[linux-2.6-block.git] / net / sunrpc / xprtrdma / xprt_rdma.h

/*
 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the BSD-type
 * license below:
 *
 * 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 the Network Appliance, Inc. 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.
 */

#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
#define _LINUX_SUNRPC_XPRT_RDMA_H

#include <linux/wait.h> 		/* wait_queue_head_t, etc */
#include <linux/spinlock.h> 		/* spinlock_t, etc */
#include <linux/atomic.h>			/* atomic_t, etc */
#include <linux/workqueue.h>		/* struct work_struct */

#include <rdma/rdma_cm.h>		/* RDMA connection api */
#include <rdma/ib_verbs.h>		/* RDMA verbs api */

#include <linux/sunrpc/clnt.h> 		/* rpc_xprt */
#include <linux/sunrpc/rpc_rdma.h> 	/* RPC/RDMA protocol */
#include <linux/sunrpc/xprtrdma.h> 	/* xprt parameters */

#define RDMA_RESOLVE_TIMEOUT	(5000)	/* 5 seconds */
#define RDMA_CONNECT_RETRY_MAX	(2)	/* retries if no listener backlog */

/*
 * Interface Adapter -- one per transport instance
 */
struct rpcrdma_ia {
	struct rdma_cm_id 	*ri_id;
	struct ib_pd		*ri_pd;
	struct ib_mr		*ri_bind_mem;
	u32			ri_dma_lkey;
	int			ri_have_dma_lkey;
	struct completion	ri_done;
	int			ri_async_rc;
	enum rpcrdma_memreg	ri_memreg_strategy;
	unsigned int		ri_max_frmr_depth;
};

/*
 * RDMA Endpoint -- one per transport instance
 */

struct rpcrdma_ep {
	atomic_t		rep_cqcount;
	int			rep_cqinit;
	int			rep_connected;
	struct rpcrdma_ia	*rep_ia;
	struct ib_cq		*rep_cq;
	struct ib_qp_init_attr	rep_attr;
	wait_queue_head_t 	rep_connect_wait;
	struct ib_sge		rep_pad;	/* holds zeroed pad */
	struct ib_mr		*rep_pad_mr;	/* holds zeroed pad */
	void			(*rep_func)(struct rpcrdma_ep *);
	struct rpc_xprt		*rep_xprt;	/* for rep_func */
	struct rdma_conn_param	rep_remote_cma;
	struct sockaddr_storage	rep_remote_addr;
	struct delayed_work	rep_connect_worker;
};

#define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit)
#define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount)

/*
 * struct rpcrdma_rep -- this structure encapsulates state required to recv
 * and complete a reply, asychronously. It needs several pieces of
 * state:
 *   o recv buffer (posted to provider)
 *   o ib_sge (also donated to provider)
 *   o status of reply (length, success or not)
 *   o bookkeeping state to get run by tasklet (list, etc)
 *
 * These are allocated during initialization, per-transport instance;
 * however, the tasklet execution list itself is global, as it should
 * always be pretty short.
 *
 * N of these are associated with a transport instance, and stored in
 * struct rpcrdma_buffer. N is the max number of outstanding requests.
 */

/* temporary static scatter/gather max */
#define RPCRDMA_MAX_DATA_SEGS	(64)	/* max scatter/gather */
#define RPCRDMA_MAX_SEGS 	(RPCRDMA_MAX_DATA_SEGS + 2) /* head+tail = 2 */
#define MAX_RPCRDMAHDR	(\
	/* max supported RPC/RDMA header */ \
	sizeof(struct rpcrdma_msg) + (2 * sizeof(u32)) + \
	(sizeof(struct rpcrdma_read_chunk) * RPCRDMA_MAX_SEGS) + sizeof(u32))

struct rpcrdma_buffer;

struct rpcrdma_rep {
	unsigned int	rr_len;		/* actual received reply length */
	struct rpcrdma_buffer *rr_buffer; /* home base for this structure */
	struct rpc_xprt	*rr_xprt;	/* needed for request/reply matching */
	void (*rr_func)(struct rpcrdma_rep *);/* called by tasklet in softint */
	struct list_head rr_list;	/* tasklet list */
	wait_queue_head_t rr_unbind;	/* optional unbind wait */
	struct ib_sge	rr_iov;		/* for posting */
	struct ib_mr	*rr_handle;	/* handle for mem in rr_iov */
	char	rr_base[MAX_RPCRDMAHDR]; /* minimal inline receive buffer */
};

