On Mon, 11 Nov 2024 08:21:45 -0600 David Lechner <dlechner@xxxxxxxxxxxx> wrote: > On 11/11/24 6:08 AM, Uwe Kleine-König wrote: > > [dropped Mircea Caprioru from Cc: as their address bounces.] > > > > Hello David, > > > > On Fri, Nov 08, 2024 at 12:52:35PM -0600, David Lechner wrote: > >> On 11/8/24 12:18 PM, Uwe Kleine-König wrote: > >>> The ad7124-4 and ad7124-8 both support 16 channel registers. Don't > >>> accept more (logical) channels from dt than that. > >> > >> Why should the devicetree be limited by the number of channel > >> registers? Channel registers are a resource than can be > >> dynamically assigned, so it doesn't seem like the devicetree > >> should be specifying that assignment. > > > > Note the device tree isn't limited as I didn't adapt the binding. It's > > just that the driver doesn't bind if too many channels are specified. > > And while your statement about the channels being a dynamic resource is > > right, currently the driver doesn't cope and allocates resources > > statically, and happily assumes there is a CHANNEL_16 register if the > > device tree specifies 17 (or more) logical channels and writes to > > CONFIG_0 then which very likely results in strange effects. > > > > So as long as the driver doesn't implement this (possible) dynamic > > mapping to the CHANNEL registers, it's IMHO right to refuse to bind (or > > alternatively only use the 16 first logical channels). > > > > Best regards > > Uwe > > Understood. It would be nice to implement such dynamic allocation > in the future but as a fix to backport to stable kernels, this makes > sense. Agreed. We do have other drivers that have internal allocators for constrained sequencer resources, but they are complex so not fix material! Jonathan
