[PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[Tom de Vries]

Hi,

Occassionally I run into the following assert:
...
(gdb) PASS: gdb.multi/multi-target-continue.exp: inferior 5
Remote debugging from host ::1, port 49990^M
Process multi-target-continue created; pid = 31241^M
src/gdb/remote-notif.c:113: internal-error: \
  void remote_async_get_pending_events_handler(gdb_client_data): \
  Assertion `target_is_non_stop_p ()' failed.^M
...

The assert checks target_is_non_stop_p, which is related to the current
target.

Fix this by changing the assert such that it checks non-stopness related to
the event it's handling.

Tested on x86_64-linux.

Any comments?

Thanks,
- Tom

[gdb] Fix assert in remote_async_get_pending_events_handler

2021-04-16  Simon Marchi  <simon.marchi@polymtl.ca>
	    Tom de Vries  <tdevries@suse.de>

	PR remote/27710
	* remote.c (remote_target_is_non_stop_p): New function.
	* remote.h (remote_target_is_non_stop_p): Declare.
	* remote-notif.c (remote_async_get_pending_events_handler): Fix assert
	to check non-stopness using notif_state->remote rather current target.

---
 gdb/remote-notif.c |  2 +-
 gdb/remote.c       | 11 +++++++++++
 gdb/remote.h       |  1 +
 3 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/gdb/remote-notif.c b/gdb/remote-notif.c
index 5a3e1395b76..4245015ae37 100644
--- a/gdb/remote-notif.c
+++ b/gdb/remote-notif.c
@@ -110,7 +110,7 @@ remote_async_get_pending_events_handler (gdb_client_data data)
 {
   remote_notif_state *notif_state = (remote_notif_state *) data;
   clear_async_event_handler (notif_state->get_pending_events_token);
-  gdb_assert (target_is_non_stop_p ());
+  gdb_assert (remote_target_is_non_stop_p (notif_state->remote));
   remote_notif_process (notif_state, NULL);
 }
 
diff --git a/gdb/remote.c b/gdb/remote.c
index 7429e1a86b3..2e365df9bba 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -14730,6 +14730,17 @@ remote_target::store_memtags (CORE_ADDR address, size_t len,
   return packet_check_result (rs->buf.data ()) == PACKET_OK;
 }
 
+/* Return true if remote target T is non-stop.  */
+
+bool
+remote_target_is_non_stop_p (remote_target *t)
+{
+  scoped_restore_current_thread restore_thread;
+  switch_to_target_no_thread (t);
+
+  return target_is_non_stop_p ();
+}
+
 #if GDB_SELF_TEST
 
 namespace selftests {
diff --git a/gdb/remote.h b/gdb/remote.h
index 18352ddb866..46bfa01fc79 100644
--- a/gdb/remote.h
+++ b/gdb/remote.h
@@ -77,4 +77,5 @@ extern int remote_register_number_and_offset (struct gdbarch *gdbarch,
 
 extern void remote_notif_get_pending_events (remote_target *remote,
 					     struct notif_client *np);
+extern bool remote_target_is_non_stop_p (remote_target *t);
 #endif

Re: [PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[Andrew Burgess]

* Tom de Vries <tdevries@suse.de> [2021-04-22 10:51:29 +0200]:

> Hi,
> 
> Occassionally I run into the following assert:
> ...
> (gdb) PASS: gdb.multi/multi-target-continue.exp: inferior 5
> Remote debugging from host ::1, port 49990^M
> Process multi-target-continue created; pid = 31241^M
> src/gdb/remote-notif.c:113: internal-error: \
>   void remote_async_get_pending_events_handler(gdb_client_data): \
>   Assertion `target_is_non_stop_p ()' failed.^M
> ...
> 
> The assert checks target_is_non_stop_p, which is related to the current
> target.
> 
> Fix this by changing the assert such that it checks non-stopness related to
> the event it's handling.
> 
> Tested on x86_64-linux.
> 
> Any comments?

