I have a source level workaround for this bug.

In the shell/ash.c

It defines a function
static int evaltree(union node *, int);

and it also adds an alias to this function with attribute __noreturn__.

static
int evaltreenr(union node *, int) __attribute__ ((alias("evaltree"),__noreturn__));


and it dies at this piece of code when I run
busybox ash shell/ash_test/ash-misc/wait4.tests 

unsigned is_or = n->type - NAND;
   status = evaltree(
                n->nbinary.ch1,
                (flags | ((is_or >> 1) - 1)) & EV_TESTED
                );
        if ((!status) == is_or || evalskip)
               break;
           n = n->nbinary.ch2;
evaln:
                evalfn = evaltree;
calleval:
                status = evalfn(n, flags);   <---dies here. wrong function pointer.
                goto setstatus;
        }

If I remove the attribute __noreturn__ from evaltreenr,
the unittest passes, recovery image works.

that attribute usage does seem evil, but does it mean it's a bug in clang ?

yes, looks very evil. 
It may be difficult to say that this is clang bug. I dont think this is the intent of the alias attribute.
can you make evaltreenr an alsways inline routine with the nr attribute? 
I think that is the correct way to do what they want.

i'm not sure it violates the intent of the alias attribute.  imo, the point of alias is to set up another symbol/entry point but with different attributes (which *don't* propagate back to the original symbol).  most often it's used to get different bindings/visibility.

in that light, adding a noreturn attribute doesn't seem like it should break things.  so if clang is propagating the noreturn attribute back to the target of the alias (evaltree in this case), seems like a bug in clang.

upstream has a fix shown below.
Mike, could you please backport it to gentoo upstream so that I can pull
it over?

Thanks

commit 619d9b5e6848a72350126ea9c1e413fd133181e3
Author: Denys Vlasenko <vda.linux@googlemail.com>
Date:   Fri Jul 28 15:28:33 2017 +0200

    ash: less hackish implementation of evaltreenr()
    
    Defining a function alias with __attribute__ ((alias("evaltree"),__noreturn__))
    is not that usual, and clang had a bug which made it misunderstand
    this construct.
    

The following revision refers to this bug:
  https://chromium.googlesource.com/chromiumos/overlays/portage-stable/+/a5a2b4a008a31ac63d7a50988aa9e01bac90f62c

commit a5a2b4a008a31ac63d7a50988aa9e01bac90f62c
Author: Yunlian Jiang <yunlian@google.com>
Date: Fri Oct 06 22:51:15 2017

busybox: upgrade to 1.27.2

This pulls upstream busybox 1.27.2 to fix a runtime bug if built
with LLVM.

BUG= chromium:659834 
TEST=emerge-terra busybox

Change-Id: I79299b5e0d893f2030211fa5eca97e93fe334f2c
Reviewed-on: https://chromium-review.googlesource.com/705395
Commit-Ready: Yunlian Jiang <yunlian@chromium.org>
Tested-by: Yunlian Jiang <yunlian@chromium.org>
Reviewed-by: Jason Clinton <jclinton@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>

[rename] https://crrev.com/a5a2b4a008a31ac63d7a50988aa9e01bac90f62c/sys-apps/busybox/busybox-1.27.2.ebuild
[modify] https://crrev.com/a5a2b4a008a31ac63d7a50988aa9e01bac90f62c/sys-apps/busybox/Manifest
[delete] https://crrev.com/793b48a7631ad6937d6c6b68e4b9ad96b5f8dbd6/sys-apps/busybox/files/busybox-1.19.0-bb.patch

The following revision refers to this bug:
  https://chromium.googlesource.com/chromiumos/overlays/portage-stable/+/7bca43b3504c491ab95c6d1865a12d4a876ff1da

commit 7bca43b3504c491ab95c6d1865a12d4a876ff1da
Author: Yunlian Jiang <yunlian@google.com>
Date: Mon Oct 23 22:22:20 2017

busybox: backport an upstream patch.

