[deliverable/linux.git] / net / rds / ib.h

#ifndef _RDS_IB_H
#define _RDS_IB_H

#include <rdma/ib_verbs.h>
#include <rdma/rdma_cm.h>
#include <linux/interrupt.h>
#include <linux/pci.h>
#include <linux/slab.h>
#include "rds.h"
#include "rdma_transport.h"

#define RDS_FMR_SIZE			256
#define RDS_FMR_POOL_SIZE		8192

#define RDS_IB_MAX_SGE			8
#define RDS_IB_RECV_SGE 		2

#define RDS_IB_DEFAULT_RECV_WR		1024
#define RDS_IB_DEFAULT_SEND_WR		256

#define RDS_IB_DEFAULT_RETRY_COUNT	2

#define RDS_IB_SUPPORTED_PROTOCOLS	0x00000003	/* minor versions supported */

#define RDS_IB_RECYCLE_BATCH_COUNT	32

#define RDS_IB_WC_MAX			32

extern struct rw_semaphore rds_ib_devices_lock;
extern struct list_head rds_ib_devices;

/*
 * IB posts RDS_FRAG_SIZE fragments of pages to the receive queues to
 * try and minimize the amount of memory tied up both the device and
 * socket receive queues.
 */
struct rds_page_frag {
	struct list_head	f_item;
	struct list_head	f_cache_entry;
	struct scatterlist	f_sg;
};

struct rds_ib_incoming {
	struct list_head	ii_frags;
	struct list_head	ii_cache_entry;
	struct rds_incoming	ii_inc;
};

struct rds_ib_cache_head {
	struct list_head *first;
	unsigned long count;
};

struct rds_ib_refill_cache {
	struct rds_ib_cache_head __percpu *percpu;
	struct list_head	 *xfer;
	struct list_head	 *ready;
};

struct rds_ib_connect_private {
	/* Add new fields at the end, and don't permute existing fields. */
	__be32			dp_saddr;
	__be32			dp_daddr;
	u8			dp_protocol_major;
	u8			dp_protocol_minor;
	__be16			dp_protocol_minor_mask; /* bitmask */
	__be32			dp_reserved1;
	__be64			dp_ack_seq;
	__be32			dp_credit;		/* non-zero enables flow ctl */
};

struct rds_ib_send_work {
	void			*s_op;
	struct ib_send_wr	s_wr;
	struct ib_sge		s_sge[RDS_IB_MAX_SGE];
	unsigned long		s_queued;
};

struct rds_ib_recv_work {
	struct rds_ib_incoming 	*r_ibinc;
	struct rds_page_frag	*r_frag;
	struct ib_recv_wr	r_wr;
	struct ib_sge		r_sge[2];
};

struct rds_ib_work_ring {
	u32		w_nr;
	u32		w_alloc_ptr;
	u32		w_alloc_ctr;
	u32		w_free_ptr;
	atomic_t	w_free_ctr;
};

/* Rings are posted with all the allocations they'll need to queue the
 * incoming message to the receiving socket so this can't fail.
 * All fragments start with a header, so we can make sure we're not receiving
 * garbage, and we can tell a small 8 byte fragment from an ACK frame.
 */
struct rds_ib_ack_state {
	u64		ack_next;
	u64		ack_recv;
	unsigned int	ack_required:1;
	unsigned int	ack_next_valid:1;
	unsigned int	ack_recv_valid:1;
};


struct rds_ib_device;

struct rds_ib_connection {

	struct list_head	ib_node;
	struct rds_ib_device	*rds_ibdev;
	struct rds_connection	*conn;

	/* alphabet soup, IBTA style */
	struct rdma_cm_id	*i_cm_id;
	struct ib_pd		*i_pd;
	struct ib_cq		*i_send_cq;
	struct ib_cq		*i_recv_cq;
	struct ib_wc		i_recv_wc[RDS_IB_WC_MAX];

	/* interrupt handling */
	struct tasklet_struct	i_recv_tasklet;