This seems fine to me.  I wonder though if you considered converting
target_is_non_stop_p into a member function on target_ops?  If we did
then we would avoid having to switch targets just to ask this
question.  All of the helper functions that target_is_non_stop_p calls
are already available as member functions so there would be no
additional changes needed I think.

Just a thought.

Thanks,
Andrew

> 
> Thanks,
> - Tom
> 
> [gdb] Fix assert in remote_async_get_pending_events_handler
> 
> 2021-04-16  Simon Marchi  <simon.marchi@polymtl.ca>
> 	    Tom de Vries  <tdevries@suse.de>
> 
> 	PR remote/27710
> 	* remote.c (remote_target_is_non_stop_p): New function.
> 	* remote.h (remote_target_is_non_stop_p): Declare.
> 	* remote-notif.c (remote_async_get_pending_events_handler): Fix assert
> 	to check non-stopness using notif_state->remote rather current target.
> 
> ---
>  gdb/remote-notif.c |  2 +-
>  gdb/remote.c       | 11 +++++++++++
>  gdb/remote.h       |  1 +
>  3 files changed, 13 insertions(+), 1 deletion(-)
> 
> diff --git a/gdb/remote-notif.c b/gdb/remote-notif.c
> index 5a3e1395b76..4245015ae37 100644
> --- a/gdb/remote-notif.c
> +++ b/gdb/remote-notif.c
> @@ -110,7 +110,7 @@ remote_async_get_pending_events_handler (gdb_client_data data)
>  {
>    remote_notif_state *notif_state = (remote_notif_state *) data;
>    clear_async_event_handler (notif_state->get_pending_events_token);
> -  gdb_assert (target_is_non_stop_p ());
> +  gdb_assert (remote_target_is_non_stop_p (notif_state->remote));
>    remote_notif_process (notif_state, NULL);
>  }
>  
> diff --git a/gdb/remote.c b/gdb/remote.c
> index 7429e1a86b3..2e365df9bba 100644
> --- a/gdb/remote.c
> +++ b/gdb/remote.c
> @@ -14730,6 +14730,17 @@ remote_target::store_memtags (CORE_ADDR address, size_t len,
>    return packet_check_result (rs->buf.data ()) == PACKET_OK;
>  }
>  
> +/* Return true if remote target T is non-stop.  */
> +
> +bool
> +remote_target_is_non_stop_p (remote_target *t)
> +{
> +  scoped_restore_current_thread restore_thread;
> +  switch_to_target_no_thread (t);
> +
> +  return target_is_non_stop_p ();
> +}
> +
>  #if GDB_SELF_TEST
>  
>  namespace selftests {
> diff --git a/gdb/remote.h b/gdb/remote.h
> index 18352ddb866..46bfa01fc79 100644
> --- a/gdb/remote.h
> +++ b/gdb/remote.h
> @@ -77,4 +77,5 @@ extern int remote_register_number_and_offset (struct gdbarch *gdbarch,
>  
>  extern void remote_notif_get_pending_events (remote_target *remote,
>  					     struct notif_client *np);
> +extern bool remote_target_is_non_stop_p (remote_target *t);
>  #endif

Re: [PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[Tom de Vries]

On 4/22/21 12:19 PM, Andrew Burgess wrote:
> * Tom de Vries <tdevries@suse.de> [2021-04-22 10:51:29 +0200]:
> 
>> Hi,
>>
>> Occassionally I run into the following assert:
>> ...
>> (gdb) PASS: gdb.multi/multi-target-continue.exp: inferior 5
>> Remote debugging from host ::1, port 49990^M
>> Process multi-target-continue created; pid = 31241^M
>> src/gdb/remote-notif.c:113: internal-error: \
>>   void remote_async_get_pending_events_handler(gdb_client_data): \
>>   Assertion `target_is_non_stop_p ()' failed.^M
>> ...
>>
>> The assert checks target_is_non_stop_p, which is related to the current
>> target.
>>
>> Fix this by changing the assert such that it checks non-stopness related to
>> the event it's handling.
>>
>> Tested on x86_64-linux.
>>
>> Any comments?
> 
> This seems fine to me.  I wonder though if you considered converting
> target_is_non_stop_p into a member function on target_ops?
> If we did
> then we would avoid having to switch targets just to ask this
> question.  All of the helper functions that target_is_non_stop_p calls
> are already available as member functions so there would be no
> additional changes needed I think.
> 