/*
 * struct rpcrdma_req -- structure central to the request/reply sequence.
 *
 * N of these are associated with a transport instance, and stored in
 * struct rpcrdma_buffer. N is the max number of outstanding requests.
 *
 * It includes pre-registered buffer memory for send AND recv.
 * The recv buffer, however, is not owned by this structure, and
 * is "donated" to the hardware when a recv is posted. When a
 * reply is handled, the recv buffer used is given back to the
 * struct rpcrdma_req associated with the request.
 *
 * In addition to the basic memory, this structure includes an array
 * of iovs for send operations. The reason is that the iovs passed to
 * ib_post_{send,recv} must not be modified until the work request
 * completes.
 *
 * NOTES:
 *   o RPCRDMA_MAX_SEGS is the max number of addressible chunk elements we
 *     marshal. The number needed varies depending on the iov lists that
 *     are passed to us, the memory registration mode we are in, and if
 *     physical addressing is used, the layout.
 */

struct rpcrdma_mr_seg {		/* chunk descriptors */
	union {				/* chunk memory handles */
		struct ib_mr	*rl_mr;		/* if registered directly */
		struct rpcrdma_mw {		/* if registered from region */
			union {
				struct ib_mw	*mw;
				struct ib_fmr	*fmr;
				struct {
					struct ib_fast_reg_page_list *fr_pgl;
					struct ib_mr *fr_mr;
					enum { FRMR_IS_INVALID, FRMR_IS_VALID  } state;
				} frmr;
			} r;
			struct list_head mw_list;
		} *rl_mw;
	} mr_chunk;
	u64		mr_base;	/* registration result */
	u32		mr_rkey;	/* registration result */
	u32		mr_len;		/* length of chunk or segment */
	int		mr_nsegs;	/* number of segments in chunk or 0 */
	enum dma_data_direction	mr_dir;	/* segment mapping direction */
	dma_addr_t	mr_dma;		/* segment mapping address */
	size_t		mr_dmalen;	/* segment mapping length */
	struct page	*mr_page;	/* owning page, if any */
	char		*mr_offset;	/* kva if no page, else offset */
};

struct rpcrdma_req {
	size_t 		rl_size;	/* actual length of buffer */
	unsigned int	rl_niovs;	/* 0, 2 or 4 */
	unsigned int	rl_nchunks;	/* non-zero if chunks */
	unsigned int	rl_connect_cookie;	/* retry detection */
	struct rpcrdma_buffer *rl_buffer; /* home base for this structure */
	struct rpcrdma_rep	*rl_reply;/* holder for reply buffer */
	struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];/* chunk segments */
	struct ib_sge	rl_send_iov[4];	/* for active requests */
	struct ib_sge	rl_iov;		/* for posting */
	struct ib_mr	*rl_handle;	/* handle for mem in rl_iov */
	char		rl_base[MAX_RPCRDMAHDR]; /* start of actual buffer */
	__u32 		rl_xdr_buf[0];	/* start of returned rpc rq_buffer */
};
#define rpcr_to_rdmar(r) \
	container_of((r)->rq_buffer, struct rpcrdma_req, rl_xdr_buf[0])

/*
 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
 * inline requests/replies, and client/server credits.
 *
 * One of these is associated with a transport instance
 */
struct rpcrdma_buffer {
	spinlock_t	rb_lock;	/* protects indexes */
	atomic_t	rb_credits;	/* most recent server credits */
	unsigned long	rb_cwndscale;	/* cached framework rpc_cwndscale */
	int		rb_max_requests;/* client max requests */
	struct list_head rb_mws;	/* optional memory windows/fmrs/frmrs */
	int		rb_send_index;
	struct rpcrdma_req	**rb_send_bufs;
	int		rb_recv_index;
	struct rpcrdma_rep	**rb_recv_bufs;
	char		*rb_pool;
};
#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)

