On Mon, May 18, 2009 at 04:21:19PM +0300, Felipe Contreras wrote: > On Mon, May 18, 2009 at 4:02 PM, Russell King - ARM Linux > <linux@xxxxxxxxxxxxxxxx> wrote: > > On Mon, May 18, 2009 at 03:46:07PM +0300, Felipe Contreras wrote: > >> On Mon, May 18, 2009 at 3:11 PM, Russell King - ARM Linux > >> <linux@xxxxxxxxxxxxxxxx> wrote: > >> > The real problem here seems to be the TI DSP bridge code, and if that's > >> > the case why can't we just avoid registering IVA2 if the TI DSP bridge > >> > code is enabled. That solves your stated problem without creating > >> > additional management issues. > >> > >> The bridgedriver is expected to move and use iommu eventually, but not > >> right now, so I guess the iva2 device should be registered only if > >> MPU_BRIDGE_IOMMU is defined. > >> > >> But then what's the point of having the isp iommu device if the camera > >> driver is disabled? Wouldn't that be wasting resources? Then if CAMERA > >> is not defined the isp device should not be registered either. > > > > So have something like: > > > > config OMAP_IOMMU > > tristate > > > > and then have both MPU_BRIDGE_IOMMU and the camera support (and whatever > > else) select it. That way, you only end up with the IOMMU support code > > built into the kernel if you have users of it. > > > > These low-level internal services drivers really don't need to be > > publically visible in the configuration system. > > Yeap, that needs to be done too. > > > As for the run-time size, that's truely minimal. > > I thought creating iommu devices involved some kind of overhead, > allocating some resources probably. That's why the iva2 iommu device > conflicts with tidspbridge custom mmu. I believe I've already said how to handle that - in fact it's in the quoted messages at the top of this mail. > >> And finally if none of the two are enabled then you don't really > >> iommu. By having omap_iommu_add all the dependencies would be handled > >> automatically, right? 'modprobe bridgedriver' would load iommu. > > > > Think about it - the dependencies _already_ have to be there to use > > the iommu services. > > Ok, yes, for iommu, but not for omap3-iommu which is a separate module. That is a point, but I think it's a relatively minor one. We could get around that by ensuring that omap3-iommu is always built-in if we have the possibility of iommu support, and leave iommu as a module. -- 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
