Re: [PATCH 1/2] uprobes: add comment with insn opcodes, mnemonics and why we dont support them

[Oleg Nesterov <oleg@redhat.com>]

Firstly, let me remind that my understanding of low lovel hardware details
is very limited.

On 05/06, Denys Vlasenko wrote:
>
> Oleg, can you clear for me the following -
>
> If the probed instruction triggers an "illegal insn" or "privileged insn"
> CPU exception - are we completely fine?

Yes I think we are fine. I assume that, say, do_debug() won't be called in this
case, and do_invalid_op()->do_trap() should trigger arch_uprobe_xol_was_trapped()
logic.

Well, actually we are not 100% fine because si_addr can be wrong. But this is
not invalid_op-specific, we need to fix this anyway and the fix is simple.

I do not want to discuss this now, but I am going to make another series later
which adds something like uprobe_instruction_pointer(regs). It can (should) be
used by DO_ERROR_INFO() (perhaps by something else, not sure about math_error())
_and_ by show_unhandled_signals users (actually the main reason to me). The only
problem is that this code should be cleanuped first. In fact I was thinking about
this change from the very beginning of the recent fixes, the "wrong" ip reported
by do_general_protection() greatly complicated the investigation of that problem.
But I need to take a rest of uprobes ;)

Oleg.