/*
 * Internal structure for transport instance creation. This
 * exists primarily for modularity.
 *
 * This data should be set with mount options
 */
struct rpcrdma_create_data_internal {
	struct sockaddr_storage	addr;	/* RDMA server address */
	unsigned int	max_requests;	/* max requests (slots) in flight */
	unsigned int	rsize;		/* mount rsize - max read hdr+data */
	unsigned int	wsize;		/* mount wsize - max write hdr+data */
	unsigned int	inline_rsize;	/* max non-rdma read data payload */
	unsigned int	inline_wsize;	/* max non-rdma write data payload */
	unsigned int	padding;	/* non-rdma write header padding */
};

#define RPCRDMA_INLINE_READ_THRESHOLD(rq) \
	(rpcx_to_rdmad(rq->rq_xprt).inline_rsize)

#define RPCRDMA_INLINE_WRITE_THRESHOLD(rq)\
	(rpcx_to_rdmad(rq->rq_xprt).inline_wsize)

#define RPCRDMA_INLINE_PAD_VALUE(rq)\
	rpcx_to_rdmad(rq->rq_xprt).padding

/*
 * Statistics for RPCRDMA
 */
struct rpcrdma_stats {
	unsigned long		read_chunk_count;
	unsigned long		write_chunk_count;
	unsigned long		reply_chunk_count;

	unsigned long long	total_rdma_request;
	unsigned long long	total_rdma_reply;

	unsigned long long	pullup_copy_count;
	unsigned long long	fixup_copy_count;
	unsigned long		hardway_register_count;
	unsigned long		failed_marshal_count;
	unsigned long		bad_reply_count;
};

/*
 * RPCRDMA transport -- encapsulates the structures above for
 * integration with RPC.
 *
 * The contained structures are embedded, not pointers,
 * for convenience. This structure need not be visible externally.
 *
 * It is allocated and initialized during mount, and released
 * during unmount.
 */
struct rpcrdma_xprt {
	struct rpc_xprt		xprt;
	struct rpcrdma_ia	rx_ia;
	struct rpcrdma_ep	rx_ep;
	struct rpcrdma_buffer	rx_buf;
	struct rpcrdma_create_data_internal rx_data;
	struct delayed_work	rdma_connect;
	struct rpcrdma_stats	rx_stats;
};

#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, xprt)
#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)

/* Setting this to 0 ensures interoperability with early servers.
 * Setting this to 1 enhances certain unaligned read/write performance.
 * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
extern int xprt_rdma_pad_optimize;

/*
 * Interface Adapter calls - xprtrdma/verbs.c
 */
int rpcrdma_ia_open(struct rpcrdma_xprt *, struct sockaddr *, int);
void rpcrdma_ia_close(struct rpcrdma_ia *);

/*
 * Endpoint calls - xprtrdma/verbs.c
 */
int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
				struct rpcrdma_create_data_internal *);
int rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
int rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);

int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
				struct rpcrdma_req *);
int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_ep *,
				struct rpcrdma_rep *);

/*
 * Buffer calls - xprtrdma/verbs.c
 */
int rpcrdma_buffer_create(struct rpcrdma_buffer *, struct rpcrdma_ep *,
				struct rpcrdma_ia *,
				struct rpcrdma_create_data_internal *);
void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);

struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
void rpcrdma_buffer_put(struct rpcrdma_req *);
void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);

int rpcrdma_register_internal(struct rpcrdma_ia *, void *, int,
				struct ib_mr **, struct ib_sge *);
int rpcrdma_deregister_internal(struct rpcrdma_ia *,
				struct ib_mr *, struct ib_sge *);

int rpcrdma_register_external(struct rpcrdma_mr_seg *,
				int, int, struct rpcrdma_xprt *);
int rpcrdma_deregister_external(struct rpcrdma_mr_seg *,
				struct rpcrdma_xprt *, void *);

/*
 * RPC/RDMA connection management calls - xprtrdma/rpc_rdma.c
 */
void rpcrdma_connect_worker(struct work_struct *);
void rpcrdma_conn_func(struct rpcrdma_ep *);
void rpcrdma_reply_handler(struct rpcrdma_rep *);

/*
 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
 */
int rpcrdma_marshal_req(struct rpc_rqst *);

