LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To: Sudeep Dutt <sudeep.dutt@intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>, Jonathan Corbet <corbet@lwn.net>,
	linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org,
	Dave Jiang <dave.jiang@intel.com>,
	Nikhil Rao <nikhil.rao@intel.com>,
	Ashutosh Dixit <ashutosh.dixit@intel.com>
Subject: Re: [PATCH char-misc-next v3 01/13] misc: mic: SCIF header file and IOCTL interface
Date: Wed, 25 Mar 2015 12:04:16 +0100	[thread overview]
Message-ID: <20150325110416.GA23335@kroah.com> (raw)
In-Reply-To: <b386e357efa0b3d646521a0ce4b305b9e4cf5343.1422317099.git.sudeep.dutt@intel.com>

On Mon, Jan 26, 2015 at 04:32:42PM -0800, Sudeep Dutt wrote:
> This patch introduces the SCIF documentation in the header file
> and describes the IOCTL interface for user mode. mic_overview.txt
> is updated with documentation on SCIF and a new document
> describing SCIF in more details is available in scif_overview.txt.
> 
> Reviewed-by: Nikhil Rao <nikhil.rao@intel.com>
> Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
> Signed-off-by: Sudeep Dutt <sudeep.dutt@intel.com>
> ---
>  Documentation/mic/mic_overview.txt  |  28 +-
>  Documentation/mic/scif_overview.txt |  98 ++++
>  include/uapi/linux/Kbuild           |   1 +
>  include/linux/scif.h                | 993 ++++++++++++++++++++++++++++++++++++
>  include/uapi/linux/scif_ioctl.h     | 215 ++++++++
>  5 files changed, 1323 insertions(+), 12 deletions(-)
>  create mode 100644 Documentation/mic/scif_overview.txt
>  create mode 100644 include/linux/scif.h
>  create mode 100644 include/uapi/linux/scif_ioctl.h
> 
> diff --git a/Documentation/mic/mic_overview.txt b/Documentation/mic/mic_overview.txt
> index 77c5418..1a2f2c8 100644
> --- a/Documentation/mic/mic_overview.txt
> +++ b/Documentation/mic/mic_overview.txt
> @@ -24,6 +24,10 @@ a virtual bus called mic bus is created and virtual dma devices are
>  created on it by the host/card drivers. On host the channels are private
>  and used only by the host driver to transfer data for the virtio devices.
>  
> +The Symmetric Communication Interface (SCIF (pronounced as skiff)) is a
> +low level communications API across PCIe currently implemented for MIC.
> +More details are available at scif_overview.txt.
> +
>  Here is a block diagram of the various components described above. The
>  virtio backends are situated on the host rather than the card given better
>  single threaded performance for the host compared to MIC, the ability of
> @@ -47,18 +51,18 @@ the fact that the virtio block storage backend can only be on the host.
>                        |               |       | Virtio over PCIe IOCTLs  |
>                        |               |       +--------------------------+
>  +-----------+         |               |                   |  +-----------+
> -| MIC DMA   |         |               |                   |  | MIC DMA   |
> -| Driver    |         |               |                   |  | Driver    |
> -+-----------+         |               |                   |  +-----------+
> -      |               |               |                   |        |
> -+---------------+     |               |                   |  +----------------+
> -|MIC virtual Bus|     |               |                   |  |MIC virtual Bus |
> -+---------------+     |               |                   |  +----------------+
> -      |               |               |                   |              |
> -      |   +--------------+            |            +---------------+     |
> -      |   |Intel MIC     |            |            |Intel MIC      |     |
> -      +---|Card Driver   |            |            |Host Driver    |     |
> -          +--------------+            |            +---------------+-----+
> +| MIC DMA   |         |  +----------+ | +-----------+     |  | MIC DMA   |
> +| Driver    |         |  |  SCIF    | | |   SCIF    |     |  | Driver    |
> ++-----------+         |  +----------+ | +-----------+     |  +-----------+
> +      |               |       |       |       |           |        |
> ++---------------+     | +-----+-----+ | +-----+-----+     | +---------------+
> +|MIC virtual Bus|     | |SCIF HW Bus| | |SCIF HW BUS|     | |MIC virtual Bus|
> ++---------------+     | +-----------+ | +-----+-----+     | +---------------+
> +      |               |       |       |       |           |              |
> +      |   +--------------+    |       |       |    +---------------+     |
> +      |   |Intel MIC     |    |       |       |    |Intel MIC      |     |
> +      +---|Card Driver   +----+       |       |    |Host Driver    |     |
> +          +--------------+            |       +----+---------------+-----+
>                        |               |                   |
>               +-------------------------------------------------------------+
>               |                                                             |
> diff --git a/Documentation/mic/scif_overview.txt b/Documentation/mic/scif_overview.txt
> new file mode 100644
> index 0000000..0a280d9
> --- /dev/null
> +++ b/Documentation/mic/scif_overview.txt
> @@ -0,0 +1,98 @@
> +The Symmetric Communication Interface (SCIF (pronounced as skiff)) is a low
> +level communications API across PCIe currently implemented for MIC. Currently
> +SCIF provides inter-node communication within a single host platform, where a
> +node is a MIC Coprocessor or Xeon based host. SCIF abstracts the details of
> +communicating over the PCIe bus while providing an API that is symmetric
> +across all the nodes in the PCIe network. An important design objective for SCIF
> +is to deliver the maximum possible performance given the communication
> +abilities of the hardware. SCIF has been used to implement an offload compiler
> +runtime and OFED support for MPI implementations for MIC coprocessors.
> +
> +==== SCIF API Components ====
> +The SCIF API has the following parts:
> +1. Connection establishment using a client server model
> +2. Byte stream messaging intended for short messages
> +3. Node enumeration to determine online nodes
> +4. Poll semantics for detection of incoming connections and messages
> +5. Memory registration to pin down pages
> +6. Remote memory mapping for low latency CPU accesses via mmap
> +7. Remote DMA (RDMA) for high bandwidth DMA transfers
> +8. Fence APIs for RDMA synchronization
> +
> +SCIF exposes the notion of a connection which can be used by peer processes on
> +nodes in a SCIF PCIe "network" to share memory "windows" and to communicate. A
> +process in a SCIF node initiates a SCIF connection to a peer process on a
> +different node via a SCIF "endpoint". SCIF endpoints support messaging APIs
> +which are similar to connection oriented socket APIs. Connected SCIF endpoints
> +can also register local memory which is followed by data transfer using either
> +DMA, CPU copies or remote memory mapping via mmap. SCIF supports both user and
> +kernel mode clients which are functionally equivalent.
> +
> +==== SCIF Performance for MIC ====
> +DMA bandwidth comparison between the TCP (over ethernet over PCIe) stack versus
> +SCIF shows the performance advantages of SCIF for HPC applications and runtimes.
> +
> +             Comparison of TCP and SCIF based BW
> +
> +  Throughput (GB/sec)
> +    8 +                                             PCIe Bandwidth ******
> +      +                                                        TCP ######
> +    7 +    **************************************             SCIF %%%%%%
> +      |                       %%%%%%%%%%%%%%%%%%%
> +    6 +                   %%%%
> +      |                 %%
> +      |               %%%
> +    5 +              %%
> +      |            %%
> +    4 +           %%
> +      |          %%
> +    3 +         %%
> +      |        %
> +    2 +      %%
> +      |     %%
> +      |    %
> +    1 +
> +      +    ######################################
> +    0 +++---+++--+--+-+--+--+-++-+--+-++-+--+-++-+-
> +      1       10     100      1000   10000   100000
> +                   Transfer Size (KBytes)
> +
> +SCIF allows memory sharing via mmap(..) between processes on different PCIe
> +nodes and thus provides bare-metal PCIe latency. The round trip SCIF mmap
> +latency from the host to an x100 MIC for an 8 byte message is 0.44 usecs.
