On Fri, May 29, 2026 at 04:39:13PM +0300, Jani Nikula wrote: > On Thu, 28 May 2026, Vinod Govindapillai <[email protected]> > wrote: > > Pcode reports the number of subchannels for LPDDR4 and LPDDR5. > > But the page mask selection table is based on the channels. So > > adjust the number of channels for LPDDR4 and LPDDR5 so that the > > correct page mask can be picked. There are two subchannels per > > channel. > > IMO the parsing of the information belongs in intel_dram.c. If you get > subchannels from pcode or where ever, intel_dram.c should figure it out, > and store the number of channels in ->num_channels.
We actually want subchannels for QGV stuff. And that's what modern pcode gives us, older pcode gave us channels instead and we convert that to subchannels in the QGV code on those platforms. I have a patch to move the adjustment into intel_dram.c for those platforms so that we get consistent data out of intel_dram.c everywhere. For this thing here I think the best thing would be to change the table to use subchannels instead. Looks like Vinod didn't file the Bspec issue asking for the table to be updated there, so I just did that. -- Ville Syrjälä Intel
