Re: [PATCH v8 0/3] mm/hmm/test: add self tests for HMM

From: Jason Gunthorpe <jgg@mellanox.com>
To: Leon Romanovsky <leon@kernel.org>
Cc: Ralph Campbell <rcampbell@nvidia.com>,
	Jerome Glisse <jglisse@redhat.com>,
	John Hubbard <jhubbard@nvidia.com>,
	Christoph Hellwig <hch@lst.de>,
	Andrew Morton <akpm@linux-foundation.org>,
	Shuah Khan <shuah@kernel.org>,
	linux-rdma@vger.kernel.org, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org
Subject: Re: [PATCH v8 0/3] mm/hmm/test: add self tests for HMM
Date: Wed, 15 Apr 2020 16:32:13 -0300	[thread overview]
Message-ID: <20200415193213.GW11945@mellanox.com> (raw)
In-Reply-To: <20200415192952.GA1309273@unreal>

On Wed, Apr 15, 2020 at 10:29:52PM +0300, Leon Romanovsky wrote:
> On Wed, Apr 15, 2020 at 10:28:23AM -0700, Ralph Campbell wrote:
> >
> > On 4/15/20 7:41 AM, Jason Gunthorpe wrote:
> > > On Fri, Mar 20, 2020 at 05:31:05PM -0700, Ralph Campbell wrote:
> > > > This series adds basic self tests for HMM and are intended for Jason
> > > > Gunthorpe's rdma tree which has a number of HMM patches applied.
> > >
> > > Here are some hunks I noticed while testing this:
> > >
> > > +++ b/lib/Kconfig.debug
> > > @@ -2201,7 +2201,8 @@ config TEST_MEMINIT
> > >   config TEST_HMM
> > >   	tristate "Test HMM (Heterogeneous Memory Management)"
> > > -	depends on DEVICE_PRIVATE
> > > +	depends on TRANSPARENT_HUGEPAGE
> > > +	select DEVICE_PRIVATE
> > >   	select HMM_MIRROR
> > >   	select MMU_NOTIFIER
> > >   	help
> > >
> > > It fails testing if TRANSPARENT_HUGEPAGE is not on
> > >
> > > @@ -1097,6 +1071,7 @@ static int dmirror_device_init(struct dmirror_device *mdevice, int id)
> > >   	spin_lock_init(&mdevice->lock);
> > >   	cdev_init(&mdevice->cdevice, &dmirror_fops);
> > > +	mdevice->cdevice.owner = THIS_MODULE;
> > >   	ret = cdev_add(&mdevice->cdevice, dev, 1);
> > >   	if (ret)
> > >   		return ret;
> > >
> > > The use of cdev without a struct device is super weird, but it still
> > > needs this
> > >
> > > diff --git a/tools/testing/selftests/vm/test_hmm.sh b/tools/testing/selftests/vm/test_hmm.sh
> > > index 461e4a99a362cf..0647b525a62564 100755
> > > +++ b/tools/testing/selftests/vm/test_hmm.sh
> > > @@ -59,7 +59,7 @@ run_smoke()
> > >   	echo "Running smoke test. Note, this test provides basic coverage."
> > >   	load_driver
> > > -	./hmm-tests
> > > +	$(dirname "${BASH_SOURCE[0]}")/hmm-tests
> > >   	unload_driver
> > >   }
> > >
> > > Make it runnably reliably
> > >
> > > Jason
> >
> > Thanks for the fixes. I'll apply these and send a v9.
> > I will also add missing calls to release_mem_region() to free the reserved device private
> > addresses.
> 
> If you decide to ignore my request to avoid addition of special header
> file to UAPI, at least don't copy and install that file without some
> special CONFIG option (TEST_HMM ???) requested by the users. It also
> will be good to get Acked-by on this change from HMM people.
> 
> However, I still think that include/uapi/linux/test_hmm.h opens
> pandora box of having UAPI files without real promise to keep it
> backward compatible.

It would be nice if we could put the header outside the uapi
directory and outside the install machinery

Maybe for now hackery some relative include like
 #include "../../../lib/hmm_test_uapi.h"

?

I don't see any sane way to avoid the dedicate module and special
ioctl though.

Jason