> +
> +SCIF has a user space library which is a thin IOCTL wrapper providing a user
> +space API similar to the kernel API in scif.h. The SCIF user space library
> +is distributed @ https://software.intel.com/en-us/mic-developer
> +
> +Here is some pseudo code for an example of how two applications on two PCIe
> +nodes would typically use the SCIF API:
> +
> +Process A (on node A)			Process B (on node B)
> +
> +/* get online node information */
> +scif_get_node_ids(..)			scif_get_node_ids(..)
> +scif_open(..)				scif_open(..)
> +scif_bind(..)				scif_bind(..)
> +scif_listen(..)
> +scif_accept(..)				scif_connect(..)
> +/* SCIF connection established */
> +
> +/* Send and receive short messages */
> +scif_send(..)/scif_recv(..)		scif_send(..)/scif_recv(..)
> +
> +/* Register memory */
> +scif_register(..)			scif_register(..)
> +
> +/* RDMA */
> +scif_readfrom(..)/scif_writeto(..)	scif_readfrom(..)/scif_writeto(..)
> +
> +/* Fence DMAs */
> +scif_fence_signal(..)			scif_fence_signal(..)
> +
> +mmap(..)				mmap(..)
> +
> +/* Access remote registered memory */
> +
> +/* Close the endpoints */
> +scif_close(..)				scif_close(..)
> diff --git a/include/uapi/linux/Kbuild b/include/uapi/linux/Kbuild
> index 00b10002..5083817 100644
> --- a/include/uapi/linux/Kbuild
> +++ b/include/uapi/linux/Kbuild
> @@ -349,6 +349,7 @@ header-y += rtc.h
>  header-y += rtnetlink.h
>  header-y += scc.h
>  header-y += sched.h
> +header-y += scif_ioctl.h
>  header-y += screen_info.h
>  header-y += sctp.h
>  header-y += sdla.h
> diff --git a/include/linux/scif.h b/include/linux/scif.h
> new file mode 100644
> index 0000000..44f4f38
> --- /dev/null
> +++ b/include/linux/scif.h
> @@ -0,0 +1,993 @@
> +/*
> + * Intel MIC Platform Software Stack (MPSS)
> + *
> + * This file is provided under a dual BSD/GPLv2 license.  When using or
> + * redistributing this file, you may do so under either license.
> + *
> + * GPL LICENSE SUMMARY
> + *
> + * Copyright(c) 2014 Intel Corporation.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of version 2 of the GNU General Public License as
> + * published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope that it will be useful, but
> + * WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * General Public License for more details.
> + *
> + * BSD LICENSE
> + *
> + * Copyright(c) 2014 Intel Corporation.
> + *
> + * Redistribution and use in source and binary forms, with or without
> + * modification, are permitted provided that the following conditions
> + * are met:
> + *
> + * * Redistributions of source code must retain the above copyright
> + *   notice, this list of conditions and the following disclaimer.
> + * * Redistributions in binary form must reproduce the above copyright
> + *   notice, this list of conditions and the following disclaimer in
> + *   the documentation and/or other materials provided with the
> + *   distribution.
> + * * Neither the name of Intel Corporation nor the names of its
> + *   contributors may be used to endorse or promote products derived
> + *   from this software without specific prior written permission.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + *
> + * Intel SCIF driver.
> + *
> + */
> +#ifndef __SCIF_H__
> +#define __SCIF_H__
> +
> +#include <linux/types.h>
> +#include <linux/poll.h>
> +#include <linux/scif_ioctl.h>
> +
> +#define SCIF_ACCEPT_SYNC	1
> +#define SCIF_SEND_BLOCK		1
> +#define SCIF_RECV_BLOCK		1
> +
> +enum {
> +	SCIF_PROT_READ = (1 << 0),
> +	SCIF_PROT_WRITE = (1 << 1)
> +};
> +
> +enum {
> +	SCIF_MAP_FIXED = 0x10,
> +	SCIF_MAP_KERNEL	= 0x20,
> +};
> +
> +enum {
> +	SCIF_FENCE_INIT_SELF = (1 << 0),
> +	SCIF_FENCE_INIT_PEER = (1 << 1),
> +	SCIF_SIGNAL_LOCAL = (1 << 4),
> +	SCIF_SIGNAL_REMOTE = (1 << 5)
> +};
> +
> +enum {
> +	SCIF_RMA_USECPU = (1 << 0),
> +	SCIF_RMA_USECACHE = (1 << 1),
> +	SCIF_RMA_SYNC = (1 << 2),
> +	SCIF_RMA_ORDERED = (1 << 3)
> +};
> +
> +/* End of SCIF Admin Reserved Ports */
> +#define SCIF_ADMIN_PORT_END	1024
> +
> +/* End of SCIF Reserved Ports */
> +#define SCIF_PORT_RSVD		1088
> +
> +typedef struct scif_endpt *scif_epd_t;
> +
> +#define SCIF_OPEN_FAILED ((scif_epd_t)-1)
> +#define SCIF_REGISTER_FAILED ((off_t)-1)
> +#define SCIF_MMAP_FAILED ((void *)-1)
> +
> +/**
> + * scif_open() - Create an endpoint
> + *
> + * Return:
> + * Upon successful completion, scif_open() returns an endpoint descriptor to
> + * be used in subsequent SCIF functions calls to refer to that endpoint;
> + * otherwise in user mode SCIF_OPEN_FAILED (that is ((scif_epd_t)-1)) is
> + * returned and errno is set to indicate the error; in kernel mode a NULL
> + * scif_epd_t is returned.
> + *
> + * Errors:
> + * ENOMEM - Insufficient kernel memory was available
> + */
> +scif_epd_t scif_open(void);
> +
> +/**
> + * scif_bind() - Bind an endpoint to a port
> + * @epd:	endpoint descriptor
> + * @pn:		port number
> + *
> + * scif_bind() binds endpoint epd to port pn, where pn is a port number on the
> + * local node. If pn is zero, a port number greater than or equal to
> + * SCIF_PORT_RSVD is assigned and returned. Each endpoint may be bound to
> + * exactly one local port. Ports less than 1024 when requested can only be bound
> + * by system (or root) processes or by processes executed by privileged users.
> + *
> + * Return:
> + * Upon successful completion, scif_bind() returns the port number to which epd
> + * is bound; otherwise in user mode -1 is returned and errno is set to
> + * indicate the error; in kernel mode the negative of one of the following
> + * errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * EINVAL - the endpoint or the port is already bound
> + * EISCONN - The endpoint is already connected
> + * ENOSPC - No port number available for assignment
> + * EACCES - The port requested is protected and the user is not the superuser
> + */
> +int scif_bind(scif_epd_t epd, u16 pn);
> +
> +/**
> + * scif_listen() - Listen for connections on an endpoint
> + * @epd:	endpoint descriptor
> + * @backlog:	maximum pending connection requests
> + *
> + * scif_listen() marks the endpoint epd as a listening endpoint - that is, as
> + * an endpoint that will be used to accept incoming connection requests. Once
> + * so marked, the endpoint is said to be in the listening state and may not be
> + * used as the endpoint of a connection.
