Re: [RFC PATCH v3 11/12] powerpc: Remove unreachable() from WARN_ON()

From
Christophe Leroy <hidden>
Date
2022-06-29 18:30:31
Also in
linux-arm-kernel, lkml
Reply to
Sathvika Vasireddy, 2022-06-27
Hi Sathvika,

Adding ARM people as they seem to face the same kind of problem (see 
https://patchwork.kernel.org/project/linux-kbuild/patch/20220623014917.199563-33-chenzhongjin@huawei.com/)

Le 27/06/2022 à 17:35, Sathvika Vasireddy a écrit :
On 25/06/22 12:16, Christophe Leroy wrote:
quoted
Le 24/06/2022 à 20:32, Sathvika Vasireddy a écrit :
quoted
objtool is throwing *unannotated intra-function call*
warnings with a few instructions that are marked
unreachable. Remove unreachable() from WARN_ON()
to fix these warnings, as the codegen remains same
with and without unreachable() in WARN_ON().
Did you try the two exemples described in commit 1e688dd2a3d6
("powerpc/bug: Provide better flexibility to WARN_ON/__WARN_FLAGS() with
asm goto") ?

Without your patch:

00000640 <test>:
   640:    81 23 00 84     lwz     r9,132(r3)
   644:    71 29 40 00     andi.   r9,r9,16384
   648:    40 82 00 0c     bne     654 <test+0x14>
   64c:    80 63 00 0c     lwz     r3,12(r3)
   650:    4e 80 00 20     blr
   654:    0f e0 00 00     twui    r0,0

00000658 <test9w>:
   658:    2c 04 00 00     cmpwi   r4,0
   65c:    41 82 00 0c     beq     668 <test9w+0x10>
   660:    7c 63 23 96     divwu   r3,r3,r4
   664:    4e 80 00 20     blr
   668:    0f e0 00 00     twui    r0,0
   66c:    38 60 00 00     li      r3,0
   670:    4e 80 00 20     blr


With your patch:

00000640 <test>:
   640:    81 23 00 84     lwz     r9,132(r3)
   644:    71 29 40 00     andi.   r9,r9,16384
   648:    40 82 00 0c     bne     654 <test+0x14>
   64c:    80 63 00 0c     lwz     r3,12(r3)
   650:    4e 80 00 20     blr
   654:    0f e0 00 00     twui    r0,0
   658:    4b ff ff f4     b       64c <test+0xc>        <==

0000065c <test9w>:
   65c:    2c 04 00 00     cmpwi   r4,0
   660:    41 82 00 0c     beq     66c <test9w+0x10>
   664:    7c 63 23 96     divwu   r3,r3,r4
   668:    4e 80 00 20     blr
   66c:    0f e0 00 00     twui    r0,0
   670:    38 60 00 00     li      r3,0            <==
   674:    4e 80 00 20     blr                <==
   678:    38 60 00 00     li      r3,0
   67c:    4e 80 00 20     blr
The builtin variant of unreachable (__builtin_unreachable()) works.

How about using that instead of unreachable() ?
In fact the problem comes from the macro annotate_unreachable() which is 
called by unreachable() before calling __build_unreachable().

Seems like this macro adds (after the unconditional trap twui) a call to 
an empty function whose address is listed in section .discard.unreachable

     1c78:       00 00 e0 0f     twui    r0,0
     1c7c:       55 e7 ff 4b     bl      3d0 
<qdisc_root_sleeping_lock.part.0>


RELOCATION RECORDS FOR [.discard.unreachable]:
OFFSET           TYPE              VALUE
0000000000000000 R_PPC64_REL32     .text+0x00000000000003d0

The problem is that that function has size 0:

00000000000003d0 l     F .text	0000000000000000 
qdisc_root_sleeping_lock.part.0


And objtool is not prepared for a function with size 0.


The following changes to objtool seem to fix the problem, most warning 
are gone with that change.
diff --git a/tools/objtool/elf.c b/tools/objtool/elf.c
index 63218f5799c2..37c0a268b7ea 100644
--- a/tools/objtool/elf.c
+++ b/tools/objtool/elf.c
@@ -77,6 +77,8 @@ static int symbol_by_offset(const void *key, const 
struct rb_node *node)

  	if (*o < s->offset)
  		return -1;
+	if (*o == s->offset && !s->len)
+		return 0;
  	if (*o >= s->offset + s->len)
  		return 1;
@@ -400,7 +402,7 @@ static void elf_add_symbol(struct elf *elf, struct 
symbol *sym)
  	 * Don't store empty STT_NOTYPE symbols in the rbtree.  They
  	 * can exist within a function, confusing the sorting.
  	 */
-	if (!sym->len)
+	if (sym->type == STT_NOTYPE && !sym->len)
  		rb_erase(&sym->node, &sym->sec->symbol_tree);
  }

---

I also had objtool running for ever on 
arch/powerpc/sysdev/xics/icp-hv.o, which I fixed with the below hack:
diff --git a/tools/objtool/check.c b/tools/objtool/check.c
index 51b6dcec8d6a..ef2303ad6381 100644
--- a/tools/objtool/check.c
+++ b/tools/objtool/check.c
@@ -529,7 +529,7 @@ static struct instruction *find_last_insn(struct 
objtool_file *file,
  	unsigned int offset;
  	unsigned int end = (sec->sh.sh_size > 10) ? sec->sh.sh_size - 10 : 0;

-	for (offset = sec->sh.sh_size - 1; offset >= end && !insn; offset--)
+	for (offset = sec->sh.sh_size - 1; offset && offset >= end && !insn; 
offset--)
  		insn = find_insn(file, sec, offset);

  	return insn;
---

Now I only have the following two warnings:

arch/powerpc/sysdev/xics/icp-hv.o: warning: objtool: can't find 
unreachable insn at .text.unlikely+0x0
drivers/crypto/vmx/aesp8-ppc.o: warning: objtool: 
aes_p8_set_encrypt_key+0x44: unannotated intra-function call

The first one is linked to the infinite loop I hacked. So I now have to 
understand what the problem really is.

The second one is an assembly file aesp8-ppc.S which lacks the changes 
you did in other assembly files in patch 12.

Christophe
Thread (41 messages) 41 messages, 7 authors, last reply 2022-07-08
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help