On 29/06/2026 06:02, Mikko Perttunen wrote: > On Friday, June 26, 2026 7:26 PM Krzysztof Kozlowski wrote: >> On Thu, Jun 25, 2026 at 06:26:43PM +0900, Mikko Perttunen wrote: >>> On Thursday, June 25, 2026 5:36 PM Krzysztof Kozlowski wrote: >>>> On Mon, Jun 22, 2026 at 05:22:51PM +0900, Mikko Perttunen wrote: >>>>> Channels and syncpoints available may be limited when other system >>>> >>>> What are channels and syncpoints? >>> >>> Very host1x-specific hardware resources. Somewhat close to GPU channels >>> and semaphores / fences if you're familiar with those. >>> >>>> >>>>> components are using them. Add properties nvidia,channels and >>>>> nvidia,syncpoints to limit the range of usable channels and/or >>>>> syncpoints. >>>> >>>> Why isn't this deducible from the compatible? >>> >>> When Host1x is partitioned, the hypervisor (or other software entity >>> with access to the non-vm host1x register regions) can configure which >>> channels and syncpoints are available to each partition arbitrarily. >> >> So you will have different boards for the same SoC with different >> values? >> > > Yes -- even the same board will have different values depending on the > use case of the system. The hypervisor or bootloader has to update the > properties according to the configuration it has programmed in the > hardware. > > I will update the commit message with clarification. > >>> >>> How that partitioning was done is not discoverable through hardware >>> directly, so we need to pass that information through device tree. >>> >>>> >>>> Also, nvidia,channels is too broad/generic. This is not ADC, right? And >>>> channels is a common term in IIO. And in few other cases. >>> >>> It's not ADC. Yes, it's a common term in a lot of contexts. I can call >>> it nvidia,host1x-channels if you prefer but that seems a bit redundant >>> given the context. >> >> host1x is indeed poor, but maybe there is some way to be a bit more >> descriptive? message-channels? > > I would prefer to keep these as is -- anyone working with host1x will > know what 'channels' and 'syncpoints' mean, and changing the name could > cause confusion since the name would not match other code or > documentation discussing these. > > I can update the binding documentation with some more explanation about > what these properties mean.
Property has only one meaning in DT schema, so basically you are now fixing nvidia,channels to uint32-array for all possible use cases and nvidia devices. Well, you have been warned. Best regards, Krzysztof
