On 01/10/2019 10:26, Julien Grall wrote: > Hi Jan, > > On 01/10/2019 10:22, Jan Beulich wrote: >> On 01.10.2019 11:17, Julien Grall wrote: >>> On 01/10/2019 00:21, Andrew Cooper wrote: >>>> On 30/09/2019 21:17, Julien Grall wrote: >>>>> My worry is this gate config gate nothing on Arm so the user may have >>>>> a false sense that it can be used (even though it is clearly BROKEN). >>>>> >>>>> Also the name is quite close to the CONFIG_HARDEN_PREDICTOR on Arm >>>>> and >>>>> may confuse user. Although, I don't have a better name so far :/ >>>> >>>> The "depends on BROKEN" means it will never show up to a user in >>>> menuconfig, which is why it was only CC to x86, and not to rest. >>> >>> What's the long term plan for this option? Are you planning to >>> remove it >>> completely or just the dependencies on BROKEN? >>> >>> My concern is if this option will ever become selectable then it >>> will not doing >>> what's expected on Arm. >>> >>> So, even if in principle it is arch agnostic, it feels to me this >>> option should >>> better be implemented in x86/Kconfig. >> >> I don't think so, no. When BROKEN is to be removed, it ought to be >> replaced by a suitable "depends on HAVE_*", which Arm could choose >> to not select. > > So there are an expectation this option will be used by common code in > the future?
It already is. ARM has stubs for evaluate_nospec() etc. My gut feeling is that the only way this is going to be resolved sanely is with a compiler feature or plugin, at which point it reasonably can be cross-arch. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