/* Temporary NFS request map cache. Created in svc_rdma.c  */
extern struct kmem_cache *svc_rdma_map_cachep;
/* WR context cache. Created in svc_rdma.c  */
extern struct kmem_cache *svc_rdma_ctxt_cachep;
/* Workqueue created in svc_rdma.c */
extern struct workqueue_struct *svc_rdma_wq;

#endif				/* _LINUX_SUNRPC_XPRT_RDMA_H */
Commit	Line	Data
f58851e6 TT	1	/*
	2	* Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the BSD-type
	8	* license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or without
	11	* modification, are permitted provided that the following conditions
	12	* are met:
	13	*
	14	* Redistributions of source code must retain the above copyright
	15	* notice, this list of conditions and the following disclaimer.
	16	*
	17	* Redistributions in binary form must reproduce the above
	18	* copyright notice, this list of conditions and the following
	19	* disclaimer in the documentation and/or other materials provided
	20	* with the distribution.
	21	*
	22	* Neither the name of the Network Appliance, Inc. nor the names of
	23	* its contributors may be used to endorse or promote products
	24	* derived from this software without specific prior written
	25	* permission.
	26	*
	27	* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
	28	* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
	29	* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
	30	* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
	31	* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
	32	* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
	33	* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
	34	* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
	35	* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
	36	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
	37	* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
	38	*/
	39
	40	#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
	41	#define _LINUX_SUNRPC_XPRT_RDMA_H
	42
	43	#include <linux/wait.h> /* wait_queue_head_t, etc */
	44	#include <linux/spinlock.h> /* spinlock_t, etc */
60063497	45	#include <linux/atomic.h> /* atomic_t, etc */
254f91e2	46	#include <linux/workqueue.h> /* struct work_struct */
f58851e6 TT	47
	48	#include <rdma/rdma_cm.h> /* RDMA connection api */
	49	#include <rdma/ib_verbs.h> /* RDMA verbs api */
	50
	51	#include <linux/sunrpc/clnt.h> /* rpc_xprt */
	52	#include <linux/sunrpc/rpc_rdma.h> /* RPC/RDMA protocol */
	53	#include <linux/sunrpc/xprtrdma.h> /* xprt parameters */
	54
5675add3 TT	55	#define RDMA_RESOLVE_TIMEOUT (5000) /* 5 seconds */
	56	#define RDMA_CONNECT_RETRY_MAX (2) /* retries if no listener backlog */
	57
f58851e6 TT	58	/*
	59	* Interface Adapter -- one per transport instance
	60	*/
	61	struct rpcrdma_ia {
	62	struct rdma_cm_id *ri_id;
	63	struct ib_pd *ri_pd;
	64	struct ib_mr *ri_bind_mem;
fe9053b3 TT	65	u32 ri_dma_lkey;
fe9053b3 TT	66	int ri_have_dma_lkey;
f58851e6 TT	67	struct completion ri_done;
	68	int ri_async_rc;
	69	enum rpcrdma_memreg ri_memreg_strategy;
0fc6c4e7	70	unsigned int ri_max_frmr_depth;
f58851e6 TT	71	};
	72
	73	/*
	74	* RDMA Endpoint -- one per transport instance
	75	*/
	76
	77	struct rpcrdma_ep {
	78	atomic_t rep_cqcount;
	79	int rep_cqinit;
	80	int rep_connected;
	81	struct rpcrdma_ia *rep_ia;
	82	struct ib_cq *rep_cq;
	83	struct ib_qp_init_attr rep_attr;
	84	wait_queue_head_t rep_connect_wait;
	85	struct ib_sge rep_pad; /* holds zeroed pad */
	86	struct ib_mr rep_pad_mr; / holds zeroed pad */
	87	void (rep_func)(struct rpcrdma_ep );
	88	struct rpc_xprt rep_xprt; / for rep_func */
	89	struct rdma_conn_param rep_remote_cma;
