On Sat, Nov 18, 2023 at 04:32:26PM +0100, Andrew Lunn wrote:

quoted

One example of not `Send` type (or `!Send`) is spinlock guard:

	let guard: Guard<..> = some_lock.lock();

creating a Guard means "spin_lock()" and dropping a Guard means
"spin_unlock()", since we cannot acquire a spinlock in one context and
release it in another context in kernel, so `Guard<..>` is `!Send`.

Thanks for the explanation. Kernel people might have a different

Surely *we* do, and looks like I created more confusion ;-) Maybe I
should say "execution context" as in include/linux/preempt.h: NMI, hard
IRQ, softirq, task.

meaning for context, especially in this example. We have process
context and atomic context. Process context you are allowed to sleep,
atomic context you cannot sleep. If you are in process context and
take a spinlock, you change into atomic context. And when you release
the spinlock you go back to process context. So with this meaning of
context, you do acquire the spinlock in one context, and release it in
another.

Also as I tried to explain previously, the type of contexts doesn't
matter. Yes, once you hold a spinlock, you enter atomic context, but you
are still in the same task execution context, so acquiring and releasing
in the same task execution doesn't count as "Sending". But if after
acquired one somehow passes the guard to another task, or an interrupt
handler, that's "Sending".

So we are going to have to think about the context the word context is
used in, and expect kernel and Rust people to maybe think of it
differently.

In Rust doc [1], `Send` means:

	Types that can be transferred across thread boundaries.

but of course, we have more "thread-like" things in kernel, so I think
"execution context" may be a better term?

[1]: https://doc.rust-lang.org/core/marker/trait.Send.html

Regards,
Boqun

	Andrew