On Thu, Mar 12, 2020 at 01:04:05PM -0400, Andrew Boyer wrote: > What would you say to a per-process env variable to disable locking in a userspace provider? We have thread-domain concept in libibverbs to achieve lockless flows. https://github.com/linux-rdma/rdma-core/blob/master/libibverbs/man/ibv_alloc_td.3 BTW, please don't use top-posting. Thanks > > -Andrew > > > On Mar 12, 2020, at 1:02 PM, Jason Gunthorpe <jgg@xxxxxxxx> wrote: > > > > On Thu, Mar 12, 2020 at 11:26:40AM +0200, Leon Romanovsky wrote: > >> On Thu, Mar 12, 2020 at 03:48:10PM +0800, Weihang Li wrote: > >>> From: Jiaran Zhang <zhangjiaran@xxxxxxxxxx> > >>> > >>> In some scenarios, ULP can ensure that there is no concurrency when > >>> processing io, thus lock free can be used to improve performance. > >>> > >>> Signed-off-by: Jiaran Zhang <zhangjiaran@xxxxxxxxxx> > >>> Signed-off-by: Yixian Liu <liuyixian@xxxxxxxxxx> > >>> Signed-off-by: Weihang Li <liweihang@xxxxxxxxxx> > >>> drivers/infiniband/hw/hns/hns_roce_device.h | 1 + > >>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 112 ++++++++++++++++++++-------- > >>> drivers/infiniband/hw/hns/hns_roce_qp.c | 1 + > >>> 3 files changed, 84 insertions(+), 30 deletions(-) > >>> > >> > >> NAK, everything in this patch is wrong, starting from exposure, > >> implementation and description. > > > > Yes, complete no on a module parameter to disable locking. > > > > Jason >
