mbox series

[0/3] crypto/rcu: suppress unnecessary CPU stall warnings

Message ID 20221219202910.3063036-1-elliott@hpe.com
Headers show
Series crypto/rcu: suppress unnecessary CPU stall warnings | expand

Message

Elliott, Robert (Servers) Dec. 19, 2022, 8:29 p.m. UTC 
Suppress RCU CPU stall warnings during tcrypt speed tests.

Robert Elliott (3):
  rcu: genericize RCU stall suppression functions
  rcu: print first CPU on expedited stall line
  crypto: tcrypt - suppress RCU stall warnings during speed tests

 .../RCU/Design/Requirements/Requirements.rst  |   6 +-
 Documentation/RCU/stallwarn.rst               |  19 ++-
 arch/parisc/kernel/process.c                  |   2 +-
 crypto/tcrypt.c                               | 113 +++++++++++++++---
 drivers/tty/sysrq.c                           |   4 +-
 include/linux/rcupdate.h                      |   8 +-
 kernel/rcu/rcu.h                              |  11 +-
 kernel/rcu/tree_exp.h                         |  22 ++--
 kernel/rcu/tree_stall.h                       |  18 +--
 kernel/rcu/update.c                           |  11 +-
 10 files changed, 160 insertions(+), 54 deletions(-)

Comments

Herbert Xu Dec. 30, 2022, 8:39 a.m. UTC | #1 
On Mon, Dec 19, 2022 at 02:29:10PM -0600, Robert Elliott wrote:
> Suppress RCU CPU stall warnings while running speed tests.
> 
> The tcrypt module is intended only for developer usage, so
> RCU stalls induced by those tests are not necessarily representative
> of real problems.
> 
> Speed tests need to disable interrupts or preemption to get results
> that are not distorted by such interruptions. This triggers more
> RCU stalls than normal invocations of the crypto functions.

Where do we disable interrupts? That would seem to break the
use of jiffies since it won't get updated anymore.

Which particular test is still triggering the RCU warnings after
fixing the hash problems that you've already identified?

Cheers,