From mboxrd@z Thu Jan 1 00:00:00 1970 From: vladimir.murzin@arm.com (Vladimir Murzin) Date: Wed, 20 Dec 2017 12:22:16 +0000 Subject: [PATCH] ARM: NOMMU: Setup VBAR/Hivecs for secondaries cores In-Reply-To: <20171220114924.GA5377@afzalpc> References: <1513679029-5915-1-git-send-email-vladimir.murzin@arm.com> <20171219112954.GA14910@afzalpc> <4e80e47a-d7b4-c09c-fe88-a7dff6522038@arm.com> <20171220045219.GA6416@afzalpc> <20171220114924.GA5377@afzalpc> Message-ID: To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On 20/12/17 11:49, afzal mohammed wrote: > Hi, > > On Wed, Dec 20, 2017 at 09:55:00AM +0000, Vladimir Murzin wrote: > >>>> I caught it when was trying to setup VBAR and after code inspection I >>>> noticed that setting of Hivecs were changed as well. >>> >>> Thinking again about this, should the Hivecs setting on secondary >>> CPU's be done (till a requirement comes) ? >>> >>> ARM ARM deprecates using Hivecs setting on ARMv7-R, so this issue >>> might not be hit in practice for R class. While pre-ARMv7, lack of >>> Hivecs setting for secondaries, it seems can affect only ARMv6k >>> (multi-processing support added here ?) and i am making a guess that >>> even if there are ARMv6k with more than one core available, they might >>> not yet have run with MMU disabled to hit this case, probably the >>> reason no one has reported issue for long. >> >> I've just reported an issue, no? :) > > By reported, i meant whether the lack of this in secondary would cause > an issue at run time in any of the platform's. You spotted it by code > inspection rather than hitting any issue in practice is what i > understood. > >>> Perhaps, we can avoid configuring Hivecs for secondaries until some >>> one needs it ? >> >> Well, before ad475117d201, Hivec would be enabled for secondaries via >> >> secondary_startup >> -> __after_proc_init >> >> after that commit it is not true, so it is kind of regression. > > Something that was done before that commit not being done later > (though unintentionally) per se doesn't count as regression in my > opinion. But if any platform really needs to gets this done or > misbehaves due to my change, then certainly it has to be counted a > regression, which i believe is not the case here. > >> >> Additionally, patch is not about Hivec only, but VBAR as well and TBH I >> don't follow what is your proposal... > > i was referring to the fact that vector remapping can't be done in > Cortex-R, as security extension is a requisite for this feature, which > Cortex-R don't have on ARMv7. For instance, just think of ARMv7A with 1:1 MMU running in SMP... Vladimir > > afzal >