Date:   Wed, 21 Dec 2022 11:26:22 -0800
From:   "Paul E. McKenney" <paulmck@...nel.org>
To:     "Zhang, Qiang1" <qiang1.zhang@...el.com>
Cc:     "frederic@...nel.org" <frederic@...nel.org>,
        "quic_neeraju@...cinc.com" <quic_neeraju@...cinc.com>,
        "joel@...lfernandes.org" <joel@...lfernandes.org>,
        "rcu@...r.kernel.org" <rcu@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] rcu: Remove duplicate settings for rcu boost kthreads
 affinity

On Sat, Dec 17, 2022 at 01:02:03PM +0000, Zhang, Qiang1 wrote:
> On Thu, Dec 15, 2022 at 05:00:29PM +0800, Zqiang wrote:
> > Currently, when CPUs going offline, the rcu boost kthreads CPU
> > affinity has been reset in rcutree_offline_cpu(), therefore this
> > commit remove duplicate settings in rcutree_dead_cpu().
> > 
> > Signed-off-by: Zqiang <qiang1.zhang@...el.com>
> >
> >One question...
> >
> >Given that rcutree_offline_cpu() causes rcu_boost_kthread_setaffinity()
> >to be invoked with the number of the outgoing CPU, but rcutree_dead_cpu()
> >instead passes in -1, are these two invocations really redundant?
> >
> >Although -1 is passed in, but at this time, the outgoing CPU has been removed from 
> >rcu_rnp_online_cpus(rnp),  in other words, we will eliminate the same CPU, which is
> >also  we passed in cpu in the rcutree_offline_cpu() before.
> >
> 
> I also used " qemu -smp 4 "for a simple test.
> 
> 1) rcutree_offline_cpu()
> 
> (gdb) bt
> #0  set_cpus_allowed_ptr (p=p@...ry=0xffff888100bd9cc0, new_mask=new_mask@...ry=0xffff888100ce7cd0)
>     at kernel/sched/core.c:2939
> #1  0xffffffff811be169 in rcu_boost_kthread_setaffinity (rnp=0xffffffff837a4c00 <rcu_state>,
>     outgoingcpu=outgoingcpu@...ry=3) at kernel/rcu/tree_plugin.h:1250
> #2  0xffffffff811cfa9f in rcutree_affinity_setting (outgoing=3, cpu=3) at kernel/rcu/tree.c:4297
> #3  rcutree_offline_cpu (cpu=3) at kernel/rcu/tree.c:4341
> #4  0xffffffff810c4e6f in cpuhp_invoke_callback (cpu=cpu@...ry=3, state=state@...ry=CPUHP_AP_RCUTREE_ONLINE,
>     bringup=bringup@...ry=false, node=<optimized out>, lastp=lastp@...ry=0xffff88815b027568) at kernel/cpu.c:192
> #5  0xffffffff810c6e24 in cpuhp_thread_fun (cpu=<optimized out>) at kernel/cpu.c:785
> #6  0xffffffff8111784e in smpboot_thread_fn (data=0xffff88810014c3c0) at kernel/smpboot.c:164
> #7  0xffffffff8110c88e in kthread (_create=0xffff8881009b8b00) at kernel/kthread.c:376
> #8  0xffffffff81002902 in ret_from_fork () at arch/x86/entry/entry_64.S:306
> #9  0x0000000000000000 in ?? ()
> 
> (gdb) p/x *0xffff888100ce7cd0
> $1 = 0x7
> 
> 2) rcutree_dead_cpu()
> 
> #0  set_cpus_allowed_ptr (p=p@...ry=0xffff888100bd9cc0, new_mask=new_mask@...ry=0xffff888109fc7948)
>     at kernel/sched/core.c:2939
> #1  0xffffffff811be169 in rcu_boost_kthread_setaffinity (rnp=0xffffffff837a4c00 <rcu_state>,
>     outgoingcpu=outgoingcpu@...ry=-1) at kernel/rcu/tree_plugin.h:1250
> #2  0xffffffff811cf4a1 in rcutree_dead_cpu (cpu=3) at kernel/rcu/tree.c:4176
> #3  0xffffffff810c4e6f in cpuhp_invoke_callback (cpu=cpu@...ry=3, state=CPUHP_RCUTREE_PREP, bringup=bringup@...ry=false,
>     node=node@...ry=0x0 <fixed_percpu_data>, lastp=lastp@...ry=0x0 <fixed_percpu_data>) at kernel/cpu.c:192
> #4  0xffffffff810c58f5 in cpuhp_invoke_callback_range (bringup=bringup@...ry=false, cpu=cpu@...ry=3,
>     st=st@...ry=0xffff88815b027540, target=target@...ry=CPUHP_OFFLINE) at kernel/cpu.c:675
> #5  0xffffffff827a5f6a in cpuhp_down_callbacks (target=CPUHP_OFFLINE, st=0xffff88815b027540, cpu=3) at kernel/cpu.c:1170
> #6  _cpu_down (cpu=cpu@...ry=3, tasks_frozen=tasks_frozen@...ry=0, target=target@...ry=CPUHP_OFFLINE) at kernel/cpu.c:1170
> #7  0xffffffff810c6ff8 in cpu_down_maps_locked (target=CPUHP_OFFLINE, cpu=3) at kernel/cpu.c:1202
> #8  cpu_down (cpu=3, target=target@...ry=CPUHP_OFFLINE) at kernel/cpu.c:1210
> #9  0xffffffff810c7906 in cpu_device_down (dev=dev@...ry=0xffff88815b026988) at kernel/cpu.c:1227
> 
> (gdb) p/x *0xffff888109fc7948
> $2 = 0x7
> 
> As can be seen from the above, we have set the same cpumask twice for the same boost kthreads.

Very good, thank you for checking!

I have queued this as shown below, so please let me know of any issues.

							Thanx, Paul

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

commit 5815245b56229198d5f5d0c48c26d3c0c53ee222
Author: Zqiang <qiang1.zhang@...el.com>
Date:   Wed Dec 21 11:15:43 2022 -0800

    rcu: Remove redundant call to rcu_boost_kthread_setaffinity()
    
    The rcu_boost_kthread_setaffinity() function is invoked at
    rcutree_online_cpu() and rcutree_offline_cpu() time, early in the online
    timeline and late in the offline timeline, respectively.  It is also
    invoked from rcutree_dead_cpu(), however, in the absence of userspace
    manipulations (for which userspace must take responsibility), this call
    is redundant with that from rcutree_offline_cpu().  This redundancy can
    be demonstrated by printing out the relevant cpumasks
    
    This commit therefore removes the call to rcu_boost_kthread_setaffinity()
    from rcutree_dead_cpu().
    
    Signed-off-by: Zqiang <qiang1.zhang@...el.com>
    Signed-off-by: Paul E. McKenney <paulmck@...nel.org>

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index c3255d27286c4..a6181b76fd5ab 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -4165,15 +4165,10 @@ static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
  */
 int rcutree_dead_cpu(unsigned int cpu)
 {
-	struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
-	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
-
 	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
 		return 0;
 
 	WRITE_ONCE(rcu_state.n_online_cpus, rcu_state.n_online_cpus - 1);
-	/* Adjust any no-longer-needed kthreads. */
-	rcu_boost_kthread_setaffinity(rnp, -1);
 	// Stop-machine done, so allow nohz_full to disable tick.
 	tick_dep_clear(TICK_DEP_BIT_RCU);
 	return 0;

Powered by blists - more mailing lists