	/* tx */
	struct rds_ib_work_ring	i_send_ring;
	struct rm_data_op	*i_data_op;
	struct rds_header	*i_send_hdrs;
	u64			i_send_hdrs_dma;
	struct rds_ib_send_work *i_sends;
	atomic_t		i_signaled_sends;

	/* rx */
	struct mutex		i_recv_mutex;
	struct rds_ib_work_ring	i_recv_ring;
	struct rds_ib_incoming	*i_ibinc;
	u32			i_recv_data_rem;
	struct rds_header	*i_recv_hdrs;
	u64			i_recv_hdrs_dma;
	struct rds_ib_recv_work *i_recvs;
	u64			i_ack_recv;	/* last ACK received */
	struct rds_ib_refill_cache i_cache_incs;
	struct rds_ib_refill_cache i_cache_frags;

	/* sending acks */
	unsigned long		i_ack_flags;
#ifdef KERNEL_HAS_ATOMIC64
	atomic64_t		i_ack_next;	/* next ACK to send */
#else
	spinlock_t		i_ack_lock;	/* protect i_ack_next */
	u64			i_ack_next;	/* next ACK to send */
#endif
	struct rds_header	*i_ack;
	struct ib_send_wr	i_ack_wr;
	struct ib_sge		i_ack_sge;
	u64			i_ack_dma;
	unsigned long		i_ack_queued;

	/* Flow control related information
	 *
	 * Our algorithm uses a pair variables that we need to access
	 * atomically - one for the send credits, and one posted
	 * recv credits we need to transfer to remote.
	 * Rather than protect them using a slow spinlock, we put both into
	 * a single atomic_t and update it using cmpxchg
	 */
	atomic_t		i_credits;

	/* Protocol version specific information */
	unsigned int		i_flowctl:1;	/* enable/disable flow ctl */

	/* Batched completions */
	unsigned int		i_unsignaled_wrs;
};

/* This assumes that atomic_t is at least 32 bits */
#define IB_GET_SEND_CREDITS(v)	((v) & 0xffff)
#define IB_GET_POST_CREDITS(v)	((v) >> 16)
#define IB_SET_SEND_CREDITS(v)	((v) & 0xffff)
#define IB_SET_POST_CREDITS(v)	((v) << 16)

struct rds_ib_ipaddr {
	struct list_head	list;
	__be32			ipaddr;
	struct rcu_head		rcu;
};

struct rds_ib_device {
	struct list_head	list;
	struct list_head	ipaddr_list;
	struct list_head	conn_list;
	struct ib_device	*dev;
	struct ib_pd		*pd;
	struct rds_ib_mr_pool	*mr_pool;
	unsigned int		fmr_max_remaps;
	unsigned int		max_fmrs;
	int			max_sge;
	unsigned int		max_wrs;
	unsigned int		max_initiator_depth;
	unsigned int		max_responder_resources;
	spinlock_t		spinlock;	/* protect the above */
	atomic_t		refcount;
	struct work_struct	free_work;
};

#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
#define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)

/* bits for i_ack_flags */
#define IB_ACK_IN_FLIGHT	0
#define IB_ACK_REQUESTED	1

/* Magic WR_ID for ACKs */
#define RDS_IB_ACK_WR_ID	(~(u64) 0)

