On Wed, Jan 27, 2016 at 09:51:12PM +1100, Balbir Singh wrote:

On Mon, 25 Jan 2016 16:38:48 +0100
Torsten Duwe [off-list ref] wrote:

quoted

Changes since v5:
  * extra "std r0,LRSAVE(r1)" for gcc-6
    This makes the code compiler-agnostic.
  * Follow Petr Mladek's suggestion to avoid
    redefinition of HAVE_LIVEPATCH

I looked at the patches - well mostly patches 1 and 2, some quick questions

1. I know -mprofile-kernel is a big optimization win, do we need it or can
we incrementally add it?

There's a reason why these are first ;-)
The following ones assume -mprofile-kernel is used.
The disadvantage is all relevant registers need to be saved before calling
further C code in between functions. On the Pro side, no stack frame has been
created at that point. These are assumptions made all over the ftrace-with-regs
and live patching code here.

2. Some of the hardcoded checks for opcode are hard to review, I know they've
been there in similar forms for a while. May be as an iterative step we should
give the numbers some meaning and use proper helpers for it.

Yes, Michael has already criticised that. No further literal hex constants, I promise.

I am going to give the patches a spin

Thanks! Make sure you use a compiler that can disable -mprofile-kernel with "notrace".

	Torsten