Re: [RFC][PATCH 1/2] x86: Allow breakpoints to emulate call functions

From: Steven Rostedt
Date: Mon May 06 2019 - 21:54:47 EST

On Mon, 6 May 2019 18:34:59 -0700
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Mon, May 6, 2019 at 6:04 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >
> > That iterator does something special for each individual record. All
> > 40,000 of them.
> .. yes, but the 'int3' only happens for *one* of them at a time.
> Why would it bother with the other 39,999 calls?
> You could easily just look up the record at the int3 time, and just
> use the record. Exactly the same way you use the one-at-a-time ones.
> Instead, you emulate a fake call to a function that *wouldn't* get
> called, which now does the lookup there. That's the part I don't get.
> Why are you emulating something else than what you'd be rewriting?

Ah, now I see what you are saying. Yes, I could pass in what it is
suppose to call. But I was trying to use the same code for all the
alternative solutions we were passing around, and this became the
"default" case that would work with any int3_emulate_call
implementation we came up with.

That is, if we call ftrace_regs_caller() for any scenario it should
work. Even if the call was suppose to be a nop, because in that case,
all the ftrace_ops registered in the iterator would refuse to have
their handler be called for that function.

I sent you a single patch, but that was really just a diff of several
applied patches against your unmodified tree. The last patch implements
the ftrace code. And I had it this way because it should work for any
of the implementations.

I could modify it so that it picks what function to call when the int3
is triggered. I think all the solutions we are down to allow that now.
Some of the early ideas had me call one function for all int3s due to
trampolines and such.

Also, I figured just calling ftrace_regs_caller() was simpler then
having that int3 handler do the hash look ups to determine what handler
it needs to call.

-- Steve