From mboxrd@z Thu Jan  1 00:00:00 1970
From: Jason Gunthorpe <jgg@ziepe.ca>
Subject: Re: [PATCH v2 05/18] mm/gup: introduce pin_user_pages*() and FOLL_PIN
Date: Mon, 4 Nov 2019 17:15:25 -0400
Message-ID: <20191104211525.GJ30938@ziepe.ca>
References: <20191103211813.213227-6-jhubbard@nvidia.com>
 <20191104173325.GD5134@redhat.com>
 <be9de35c-57e9-75c3-2e86-eae50904bbdf@nvidia.com>
 <20191104191811.GI5134@redhat.com>
 <e9656d47-b4a1-da8a-e8cc-ebcfb8cc06d6@nvidia.com>
 <20191104195248.GA7731@redhat.com>
 <25ec4bc0-caaa-2a01-2ae7-2d79663a40e1@nvidia.com>
 <20191104203153.GB7731@redhat.com>
 <20191104203702.GG30938@ziepe.ca>
 <d0890a8b-c349-0515-2570-10e83979836b@nvidia.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Return-path: <linux-kernel-owner@vger.kernel.org>
Content-Disposition: inline
In-Reply-To: <d0890a8b-c349-0515-2570-10e83979836b@nvidia.com>
Sender: linux-kernel-owner@vger.kernel.org
To: John Hubbard <jhubbard@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>, Andrew Morton <akpm@linux-foundation.org>, Al Viro <viro@zeniv.linux.org.uk>, Alex Williamson <alex.williamson@redhat.com>, Benjamin Herrenschmidt <benh@kernel.crashing.org>, =?utf-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn.topel@intel.com>, Christoph Hellwig <hch@infradead.org>, Dan Williams <dan.j.williams@intel.com>, Daniel Vetter <daniel@ffwll.ch>, Dave Chinner <david@fromorbit.com>, David Airlie <airlied@linux.ie>, "David S . Miller" <davem@davemloft.net>, Ira Weiny <ira.weiny@intel.com>, Jan Kara <jack@suse.cz>, Jens Axboe <axboe@kernel.dk>, Jonathan Corbet <corbet@lwn.net>, Magnus Karlsson <magnus.karlsson@intel.com>, Mauro Carvalho Chehab <mchehab@kernel.org>, Michael Ellerman <mpe@ellerman.id.au>, Michal Hocko <mho>
List-Id: dri-devel@lists.freedesktop.org

On Mon, Nov 04, 2019 at 12:57:59PM -0800, John Hubbard wrote:
> On 11/4/19 12:37 PM, Jason Gunthorpe wrote:
> > On Mon, Nov 04, 2019 at 03:31:53PM -0500, Jerome Glisse wrote:
> >>> Note for Jason: the (a) or (b) items are talking about the vfio case, which is
> >>> one of the two call sites that now use pin_longterm_pages_remote(), and the
> >>> other one is infiniband:
> >>>
> >>> drivers/infiniband/core/umem_odp.c:646:         npages = pin_longterm_pages_remote(owning_process, owning_mm,
> >>> drivers/vfio/vfio_iommu_type1.c:353:            ret = pin_longterm_pages_remote(NULL, mm, vaddr, 1,
> >>
> >> vfio should be reverted until it can be properly implemented.
> >> The issue is that when you fix the implementation you might
> >> break vfio existing user and thus regress the kernel from user
> >> point of view. So i rather have the change to vfio reverted,
> >> i believe it was not well understood when it got upstream,
> >> between in my 5.4 tree it is still gup_remote not longterm.
> > 
> > It is clearly a bug, vfio must use LONGTERM, and does right above this
> > remote call:
> > 
> >         if (mm == current->mm) {
> >                 ret = get_user_pages(vaddr, 1, flags | FOLL_LONGTERM, page,
> >                                      vmas);
> >         } else {
> >                 ret = get_user_pages_remote(NULL, mm, vaddr, 1, flags, page,
> >                                             vmas, NULL);
> > 
> > 
> > I'm not even sure that it really makes any sense to build a 'if' like
> > that, surely just always call remote??
> > 
> 
> 
> Right, and I thought about this when converting, and realized that the above 
> code is working around the current gup.c limitations, which are "cannot support
> gup remote with FOLL_LONGTERM".

But AFAICT it doesn't have a problem, the protection test is just too
strict, and I guess the control flow needs a bit of fixing..

The issue is this:

static __always_inline long __get_user_pages_locked():
{
        if (locked) {
                /* if VM_FAULT_RETRY can be returned, vmas become invalid */
                BUG_ON(vmas);
                /* check caller initialized locked */
                BUG_ON(*locked != 1);
        }


so remote could be written as:

if (gup_flags & FOLL_LONGTERM) {
   if (WARN_ON_ONCE(locked))
        return -EINVAL;
   return __gup_longterm_locked(...)
}

return __get_user_pages_locked(...)

??

Jason