next_tick(). If writing to TICK_CMPR fails, we lose hardclock interrupt
on secondary CPUs.
About BB_ERRATA_1 from comment in OpenSolaris:
/*
* Writes to the TICK_COMPARE register sometimes fail on blackbird modules.
* The failure occurs only when the following instruction decodes to wr or
* wrpr. The workaround is to immediately follow writes to TICK_COMPARE
* with a read, thus stalling the pipe and keeping following instructions
* from causing data corruption. Aligning to a quadword will ensure these
* two instructions are not split due to i$ misses.
*/
appended while the receiver is blocked, the sockbuf will be corrupted.
Dequeue control messages from the sockbuf and sync its state in one
pass. Only then process the control messages. From FreeBSD.
All Reference Objects returned via the AcpiEvaluateObject interface are now
marked as type "REFERENCE" instead of "ANY". The type ANY is now reserved for
NULL objects - either NULL package elements or unresolved named references.
A device_t should never be cast with void*. This fixes reboot on acorn32.
I also suspect it'll fix a panic mjf was seeing due to corruption of the
alldevs list.
put in the right spot. The 'next' link in the new entry must become globally
visible before the list head is updated. This could have affected systems
with weak memory ordering like the alpha.
mandatory. Remove the 4BSD run queue code. Effects:
- Pluggable scheduler is only responsible for co-ordinating timeshared jobs.
- All systems run with per-CPU run queues.
- 4BSD scheduler gets processor sets / affinity.
- 4BSD scheduler gets a significant peformance boost on some workloads.
Discussed on tech-kern@.
but only the processes that emulate threads (with LINUX_CLONE_THREAD
set).
This fix a problem for child processes that share address space with
the parent. At exit, the child will die silently, leaving the parent
waiting indefinitely for its end ...
pushing the syncer before considering rate limiting the deletes. We hold
vnodes locked and it's likely that the syncer will try to lock them while
flushing, leading to the syncer and remover proceeding in lockstep and
making very little forward progress. XXX this is not a solution.
- Use atomic ops directly, since rwlocks work the same way on all platforms.
- Try to make it a bit more cache efficient, and use branch hints.
- Fix a bug in rw_downgrade() where the turnstile lock was not released.
- Remove a couple of redundant assertions.
- Use atomic_swap instead of atomic_cas where it's safe to do so.
- After acquiring the turnstile lock in rw_vector_enter, check if the
owner is running again and spin if so.
- Introduce and use rw_onproc() instead of abusing mutex_onproc().
- Change the handoff/release algorithm to reduce the window when a rwlock
can held, but the owner not on a CPU.
* Asynchronous operation with result retrieval via select/poll
* Mutliple-request submit/retrieve ioctls
* Mutliple-session create-destroy ioctls
Revise/rewrite crypto.4 manual page. It should now be much easier to write
new applications to this API.
Measured performance for trivial requests: 84,000 very short modular math
operations/sec, 120,000 very short md5 hashes per sec (with a hardware
accellerator of moderate performance but very low latency, whose driver
will be contributed at a later date).
Contributed to TNF by Coyote Point Systems, Inc.
- use partition 'e' for a disk with no NetBSD slice.
- add Linux ATAG to feature ELF Linux kernel booting. Commented out for now.
- fix errors to retrieve iospace BARs.
- don't use separate statintr on primary CPU.
- invoke clockintr instead of statintr on secondary CPUs.
This change fixes the problem setitimer(2) didn't work on secodary CPUs.
- Make ipflow_reap() private to ip_flow.c, and introduce ipflow_prune()
for external callers to use (avoids returning an ipflow * that is never
actually used anyway).
- percpu_getptr() is now called percpu_getref() and implicitly disables
preemption (via crit_enter()) when it is called.
- Added percpu_putref() which implicitly reenables preemption (via
crit_exit()).
at elansc(4). Take advantage of a suspend/resume cycle to reconfigure
the SC520's PCI host-bridge bus for higher performance. Update
the manual pages and the NET4501 kernel.
figuring out all the crazy nuances of getting this working, and to
Michael Lorenz for testing/fixing my changes on macppc. Tested with a
quad-proc 7044-270.
Summary of changes:
Bumped CPU_MAXNUM to 16 on ofppc.
Added md_* routines to ofppc/cpu.c, to sync the timebase, and awaken the CPUs.
Fixed a bug in the test for a 64bit bridge cpu early in locore.S
Added code to set the interrupt priority for all CPUs with an openpic.
Change rtas to probe before cpus, to allow use of the rtas freeze/thaw
timebase code routines.
Fix CPU_INFO_FOREACH macro to iterate through detected cpus, not CPU_MAXNUM.
Change most uses of ci_cpuid to ci_index, to deal with CPUs that do not allow
writing to SPR_PIR. Don't write SPR_PIR unless the secondary cpu identifies
itself as 0.
Change the hatchstack/interrupt stack allocations to allocate a 8192byte
interrupt stack, and a 4096 byte hatch stack, align them to 16 bytes, and
allocate them no lower than 0x10000. Allocate them separately to prevent the
hatch stack corrupting the interrupt stack later on.
If the CPU is a 64bit cpu, copy SPR_ASR in cpu_hatch()
Set the idle stack to ci->ci_data.cpu_idlelwp->l_addr->u_pcb.pcb_sp.
Add OF_start_cpu(). Add a routine to ofwoea_initppc to spin up secondary
procs early, and place them into a spinloop waiting for the hatch routines
to be ready.
Modify the ipi routines to deal with openpics that reverse byte order on read
from an ipi register. (such as on the 7044)
Change the rtas setup to allocate the rtas physical base address above
the kernel, to avoid mucking up the hatch/interrupt stacks.
