On Mon, Jul 01, 2019 at 01:24:39PM +0200, Maxime Chevallier wrote: > I agree that this the real solution, it just seemed a bit overwhelming > to me. I'll be happy to help with this though, now that you took a big > first step.
I think the first step is to resurrect my original patch to default to a 32-bit DMA mask for platform devices, as that will cut a lot of crap from the platform device declarations. IIRC the problem back then was that USB uses the fact that a DMA mask exist to decide if it uses a DMA vs PIO path in the HCD core. So I'll need some help from Greg or other USB folks to clean that up, after that we can try to apply my patch again (preferably early in the next merge window), and once that sticks clean up all the 32-bit dma mask initialization for platform devices, and then turn the dma_mask into a scalar.
