PERFORCE change 63836 for review
John Baldwin
jhb at FreeBSD.org
Wed Oct 27 10:47:52 PDT 2004
http://perforce.freebsd.org/chv.cgi?CH=63836
Change 63836 by jhb at jhb_slimer on 2004/10/27 17:47:26
Update notes.
Affected files ...
.. //depot/projects/smpng/sys/notes#22 edit
.. //depot/user/jhb/needresched/notes#7 edit
Differences ...
==== //depot/projects/smpng/sys/notes#22 (text+ko) ====
@@ -53,5 +53,16 @@
go figure)
- Optimize spin locks on UP such that they don't do atomic operations
- Untested
+ - Unbenchmarked
+- Add support for disabling CPUs on x86 via APIC ID (hint.lapic.X.disabled)
+ - need to separate mp_ncpus a bit, maybe mp_ncpus stays as count of
+ present CPUs, but a separate mp_activecpus is count of CPUs actually in
+ use. Hmm, CPUs really should just not show up when they are disabled
+ maybe, but that means fudging with mp_ncpus.
+- Fix boot_cpu_id to be a logical FreeBSD cpuid that is the BSP on all archs
+ - Maybe hardcode FreeBSD cpuid 0 as BSP instead
+- Change alpha to use logical FreeBSD cpuid's rather than mapping 1:1 to
+ physical IDs
+- Revert bde's changes to the interrupt storm code.
Space reserved for child branches:
==== //depot/user/jhb/needresched/notes#7 (text+ko) ====
@@ -74,3 +74,7 @@
- ULE
- Move the postsig() loop in ast() into userret(), possibly conditional
on needresched? If so, it goes into sched_userret().
+- Need lots of thought on realtime and idle time priorities and handling
+ tsleep() priority "boosts" for such processes. Might need to make
+ the priority boost be something passsed to sched_sleep() rather than
+ a sched_prio() call.
More information about the p4-projects
mailing list