On Mon 07-10-13 15:38:24, Marciniszyn, Mike wrote:
> > > This patch and the sibling ipath patch will nominally take the mmap_sem
> > > twice where the old routine only took it once. This is a performance
> > > issue.
> > It will take mmap_sem only once during normal operation. Only if
> > get_user_pages_unlocked() fail, we have to take mmap_sem again to undo
> > the change of mm->pinned_vm.
> >
> > > Is the intent here to deprecate get_user_pages()?
>
> The old code looked like:
> __qib_get_user_pages()
> (broken) ulimit test
> for (...)
> get_user_pages()
>
> qib_get_user_pages()
> mmap_sem lock
> __qib_get_user_pages()
> mmap_sem() unlock
>
> The new code is:
>
> get_user_pages_unlocked()
> mmap_sem lock
> get_user_pages()
> mmap_sem unlock
>
> qib_get_user_pages()
> mmap_sem lock
> ulimit test and locked pages maintenance
> mmap_sem unlock
> for (...)
> get_user_pages_unlocked()
>
> I count an additional pair of mmap_sem transactions in the normal case.
Ah, sorry, you are right.
> > > Could the lock limit test be pushed into another version of the
> > > wrapper so that there is only one set of mmap_sem transactions?
> > I'm sorry, I don't understand what you mean here...
> >
>
> This is what I had in mind:
>
> get_user_pages_ulimit_unlocked()
> mmap_sem lock
> ulimit test and locked pages maintenance (from qib/ipath)
> for (...)
> get_user_pages_unlocked()
> mmap_sem unlock
>
> qib_get_user_pages()
> get_user_pages_ulimit_unlocked()
>
> This really pushes the code into a new wrapper common to ipath/qib and
> any others that might want to combine locking with ulimit enforcement.
We could do that but frankly, I'd rather change ulimit enforcement to not
require mmap_sem and use atomic counter instead. I'll see what I can do.
Honza
--
Jan Kara <j...@suse.cz>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