Um, I'm the one who ran into the problem, Simon is the one who came up
with the fix, so I guess this is a question for him.  I'm afraid I'm not
familiar with this code at all.

Thanks,
- Tom

> Just a thought.
> 
> Thanks,
> Andrew
> 
>>
>> Thanks,
>> - Tom
>>
>> [gdb] Fix assert in remote_async_get_pending_events_handler
>>
>> 2021-04-16  Simon Marchi  <simon.marchi@polymtl.ca>
>> 	    Tom de Vries  <tdevries@suse.de>
>>
>> 	PR remote/27710
>> 	* remote.c (remote_target_is_non_stop_p): New function.
>> 	* remote.h (remote_target_is_non_stop_p): Declare.
>> 	* remote-notif.c (remote_async_get_pending_events_handler): Fix assert
>> 	to check non-stopness using notif_state->remote rather current target.
>>
>> ---
>>  gdb/remote-notif.c |  2 +-
>>  gdb/remote.c       | 11 +++++++++++
>>  gdb/remote.h       |  1 +
>>  3 files changed, 13 insertions(+), 1 deletion(-)
>>
>> diff --git a/gdb/remote-notif.c b/gdb/remote-notif.c
>> index 5a3e1395b76..4245015ae37 100644
>> --- a/gdb/remote-notif.c
>> +++ b/gdb/remote-notif.c
>> @@ -110,7 +110,7 @@ remote_async_get_pending_events_handler (gdb_client_data data)
>>  {
>>    remote_notif_state *notif_state = (remote_notif_state *) data;
>>    clear_async_event_handler (notif_state->get_pending_events_token);
>> -  gdb_assert (target_is_non_stop_p ());
>> +  gdb_assert (remote_target_is_non_stop_p (notif_state->remote));
>>    remote_notif_process (notif_state, NULL);
>>  }
>>  
>> diff --git a/gdb/remote.c b/gdb/remote.c
>> index 7429e1a86b3..2e365df9bba 100644
>> --- a/gdb/remote.c
>> +++ b/gdb/remote.c
>> @@ -14730,6 +14730,17 @@ remote_target::store_memtags (CORE_ADDR address, size_t len,
>>    return packet_check_result (rs->buf.data ()) == PACKET_OK;
>>  }
>>  
>> +/* Return true if remote target T is non-stop.  */
>> +
>> +bool
>> +remote_target_is_non_stop_p (remote_target *t)
>> +{
>> +  scoped_restore_current_thread restore_thread;
>> +  switch_to_target_no_thread (t);
>> +
>> +  return target_is_non_stop_p ();
>> +}
>> +
>>  #if GDB_SELF_TEST
>>  
>>  namespace selftests {
>> diff --git a/gdb/remote.h b/gdb/remote.h
>> index 18352ddb866..46bfa01fc79 100644
>> --- a/gdb/remote.h
>> +++ b/gdb/remote.h
>> @@ -77,4 +77,5 @@ extern int remote_register_number_and_offset (struct gdbarch *gdbarch,
>>  
>>  extern void remote_notif_get_pending_events (remote_target *remote,
>>  					     struct notif_client *np);
>> +extern bool remote_target_is_non_stop_p (remote_target *t);
>>  #endif

Re: [PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[Simon Marchi]

