cap but to stop booting iOS for a measurement at+ all: `init` runs programs itself, before `launchd`, where the log is nearly empty and nothing+ can push the output out of it (`RC=<file>`, as `rc.ios` does it). The GPU is up by then, since+ the kernel brings it up, and a question about what the hardware does to a number does not need+ a window server. That is the shape this suite should take if it ever outgrows `--cases 4`; it+ has not yet, and the 225 modules above all arrived.5. **The atomics, with a kernel written to make a lost update show.** This document used to saythat none of Apple's kernels could test an atomic and why: "a sum written by threads that never