Messages in this thread | | | Date | Tue, 7 Apr 2015 11:25:27 +0100 | From | Mark Rutland <> | Subject | Re: [PATCH v7 1/9] arm64: dts: exynos: Add dts files for 64-bit Exynos5433 SoC |
| |
> >>> I'm very worried about adding a DT that's known broken, especially when > >>> we have no idea as to if/when the FW will be fixed judging from prior > >>> replies. > >> > >> As I replied, I can not fix the FW because I don't have any code of FW. > > > > Surely you are able to contact those who do? > > > >> I don't have any solution to fix it on Linux kernel level. > >> > >> So, If you agree, I can add the comment of CPU0 hotplug issue on DT file. > > > > I disagree. I do not want to add a DT that is known to be broken; > > especially when we have no idea how to fix it. It creates long-term > > maintenance pain for everyone, and marginal gain for few. A comment does > > nothing to aid the end-user. > > > > So NAK to the PSCI node and PSCI enable method in this dts until we > > either have a working firmware, or a reasonable mechanism to handle the > > deficiency. > > There is only CPU0 hotplug issue. Excpet CPU{1-7} are well working.
I understand that, but the issue with CPU0 is still a blocker from my PoV.
> To fix this issue, we must need the help of firmware developer. > But, We never get the any help.
Previously you said that you did not have access to the source code rather than not having help from the relevant firmware engineers. I take it you have informed them of the issue with CPU0?
> Also, as I mentioned on previous mail, all Exynos SoCs can not turn > off the CPU0. I've never seen Exynos SoC that CP0 hotplug is possible.
While that may be the case, PSCI is a more generic standard, and it is used on systems where CPU0 can be hot unplugged. So Exynos-specific details cannot dictate how the kernel PSCI driver should behave.
Is there a particular reason that CPU0 cannot be hotplugged?
In PSCI 0.2 and later it's possible to determine whether a trusted OS prohibits a core from being hotplugged, but this mechanism doesn't exist in earlier versions. I am not averse to adding a property to PSCI 0.1 to mark a CPU as not being hotpluggable if there is a fundamental reason (i.e. not simply a bug) for this being the case.
Thanks, Mark.
| |