On 2021-04-22 7:03 a.m., Tom de Vries wrote:
> On 4/22/21 12:19 PM, Andrew Burgess wrote:
>> * Tom de Vries <tdevries@suse.de> [2021-04-22 10:51:29 +0200]:
>>
>>> Hi,
>>>
>>> Occassionally I run into the following assert:
>>> ...
>>> (gdb) PASS: gdb.multi/multi-target-continue.exp: inferior 5
>>> Remote debugging from host ::1, port 49990^M
>>> Process multi-target-continue created; pid = 31241^M
>>> src/gdb/remote-notif.c:113: internal-error: \
>>>   void remote_async_get_pending_events_handler(gdb_client_data): \
>>>   Assertion `target_is_non_stop_p ()' failed.^M
>>> ...
>>>
>>> The assert checks target_is_non_stop_p, which is related to the current
>>> target.
>>>
>>> Fix this by changing the assert such that it checks non-stopness related to
>>> the event it's handling.
>>>
>>> Tested on x86_64-linux.
>>>
>>> Any comments?
>>
>> This seems fine to me.  I wonder though if you considered converting
>> target_is_non_stop_p into a member function on target_ops?
>> If we did
>> then we would avoid having to switch targets just to ask this
>> question.  All of the helper functions that target_is_non_stop_p calls
>> are already available as member functions so there would be no
>> additional changes needed I think.
>>
> 
> Um, I'm the one who ran into the problem, Simon is the one who came up
> with the fix, so I guess this is a question for him.  I'm afraid I'm not
> familiar with this code at all.

This was proposed in the bug as well:

  https://sourceware.org/bugzilla/show_bug.cgi?id=27710#c16

I am not against it, but I think Tom's patch is OK, given it follows
current practices.

Simon

Re: [PATCH][gdb] Fix assert in remote_async_get_pending_events_handler

[Andrew Burgess]

* Simon Marchi <simon.marchi@polymtl.ca> [2021-04-22 09:33:28 -0400]:

> On 2021-04-22 7:03 a.m., Tom de Vries wrote:
> > On 4/22/21 12:19 PM, Andrew Burgess wrote:
> >> * Tom de Vries <tdevries@suse.de> [2021-04-22 10:51:29 +0200]:
> >>
> >>> Hi,
> >>>
> >>> Occassionally I run into the following assert:
> >>> ...
> >>> (gdb) PASS: gdb.multi/multi-target-continue.exp: inferior 5
> >>> Remote debugging from host ::1, port 49990^M
> >>> Process multi-target-continue created; pid = 31241^M
> >>> src/gdb/remote-notif.c:113: internal-error: \
> >>>   void remote_async_get_pending_events_handler(gdb_client_data): \
> >>>   Assertion `target_is_non_stop_p ()' failed.^M
> >>> ...
> >>>
> >>> The assert checks target_is_non_stop_p, which is related to the current
> >>> target.
> >>>
> >>> Fix this by changing the assert such that it checks non-stopness related to
> >>> the event it's handling.
> >>>
> >>> Tested on x86_64-linux.
> >>>
> >>> Any comments?
> >>
> >> This seems fine to me.  I wonder though if you considered converting
> >> target_is_non_stop_p into a member function on target_ops?
> >> If we did
> >> then we would avoid having to switch targets just to ask this
> >> question.  All of the helper functions that target_is_non_stop_p calls
> >> are already available as member functions so there would be no
> >> additional changes needed I think.
> >>
> > 
> > Um, I'm the one who ran into the problem, Simon is the one who came up
> > with the fix, so I guess this is a question for him.  I'm afraid I'm not
> > familiar with this code at all.
> 
> This was proposed in the bug as well:
> 
>   https://sourceware.org/bugzilla/show_bug.cgi?id=27710#c16
> 
> I am not against it, but I think Tom's patch is OK, given it follows
> current practices.

I agree, I'm certainly not blocking the patch.

Thanks,
Andrew