On 26/10/2023 09:19, Huacai Chen wrote: > Hi, Krzysztof > > On Wed, Oct 25, 2023 at 3:16 PM Krzysztof Kozlowski > <krzysztof.kozlowski@xxxxxxxxxx> wrote: >> >> On 25/10/2023 03:56, Huacai Chen wrote: >>> Hi, Krzysztof, >>> >>> On Fri, Oct 20, 2023 at 8:18 PM Marc Zyngier <maz@xxxxxxxxxx> wrote: >>>> >>>> On Fri, 20 Oct 2023 10:51:35 +0100, >>>> Binbin Zhou <zhoubb.aaron@xxxxxxxxx> wrote: >>>>> >>>>> Hi Krzysztof & Marc: >>>>> >>>>> Sorry for the interruption. >>>>> As said before, we tried to use the 'interrupt-map attribute' in our >>>>> Loongson liointc dts(i), but there are some unfriendly points. >>>>> Do you have any other different suggestions? >>>> >>>> I don't have any suggestion, but if you are still thinking of adding >>>> some extra crap to the of_irq_imap_abusers[] array, the answer is a >>>> firm 'NO'. >>> Excuse me, but as described before, 'interrupt-map' cannot be used for >>> liointc unless adding it to of_irq_imap_abusers[], can we still use >>> 'parent_int_map' in this case? Or just change it to 'parent-int-map' >>> to satisfy the naming style? >> >> Why do you respond to me? You received firm 'NO' about >> of_irq_imap_abusers, so how adhering to naming style or violating naming >> style has anything to do with it? > I'm sorry but of_irq_imap_abusers is to make 'interrupt-map' to work, > without of_irq_imap_abusers we can only use the existing > 'parent_int_map'. We need your response because we want to know > whether you can accept the existing method since the other approach > has received 'NO'. And, changing 'parent_int_map' to 'parent-int-map' > can be a little better, at least it satisfies the naming style. Indeed, interrupt-map might not fit here. I don't know whether your custom property - purely for runtime performance purpose - will be accepted. Initial description of this field suggested that it is OS policy, not hardware choice. But sure, propose something with justification, so we can review it. The proposal must not break ABI, so you must support both parent_int_map and parent-int-map (or whatever we call it) properties. The first we will probably deprecate. The way this property was sneaked into kernel bypassing review is still disappointing. Best regards, Krzysztof
