Re: [PATCH 2/2] srcu: reduce scans of srcu_data in counter wrap check

[Paul E. McKenney]

On Wed, Jan 31, 2018 at 10:42:21PM +0300, Ildar Ismagilov wrote:
> This patch reduce scans of srcu_data in case of a srcu_struct
> tree have more than one levels

Good catch!!!  Applied for testing and review.

I had to hand-apply this patch as well, and also fill out the commit log.
In addition, in the future, please capitalize the word in the Subject
line following the colon (":").  Please see below for the updated commit
and again please let me know if I messed something up.

                                                        Thanx, Paul

------------------------------------------------------------------------

commit bcec03ee2e43f2a24c61a99709d4dd1aa90438c2
Author: Ildar Ismagilov <devi...@gmail.com>
Date:   Wed Jan 31 22:42:21 2018 +0300

    srcu: Reduce scans of srcu_data in counter wrap check
    
    Currently, given a multi-level srcu_node tree, SRCU can scan the full
    set of srcu_data structures at each level when cleaning up after a grace
    period.  This, though harmless otherwise, represents pointless overhead.
    This commit therefore eliminates this overhead by scanning the srcu_data
    structures only when traversing the leaf srcu_node structures.
    
    Signed-off-by: Ildar Ismagilov <devi...@gmail.com>
    Signed-off-by: Paul E. McKenney <paul...@linux.vnet.ibm.com>

diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
index ad3e1aa5e6ea..72dc6e39be14 100644
--- a/kernel/rcu/srcutree.c
+++ b/kernel/rcu/srcutree.c
@@ -527,6 +527,7 @@ static void srcu_gp_end(struct srcu_struct *sp)
 {
        unsigned long cbdelay;
        bool cbs;
+       bool last_lvl;
        int cpu;
        unsigned long flags;
        unsigned long gpseq;
@@ -559,7 +560,8 @@ static void srcu_gp_end(struct srcu_struct *sp)
        rcu_for_each_node_breadth_first(sp, snp) {
                spin_lock_irq_rcu_node(snp);
                cbs = false;
-               if (snp >= sp->level[rcu_num_lvls - 1])
+               last_lvl = snp >= sp->level[rcu_num_lvls - 1];
+               if (last_lvl)
                        cbs = snp->srcu_have_cbs[idx] == gpseq;
                snp->srcu_have_cbs[idx] = gpseq;
                rcu_seq_set_state(&snp->srcu_have_cbs[idx], 1);
@@ -572,7 +574,7 @@ static void srcu_gp_end(struct srcu_struct *sp)
                        srcu_schedule_cbs_snp(sp, snp, mask, cbdelay);
 
                /* Occasionally prevent srcu_data counter wrap. */
-               if (!(gpseq & counter_wrap_check))
+               if (!(gpseq & counter_wrap_check) && last_lvl)
                        for (cpu = snp->grplo; cpu <= snp->grphi; cpu++) {
                                sdp = per_cpu_ptr(sp->sda, cpu);
                                spin_lock_irqsave_rcu_node(sdp, flags);