On Wed, 9 Feb 2022 at 15:41, Sui Jingfeng <15330273260@xxxxxx> wrote: > On 2022/2/9 16:43, Maxime Ripard wrote: > > More fundamentally (and this extends to the CMA, caching and VRAM stuff > > you explained above), why can't the driver pick the right decision all > > the time and why would that be under the user control? > > The right decision for ls7a1000 is to use VRAM based helper, But sometimes > we need CMA helper based solution. Because: The PRIME support is lost, use > lsdc with etnaviv is not possible any more. > > Buffer sharing with etnaviv is no longer possible, loongson display controllers > are simple which require scanout buffers to be physically contiguous. Other systems have this limitation, and Mesa's 'kmsro' concept makes this work transparently, as long as your driver can export dmabufs when running in 'VRAM' mode. Cheers, Daniel
