linux-kernel - Re: [RFC PATCH 03/10] timer: Simplify LVL_START() and calc

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: <20200702122659.GB17670@lenoir>
Date:   Thu, 2 Jul 2020 14:27:01 +0200
From:   Frederic Weisbecker <frederic@...nel.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        Anna-Maria Gleixner <anna-maria@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>,
        Juri Lelli <juri.lelli@...hat.com>
Subject: Re: [RFC PATCH 03/10] timer: Simplify LVL_START() and calc_index()

On Thu, Jul 02, 2020 at 01:59:17PM +0200, Thomas Gleixner wrote:
> Frederic Weisbecker <frederic@...nel.org> writes:
> > LVL_START() makes the first index of a level to start with what would be
> > the value of all bits set of the previous level.
> >
> > For example level 1 starts at 63 instead of 64.
> >
> > To cope with that, calc_index() always adds one offset for the level
> > granularity to the expiry passed in parameter.
> >
> > Yet there is no apparent reason for such fixups so simplify the whole
> > thing.
> 
> You sure?

No :o)

> 
> > @@ -158,7 +158,7 @@ EXPORT_SYMBOL(jiffies_64);
> >   * The time start value for each level to select the bucket at enqueue
> >   * time.
> >   */
> > -#define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
> > +#define LVL_START(n)	(LVL_SIZE << (((n) - 1) * LVL_CLK_SHIFT))
> 
> >  /* Size of each clock level */
> >  #define LVL_BITS	6
> > @@ -489,7 +489,7 @@ static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
> >   */
> >  static inline unsigned calc_index(unsigned expires, unsigned lvl)
> >  {
> > -	expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
> > +	expires >>= LVL_SHIFT(lvl);
> >  	return LVL_OFFS(lvl) + (expires & LVL_MASK);
> >  }
> 
> So with that you move the expiry of each timer one jiffie ahead vs. the
> original code, which violates the guarantee that a timer sleeps at least
> for one jiffie for real and not measured in jiffies.
> 
> base->clk = 1
> jiffies = 0
>  local_irq_disable()
>                         -> timer interrupt is raised in HW
>  timer->expires = 1
>  add_timer(timer)
>         ---> index == 1
>  local_irq_enable()
>  timer interrupt
>     jiffies++;
>     softirq()
>         expire(timer);
> 
> So the off by one has a reason.

Fair enough. I didn't know about the one jiffy sleep guarantee.
I'll convert this patch to a comment explaining that off by one,
using your above example.

Thanks!