linux-kernel - Re: [PATCH v6 1/4] rcu: Make call

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <BE2B629D-B5D2-4ED0-944E-2F13E846047E@joelfernandes.org>
Date:   Mon, 26 Sep 2022 19:44:19 -0400
From:   Joel Fernandes <joel@...lfernandes.org>
To:     paulmck@...nel.org
Cc:     Uladzislau Rezki <urezki@...il.com>, rcu@...r.kernel.org,
        linux-kernel@...r.kernel.org, rushikesh.s.kadam@...el.com,
        neeraj.iitr10@...il.com, frederic@...nel.org, rostedt@...dmis.org
Subject: Re: [PATCH v6 1/4] rcu: Make call_rcu() lazy to save power



> On Sep 26, 2022, at 6:35 PM, Paul E. McKenney <paulmck@...nel.org> 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