Found some interesting things, the delays are caused by
1) throttle_vm_writeout()
I removed it, Andrew worries about that, but hopefully there's a
better solution to his worries
2) atime updates
my uml image did not have them turned off
3) UML timer tick does not seem very reliable, here's what the below
patch outputs during writing. It shows the delay between ticks in
milliseconds:
[42949415.110000] zzz - 267
[42949415.120000] zzz - 862
[42949415.130000] zzz - 71
[42949415.140000] zzz - 599
[42949415.170000] zzz - 3746
[42949415.190000] zzz - 2961
[42949415.200000] zzz - 3272
[42949415.210000] zzz - 253
[42949415.220000] zzz - 1692
[42949415.230000] zzz - 823
[42949415.280000] zzz - 344
[42949415.310000] zzz - 224
I guess, the ubd interrupts confuse the timekeeping somehow, but the
details are unclear.
Miklos
Index: linux/arch/um/kernel/time.c
===================================================================
--- linux.orig/arch/um/kernel/time.c 2007-11-30 20:19:17.000000000 +0100
+++ linux/arch/um/kernel/time.c 2007-11-30 20:21:24.000000000 +0100
@@ -23,6 +23,12 @@ unsigned long long sched_clock(void)
void timer_handler(int sig, struct uml_pt_regs *regs)
{
unsigned long flags;
+ static long long last;
+ long long now = os_nsecs() / 1000000;
+
+ if (now - last > 50)
+ printk("zzz - %lli\n", now-last);
+ last = now;
local_irq_save(flags);
do_IRQ(TIMER_IRQ, regs);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
