mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Samir M <samir@linux.ibm.com>
To: Shrikanth Hegde <sshegde@linux.ibm.com>,
	Joel Fernandes <joelagnelf@nvidia.com>
Cc: Uladzislau Rezki <urezki@gmail.com>,
	Paul E McKenney <paulmck@kernel.org>,
	Vishal Chourasia <vishalc@linux.ibm.com>,
	Neeraj upadhyay <Neeraj.Upadhyay@amd.com>,
	RCU <rcu@vger.kernel.org>, LKML <linux-kernel@vger.kernel.org>,
	Frederic Weisbecker <frederic@kernel.org>
Subject: Re: [PATCH] rcu: Latch normal synchronize_rcu() path on flood
Date: Sun, 18 Jan 2026 16:54:46 +0530	[thread overview]
Message-ID: <9c544560-1448-42a4-b16d-000f06733d0b@linux.ibm.com> (raw)
In-Reply-To: <00e91ebc-0783-4519-9727-53dd3a625298@linux.ibm.com>


On 17/01/26 5:48 pm, Shrikanth Hegde wrote:
>
>
> On 1/17/26 2:18 PM, Joel Fernandes wrote:
>>
>>
>>> On Jan 17, 2026, at 1:17 AM, Samir M <samir@linux.ibm.com> wrote:
>>>
>>> 
>>>> On 15/01/26 12:04 am, Uladzislau Rezki (Sony) wrote:
>>>> Currently, rcu_normal_wake_from_gp is only enabled by default
>>>> on small systems(<= 16 CPUs) or when a user explicitly set it
>>>> enabled.
>>>>
>>>> This patch introduces an adaptive latching mechanism:
>>>>   * Tracks the number of in-flight synchronize_rcu() requests
>>>>     using a new atomic_t counter(rcu_sr_normal_count);
>>>>
>>>>   * If the count exceeds RCU_SR_NORMAL_LATCH_THR(64), it sets
>>>>     the rcu_sr_normal_latched, reverting new requests onto the
>>>>     scaled wait_rcu_gp() path;
>>>>
>>>>   * The latch is cleared only when the pending requests are fully
>>>>     drained(nr == 0);
>>>>
>>>>   * Enables rcu_normal_wake_from_gp by default for all systems,
>>>>     relying on this dynamic throttling instead of static CPU
>>>>     limits.
>>>>
>>>> Suggested-by: Joel Fernandes <joelagnelf@nvidia.com>
>>>> Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
>>>> ---
>>>>   kernel/rcu/tree.c | 37 ++++++++++++++++++++++++++-----------
>>>>   1 file changed, 26 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
>>>> index 293bbd9ac3f4..c42d480d6e0b 100644
>>>> --- a/kernel/rcu/tree.c
>>>> +++ b/kernel/rcu/tree.c
>>>> @@ -1631,17 +1631,21 @@ static void rcu_sr_put_wait_head(struct 
>>>> llist_node *node)
>>>>       atomic_set_release(&sr_wn->inuse, 0);
>>>>   }
>>>>   -/* Enable rcu_normal_wake_from_gp automatically on small 
>>>> systems. */
>>>> -#define WAKE_FROM_GP_CPU_THRESHOLD 16
>>>> -
>>>> -static int rcu_normal_wake_from_gp = -1;
>>>> +static int rcu_normal_wake_from_gp = 1;
>>>>   module_param(rcu_normal_wake_from_gp, int, 0644);
>>>>   static struct workqueue_struct *sync_wq;
>>>>   +#define RCU_SR_NORMAL_LATCH_THR 64
>>>> +
>>>> +/* Number of in-flight synchronize_rcu() calls queued on srs_next. */
>>>> +static atomic_long_t rcu_sr_normal_count;
>>>> +static atomic_t rcu_sr_normal_latched;
>>>> +
>>>>   static void rcu_sr_normal_complete(struct llist_node *node)
>>>>   {
>>>>       struct rcu_synchronize *rs = container_of(
>>>>           (struct rcu_head *) node, struct rcu_synchronize, head);
>>>> +    long nr;
>>>>         WARN_ONCE(IS_ENABLED(CONFIG_PROVE_RCU) &&
>>>> !poll_state_synchronize_rcu_full(&rs->oldstate),
>>>> @@ -1649,6 +1653,15 @@ static void rcu_sr_normal_complete(struct 
>>>> llist_node *node)
>>>>         /* Finally. */
>>>>       complete(&rs->completion);
>>>> +    nr = atomic_long_dec_return(&rcu_sr_normal_count);
>>>> +    WARN_ON_ONCE(nr < 0);
>>>> +
>>>> +    /*
>>>> +     * Unlatch: switch back to normal path when fully
>>>> +     * drained and if it has been latched.
>>>> +     */
>>>> +    if (nr == 0)
>>>> +        (void)atomic_cmpxchg(&rcu_sr_normal_latched, 1, 0);
>>>>   }
>>>>     static void rcu_sr_normal_gp_cleanup_work(struct work_struct 
>>>> *work)
>>>> @@ -1794,7 +1807,14 @@ static bool rcu_sr_normal_gp_init(void)
>>>>     static void rcu_sr_normal_add_req(struct rcu_synchronize *rs)
>>>>   {
>>>> +    long nr;
>>>> +
>>>>       llist_add((struct llist_node *) &rs->head, &rcu_state.srs_next);
>>>> +    nr = atomic_long_inc_return(&rcu_sr_normal_count);
>>>> +
>>>> +    /* Latch: only when flooded and if unlatched. */
>>>> +    if (nr >= RCU_SR_NORMAL_LATCH_THR)
>>>> +        (void)atomic_cmpxchg(&rcu_sr_normal_latched, 0, 1);
>>>>   }
>>>>     /*
>>>> @@ -3268,7 +3288,8 @@ static void synchronize_rcu_normal(void)
>>>>         trace_rcu_sr_normal(rcu_state.name, &rs.head, TPS("request"));
>>>>   -    if (READ_ONCE(rcu_normal_wake_from_gp) < 1) {
>>>> +    if (READ_ONCE(rcu_normal_wake_from_gp) < 1 ||
>>>> +            atomic_read(&rcu_sr_normal_latched)) {
>>>>           wait_rcu_gp(call_rcu_hurry);
>>>>           goto trace_complete_out;
>>>>       }
>>>> @@ -4892,12 +4913,6 @@ void __init rcu_init(void)
>>>>       sync_wq = alloc_workqueue("sync_wq", WQ_MEM_RECLAIM | 
>>>> WQ_UNBOUND, 0);
>>>>       WARN_ON(!sync_wq);
>>>>   -    /* Respect if explicitly disabled via a boot parameter. */
>>>> -    if (rcu_normal_wake_from_gp < 0) {
>>>> -        if (num_possible_cpus() <= WAKE_FROM_GP_CPU_THRESHOLD)
>>>> -            rcu_normal_wake_from_gp = 1;
>>>> -    }
>>>> -
>>>>       /* Fill in default value for rcutree.qovld boot parameter. */
>>>>       /* -After- the rcu_node ->lock fields are initialized! */
>>>>       if (qovld < 0)
>>>
>>>
>>> Hi Uladzislau,
>>>
>>> I verified this patch using the configuration described below.
>>> Configuration:
>>>      •    Kernel version: 6.19.0-rc5
>>>      •    Number of CPUs: 2048
>>>
>>> Using this setup, I evaluated the patch with both SMT enabled and 
>>> SMT disabled. The results indicate that when SMT is enabled, the 
>>> system time is noticeably higher. In contrast, with SMT disabled, no 
>>> significant increase in system time is observed.
>>>
>>> SMT=ON  -> sys 31m22.922s
>>> SMT=OFF -> sys 0m0.046s
>>>
>>>
>>> SMT Mode    | Without Patch    | With Patch   | % Improvement    |
>>> ------------------------------------------------------------------
>>> SMT=off     | 30m 53.194s      | 26m 24.009s  | +14.53%          |
>>> SMT=on      | 49m 5.920s       | 47m 5.513s   | +4.09%
>>
>> So it takes you 47 minutes to offline CPUs and you are Ok with that?
>>
>> - Joel
>>
>
>
> This is certainly quite long. IMO not worth the added complexity
> of atomic inc/dec reads happening(even though till 64 CPUs)
>
> Samir,
> can you post number's with vishal's patch of using rcu expedited?
>
>
>>
>>
>>>         |
>>>
>>>
>>> Please add below tag: Tested-by: Samir M <samir@linux.ibm.com>
>>>
>>> Regards,
>>> Samir
>>>
>>>
>
Shrikanth,

