Re: [lkp-robot] [fs] 3deb642f0d: will-it-scale.per_process_ops -8.8% regression

[Al Viro <viro@ZenIV.linux.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 2521 bytes --]

On Fri, Jun 22, 2018 at 12:00:14PM +0200, Christoph Hellwig wrote:
> And a version with select() also covered:

For fuck sake, if you want vfs_poll() inlined, *make* *it* *inlined*.
Is there any reason for not doing that other than EXPORT_SYMBOL_GPL
fetish?  Because if there isn't, I would like to draw your attention
to the fact that _this_ pwecious inchewlekshul pwopewty can be trivially
open-coded by out-of-tree shite even if it happens to be non-GPL one.

>  					mask = vfs_poll(f.file, wait);
> +					if (f.file->f_op->poll) {

... not to mention that here you forgot to remove the call itself while
expanding it.

Said that, you are not attacking the worst part of it - it's a static
branch, not the considerably more costly indirect ones.  Remember when
I asked you about the price of those?  Method calls are costly.

Another problem with with ->get_poll_head() calling conventions is
that originally you wanted to return ERR_PTR(-mask) as a way to report
not needing to call ->poll_mask(); that got shot down since quite
a few of those don't fit into 12 bits that ERR_PTR() gives us.

IIRC, the real reason for non-constant ->get_poll_head() was the sockets,
with

static struct wait_queue_head *sock_get_poll_head(struct file *file,
                __poll_t events)
{
        struct socket *sock = file->private_data;

        if (!sock->ops->poll_mask)
                return NULL;
        sock_poll_busy_loop(sock, events);
        return sk_sleep(sock->sk); 
}

The first part isn't a problem (it is constant).  The second is
static inline void sock_poll_busy_loop(struct socket *sock, __poll_t events)
{
        if (sk_can_busy_loop(sock->sk) &&
            events && (events & POLL_BUSY_LOOP)) {
                /* once, only if requested by syscall */
                sk_busy_loop(sock->sk, 1);
        }
} 

and the third -

static inline wait_queue_head_t *sk_sleep(struct sock *sk)
{
        BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
        return &rcu_dereference_raw(sk->sk_wq)->wait; 
} 

Now, ->sk_wq is modified only in sock_init_data() and sock_graft();
the latter, IIRC, is ->accept() helper.  Do we ever call either of
those on a sock of already opened file?  IOW, is there any real
reason for socket ->get_poll_head() not to be constant, other
than wanting to keep POLL_BUSY_LOOP handling out of ->poll_mask()?
I agree that POLL_BUSY_LOOP is ugly as hell, but you *still* have
sock_poll_mask() not free from it...