[PATCH] ANDROID: binder: correct the cmd print for BINDER_WORK_RETURN_ERROR

[PATCH] ANDROID: binder: correct the cmd print for BINDER_WORK_RETURN_ERROR

[宋金时]

In case of the BINDER_WORK_RETURN_ERROR the cmd is no assignment,
so it's value will be old value or dirty value, before modifying
e->cmd, assign the value of the e->cmd to cmd to ensure the correct
print of binder_stat_br.

Signed-off-by: songjinshi <songjinshi@xiaomi.com>
---

diff --git a/drivers/android/binder.c b/drivers/android/binder.c
index 4ffc84d..acd7d44a 100644
--- a/drivers/android/binder.c
+++ b/drivers/android/binder.c
@@ -4086,6 +4086,7 @@
 			struct binder_error *e = container_of(
 					w, struct binder_error, work);
 
+			cmd = e->cmd;
 			WARN_ON(e->cmd == BR_OK);
 			binder_inner_proc_unlock(proc);
 			if (put_user(e->cmd, (uint32_t __user *)ptr))

Re: [PATCH] ANDROID: binder: correct the cmd print for BINDER_WORK_RETURN_ERROR

[Todd Kjos]

On Wed, May 9, 2018 at 12:31 AM 宋金时 <songjinshi@xiaomi.com> wrote:

> In case of the BINDER_WORK_RETURN_ERROR the cmd is no assignment,
> so it's value will be old value or dirty value, before modifying
> e->cmd, assign the value of the e->cmd to cmd to ensure the correct
> print of binder_stat_br.


That's not the problem... the problem is that e->cmd is unconditionally set
to BR_OK prior to the call to binder_stat_br(proc, thread, e->cmd) so the
reported cmd is always BR_OK.


> Signed-off-by: songjinshi <songjinshi@xiaomi.com>
> ---

> diff --git a/drivers/android/binder.c b/drivers/android/binder.c
> index 4ffc84d..acd7d44a 100644
> --- a/drivers/android/binder.c
> +++ b/drivers/android/binder.c
> @@ -4086,6 +4086,7 @@
>                          struct binder_error *e = container_of(
>                                          w, struct binder_error, work);

> +                       cmd = e->cmd;
>                          WARN_ON(e->cmd == BR_OK);
>                          binder_inner_proc_unlock(proc);
>                          if (put_user(e->cmd, (uint32_t __user *)ptr))


Don't you also need to use cmd in the call to binder_state_br() below?

-                       binder_stat_br(proc, thread, e->cmd);
+                       binder_stat_br(proc, thread, cmd);

-Todd

答复: [PATCH] ANDROID: binder: correct the cmd print for BINDER_WORK_RETURN_ERROR

[宋金时]

OK，this patch is based on the downstream android kernel4.4 version, not yet include the latest changes, let me update the patch to base on the latest upstream, Tks.

________________________________________
发件人: Todd Kjos <tkjos@google.com>
发送时间: 2018年5月10日 2:08
收件人: 宋金时
抄送: Greg Kroah-Hartman; Martijn Coenen; Arve Hjønnevåg; LKML
主题: Re: [PATCH] ANDROID: binder: correct the cmd print for BINDER_WORK_RETURN_ERROR

On Wed, May 9, 2018 at 12:31 AM 宋金时 <songjinshi@xiaomi.com> wrote:

> In case of the BINDER_WORK_RETURN_ERROR the cmd is no assignment,
> so it's value will be old value or dirty value, before modifying
> e->cmd, assign the value of the e->cmd to cmd to ensure the correct
> print of binder_stat_br.


That's not the problem... the problem is that e->cmd is unconditionally set
to BR_OK prior to the call to binder_stat_br(proc, thread, e->cmd) so the
reported cmd is always BR_OK.


> Signed-off-by: songjinshi <songjinshi@xiaomi.com>
> ---

> diff --git a/drivers/android/binder.c b/drivers/android/binder.c
> index 4ffc84d..acd7d44a 100644
> --- a/drivers/android/binder.c
> +++ b/drivers/android/binder.c
> @@ -4086,6 +4086,7 @@
>                          struct binder_error *e = container_of(
>                                          w, struct binder_error, work);

> +                       cmd = e->cmd;
>                          WARN_ON(e->cmd == BR_OK);
>                          binder_inner_proc_unlock(proc);
>                          if (put_user(e->cmd, (uint32_t __user *)ptr))


Don't you also need to use cmd in the call to binder_state_br() below?

-                       binder_stat_br(proc, thread, e->cmd);
+                       binder_stat_br(proc, thread, cmd);

-Todd