From: Rob Herring <robh@kernel.org> To: Kuldeep Singh <singh.kuldeep87k@gmail.com> Cc: Robin Murphy <robin.murphy@arm.com>, Marc Zyngier <maz@kernel.org>, Daniel Lezcano <daniel.lezcano@linaro.org>, Thomas Gleixner <tglx@linutronix.de>, Marc Zyngier <marc.zyngier@arm.com>, Mark Rutland <mark.rutland@arm.com>, linux-kernel@vger.kernel.org, devicetree@vger.kernel.org, linux-arm-kernel@lists.infradead.org Subject: Re: [PATCH v2 2/3] dt-bindings: timer: Document arm, cortex-a7-timer in arch timer Date: Fri, 25 Mar 2022 16:23:18 -0500 [thread overview] Message-ID: <Yj4yxjQ2rQ4HW4uR@robh.at.kernel.org> (raw) In-Reply-To: <20220323183544.GA95717@9a2d8922b8f1> On Thu, Mar 24, 2022 at 12:05:44AM +0530, Kuldeep Singh wrote: > On Mon, Mar 21, 2022 at 11:52:27AM +0000, Robin Murphy wrote: > > On 2022-03-20 18:47, Rob Herring wrote: > > > On Fri, Mar 18, 2022 at 02:55:08AM +0530, Kuldeep Singh wrote: > > > > On Thu, Mar 17, 2022 at 08:25:12PM +0000, Robin Murphy wrote: > > > > > On 2022-03-17 19:15, Kuldeep Singh wrote: > > > > > > Renesas RZ/N1D platform uses compatible "arm,cortex-a7-timer" in > > > > > > conjugation with "arm,armv7-timer". Since, initial entry is not > > > > > > documented, it start raising dtbs_check warnings. > > > > > > > > > > > > ['arm,cortex-a7-timer', 'arm,armv7-timer'] is too long > > > > > > 'arm,cortex-a7-timer' is not one of ['arm,armv7-timer', 'arm,armv8-timer'] > > > > > > 'arm,cortex-a7-timer' is not one of ['arm,cortex-a15-timer'] > > > > > > > > > > > > Document this compatible to address it. The motivation to add this > > > > > > change is taken from an already existing entry "arm,cortex-a15-timer". > > > > > > Please note, this will not hurt any arch timer users. > > > > > > > > > > Eh, if it's never been documented or supported, I say just get rid of it. > > > > > The arch timer interface is by definition part of a CPU, and we can tell > > > > > what the CPU is by reading its ID registers. Indeed that's how the driver > > > > > handles the non-zero number of CPU-specific errata that already exist - we > > > > > don't need compatibles for that. > > > > > > > > > > In some ways it might have been nice to have *SoC-specific* compatibles > > > > > given the difficulty some integrators seem to have had in wiring up a stable > > > > > count *to* the interface, but it's not like they could be magically added to > > > > > already-deployed DTs after a bug is discovered, and nor could we have > > > > > mandated them from day 1 just in case and subsequently maintained a binding > > > > > that is just an ever-growing list of every SoC. Oh well. > > > > > > > > Robin, A similar discussion was already done on v1 thread. Please see > > > > below for details: > > > > https://lore.kernel.org/linux-devicetree/20220317065925.GA9158@9a2d8922b8f1/ > > > > https://lore.kernel.org/linux-devicetree/726bde76-d792-febf-d364-6eedeb748c3b@canonical.com/ > > > > > > > > And final outcome of discussion turns out to add this compatible string. > > > > > > I agree with Robin on dropping. More specific here is not useful. If > > > we're going to add some cores, then we should add every core > > > implementation. > > Sure Rob, I will drop A7/15-timer entry from compatibles. > This means only two entries i.e arm,armv7/8-timer will be there under > compatibles now. > > I actually added A7-timer because A15-timer was already present in > binding. Since, it was added by you that's why I added this one. > I will update compatibles accordingly as you said above. The A15 compatible is likely there because upstream dts files used it and it's a judgement call of supporting in the schema vs. making dts changes. Just like the PL022. Maybe there are A7 cases, but fewer to fix. I don't remember. So no real object to removing it, but I can think of better things to work on. Here's a list of most occurring compatibles with no schema[1]. Or find a platform and work towards getting 0 