Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 15 Aug 2019 15:01:59 -0300
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     Jerome Glisse <jglisse@redhat.com>
Cc:     Daniel Vetter <daniel.vetter@ffwll.ch>,
        Michal Hocko <mhocko@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux MM <linux-mm@kvack.org>,
        DRI Development <dri-devel@lists.freedesktop.org>,
        Intel Graphics Development <intel-gfx@lists.freedesktop.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        David Rientjes <rientjes@google.com>,
        Christian =?utf-8?B?S8O2bmln?= <christian.koenig@amd.com>,
        Masahiro Yamada <yamada.masahiro@socionext.com>,
        Wei Wang <wvw@google.com>,
        Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Jann Horn <jannh@google.com>, Feng Tang <feng.tang@intel.com>,
        Kees Cook <keescook@chromium.org>,
        Randy Dunlap <rdunlap@infradead.org>,
        Daniel Vetter <daniel.vetter@intel.com>
Subject: Re: [PATCH 2/5] kernel.h: Add non_block_start/end()
Message-ID: <20190815180159.GO21596@ziepe.ca>
References: <20190814134558.fe659b1a9a169c0150c3e57c@linux-foundation.org>
 <20190815084429.GE9477@dhcp22.suse.cz>
 <20190815130415.GD21596@ziepe.ca>
 <CAKMK7uE9zdmBuvxa788ONYky=46GN=5Up34mKDmsJMkir4x7MQ@mail.gmail.com>
 <20190815143759.GG21596@ziepe.ca>
 <CAKMK7uEJQ6mPQaOWbT_6M+55T-dCVbsOxFnMC6KzLAMQNa-RGg@mail.gmail.com>
 <20190815151028.GJ21596@ziepe.ca>
 <CAKMK7uG33FFCGJrDV4-FHT2FWi+Z5SnQ7hoyBQd4hignzm1C-A@mail.gmail.com>
 <20190815173557.GN21596@ziepe.ca>
 <20190815173922.GH30916@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190815173922.GH30916@redhat.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Aug 15, 2019 at 01:39:22PM -0400, Jerome Glisse wrote:
> On Thu, Aug 15, 2019 at 02:35:57PM -0300, Jason Gunthorpe wrote:
> > On Thu, Aug 15, 2019 at 06:25:16PM +0200, Daniel Vetter wrote:
> > 
> > > I'm not really well versed in the details of our userptr, but both
> > > amdgpu and i915 wait for the gpu to complete from
> > > invalidate_range_start. Jerome has at least looked a lot at the amdgpu
> > > one, so maybe he can explain what exactly it is we're doing ...
> > 
> > amdgpu is (wrongly) using hmm for something, I can't really tell what
> > it is trying to do. The calls to dma_fence under the
> > invalidate_range_start do not give me a good feeling.
> > 
> > However, i915 shows all the signs of trying to follow the registration
> > cache model, it even has a nice comment in
> > i915_gem_userptr_get_pages() explaining that the races it has don't
> > matter because it is a user space bug to change the VA mapping in the
> > first place. That just screams registration cache to me.
> > 
> > So it is fine to run HW that way, but if you do, there is no reason to
> > fence inside the invalidate_range end. Just orphan the DMA buffer and
> > clean it up & release the page pins when all DMA buffer refs go to
> > zero. The next access to that VA should get a new DMA buffer with the
> > right mapping.
> > 
> > In other words the invalidation should be very simple without
> > complicated locking, or wait_event's. Look at hfi1 for example.
> 
> This would break the today usage model of uptr and it will
> break userspace expectation ie if GPU is writting to that
> memory and that memory then the userspace want to make sure
> that it will see what the GPU write.

How exactly? This is holding the page pin, so the only way the VA
mapping can be changed is via explicit user action.

ie:

   gpu_write_something(va, size)
   mmap(.., va, size, MMAP_FIXED);
   gpu_wait_done()

This is racy and indeterminate with both models.

Based on the comment in i915 it appears to be going on the model that
changes to the mmap by userspace when the GPU is working on it is a
programming bug. This is reasonable, lots of systems use this kind of
consistency model.

Since the driver seems to rely on a dma_fence to block DMA access, it
looks to me like the kernel has full visibility to the
'gpu_write_something()' and 'gpu_wait_done()' markers.

I think trying to use hmm_range_fault on HW that can't do HW page
faulting and HW 'TLB shootdown' is a very, very bad idea. I fear that
is what amd gpu is trying to do.

I haven't yet seen anything that looks like 'TLB shootdown' in i915??

Jason