linux-kernel - Re: Early timeouts due to inaccurate jiffies during system suspend/resume

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 for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20180424142101.GO13908@intel.com>
Date:   Tue, 24 Apr 2018 17:21:01 +0300
From:   Ville Syrjälä <ville.syrjala@...ux.intel.com>
To:     Imre Deak <imre.deak@...el.com>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        LKML <linux-kernel@...r.kernel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Mika Kuoppala <mika.kuoppala@...el.com>,
        Chris Wilson <chris@...is-wilson.co.uk>
Subject: Re: Early timeouts due to inaccurate jiffies during system
 suspend/resume

On Tue, Apr 24, 2018 at 05:07:41PM +0300, Imre Deak wrote:
> On Mon, Apr 23, 2018 at 08:01:28PM +0300, Imre Deak wrote:
> > On Thu, Apr 19, 2018 at 01:05:39PM +0200, Thomas Gleixner wrote:
> > > On Thu, 19 Apr 2018, Imre Deak wrote:
> > > > Hi,
> > > > 
> > > > while checking bug [1], I noticed that jiffies based timing loops like
> > > > 
> > > > 	expire = jiffies + timeout + 1;
> > > > 	while (!time_after(jiffies, expire))
> > > > 		do_something;
> > > > 
> > > > can last shorter than expected (that is less than timeout).
> > > 
> > > Yes, that can happen when the timer interrupt is delayed long enough for
> > > whatever reason. If you need accurate timing then you need to use
> > > ktime_get().
> > 
> > Thanks. I always regarded jiffies as non-accurate, but something that
> > gives a minimum time delay guarantee (when adjusted by +1 as above). I
> > wonder if there are other callers in kernel that don't expect an early
> > timeout.
> 
> msleep and any other schedule_timeout based waits are also affected. At the
> same time for example msleep's documentation says:
> "msleep - sleep safely even with waitqueue interruptions".
> 
> To me that suggests a wait with a minimum guaranteed delay.
> 
> Ville had an idea to make the behavior more deterministic by clamping
> the jiffies increment to 1 for each timer interrupt. Would that work?

Another observation is that this is basically the same problem
that we had with the drm vblank counter. I solved that by introducing
drm_accurate_vblank_count() which makes sure the counter is up to
date before sampling it. Then we can safely do stuff like:

count = drm_accurate_vblank_count();
while (drm_vblank_count() == count)
	...;

As long as we don't lose all vblank interrupts that will work and never
complete prematurely. And we still allow the vblank counter to increment
by >1.

I suppose doing something similar for jiffies would be nice as well,
but I'm not sure how feasible that would be. At the very least it
would involve patching a lot of code.

> 
> > 
> > We switched now to using ktime_get_raw() in the i915 driver.
> > 
> > > 
> > > > After some ftracing it seems like jiffies gets stale due to a missed
> > > > LAPIC timer interrupt after the interrupt is armed in
> > > > lapic_next_deadline() and before jiffies is sampled at 2. above.
> > > > Eventually the interrupt does get delivered, at which point jiffies gets
> > > > updated via tick_do_update_jiffies64() with a >1 ticks increment.
> > > > Between lapic_next_deadline() and the - late - delivery of the interrupt
> > > > the CPU on which the interrupt is armed doesn't go idle.
> > > 
> > > That's odd. I have no real explanation for that.
> > 
> > Looks like the reason is IRQ latency. For reference here are the
> > longest ones I found with irqsoff ftracing, all running with IRQs disabled
> > during system resume:
> > 
> > hpet_rtc_interrupt()->hpet_rtc_timer_reinit():
> > do { ... } while(!hpet_cnt_ahead(...));
> > takes sometimes up to ~40msec for me.
> > 
> > hpet_rtc_interrupt()->mc146818_get_time():
> > if (mc146818_is_updating()) mdelay(20);
> > 
> > driver_probe_device->atkbd_connect()->i8042_port_close()->__i8042_command()->i8042_wait_write():
> > takes sometimes up to ~10msec for me.
> > 
> > All the above paired with asynchronous calling of the drivers' resume
> > hooks may result in the jumps in jiffies I saw.
> > 
> > --Imre

-- 
Ville Syrjälä
Intel