Re: [GIT PULL] irq fix for 3.12-rc

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Sep 30, 2013 at 09:07:19AM -0700, Linus Torvalds wrote:
> On Mon, Sep 30, 2013 at 7:55 AM, Frederic Weisbecker <fweisbec@gmail.com> wrote:
> > ...
> >     the chances for a stack overrun as reported in powerpc for example:
> >
> >     [ 1002.364577] do_IRQ: stack overflow: 1920
> >     [ 1002.364653] CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1
> >     [ 1002.364734] Call Trace:
> >     [ 1002.364770] [c0000000050a8740] [c0000000000157c0] .show_stack+0x130/0x200 (unreliable)
> >     [ 1002.364872] [c0000000050a8810] [c00000000082f6d0] .dump_stack+0x28/0x3c
> >     [ 1002.364955] [c0000000050a8880] [c000000000010b98] .do_IRQ+0x2b8/0x2c0
> >     [ 1002.365039] [c0000000050a8930] [c0000000000026d4] hardware_interrupt_common+0x154/0x180
> >     [ 1002.365148] --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp]
> >     [ 1002.365148]     LR = .cp_start_xmit+0x390/0x820 [8139cp]
> >     [ 1002.365359] [c0000000050a8d40] [c0000000006d7f14] .dev_hard_start_xmit+0x394/0x640
> >  ...
> 
> Btw, I'd really wish people edited things like this when putting them
> in the commit logs. I try to do it when I get them (usually though
> Andrew's patch-bombs), just because there's just a ton of detail there
> that just isn't relevant for the actual issue at hand.
> 
> The kernel oops messages try to contain all kinds of possibly-relevant
> data, which makes them useful for a wide range of situations ("oh, it
> looks like a single-bit flip"), but at the same time means that once
> you know what the problem is, 90% of the data printed out is just pure
> noise and at that point no longer helpful, but just makes it harder to
> see what's actually the issue.
> 
> So please, after you've analyzed an oops, don't use the raw oops data
> any more. Usually what remains relevant is the actual oops message
> itself, and the backtrace.I try to generally edit out the hex
> representation of the symbol information, and obviously stale entries
> from the backtrace. I'm not consistent, see for example commit
> 6f6b8951897e (register info remains) vs commit d6394b590029 (mainly
> just call trace) vs commit 3e6b11df2451 (where I just truncated it
> mercilessly). And no, I don't always clean things up (it can be a
> bother), but I generally try, so now I'm just trying to spread the
> word..
> 
> Because at some point the excess verbiage really goes from "that's
> useful" to being a blob of noise that actually takes away from the
> message.

Yeah, I did such things sporadically before. Well, it summed up to
simply remove the timestamps from backtraces but yeah, then I've become
less patient about that and now I simply paste the raw thing.

I'll take care of that and prune these things on my future patches.