On Thu, Mar 5, 2020 at 1:16 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>
> On Thu, Mar 05, 2020 at 01:01:08PM +0100, Danil Kipnis wrote:
> > On Thu, 5 Mar 2020, 09:00 Leon Romanovsky, <leon@xxxxxxxxxx> wrote:
> >
> > > 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
> > >
> > Hi Leon,
> >
> > Queue_depth and max_chunksize parameters control the tradeoff between
> > throuput performance and memory consumption. We do use them to set
> > different values for storages equipped with SSDs (fast) and on storages
> > equipped with HDDs (slow). The last parameter always_invaldate enforces the
> > invalidation of an rdma buffer before its hand over to the block layer. We
> > set it to no in our datacenters, since they are closed and malicious
> > clients are not a threat in our scenario. In general case it defaults to
> > yes, as requested by Jason. Our admins need to have control over those
> > control knobs somehow... We could make sysfs entries out of them or
> > something, but would it really make sense?
>
> blk_queue_nonrot() inside your code?
It's exported function, and also used by other drivers like
md/dm/target core, right?
Thanks
