Re: [PATCH v9 07/10] mm: Device exclusive memory access

[Alistair Popple <apopple@nvidia.com>]

On Thursday, 3 June 2021 12:37:30 AM AEST Peter Xu wrote:
> External email: Use caution opening links or attachments
> 
> On Wed, Jun 02, 2021 at 06:50:37PM +1000, Balbir Singh wrote:
> > On Wed, May 26, 2021 at 12:17:18AM -0700, John Hubbard wrote:
> > > On 5/25/21 4:51 AM, Balbir Singh wrote:
> > > ...
> > > 
> > > > > How beneficial is this code to nouveau users?  I see that it permits
> > > > > a
> > > > > part of OpenCL to be implemented, but how useful/important is this
> > > > > in
> > > > > the real world?
> > > > 
> > > > That is a very good question! I've not reviewed the code, but a sample
> > > > program with the described use case would make things easy to parse.
> > > > I suspect that is not easy to build at the moment?
> > > 
> > > The cover letter says this:
> > > 
> > > This has been tested with upstream Mesa 21.1.0 and a simple OpenCL
> > > program
> > > which checks that GPU atomic accesses to system memory are atomic.
> > > Without
> > > this series the test fails as there is no way of write-protecting the
> > > page
> > > mapping which results in the device clobbering CPU writes. For reference
> > > the test is available at https://ozlabs.org/~apopple/opencl_svm_atomics/
> > > 
> > > Further testing has been performed by adding support for testing
> > > exclusive
> > > access to the hmm-tests kselftests.
> > > 
> > > ...so that seems to cover the "sample program" request, at least.
> > 
> > Thanks, I'll take a look
> > 
> > > > I wonder how we co-ordinate all the work the mm is doing, page
> > > > migration,
> > > > reclaim with device exclusive access? Do we have any numbers for the
> > > > worst
> > > > case page fault latency when something is marked away for exclusive
> > > > access?
> > > 
> > > CPU page fault latency is approximately "terrible", if a page is
> > > resident on the GPU. We have to spin up a DMA engine on the GPU and
> > > have it copy the page over the PCIe bus, after all.
> > > 
> > > > I presume for now this is anonymous memory only? SWP_DEVICE_EXCLUSIVE
> > > > would
> > > 
> > > Yes, for now.
> > > 
> > > > only impact the address space of programs using the GPU. Should the
> > > > exclusively marked range live in the unreclaimable list and recycled
> > > > back to active/in-active to account for the fact that
> > > > 
> > > > 1. It is not reclaimable and reclaim will only hurt via page faults?
> > > > 2. It ages the page correctly or at-least allows for that possibility
> > > > when the> > > 
> > > >     page is used by the GPU.
> > > 
> > > I'm not sure that that is *necessarily* something we can conclude. It
> > > depends upon access patterns of each program. For example, a
> > > "reduction" parallel program sends over lots of data to the GPU, and
> > > only a tiny bit of (reduced!) data comes back to the CPU. In that case,
> > > freeing the physical page on the CPU is actually the best decision for
> > > the OS to make (if the OS is sufficiently prescient).> 
> > With a shared device or a device exclusive range, it would be good to get
> > the device usage pattern and update the mm with that knowledge, so that
> > the LRU can be better maintained. With your comment you seem to suggest
> > that a page used by the GPU might be a good candidate for reclaim based
> > on the CPU's understanding of the age of the page should not account for
> > use by the device
> > (are GPU workloads - access once and discard?)
> 
> Hmm, besides the aging info, this reminded me: do we need to isolate the
> page from lru too when marking device exclusive access?
> 
> Afaict the current patch didn't do that so I think it's reclaimable.  If we
> still have the rmap then we'll get a mmu notify CLEAR when unmapping that
> special pte, so device driver should be able to drop the ownership.  However
> we dropped the rmap when marking exclusive.  Now I don't know whether and
> how it'll work if page reclaim runs with the page being exclusively owned
> if without isolating the page..

Reclaim won't run on the page due to the extra references from the special 
swap entries.

> --
> Peter Xu