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: <20211027073458.GA22231@willie-the-truck>
Date:   Wed, 27 Oct 2021 08:34:58 +0100
From:   Will Deacon <will@...nel.org>
To:     Youngmin Nam <youngmin.nam@...sung.com>
Cc:     krzysztof.kozlowski@...onical.com, mark.rutland@....com,
        daniel.lezcano@...aro.org, tglx@...utronix.de,
        linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        linux-samsung-soc@...r.kernel.org, pullip.cho@...sung.com,
        hoony.yu@...sung.com, hajun.sung@...sung.com,
        myung-su.cha@...sung.com, kgene@...nel.org
Subject: Re: [PATCH v1 1/2] clocksource/drivers/exynos_mct_v2: introduce
 Exynos MCT version 2 driver for next Exynos SoC

On Wed, Oct 27, 2021 at 10:38:37AM +0900, Youngmin Nam wrote:
> On Tue, Oct 26, 2021 at 01:00:51PM +0200, Krzysztof Kozlowski wrote:
> > On 26/10/2021 12:45, Youngmin Nam wrote:
> > > On Tue, Oct 26, 2021 at 09:10:28AM +0200, Krzysztof Kozlowski wrote:
> > >> On 26/10/2021 03:47, Youngmin Nam wrote:
> > >>>> If everyone added a new driver to avoid integrating with existing code,
> > >>>> we would have huge kernel with thousands of duplicated solutions. The
> > >>>> kernel also would be unmaintained.
> > >>>>
> > >>>> Such arguments were brought before several times - "I don't want to
> > >>>> integrating with existing code", "My use case is different", "I would
> > >>>> need to test the other cases", "It's complicated for me".
> > >>>>
> > >>>> Instead of pushing a new vendor driver you should integrate it with
> > >>>> existing code.
> > >>>>
> > >>> Let me ask you one question.
> > >>> If we maintain as one driver, how can people who don't have the new MCT test the new driver?
> > >>
> > >> I assume you talk about a case when someone else later changes something
> > >> in the driver. Such person doesn't necessarily have to test it. The same
> > >> as in all other cases (Exynos MCT is not special here): just ask for
> > >> testing on platform one doesn't have.
> > >>
> > >> Even if you submit this as separate driver, there is the exact same
> > >> problem. People will change the MCTv2 driver without access to hardware.
> > >>
> > > Yes, I can test the new MCT driver if someone ask for testing after modifying the new driver.
> > > But in this case, we don't need to test the previous MCT driver. We have only to test the new MCT driver.
> > 
> > Like with everything in Linux kernel. We merge instead of duplicate.
> > It's not an argument.
> > 
> > >> None of these differ for Exynos MCT from other drivers, e.g. mentioned
> > >> Samsung PMIC drivers, recently modified (by Will and Sam) the SoC clock
> > >> drivers or the ChipID drivers (changed by Chanho).
> > > From HW point of view, the previous MCT is almost 10-year-old IP without any major change and
> > > it will not be used on next new Exynos SoC.
> > > MCTv2 is the totally newly designed IP and it will replace the Exynos system timer.
> > > Device driver would be dependent with H/W. We are going to apply a lot of changes for this new MCT.
> > > For maintenance, I think we should separate the new MCT driver for maintenance.
> > > 
> > 
> > There are several similarities which actually suggest that you
> > exaggerate the differences.
> > 
> > The number of interrupts is the same (4+8 in older one, 12 in new one...).
> 
> I didn't "exaggerate" at all.
> The numer of interrups is the same. But their usage is completely different.
> The type of each timer is different.
> And previous MCT can only support upto 8 cores.
> 
> * MCTv1 (Let me call previous MCT as MCTv1)
>  - 4 global timer + 8 local timer
>  - Global timer and local timer are totally different.
>  - 4 global timer have only one 64bit FRC that serves as the "up-counter" with 4 "comparators"
>  - 8 local timer have 8 of 32bit FRC that serves as the "down-counter" without any "comparators".(just expire timer)
>  - local timer can be used as per-cpu event timer, so it can only support upto 8 cores.
> 
> * MCTv2
>  - There are no global timer and local timer anymore.
>  - 1 of 64bit FRC that serves as "up-counter" (just counter without "comparators")
>  - 12 comaprators (These are not "counter") can be used as per-cpu event timer so that it can support upto 12 cores.
>  - RTC source can be used as backup source.
> 
> > You assign the MCT priority also as higher than Architected Timer
> > (+Cc Will and Mark - is it ok for you?)
> >     evt->rating = 500;      /* use value higher than ARM arch timer *
> > 
> Yes, this is absolutely correct on event timer.
> We cannot use arm arch timer which is operating based on PPI as per-cpu event timer because of poewr mode.

You should be able to now that I've added support for per-cpu wakeup timers.

As long as the Arm arch timer is marked as C3STOP (e.g. by sticking the
"local-timer-stop" property in the DT notes), then the MCT will be used
as the wakeup source if you set the CLOCK_EVT_FEAT_PERCPU feature flag.

Give it a try.

Will

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