Hi, On Fri, Apr 25, 2014 at 10:01:51PM +0200, Arnd Bergmann wrote: > On Friday 25 April 2014 16:07:01 Gregory CLEMENT wrote: > > @@ -148,6 +149,9 @@ static int xhci_plat_probe(struct platform_device *pdev) > > if (ret) > > return ret; > > > > + if (of_device_is_compatible(pdev->dev.of_node, "marvell,armada-380-xhci")) > > + xhci_mvebu_mbus_init_quirk(pdev); > > + > > /* Initialize dma_mask and coherent_dma_mask to 32-bits */ > > ret = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32)); > > if (ret) > > I think you're doing it the wrong way around: You have a specialized no, it's the correct way around. > version of the generic xhci-plat driver. The normal way to handle this > is to have a loadable module that contains all the Armada specific > code and that registers a platform_driver. In the probe() function of no, that'll cause too much duplication of boilerplate code. I asked him to write it as a quirk because then we won't end up with and xhci platform_driver for each of the ARM licensees when all they have are small quirks here and there. Imagine if you had one PCI driver for each possible XHCI PCI controller out there. It's pointless maintainenance burden. > that driver, you can do the platform specific setup and then call > the generic xhci_plat_probe() function, which of course has to > be provided using EXPORT_SYMBOL_GPL. yeah, that sucks. Exposing a probe() function like that. I rather have probe() actually *probe* for the HW differences. -- balbi
Attachment:
signature.asc
Description: Digital signature
