TODO list item - multiple poll waitqueues
Jens Axboe
maintainerHi,
This has been on my TODO list for a while, just haven't gotten around to it.
The issue is that some drivers use multiple waitqueues for poll, which doesn't work with POLL_ADD. io_poll_queue_proc() checks for this and fails it:
static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
struct poll_table_struct *p)
{
struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
if (unlikely(pt->req->poll.head)) {
pt->error = -EINVAL;
return;
}
pt->error = 0;
pt->req->poll.head = head;
add_wait_queue(head, &pt->req->poll.wait);
}since we just have the one waitqueue on the io_uring side. Most notably affected are TTYs, I've also noticed that /dev/random does the same thing, and recently pipes as well.
This is a problem for event handlers, in that not all file types work reliably with POLL_ADD. Note that this also affects the aio poll implementation, unsurprisingly.
If anyone has the inclination to look into this, that'd be great.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerQuoted 28 lines · Jens Axboe
Hi,
This has been on my TODO list for a while, just haven't gotten around to it.
The issue is that some drivers use multiple waitqueues for poll, which doesn't work with
POLL_ADD.io_poll_queue_proc()checks for this and fails it:static void io_poll_queue_proc(struct file *file, struct
wait_queue_head*head,
structpoll_table_struct*p){ struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); if (unlikely(pt->req->poll.head)) { pt->error = -EINVAL; return; } pt->error = 0; pt->req->poll.head = head; add_wait_queue(head, &pt->req->poll.wait); }since we just have the one waitqueue on the
io_uringside. Most notably affected are TTYs, I've also noticed that/dev/randomdoes the same thing, and recently pipes as well.
Quoted 5 lines
This is a problem for event handlers, in that not all file types work reliably with
POLL_ADD. Note that this also affects the aio poll implementation, unsurprisingly.If anyone has the inclination to look into this, that'd be great.
Of course after sending this one, I did take a look and hacked up some support for it. Not super happy with it, but I'll send it out for proper review and hopefully we can turn it into something that's 5.6 material.
Forgot to mention that this is particularly sucky since pipes were recently switched to using multiple waitqueues.
https://git.kernel.dk/cgit/linux-block/log/?h=io_uring-5.6-poll
Signature · 1 line
Jens Axboe