[linux-audio-dev] lowish-lat for 2.4.0-test10-pre3

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: [linux-audio-dev] lowish-lat for 2.4.0-test10-pre3
From: Andrew Morton (andrewm_AT_uow.edu.au)
Date: Sat Oct 21 2000 - 16:58:00 EEST

Patch is at

http://www.uow.edu.au/~andrewm/linux/2.4.0-test10-pre3-low-latency.patch

Changes:

- Non-inlined the set_current_state()/schedule() code. Saves
  a couple of hundred bytes.

- Simplified some code in dcache.c - the performance benefit
  wasn't worth the ugliness.

- Put a couple of rescheduling points in the /proc files. This
  is because reading /proc/meminfo is part of Benno's test suite :)

- There were some complaints about the decision to disable this
  patch for SMP. So it can be reenabled via CONFIG_LOLAT_SMP (Kernel
  Hacking menu).

  But no promises here at all. It may seem to work, but if you
  start getting heavy spinlock contention in the kernel it
  goes bad.

- Fixed (I believe) the reschedule and signal race in
  arch/i386/kernel/entry.S. This is the only x86-specific
  part of this patch.

  I simply stuck a `cli' in there:

--- linux-2.4.0-test10-pre3/arch/i386/kernel/entry.S Sat Oct 14 17:02:03 2000
+++ linux-akpm/arch/i386/kernel/entry.S Sat Oct 21 22:15:47 2000
@@ -215,21 +215,27 @@
         jne handle_softirq
         
 ret_with_reschedule:
- cmpl $0,need_resched(%ebx)
- jne reschedule
- cmpl $0,sigpending(%ebx)
- jne signal_return
+ cli
+ movl need_resched(%ebx),%eax
+ orl sigpending(%ebx),%eax
+ jne signal_or_resched
 restore_all:
         RESTORE_ALL
 
         ALIGN
-signal_return:
+signal_or_resched:
+ cmpl $0,need_resched(%ebx)
+ jne reschedule
+ # Must be a pending signal
         sti # we can get here from an interrupt handler
         testl $(VM_MASK),EFLAGS(%esp)
         movl %esp,%eax
         jne v86_signal_return
         xorl %edx,%edx
         call SYMBOL_NAME(do_signal)
+ cli
+ cmpl $0,need_resched(%ebx)
+ jne reschedule
         jmp restore_all
 
         ALIGN
@@ -285,6 +291,7 @@
         
         ALIGN
 reschedule:
+ sti
         call SYMBOL_NAME(schedule) # test
         jmp ret_from_sys_call

  On the Mendocino the `cli' adds ~15 cycles to system calls. The
  cunning removal of a conditional jump from the syscall and interrupt
  fastpath reduces this to 13 cycles.

  So `getpid()' now takes 1.005x as long as it used to. So shoot me.

- Some testing results on 2.4.0-test9.

  These tests were run with the separate tcp_minisocks patch because
  the reaping of timed-wait sockets gets in the way with lmbench.
  See http://www.uow.edu.au/~andrewm/linux/schedlat.html#ddt

  Machine: 566 MHz UP Celeron, 256M RAM, UDMA66
  Workload: 2 instances of bonnie++
            2 instances of lmbench
            1 instance of mmap001, mmap002, mmap001, mmap002, ...
            1 instance of netperf
            1 SCHED_FIFO process handling a 1024 Hz signal stream

  After 20 hours the latency histogram was:

  0-1 milliseconds: ~78,000,000
  1-2 milliseconds: 3
  5-8 milliseconds: 8 (kmem_cache_reap->kmem_slab_destroy->avl_remove)
  9-10 milliseconds: 3 (ide_intr)

  The kmem_cache_reap thing is due to the insane number of
  inodes and dentries which bonnie++ leaves around. It doesn't
  happen normally.

  Not sure why the IDE ISR had those glitches.

New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Sat Oct 21 2000 - 17:34:51 EEST