On 03/18/2014 08:35 AM, Ohad Ben-Cohen wrote: > Hi Suman, > > On Tue, Mar 18, 2014 at 1:46 AM, Suman Anna <s-anna@xxxxxx> wrote: >> So far, we have not come across multiple controllers. I see your point, >> and I think this also depends on the semantics of how you exchange the >> lock id number. The agreement at the moment is on base_ids across >> multiple SoC components. If the semantics involve exchanging the >> controller instance, for example, then we might get away with it. But >> that probably involves adding additional helpers to retrieve controller >> instance in addition to lock number, or some other similar functions. > > Yes, this could be done too, but I agree it is less simple with no real win. > >> Sorry, I should have rephrased it better - by order, I meant the >> inherent order between board early code and other drivers. With DT, we >> cannot guarantee that right, as specific locks are requested from drivers. > > Yeah. > >> Understood. And we may have to assign the client association with a lock >> as well. These are core changes that were actually not needed in the >> non-DT case due to the inherent order as stated above. So, are you >> suggesting that we add one more property to the controller node to mark >> which are reserved, or rely on constructing this through DT tree parsing? > > I guess this is a question to the DT folks; both approaches work from > hwspinlock perspective. > > In the past Arnd Benoit and myself were happy with adding one more > property to the controller node, but this might be somewhat error > prone as it leaves room for mistakes - developers can add hwlock > phandles and forget to update the reserved property in the controller > node. Ohad, I agree that this is the most simplest form (either a reserved number starting from base, or a reserved range - I prefer the first). The developer errors can be restricted by having the of_hwspin_lock_request_specific() return an error if anything outside this reserved range is requested. Mark, Kumar, Any recommendations/objections on this problem/approach? I also have to bring back the hwlock-base-id property (dropped in v3) for registration purposes, so that the registration does not change based on the probe order of the multiple controller nodes. regards Suman -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
