On Mon, 19 Mar 2018 14:43:00 +0530
"Naveen N. Rao" <naveen.n....@linux.vnet.ibm.com> wrote:
> diff --git a/arch/powerpc/kernel/trace/ftrace_64_mprofile.S
> b/arch/powerpc/kernel/trace/ftrace_64_mprofile.S
> index 3f3e81852422..fdf702b4df25 100644
> --- a/arch/powerpc/kernel/trace/ftrace_64_mprofile.S
> +++ b/arch/powerpc/kernel/trace/ftrace_64_mprofile.S
> @@ -60,6 +60,19 @@ _GLOBAL(ftrace_caller)
> mfxer r10
> mfcr r11
>
> +#ifdef CONFIG_KVM
> + lbz r3, PACA_FTRACE_DISABLED(r13)
> + cmpdi r3, 0
> + beq 1f
> + mflr r3
> + mtctr r3
> + REST_GPR(3, r1)
> + addi r1, r1, SWITCH_FRAME_SIZE
> + mtlr r0
> + bctr
> +1:
> +#endif
I wonder if we should try to move the return out of the fast path (for
cache reasons), as most of the time the above compare will be true. That
is:
#ifdef CONFIG_KVM
lbz r3, PACA_FTRACE_DISABLED(r13)
cmpdi r3, 0
bne no_trace
#endif
/* rest of ftrace_caller code */
/* after ftrace_caller... */
bctr /* jump after _mcount site */
#ifdef CONFIG_KVM
no_trace:
mflr r3
mtctr r3
REST_GPR(3, r1)
addi r1, r1, SWITCH_FRAME_SIZE
mtlr r0
bctr
#endif
-- Steve
> +
> /* Get the _mcount() call site out of LR */
> mflr r7
> /* Save it as pt_regs->nip */
