Skip to contentSkip to conversation

s390/test_unwind: use raw opcode instead of invalid instruction

PATCH In review Ilie Halip 4 participants 9 messages Nov 17, 2021 – Nov 19, 2021 s390

lore
Share & tools
lore

Ilie Halip

author
#0 lore raw

PATCH 1/1 s390/test_unwind: use raw opcode instead of invalid instruction

Building with clang & LLVM_IAS=1 leads to an error:

    arch/s390/lib/test_unwind.c:179:4: error: invalid register pair
                        "       mvcl    %%r1,%%r1\n"
                        ^

The test creates an invalid instruction that would trap at runtime, but the LLVM inline assembler tries to validate it at compile time too.

Use the raw instruction opcode instead.

---
 arch/s390/lib/test_unwind.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/s390/lib/test_unwind.c +1 -1
index cfc5f5557c06..d342bc884b94 100644
@@ -176,7 +176,7 @@ static noinline int unwindme_func4(struct unwindme *u)
 		 * trigger specification exception
 		 */
 		asm volatile(
-			"	mvcl	%%r1,%%r1\n"
+			"	.insn e,0x0e11\n"	/* mvcl	%%r1,%%r1" */
 			"0:	nopr	%%r7\n"
 			EX_TABLE(0b, 0b)
 			:);
Signature · 1 line
2.25.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Ilie Halip

Building with clang & LLVM_IAS=1 leads to an error:

    arch/s390/lib/test_unwind.c:179:4: error: invalid register pair
                        "       mvcl    %%r1,%%r1\n"
                        ^
Quoted 8 lines

The test creates an invalid instruction that would trap at runtime, but the LLVM inline assembler tries to validate it at compile time too.

Use the raw instruction opcode instead.

Ilie, thanks for the patch!

So if I understand
https://sourceware.org/binutils/docs/as/s390-Directives.html#s390-Directives
https://sourceware.org/binutils/docs/as/s390-Formats.html
that e, prefix is for 16B opcodes?

LGTM, thanks again.

I triple checked that GAS, clang, and GNU objdump are in agreement in terms of encoding here.

Quoted 20 lines
---
 arch/s390/lib/test_unwind.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/s390/lib/test_unwind.c +1 -2
index cfc5f5557c06..d342bc884b94 100644
@@ -176,7 +176,7 @@ static noinline int unwindme_func4(struct unwindme *u)
                 * trigger specification exception
                 */
                asm volatile(
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */
                        "0:     nopr    %%r7\n"
                        EX_TABLE(0b, 0b)
                        :);
--

2.25.1

Signature · 2 lines
Thanks,
~Nick Desaulniers
#2 lore raw

Am 19.11.21 um 02:10 schrieb Nick Desaulniers:

Quoted 16 lines · Ilie Halip
Quoted 6 lines

Building with clang & LLVM_IAS=1 leads to an error:

     arch/s390/lib/test_unwind.c:179:4: error: invalid register pair
                         "       mvcl    %%r1,%%r1\n"
                         ^
Quoted 8 lines

The test creates an invalid instruction that would trap at runtime, but the LLVM inline assembler tries to validate it at compile time too.

Use the raw instruction opcode instead.

Quoted 6 lines

Ilie, thanks for the patch!

So if I understand
https://sourceware.org/binutils/docs/as/s390-Directives.html#s390-Directives
https://sourceware.org/binutils/docs/as/s390-Formats.html
that e, prefix is for 16B opcodes?

e is an instruction format as specified by the architecture. See http://publibfp.dhe.ibm.com/epubs/pdf/a227832c.pdf
without any parameters.
Normally RR would be the right thing for MVCL, but since we try to build an invalid opcode without the assembler noticing (ab)using e seem like a safer approach.

Quoted 4 lines

LGTM, thanks again.

added those and added my RB. applied to the s390 tree. Thanks

Quoted 26 lines

I triple checked that GAS, clang, and GNU objdump are in agreement in terms of encoding here.

Quoted 20 lines
---
  arch/s390/lib/test_unwind.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/s390/lib/test_unwind.c +1 -2
index cfc5f5557c06..d342bc884b94 100644
@@ -176,7 +176,7 @@ static noinline int unwindme_func4(struct unwindme *u)
                  * trigger specification exception
                  */
                 asm volatile(
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */
                         "0:     nopr    %%r7\n"
                         EX_TABLE(0b, 0b)
                         :);
--

2.25.1

#3 lore raw

Am 19.11.21 um 10:39 schrieb Christian Borntraeger:

Quoted 21 lines

Am 19.11.21 um 02:10 schrieb Nick Desaulniers:

Quoted 18 lines · Ilie Halip
Quoted 6 lines

Building with clang & LLVM_IAS=1 leads to an error:
     arch/s390/lib/test_unwind.c:179:4: error: invalid register pair "       mvcl    %%r1,%%r1\n"
                         ^

Quoted 8 lines

The test creates an invalid instruction that would trap at runtime, but the LLVM inline assembler tries to validate it at compile time too.

Use the raw instruction opcode instead.

Ilie, thanks for the patch!

Quoted 7 lines
Quoted 4 lines

So if I understand
https://sourceware.org/binutils/docs/as/s390-Directives.html#s390-Directives
https://sourceware.org/binutils/docs/as/s390-Formats.html
that e, prefix is for 16B opcodes?

e is an instruction format as specified by the architecture. See http://publibfp.dhe.ibm.com/epubs/pdf/a227832c.pdf

(page 5-3 for the instruction formats and page 7-289 for MVCL)

maintainer
#4 lore raw
In reply to Christian Borntraeger
Quoted 17 lines · Christian Borntraeger
Quoted 4 lines

So if I understand
https://sourceware.org/binutils/docs/as/s390-Directives.html#s390-Directives
https://sourceware.org/binutils/docs/as/s390-Formats.html
that e, prefix is for 16B opcodes?

e is an instruction format as specified by the architecture. See http://publibfp.dhe.ibm.com/epubs/pdf/a227832c.pdf
without any parameters.
Normally RR would be the right thing for MVCL, but since we try to build an invalid opcode without the assembler noticing (ab)using e seem like a safer approach.

