On 08/11/2015 12:06 PM, Russell King - ARM Linux wrote: > I think what people need to learn is that an API in the kernel which > returns an int _can_ fail - it returns an int so it _can_ return an > error code. If it _can_ return an error code, there _will_ be > implementations which _do_. > > If you don't check the return code, either your code doesn't care whether > the function was successful or not, or you're playing with fire. This is > a prime example of playing with fire. > > Let's leave the crappy userspace laziness with regard to error checking > to userspace, and keep it out of the kernel. > > Yes, the DMA engine capabilities may not be sufficient to describe every > detail of DMA engines, but that's absolutely no reason to skimp on error > checking. Had there been some kind of error checking at the site, this > problem would have been spotted before the 8250-omap driver was merged. Let me disable RX-DMA in 8250-omap code and push that stable. Then we won't need a special annotation for pause support because it remains off and is currently about one user. I browsed each driver in drivers/dma each one which does support pause supports it and all of them implement it unconditionally (ipu_idmac grabs a mutex first but this is another story). Adding error checking to 8250-omap like I have it in #1 and disabling RX-DMA in case pause fails looks be reasonable since there is nothing else that can be done I guess. Once we have the missing piece in omap-dma the RX-DMA can be enabled in 8250-omap. Does this sound like a plan we can agree on? Sebastian -- To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
