Re: [PATCH 4/5 v3] ftrace/x86_32: Clean up ftrace_regs_caller

[Steven Rostedt]

On Thu, 16 Mar 2017 11:19:44 -0700
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Thu, Mar 16, 2017 at 11:09 AM, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> > +
> > +       /* Since we don't care about cs, move flags there to simplify return */
> > +       movl    14*4(%esp), %eax
> > +       movl    %eax, 13*4(%esp)
> > +
> > +       /* Move return ip back to its original location */
> > +       movl    12*4(%esp), %eax
> > +       movl    %eax, 14*4(%esp)  
> 
> Could this perhaps be removed entirely?
> 
> The return code could instead do:
> 
>         ... restore all the normal registers ..
> 
>         # Now restore flags that is under the return address and our
> fake __KERNEL_CS
>         pushl 8(%esp)
>         popfl
> 
>         # and then return, skipping __KERNEL_CS and %flasg
>         ret $8
> 
> which is smaller and simpler than (again) playing games with stack entries.
> 
>                 Linus


The thing is we don't return, we jump to the location that may be
modified to run the function graph tracer.

.Lftrace_ret:
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
.globl ftrace_graph_call
ftrace_graph_call:
	jmp	ftrace_stub  <- this can turn to a jump to function graph tracing
#endif

WEAK(ftrace_stub)
	ret
END(ftrace_caller)


[...]

	popf					/* flags is in the cs location */
	jmp	.Lftrace_ret


-- Steve