Hi Alice,
On Fri, 30 Jan 2026 14:24:10 +0000
Alice Ryhl <[email protected]> wrote:
> +/// A DRM GPU VA manager.
> +///
> +/// This object is refcounted, but the "core" is only accessible using a
> special unique handle. The
> +/// core consists of the `core` field and the GPUVM's interval tree.
> +///
> +/// # Invariants
> +///
> +/// * Stored in an allocation managed by the refcount in `self.vm`.
> +/// * Access to `data` and the gpuvm interval tree is controlled via the
> [`GpuVmCore`] type.
> +#[pin_data]
> +pub struct GpuVm<T: DriverGpuVm> {
> + #[pin]
> + vm: Opaque<bindings::drm_gpuvm>,
> + /// Accessed only through the [`GpuVmCore`] reference.
> + data: UnsafeCell<T>,
> +}
> +
> +// SAFETY: By type invariants, the allocation is managed by the refcount in
> `self.vm`.
> +unsafe impl<T: DriverGpuVm> AlwaysRefCounted for GpuVm<T> {
> + fn inc_ref(&self) {
> + // SAFETY: By type invariants, the allocation is managed by the
> refcount in `self.vm`.
> + unsafe { bindings::drm_gpuvm_get(self.vm.get()) };
> + }
> +
> + unsafe fn dec_ref(obj: NonNull<Self>) {
> + // SAFETY: By type invariants, the allocation is managed by the
> refcount in `self.vm`.
> + unsafe { bindings::drm_gpuvm_put((*obj.as_ptr()).vm.get()) };
> + }
> +}
As discussed on Zulip, in Tyr, we're gonna need Sync+Send on GpuVm<T> if
we want to be able to call some of the thread-safe functions
concurrently (thinking of obtain(), but other read-only info might be
needed to).
Regards,
Boris