warnings. > > Yeah, what I was trying to convey is that a compatible like > > "arm,cortex-a76-timer" has the problem of being both too specific *and* not > > specific enough to be genuinely useful *for the particular case of the arch > > timer*. It's an architectural interface, where the actual functional > > features are described through the interface itself, so the purpose of the > > DT entry is really just to indicate that the standard interface is present > > and describe how its externally-routed interrupts are wired up. > > > > However, it's also true that implementations of standard functionality > > sometimes have bugs that software needs to know about, but in order for > > specific DT compatibles to be useful in that respect they really need to > > identify the *exact* implementation, e.g. to know that > > "arm,cortex-a76-r0p0-timer" has a bug which needs working around, but > > "arm,cortex-a76-r4p0-timer" does not. There might be cases where every known > > version of a CPU is equally affected (e.g. Cortex-A73), but it doesn't hold > > as a general assumption. Furthermore as mentioned, the other class of bugs > > which affect this interface are not in the CPU's implementation of the > > interface at all, but in the external SoC logic that provides the counter > > value, and therefore it can be identification of the overall SoC that > > matters regardless of which CPU IP(s) may be present. > > > > If we'd had the benefit of 10 years worth of hindsight 10 years ago, we > > probably wouldn't have defined "arm,cortex-a15-timer" either. However the > > fact that we can't erase the legacy of that decision doesn't make an > > argument for repeating it now. > > > > > If one has a big.LITTLE system with A15/A7 what would be the right > > > compatible value? > > > > > > > > > > > I see people with different set of perspective in regard to whether keep > > > > compatible string or not. We should have some sort of evidences to > > > > support claims so that next time when similar situation arises, we'll be > > > > aware beforehand how to proceed. > > > > > > Every situation tends to be different. > > > > Indeed, I certainly don't have a personal perspective of "delete all the > > bindings!" in general - only when they're truly redundant (functionally, any > > driver that can touch the arch timer registers can also read the CPU ID > > registers, but even in the DT there should already be compatibles for the > > CPUs themselves). > > Thanks Robin for providing inputs. > I agree with your opinion of having soc specific compatibles which is > also mentioned under dos and dont's of bindings and other cases will > require investigation though. > https://www.kernel.org/doc/html/latest/devicetree/bindings/writing-bindings.html There's always exceptions to guidelines. This is one of them. Rob [1] https://gitlab.com/robherring/linux-dt/-/jobs/2250856818#L7769
WARNING: multiple messages have this Message-ID (diff)
From: Rob Herring <robh@kernel.org> To: Kuldeep Singh <singh.kuldeep87k@gmail.com> Cc: Robin Murphy <robin.murphy@arm.com>, Marc Zyngier <maz@kernel.org>, Daniel Lezcano <daniel.lezcano@linaro.org>, Thomas Gleixner <tglx@linutronix.de>, Marc Zyngier <marc.zyngier@arm.com>, Mark Rutland <mark.rutland@arm.com>, linux-kernel@vger.kernel.org, devicetree@vger.kernel.org, linux-arm-kernel@lists.infradead.org Subject: Re: [PATCH v2 2/3] dt-bindings: timer: Document arm, cortex-a7-timer in arch timer Date: Fri, 25 Mar 2022 16:23:18 -0500 [thread overview] Message-ID: <Yj4yxjQ2rQ4HW4uR@robh.at.kernel.org> (raw) In-Reply-To: <20220323183544.GA95717@9a2d8922b8f1> On Thu, Mar 24, 2022 at 12:05:44AM +0530, Kuldeep Singh wrote: > On Mon, Mar 21, 2022 at 11:52:27AM +0000, Robin Murphy wrote: > > On 2022-03-20 18:47, Rob Herring wrote: > > > On Fri, Mar 18, 2022 at 02:55:08AM +0530, Kuldeep Singh wrote: > > > > On Thu, Mar 17, 2022 at 08:25:12PM +0000, Robin Murphy wrote: > > > > > On 2022-03-17 