Re: [PATCH v6 1/4] rcu: Make call_rcu() lazy to save power

From: Joel Fernandes
Date: Mon Sep 26 2022 - 16:54:37 EST


Hi Vlad,

On Mon, Sep 26, 2022 at 09:39:23PM +0200, Uladzislau Rezki wrote:
[...]
> > > On my KVM machine the boot time is affected:
> > >
> > > <snip>
> > > [ 2.273406] e1000 0000:00:03.0 eth0: Intel(R) PRO/1000 Network Connection
> > > [ 11.945283] e1000 0000:00:03.0 ens3: renamed from eth0
> > > [ 22.165198] sr 1:0:0:0: [sr0] scsi3-mmc drive: 4x/4x cd/rw xa/form2 tray
> > > [ 22.165206] cdrom: Uniform CD-ROM driver Revision: 3.20
> > > [ 32.406981] sr 1:0:0:0: Attached scsi CD-ROM sr0
> > > [ 104.115418] process '/usr/bin/fstype' started with executable stack
> > > [ 104.170142] EXT4-fs (sda1): mounted filesystem with ordered data mode. Quota mode: none.
> > > [ 104.340125] systemd[1]: systemd 241 running in system mode. (+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +SECCOMP +BLKID +ELFUTILS +KMOD -IDN2 +IDN -PCRE2 default-hierarchy=hybrid)
> > > [ 104.340193] systemd[1]: Detected virtualization kvm.
> > > [ 104.340196] systemd[1]: Detected architecture x86-64.
> > > [ 104.359032] systemd[1]: Set hostname to <pc638>.
> > > [ 105.740109] random: crng init done
> > > [ 105.741267] systemd[1]: Reached target Remote File Systems.
> > > <snip>
> > >
> > > 2 - 11 and second delay is between 32 - 104. So there are still users which must
> > > be waiting for "RCU" in a sync way.
> >
> > I was wondering if you can compare boot logs and see which timestamp does the
> > slow down start from. That way, we can narrow down the callback. Also another
> > idea is, add "trace_event=rcu:rcu_callback,rcu:rcu_invoke_callback
> > ftrace_dump_on_oops" to the boot params, and then manually call
> > "tracing_off(); panic();" from the code at the first printk that seems off in
> > your comparison of good vs bad. For example, if "crng init done" timestamp is
> > off, put the "tracing_off(); panic();" there. Then grab the serial console
> > output to see what were the last callbacks that was queued/invoked.

Would you be willing to try these steps? Meanwhile I will try on my side as
well with the .config you sent me in another email.

> > > > diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> > > > index 08605ce7379d..40ae36904825 100644
> > > > --- a/include/linux/rcupdate.h
> > > > +++ b/include/linux/rcupdate.h
> > > > @@ -108,6 +108,13 @@ static inline int rcu_preempt_depth(void)
> > > >
> > > > #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
> > > >
> > > > +#ifdef CONFIG_RCU_LAZY
> > > > +void call_rcu_flush(struct rcu_head *head, rcu_callback_t func);
> > > > +#else
> > > > +static inline void call_rcu_flush(struct rcu_head *head,
> > > > + rcu_callback_t func) { call_rcu(head, func); }
> > > > +#endif
> > > > +
> > > > /* Internal to kernel */
> > > > void rcu_init(void);
> > > > extern int rcu_scheduler_active;
> > > > diff --git a/kernel/rcu/Kconfig b/kernel/rcu/Kconfig
> > > > index f53ad63b2bc6..edd632e68497 100644
> > > > --- a/kernel/rcu/Kconfig
> > > > +++ b/kernel/rcu/Kconfig
> > > > @@ -314,4 +314,12 @@ config TASKS_TRACE_RCU_READ_MB
> > > > Say N here if you hate read-side memory barriers.
> > > > Take the default if you are unsure.
> > > >
> > > > +config RCU_LAZY
> > > > + bool "RCU callback lazy invocation functionality"
> > > > + depends on RCU_NOCB_CPU
> > > > + default n
> > > > + help
> > > > + To save power, batch RCU callbacks and flush after delay, memory
> > > > + pressure or callback list growing too big.
> > > > +
> > > >
> > > Do you think you need this kernel option? Can we just consider and make
> > > it a run-time configurable? For example much more users will give it a try,
> > > so it will increase a coverage. By default it can be off.
> > >
> > > Also you do not need to do:
> > >
> > > #ifdef LAZY
> >
> > How does the "LAZY" macro end up being runtime-configurable? That's static /
> > compile time. Did I miss something?
> >
> I am talking about removing if:
>
> config RCU_LAZY
>
> we might run into issues related to run-time switching though.

When we started off, Paul said he wanted it kernel CONFIGurable. I will defer
to Paul on a decision for that. I prefer kernel CONFIG so people don't forget
to pass a boot param.

thanks,

- Joel