Re: [PATCH v4 01/13] objtool: Remove CFI save/restore special case

[Peter Zijlstra]

On Thu, Mar 26, 2020 at 08:44:48AM -0500, Josh Poimboeuf wrote:
> On Thu, Mar 26, 2020 at 01:58:44PM +0100, Peter Zijlstra wrote:
> > So instr_begin() / instr_end() have this exact problem, but worse. Those
> > actually do nest and I've ran into the following situation:
> > 
> > 	if (cond1) {
> > 		instr_begin();
> > 		// code1
> > 		instr_end();
> > 	}
> > 	// code
> > 
> > 	if (cond2) {
> > 		instr_begin();
> > 		// code2
> > 		instr_end();
> > 	}
> > 	// tail
> > 
> > Where objtool then finds the path: !cond1, cond2, which ends up at code2
> > with 0, instead of 1.
> 
> Hm, I don't see the nesting in this example, can you clarify?

Indeed no nesting here, but because they can nest we have that begin as
+1, end as -1 and then we sum it over the code flow.

Then given that, the above, ends up as -1 + 1 in the !cond1,cond2 case,
because that -1 escapes the cond1 block.

> > I've also seen:
> > 
> > 	if (cond) {
> > 		instr_begin();
> > 		// code1
> > 		instr_end();
> > 	}
> > 	instr_begin();
> > 	// code2
> > 	instr_end();
> > 
> > Where instr_end() and instr_begin() merge onto the same instruction of
> > code2 as a 0, and again code2 will issue a false warning.
> > 
> > You can also not make objtool lift the end marker to the previous
> > instruction, because then:
> > 
> > 	if (cond1) {
> > 		instr_begin();
> > 		if (cond2) {
> > 			// code2
> > 		}
> > 		instr_end();
> > 	}
> > 
> > Suffers the reverse problem, instr_end() becomes part of the @cond2
> > block and cond1 grows a path that misses it entirely.
> > 
> > So far I've not had any actual solution except adding a NOP to anchor
> > the annotation on.
> 
> Are you adding the NOP to the instr_end() annotation itself?  Seems like
> that would be the cleanest/easiest.

That actually generates a whole bunch of 'stupid' unreachable warnings.
Also, in the hope of still coming up with something saner, we've been
carrying a minimal set of explicit nop()s.

> Though it is sad that we have to change the code to make objtool happy
> -- would be nice if we could come up with something less intrusive.

Very much yes, but so far that's been eluding me.