	90	struct sockaddr_storage rep_remote_addr;
254f91e2	91	struct delayed_work rep_connect_worker;
f58851e6 TT	92	};
	93
	94	#define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit)
	95	#define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount)
	96
	97	/*
	98	* struct rpcrdma_rep -- this structure encapsulates state required to recv
	99	* and complete a reply, asychronously. It needs several pieces of
	100	* state:
	101	* o recv buffer (posted to provider)
	102	* o ib_sge (also donated to provider)
	103	* o status of reply (length, success or not)
	104	* o bookkeeping state to get run by tasklet (list, etc)
	105	*
	106	* These are allocated during initialization, per-transport instance;
	107	* however, the tasklet execution list itself is global, as it should
	108	* always be pretty short.
	109	*
	110	* N of these are associated with a transport instance, and stored in
	111	* struct rpcrdma_buffer. N is the max number of outstanding requests.
	112	*/
	113
	114	/* temporary static scatter/gather max */
2773395b	115	#define RPCRDMA_MAX_DATA_SEGS (64) /* max scatter/gather */
f58851e6 TT	116	#define RPCRDMA_MAX_SEGS (RPCRDMA_MAX_DATA_SEGS + 2) /* head+tail = 2 */
	117	#define MAX_RPCRDMAHDR (\
	118	/* max supported RPC/RDMA header */ \
	119	sizeof(struct rpcrdma_msg) + (2 * sizeof(u32)) + \
	120	(sizeof(struct rpcrdma_read_chunk) * RPCRDMA_MAX_SEGS) + sizeof(u32))
	121
	122	struct rpcrdma_buffer;
	123
	124	struct rpcrdma_rep {
	125	unsigned int rr_len; /* actual received reply length */
	126	struct rpcrdma_buffer rr_buffer; / home base for this structure */
	127	struct rpc_xprt rr_xprt; / needed for request/reply matching */
	128	void (rr_func)(struct rpcrdma_rep );/* called by tasklet in softint */
	129	struct list_head rr_list; /* tasklet list */
	130	wait_queue_head_t rr_unbind; /* optional unbind wait */
	131	struct ib_sge rr_iov; /* for posting */
	132	struct ib_mr rr_handle; / handle for mem in rr_iov */
	133	char rr_base[MAX_RPCRDMAHDR]; /* minimal inline receive buffer */
	134	};
	135
	136	/*
	137	* struct rpcrdma_req -- structure central to the request/reply sequence.
	138	*
	139	* N of these are associated with a transport instance, and stored in
	140	* struct rpcrdma_buffer. N is the max number of outstanding requests.
	141	*
	142	* It includes pre-registered buffer memory for send AND recv.
	143	* The recv buffer, however, is not owned by this structure, and
	144	* is "donated" to the hardware when a recv is posted. When a
	145	* reply is handled, the recv buffer used is given back to the
	146	* struct rpcrdma_req associated with the request.
	147	*
	148	* In addition to the basic memory, this structure includes an array
	149	* of iovs for send operations. The reason is that the iovs passed to
	150	* ib_post_{send,recv} must not be modified until the work request
	151	* completes.
	152	*
	153	* NOTES:
	154	* o RPCRDMA_MAX_SEGS is the max number of addressible chunk elements we
	155	* marshal. The number needed varies depending on the iov lists that
	156	* are passed to us, the memory registration mode we are in, and if
	157	* physical addressing is used, the layout.
	158	*/
	159
	160	struct rpcrdma_mr_seg { /* chunk descriptors */
	161	union { /* chunk memory handles */
	162	struct ib_mr rl_mr; / if registered directly */
	163	struct rpcrdma_mw { /* if registered from region */
	164	union {
	165	struct ib_mw *mw;
	166	struct ib_fmr *fmr;
fe9053b3 TT	167	struct {
	168	struct ib_fast_reg_page_list *fr_pgl;
	169	struct ib_mr *fr_mr;
5c635e09	170	enum { FRMR_IS_INVALID, FRMR_IS_VALID } state;
fe9053b3	171	} frmr;
f58851e6 TT	172	} r;
	173	struct list_head mw_list;
