lkml.org 
[lkml]   [2006]   [Sep]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: 2.6.18-rc6-mm1: GPF loop on early boot

* Andi Kleen <ak@suse.de> wrote:

> > ugh, "having a working current" is "so much infrastructure" ??
>
> Together with stacktrace the infrastructure needed is quite
> considerable.

Having the ability to produce a backtrace _anywhere within the kernel_
is a pretty fundamental thing too. So IMO the problems with stacktrace
were the stack dumping code's weaknesses, independent of lockdep, and
they would have triggered in one way or another as well. Not that i
think that doing a dwarf unwinder is simple - and you certainly did a
great job with it. Now with stacktrace we do have a pretty powerful
infrastructure to do transparent execution tracing: for example the
page-tracing code could be changed to use stacktrace.

> > the i686 PDA patches introduce tons of early_current() uses. While i
> > like the new PDA code, its bootstrap (like x86_64's PDA bootstrap) is
> > too fragile in my opinion, and it will regularly hit instrumenting
> > patches.
>
> Or the instrumentation patches just always need to check some global
> variable. Maybe system_state could be extended or something.

maybe, but i find that unrobust too. Example: i debugged many
early-bootup hangs via the use of an early_printk() based function
execution tracer (a feature of the latency tracing patchset) - the lack
of working 'current' would hit that code too.

Yes, it can all be worked around, but the fundamental weakness IMO is
that we dont have a working 'current' in all situations.

> > should be moved into 32-bit early boot userspace code (like
> > compressed/misc.c) and it will thus not depend on any kernel
> > infrastructure.
>
> Ok I guess it would make sense to add a i386_start_kernel to i386 and
> initialize the boot PDA there. I would also move early_cpu_init into
> there because that also avoids quite some mess later.

yes. And move all of this _outside_ the normal kernel build process, so
that whatever ad-hoc or less ad-hoc instrumentation is added (be that
PROFILE_UNLIKELY, function tracing, 'get_current' based hacks, irqtrace,
lockdep or whatever) it wont be included in that code. That gets rid of
most of the dependency and bootstrap problems. It's hard enough already
to debug early boot code.

Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-09-10 16:13    [W:0.170 / U:0.068 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site