> On Sep 26, 2022, at 6:35 PM, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
>
> On Mon, Sep 26, 2022 at 08:54:27PM +0000, Joel Fernandes wrote:
>> 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.
>
> I am fine with a kernel boot parameter for this one. You guys were the
> ones preferring Kconfig options. ;-)
Yes I still prefer that.. ;-)
> But in that case, the CONFIG_RCU_NOCB_CPU would come into play to handle
> the case where there is no bypass.
If you don’t mind, let’s do both like we did for NOCB_CPU_ALL. In which case, Vlad since this was your suggestion, would you be so kind to send a patch adding a boot parameter on top of the series? ;-). I’ll include it in the next version. I’d suggest keep the boot param default off and add a CONFIG option that forces the boot param to be turned on.
Thanks,
- Joel
>
> Thanx, Paul