> + *
> + * The endpoint, epd, must have been bound to a port.
> + *
> + * The backlog argument defines the maximum length to which the queue of
> + * pending connections for epd may grow. If a connection request arrives when
> + * the queue is full, the client may receive an error with an indication that
> + * the connection was refused.
> + *
> + * Return:
> + * Upon successful completion, scif_listen() returns 0; otherwise in user mode
> + * -1 is returned and errno is set to indicate the error; in kernel mode the
> + * negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * EINVAL - the endpoint is not bound to a port
> + * EISCONN - The endpoint is already connected or listening
> + */
> +int scif_listen(scif_epd_t epd, int backlog);
> +
> +/**
> + * scif_connect() - Initiate a connection on a port
> + * @epd:	endpoint descriptor
> + * @dst:	global id of port to which to connect
> + *
> + * The scif_connect() function requests the connection of endpoint epd to remote
> + * port dst. If the connection is successful, a peer endpoint, bound to dst, is
> + * created on node dst.node. On successful return, the connection is complete.
> + *
> + * If the endpoint epd has not already been bound to a port, scif_connect()
> + * will bind it to an unused local port.
> + *
> + * A connection is terminated when an endpoint of the connection is closed,
> + * either explicitly by scif_close(), or when a process that owns one of the
> + * endpoints of the connection is terminated.
> + *
> + * In user space, scif_connect() supports an asynchronous connection mode
> + * if the application has set the O_NONBLOCK flag on the endpoint via the
> + * fcntl() system call. Setting this flag will result in the calling process
> + * not to wait during scif_connect().
> + *
> + * Return:
> + * Upon successful completion, scif_connect() returns the port ID to which the
> + * endpoint, epd, is bound; otherwise in user mode -1 is returned and errno is
> + * set to indicate the error; in kernel mode the negative of one of the
> + * following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNREFUSED - The destination was not listening for connections or refused
> + * the connection request
> + * EINVAL - dst.port is not a valid port ID
> + * EISCONN - The endpoint is already connected
> + * ENOMEM - No buffer space is available
> + * ENODEV - The destination node does not exist, or the node is lost or existed,
> + * but is not currently in the network since it may have crashed
> + * ENOSPC - No port number available for assignment
> + * EOPNOTSUPP - The endpoint is listening and cannot be connected
> + */
> +int scif_connect(scif_epd_t epd, struct scif_port_id *dst);
> +
> +/**
> + * scif_accept() - Accept a connection on an endpoint
> + * @epd:	endpoint descriptor
> + * @peer:	global id of port to which connected
> + * @newepd:	new connected endpoint descriptor
> + * @flags:	flags
> + *
> + * The scif_accept() call extracts the first connection request from the queue
> + * of pending connections for the port on which epd is listening. scif_accept()
> + * creates a new endpoint, bound to the same port as epd, and allocates a new
> + * SCIF endpoint descriptor, returned in newepd, for the endpoint. The new
> + * endpoint is connected to the endpoint through which the connection was
> + * requested. epd is unaffected by this call, and remains in the listening
> + * state.
> + *
> + * On successful return, peer holds the global port identifier (node id and
> + * local port number) of the port which requested the connection.
> + *
> + * A connection is terminated when an endpoint of the connection is closed,
> + * either explicitly by scif_close(), or when a process that owns one of the
> + * endpoints of the connection is terminated.
> + *
> + * The number of connections that can (subsequently) be accepted on epd is only
> + * limited by system resources (memory).
> + *
> + * The flags argument is formed by OR'ing together zero or more of the
> + * following values.
> + * SCIF_ACCEPT_SYNC - block until a connection request is presented. If
> + *			SCIF_ACCEPT_SYNC is not in flags, and no pending
> + *			connections are present on the queue, scif_accept()
> + *			fails with an EAGAIN error
> + *
> + * In user mode, the select() and poll() functions can be used to determine
> + * when there is a connection request. In kernel mode, the scif_poll()
> + * function may be used for this purpose. A readable event will be delivered
> + * when a connection is requested.
> + *
> + * Return:
> + * Upon successful completion, scif_accept() returns 0; otherwise in user mode
> + * -1 is returned and errno is set to indicate the error; in kernel mode the
> + *	negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EAGAIN - SCIF_ACCEPT_SYNC is not set and no connections are present to be
> + * accepted or SCIF_ACCEPT_SYNC is not set and remote node failed to complete
> + * its connection request
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * EINTR - Interrupted function
> + * EINVAL - epd is not a listening endpoint, or flags is invalid, or peer is
> + * NULL, or newepd is NULL
> + * ENODEV - The requesting node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOMEM - Not enough space
> + * ENOENT - Secondary part of epd registration failed
> + */
> +int scif_accept(scif_epd_t epd, struct scif_port_id *peer, scif_epd_t
> +		*newepd, int flags);
> +
> +/**
> + * scif_close() - Close an endpoint
> + * @epd:	endpoint descriptor
> + *
> + * scif_close() closes an endpoint and performs necessary teardown of
> + * facilities associated with that endpoint.
> + *
> + * If epd is a listening endpoint then it will no longer accept connection
> + * requests on the port to which it is bound. Any pending connection requests
> + * are rejected.
> + *
> + * If epd is a connected endpoint, then its peer endpoint is also closed. RMAs
> + * which are in-process through epd or its peer endpoint will complete before
> + * scif_close() returns. Registered windows of the local and peer endpoints are
> + * released as if scif_unregister() was called against each window.
> + *
> + * Closing a SCIF endpoint does not affect local registered memory mapped by
> + * a SCIF endpoint on a remote node. The local memory remains mapped by the peer
> + * SCIF endpoint explicitly removed by calling munmap(..) by the peer.
> + *
> + * If the peer endpoint's receive queue is not empty at the time that epd is
> + * closed, then the peer endpoint can be passed as the endpoint parameter to
> + * scif_recv() until the receive queue is empty.
> + *
> + * epd is freed and may no longer be accessed.
> + *
> + * Return:
> + * Upon successful completion, scif_close() returns 0; otherwise in user mode
> + * -1 is returned and errno is set to indicate the error; in kernel mode the
> + * negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + */
> +int scif_close(scif_epd_t epd);
> +
> +/**
> + * scif_send() - Send a message
> + * @epd:	endpoint descriptor
> + * @msg:	message buffer address
> + * @len:	message length
> + * @flags:	blocking mode flags
> + *
> + * scif_send() sends data to the peer of endpoint epd. Up to len bytes of data
> + * are copied from memory starting at address msg. On successful execution the
> + * return value of scif_send() is the number of bytes that were sent, and is
> + * zero if no bytes were sent because len was zero. scif_send() may be called
> + * only when the endpoint is in a connected state.
> + *
> + * If a scif_send() call is non-blocking, then it sends only those bytes which
> + * can be sent without waiting, up to a maximum of len bytes.
> + *
> + * If a scif_send() call is blocking, then it normally returns after sending
> + * all len bytes. If a blocking call is interrupted or the connection is
> + * reset, the call is considered successful if some bytes were sent or len is
> + * zero, otherwise the call is considered unsuccessful.
> + *
> + * In user mode, the select() and poll() functions can be used to determine
> + * when the send queue is not full. In kernel mode, the scif_poll() function
> + * may be used for this purpose.
