2022-12-01 14:51:28

by Oleg Nesterov

Subject: [PATCH] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

From: Denys Vlasenko <[email protected]>

Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
function to reserve extra space for hot-patching, and currently it is not
possible to probe these functions because branch_setup_xol_ops() wrongly
nacks NOP with REP prefix.

Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
Reported-by: Seiji Nishikawa <[email protected]>
Signed-off-by: Oleg Nesterov <[email protected]>
---
arch/x86/kernel/uprobes.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index b63cf8f7745e..6c07f6daaa22 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -722,8 +722,9 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
switch (opc1) {
case 0xeb: /* jmp 8 */
case 0xe9: /* jmp 32 */
- case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
break;
+ case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
+ goto setup;

case 0xe8: /* call relative */
branch_clear_offset(auprobe, insn);
@@ -753,6 +754,7 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
return -ENOTSUPP;
}

+setup:
auprobe->branch.opc1 = opc1;
auprobe->branch.ilen = insn->length;
auprobe->branch.offs = insn->immediate.value;
--
2.25.1.362.g51ebf55

2022-12-01 22:42:28

by Masami Hiramatsu

Subject: Re: [PATCH] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

On Thu, 1 Dec 2022 15:26:44 +0100
Oleg Nesterov <[email protected]> wrote:

> From: Denys Vlasenko <[email protected]>
>
> Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
> function to reserve extra space for hot-patching, and currently it is not
> possible to probe these functions because branch_setup_xol_ops() wrongly
> nacks NOP with REP prefix.

Looks good to me.

Acked-by: Masami Hiramatsu (Google) <[email protected]>

Thank you!

>
> Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
> Reported-by: Seiji Nishikawa <[email protected]>
> Signed-off-by: Oleg Nesterov <[email protected]>
> ---
> arch/x86/kernel/uprobes.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
> index b63cf8f7745e..6c07f6daaa22 100644
> --- a/arch/x86/kernel/uprobes.c
> +++ b/arch/x86/kernel/uprobes.c
> @@ -722,8 +722,9 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
> switch (opc1) {
> case 0xeb: /* jmp 8 */
> case 0xe9: /* jmp 32 */
> - case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
> break;
> + case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
> + goto setup;
>
> case 0xe8: /* call relative */
> branch_clear_offset(auprobe, insn);
> @@ -753,6 +754,7 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
> return -ENOTSUPP;
> }
>
> +setup:
> auprobe->branch.opc1 = opc1;
> auprobe->branch.ilen = insn->length;
> auprobe->branch.offs = insn->immediate.value;
> --
> 2.25.1.362.g51ebf55
>
>


--
Masami Hiramatsu (Google) <[email protected]>

2022-12-02 12:11:08

by Thomas Gleixner

Subject: Re: [PATCH] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

On Thu, Dec 01 2022 at 15:26, Oleg Nesterov wrote:

> From: Denys Vlasenko <[email protected]>
>
> Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
> function to reserve extra space for hot-patching, and currently it is not
> possible to probe these functions because branch_setup_xol_ops() wrongly
> nacks NOP with REP prefix.
>
> Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
> Reported-by: Seiji Nishikawa <[email protected]>
> Signed-off-by: Oleg Nesterov <[email protected]>

The SOB chain is invalid as it lacks the author's signoff.

2022-12-02 13:56:09

by Oleg Nesterov

Subject: Re: [PATCH] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

On 12/02, Thomas Gleixner wrote:
>
> On Thu, Dec 01 2022 at 15:26, Oleg Nesterov wrote:
>
> > From: Denys Vlasenko <[email protected]>
> >
> > Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
> > function to reserve extra space for hot-patching, and currently it is not
> > possible to probe these functions because branch_setup_xol_ops() wrongly
> > nacks NOP with REP prefix.
> >
> > Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
> > Reported-by: Seiji Nishikawa <[email protected]>
> > Signed-off-by: Oleg Nesterov <[email protected]>
>
> The SOB chain is invalid as it lacks the author's signoff.

Denys, can we have your signed-off-by?

