On Fri, Nov 13, 2026 at 3:24 AM Patrik Jakobsson wrote: > Clover Trail is not supported in gma500 so perhaps it's better to keep > the SGX driver separated? You and Joseph should try to figure out how > you want to do this together.
Agreed. With the SGX in its own render-only driver, it doesn't matter which display driver sits next to it, so Clover Trail+ wouldn't need gma500 at all. Gabriel, this is how I'd see it, but I'm open to other ideas. The SGX5xx cores have a lot in common: the MMU, the command and job model, the parameter buffer, the PDS. powervr_sgx keeps a table of cores, so the SGX535 on Poulsbo and Oaktrail, and later the SGX544MP2, could be entries in it rather than separate drivers, and share that common part. You've done the SGX535 work and know it far better than I do, so it would make sense for you to maintain that core in the driver. I'd work on the shared parts and on the multi-core support the SGX544MP2 needs. The gma500 side of your series still stands either way: the dumb-buffer size helper and the unpin underflow fix are good gma500 fixes on their own. What would change is that gma500 hands the SGX over to the GPU driver, the way the Cedarview patches in my series do, instead of managing its MMU itself. What do you think? Happy to go through the details here. Joseph