> + *
> + * It is recommended that scif_send()/scif_recv() only be used for short
> + * control-type message communication between SCIF endpoints. The SCIF RMA
> + * APIs are expected to provide better performance for transfer sizes of
> + * 1024 bytes or longer for the current MIC hardware and software
> + * implementation.
> + *
> + * scif_send() will block until the entire message is sent if SCIF_SEND_BLOCK
> + * is passed as the flags argument.
> + *
> + * Return:
> + * Upon successful completion, scif_send() returns the number of bytes sent;
> + * otherwise in user mode -1 is returned and errno is set to indicate the
> + * error; in kernel mode the negative of one of the following errors is
> + * returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EFAULT - An invalid address was specified for a parameter
> + * EINVAL - flags is invalid, or len is negative
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOMEM - Not enough space
> + * ENOTCONN - The endpoint is not connected
> + */
> +int scif_send(scif_epd_t epd, void *msg, int len, int flags);
> +
> +/**
> + * scif_recv() - Receive a message
> + * @epd:	endpoint descriptor
> + * @msg:	message buffer address
> + * @len:	message buffer length
> + * @flags:	blocking mode flags
> + *
> + * scif_recv() receives data from the peer of endpoint epd. Up to len bytes of
> + * data are copied to memory starting at address msg. On successful execution
> + * the return value of scif_recv() is the number of bytes that were received,
> + * and is zero if no bytes were received because len was zero. scif_recv() may
> + * be called only when the endpoint is in a connected state.
> + *
> + * If a scif_recv() call is non-blocking, then it receives only those bytes
> + * which can be received without waiting, up to a maximum of len bytes.
> + *
> + * If a scif_recv() call is blocking, then it normally returns after receiving
> + * all len bytes. If the blocking call was interrupted due to a disconnection,
> + * subsequent calls to scif_recv() will copy all bytes received upto the point
> + * of disconnection.
> + *
> + * In user mode, the select() and poll() functions can be used to determine
> + * when data is available to be received. In kernel mode, the scif_poll()
> + * function may be used for this purpose.
> + *
> + * It is recommended that scif_send()/scif_recv() only be used for short
> + * control-type message communication between SCIF endpoints. The SCIF RMA
> + * APIs are expected to provide better performance for transfer sizes of
> + * 1024 bytes or longer for the current MIC hardware and software
> + * implementation.
> + *
> + * scif_recv() will block until the entire message is received if
> + * SCIF_RECV_BLOCK is passed as the flags argument.
> + *
> + * Return:
> + * Upon successful completion, scif_recv() returns the number of bytes
> + * received; otherwise in user mode -1 is returned and errno is set to
> + * indicate the error; in kernel mode the negative of one of the following
> + * errors is returned.
> + *
> + * Errors:
> + * EAGAIN - The destination node is returning from a low power state
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EFAULT - An invalid address was specified for a parameter
> + * EINVAL - flags is invalid, or len is negative
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOMEM - Not enough space
> + * ENOTCONN - The endpoint is not connected
> + */
> +int scif_recv(scif_epd_t epd, void *msg, int len, int flags);
> +
> +/**
> + * scif_register() - Mark a memory region for remote access.
> + * @epd:		endpoint descriptor
> + * @addr:		starting virtual address
> + * @len:		length of range
> + * @offset:		offset of window
> + * @prot_flags:		read/write protection flags
> + * @map_flags:		mapping flags
> + *
> + * The scif_register() function opens a window, a range of whole pages of the
> + * registered address space of the endpoint epd, starting at offset po and
> + * continuing for len bytes. The value of po, further described below, is a
> + * function of the parameters offset and len, and the value of map_flags. Each
> + * page of the window represents the physical memory page which backs the
> + * corresponding page of the range of virtual address pages starting at addr
> + * and continuing for len bytes. addr and len are constrained to be multiples
> + * of the page size. A successful scif_register() call returns po.
> + *
> + * When SCIF_MAP_FIXED is set in the map_flags argument, po will be offset
> + * exactly, and offset is constrained to be a multiple of the page size. The
> + * mapping established by scif_register() will not replace any existing
> + * registration; an error is returned if any page within the range [offset,
> + * offset + len - 1] intersects an existing window.
> + *
> + * When SCIF_MAP_FIXED is not set, the implementation uses offset in an
> + * implementation-defined manner to arrive at po. The po value so chosen will
> + * be an area of the registered address space that the implementation deems
> + * suitable for a mapping of len bytes. An offset value of 0 is interpreted as
> + * granting the implementation complete freedom in selecting po, subject to
> + * constraints described below. A non-zero value of offset is taken to be a
> + * suggestion of an offset near which the mapping should be placed. When the
> + * implementation selects a value for po, it does not replace any extant
> + * window. In all cases, po will be a multiple of the page size.
> + *
> + * The physical pages which are so represented by a window are available for
> + * access in calls to mmap(), scif_readfrom(), scif_writeto(),
> + * scif_vreadfrom(), and scif_vwriteto(). While a window is registered, the
> + * physical pages represented by the window will not be reused by the memory
> + * subsystem for any other purpose. Note that the same physical page may be
> + * represented by multiple windows.
> + *
> + * Subsequent operations which change the memory pages to which virtual
> + * addresses are mapped (such as mmap(), munmap()) have no effect on
> + * existing window.
> + *
> + * If the process will fork(), it is recommended that the registered
> + * virtual address range be marked with MADV_DONTFORK. Doing so will prevent
> + * problems due to copy-on-write semantics.
> + *
> + * The prot_flags argument is formed by OR'ing together one or more of the
> + * following values.
> + * SCIF_PROT_READ - allow read operations from the window
> + * SCIF_PROT_WRITE - allow write operations to the window
> + *
> + * The map_flags argument can be set to SCIF_MAP_FIXED which interprets a
> + * fixed offset.
> + *
> + * Return:
> + * Upon successful completion, scif_register() returns the offset at which the
> + * mapping was placed (po); otherwise in user mode SCIF_REGISTER_FAILED (that
> + * is (off_t *)-1) is returned and errno is set to indicate the error; in
> + * kernel mode the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EADDRINUSE - SCIF_MAP_FIXED is set in map_flags, and pages in the range
> + * [offset, offset + len -1] are already registered
> + * EAGAIN - The mapping could not be performed due to lack of resources
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EFAULT - Addresses in the range [addr, addr + len - 1] are invalid
> + * EINVAL - map_flags is invalid, or prot_flags is invalid, or SCIF_MAP_FIXED is
> + * set in flags, and offset is not a multiple of the page size, or addr is not a
> + * multiple of the page size, or len is not a multiple of the page size, or is
> + * 0, or offset is negative
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOMEM - Not enough space
> + * ENOTCONN -The endpoint is not connected
> + */
> +off_t scif_register(scif_epd_t epd, void *addr, size_t len, off_t offset,
> +		    int prot_flags, int map_flags);
> +
> +/**
> + * scif_unregister() - Mark a memory region for remote access.
> + * @epd:	endpoint descriptor
> + * @offset:	start of range to unregister
> + * @len:	length of range to unregister
> + *
> + * The scif_unregister() function closes those previously registered windows
> + * which are entirely within the range [offset, offset + len - 1]. It is an
> + * error to specify a range which intersects only a subrange of a window.