struct rds_ib_statistics {
	uint64_t	s_ib_connect_raced;
	uint64_t	s_ib_listen_closed_stale;
	uint64_t	s_ib_tx_cq_call;
	uint64_t	s_ib_evt_handler_call;
	uint64_t	s_ib_tasklet_call;
	uint64_t	s_ib_tx_cq_event;
	uint64_t	s_ib_tx_ring_full;
	uint64_t	s_ib_tx_throttle;
	uint64_t	s_ib_tx_sg_mapping_failure;
	uint64_t	s_ib_tx_stalled;
	uint64_t	s_ib_tx_credit_updates;
	uint64_t	s_ib_rx_cq_event;
	uint64_t	s_ib_rx_ring_empty;
	uint64_t	s_ib_rx_refill_from_cq;
	uint64_t	s_ib_rx_refill_from_thread;
	uint64_t	s_ib_rx_alloc_limit;
	uint64_t	s_ib_rx_credit_updates;
	uint64_t	s_ib_ack_sent;
	uint64_t	s_ib_ack_send_failure;
	uint64_t	s_ib_ack_send_delayed;
	uint64_t	s_ib_ack_send_piggybacked;
	uint64_t	s_ib_ack_received;
	uint64_t	s_ib_rdma_mr_alloc;
	uint64_t	s_ib_rdma_mr_free;
	uint64_t	s_ib_rdma_mr_used;
	uint64_t	s_ib_rdma_mr_pool_flush;
	uint64_t	s_ib_rdma_mr_pool_wait;
	uint64_t	s_ib_rdma_mr_pool_depleted;
	uint64_t	s_ib_atomic_cswp;
	uint64_t	s_ib_atomic_fadd;
};

extern struct workqueue_struct *rds_ib_wq;

/*
 * Fake ib_dma_sync_sg_for_{cpu,device} as long as ib_verbs.h
 * doesn't define it.
 */
static inline void rds_ib_dma_sync_sg_for_cpu(struct ib_device *dev,
					      struct scatterlist *sglist,
					      unsigned int sg_dma_len,
					      int direction)
{
	struct scatterlist *sg;
	unsigned int i;

	for_each_sg(sglist, sg, sg_dma_len, i) {
		ib_dma_sync_single_for_cpu(dev,
				ib_sg_dma_address(dev, sg),
				ib_sg_dma_len(dev, sg),
				direction);
	}
}
#define ib_dma_sync_sg_for_cpu	rds_ib_dma_sync_sg_for_cpu

static inline void rds_ib_dma_sync_sg_for_device(struct ib_device *dev,
						 struct scatterlist *sglist,
						 unsigned int sg_dma_len,
						 int direction)
{
	struct scatterlist *sg;
	unsigned int i;

	for_each_sg(sglist, sg, sg_dma_len, i) {
		ib_dma_sync_single_for_device(dev,
				ib_sg_dma_address(dev, sg),
				ib_sg_dma_len(dev, sg),
				direction);
	}
}
#define ib_dma_sync_sg_for_device	rds_ib_dma_sync_sg_for_device


/* ib.c */
extern struct rds_transport rds_ib_transport;
struct rds_ib_device *rds_ib_get_client_data(struct ib_device *device);
void rds_ib_dev_put(struct rds_ib_device *rds_ibdev);
extern struct ib_client rds_ib_client;

extern unsigned int fmr_message_size;
extern unsigned int rds_ib_retry_count;

extern spinlock_t ib_nodev_conns_lock;
extern struct list_head ib_nodev_conns;

/* ib_cm.c */
int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp);
void rds_ib_conn_free(void *arg);
int rds_ib_conn_connect(struct rds_connection *conn);
void rds_ib_conn_shutdown(struct rds_connection *conn);
void rds_ib_state_change(struct sock *sk);
int rds_ib_listen_init(void);
void rds_ib_listen_stop(void);
void __rds_ib_conn_error(struct rds_connection *conn, const char *, ...);
int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
			     struct rdma_cm_event *event);
int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id);
void rds_ib_cm_connect_complete(struct rds_connection *conn,
				struct rdma_cm_event *event);


#define rds_ib_conn_error(conn, fmt...) \
	__rds_ib_conn_error(conn, KERN_WARNING "RDS/IB: " fmt)

/* ib_rdma.c */
int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr);
void rds_ib_add_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn);
void rds_ib_remove_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn);
void rds_ib_destroy_nodev_conns(void);
struct rds_ib_mr_pool *rds_ib_create_mr_pool(struct rds_ib_device *);
void rds_ib_get_mr_info(struct rds_ib_device *rds_ibdev, struct rds_info_rdma_connection *iinfo);
void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *);
void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
		    struct rds_sock *rs, u32 *key_ret);
void rds_ib_sync_mr(void *trans_private, int dir);
void rds_ib_free_mr(void *trans_private, int invalidate);
void rds_ib_flush_mrs(void);
int rds_ib_fmr_init(void);
void rds_ib_fmr_exit(void);

