linux-kernel - Re: [PATCH v11 06/12] pwm: imx27: Use 64-bit division macro and function

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: <CAK8P3a2Ua5v67vLLzQxsLPZfX9z5D7o-sfFpurQVWVykRK8sEQ@mail.gmail.com>
Date:   Fri, 3 Apr 2020 21:13:18 +0200
From:   Arnd Bergmann <arnd@...db.de>
To:     Guru Das Srinagesh <gurus@...eaurora.org>
Cc:     Thierry Reding <thierry.reding@...il.com>,
        Linux PWM List <linux-pwm@...r.kernel.org>,
        Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>,
        Subbaraman Narayanamurthy <subbaram@...eaurora.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Shawn Guo <shawnguo@...nel.org>,
        Sascha Hauer <s.hauer@...gutronix.de>,
        Pengutronix Kernel Team <kernel@...gutronix.de>,
        Fabio Estevam <festevam@...il.com>,
        NXP Linux Team <linux-imx@....com>,
        David Collins <collinsd@...eaurora.org>
Subject: Re: [PATCH v11 06/12] pwm: imx27: Use 64-bit division macro and function

On Fri, Apr 3, 2020 at 7:37 PM Guru Das Srinagesh <gurus@...eaurora.org> wrote:
>
> On Thu, Apr 02, 2020 at 11:16:22PM +0200, Arnd Bergmann wrote:
> > This looks correct, but very expensive, and you don't really have to
> > go this far, given that c1 is guaranteed to be a 32-bit number, and
> > you divide by a constant in the end.
> >
> > Why not do something like
> >
> > #define SHIFT 41 /* arbitrarily picked, not too big, not too small */
> > #define MUL 2199 /* 2^SHIFT / NSEC_PER_SEC */
> > period_cycles = clk_get_rate(imx->clk_per) * ((state->period * MUL) >> SHIFT);
>
> I have two concerns with this:
>
> 1. This actually results in the division by 1000010575.5125057 instead
>    of NSECS_PER_SEC whereas both the existing as well as the proposed logic
>    divide exactly by NSECS_PER_SEC.
> 2. What method shall be used to pick the SHIFT value? How is this to be
>    chosen?

I picked one that would result in a fairly accurate rounding and did not
seem too likely to overflow.

> Also, this seems sort of similar to my initial attempt at this
> problem, where period was being pre-divided prior to the multiplication,
> which was (rightly) NACKed.
>
>         c *= div_u64(state->period, 1000000000);

Ah, right, I did make the same mistake here. What I actually indended
was to do two separate shifts that add up to 41. I suppose there is also still
the fundamental problem that without bounds checking the calculation can
always overflow for large inputs.

       Arnd