On Wed, Mar 22, 2023 at 11:34:40AM +0200, Andy Shevchenko wrote: > On Wed, Mar 22, 2023 at 07:56:11AM +0000, Niyas Sait wrote: > > On 21/03/2023 17:53, Andy Shevchenko wrote: > > > > > can_we_avoid_long_name_of_the_functions_please() ? > > > > Sure, will do that. > > > > > Also is this renaming is a must? > > > > It is not a must. I considered the existing method with shared info > > as a special case as it uses non standard descriptors from CSRT table > > and introduced the new function to handle it. > > > > > Btw, what is the real argument of not using this table? > > > > > > Yes, I know that this is an MS extension, but why ARM needs something else and > > > why even that is needed at all? CSRT is only for the _shared_ DMA resources > > > and I think most of the IPs nowadays are using private DMA engines (or > > > semi-private when driver based on ID can know which channel services which > > > device). > > > > The issue is that shared info descriptor is not part of CSRT definition [1] > > and I think it is not standardized or documented anywhere. > > > > I was specifically looking at NXP I.MX8MP platform and the DMA lines for > > devices are specified using FixedDMA resource descriptor. I think other Arm > > platforms like RPi have similar requirement. > > Perhaps, but my question is _why_ is it so? > I.o.w. what is the technical background for this solution. > > > [1] https://uefi.org/sites/default/files/resources/CSRT%20v2.pdf JFYI: ARM platform(s) use SPCR, which is also not a part of the specification. -- With Best Regards, Andy Shevchenko