/* ib_recv.c */
int rds_ib_recv_init(void);
void rds_ib_recv_exit(void);
int rds_ib_recv(struct rds_connection *conn);
int rds_ib_recv_alloc_caches(struct rds_ib_connection *ic);
void rds_ib_recv_free_caches(struct rds_ib_connection *ic);
void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp);
void rds_ib_inc_free(struct rds_incoming *inc);
int rds_ib_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to);
void rds_ib_recv_cqe_handler(struct rds_ib_connection *ic, struct ib_wc *wc,
			     struct rds_ib_ack_state *state);
void rds_ib_recv_tasklet_fn(unsigned long data);
void rds_ib_recv_init_ring(struct rds_ib_connection *ic);
void rds_ib_recv_clear_ring(struct rds_ib_connection *ic);
void rds_ib_recv_init_ack(struct rds_ib_connection *ic);
void rds_ib_attempt_ack(struct rds_ib_connection *ic);
void rds_ib_ack_send_complete(struct rds_ib_connection *ic);
u64 rds_ib_piggyb_ack(struct rds_ib_connection *ic);
void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq, int ack_required);

/* ib_ring.c */
void rds_ib_ring_init(struct rds_ib_work_ring *ring, u32 nr);
void rds_ib_ring_resize(struct rds_ib_work_ring *ring, u32 nr);
u32 rds_ib_ring_alloc(struct rds_ib_work_ring *ring, u32 val, u32 *pos);
void rds_ib_ring_free(struct rds_ib_work_ring *ring, u32 val);
void rds_ib_ring_unalloc(struct rds_ib_work_ring *ring, u32 val);
int rds_ib_ring_empty(struct rds_ib_work_ring *ring);
int rds_ib_ring_low(struct rds_ib_work_ring *ring);
u32 rds_ib_ring_oldest(struct rds_ib_work_ring *ring);
u32 rds_ib_ring_completed(struct rds_ib_work_ring *ring, u32 wr_id, u32 oldest);
extern wait_queue_head_t rds_ib_ring_empty_wait;

/* ib_send.c */
void rds_ib_xmit_complete(struct rds_connection *conn);
int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm,
		unsigned int hdr_off, unsigned int sg, unsigned int off);
void rds_ib_send_cq_comp_handler(struct ib_cq *cq, void *context);
void rds_ib_send_init_ring(struct rds_ib_connection *ic);
void rds_ib_send_clear_ring(struct rds_ib_connection *ic);
int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op);
void rds_ib_send_add_credits(struct rds_connection *conn, unsigned int credits);
void rds_ib_advertise_credits(struct rds_connection *conn, unsigned int posted);
int rds_ib_send_grab_credits(struct rds_ib_connection *ic, u32 wanted,
			     u32 *adv_credits, int need_posted, int max_posted);
int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op);

/* ib_stats.c */
DECLARE_PER_CPU(struct rds_ib_statistics, rds_ib_stats);
#define rds_ib_stats_inc(member) rds_stats_inc_which(rds_ib_stats, member)
unsigned int rds_ib_stats_info_copy(struct rds_info_iterator *iter,
				    unsigned int avail);

/* ib_sysctl.c */
int rds_ib_sysctl_init(void);
void rds_ib_sysctl_exit(void);
extern unsigned long rds_ib_sysctl_max_send_wr;
extern unsigned long rds_ib_sysctl_max_recv_wr;
extern unsigned long rds_ib_sysctl_max_unsig_wrs;
extern unsigned long rds_ib_sysctl_max_unsig_bytes;
extern unsigned long rds_ib_sysctl_max_recv_allocation;
extern unsigned int rds_ib_sysctl_flow_control;

