Re: [PATCH] io_uring: save repeated issue_flags

[Pavel Begunkov <asml.silence@gmail.com>]

On 12/5/23 23:11, Jens Axboe wrote:
> On 12/5/23 4:02 PM, Keith Busch wrote:
>> On Tue, Dec 05, 2023 at 03:00:52PM -0700, Jens Axboe wrote:
>>>>   		if (!file->f_op->uring_cmd_iopoll)
>>>>   			return -EOPNOTSUPP;
>>>> -		issue_flags |= IO_URING_F_IOPOLL;
>>>>   		req->iopoll_completed = 0;
>>>>   	}
>>>>   
>>>> +	issue_flags |= ctx->issue_flags;
>>>>   	ret = file->f_op->uring_cmd(ioucmd, issue_flags);
>>>>   	if (ret == -EAGAIN) {
>>>>   		if (!req_has_async_data(req)) {
>>>
>>> I obviously like this idea, but it should be accompanied by getting rid
>>> of ->compat and ->syscall_iopoll in the ctx as well?
>>
>> Yeah, I considered that, and can incorporate it here. Below is a snippet
>> of what I had earlier to make that happen, but felt the purpose for the
>> "issue_flags" was uring_cmd specific and disconnected from everyone
>> else. Maybe I'm overthinking it...
> 
> I'd just do a patch 2 that does compat and syscall_iopoll. And then if

I don't understand why and how you'd get rid of syscall_iopoll,
considering it's not exposed to cmds, and if it is, then we have
a bigger problem. It's a bit sleazily defined and is not just
SETUP_IOPOLL.

> we ever have a new issue flags (as your other series), then it'd become
> natural to add that flag too.
> 
> It's not a hard requirement, but it's somewhat ugly to have the same
> state in two spots. Which is why I'd prefer if we got rid of the actual
> compat/syscall_iopoll members as well, after the conversion is done.

-- 
Pavel Begunkov