Quoted 4 lines

LGTM, thanks again.

added those and added my RB. applied to the s390 tree. Thanks

..

Quoted 10 lines
Quoted 10 lines

arch/s390/lib/test_unwind.c static noinline int unwindme_func4(struct unwindme *u) · line 179

Quoted 10 lines
arch/s390/lib/test_unwind.c +1 -1
index cfc5f5557c06..d342bc884b94 100644
@@ -176,7 +176,7 @@ static noinline int unwindme_func4(struct unwindme *u)
                  * trigger specification exception
                  */
                 asm volatile(
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */

Sorry, I disagree with this. As you said above rr would be the correct format for this instruction. If we go for the e format then we should also use an instruction with e format.
Which in this case would simply be an illegal opcode, which would be sufficient for what this code is good for: ".insn e,0x0000".

Plus a fixup of the comment above, since this would generate an operation insteand of a specification exception. Just a generic "exception" would be good enough for the comment.

#5 lore raw

Am 19.11.21 um 11:54 schrieb Heiko Carstens:

Quoted 19 lines · Christian Borntraeger
Quoted 17 lines
Quoted 4 lines

So if I understand
https://sourceware.org/binutils/docs/as/s390-Directives.html#s390-Directives
https://sourceware.org/binutils/docs/as/s390-Formats.html
that e, prefix is for 16B opcodes?

e is an instruction format as specified by the architecture. See http://publibfp.dhe.ibm.com/epubs/pdf/a227832c.pdf
without any parameters.
Normally RR would be the right thing for MVCL, but since we try to build an invalid opcode without the assembler noticing (ab)using e seem like a safer approach.

Quoted 4 lines

LGTM, thanks again.

added those and added my RB. applied to the s390 tree. Thanks

..

Quoted 16 lines
Quoted 10 lines
Quoted 10 lines

arch/s390/lib/test_unwind.c static noinline int unwindme_func4(struct unwindme *u) · line 179

Quoted 10 lines
arch/s390/lib/test_unwind.c +1 -1
index cfc5f5557c06..d342bc884b94 100644
@@ -176,7 +176,7 @@ static noinline int unwindme_func4(struct unwindme *u)
                   * trigger specification exception
                   */
                  asm volatile(
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */

Sorry, I disagree with this. As you said above rr would be the correct format for this instruction. If we go for the e format then we should also use an instruction with e format.
Which in this case would simply be an illegal opcode, which would be sufficient for what this code is good for: ".insn e,0x0000".

Why not simply use .short then?

maintainer
#6 lore raw
Quoted 10 lines · Christian Borntraeger
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

arch/s390/lib/test_unwind.c static noinline int unwindme_func4(struct unwindme *u) · line 179

Quoted 2 lines
arch/s390/lib/test_unwind.c +1 -1
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */

Sorry, I disagree with this. As you said above rr would be the correct format for this instruction. If we go for the e format then we should also use an instruction with e format.
Which in this case would simply be an illegal opcode, which would be sufficient for what this code is good for: ".insn e,0x0000".

Why not simply use .short then?

.short bypasses all sanity checks while .insn does not, so I think that should be preferred. But I don't care too much.

#7 lore raw

Am 19.11.21 um 12:09 schrieb Heiko Carstens:

Quoted 10 lines · Christian Borntraeger
Quoted 9 lines
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

arch/s390/lib/test_unwind.c static noinline int unwindme_func4(struct unwindme *u) · line 179

Quoted 2 lines
arch/s390/lib/test_unwind.c +1 -1
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */

Sorry, I disagree with this. As you said above rr would be the correct format for this instruction. If we go for the e format then we should also use an instruction with e format.
Which in this case would simply be an illegal opcode, which would be sufficient for what this code is good for: ".insn e,0x0000".

Quoted 4 lines
Quoted 1 line

Why not simply use .short then?

.short bypasses all sanity checks while .insn does not, so I think that should be preferred. But I don't care too much.

Heiko,
I am fine with ".insn e,0x0000" and the a changed comment that changes "specification exception" to "operation exception". Do you want Ilie to resend or simply fixup?

maintainer
#8 lore raw
Quoted 13 lines · Christian Borntraeger

Am 19.11.21 um 12:09 schrieb Heiko Carstens:

Quoted 12 lines · Christian Borntraeger
Quoted 10 lines
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

Review

Quoted 2 lines
(quoted diff) +1 -1
-                       "       mvcl    %%r1,%%r1\n"
+                       "       .insn e,0x0e11\n"       /* mvcl %%r1,%%r1" */

Sorry, I disagree with this. As you said above rr would be the correct format for this instruction. If we go for the e format then we should also use an instruction with e format.
Which in this case would simply be an illegal opcode, which would be sufficient for what this code is good for: ".insn e,0x0000".

Why not simply use .short then?

Quoted 7 lines
Quoted 2 lines

.short bypasses all sanity checks while .insn does not, so I think that should be preferred. But I don't care too much.

Heiko,
I am fine with ".insn e,0x0000" and the a changed comment that changes "specification exception" to "operation exception". Do you want Ilie to resend or simply fixup?

I'll simply change it. Let's don't spend more time on this.