Re: perf: perf_fuzzer quickly locks up on 4.15-rc7

[Ingo Molnar]

* Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Tue, Jan 09, 2018 at 10:24:55AM -0500, Vince Weaver wrote:
> > On Tue, 9 Jan 2018, Peter Zijlstra wrote:
> > 
> > > > I'll try your patch and see if it makes a difference.
> > > 
> > > I suspect not, it shouldn't be PTI specific.
> > 
> > yes, applying your patch didn't help, still locks up on the Haswell 
> > machine.
> 
> So CONFIG_PAGE_TABLE_ISOLATION=y and booting with "pti=off" makes it
> 'work', right?
> 
> > Is there any debugging I could turn on that would help?  I tried KASAN 
> > but it didn't help.  I think I have the regular lockdep stuff enabled.
> > 
> > alt-sysrq doesn't work either (or at least, the version using BREAK over 
> > the serial console doesn't, I can maybe try hooking up a keyboard/display 
> > to see if that helps).
> 
> The below is always my first try to get something out of the machine,
> after that its printk() stuffing code to see how far we get..
> 
> In particular I'd start instrumenting the NMI entry_64.S code, because
> that's really the biggest difference between PTI and !PTI :/ all rather
> bothersome I'm afraid.
> 
> Really sucks I cannot as yet reproduce.
> 
> ---
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 5e486b6509e5..fc9021fd6e3c 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -1172,6 +1172,11 @@
>  			parameter will force ia64_sal_cache_flush to call
>  			ia64_pal_cache_flush instead of SAL_CACHE_FLUSH.
>  
> +	force_early_printk
> +			Forcefully uses early_console (as per earlyprintk=)
> +			usage for regular printk, bypassing everything,
> +			including the syslog (dmesg will be empty).

BTW., this looks very useful - please consider pushing it upstream if it does not 
hurt any of the other syslog usecases.

Thanks,

	Ingo