Re: [PATCH 2/2] rust: time: Implement basic arithmetic operations for Delta

[Alice Ryhl <aliceryhl@google.com>]

On Mon, Jul 28, 2025 at 02:36:50PM -0400, Lyude Paul wrote:
> On Sun, 2025-07-27 at 07:31 +0000, Alice Ryhl wrote:
> > The CONFIG_64BIT restriction seems annoying. Could we not support 32-bit
> > from the get-go? Where is this going to be used?
> > 
> > After all, we have stuff like this:
> > https://lore.kernel.org/r/20250724165441.2105632-1-ojeda@kernel.org
> 
> I'm not really sure how the example is relevant here since we're dealing with
> a different problem. The issue is that division and remainders for u64s are
> not implemented natively on many 32 bit architectures. Also for where it's
> going to be used: currently I'm using it in rvkms for calculating the time of
> the next vblank when we enable/disable our vblank event timer. We basically
> use the duration of a single frame and divide the time elapsed since our
> emulated display was turned on, then use the remainder to figure out how many
> nanoseconds have passed since the beginning of the current scanout frame -
> which we then can just use to figure out the time of the next vblank event.

The reason I bring up the example is that once you add code using these
impls, you're going to get kernel build bot errors from your code not
compiling on 32-bit. And as seen in the linked one, code may be compiled
for 32-bit when setting CONFIG_COMPILE_TEST even if you don't support it
for real.

> This being said, the kernel does have a math library that we can call into
> that emulates operations like this on 32 bit - which I'd be willing to convert
> these implementations over to using. I just put the CONFIG_64BIT there because
> if we do use the kernel math library, I just want to make sure I don't end up
> being the oen who has to figure out how to hook up the kernel math library for
> 64 bit division outside of simple time value manipulation. I've got enough
> dependencies on my plate to get upstream as it is :P

If you just want to call the relevant bindings:: method directly without
any further logic that seems fine to me.

Alice