On 1/2/26 1:59 PM, Bryan O'Donoghue wrote: > On 02/01/2026 12:00, Konrad Dybcio wrote: >>> Is there an io-fabric in the world which exceeds 1 microsecond to perform a >>> write transaction ? >> Writes on arm64 aren't usually observable from the remote endpoint when >> you would expect them to, they can be buffered unless there's an explicit >> readback right afterwards (which creates a dependency that the processor >> will fulfill) > > I don't mean write-combining cache, I mean posted versus non-posted writes > which is a feature of the front-side-bus :)
Writes are posted (Device-nGnRE, note the E attribute is set) https://documentation-service.arm.com/static/63a43e333f28e5456434e18b Konrad
