6020484453
mm/zone: switch to linear-log bucketing
...
While we're here, also shrink some types.
2026-10-25 00:08:29 -04:00
8958854a28
rtl: add racy_read
2026-10-24 22:20:21 -04:00
0ffaa8df5b
mm/zone: remove all spinlocks
2026-10-24 22:12:56 -04:00
09762e88ae
ke/mutex: add try_acquire
2026-10-24 22:12:44 -04:00
664b621a16
ke/amd64: use movb instead of xchg
2026-10-24 19:01:45 -04:00
b9261c84de
ke: optimize mutex
...
Optimize the mutex code in a few ways:
- Split the fast and slow paths into their own functions, inline the fast path.
- Raise IPL only when acquiring a turnstile lock.
- Avoid taking the turnstile lock if the "waiters" bit is not set.
- Do adaptive spinning only when the lock owner is running.
2026-10-24 16:12:30 -04:00
caaab1e96f
ke/turnstile: rework API
2026-10-24 16:12:18 -04:00
3c5855efc7
ke/sched: don't clear td.cpu
2026-10-24 16:10:51 -04:00
13a7daad4c
ex/string: make string entries smaller
2026-10-23 23:47:35 -04:00
76147556a3
ke: make getting the current thread atomic
2026-10-23 23:47:23 -04:00
0a843157c2
ke: remove extraneous Type suffix
2026-10-23 23:46:50 -04:00
29e678a40b
meta: update zig
2026-10-23 23:46:01 -04:00
6a63714e6c
all: rework source tree structure
2026-10-12 15:02:48 -04:00
2b1b7a7c84
firmware: add FDT parser
2026-10-12 14:46:20 -04:00
a9ecd069a3
rtl: add read_struct
2026-10-12 14:45:40 -04:00
c55beec893
ke,ex: rework kernel queue
...
There was previously a lock inversion in the queue code where we acquired the
queue lock under the thread lock (in the wait_any committing path). This patch
attempts to fix this by instead using an atomic count and deferring the waking
of other threads to a DPC.
2026-10-10 09:37:43 -04:00
ad7f14ecb8
ke: detach from queue on thread exit
2026-10-10 09:37:20 -04:00
923e119756
ke: rework DPC memory ordering
...
swapping with acq_rel here ensures that all previous stores are visible to the
DPC when it's running
2026-10-10 09:23:24 -04:00
fee5d205a9
ke: clean up wait code
2026-10-09 16:55:30 -04:00
cefc69e1b5
all: more tidying
2026-10-08 18:01:42 -04:00
dc39456d6e
tests: oops fix fireworks
2026-10-07 18:08:16 -04:00
619c373d47
all: tidy
2026-10-07 18:03:06 -04:00
6cd9117a05
Remove bad tests
2026-10-07 18:02:17 -04:00
5317cb54b4
ke+loader: rework kernel entry
...
oops, didn't test last commit on debug runs. Now this starts the kernel on its 16
kib stack.
2026-10-05 13:03:16 -04:00
f53a596fb9
loader: boot kernel with a bootstrap stack (oops)
2026-10-05 12:56:36 -04:00
12d0e65fd5
all: more 0.17 migrations
2026-10-04 11:38:11 -04:00
6701f17fb1
ke: clean up wait after continuation
2026-10-04 11:37:52 -04:00
8815e5e52c
ke: clean up wait code
2026-10-04 11:36:37 -04:00
6f07f181db
ke: rework timer cancellation
2026-10-04 11:36:37 -04:00
cbe823ce6e
loader: pass more things to the kernel
2026-10-04 11:36:37 -04:00
72ff4882e8
loader: fix compile error
2026-10-04 11:36:37 -04:00
b802919fc2
ke: add lock warden
2026-10-04 11:36:28 -04:00
dd2161179d
build: update tidy script
2026-10-03 17:08:43 -04:00
10c6c3d27f
all: upgrade to zig 0.17
2026-10-03 17:00:13 -04:00
a49a738d69
riscv64: fix build
2026-09-27 21:41:08 -04:00
7d83278bc6
mm/zone: formatting
2026-09-27 21:36:45 -04:00
26084dbb99
riscv64: move trap code into ke
2026-09-27 21:36:23 -04:00
ec05f050fb
ex: remove extraneous vmem check
2026-09-27 21:32:19 -04:00
marv7000
2c20f0b456
riscv64: initial risc-v port
2026-09-27 21:32:00 -04:00
d03251d3e0
ex: fix vmem bug
2026-09-27 17:40:55 -04:00
672f04a833
add tidy script
2026-09-27 16:21:40 -04:00
c7673dc438
ke: fix shootdown bugs
2026-09-26 13:11:20 -04:00
021e213fb2
ke/turnstile: make turnstiles more finely locked
...
Add a per-turnstile lock which is now held across propagation, instead of a
per-chain lock. This should hopefully
avoid scalability problems if many different turnstiles point to the same
bucket.
2026-09-20 01:18:02 -04:00
8013e84d9f
ke/sched: style fixes
2026-09-20 00:24:44 -04:00
7d871ca75f
ke/sched: fix IPL window when stealing threads.
...
There was a window where a thread could get stolen from a CPU but never
enqueued, since IPL doesn't stay raised the whole time, and the CPU could
therefore get preempted. Keep IPL raised across the whole operation until the
thread is enqueued.
2026-09-20 00:12:54 -04:00
a9028e53ff
ke/smr: allow invalid sequences in deferred_advance_commit
...
This was buggy when the race was lost and it ended up trying to commit a
deferred advance of 0, now explicitly make the magazine's sequence invalid and
do nothing in deferred_advance_commit, avoiding an integer underflow.
2026-09-19 23:03:39 -04:00
e7655e9725
ke: fix scheduler deadlock on switching
2026-09-19 22:19:30 -04:00
472cecea08
ex: move vmem to ex
2026-09-16 16:59:22 -04:00
33ea78a7dd
ke: refine turnstile comments
2026-09-16 16:46:04 -04:00
eeadc1420a
mm/vmem: add importing and rework API
2026-09-13 22:13:18 -04:00
Loïc Grégoire
marv7000