> + *
> + * On a successful return, pages within the window may no longer be specified
> + * in calls to mmap(), scif_readfrom(), scif_writeto(), scif_vreadfrom(),
> + * scif_vwriteto(), scif_get_pages, and scif_fence_signal(). The window,
> + * however, continues to exist until all previous references against it are
> + * removed. A window is referenced if there is a mapping to it created by
> + * mmap(), or if scif_get_pages() was called against the window
> + * (and the pages have not been returned via scif_put_pages()). A window is
> + * also referenced while an RMA, in which some range of the window is a source
> + * or destination, is in progress. Finally a window is referenced while some
> + * offset in that window was specified to scif_fence_signal(), and the RMAs
> + * marked by that call to scif_fence_signal() have not completed. While a
> + * window is in this state, its registered address space pages are not
> + * available for use in a new registered window.
> + *
> + * When all such references to the window have been removed, its references to
> + * all the physical pages which it represents are removed. Similarly, the
> + * registered address space pages of the window become available for
> + * registration in a new window.
> + *
> + * Return:
> + * Upon successful completion, scif_unregister() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned. In the event of an
> + * error, no windows are unregistered.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EINVAL - the range [offset, offset + len - 1] intersects a subrange of a
> + * window, or offset is negative
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - Offsets in the range [offset, offset + len - 1] are invalid for the
> + * registered address space of epd
> + */
> +int scif_unregister(scif_epd_t epd, off_t offset, size_t len);
> +
> +/**
> + * scif_readfrom() - Copy from a remote address space
> + * @epd:	endpoint descriptor
> + * @loffset:	offset in local registered address space to
> + *		which to copy
> + * @len:	length of range to copy
> + * @roffset:	offset in remote registered address space
> + *		from which to copy
> + * @rma_flags:	transfer mode flags
> + *
> + * scif_readfrom() copies len bytes from the remote registered address space of
> + * the peer of endpoint epd, starting at the offset roffset to the local
> + * registered address space of epd, starting at the offset loffset.
> + *
> + * Each of the specified ranges [loffset, loffset + len - 1] and [roffset,
> + * roffset + len - 1] must be within some registered window or windows of the
> + * local and remote nodes. A range may intersect multiple registered windows,
> + * but only if those windows are contiguous in the registered address space.
> + *
> + * If rma_flags includes SCIF_RMA_USECPU, then the data is copied using
> + * programmed read/writes. Otherwise the data is copied using DMA. If rma_-
> + * flags includes SCIF_RMA_SYNC, then scif_readfrom() will return after the
> + * transfer is complete. Otherwise, the transfer may be performed asynchron-
> + * ously. The order in which any two asynchronous RMA operations complete
> + * is non-deterministic. The synchronization functions, scif_fence_mark()/
> + * scif_fence_wait() and scif_fence_signal(), can be used to synchronize to
> + * the completion of asynchronous RMA operations on the same endpoint.
> + *
> + * The DMA transfer of individual bytes is not guaranteed to complete in
> + * address order. If rma_flags includes SCIF_RMA_ORDERED, then the last
> + * cacheline or partial cacheline of the source range will become visible on
> + * the destination node after all other transferred data in the source
> + * range has become visible on the destination node.
> + *
> + * The optimal DMA performance will likely be realized if both
> + * loffset and roffset are cacheline aligned (are a multiple of 64). Lower
> + * performance will likely be realized if loffset and roffset are not
> + * cacheline aligned but are separated by some multiple of 64. The lowest level
> + * of performance is likely if loffset and roffset are not separated by a
> + * multiple of 64.
> + *
> + * The rma_flags argument is formed by ORing together zero or more of the
> + * following values.
> + * SCIF_RMA_USECPU - perform the transfer using the CPU, otherwise use the DMA
> + *	engine.
> + * SCIF_RMA_SYNC - perform the transfer synchronously, returning after the
> + *		transfer has completed. Passing this flag results in the
> + *		current implementation busy waiting and consuming CPU cycles
> + *		while the DMA transfer is in progress for best performance by
> + *		avoiding the interrupt latency.
> + * SCIF_RMA_ORDERED - ensure that the last cacheline or partial cacheline of
> + *		the source range becomes visible on the destination node
> + *		after all other transferred data in the source range has
> + *		become visible on the destination
> + *
> + * Return:
> + * Upon successful completion, scif_readfrom() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EACCESS - Attempt to write to a read-only range
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EINVAL - rma_flags is invalid
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - The range [loffset, loffset + len - 1] is invalid for the registered
> + * address space of epd, or, The range [roffset, roffset + len - 1] is invalid
> + * for the registered address space of the peer of epd, or loffset or roffset
> + * is negative
> + */
> +int scif_readfrom(scif_epd_t epd, off_t loffset, size_t len, off_t
> +		  roffset, int rma_flags);
> +
> +/**
> + * scif_writeto() - Copy to a remote address space
> + * @epd:	endpoint descriptor
> + * @loffset:	offset in local registered address space
> + *		from which to copy
> + * @len:	length of range to copy
> + * @roffset:	offset in remote registered address space to
> + *		which to copy
> + * @rma_flags:	transfer mode flags
> + *
> + * scif_writeto() copies len bytes from the local registered address space of
> + * epd, starting at the offset loffset to the remote registered address space
> + * of the peer of endpoint epd, starting at the offset roffset.
> + *
> + * Each of the specified ranges [loffset, loffset + len - 1] and [roffset,
> + * roffset + len - 1] must be within some registered window or windows of the
> + * local and remote nodes. A range may intersect multiple registered windows,
> + * but only if those windows are contiguous in the registered address space.
> + *
> + * If rma_flags includes SCIF_RMA_USECPU, then the data is copied using
> + * programmed read/writes. Otherwise the data is copied using DMA. If rma_-
> + * flags includes SCIF_RMA_SYNC, then scif_writeto() will return after the
> + * transfer is complete. Otherwise, the transfer may be performed asynchron-
> + * ously. The order in which any two asynchronous RMA operations complete
> + * is non-deterministic. The synchronization functions, scif_fence_mark()/
> + * scif_fence_wait() and scif_fence_signal(), can be used to synchronize to
> + * the completion of asynchronous RMA operations on the same endpoint.
> + *
> + * The DMA transfer of individual bytes is not guaranteed to complete in
> + * address order. If rma_flags includes SCIF_RMA_ORDERED, then the last
> + * cacheline or partial cacheline of the source range will become visible on
> + * the destination node after all other transferred data in the source
> + * range has become visible on the destination node.
> + *
> + * The optimal DMA performance will likely be realized if both
> + * loffset and roffset are cacheline aligned (are a multiple of 64). Lower
> + * performance will likely be realized if loffset and roffset are not cacheline
> + * aligned but are separated by some multiple of 64. The lowest level of
> + * performance is likely if loffset and roffset are not separated by a multiple
> + * of 64.
> + *
> + * The rma_flags argument is formed by ORing together zero or more of the
> + * following values.
> + * SCIF_RMA_USECPU - perform the transfer using the CPU, otherwise use the DMA
> + *			engine.
> + * SCIF_RMA_SYNC - perform the transfer synchronously, returning after the
> + *		transfer has completed. Passing this flag results in the
> + *		current implementation busy waiting and consuming CPU cycles
> + *		while the DMA transfer is in progress for best performance by
> + *		avoiding the interrupt latency.
