[PATCH v3 0/7] RCU changes for PREEMPT_LAZY

Ankur Arora posted 7 patches 1 year ago
include/linux/rcupdate.h     |  2 +-
include/linux/rcutree.h      |  2 +-
include/linux/srcutiny.h     |  2 +-
kernel/rcu/Kconfig           |  4 ++--
kernel/rcu/srcutiny.c        | 14 +++++++-------
kernel/rcu/tree_plugin.h     | 22 +++++++++++++++++-----
kernel/sched/core.c          |  4 +++-
kernel/trace/trace_osnoise.c | 32 +++++++++++++++-----------------
8 files changed, 47 insertions(+), 35 deletions(-)
[PATCH v3 0/7] RCU changes for PREEMPT_LAZY
Posted by Ankur Arora 1 year ago
This series adds RCU bits for lazy preemption.

The problem addressed is that pre-PREEMPT_LAZY, PREEMPTION=y implied
PREEMPT_RCU=y. With PREEMPT_LAZY, that's no longer true. 

That's because PREEMPT_RCU makes some trade-offs to optimize for
latency as opposed to throughput, and configurations with limited
preemption might prefer the stronger forward-progress guarantees of
PREEMPT_RCU=n.

Accordingly, with standalone PREEMPT_LAZY (much like PREEMPT_NONE,
PREEMPT_VOLUNTARY) we want to use PREEMPT_RCU=n. And, when used in
conjunction with PREEMPT_DYNAMIC, we continue to use PREEMPT_RCU=y.

Patches 1-3  are cleanup patches:
  "rcu: fix header guard for rcu_all_qs()"
  "rcu: rename PREEMPT_AUTO to PREEMPT_LAZY"
  "sched: update __cond_resched comment about RCU quiescent states"

Patch 4,
  "rcu: handle unstable rdp in rcu_read_unlock_strict()"

handles a latent RCU bug rcu_report_qs_rdp() could be called with
an unstable rdp.

Patches 5 and 6,
  "rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y"
  "osnoise: provide quiescent states"

handle quiescent states for the (PREEMPT_LAZY=y, PREEMPT_RCU=n)
configuration.

And, finally patch 7, "rcu: limit PREEMPT_RCU configurations",
explicitly limits PREEMPT_RCU=y to the PREEMPT_DYNAMIC or the latency
oriented models.

Changelog:

v3:
 - moved patch-3 to be the last one in the series (suggested by Sebastian)
 - added "rcu: handle unstable rdp in rcu_read_unlock_strict()"
   (suggested by Frederic Weisbecker).
 - switched to a more robust check in rcu_flavor_sched_clock_irq()
   (suggested by Frederic Weisbecker).
 - simplified check in osnoise (suggested by Frederic Weisbecker).
 - dropped an unrelated scheduler patch.

v2:
  - fixup incorrect usage of tif_need_resched_lazy() (comment from
    from Sebastian Andrzej Siewior)
  - massaged the commit messages a bit
  - drops the powerpc support for PREEMPT_LAZY as that was orthogonal
    to this series (Shrikanth will send that out separately.)

Please review.

Ankur Arora (7):
  rcu: fix header guard for rcu_all_qs()
  rcu: rename PREEMPT_AUTO to PREEMPT_LAZY
  sched: update __cond_resched comment about RCU quiescent states
  rcu: handle unstable rdp in rcu_read_unlock_strict()
  rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y
  osnoise: provide quiescent states
  rcu: limit PREEMPT_RCU configurations

 include/linux/rcupdate.h     |  2 +-
 include/linux/rcutree.h      |  2 +-
 include/linux/srcutiny.h     |  2 +-
 kernel/rcu/Kconfig           |  4 ++--
 kernel/rcu/srcutiny.c        | 14 +++++++-------
 kernel/rcu/tree_plugin.h     | 22 +++++++++++++++++-----
 kernel/sched/core.c          |  4 +++-
 kernel/trace/trace_osnoise.c | 32 +++++++++++++++-----------------
 8 files changed, 47 insertions(+), 35 deletions(-)

