Re: [patch] epoll use a single inode ...

[Eric Dumazet]

Eric Dumazet a écrit :
> I would definitly *love* saving dentries for pipes (and sockets too), 
> but how are you going to get the inode ?
>
> pipes()/sockets() can use read()/write()/rw_verify_area() and thus need 
> file->f_path.dentry->d_inode (so each pipe needs a separate dentry)
>
> Are you suggesting adding a new "struct file_operations" member to get 
> the inode ?
> Or re-intoducing an inode pointer in struct file ?

Crazy ideas : (some readers are going to kill me)

1) Use the low order bit of f_path.dentry to say : this pointer is not a 
pointer to a dentry but the inode pointer (with the low order bit set to 1)

OR

2) file->f_path.dentry set to NULL for this special files (so that we dont 
need to dput() and cache line ping pong the common dentry each time we 
__fput() a pipe/socket.

Same trick could be used for file->f_path.mnt, because there is a big SMP 
cache line ping/pong to maintain a mnt_count on pipe/sockets mountpoint while 
these file systems cannot be un-mounted)

If dentry is NULL, we get the inode pointer from an overlay of struct 
file_ra_state    f_ra; (because for this special files readahead is unused)

This adds some conditional branches of course, but being able to save ram and 
better use cpu caches might be worth them.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/