i've attached the next round of softirq-handling fixes.
correctness fixes:
- check for softirqs in the signal return path(s)
- make sure all the entry.S return paths check for softirqs with
interrupts disabled, otherwise we can end up getting another softirq
right after the test. (and this causes a missed softirq.)
- add softirq handling to the ACPI and APM code (untested)
performance tweaks:
- remove __cli() from idle-poll, it's not needed.
- separate exception handling and irq-ret path to avoid double-checking
for softirqs
with this -A1 patch applied, softirqs should be executed by the kernel at
the first possible place, and there should be no unlimited softirq
latencies anymore.
patch is against vanilla 2.4.5 (patch includes the previous fixes as
well). The patch compiles, boots and works just fine on x86 SMP.
Ingo
This archive was generated by hypermail 2b29 : Thu May 31 2001 - 21:00:29 EST