On Wed, Mar 04, 2020 at 12:03:32PM +0100, Jinpu Wang wrote:
> On Tue, Mar 3, 2020 at 5:59 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> >
> > On Tue, Mar 03, 2020 at 05:41:27PM +0100, Jinpu Wang wrote:
> > > On Tue, Mar 3, 2020 at 12:37 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, Feb 21, 2020 at 11:47:06AM +0100, Jack Wang wrote:
> > > > > From: Jack Wang <jinpu.wang@xxxxxxxxxxxxxxx>
> > > > >
> > > > > This is main functionality of rtrs-server module, which accepts
> > > > > set of RDMA connections (so called rtrs session), creates/destroys
> > > > > sysfs entries associated with rtrs session and notifies upper layer
> > > > > (user of RTRS API) about RDMA requests or link events.
> > > > >
> > > > > Signed-off-by: Danil Kipnis <danil.kipnis@xxxxxxxxxxxxxxx>
> > > > > Signed-off-by: Jack Wang <jinpu.wang@xxxxxxxxxxxxxxx>
> > > > > ---
> > > > > drivers/infiniband/ulp/rtrs/rtrs-srv.c | 2164 ++++++++++++++++++++++++
> > > > > 1 file changed, 2164 insertions(+)
> > > > > create mode 100644 drivers/infiniband/ulp/rtrs/rtrs-srv.c
> > > > >
> > > > > diff --git a/drivers/infiniband/ulp/rtrs/rtrs-srv.c b/drivers/infiniband/ulp/rtrs/rtrs-srv.c
> > > > > new file mode 100644
> > > > > index 000000000000..e60ee6dd675d
> > > > > --- /dev/null
> > > > > +++ b/drivers/infiniband/ulp/rtrs/rtrs-srv.c
> > > > > @@ -0,0 +1,2164 @@
> > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > +/*
> > > > > + * RDMA Transport Layer
> > > > > + *
> > > > > + * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved.
> > > > > + * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved.
> > > > > + * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved.
> > > > > + */
> > > > > +
> > > > > +#undef pr_fmt
> > > > > +#define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt
> > > > > +
> > > > > +#include <linux/module.h>
> > > > > +#include <linux/mempool.h>
> > > > > +
> > > > > +#include "rtrs-srv.h"
> > > > > +#include "rtrs-log.h"
> > > > > +
> > > > > +MODULE_DESCRIPTION("RDMA Transport Server");
> > > > > +MODULE_LICENSE("GPL");
> > > > > +
> > > > > +/* Must be power of 2, see mask from mr->page_size in ib_sg_to_pages() */
> > > > > +#define DEFAULT_MAX_CHUNK_SIZE (128 << 10)
> > > > > +#define DEFAULT_SESS_QUEUE_DEPTH 512
> > > > > +#define MAX_HDR_SIZE PAGE_SIZE
> > > > > +#define MAX_SG_COUNT ((MAX_HDR_SIZE - sizeof(struct rtrs_msg_rdma_read)) \
> > > > > + / sizeof(struct rtrs_sg_desc))
> > > > > +
> > > > > +/* We guarantee to serve 10 paths at least */
> > > > > +#define CHUNK_POOL_SZ 10
> > > > > +
> > > > > +static struct rtrs_rdma_dev_pd dev_pd;
> > > > > +static mempool_t *chunk_pool;
> > > > > +struct class *rtrs_dev_class;
> > > > > +
> > > > > +static int __read_mostly max_chunk_size = DEFAULT_MAX_CHUNK_SIZE;
> > > > > +static int __read_mostly sess_queue_depth = DEFAULT_SESS_QUEUE_DEPTH;
> > > > > +
> > > > > +static bool always_invalidate = true;
> > > > > +module_param(always_invalidate, bool, 0444);
> > > > > +MODULE_PARM_DESC(always_invalidate,
> > > > > + "Invalidate memory registration for contiguous memory regions before accessing.");
> > > > > +
> > > > > +module_param_named(max_chunk_size, max_chunk_size, int, 0444);
> > > > > +MODULE_PARM_DESC(max_chunk_size,
> > > > > + "Max size for each IO request, when change the unit is in byte (default: "
> > > > > + __stringify(DEFAULT_MAX_CHUNK_SIZE) "KB)");
> > > > > +
> > > > > +module_param_named(sess_queue_depth, sess_queue_depth, int, 0444);
> > > > > +MODULE_PARM_DESC(sess_queue_depth,
> > > > > + "Number of buffers for pending I/O requests to allocate per session. Maximum: "
> > > > > + __stringify(MAX_SESS_QUEUE_DEPTH) " (default: "
> > > > > + __stringify(DEFAULT_SESS_QUEUE_DEPTH) ")");
> > > >
> > > > We don't like module parameters in the RDMA.
> > > Hi Leon,
> > >
> > > These paramters are affecting resouce usage/performance, I think would
> > > be good to have them as module parameters,
> > > so admin could choose based their needs.
> >
> > It is premature optimization before second user comes, also it is
> > based on the assumption that everyone uses modules, which is not true.
> The idea to have module parameters is to cover more use cases, IMHO.
>
> Even you builtin the module to the kernel, you can still change the
> module parameters
> by passing the "moduls_name.paramters" in kernel command line, eg:
> kvm.nx_huge_pages=true
I know about that, but it doesn't make them helpful.
Thanks
> >
> > Thanks
> Thanks
