Re: [PATCH v6 07/12] usb: chipidea: add a usb2 driver for ci13xxx

[Peter Chen <peter.chen@freescale.com>]

On Thu, Sep 25, 2014 at 09:11:35AM +0200, Arnd Bergmann wrote:
> On Thursday 25 September 2014 09:16:48 Peter Chen wrote:
> > > +     }
> > > +
> > > +     if (dev->of_node) {
> > > +             ret = ci_hdrc_usb2_dt_probe(dev, ci_pdata);
> > > +             if (ret)
> > > +                     goto clk_err;
> > > +     } else {
> > > +             ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
> > > +             if (ret)
> > > +                     goto clk_err;
> > > +     }
> > 
> > Hi Antoine, the above code may not be needed, since get phy and set dma
> > mask are common operation, we can do it at core code, the only thing you
> > need to do is something like: dev->dma_mask = DMA_BIT_MASK(32).
> > 
> 
> Certainly not, doing that would be broken for a number of reasons:
> 
> - dev->dma_mask is a pointer, a driver should not reassign that
> - the device may have been set up for a bus that requires a smaller mask
>   that is already set, changing the mask would cause data corruption
> - setting just the dma mask but not coherent mask is wrong
> - setting the dma mask can fail, e.g. if the mask is smaller than the
>   smallest memory zone, so you have to check the return value.
> 

In current chipidea structure, the parent (glue layer) driver will not be
used for dma, udc/host driver uses dma mask from child (core layer), at core
layer we will do:


	pdev->dev.dma_mask = dev->dma_mask; /* this device is parent */
	dma_set_coherent_mask(&pdev->dev, dev->coherent_dma_mask);

Would you suggestion us a suitable way? Or it is ok we use just Antoine's way that
call dma_coerce_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) at parent
driver no matter dt or non-dt? Thanks.

-- 
Best Regards,
Peter Chen