> + * SCIF_RMA_ORDERED - ensure that the last cacheline or partial cacheline of
> + *		the source range becomes visible on the destination node
> + *		after all other transferred data in the source range has
> + *		become visible on the destination
> + *
> + * Return:
> + * Upon successful completion, scif_readfrom() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EACCESS - Attempt to write to a read-only range
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EINVAL - rma_flags is invalid
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - The range [loffset, loffset + len - 1] is invalid for the registered
> + * address space of epd, or, The range [roffset , roffset + len -1] is invalid
> + * for the registered address space of the peer of epd, or loffset or roffset
> + * is negative
> + */
> +int scif_writeto(scif_epd_t epd, off_t loffset, size_t len, off_t
> +		 roffset, int rma_flags);
> +
> +/**
> + * scif_vreadfrom() - Copy from a remote address space
> + * @epd:	endpoint descriptor
> + * @addr:	address to which to copy
> + * @len:	length of range to copy
> + * @roffset:	offset in remote registered address space
> + *		from which to copy
> + * @rma_flags:	transfer mode flags
> + *
> + * scif_vreadfrom() copies len bytes from the remote registered address
> + * space of the peer of endpoint epd, starting at the offset roffset, to local
> + * memory, starting at addr.
> + *
> + * The specified range [roffset, roffset + len - 1] must be within some
> + * registered window or windows of the remote nodes. The range may
> + * intersect multiple registered windows, but only if those windows are
> + * contiguous in the registered address space.
> + *
> + * If rma_flags includes SCIF_RMA_USECPU, then the data is copied using
> + * programmed read/writes. Otherwise the data is copied using DMA. If rma_-
> + * flags includes SCIF_RMA_SYNC, then scif_vreadfrom() will return after the
> + * transfer is complete. Otherwise, the transfer may be performed asynchron-
> + * ously. The order in which any two asynchronous RMA operations complete
> + * is non-deterministic. The synchronization functions, scif_fence_mark()/
> + * scif_fence_wait() and scif_fence_signal(), can be used to synchronize to
> + * the completion of asynchronous RMA operations on the same endpoint.
> + *
> + * The DMA transfer of individual bytes is not guaranteed to complete in
> + * address order. If rma_flags includes SCIF_RMA_ORDERED, then the last
> + * cacheline or partial cacheline of the source range will become visible on
> + * the destination node after all other transferred data in the source
> + * range has become visible on the destination node.
> + *
> + * If rma_flags includes SCIF_RMA_USECACHE, then the physical pages which back
> + * the specified local memory range may be remain in a pinned state even after
> + * the specified transfer completes. This may reduce overhead if some or all of
> + * the same virtual address range is referenced in a subsequent call of
> + * scif_vreadfrom() or scif_vwriteto().
> + *
> + * The optimal DMA performance will likely be realized if both
> + * addr and roffset are cacheline aligned (are a multiple of 64). Lower
> + * performance will likely be realized if addr and roffset are not
> + * cacheline aligned but are separated by some multiple of 64. The lowest level
> + * of performance is likely if addr and roffset are not separated by a
> + * multiple of 64.
> + *
> + * The rma_flags argument is formed by ORing together zero or more of the
> + * following values.
> + * SCIF_RMA_USECPU - perform the transfer using the CPU, otherwise use the DMA
> + *	engine.
> + * SCIF_RMA_USECACHE - enable registration caching
> + * SCIF_RMA_SYNC - perform the transfer synchronously, returning after the
> + *		transfer has completed. Passing this flag results in the
> + *		current implementation busy waiting and consuming CPU cycles
> + *		while the DMA transfer is in progress for best performance by
> + *		avoiding the interrupt latency.
> + * SCIF_RMA_ORDERED - ensure that the last cacheline or partial cacheline of
> + *	the source range becomes visible on the destination node
> + *	after all other transferred data in the source range has
> + *	become visible on the destination
> + *
> + * Return:
> + * Upon successful completion, scif_vreadfrom() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EACCESS - Attempt to write to a read-only range
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EFAULT - Addresses in the range [addr, addr + len - 1] are invalid
> + * EINVAL - rma_flags is invalid
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - Offsets in the range [roffset, roffset + len - 1] are invalid for the
> + * registered address space of epd
> + */
> +int scif_vreadfrom(scif_epd_t epd, void *addr, size_t len, off_t roffset,
> +		   int rma_flags);
> +
> +/**
> + * scif_vwriteto() - Copy to a remote address space
> + * @epd:	endpoint descriptor
> + * @addr:	address from which to copy
> + * @len:	length of range to copy
> + * @roffset:	offset in remote registered address space to
> + *		which to copy
> + * @rma_flags:	transfer mode flags
> + *
> + * scif_vwriteto() copies len bytes from the local memory, starting at addr, to
> + * the remote registered address space of the peer of endpoint epd, starting at
> + * the offset roffset.
> + *
> + * The specified range [roffset, roffset + len - 1] must be within some
> + * registered window or windows of the remote nodes. The range may intersect
> + * multiple registered windows, but only if those windows are contiguous in the
> + * registered address space.
> + *
> + * If rma_flags includes SCIF_RMA_USECPU, then the data is copied using
> + * programmed read/writes. Otherwise the data is copied using DMA. If rma_-
> + * flags includes SCIF_RMA_SYNC, then scif_vwriteto() will return after the
> + * transfer is complete. Otherwise, the transfer may be performed asynchron-
> + * ously. The order in which any two asynchronous RMA operations complete
> + * is non-deterministic. The synchronization functions, scif_fence_mark()/
> + * scif_fence_wait() and scif_fence_signal(), can be used to synchronize to
> + * the completion of asynchronous RMA operations on the same endpoint.
> + *
> + * The DMA transfer of individual bytes is not guaranteed to complete in
> + * address order. If rma_flags includes SCIF_RMA_ORDERED, then the last
> + * cacheline or partial cacheline of the source range will become visible on
> + * the destination node after all other transferred data in the source
> + * range has become visible on the destination node.
> + *
> + * If rma_flags includes SCIF_RMA_USECACHE, then the physical pages which back
> + * the specified local memory range may be remain in a pinned state even after
> + * the specified transfer completes. This may reduce overhead if some or all of
> + * the same virtual address range is referenced in a subsequent call of
> + * scif_vreadfrom() or scif_vwriteto().
> + *
> + * The optimal DMA performance will likely be realized if both
> + * addr and offset are cacheline aligned (are a multiple of 64). Lower
> + * performance will likely be realized if addr and offset are not cacheline
> + * aligned but are separated by some multiple of 64. The lowest level of
> + * performance is likely if addr and offset are not separated by a multiple of
> + * 64.
> + *
> + * The rma_flags argument is formed by ORing together zero or more of the
> + * following values.
> + * SCIF_RMA_USECPU - perform the transfer using the CPU, otherwise use the DMA
> + *	engine.
> + * SCIF_RMA_USECACHE - allow registration caching
> + * SCIF_RMA_SYNC - perform the transfer synchronously, returning after the
> + *		transfer has completed. Passing this flag results in the
> + *		current implementation busy waiting and consuming CPU cycles
> + *		while the DMA transfer is in progress for best performance by
> + *		avoiding the interrupt latency.
