Hi Tomi, On Wed, Jan 5, 2011 at 4:57 PM, Tomi Valkeinen <tomi.valkeinen@xxxxxxxxx> wrote: > Hi, > > On Mon, 2011-01-03 at 18:21 +0530, ext Guruswamy Senthilvadivu wrote: >> From: Senthilvadivu Guruswamy <svadivu@xxxxxx> >> >> Hwmod adaptation design requires each of the DSS HW IP to be a platform driver. >> Platform driver of dsshw has to be registered before of dispc, rfbi, dsi1, >> venc and omapdisplay driver should be after all the HW IPs. Sequence it with >> arch_initcall and device_initcall_sync. > > Do you know if there some extra meaning for the _sync initcalls? The > name would imply something else that just init ordering. [Sumit]:Looking at include/linux/init.h and some web-references [1] [2] [3], our understanding is that all inits called via device_initcall_sync() would be called later than the ones called by device_initcall() in an SMP environment. [1] points to the patch for enabling all probes on one level to finish before initcall for next level are processed. > > I don't like using multiple initcall levels in the same driver. It's > already quite complicated with omapdss, omapfb and the panel drivers. > What were the dependencies here? Could we have just one initcall which > would do the registration of platform drivers in the correct order? [Sumit]: With this patchset, dss would have platform drivers for each of the DSS h/w IPs - core would only do panel registration and s/w inits (like dpi and sdi). dispc, dss, rfbi, venc each of these are separate platform drivers which will have their own init(). The idea was to order the inits like this: [bus, then h/w IPs, then s/w blocks]. So, omap_dss_bus_init first => core_initcall() then omap_dss for dss hw ip init => arch_initcall() then any of the other h/w IPs (dsi, rfbi, venc, dispc) in any order => device_initcall(). omap_display device in the end => device_initcall_sync(); this would also call s/w blocks init, and the panel devices. (this should wait for all others to have finished). omapfb is a lateinit call, so will happen after all other dss inits have happened. > > Does this even work if the DSS is compiled as a module? I have the > recollection that a module can only have one initcall, and in this case > omapdss would have many. [Sumit]: we haven't tried compiling DSS as module - I guess we will have to create each dss hw IP also as a separate module. About making one common initcall, then I think we would need to have all platform_driver related functions either as global or in one file. is this a better option than having each as a separate module? I guess I will send the updated patch series once we agree on this. [I have incorporated all other comments from Tony, Kevin and you on the other patches]. Best regards, ~Sumit. > > Tomi > > > -- > 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 > [1]: http://readlist.com/lists/vger.kernel.org/linux-kernel/54/273365.html [2]: http://www.webservertalk.com/archive242-2006-10-1713467.html [3]: http://linux.derkeiler.com/Mailing-Lists/Kernel/2009-01/msg11683.html -- 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