-- 
2.43.5
Re: [PATCH v3 0/7] RCU changes for PREEMPT_LAZY
Posted by Paul E. McKenney 11 months, 1 week ago
On Thu, Dec 12, 2024 at 08:06:51PM -0800, Ankur Arora wrote:
> This series adds RCU bits for lazy preemption.
> 
> The problem addressed is that pre-PREEMPT_LAZY, PREEMPTION=y implied
> PREEMPT_RCU=y. With PREEMPT_LAZY, that's no longer true. 
> 
> That's because PREEMPT_RCU makes some trade-offs to optimize for
> latency as opposed to throughput, and configurations with limited
> preemption might prefer the stronger forward-progress guarantees of
> PREEMPT_RCU=n.
> 
> Accordingly, with standalone PREEMPT_LAZY (much like PREEMPT_NONE,
> PREEMPT_VOLUNTARY) we want to use PREEMPT_RCU=n. And, when used in
> conjunction with PREEMPT_DYNAMIC, we continue to use PREEMPT_RCU=y.
> 
> Patches 1-3  are cleanup patches:
>   "rcu: fix header guard for rcu_all_qs()"
>   "rcu: rename PREEMPT_AUTO to PREEMPT_LAZY"
>   "sched: update __cond_resched comment about RCU quiescent states"
> 
> Patch 4,
>   "rcu: handle unstable rdp in rcu_read_unlock_strict()"
> 
> handles a latent RCU bug rcu_report_qs_rdp() could be called with
> an unstable rdp.
> 
> Patches 5 and 6,
>   "rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y"
>   "osnoise: provide quiescent states"
> 
> handle quiescent states for the (PREEMPT_LAZY=y, PREEMPT_RCU=n)
> configuration.
> 
> And, finally patch 7, "rcu: limit PREEMPT_RCU configurations",
> explicitly limits PREEMPT_RCU=y to the PREEMPT_DYNAMIC or the latency
> oriented models.

Pulled into my -rcu tree for further review and testing, with initial
tests passing.  Apologies for the delay!

							Thanx, Paul

> Changelog:
> 
> v3:
>  - moved patch-3 to be the last one in the series (suggested by Sebastian)
>  - added "rcu: handle unstable rdp in rcu_read_unlock_strict()"
>    (suggested by Frederic Weisbecker).
>  - switched to a more robust check in rcu_flavor_sched_clock_irq()
>    (suggested by Frederic Weisbecker).
>  - simplified check in osnoise (suggested by Frederic Weisbecker).
>  - dropped an unrelated scheduler patch.
> 
> v2:
>   - fixup incorrect usage of tif_need_resched_lazy() (comment from
>     from Sebastian Andrzej Siewior)
>   - massaged the commit messages a bit
>   - drops the powerpc support for PREEMPT_LAZY as that was orthogonal
>     to this series (Shrikanth will send that out separately.)
> 
> Please review.
> 
> Ankur Arora (7):
>   rcu: fix header guard for rcu_all_qs()
>   rcu: rename PREEMPT_AUTO to PREEMPT_LAZY
>   sched: update __cond_resched comment about RCU quiescent states
>   rcu: handle unstable rdp in rcu_read_unlock_strict()
>   rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y
>   osnoise: provide quiescent states
>   rcu: limit PREEMPT_RCU configurations
> 
>  include/linux/rcupdate.h     |  2 +-
>  include/linux/rcutree.h      |  2 +-
>  include/linux/srcutiny.h     |  2 +-
>  kernel/rcu/Kconfig           |  4 ++--
>  kernel/rcu/srcutiny.c        | 14 +++++++-------
>  kernel/rcu/tree_plugin.h     | 22 +++++++++++++++++-----
>  kernel/sched/core.c          |  4 +++-
>  kernel/trace/trace_osnoise.c | 32 +++++++++++++++-----------------
>  8 files changed, 47 insertions(+), 35 deletions(-)
> 
> -- 
> 2.43.5
>
Re: [PATCH v3 0/7] RCU changes for PREEMPT_LAZY
Posted by Ankur Arora 11 months, 1 week ago
Paul E. McKenney <paulmck@kernel.org> writes:

