2021-01-23 11:36:42

by Yuxuan Shui

[permalink] [raw]
Subject: [PATCH] ptrace: restore the previous single step reporting behavior

Commit 64eb35f701f04b30706e21d1b02636b5d31a37d2 changed when single step
is reported.

Specifically, the report_single_step is changed so that single steps are
only reported when both SYSCALL_EMU and _TIF_SINGLESTEP are set, while
previously they are reported when _TIF_SINGLESTEP is set without
_TIF_SYSCALL_EMU being set.

This behavior change breaks rr [1]

This commit restores the old behavior.

[1]: https://github.com/rr-debugger/rr/issues/2793

Signed-off-by: Yuxuan Shui <[email protected]>
---
kernel/entry/common.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/entry/common.c b/kernel/entry/common.c
index 90533f34ea99c..ad3b17fcde782 100644
--- a/kernel/entry/common.c
+++ b/kernel/entry/common.c
@@ -220,7 +220,7 @@ static inline bool rep:ort_single_step(unsigned long work)
*/
static inline bool report_single_step(unsigned long work)
{
- if (!(work & SYSCALL_WORK_SYSCALL_EMU))
+ if (!!(work & SYSCALL_WORK_SYSCALL_EMU))
return false;

return !!(current_thread_info()->flags & _TIF_SINGLESTEP);
--
2.30.0


Attachments:
signature.asc (847.00 B)

2021-01-28 00:47:40

by Kyle Huey

[permalink] [raw]
Subject: Re: [PATCH] ptrace: restore the previous single step reporting behavior

Hey everyone,

This regression[0] has totally broken rr on 5.11. Could we get someone
to look at and merge Yuxuan's patch here?

- Kyle

[0] https://github.com/rr-debugger/rr/issues/2793

2021-01-28 00:49:50

by Gabriel Krisman Bertazi

[permalink] [raw]
Subject: Re: [PATCH] ptrace: restore the previous single step reporting behavior

Yuxuan Shui <[email protected]> writes:

> Commit 64eb35f701f04b30706e21d1b02636b5d31a37d2 changed when single step
> is reported.
>
> Specifically, the report_single_step is changed so that single steps are
> only reported when both SYSCALL_EMU and _TIF_SINGLESTEP are set, while
> previously they are reported when _TIF_SINGLESTEP is set without
> _TIF_SYSCALL_EMU being set.
>
> This behavior change breaks rr [1]
>
> This commit restores the old behavior.
>
> [1]: https://github.com/rr-debugger/rr/issues/2793
>
> Signed-off-by: Yuxuan Shui <[email protected]>

Looks correct to me.

To gather the right attention, you should directly CC the correct maintainers.

Fixes: 64eb35f701f0 ("ptrace: Migrate TIF_SYSCALL_EMU to use SYSCALL_WORK flag")
Reviewed-by: Gabriel Krisman Bertazi <[email protected]>

--
Gabriel Krisman Bertazi

2021-01-28 00:52:57

by Yuxuan Shui

[permalink] [raw]
Subject: Re: [PATCH] ptrace: restore the previous single step reporting behavior

Hi,

On Wed, Jan 27, 2021 at 11:55 PM Gabriel Krisman Bertazi
<[email protected]> wrote:
>
> Yuxuan Shui <[email protected]> writes:
>
> > Commit 64eb35f701f04b30706e21d1b02636b5d31a37d2 changed when single step
> > is reported.
> >
> > Specifically, the report_single_step is changed so that single steps are
> > only reported when both SYSCALL_EMU and _TIF_SINGLESTEP are set, while
> > previously they are reported when _TIF_SINGLESTEP is set without
> > _TIF_SYSCALL_EMU being set.
> >
> > This behavior change breaks rr [1]
> >
> > This commit restores the old behavior.
> >
> > [1]: https://github.com/rr-debugger/rr/issues/2793
> >
> > Signed-off-by: Yuxuan Shui <[email protected]>
>
> Looks correct to me.
>
> To gather the right attention, you should directly CC the correct maintainers.

Thanks, will do.

>
> Fixes: 64eb35f701f0 ("ptrace: Migrate TIF_SYSCALL_EMU to use SYSCALL_WORK flag")
> Reviewed-by: Gabriel Krisman Bertazi <[email protected]>
>
> --
> Gabriel Krisman Bertazi



--

Regards
Yuxuan Shui

2021-01-28 12:42:51

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH] ptrace: restore the previous single step reporting behavior

On Wed, Jan 27 2021 at 20:55, Gabriel Krisman Bertazi wrote:
> Yuxuan Shui <[email protected]> writes:
>
> To gather the right attention, you should directly CC the correct
> maintainers.

You could have cc'ed them on your reply ....

> Fixes: 64eb35f701f0 ("ptrace: Migrate TIF_SYSCALL_EMU to use SYSCALL_WORK flag")
> Reviewed-by: Gabriel Krisman Bertazi <[email protected]>

Subject: [tip: core/urgent] entry: Unbreak single step reporting behaviour

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

Commit-ID: 41c1a06d1d1544bed9692ba72a5692454eee1945
Gitweb: https://git.kernel.org/tip/41c1a06d1d1544bed9692ba72a5692454eee1945
Author: Yuxuan Shui <[email protected]>
AuthorDate: Sat, 23 Jan 2021 03:21:32 -08:00
Committer: Thomas Gleixner <[email protected]>
CommitterDate: Thu, 28 Jan 2021 13:46:55 +01:00

entry: Unbreak single step reporting behaviour

The move of TIF_SYSCALL_EMU to SYSCALL_WORK_SYSCALL_EMU broke single step
reporting. The original code reported the single step when TIF_SINGLESTEP
was set and TIF_SYSCALL_EMU was not set. The SYSCALL_WORK conversion got
the logic wrong and now the reporting only happens when both bits are set.

Restore the original behaviour.

[ tglx: Massaged changelog and dropped the pointless double negation ]

Fixes: 64eb35f701f0 ("ptrace: Migrate TIF_SYSCALL_EMU to use SYSCALL_WORK flag")
Signed-off-by: Yuxuan Shui <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Reviewed-by: Gabriel Krisman Bertazi <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
kernel/entry/common.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/entry/common.c b/kernel/entry/common.c
index 3783416..6dd82be 100644
--- a/kernel/entry/common.c
+++ b/kernel/entry/common.c
@@ -222,7 +222,7 @@ static inline bool report_single_step(unsigned long work)
*/
static inline bool report_single_step(unsigned long work)
{
- if (!(work & SYSCALL_WORK_SYSCALL_EMU))
+ if (work & SYSCALL_WORK_SYSCALL_EMU)
return false;

return !!(current_thread_info()->flags & _TIF_SINGLESTEP);

2021-01-28 20:19:59

by Kyle Huey

[permalink] [raw]
Subject: Re: [PATCH] ptrace: restore the previous single step reporting behavior

On Thu, Jan 28, 2021 at 11:10 AM Linus Torvalds
<[email protected]> wrote:
> You should have pointed to the actual patch.

Sorry, I broke the reply threading in my mail client.

- Kyle