PERFORCE change 45020 for review
John Baldwin
jhb at FreeBSD.org
Fri Jan 9 11:01:54 PST 2004
http://perforce.freebsd.org/chv.cgi?CH=45020
Change 45020 by jhb at jhb_slimer on 2004/01/09 11:01:31
Update todo lists.
Affected files ...
.. //depot/projects/smpng/sys/notes#3 edit
.. //depot/user/jhb/acpipci/notes#78 edit
Differences ...
==== //depot/projects/smpng/sys/notes#3 (text+ko) ====
@@ -36,14 +36,11 @@
- Remove sched_lock from lockmgr TDF_DEADLKTREAT flag check. Perhaps we
can move this flag to TDP_DEADLKTREAT even.
- Mostly untested
-- Allow mtx_trylock() to recurse on a recursive mutex.
- - Needs strict testing
- Use panic_thread pointer instead of panic_cpu to allow for migration
of the panic'ing thread
- Untested
-+ Move schedcpu() into its own kthread.
- + Committed
-- Move loadav() into the schedcpu kthread.
+- Move loadav() into its own thread.
+ - Untested
- Remove some bogus atomic_load_acq()'s and add ia32_pause()'s to
stop_cpus() and restart_cpus().
- Untested
==== //depot/user/jhb/acpipci/notes#78 (text+ko) ====
@@ -5,12 +5,14 @@
+ Allow loader tunables to override the IRQ for each link device.
+ Test routing of an unrouted link.
- Trim includes in sys/i386/pci/pci_*.
+ - Only parse the PIR and create link devices if a PCIBIOS host-pci bridge
+ attaches.
- Export PA's of APICs for vmware
+ Handle MADT and IRQ 9 special case
- This is gross, we might want a redirection layer in nexus(4)
- - Linux does this. We just need to read the SCI int out of the FADT
+ + Linux does this. We just need to read the SCI int out of the FADT
instead of hard-coding 9.
- - Use level/low if SCI routed to IRQ 9 as active/hi or if routed to
+ + Use level/low if SCI routed to IRQ 9 as active/hi or if routed to
PCI interrupt
+ Add dev/ic/i8259A.h
+ Make acpi_cpu_c1() MD.
@@ -21,9 +23,11 @@
- Allow SCI to be routed via mixed-mode?
- Add a MI interrupt source abstraction and move handlers out of ithd and
into that
+ - Have suspend/resume operate on pics, not individual sources
- Defer creation of ithreads until a non-fast handler is added to an
interrupt source
- Kill isa_irq_pending() (maybe, is this an MI function?)
+ - Lazily allocate PCI lapic interrupt vectors.
- DDB
- Add 'show ithread'
+ Add 'show irqs'
@@ -42,8 +46,8 @@
+ Implement mixed-mode and use it for IRQ0 unless no ExtINT pin or
NO_MIXED_MODE
+ Add NO_MIXED_MODE option
+ - Make NO_MIXED_MODE a loader tunable as well. (hw.apic.mixed_mode)
- Add runtime decision for IRQ0 mixed mode?
- - Don't use mixed mode with MADT?
- Add handling of local APIC ERROR lvt.
- Use the local apic timer to drive hard/stat/profclock (maybe too hard)
- Do we need to be able to renumber I/O APICs?
@@ -52,7 +56,6 @@
- Renumber I/O APIC IDs. Ugh. Not sure if this will have to be done for
the general case or just for the mptable case. Update: may not have to,
it was a BIOS bug. May have to bump NAPICID to 256 though.
-- Flesh out cpu topology map
- Suspend/resume support for I/O APICs
- reimplement lazy masking of interrupts for critical sections w/o bitmasks??
- Support bus_config_intr()?
@@ -65,3 +68,6 @@
Unrelated to this branch:
- Work on cleaning up nested includes in sys/* headers. According to Bruce,
the only nested sys/* headers should be _*.h and queue.h.
+- Add commands to the loader to allow one to set hints via an intuitive
+ inteface, like 'disable apic0' sets 'hint.apic.0.disabled=1' or maybe
+ 'devcfg disable apic0' 'devcfg set sio0 irq 4', 'devcfg delete foo0'
More information about the p4-projects
mailing list