On Monday 15 December 2014 11:16:31 Ray Jui wrote: > On 12/12/2014 9:21 AM, Arnd Bergmann wrote: > > On Friday 12 December 2014 09:08:48 Ray Jui wrote: > > One way to solve this would be by turning the driver into a library > > the same way as the pcie-dw driver, and have separate front-ends > > for it for platform_device and bcma_device. > > > I'm fine with this solution, i.e., to introduce a common pcie-iproc core > driver (just like pcie-designware) and have different front-ends > depending on the device/bus type. If we end up deciding to go with this > solution, I need to discuss with Hauke to come up with a plan to > collaborate. Ok > But before we choose to go with that route, may I ask, what is the > purpose of tying a PCIe host driver to BCMA? What benefit does BCMA give > us? If we have a generic platform based PCIe driver that can work on all > iProc SoCs + BCM4708 and BCM5301X with all HW specific configurations > taken care of by device tree, why do we still need to use BCMA? > > I thought all a BCMA device here does is to auto-instantiate based on > some register readings? Basically, DT is a hack that we only need for nondiscoverable buses. As BCMA is (mostly) discoverable, we should use that, just like we do for things like PCI and USB. There are clearly some limitations to BCMA as well, e.g. on bcm4708 we don't have proper IRQ numbers in the device list and we need to work around that using DT, but overall it still seems to have more upsides than downsides to use it. It's also good to point other SoC makers to Broadcom as a good example for how to do things they claim are impossible to do. Arnd -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
