linux-kernel - Re: [PATCH V3 2/3] tracing/osnoise: Add preempt/irq disable options

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] [day] [month] [year] [list]

Message-ID: <20221130111026.47b9c456@gandalf.local.home>
Date:   Wed, 30 Nov 2022 11:10:26 -0500
From:   Steven Rostedt <rostedt@...dmis.org>
To:     Daniel Bristot de Oliveira <bristot@...nel.org>
Cc:     Masami Hiramatsu <mhiramat@...nel.org>,
        Jonathan Corbet <corbet@....net>,
        Juri Lelli <juri.lelli@...hat.com>,
        Clark Williams <williams@...hat.com>,
        linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH V3 2/3] tracing/osnoise: Add preempt/irq disable options

On Wed, 30 Nov 2022 16:47:29 +0100
Daniel Bristot de Oliveira <bristot@...nel.org> wrote:

> On 11/28/22 21:39, Steven Rostedt wrote:
> > On Fri, 25 Nov 2022 22:20:23 +0100
> > Daniel Bristot de Oliveira <bristot@...nel.org> wrote:
> >   
> >> @@ -1308,6 +1315,8 @@ static void notify_new_max_latency(u64 latency)
> >>   */
> >>  static int run_osnoise(void)
> >>  {
> >> +	bool preempt_disable = test_bit(OSN_PREEMPT_DISABLE, &osnoise_options);
> >> +	bool irq_disable = test_bit(OSN_IRQ_DISABLE, &osnoise_options);  
> > 
> > 	bool irq_disable = test_bit(OSN_IRQ_DISABLE, &osnoise_options);
> > 	bool preempt_disable = IS_ENABLED(CONFIG_PREEMPT) &&
> > 			!irq_disable && test_bit(OSN_PREEMPT_DISABLE, &osnoise_options);  
> 
> Ooops again, that is not exactly what I wanted, because...

Then just remove the "IS_ENABLED()" part and it should work just fine.

> 
> >>  	struct osnoise_variables *osn_var = this_cpu_osn_var();
> >>  	u64 start, sample, last_sample;
> >>  	u64 last_int_count, int_count;
> >> @@ -1335,6 +1344,14 @@ static int run_osnoise(void)
> >>  	 */
> >>  	threshold = tracing_thresh ? : 5000;
> >>  
> >> +	/*
> >> +	 * IRQ disable also implies in preempt disable.
> >> +	 */
> >> +	if (irq_disable)
> >> +		local_irq_disable();  
> > 
> > 	if (preempt_disable)  
> >> +		preempt_disable();

The above is a nop when CONFIG_PREEMPT is false.

> >> +
> >>  	/*
> >>  	 * Make sure NMIs see sampling first
> >>  	 */
> >> @@ -1422,16 +1439,21 @@ static int run_osnoise(void)
> >>  		 * cond_resched()
> >>  		 */
> >>  		if (IS_ENABLED(CONFIG_PREEMPT_RCU)) {
> >> -			local_irq_disable();
> >> +			if (!irq_disable)
> >> +				local_irq_disable();
> >> +
> >>  			rcu_momentary_dyntick_idle();
> >> -			local_irq_enable();
> >> +
> >> +			if (!irq_disable)
> >> +				local_irq_enable();
> >>  		}
> >>  
> >>  		/*
> >>  		 * For the non-preemptive kernel config: let threads runs, if
> >> -		 * they so wish.
> >> +		 * they so wish, unless set not do to so.
> >>  		 */  
> 
> Then I end up cond_resched'ing here in the non-preemptive kernel.

Sorry, I missed the point that you want to *not* cond_resched() even in a
CONFIG_PREEMPT is false situation, if preempt_disable flag is set. That's
the reason I added the IS_ENABLED(CONFIG_PREEMPT) check at the top. I
originally didn't have that, but then thought this should always happen in
that case.

> 
> >> -		cond_resched();
> >> +		if (!irq_disable && !preempt_disable)
> >> +			cond_resched();  
> 
> But I also want to avoid this cond_resched if preempt_disable is set.

Right, so just remove the IS_ENABLED() part in the beginning.

> 
> So, I will merge both things:
> 
> 	- change the preempt_disable assignment to check !irq_disabled, like:
> 
> 		/*
> 		 * Disabling preemption is only required if IRQs are enabled, and the options is set on.
> 		 */
> 		preempt_disable = !irq_disable && test_bit(OSN_PREEMPT_DISABLE, &osnoise_options);

Yep (that's what I original had until I changed it)

> 
> 	- change the preempt disabled if to
> 		if (IS_ENABLED(CONFIG_PREEMPT) && preempt_disabled)
> 			preempt_disable();

No need, preempt_disable() is a nop when CONFIG_PREEMPT is false.

> 
> I tested with both preemption models (preemptive and not preemptive) and it works fine.
> 
> am I missing something?

Just that you don't need to add the IS_ENABLED() part at all.

-- Steve