On Sat, Nov 04, 2017 at 06:51:34PM +0100, Martin Pieuchot wrote:
> On 04/11/17(Sat) 17:20, Paul Irofti wrote:
> > Hi,
> >
> > Someone, somewhere (perhaps on bugs@?) said they would like to see
> > something useful when typing show panic on a page fault'd kernel. That
> > struck very close to home so I went ahead and did it.
> >
> > The following diff shows the uvm_fault and the first trace entry in an
> > attempt to mimic a real show panic. The uvm_fault entry is extra.
> >
> > Currently I only wrote support for amd64 but I can add support for
> > other architectures if you guys like it.
> >
> > Thoughts?
>
> I think you should just change the printf("uvm_fault...) into a panic.
>
> This introduces a behavior change when DDB is not compiled in. In such
> kernels trap informations from 'we_re_toast' won't be printed. I doubt
> we care about such kernels, but if we do, when uvm_fault(9) returns an
> error its arguments and the traceback are generally enough to debug the
> problem.
I completely agree. It is easier and more clear that way.
The reason I did not propose that in my original diff is because I did
not want to change existing behaviour and get dragged into politics.
If the general consent is that is the way to go I will come back with a
simpler diff.
