Re: [PATCH 3/6] drivers/IB,qib: do not use mmap_sem

[Jason Gunthorpe <jgg@ziepe.ca>]

On Mon, Jan 28, 2019 at 04:31:40PM -0700, Jason Gunthorpe wrote:
> On Mon, Jan 21, 2019 at 09:42:17AM -0800, Davidlohr Bueso wrote:
> > The driver uses mmap_sem for both pinned_vm accounting and
> > get_user_pages(). By using gup_fast() and letting the mm handle
> > the lock if needed, we can no longer rely on the semaphore and
> > simplify the whole thing as the pinning is decoupled from the lock.
> > 
> > This also fixes a bug that __qib_get_user_pages was not taking into
> > account the current value of pinned_vm.
> > 
> > Cc: dennis.dalessandro@intel.com
> > Cc: mike.marciniszyn@intel.com
> > Reviewed-by: Ira Weiny <ira.weiny@intel.com>
> > Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
> >  drivers/infiniband/hw/qib/qib_user_pages.c | 67 ++++++++++--------------------
> >  1 file changed, 22 insertions(+), 45 deletions(-)
> 
> I need you to respin this patch/series against the latest rdma tree:
> 
> git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git
> 
> branch for-next
> 
> > diff --git a/drivers/infiniband/hw/qib/qib_user_pages.c b/drivers/infiniband/hw/qib/qib_user_pages.c
> > -static int __qib_get_user_pages(unsigned long start_page, size_t num_pages,
> > -				struct page **p)
> > -{
> > -	unsigned long lock_limit;
> > -	size_t got;
> > -	int ret;
> > -
> > -	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> > -
> > -	if (num_pages > lock_limit && !capable(CAP_IPC_LOCK)) {
> > -		ret = -ENOMEM;
> > -		goto bail;
> > -	}
> > -
> > -	for (got = 0; got < num_pages; got += ret) {
> > -		ret = get_user_pages(start_page + got * PAGE_SIZE,
> > -				     num_pages - got,
> > -				     FOLL_WRITE | FOLL_FORCE,
> > -				     p + got, NULL);
> 
> As this has been rightly changed to get_user_pages_longterm, and I
> think the right answer to solve the conflict is to discard some of
> this patch?

.. and I'm looking at some of the other conversions here.. *most
likely* any caller that is manipulating rlimit for get_user_pages
should really be calling get_user_pages_longterm, so they should not
be converted to use _fast?

Jason