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: <20220217223736.DFC2EC340E8@smtp.kernel.org>
Date:   Thu, 17 Feb 2022 14:37:35 -0800
From:   Stephen Boyd <sboyd@...nel.org>
To:     Nikita Travkin <nikita@...n.ru>
Cc:     linus.walleij@...aro.org, mturquette@...libre.com,
        bjorn.andersson@...aro.org, agross@...nel.org, tdas@...eaurora.org,
        svarbanov@...sol.com, linux-arm-msm@...r.kernel.org,
        linux-clk@...r.kernel.org, linux-gpio@...r.kernel.org,
        linux-kernel@...r.kernel.org, ~postmarketos/upstreaming@...ts.sr.ht
Subject: Re: [PATCH 1/4] clk: qcom: clk-rcg2: Fail Duty-Cycle configuration if MND divider is not enabled.

Quoting Nikita Travkin (2022-01-26 07:14:21)
> Stephen Boyd писал(а) 11.01.2022 01:14:
> > Quoting Nikita Travkin (2022-01-07 23:25:19)
> >> Hi,
> >>
> >> Stephen Boyd писал(а) 08.01.2022 05:52:
> >> > Quoting Nikita Travkin (2021-12-09 08:37:17)
> >> I'm adding this error here primarily to bring attention of the
> >> user (e.g. developer enabling some peripheral that needs
> >> duty cycle control) who might have to change their clock tree
> >> to make this control effective. So, assuming that if someone
> >> sets the duty cycle to 50% then they might set it to some other
> >> value later, it makes sense to fail the first call anyway.
> >>
> >> If you think there are some other possibilities for this call
> >> to happen specifically with 50% duty cycle (e.g. some
> >> preparations or cleanups in the clk subsystem or some drivers
> >> that I'm not aware of) then I can make an exemption in the check
> >> for that.
> >>
> > 
> > I don't see anywhere in clk_set_duty_cycle() where it would bail out
> > early if the duty cycle was set to what it already is. The default for
> > these clks is 50%, so I worry that some driver may try to set the duty
> > cycle to 50% and then fail now. Either we need to check the duty cycle
> > in the core before calling down into the driver or we need to check it
> > here in the driver. Can you send a patch to check the current duty cycle
> > in the core before calling down into the clk ops?
> 
> Hi, sorry for a rather delayed response,
> I spent a bit of time looking at how to make the clk core be
> careful with ineffective duty-cycle calls and I can't find a
> nice way to do this... My idea was something like this:
> 
> static int clk_core_set_duty_cycle_nolock(struct clk_core *core,
>                                           struct clk_duty *duty)
> {       /* ... */
> 
>         /* Update core->duty values */
>         clk_core_update_duty_cycle_nolock(core);
> 
>         if ( /* duty doesn't match core->duty */ ) {
>                 ret = core->ops->set_duty_cycle(core->hw, duty);
>         /* ... */
> }
> 
> However there seem to be drawbacks to any variant of the
> comparison that I could come up with:
> 
> Naive one would be to do
>     if (duty->num != core->duty->num || duty->den != core->duty->den)
> but it won't correctly compare e.g. 1/2 and 10/20.
> 
> Other idea was to do
>     if (duty->den / duty->num != core->duty->den / core->duty->num)
> but it will likely fail with very close values (e.g. 100/500 and 101/500)
> 
> I briefly thought of some more sophisticated math but I don't
> like the idea of complicating this too far.
> 
> I briefly grepped the kernel sources for duty-cycle related methods
> and I saw only one user of the clk_set_duty_cycle:
>     sound/soc/meson/axg-tdm-interface.c
> Notably it sets the cycle to 1/2 in some cases, though it seems to
> be tied to the drivers/clk/meson/sclk-div.c clock driver by being
> the blocks of the same SoC.

Indeed, so this patch is untested? I doubt the qcom driver is being used
with the one caller of clk_set_duty_cycle() in the kernel.

> 
> Thinking of it a bit more, I saw another approach to the problem
> I want to solve: Since I just want to make developers aware of the
> hardware quirk, maybe I don't need to fail the set but just put a
> WARN or even WARN_ONCE there? This way the behavior will be unchanged.
> 

I don't like the idea of a WARN or a WARN_ONCE as most likely nobody is
going to read it or do anything about it. Returning an error should be
fine then. If the duty cycle call fails for 50% then that's something we
have to live with.

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