> + * SCIF_RMA_ORDERED - ensure that the last cacheline or partial cacheline of
> + *		the source range becomes visible on the destination node
> + *		after all other transferred data in the source range has
> + *		become visible on the destination
> + *
> + * Return:
> + * Upon successful completion, scif_vwriteto() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EACCESS - Attempt to write to a read-only range
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EFAULT - Addresses in the range [addr, addr + len - 1] are invalid
> + * EINVAL - rma_flags is invalid
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - Offsets in the range [roffset, roffset + len - 1] are invalid for the
> + * registered address space of epd
> + */
> +int scif_vwriteto(scif_epd_t epd, void *addr, size_t len, off_t roffset,
> +		  int rma_flags);
> +
> +/**
> + * scif_fence_mark() - Mark previously issued RMAs
> + * @epd:	endpoint descriptor
> + * @flags:	control flags
> + * @mark:	marked value returned as output.
> + *
> + * scif_fence_mark() returns after marking the current set of all uncompleted
> + * RMAs initiated through the endpoint epd or the current set of all
> + * uncompleted RMAs initiated through the peer of endpoint epd. The RMAs are
> + * marked with a value returned at mark. The application may subsequently call
> + * scif_fence_wait(), passing the value returned at mark, to await completion
> + * of all RMAs so marked.
> + *
> + * The flags argument has exactly one of the following values.
> + * SCIF_FENCE_INIT_SELF - RMA operations initiated through endpoint
> + *	epd are marked
> + * SCIF_FENCE_INIT_PEER - RMA operations initiated through the peer
> + *	of endpoint epd are marked
> + *
> + * Return:
> + * Upon successful completion, scif_fence_mark() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EINVAL - flags is invalid
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENOMEM - Insufficient kernel memory was available
> + */
> +int scif_fence_mark(scif_epd_t epd, int flags, int *mark);
> +
> +/**
> + * scif_fence_wait() - Wait for completion of marked RMAs
> + * @epd:	endpoint descriptor
> + * @mark:	mark request
> + *
> + * scif_fence_wait() returns after all RMAs marked with mark have completed.
> + * The value passed in mark must have been obtained in a previous call to
> + * scif_fence_mark().
> + *
> + * Return:
> + * Upon successful completion, scif_fence_wait() returns 0; otherwise in user
> + * mode -1 is returned and errno is set to indicate the error; in kernel mode
> + * the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENOMEM - Insufficient kernel memory was available
> + */
> +int scif_fence_wait(scif_epd_t epd, int mark);
> +
> +/**
> + * scif_fence_signal() - Request a memory update on completion of RMAs
> + * @epd:	endpoint descriptor
> + * @loff:	local offset
> + * @lval:	local value to write to loffset
> + * @roff:	remote offset
> + * @rval:	remote value to write to roffset
> + * @flags:	flags
> + *
> + * scif_fence_signal() returns after marking the current set of all uncompleted
> + * RMAs initiated through the endpoint epd or marking the current set of all
> + * uncompleted RMAs initiated through the peer of endpoint epd.
> + *
> + * If flags includes SCIF_SIGNAL_LOCAL, then on completion of the RMAs in the
> + * marked set, lval is written to memory at the address corresponding to offset
> + * loff in the local registered address space of epd. loff must be within a
> + * registered window. If flags includes SCIF_SIGNAL_REMOTE, then on completion
> + * of the RMAs in the marked set, rval is written to memory at the address
> + * corresponding to offset roff in the remote registered address space of epd.
> + * roff must be within a remote registered window of the peer of epd. Note
> + * that any specified offset must be DWORD (4 byte / 32 bit) aligned.
> + *
> + * The flags argument is formed by OR'ing together the following.
> + * Exactly one of the following values.
> + * SCIF_FENCE_INIT_SELF - RMA operations initiated through endpoint
> + *	epd are marked
> + * SCIF_FENCE_INIT_PEER - RMA operations initiated through the peer
> + *	of endpoint epd are marked
> + * One or more of the following values.
> + * SCIF_SIGNAL_LOCAL - On completion of the marked set of RMAs, write lval to
> + *	memory at the address corresponding to offset loff in the local
> + *	registered address space of epd.
> + * SCIF_SIGNAL_REMOTE - On completion of the marked set of RMAs, write rval to
> + *	memory at the address corresponding to offset roff in the remote
> + *	registered address space of epd.
> + *
> + * Return:
> + * Upon successful completion, scif_fence_signal() returns 0; otherwise in
> + * user mode -1 is returned and errno is set to indicate the error; in kernel
> + * mode the negative of one of the following errors is returned.
> + *
> + * Errors:
> + * EBADF, ENOTTY - epd is not a valid endpoint descriptor
> + * ECONNRESET - Connection reset by peer
> + * EINVAL - flags is invalid, or loff or roff are not DWORD aligned
> + * ENODEV - The remote node is lost or existed, but is not currently in the
> + * network since it may have crashed
> + * ENOTCONN - The endpoint is not connected
> + * ENXIO - loff is invalid for the registered address of epd, or roff is invalid
> + * for the registered address space, of the peer of epd
> + */
> +int scif_fence_signal(scif_epd_t epd, off_t loff, u64 lval, off_t roff,
> +		      u64 rval, int flags);
> +
> +/**
> + * scif_get_node_ids() - Return information about online nodes
> + * @nodes:	array in which to return online node IDs
> + * @len:	number of entries in the nodes array
> + * @self:	address to place the node ID of the local node
> + *
> + * scif_get_node_ids() fills in the nodes array with up to len node IDs of the
> + * nodes in the SCIF network. If there is not enough space in nodes, as
> + * indicated by the len parameter, only len node IDs are returned in nodes. The
> + * return value of scif_get_node_ids() is the total number of nodes currently in
> + * the SCIF network. By checking the return value against the len parameter,
> + * the user may determine if enough space for nodes was allocated.
> + *
> + * The node ID of the local node is returned at self.
> + *
> + * Return:
> + * Upon successful completion, scif_get_node_ids() returns the actual number of
> + * online nodes in the SCIF network including 'self'; otherwise in user mode
> + * -1 is returned and errno is set to indicate the error; in kernel mode no
> + * errors are returned.
> + *
> + * Errors:
> + * EFAULT - Bad address
> + */
> +int scif_get_node_ids(u16 *nodes, int len, u16 *self);
> +
> +#endif /* __SCIF_H__ */
> diff --git a/include/uapi/linux/scif_ioctl.h b/include/uapi/linux/scif_ioctl.h
> new file mode 100644
> index 0000000..32449de
> --- /dev/null
> +++ b/include/uapi/linux/scif_ioctl.h
> @@ -0,0 +1,215 @@
> +/*
> + * Intel MIC Platform Software Stack (MPSS)
> + *
> + * This file is provided under a dual BSD/GPLv2 license.  When using or
> + * redistributing this file, you may do so under either license.
> + *
> + * GPL LICENSE SUMMARY
> + *
> + * Copyright(c) 2014 Intel Corporation.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of version 2 of the GNU General Public License as
> + * published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope that it will be useful, but
> + * WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * General Public License for more details.
> + *
> + * BSD LICENSE
> + *
> + * Copyright(c) 2014 Intel Corporation.
> + *
> + * Redistribution and use in source and binary forms, with or without
> + * modification, are permitted provided that the following conditions
> + * are met:
> + *
> + * * Redistributions of source code must retain the above copyright
> + *   notice, this list of conditions and the following disclaimer.
> + * * Redistributions in binary form must reproduce the above copyright
> + *   notice, this list of conditions and the following disclaimer in
> + *   the documentation and/or other materials provided with the
> + *   distribution.
> + * * Neither the name of Intel Corporation nor the names of its
> + *   contributors may be used to endorse or promote products derived
> + *   from this software without specific prior written permission.