19:15, Kuldeep Singh wrote: > > > > > > Renesas RZ/N1D platform uses compatible "arm,cortex-a7-timer" in > > > > > > conjugation with "arm,armv7-timer". Since, initial entry is not > > > > > > documented, it start raising dtbs_check warnings. > > > > > > > > > > > > ['arm,cortex-a7-timer', 'arm,armv7-timer'] is too long > > > > > > 'arm,cortex-a7-timer' is not one of ['arm,armv7-timer', 'arm,armv8-timer'] > > > > > > 'arm,cortex-a7-timer' is not one of ['arm,cortex-a15-timer'] > > > > > > > > > > > > Document this compatible to address it. The motivation to add this > > > > > > change is taken from an already existing entry "arm,cortex-a15-timer". > > > > > > Please note, this will not hurt any arch timer users. > > > > > > > > > > Eh, if it's never been documented or supported, I say just get rid of it. > > > > > The arch timer interface is by definition part of a CPU, and we can tell > > > > > what the CPU is by reading its ID registers. Indeed that's how the driver > > > > > handles the non-zero number of CPU-specific errata that already exist - we > > > > > don't need compatibles for that. > > > > > > > > > > In some ways it might have been nice to have *SoC-specific* compatibles > > > > > given the difficulty some integrators seem to have had in wiring up a stable > > > > > count *to* the interface, but it's not like they could be magically added to > > > > > already-deployed DTs after a bug is discovered, and nor could we have > > > > > mandated them from day 1 just in case and subsequently maintained a binding > > > > > that is just an ever-growing list of every SoC. Oh well. > > > > > > > > Robin, A similar discussion was already done on v1 thread. Please see > > > > below for details: > > > > https://lore.kernel.org/linux-devicetree/20220317065925.GA9158@9a2d8922b8f1/ > > > > https://lore.kernel.org/linux-devicetree/726bde76-d792-febf-d364-6eedeb748c3b@canonical.com/ > > > > > > > > And final outcome of discussion turns out to add this compatible string. > > > > > > I agree with Robin on dropping. More specific here is not useful. If > > > we're going to add some cores, then we should add every core > > > implementation. > > Sure Rob, I will drop A7/15-timer entry from compatibles. > This means only two entries i.e arm,armv7/8-timer will be there under > compatibles now. > > I actually added A7-timer because A15-timer was already present in > binding. Since, it was added by you that's why I added this one. > I will update compatibles accordingly as you said above. The A15 compatible is likely there because upstream dts files used it and it's a judgement call of supporting in the schema vs. making dts changes. Just like the PL022. Maybe there are A7 cases, but fewer to fix. I don't remember. So no real object to removing it, but I can think of better things to work on. Here's a list of most occurring compatibles with no schema[1]. Or find a platform and work towards getting 0 warnings. > > Yeah, what I was trying to convey is that a compatible like > > "arm,cortex-a76-timer" has the problem of being both too specific *and* not > > specific enough to be genuinely useful *for the particular case of the arch > > timer*. It's an architectural interface, where the actual functional > > features are described through the interface itself, so the purpose of the > > DT entry is really just to indicate that the standard interface is present > > and describe how its externally-routed interrupts are wired up. > > > > However, it's also true that implementations of standard functionality > > sometimes have bugs that software needs to know about, but in order for > > specific DT compatibles to be useful in that respect they really need to > > identify the *exact* implementation, e.g. to know that > > "arm,cortex-a76-r0p0-timer" has a bug which needs working around, but > > "arm,cortex-a76-r4p0-timer" does not. There might be cases where every known > > version of a CPU is