RE: [PATCH 0/5] add printk specifier %px, unique identifier

From: David Laight
Date: Tue Nov 28 2017 - 05:12:16 EST


From: Eric W. Biederman
> Sent: 28 November 2017 06:27
> Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> writes:
>
> > On Mon, Nov 27, 2017 at 4:03 PM, Linus Torvalds
> > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >>
> >> So the big remaining ones for me are the /proc/<pid>/stack (stack
> >> pointers) and the /proc/net/* ones.
> >>
> >> I'm a bit disappointed that those haven't been fixed already and
> >> aren't even in this series..
> >
> > Oh well, I just did /proc/<pid>/stack by making it just print 0
> > unconditionally rather than the hex number.
>
> Patch?
>
> I know I have used /proc/<pid>/stack manually many times when looking
> at a system where something is hung/weird and I needed to see what is
> going on. The backtrace inside the kernel can be invaluable.

Ditto - after I spotted it.
Also the similar tracebacks from echo t >/proc/sysrq-trigger
although they are less useful unless you've a big kernel message buffer.
Although they can be requested from a keyboard if everything except the
keyboard interrupt is borked.

> At the same time I don't know if we actually need the hex address.
> But please don't break that interface it is very useful.

Definitely need to know which addresses are zero (or near zero).
I will have tied the addresses there to ones available elsewhere.
(In private trace that won't be affected by whatever kernel printf
does with %p.)

If you want to hide addresses, then maybe use a write-only sysctl.

David