	174	} *rl_mw;
	175	} mr_chunk;
	176	u64 mr_base; /* registration result */
	177	u32 mr_rkey; /* registration result */
	178	u32 mr_len; /* length of chunk or segment */
	179	int mr_nsegs; /* number of segments in chunk or 0 */
	180	enum dma_data_direction mr_dir; /* segment mapping direction */
	181	dma_addr_t mr_dma; /* segment mapping address */
	182	size_t mr_dmalen; /* segment mapping length */
	183	struct page mr_page; / owning page, if any */
	184	char mr_offset; / kva if no page, else offset */
	185	};
	186
	187	struct rpcrdma_req {
	188	size_t rl_size; /* actual length of buffer */
	189	unsigned int rl_niovs; /* 0, 2 or 4 */
	190	unsigned int rl_nchunks; /* non-zero if chunks */
575448bd	191	unsigned int rl_connect_cookie; /* retry detection */
f58851e6 TT	192	struct rpcrdma_buffer rl_buffer; / home base for this structure */
	193	struct rpcrdma_rep rl_reply;/ holder for reply buffer */
	194	struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];/* chunk segments */
	195	struct ib_sge rl_send_iov[4]; /* for active requests */
	196	struct ib_sge rl_iov; /* for posting */
	197	struct ib_mr rl_handle; / handle for mem in rl_iov */
	198	char rl_base[MAX_RPCRDMAHDR]; /* start of actual buffer */
	199	__u32 rl_xdr_buf[0]; /* start of returned rpc rq_buffer */
	200	};
	201	#define rpcr_to_rdmar(r) \
	202	container_of((r)->rq_buffer, struct rpcrdma_req, rl_xdr_buf[0])
	203
	204	/*
	205	* struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
	206	* inline requests/replies, and client/server credits.
	207	*
	208	* One of these is associated with a transport instance
	209	*/
	210	struct rpcrdma_buffer {
	211	spinlock_t rb_lock; /* protects indexes */
	212	atomic_t rb_credits; /* most recent server credits */
	213	unsigned long rb_cwndscale; /* cached framework rpc_cwndscale */
	214	int rb_max_requests;/* client max requests */
fe9053b3	215	struct list_head rb_mws; /* optional memory windows/fmrs/frmrs */
f58851e6 TT	216	int rb_send_index;
	217	struct rpcrdma_req **rb_send_bufs;
	218	int rb_recv_index;
	219	struct rpcrdma_rep **rb_recv_bufs;
	220	char *rb_pool;
	221	};
	222	#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
	223
	224	/*
	225	* Internal structure for transport instance creation. This
	226	* exists primarily for modularity.
	227	*
	228	* This data should be set with mount options
	229	*/
	230	struct rpcrdma_create_data_internal {
	231	struct sockaddr_storage addr; /* RDMA server address */
	232	unsigned int max_requests; /* max requests (slots) in flight */
	233	unsigned int rsize; /* mount rsize - max read hdr+data */
	234	unsigned int wsize; /* mount wsize - max write hdr+data */
	235	unsigned int inline_rsize; /* max non-rdma read data payload */
	236	unsigned int inline_wsize; /* max non-rdma write data payload */
	237	unsigned int padding; /* non-rdma write header padding */
	238	};
	239
	240	#define RPCRDMA_INLINE_READ_THRESHOLD(rq) \
a4f0835c	241	(rpcx_to_rdmad(rq->rq_xprt).inline_rsize)
f58851e6 TT	242
f58851e6 TT	243	#define RPCRDMA_INLINE_WRITE_THRESHOLD(rq)\
a4f0835c	244	(rpcx_to_rdmad(rq->rq_xprt).inline_wsize)
f58851e6 TT	245
f58851e6 TT	246	#define RPCRDMA_INLINE_PAD_VALUE(rq)\
a4f0835c	247	rpcx_to_rdmad(rq->rq_xprt).padding
f58851e6 TT	248
	249	/*
	250	* Statistics for RPCRDMA
	251	*/
	252	struct rpcrdma_stats {
	253	unsigned long read_chunk_count;
	254	unsigned long write_chunk_count;
	255	unsigned long reply_chunk_count;
	256
	257	unsigned long long total_rdma_request;
	258	unsigned long long total_rdma_reply;
	259
	260	unsigned long long pullup_copy_count;
	261	unsigned long long fixup_copy_count;
