Re: [PATCH tip/core/rcu 0/4] RCU debug-objects improvements

Re: [PATCH tip/core/rcu 0/4] RCU debug-objects improvements

[Sedat Dilek]

[-- Attachment #1: Type: text/plain, Size: 1564 bytes --]

[ QUOTE ]

Hello!

RCU's current debug-objects fixups are complex and heavyweight, plus the
fixups are not complete, because even with the fixups, RCU's callback
lists can still be corrupted.  This series reworks RCU's use of the
debug-objects facility and the facility itself to allow improved handling
of things like duplicate invocations of call_rcu().  This series includes:

1.	Simplify RCU's debug-objects fixups.

2.	Make debug_object_activate() return status.

3.	Make __call_rcu() leak callbacks when debug_object_activate()
	notes suspicious activity.  Leaking has its problems, but it
	is far better than silent corruption of RCU's data structures.

4.	Add a new object_debug option to rcutorture to allow
	deterministic testing of #3 above.

							Thanx, Paul


 b/include/linux/debugobjects.h |    6 +-
 b/kernel/rcu.h                 |   10 ++--
 b/kernel/rcupdate.c            |  100 -----------------------------------------
 b/kernel/rcutorture.c          |   37 +++++++++++++++
 b/kernel/rcutree.c             |   14 +++++
 b/lib/debugobjects.c           |   20 +++++---
 6 files changed, 74 insertions(+), 113 deletions(-)

[ /QUOTE ]

I applied your patchset against the "buggy" Linux-Next (next-20130423)
with none of the ipc-sem-next patches (even w/o the suggested Linus
patch) to see if I see more debug-outputs,

I have attached my syslog while building my kernel.
Please, have a look!

Thanks for the patches.

Feel free to add to all of the 4 patches my Tested-by.

- Sedat -

[1] http://marc.info/?l=linux-kernel&m=136676361101735&w=2

[-- Attachment #2: syslog.txt.gz --]
[-- Type: application/x-gzip, Size: 54126 bytes --]

[-- Attachment #3: syslog.txt.gz.sha256sum --]
[-- Type: application/octet-stream, Size: 80 bytes --]

5fcc17106d02a9ec0b936543cf2a7b2aa57ad2b1efdbdcb5c7b1d7c9a7248f2c  syslog.txt.gz

Re: [PATCH tip/core/rcu 0/4] RCU debug-objects improvements

[Paul E. McKenney]

On Wed, Apr 24, 2013 at 03:45:20AM +0200, Sedat Dilek wrote:
> [ QUOTE ]
> 
> Hello!
> 
> RCU's current debug-objects fixups are complex and heavyweight, plus the
> fixups are not complete, because even with the fixups, RCU's callback
> lists can still be corrupted.  This series reworks RCU's use of the
> debug-objects facility and the facility itself to allow improved handling
> of things like duplicate invocations of call_rcu().  This series includes:
> 
> 1.	Simplify RCU's debug-objects fixups.
> 
> 2.	Make debug_object_activate() return status.
> 
> 3.	Make __call_rcu() leak callbacks when debug_object_activate()
> 	notes suspicious activity.  Leaking has its problems, but it
> 	is far better than silent corruption of RCU's data structures.
> 
> 4.	Add a new object_debug option to rcutorture to allow
> 	deterministic testing of #3 above.
> 
> 							Thanx, Paul
> 
> 
>  b/include/linux/debugobjects.h |    6 +-
>  b/kernel/rcu.h                 |   10 ++--
>  b/kernel/rcupdate.c            |  100 -----------------------------------------
>  b/kernel/rcutorture.c          |   37 +++++++++++++++
>  b/kernel/rcutree.c             |   14 +++++
>  b/lib/debugobjects.c           |   20 +++++---
>  6 files changed, 74 insertions(+), 113 deletions(-)
> 
> [ /QUOTE ]
> 
> I applied your patchset against the "buggy" Linux-Next (next-20130423)
> with none of the ipc-sem-next patches (even w/o the suggested Linus
> patch) to see if I see more debug-outputs,
> 
> I have attached my syslog while building my kernel.
> Please, have a look!

The warnings and the leaking of the offending callback are all present,
thank you for testing this!

> Thanks for the patches.
> 
> Feel free to add to all of the 4 patches my Tested-by.

Will do!

						Thanx, Paul

> - Sedat -
> 
> [1] http://marc.info/?l=linux-kernel&m=136676361101735&w=2

[PATCH tip/core/rcu 0/4] RCU debug-objects improvements

[Paul E. McKenney]

Hello!

RCU's current debug-objects fixups are complex and heavyweight, plus the
fixups are not complete, because even with the fixups, RCU's callback
lists can still be corrupted.  This series reworks RCU's use of the
debug-objects facility and the facility itself to allow improved handling
of things like duplicate invocations of call_rcu().  This series includes:

1.	Simplify RCU's debug-objects fixups.

2.	Make debug_object_activate() return status.

3.	Make __call_rcu() leak callbacks when debug_object_activate()
	notes suspicious activity.  Leaking has its problems, but it
	is far better than silent corruption of RCU's data structures.

4.	Add a new object_debug option to rcutorture to allow
	deterministic testing of #3 above.

							Thanx, Paul


 b/include/linux/debugobjects.h |    6 +-
 b/kernel/rcu.h                 |   10 ++--
 b/kernel/rcupdate.c            |  100 -----------------------------------------
 b/kernel/rcutorture.c          |   37 +++++++++++++++
 b/kernel/rcutree.c             |   14 +++++
 b/lib/debugobjects.c           |   20 +++++---
 6 files changed, 74 insertions(+), 113 deletions(-)