[RFC PATCH 1/6] server: Allow calling async_handoff() with status code STATUS_ALERTED.

[Jinoh Kang]

On 1/27/22 08:52, Zebediah Figura (she/her) wrote:
> On 1/23/22 11:29, Jinoh Kang wrote:
>> +static int async_add_queue( struct object *obj, struct wait_queue_entry *entry )
>> +{
>> +    struct async *async = (struct async *)obj;
>> +    assert( obj->ops == &async_ops );
>> +
>> +    if (!async->pending && async->terminated && async->alerted)
>> +    {
>> +        /* The client has failed to complete synchronously (e.g. EWOULDBLOCK).
>> +         * Restart the async as fully fledged asynchronous I/O, where
>> +         * the completion port notification and APC call will be triggered
>> +         * appropriately. */
>> +        async->pending = 1;
>> +
>> +        /* Unset the signaled flag if the client wants to block on this async. */
>> +        if (async->blocking) async->signaled = 0;
>> +
>> +        async_set_result( obj, STATUS_PENDING, 0 );  /* kick it off */
>> +    }
>> +
>> +    return add_queue( obj, entry );
>> +}
>> +
> 
> I'll admit, this kind of thing is why I didn't really want to have to try to optimize 3 server calls into 2.

I concur.  Hence,
> (However, if it turns out that this goal is not of utmost significance, then
>  this patch serie can be easily modified so that it issues separate server
>  calls.)

> Asyncs are already really complicated,

To be fair, asynchronous I/O is inherently complicated in itself.

> in terms of the many paths they can take,

Although that one has a lot of room for improvement, yes.

> and it seems like no matter what we do they're going to get worse.
> 
> Still, I have a potential idea.
> 
> What we need to do here is similar to the infrastructure that already exists for device asyncs, namely "unknown_status" etc. It would be nice to use that instead of reinventing it, and although I haven't tried, it seems possible.

That one was on the table, too.  In fact it can also help eliminate the initial_status == STATUS_ALERTED check.

One catch is that async_set_unknown_status also sets direct_result to 0, which means to always fire off APC on completion.
I wasn't entirely sure of what the effects of { .unknown_status = 1, .direct_result = 1 } would be.

> 
> async_add_queue() as it is above is not great. I'm not sure that code actually works in every case;

!pending && terminated && alerted was the condition I was able to deduce to detect this exact condition.
It does sound a little arbitrary though, especially since it's testing for three unrelated conditions.

> it definitely increases the mental burden even if it does. (Consider for instance that it would be triggered for *every* async).
> 
> Instead what I'd suggest is to use the request introduced here in every case, even if the initial status was pending.

You mean, along with use of unknown_status?

> This introduces a new server call, but it only does so in cases where we already have to wait.

Sounds reasonable.

> 
> The end result would be not unlike get_next_device_request, which is essentially that request combined with some other things (and it doesn't deal in async object handles).


-- 
Sincerely,
Jinoh Kang