On 11/24/19 4:37 PM, John Hubbard wrote: > Hi Leon, Jason, Christoph, > > Maybe I'm overlooking something, but as I wrote in patch 1, it looks > like we can simply allow FOLL_FORCE to be passed to gup_fast(). > > This should fix Leon's reported RDMA failure [1] when using patch 2 by > itself. (I've compile- and boot-tested these, and also did short LTP > and fio with direct IO tests, but I don't have an Infiniband runtime > setup that exercises the umem.c code.) > > [1] https://lore.kernel.org/r/20191124100724.GH136476@unreal > > John Hubbard (2): > mm/gup: allow FOLL_FORCE for get_user_pages_fast() > IB/umem: use get_user_pages_fast() to pin DMA pages > > drivers/infiniband/core/umem.c | 17 ++++++----------- > mm/gup.c | 3 ++- > 2 files changed, 8 insertions(+), 12 deletions(-) > OK, based on Jason's response [1] that it's too late to put this into 5.5, let's withdraw this, and I'll resend when it's time to send out patches for 5.6. [1] https://lore.kernel.org/r/20191125005339.GC5634@xxxxxxxx thanks, -- John Hubbard NVIDIA