2022-12-04 17:56:43

by Oleg Nesterov

Subject: Re: [PATCH] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

On 12/02, Oleg Nesterov wrote:
>
> On 12/02, Thomas Gleixner wrote:
> >
> > On Thu, Dec 01 2022 at 15:26, Oleg Nesterov wrote:
> >
> > > From: Denys Vlasenko <[email protected]>
> > >
> > > Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
> > > function to reserve extra space for hot-patching, and currently it is not
> > > possible to probe these functions because branch_setup_xol_ops() wrongly
> > > nacks NOP with REP prefix.
> > >
> > > Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
> > > Reported-by: Seiji Nishikawa <[email protected]>
> > > Signed-off-by: Oleg Nesterov <[email protected]>
> >
> > The SOB chain is invalid as it lacks the author's signoff.
>
> Denys, can we have your signed-off-by?

OK, let me turn "From" into "Suggested-by" if this can help to merge this fix.

After all, the patch is really simple, I don't think Denys will object when he
returns.

I am sending v2 with Masami's ack applied.

Oleg.

2022-12-04 18:32:05

by Oleg Nesterov

Subject: [PATCH v2] uprobes/x86: allow to probe a "nop" insn with 0x66 prefix

Intel icc -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
function to reserve extra space for hot-patching, and currently it is not
possible to probe these functions because branch_setup_xol_ops() wrongly
nacks NOP with REP prefix.

Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
Reported-by: Seiji Nishikawa <[email protected]>
Suggested-by: Denys Vlasenko <[email protected]>
Acked-by: Masami Hiramatsu (Google) <[email protected]>
Signed-off-by: Oleg Nesterov <[email protected]>
---
arch/x86/kernel/uprobes.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index b63cf8f7745e..6c07f6daaa22 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -722,8 +722,9 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
switch (opc1) {
case 0xeb: /* jmp 8 */
case 0xe9: /* jmp 32 */
- case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
break;
+ case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
+ goto setup;

case 0xe8: /* call relative */
branch_clear_offset(auprobe, insn);
@@ -753,6 +754,7 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
return -ENOTSUPP;
}

+setup:
auprobe->branch.opc1 = opc1;
auprobe->branch.ilen = insn->length;
auprobe->branch.offs = insn->immediate.value;
--
2.25.1.362.g51ebf55

2022-12-05 11:12:07

by tip-bot2 for Alexey Makhalov

Subject: [tip: x86/urgent] uprobes/x86: Allow to probe a NOP instruction with 0x66 prefix

The following commit has been merged into the x86/urgent branch of tip:

Commit-ID: cefa72129e45313655d53a065b8055aaeb01a0c9
Gitweb: https://git.kernel.org/tip/cefa72129e45313655d53a065b8055aaeb01a0c9
Author: Oleg Nesterov <[email protected]>
AuthorDate: Sun, 04 Dec 2022 18:39:33 +01:00
Committer: Thomas Gleixner <[email protected]>
CommitterDate: Mon, 05 Dec 2022 11:55:18 +01:00

uprobes/x86: Allow to probe a NOP instruction with 0x66 prefix

Intel ICC -hotpatch inserts 2-byte "0x66 0x90" NOP at the start of each
function to reserve extra space for hot-patching, and currently it is not
possible to probe these functions because branch_setup_xol_ops() wrongly
rejects NOP with REP prefix as it treats them like word-sized branch
instructions.

Fixes: 250bbd12c2fe ("uprobes/x86: Refuse to attach uprobe to "word-sized" branch insns")
Reported-by: Seiji Nishikawa <[email protected]>
Suggested-by: Denys Vlasenko <[email protected]>
Signed-off-by: Oleg Nesterov <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Acked-by: Masami Hiramatsu (Google) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

---
arch/x86/kernel/uprobes.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index b63cf8f..6c07f6d 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -722,8 +722,9 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
switch (opc1) {
case 0xeb: /* jmp 8 */
case 0xe9: /* jmp 32 */
- case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
break;
+ case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
+ goto setup;

case 0xe8: /* call relative */
branch_clear_offset(auprobe, insn);
@@ -753,6 +754,7 @@ static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn)
return -ENOTSUPP;
}

+setup:
auprobe->branch.opc1 = opc1;
auprobe->branch.ilen = insn->length;
auprobe->branch.offs = insn->immediate.value;