Re: pre-patch-2.0.31 try_to_free_page() messages

Ingo Molnar (mingo@pc7537.hil.siemens.at)
Sun, 1 Jun 1997 22:02:59 +0200 (MET DST)


On Sun, 1 Jun 1997, Leonard N. Zubkoff wrote:

> I assume you're referring to my SMP deadlock patch?
>
> My 2.0.30 SMP deadlock patch has solved the problem 100% for my systems but
> some other people still reported problems. However, despite Ingo's posting a
> clever new deadlock detection patch, no one has reported tracking down any of
> the remaining deadlocks.

i was able to produce and trace hard deadlocks in 2.1.36-40 kernels, and i
would say 20 out of those 21 traced down deadlocks [5-6 distinct types of
deadlocks] should be gone in 2.1.42, judging from the fixes seen. Maybe
all 21, have to recheck ;)

like Linus has written recently, 'normal' oopses look like deadlocks, due
to the delay loops in arch/i386/kernel/traps.c, remove them with the
attached patch.

i'd be interested if anyone still sees real SMP deadlocks with 2.1.42 and
this patch.

--- traps.c.original Sun Jun 1 21:56:43 1997
+++ traps.c Sun Jun 1 21:56:48 1997
@@ -191,8 +191,6 @@
spin_lock_irq(&die_lock);
printk("%s: %04lx\n", str, err & 0xffff);
show_registers(regs);
-do { int i=2000000000; while (i) i--; } while (0);
-do { int i=2000000000; while (i) i--; } while (0);
spin_unlock_irq(&die_lock);
do_exit(SIGSEGV);
}