equally affected (e.g. Cortex-A73), but it doesn't hold > > as a general assumption. Furthermore as mentioned, the other class of bugs > > which affect this interface are not in the CPU's implementation of the > > interface at all, but in the external SoC logic that provides the counter > > value, and therefore it can be identification of the overall SoC that > > matters regardless of which CPU IP(s) may be present. > > > > If we'd had the benefit of 10 years worth of hindsight 10 years ago, we > > probably wouldn't have defined "arm,cortex-a15-timer" either. However the > > fact that we can't erase the legacy of that decision doesn't make an > > argument for repeating it now. > > > > > If one has a big.LITTLE system with A15/A7 what would be the right > > > compatible value? > > > > > > > > > > > I see people with different set of perspective in regard to whether keep > > > > compatible string or not. We should have some sort of evidences to > > > > support claims so that next time when similar situation arises, we'll be > > > > aware beforehand how to proceed. > > > > > > Every situation tends to be different. > > > > Indeed, I certainly don't have a personal perspective of "delete all the > > bindings!" in general - only when they're truly redundant (functionally, any > > driver that can touch the arch timer registers can also read the CPU ID > > registers, but even in the DT there should already be compatibles for the > > CPUs themselves). > > Thanks Robin for providing inputs. > I agree with your opinion of having soc specific compatibles which is > also mentioned under dos and dont's of bindings and other cases will > require investigation though. > https://www.kernel.org/doc/html/latest/devicetree/bindings/writing-bindings.html There's always exceptions to guidelines. This is one of them. Rob [1] https://gitlab.com/robherring/linux-dt/-/jobs/2250856818#L7769 _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
next prev parent reply other threads:[~2022-03-25 21:23 UTC|newest] Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-03-17 19:15 [PATCH v2 0/3] Fix for arch timer users Kuldeep Singh 2022-03-17 19:15 ` Kuldeep Singh 2022-03-17 19:15 ` [PATCH v2 1/3] dt-bindings: timer: Rearrange compatible entries of arch timer Kuldeep Singh 2022-03-17 19:15 ` Kuldeep Singh 2022-03-17 19:15 ` [PATCH v2 2/3] dt-bindings: timer: Document arm,cortex-a7-timer in " Kuldeep Singh 2022-03-17 19:15 ` [PATCH v2 2/3] dt-bindings: timer: Document arm, cortex-a7-timer " Kuldeep Singh 2022-03-17 20:25 ` Robin Murphy 2022-03-17 20:25 ` Robin Murphy 2022-03-17 21:25 ` Kuldeep Singh 2022-03-17 21:25 ` Kuldeep Singh 2022-03-20 18:47 ` Rob Herring 2022-03-20 18:47 ` Rob Herring 2022-03-21 11:52 ` Robin Murphy 2022-03-21 11:52 ` Robin Murphy 2022-03-23 18:35 ` Kuldeep Singh 2022-03-23 18:35 ` Kuldeep Singh 2022-03-25 21:23 ` Rob Herring [this message] 2022-03-25 21:23 ` Rob Herring 2022-04-11 12:35 ` Geert Uytterhoeven 2022-04-11 12:35 ` Geert Uytterhoeven 2022-03-17 19:15 ` [PATCH v2 3/3] ARM: dts: aspeed: Remove arch timer clocks property Kuldeep Singh 2022-03-17 19:15 ` Kuldeep Singh 2022-03-17 19:54 ` Marc Zyngier 2022-03-17 19:54 ` Marc Zyngier 2022-03-17 21:10 ` Kuldeep Singh 2022-03-17 21:10 ` Kuldeep Singh 2022-03-17 21:46 ` Marc Zyngier 2022-03-17 21:46 ` Marc Zyngier 2022-03-18 6:18 ` Joel Stanley 2022-03-18 6:18 ` Joel Stanley 2022-03-18 13:44 ` Krzysztof Kozlowski 2022-03-18 13:44 ` Krzysztof Kozlowski
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=Yj4yxjQ2rQ4HW4uR@robh.at.kernel.org \ --to=robh@kernel.org \ --cc=daniel.lezcano@linaro.org \ --cc=devicetree@vger.kernel.org \ --cc=linux-arm-kernel@lists.infradead.org \ --cc=linux-kernel@vger.kernel.org \ --cc=marc.zyngier@arm.com \ --cc=mark.rutland@arm.com \ --cc=maz@kernel.org \ --cc=robin.murphy@arm.com \ --cc=singh.kuldeep87k@gmail.com \ --cc=tglx@linutronix.de \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.