On Thu, Feb 23, 2012 at 1:28 AM, Tomi Valkeinen <tomi.valkeinen@xxxxxx> wrote: > > Hi, > > This is an old issue, and I've mentioned earlier, but I think we should > finally do something about this. Or perhaps this can already be done, > but I don't know the solution. I agree with you. We need to address the L3 bandwidth requirement for DSS as needed by high resolution panels. > > The problem is that various DSS clocks have max frequencies that depend > on whether we are at OPP100 or OPP50 (e.g. table 10-4. DSS Clock > Frequencies in OMAP4 TRM). Some of these clocks come from the PRCM (like > DSS_CLK on OMAP4), but some are generated internally by DSS (like pixel > clocks). > > While the clocks coming from PRCM could perhaps be handled automatically > (from DSS's point of view) by the PM/clock framework so that OPP50 would > be disabled if the clocks are over the OPP50-limit, I guess that's not > quite correct. The max freq limitation is inside DSS HW, not in the PRCM > side (well, I'm guessing so). And as we anyway need to handle some of > the clock limits inside DSS, I guess it's simpler to handle them all in > DSS. > > But the main question is, how should it be done? Afaik the dss driver > cannot say "stay in OPP100" to the PM framework. I guess the closest > thing is to set a constraint to memory throughput with an arbitrarily > high value, thus forcing OPP100. > > But even if it's quite easy to come up with an arbitrary value that > should be high enough for years to come, this approach feels rather > hacky. Can it have side effects? May it cause overclocking, as the PM > framework thinks we need as much bandwidth as possible? Does it cause > higher power consumption because something in L3/Core/somewhere is kept > in a high perf mode for the high mem throughput, even if we don't > actually need high throughput? I have been thinking to determine the L3 bandwidth requirement based on the number of overlays/pipes enabled, width, height and color format and use the PM API omap_pm_set_min_bus_tput() to request the required L3 bandwidth. This bandwidth request and release has to be done for each frame. I have been looking for right entry and exit functions in DSS drivers to add this L3 bandwidth request/release. Regards, Ragha -- 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