This backports an upstream busybox patch to make it work with
LLVM.

BUG= chromium:659834 
TEST=build busybox with llvm, recovery image works.

Change-Id: I43256f2f604e15204c02899a32f7c732fad562e4
Reviewed-on: https://chromium-review.googlesource.com/710502
Commit-Ready: Yunlian Jiang <yunlian@chromium.org>
Tested-by: Yunlian Jiang <yunlian@chromium.org>
Reviewed-by: Manoj Gupta <manojgupta@chromium.org>
Reviewed-by: Jason Clinton <jclinton@chromium.org>

[add] https://crrev.com/7bca43b3504c491ab95c6d1865a12d4a876ff1da/sys-apps/busybox/files/busybox-1.27.2-clang.patch
[rename] https://crrev.com/7bca43b3504c491ab95c6d1865a12d4a876ff1da/sys-apps/busybox/busybox-1.27.2-r1.ebuild

The following revision refers to this bug:
  https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/c22592daf9b2649462f6e91021831975d0bfe37d

commit c22592daf9b2649462f6e91021831975d0bfe37d
Author: Yunlian Jiang <yunlian@google.com>
Date: Wed Oct 25 19:13:31 2017

busybox: use clang to build it.

This enables us to use clang to build busybox.

CQ-DEPEND=CL:710502
BUG= chromium:659834 
TEST=build a recovey image for falco and it works.

Change-Id: Ia682311772d715d87c4bd9b2cd2db46d8e8fb75c
Reviewed-on: https://chromium-review.googlesource.com/729498
Commit-Ready: Yunlian Jiang <yunlian@chromium.org>
Tested-by: Yunlian Jiang <yunlian@chromium.org>
Reviewed-by: Jason Clinton <jclinton@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>

[delete] https://crrev.com/a12f14463eb39129ef8fdc553f1c6a738fe71974/chromeos/config/env/sys-apps/busybox

The following revision refers to this bug:
  https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/497153a1968d96752cead817eee81023e73645d8

commit 497153a1968d96752cead817eee81023e73645d8
Author: Yunlian Jiang <yunlian@chromium.org>
Date: Wed Nov 01 07:00:13 2017

Revert "busybox: use clang to build it."

This reverts commit c22592daf9b2649462f6e91021831975d0bfe37d.

Reason for revert: chromium:779029

This might break the USB installation for minnie. Revert it first to see whether it helps.

Original change's description:
> busybox: use clang to build it.
>
> This enables us to use clang to build busybox.
>
> CQ-DEPEND=CL:710502
> BUG= chromium:659834 
> TEST=build a recovey image for falco and it works.
>
> Change-Id: Ia682311772d715d87c4bd9b2cd2db46d8e8fb75c
> Reviewed-on: https://chromium-review.googlesource.com/729498
> Commit-Ready: Yunlian Jiang <yunlian@chromium.org>
> Tested-by: Yunlian Jiang <yunlian@chromium.org>
> Reviewed-by: Jason Clinton <jclinton@chromium.org>
> Reviewed-by: Mike Frysinger <vapier@chromium.org>

Bug:  chromium:659834 
Change-Id: Ied3126c9ac6b1a20e52e7606c159b47f3d52c120
Reviewed-on: https://chromium-review.googlesource.com/747821
Commit-Ready: Yunlian Jiang <yunlian@chromium.org>
Tested-by: Yunlian Jiang <yunlian@chromium.org>
Reviewed-by: Jason Clinton <jclinton@chromium.org>
Reviewed-by: Yunlian Jiang <yunlian@chromium.org>

[add] https://crrev.com/497153a1968d96752cead817eee81023e73645d8/chromeos/config/env/sys-apps/busybox

[It appears that a bunch of old cros issues bulk-added the "Infra" component recently, but they should probably be "Infra>Client>ChromeOS".]