> On Thu, Dec 12, 2024 at 08:06:51PM -0800, Ankur Arora wrote:
>> This series adds RCU bits for lazy preemption.
>>
>> The problem addressed is that pre-PREEMPT_LAZY, PREEMPTION=y implied
>> PREEMPT_RCU=y. With PREEMPT_LAZY, that's no longer true.
>>
>> That's because PREEMPT_RCU makes some trade-offs to optimize for
>> latency as opposed to throughput, and configurations with limited
>> preemption might prefer the stronger forward-progress guarantees of
>> PREEMPT_RCU=n.
>>
>> Accordingly, with standalone PREEMPT_LAZY (much like PREEMPT_NONE,
>> PREEMPT_VOLUNTARY) we want to use PREEMPT_RCU=n. And, when used in
>> conjunction with PREEMPT_DYNAMIC, we continue to use PREEMPT_RCU=y.
>>
>> Patches 1-3  are cleanup patches:
>>   "rcu: fix header guard for rcu_all_qs()"
>>   "rcu: rename PREEMPT_AUTO to PREEMPT_LAZY"
>>   "sched: update __cond_resched comment about RCU quiescent states"
>>
>> Patch 4,
>>   "rcu: handle unstable rdp in rcu_read_unlock_strict()"
>>
>> handles a latent RCU bug rcu_report_qs_rdp() could be called with
>> an unstable rdp.
>>
>> Patches 5 and 6,
>>   "rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y"
>>   "osnoise: provide quiescent states"
>>
>> handle quiescent states for the (PREEMPT_LAZY=y, PREEMPT_RCU=n)
>> configuration.
>>
>> And, finally patch 7, "rcu: limit PREEMPT_RCU configurations",
>> explicitly limits PREEMPT_RCU=y to the PREEMPT_DYNAMIC or the latency
>> oriented models.
>
> Pulled into my -rcu tree for further review and testing, with initial
> tests passing.  Apologies for the delay!

Great. Thanks Paul!

Ankur

>> Changelog:
>>
>> v3:
>>  - moved patch-3 to be the last one in the series (suggested by Sebastian)
>>  - added "rcu: handle unstable rdp in rcu_read_unlock_strict()"
>>    (suggested by Frederic Weisbecker).
>>  - switched to a more robust check in rcu_flavor_sched_clock_irq()
>>    (suggested by Frederic Weisbecker).
>>  - simplified check in osnoise (suggested by Frederic Weisbecker).
>>  - dropped an unrelated scheduler patch.
>>
>> v2:
>>   - fixup incorrect usage of tif_need_resched_lazy() (comment from
>>     from Sebastian Andrzej Siewior)
>>   - massaged the commit messages a bit
>>   - drops the powerpc support for PREEMPT_LAZY as that was orthogonal
>>     to this series (Shrikanth will send that out separately.)
>>
>> Please review.
>>
>> Ankur Arora (7):
>>   rcu: fix header guard for rcu_all_qs()
>>   rcu: rename PREEMPT_AUTO to PREEMPT_LAZY
>>   sched: update __cond_resched comment about RCU quiescent states
>>   rcu: handle unstable rdp in rcu_read_unlock_strict()
>>   rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y
>>   osnoise: provide quiescent states
>>   rcu: limit PREEMPT_RCU configurations
>>
>>  include/linux/rcupdate.h     |  2 +-
>>  include/linux/rcutree.h      |  2 +-
>>  include/linux/srcutiny.h     |  2 +-
>>  kernel/rcu/Kconfig           |  4 ++--
>>  kernel/rcu/srcutiny.c        | 14 +++++++-------
>>  kernel/rcu/tree_plugin.h     | 22 +++++++++++++++++-----
>>  kernel/sched/core.c          |  4 +++-
>>  kernel/trace/trace_osnoise.c | 32 +++++++++++++++-----------------
>>  8 files changed, 47 insertions(+), 35 deletions(-)
>>
>> --
>> 2.43.5
>>


--
ankur