| 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
| ||
|
Message-ID: <54FA08F1.7090502@broadcom.com> Date: Fri, 6 Mar 2015 12:07:13 -0800 From: Ray Jui <rjui@...adcom.com> To: Mike Turquette <mturquette@...aro.org>, Sascha Hauer <s.hauer@...gutronix.de> CC: Stephen Boyd <sboyd@...eaurora.org>, Matt Porter <mporter@...aro.org>, Alex Elder <elder@...aro.org>, Rob Herring <robh+dt@...nel.org>, Pawel Moll <pawel.moll@....com>, Mark Rutland <mark.rutland@....com>, Ian Campbell <ijc+devicetree@...lion.org.uk>, Russell King <linux@....linux.org.uk>, Arnd Bergmann <arnd@...db.de>, <linux-arm-kernel@...ts.infradead.org>, <devicetree@...r.kernel.org>, Scott Branden <sbranden@...adcom.com>, "Dmitry Torokhov" <dtor@...gle.com>, Anatol Pomazau <anatol@...gle.com>, <linux-kernel@...r.kernel.org>, <bcm-kernel-feedback-list@...adcom.com> Subject: Re: [PATCH v5 1/6] clk: add of_clk_get_parent_rate function Hi Mike, On 3/6/2015 11:55 AM, Mike Turquette wrote: > Quoting Sascha Hauer (2015-02-26 00:43:19) >> On Wed, Feb 25, 2015 at 11:42:44PM -0800, Ray Jui wrote: >>> On 2/25/2015 10:51 PM, Sascha Hauer wrote: >>>> On Wed, Feb 25, 2015 at 10:13:15PM -0800, Ray Jui wrote: >>>>> Hi Sascha, >>>>> >>>>> On 2/25/2015 9:54 PM, Sascha Hauer wrote: >>>>>> Hi Ray, >>>>>> >>>>>> On Wed, Feb 04, 2015 at 04:55:00PM -0800, Ray Jui wrote: >>>>>>> Sometimes a clock needs to know the rate of its parent before itself is >>>>>>> registered to the framework. An example is that a PLL may need to >>>>>>> initialize itself to a specific VCO frequency, before registering to the >>>>>>> framework. The parent rate needs to be known, for PLL multipliers and >>>>>>> divisors to be configured properly. >>>>>>> >>>>>>> Introduce helper function of_clk_get_parent_rate, which can be used to >>>>>>> obtain the parent rate of a clock, given a device node and index. >>>>>> >>>>>> I can't see how this patch helps you. First it's not guaranteed that >>>>>> the parent is already registered, what do you do in this case? >>>>> >>>>> In the case when clock parent is not found, as you can see from the >>>>> code, it simply returns zero, just like other clk get rate APIs. >>>> >>>> Yes, but what do you do with the 0 result then in your PLL initialization? >>>> >>> >>> As of the current code, it fails the PLL frequency initialization and >>> bails out. Thinking about it more, it actually makes more sense to just >>> warn and still go ahead to register the clock, in which case it will use >>> whatever default frequency after chip power on reset or a frequency >>> configured in the bootloader. >>> >>>>> >>>>> I thought the order of clock registration is based on order of the clock >>>>> nodes in device tree. It makes sense to me to declare the parent clock >>>>> before a child clock, so it's guaranteed that the parent is registered >>>>> before the child. >>>> >>>> No, you can't rely on that. The order of the device nodes may happen to >>>> define the order of clock initialization now, but that may change. >>>> device nodes are usually ordered by bus addresses, not by intended >>>> initialization order. Even if you reorder them everything must still >>>> work. >>>> >>> >>> Okay I get your point that the order of device nodes may not be relied >>> on for device initialization order. But then another mechanism should be >>> deployed to give developers the option to decide on the clock >>> initialization sequence. It can be optional but it should be there. >>> >>>>> >>>>>> Then the clock framework doesn't require that you initialize the PLL >>>>>> before registering. That can be done in the clk ops later. >>>>> >>>>> Sure it's not mandatory. But what's wrong with me choosing to initialize >>>>> the PLL clock to a known frequency before registering it to the framework? >>>> >>>> Appearantly you don't know the (input) frequency of the PLL when >>>> registering it to the framework, so the question must be: What's wrong >>>> with keeping it uninitialized? >>>> >>>> If the PLL is unused then you don't care about it's initialization >>>> status. If it happens to be enabled by a bootloader and still unused >>>> at late_initcall time the clock framework will disable it so you >>>> have a known state then. If a consumer for the PLL appears it's its >>>> job to initialize it through the clk api. >>>> >>>> Sascha >>>> >>> >>> Okay, what we need here is to initialize the PLL to a desired frequency, >>> based on device tree settings (since it will be configured differently, >>> among different boards). This is a PLL that 1) has limited options of >>> frequencies which it can be configured to, and 2) has multiple child >>> clocks, where is a more suitable place to initialize it to the desired >>> frequency than right before registering it to the framework? I know a >>> lot of people do it in the bootloader, but I thought we should be given >>> the flexibility of configuring it in the kernel. >>> >>> When you say "consumers", do you mean 1) the device driver that uses the >>> PLL; or 2) the device driver that use the child clock of the PLL? If >>> it's case 1), then we don't really have a device driver that directly >>> uses the PLL, and I thought that's quite normal, as most PLLs don't >>> directly feed into any peripherals. >> >> I meant 1) and 2). Before a consumer comes along the state of the PLL >> doesn't matter. When a consumer shows up it has to call >> clk_prepare_enable which (directly or indirectly) will enable your PLL. >> Then it's still time to apply the default settings you found out during >> probe of the PLL. > > My review comments are really for iproc_pll_setup() in patch #3, but the > discussion is here so I'll respond to this thread. > > I think the root of this problem is that your pll clk_ops does not > support .set_rate. That is why your clock driver hacks in a call to > pll_set_rate in iproc_pll_setup. > > Due to the above shortcoming you also do not use the assigned-clock-rate > infrastructure to set your pll rate at registration-time. There is no > reason for your driver to re-invent this logic. iproc_pll_setup is > fetching the clock-frequency property from DT and then trying to set > that rate. Instead please use the generic code. > > The right way to handle this is to support a .set_rate callback (looks > like you're 90% of the way there with pll_set_rate) and then use the > assigned-clock-rates property to specify this from DT. > > Regards, > Mike > Okay. It's good to know that "assigned-clock-rate" can be used and serve exactly what we need here. I'll update my patch series to use that instead. In this case, do you think I should still keep of_clk_get_parent_rate in the patch series? Thanks, Ray >> >> Sascha >> >> -- >> Pengutronix e.K. | | >> Industrial Linux Solutions | http://www.pengutronix.de/ | >> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | >> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@...r.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists