On 7/8/20 9:02 AM, Matthew Wilcox wrote:
> On Wed, Jul 08, 2020 at 08:59:50AM -0600, Jens Axboe wrote:
>> On 7/8/20 8:58 AM, Matthew Wilcox wrote:
>>> On Wed, Jul 08, 2020 at 08:54:07AM -0600, Jens Axboe wrote:
>>>> On 7/8/20 6:58 AM, Kanchan Joshi wrote:
>>>>>>> +#define IOCB_NO_CMPL (15 << 28)
>>>>>>>
>>>>>>> struct kiocb {
>>>>>>> [...]
>>>>>>> - void (*ki_complete)(struct kiocb *iocb, long ret, long ret2);
>>>>>>> + loff_t __user *ki_uposp;
>>>>>>> - int ki_flags;
>>>>>>> + unsigned int ki_flags;
>>>>>>>
>>>>>>> +typedef void ki_cmpl(struct kiocb *, long ret, long ret2);
>>>>>>> +static ki_cmpl * const ki_cmpls[15];
>>>>>>>
>>>>>>> +void ki_complete(struct kiocb *iocb, long ret, long ret2)
>>>>>>> +{
>>>>>>> + unsigned int id = iocb->ki_flags >> 28;
>>>>>>> +
>>>>>>> + if (id < 15)
>>>>>>> + ki_cmpls[id](iocb, ret, ret2);
>>>>>>> +}
>>>>>>>
>>>>>>> +int kiocb_cmpl_register(void (*cb)(struct kiocb *, long, long))
>>>>>>> +{
>>>>>>> + for (i = 0; i < 15; i++) {
>>>>>>> + if (ki_cmpls[id])
>>>>>>> + continue;
>>>>>>> + ki_cmpls[id] = cb;
>>>>>>> + return id;
>>>>>>> + }
>>>>>>> + WARN();
>>>>>>> + return -1;
>>>>>>> +}
>>>>>>
>>>>>> That could work, we don't really have a lot of different completion
>>>>>> types in the kernel.
>>>>>
>>>>> Thanks, this looks sorted.
>>>>
>>>> Not really, someone still needs to do that work. I took a quick look, and
>>>> most of it looks straight forward. The only potential complication is
>>>> ocfs2, which does a swap of the completion for the kiocb. That would just
>>>> turn into an upper flag swap. And potential sync kiocb with NULL
>>>> ki_complete. The latter should be fine, I think we just need to reserve
>>>> completion nr 0 for being that.
>>>
>>> I was reserving completion 15 for that ;-)
>>>
>>> +#define IOCB_NO_CMPL (15 << 28)
>>> ...
>>> + if (id < 15)
>>> + ki_cmpls[id](iocb, ret, ret2);
>>>
>>> Saves us one pointer in the array ...
>>
>> That works. Are you going to turn this into an actual series of patches,
>> adding the functionality and converting users?
>
> I was under the impression Kanchan was going to do that, but I can run it
> off quickly ...
I just wanted to get clarification there, because to me it sounded like
you expected Kanchan to do it, and Kanchan assuming it "was sorted". I'd
consider that a prerequisite for the append series as far as io_uring is
concerned, hence _someone_ needs to actually do it ;-)
--
Jens Axboe
