Re: [RFC PATCH 00/11] printk: safe printing in NMI context

[Frederic Weisbecker]

On Fri, May 30, 2014 at 10:13:28AM +0200, Jan Kara wrote:
> On Thu 29-05-14 02:09:11, Frederic Weisbecker wrote:
> > On Thu, May 29, 2014 at 12:02:30AM +0200, Jiri Kosina wrote:
> > > On Fri, 9 May 2014, Petr Mladek wrote:
> > > 
> > > > printk() cannot be used safely in NMI context because it uses internal locks
> > > > and thus could cause a deadlock. Unfortunately there are circumstances when
> > > > calling printk from NMI is very useful. For example, all WARN.*(in_nmi())
> > > > would be much more helpful if they didn't lockup the machine.
> > > > 
> > > > Another example would be arch_trigger_all_cpu_backtrace for x86 which uses NMI
> > > > to dump traces on all CPU (either triggered by sysrq+l or from RCU stall
> > > > detector).
> > > 
> > > I am rather surprised that this patchset hasn't received a single review 
> > > comment for 3 weeks.
> > > 
> > > Let me point out that the issues Petr is talking about in the cover letter 
> > > are real -- we've actually seen the lockups triggered by RCU stall 
> > > detector trying to dump stacks on all CPUs, and hard-locking machine up 
> > > while doing so.
> > > 
> > > So this really needs to be solved.
> > 
> > The lack of review may be partly due to a not very appealing changestat on an
> > old codebase that is already unpopular:
> > 
> >  Documentation/kernel-parameters.txt |   19 +-
> >  kernel/printk/printk.c              | 1218 +++++++++++++++++++++++++----------
> >  2 files changed, 878 insertions(+), 359 deletions(-)
> > 
> > 
> > Your patches look clean and pretty nice actually. They must be seriously
> > considered if we want to keep the current locked ring buffer design and
> > extend it to multiple per context buffers. But I wonder if it's worth to
> > continue that way with the printk ancient design.
> > 
> > If it takes more than 1000 line changes (including 500 added) to make it
> > finally work correctly with NMIs by working around its fundamental flaws,
> > shouldn't we rather redesign it to use a lockless ring buffer like ftrace
> > or perf ones?
>   I agree that lockless ringbuffer would be a more elegant solution but a
> much more intrusive one and complex as well. Petr's patch set basically
> leaves ordinary printk path intact to avoid concerns about regressions
> there.
> 
> Given how difficult / time consuming is it to push any complex changes to
> printk I'd push for fixing printk from NMI in this inelegant but relatively
> non-contentious way and work on converting printk to lockless
> implementation long term. But before spending huge amount of time on that
> I'd like to get some wider concensus that this is really the way we want to
> go - at least AKPM and Steven - something for discussion in the KS topic I'd
> proposed I think [1].

Agreed, lets wait for others opinion. Andrew, Steve?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/