Hi, Z qiang. > Maybe this "start new gp" note misunderstood you. > For built with CONFIG_RCU_STRICT_GRACE_PERIOD=y and CONFIG_PREEMPT=n kernels， > if the gp kthread start a new GP before we exit the RCU read critical section, > and just before we call rcu_report_qs_rdp() in > rcu_read_unlock_strict(), at this time if the clock irq > happens and find that the "rcu_seq_current(&rnp->gp_seq) != > rdp->gp_seq" is true in rcu_pening(), > will trigger RCU softirq and find that the rcu_seq_new_gp(rdp->gp_seq, > rnp->gp_seq) is true, > will set rdp->cpu_no_qs.b.norm is true. when we return from the > softirq and call rcu_report_qs_rdp() > in rcu_read_unlock_strict(), find that the rdp->cpu_no_qs.b.norm is true. > so there is a situation where the rdp->cpu_no_qs.b.norm is true. Thanks for making me out from my misunderstanding..! and even the not in STRICT_GRACE_PERIOD, old grace period's qs could be reported though rcu_gp_kthread start new gp by first rcu_report_qs_rdp, will be clear soon by continuous rcu_report_qs_rdp which is triggered by interrupting before the first rcu_report_qs_rdp. Many thanks..! And sorry to make noise...!