> + *
> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + *
> + * Intel SCIF driver.
> + *
> + */
> +/*
> + * -----------------------------------------
> + * SCIF IOCTL interface information
> + * -----------------------------------------
> + */
> +#ifndef SCIF_IOCTL_H
> +#define SCIF_IOCTL_H
> +
> +#include <linux/types.h>
> +
> +/**
> + * struct scif_port_id - SCIF port information
> + * @node:	node on which port resides
> + * @port:	local port number
> + */
> +struct scif_port_id {
> +	__u16 node;
> +	__u16 port;
> +};
> +
> +/**
> + * struct scifioctl_connect - used for SCIF_CONNECT IOCTL
> + * @self:	used to read back the assigned port_id
> + * @peer:	destination node and port to connect to
> + */
> +struct scifioctl_connect {
> +	struct scif_port_id	self;
> +	struct scif_port_id	peer;
> +};
> +
> +/**
> + * struct scifioctl_accept - used for SCIF_ACCEPTREQ IOCTL
> + * @flags:	flags
> + * @peer:	global id of peer endpoint
> + * @endpt:	new connected endpoint descriptor
> + */
> +struct scifioctl_accept {
> +	__s32			flags;
> +	struct scif_port_id	peer;
> +	void			*endpt;
> +};
> +
> +/**
> + * struct scifioctl_msg - used for SCIF_SEND/SCIF_RECV IOCTL
> + * @msg:	message buffer address
> + * @len:	message length
> + * @flags:	flags
> + * @out_len:	number of bytes sent/received
> + */
> +struct scifioctl_msg {
> +	void	__user *msg;
> +	__s32	len;
> +	__s32	flags;
> +	__s32	out_len;
> +};
> +
> +/**
> + * struct scifioctl_reg - used for SCIF_REG IOCTL
> + * @addr:	starting virtual address
> + * @len:	length of range
> + * @offset:	offset of window
> + * @prot:	read/write protection
> + * @flags:	flags
> + * @out_offset:	offset returned
> + */
> +struct scifioctl_reg {
> +	void		*addr;
> +	__u64		len;
> +	__s64		offset;
> +	__s32		prot;
> +	__s32		flags;
> +	__s64		out_offset;
> +};
> +
> +/**
> + * struct scifioctl_unreg - used for SCIF_UNREG IOCTL
> + * @offset:	start of range to unregister
> + * @len:	length of range to unregister
> + */
> +struct scifioctl_unreg {
> +	__s64		offset;
> +	__u64		len;
> +};
> +
> +/**
> + * struct scifioctl_copy - used for SCIF DMA copy IOCTLs
> + *
> + * @loffset:	offset in local registered address space to/from
> + *		which to copy
> + * @len:	length of range to copy
> + * @roffset:	offset in remote registered address space to/from
> + *		which to copy
> + * @addr:	user virtual address to/from which to copy
> + * @flags:	flags
> + *
> + * This structure is used for SCIF_READFROM, SCIF_WRITETO, SCIF_VREADFROM
> + * and SCIF_VREADFROM IOCTL's.
> + */
> +struct scifioctl_copy {
> +	__s64		loffset;
> +	__u64		len;
> +	__s64		roffset;
> +	__u8		__user *addr;
> +	__s32		flags;
> +};
> +
> +/**
> + * struct scifioctl_fence_mark  - used for SCIF_FENCE_MARK IOCTL
> + * @flags:	flags
> + * @mark:	fence handle returned by reference
> + */
> +struct scifioctl_fence_mark {
> +	__s32	flags;
> +	__s32	__user *mark;
> +};
> +
> +/**
> + * struct scifioctl_fence_signal - used for SCIF_FENCE_SIGNAL IOCTL
> + * @loff:	local offset
> + * @lval:	local value to write to loffset
> + * @roff:	remote offset
> + * @rval:	remote value to write to roffset
> + * @flags:	flags
> + */
> +struct scifioctl_fence_signal {
> +	__s64		loff;
> +	__u64		lval;
> +	__s64		roff;
> +	__u64		rval;
> +	__s32		flags;
> +};
> +
> +/**
> + * struct scifioctl_node_ids - used for SCIF_GET_NODEIDS IOCTL
> + * @nodes:	pointer to an array of node_ids
> + * @len:	length of array
> + * @self:	ID of the current node
> + */
> +struct scifioctl_node_ids {
> +	__u16	__user *nodes;
> +	__s32	len;
> +	__u16	__user *self;
> +};
> +
> +#define SCIF_BIND		_IOWR('s', 1, __s32 *)
> +#define SCIF_LISTEN		_IOW('s', 2, __s32)
> +#define SCIF_CONNECT		_IOWR('s', 3, struct scifioctl_connect *)
> +#define SCIF_ACCEPTREQ		_IOWR('s', 4, struct scifioctl_accept *)
> +#define SCIF_ACCEPTREG		_IOWR('s', 5, void *)
> +#define SCIF_SEND		_IOWR('s', 6, struct scifioctl_msg *)
> +#define SCIF_RECV		_IOWR('s', 7, struct scifioctl_msg *)
> +#define SCIF_REG		_IOWR('s', 8, struct scifioctl_reg *)

I just picked one random ioctl, this one, and I don't see it used
anywhere in this series.  Why are you defining it if you don't use it?

Also, these are not the correct way to define ioctls, someone at Intel
should have caught this, which makes me really worry about the
user/kernel api here :(

Please fix.

greg k-h

  reply	other threads:[~2015-03-25 11:04 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-01-27  0:32 [PATCH char-misc-next v3 00/13] misc: mic: SCIF driver Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 01/13] misc: mic: SCIF header file and IOCTL interface Sudeep Dutt
2015-03-25 11:04   ` Greg Kroah-Hartman [this message]
2015-03-26 17:53     ` Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 02/13] misc: mic: SCIF ring buffer infrastructure Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 03/13] misc: mic: SCIF Hardware Bus Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 04/13] misc: mic: SCIF Peer Bus Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 05/13] misc: mic: Common MIC header file changes in preparation for SCIF Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 06/13] misc: mic: SCIF module initialization Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 07/13] misc: mic: SCIF node queue pair setup management Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 08/13] misc: mic: SCIF open close bind and listen APIs Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 09/13] misc: mic: SCIF connections APIs i.e. accept and connect Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 10/13] misc: mic: SCIF messaging and node enumeration APIs Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 11/13] misc: mic: MIC host driver specific changes to enable SCIF Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 12/13] misc: mic: MIC card " Sudeep Dutt
2015-01-27  0:32 ` [PATCH char-misc-next v3 13/13] misc: mic: add support for loading/unloading SCIF driver Sudeep Dutt
2015-02-24  0:45 ` [PATCH char-misc-next v3 00/13] misc: mic: " Sudeep Dutt
2015-03-09 16:13   ` Sudeep Dutt
2015-03-23 16:44     ` Sudeep Dutt

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150325110416.GA23335@kroah.com \
    --to=gregkh@linuxfoundation.org \
    --cc=arnd@arndb.de \
    --cc=ashutosh.dixit@intel.com \
    --cc=corbet@lwn.net \
    --cc=dave.jiang@intel.com \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=nikhil.rao@intel.com \
    --cc=sudeep.dutt@intel.com \
    --subject='Re: [PATCH char-misc-next v3 01/13] misc: mic: SCIF header file and IOCTL interface' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).