Re: [PATCH v2] xen/evtchn: Add design for static event channel signaling

[Julien Grall <julien@xen.org>]

Hi Rahul,

On 04/05/2022 18:34, Rahul Singh wrote:
> This patch introduces a new feature to support the signaling between
> two domains in dom0less system.
> 
> Signed-off-by: Rahul Singh <rahul.singh@arm.com>
> ---
> v2 changes:
> - switch to the one-subnode-per-evtchn under xen,domain" compatible node.
> - Add more detail about event-channel
> ---
>   docs/designs/dom0less-evtchn.md | 126 ++++++++++++++++++++++++++++++++

Answering here to also keep the history. On IRC, Bertrand was asking 
whether we merge design proposal.

We have merged proposal in the past (e.g. non-cooperative migration) and 
I would be ready to do it again as it is easier to find them afterwards.

However, I wonder whether it would be better to turn this proposal to a 
binding change in misc/arm/device-tree/. Any thoughts?

>   1 file changed, 126 insertions(+)
>   create mode 100644 docs/designs/dom0less-evtchn.md
> 
> diff --git a/docs/designs/dom0less-evtchn.md b/docs/designs/dom0less-evtchn.md
> new file mode 100644
> index 0000000000..62ec8a4009
> --- /dev/null
> +++ b/docs/designs/dom0less-evtchn.md
> @@ -0,0 +1,126 @@
> +# Signaling support between two domUs on dom0less system
> +
> +## Current state: Draft version
> +
> +## Proposer(s): Rahul Singh, Bertrand Marquis
> +
> +## Problem Statement:
> +
> +Dom0less guests would benefit from a statically-defined memory sharing and
> +signally system for communication. One that would be immediately available at
> +boot without any need for dynamic configurations.
> +
> +In embedded a great variety of guest operating system kernels exist, many of
> +which don't have support for xenstore, grant table, or other complex drivers.

I am not sure I would consider event channel FIFO a "trival" drivers :).

> +Some of them are small kernel-space applications (often called "baremetal",
> +not to be confused with the term "baremetal" used in the data center which
> +means "without hypervisors") or RTOSes. Additionally, for safety reasons, users
> +often need to be able to configure the full system statically so that it can
> +be verified statically.
> +
> +Event channels are very simple and can be added even to baremetal applications.
> +This proposal introduces a way to define them statically to make them suitable
> +for dom0less embedded deployments.
> +
> +## Proposal:
> +
> +Event channels are the basic primitive provided by Xen for event notifications.
> +An event channel is a logical connection between 2 domains (more specifically
> +between dom1,port1, and dom2,port2). Each event has a pending and a masked bit.
> +The pending bit indicates the event has been raised. The masked bit is used by
> +the domain to prevent the delivery of that specific event. Xen only performs a
> +0 → 1 transition on the pending bits and does not touch the mask bit. The

NIT: I think → is not an ascii character. Can you use "->"?

> +domain may toggle masked bits in the masked bit field and should clear the
> +pending bit when an event has been processed
> +
> +Events are received by a domain via an interrupt from Xen to the domain,
> +indicating when an event arrives (setting the bit). Further notifications are
> +blocked until the bit is cleared again. Events are delivered asynchronously to
> +a domain and are enqueued when the domain is not running.
> +More information about FIFO based event channel can be found at:

I think the explanation is fine for a design proposal. If you want to 
use it as documentation, then I would suggest to clarify there are two 
different ABI for event channel: FIFO and 2L.

2L is the easiest one to implement and for embedded we may want to steer 
the users towards it.

> +https://xenbits.xen.org/people/dvrabel/event-channels-H.pdf

It is quite unfortunate that this wasn't merged in docs/. Oh well, no 
action for you here.

> +
> +The event channel communication will be established statically between two
> +domains (dom0 and domU also) before unpausing the domains after domain creation.
> +Event channel connection information between domains will be passed to XEN via

NIT: above you are using "Xen". So s/XEN/Xen/ for consistency.

> +the device tree node. The event channel will be created and established
> +beforehand in XEN before the domain started. The domain doesn’t need to do any

Same here.

NIT: I think "beforehand" and "before" is redundant.

> +operation to establish a connection. Domain only needs hypercall
> +EVTCHNOP_send(local port) to send notifications to the remote guest.
> +
> +There is no need to describe the static event channel info in the domU device
> +tree. Static event channels are only useful in fully static configurations,
> +and in those configurations the domU device tree dynamically generated by Xen
> +is not needed.
> +
> +Under the "xen,domain" compatible node, there need to be sub-nodes with
> +compatible "xen,evtchn" that describe the event channel connection between two
> +domains(dom0 and domU also).

Below you provided an example between two domUs. Can you provide one 
between dom0 and a domU?

> +
> +The event channel sub-node has the following properties:
> +
> +- compatible
> +
> +    "xen,evtchn"
> +
> +- xen,evtchn
> +
> +    The property is tuples of two numbers
> +    (local-evtchn link-to-foreign-evtchn) where:
> +
> +    local-evtchn is an integer value that will be used to allocate local port
> +    for a domain to send and receive event notifications to/from the remote
> +    domain.
Port 0 is reserved and both FIFO/2L have limit on the port numbers.

I think we should let know the users about those limitations but I am 
not sure whether the binding is the right place for that.

> +
> +    link-to-foreign-evtchn is a single phandle to a remote evtchn to which
> +    local-evtchn will be connected.

I would consider to relax the wording so a user can create an event 
channel with the both end in the same domain.

Implementation wise, it should make no difference as you still need to 
lookup the domain.

> +
> +
> +Example:
> +
> +    chosen {
> +        ....
> +
> +        domU1: domU1 {
> +            compatible = "xen,domain";
> +
> +            /* one sub-node per local event channel */
> +            ec1: evtchn@1 {
> +                compatible = "xen,evtchn-v1";
> +                /* local-evtchn link-to-foreign-evtchn */
> +                xen,evtchn = <0xa &ec3>;
> +            };
> +
> +            ec2: evtchn@2 {
> +                compatible = "xen,evtchn-v1";
> +                xen,evtchn = <0xc &ec4>;
> +            };
> +            ....
> +        };
> +
> +        domU2: domU2 {
> +            compatible = "xen,domain";
> +
> +            /* one sub-node per local event channel */
> +            ec3: evtchn@3 {
> +                compatible = "xen,evtchn-v1";
> +                /* local-evtchn link-to-foreign-evtchn */
> +                xen,evtchn = <0xb &ec1>;
> +            };
> +
> +            ec4: evtchn@4 {
> +                compatible = "xen,evtchn-v1";
> +                xen,evtchn = <0xd &ec2>;
> +            };
> +            ....
> +        };
> +    };
> +
> +In above example two event channel comunication will be established between
> +domU1 and domU2.
> +
> +    domU1 (port 0xa) <-----------------> domU2 (port 0xb)
> +    domU1 (port 0xc) <-----------------> domU2 (port 0xd)
> +
> +domU1 and domU2 can send the signal to remote domain via hypercall
> +EVTCHNOP_send(.) on local port.

Cheers,

-- 
Julien Grall