I verified Vishal’s patch as described below, using the following 
configuration.
Configuration:
     •    Kernel version: 6.19.0-rc5
     •    Number of CPUs: 2048

Vishal's patch link: 
https://lore.kernel.org/all/20260112094332.66006-2-vishalc@linux.ibm.com/
Verification of Vishal’s patch shows a significant improvement in the 
*SMT=off* case and a measurable improvement in the *SMT=on* case.

SMT Mode    | Without Patch    | With Patch | % Improvement    |
------------------------------------------------------------------
SMT=off     | 30m 53.194s      |  6m 4.250s  | +80.40%          |
SMT=on      | 49m 5.920s       | 36m 50.386s | +25.01%          |


Regards,
Samir


  reply	other threads:[~2026-01-18 11:25 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-01-14 18:34 Uladzislau Rezki (Sony)
2026-01-15 17:23 ` Shrikanth Hegde
2026-01-15 21:39   ` Uladzislau Rezki
2026-01-17  6:17 ` Samir M
2026-01-17  8:48   ` Joel Fernandes
2026-01-17 12:18     ` Shrikanth Hegde
2026-01-18 11:24       ` Samir M [this message]
2026-01-25 14:22       ` Uladzislau Rezki
2026-01-29 19:33         ` Samir M
2026-02-02 11:25           ` Uladzislau Rezki
2026-01-19 11:24   ` Uladzislau Rezki
2026-01-17 15:16 ` Markus Elfring

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=9c544560-1448-42a4-b16d-000f06733d0b@linux.ibm.com \
    --to=samir@linux.ibm.com \
    --cc=Neeraj.Upadhyay@amd.com \
    --cc=frederic@kernel.org \
    --cc=joelagnelf@nvidia.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=paulmck@kernel.org \
    --cc=rcu@vger.kernel.org \
    --cc=sshegde@linux.ibm.com \
    --cc=urezki@gmail.com \
    --cc=vishalc@linux.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®