	262	unsigned long hardway_register_count;
	263	unsigned long failed_marshal_count;
	264	unsigned long bad_reply_count;
	265	};
	266
	267	/*
	268	* RPCRDMA transport -- encapsulates the structures above for
	269	* integration with RPC.
	270	*
	271	* The contained structures are embedded, not pointers,
	272	* for convenience. This structure need not be visible externally.
	273	*
	274	* It is allocated and initialized during mount, and released
	275	* during unmount.
	276	*/
	277	struct rpcrdma_xprt {
	278	struct rpc_xprt xprt;
	279	struct rpcrdma_ia rx_ia;
	280	struct rpcrdma_ep rx_ep;
	281	struct rpcrdma_buffer rx_buf;
	282	struct rpcrdma_create_data_internal rx_data;
	283	struct delayed_work rdma_connect;
	284	struct rpcrdma_stats rx_stats;
	285	};
	286
	287	#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, xprt)
	288	#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
	289
9191ca3b TT	290	/* Setting this to 0 ensures interoperability with early servers.
	291	* Setting this to 1 enhances certain unaligned read/write performance.
	292	* Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
	293	extern int xprt_rdma_pad_optimize;
	294
f58851e6 TT	295	/*
	296	* Interface Adapter calls - xprtrdma/verbs.c
	297	*/
	298	int rpcrdma_ia_open(struct rpcrdma_xprt , struct sockaddr , int);
	299	void rpcrdma_ia_close(struct rpcrdma_ia *);
	300
	301	/*
	302	* Endpoint calls - xprtrdma/verbs.c
	303	*/
	304	int rpcrdma_ep_create(struct rpcrdma_ep , struct rpcrdma_ia ,
	305	struct rpcrdma_create_data_internal *);
	306	int rpcrdma_ep_destroy(struct rpcrdma_ep , struct rpcrdma_ia );
	307	int rpcrdma_ep_connect(struct rpcrdma_ep , struct rpcrdma_ia );
	308	int rpcrdma_ep_disconnect(struct rpcrdma_ep , struct rpcrdma_ia );
	309
	310	int rpcrdma_ep_post(struct rpcrdma_ia , struct rpcrdma_ep ,
	311	struct rpcrdma_req *);
	312	int rpcrdma_ep_post_recv(struct rpcrdma_ia , struct rpcrdma_ep ,
	313	struct rpcrdma_rep *);
	314
	315	/*
	316	* Buffer calls - xprtrdma/verbs.c
	317	*/
	318	int rpcrdma_buffer_create(struct rpcrdma_buffer , struct rpcrdma_ep ,
	319	struct rpcrdma_ia *,
	320	struct rpcrdma_create_data_internal *);
	321	void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
	322
	323	struct rpcrdma_req rpcrdma_buffer_get(struct rpcrdma_buffer );
	324	void rpcrdma_buffer_put(struct rpcrdma_req *);
	325	void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
	326	void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
	327
	328	int rpcrdma_register_internal(struct rpcrdma_ia , void , int,
	329	struct ib_mr *, struct ib_sge );
	330	int rpcrdma_deregister_internal(struct rpcrdma_ia *,
	331	struct ib_mr , struct ib_sge );
	332
	333	int rpcrdma_register_external(struct rpcrdma_mr_seg *,
	334	int, int, struct rpcrdma_xprt *);
	335	int rpcrdma_deregister_external(struct rpcrdma_mr_seg *,
	336	struct rpcrdma_xprt , void );
	337
	338	/*
	339	* RPC/RDMA connection management calls - xprtrdma/rpc_rdma.c
	340	*/
254f91e2	341	void rpcrdma_connect_worker(struct work_struct *);
f58851e6 TT	342	void rpcrdma_conn_func(struct rpcrdma_ep *);
	343	void rpcrdma_reply_handler(struct rpcrdma_rep *);
	344
	345	/*
	346	* RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
	347	*/
	348	int rpcrdma_marshal_req(struct rpc_rqst *);
	349
cec56c8f TT	350	/* Temporary NFS request map cache. Created in svc_rdma.c */
	351	extern struct kmem_cache *svc_rdma_map_cachep;
	352	/* WR context cache. Created in svc_rdma.c */
	353	extern struct kmem_cache *svc_rdma_ctxt_cachep;
	354	/* Workqueue created in svc_rdma.c */
	355	extern struct workqueue_struct *svc_rdma_wq;
	356
f58851e6	357	#endif /* _LINUX_SUNRPC_XPRT_RDMA_H */