On Thu Feb 12, 2026 at 10:23 AM CET, Danilo Krummrich wrote: > In particular, I don't think we need an unsafe API until we see a concrete > example where the proposed safe API does not work (and no other safe API would > work either).
One additional note for clarification, I'm not saying that I'm against an unsafe API if it is necessary. In fact, I have added an unsafe API for drivers myself with the dma::Device trait [1]. This was not an easy decision and I discussed this back and forth with a lot of people and, unfortunately, we had to come to the conclusion that any attempt to make this safe from the Rust side of things would cause unreasonable overhead and can't be considered an overall improvement. But as I said, I am convinced that need clear evidence that an unsafe API is actually needed and that there is no reasonable alternative. [1] https://rust.docs.kernel.org/kernel/dma/trait.Device.html
