[coyotos-dev] driver separation - most of the time?

[Jonathan S. Shapiro]

On Sat, 2005-10-15 at 09:38 -0700, Reuven R. wrote:
> but if separating the drivers from the kernel comes at
> a serious performance cost...

Eric has unintentionally led this discussion in a bad direction.

There is absolutely no evidence that user-mode drivers carries an
intrinsic performance penalty.

There *is* evidence that in some cases a high-performance user-mode
driver needs to be designed differently from the corresponding in-kernel
driver.

I am not aware of *any* justification for an in-kernel driver based on
where the software and hardware architectures stand today.

shap