#endif
Commit	Line	Data
ec16227e AG	1	#ifndef _RDS_IB_H
	2	#define _RDS_IB_H
	3
	4	#include <rdma/ib_verbs.h>
	5	#include <rdma/rdma_cm.h>
a6b7a407	6	#include <linux/interrupt.h>
e4c52c98 AG	7	#include <linux/pci.h>
e4c52c98 AG	8	#include <linux/slab.h>
ec16227e AG	9	#include "rds.h"
	10	#include "rdma_transport.h"
	11
	12	#define RDS_FMR_SIZE 256
eabb7322	13	#define RDS_FMR_POOL_SIZE 8192
ec16227e AG	14
	15	#define RDS_IB_MAX_SGE 8
	16	#define RDS_IB_RECV_SGE 2
	17
	18	#define RDS_IB_DEFAULT_RECV_WR 1024
	19	#define RDS_IB_DEFAULT_SEND_WR 256
	20
3ba23ade AG	21	#define RDS_IB_DEFAULT_RETRY_COUNT 2
3ba23ade AG	22
ec16227e AG	23	#define RDS_IB_SUPPORTED_PROTOCOLS 0x00000003 /* minor versions supported */
ec16227e AG	24
33244125 CM	25	#define RDS_IB_RECYCLE_BATCH_COUNT 32
33244125 CM	26
f4f943c9 SS	27	#define RDS_IB_WC_MAX 32
f4f943c9 SS	28
ea819867	29	extern struct rw_semaphore rds_ib_devices_lock;
ec16227e AG	30	extern struct list_head rds_ib_devices;
	31
	32	/*
	33	* IB posts RDS_FRAG_SIZE fragments of pages to the receive queues to
	34	* try and minimize the amount of memory tied up both the device and
	35	* socket receive queues.
	36	*/
ec16227e AG	37	struct rds_page_frag {
ec16227e AG	38	struct list_head f_item;
33244125	39	struct list_head f_cache_entry;
0b088e00	40	struct scatterlist f_sg;
ec16227e AG	41	};
	42
	43	struct rds_ib_incoming {
	44	struct list_head ii_frags;
33244125	45	struct list_head ii_cache_entry;
ec16227e AG	46	struct rds_incoming ii_inc;
	47	};
	48
33244125 CM	49	struct rds_ib_cache_head {
	50	struct list_head *first;
	51	unsigned long count;
	52	};
	53
	54	struct rds_ib_refill_cache {
ae4b46e9	55	struct rds_ib_cache_head __percpu *percpu;
33244125 CM	56	struct list_head *xfer;
	57	struct list_head *ready;
	58	};
	59
ec16227e AG	60	struct rds_ib_connect_private {
	61	/* Add new fields at the end, and don't permute existing fields. */
	62	__be32 dp_saddr;
	63	__be32 dp_daddr;
	64	u8 dp_protocol_major;
	65	u8 dp_protocol_minor;
	66	__be16 dp_protocol_minor_mask; /* bitmask */
	67	__be32 dp_reserved1;
	68	__be64 dp_ack_seq;
	69	__be32 dp_credit; /* non-zero enables flow ctl */
	70	};
	71
	72	struct rds_ib_send_work {
ff3d7d36	73	void *s_op;
ec16227e AG	74	struct ib_send_wr s_wr;
	75	struct ib_sge s_sge[RDS_IB_MAX_SGE];
	76	unsigned long s_queued;
	77	};
	78
	79	struct rds_ib_recv_work {
	80	struct rds_ib_incoming *r_ibinc;
	81	struct rds_page_frag *r_frag;
	82	struct ib_recv_wr r_wr;
	83	struct ib_sge r_sge[2];
	84	};
	85
	86	struct rds_ib_work_ring {
	87	u32 w_nr;
	88	u32 w_alloc_ptr;
	89	u32 w_alloc_ctr;
	90	u32 w_free_ptr;
	91	atomic_t w_free_ctr;
	92	};
	93
f4f943c9 SS	94	/* Rings are posted with all the allocations they'll need to queue the
	95	* incoming message to the receiving socket so this can't fail.
	96	* All fragments start with a header, so we can make sure we're not receiving
	97	* garbage, and we can tell a small 8 byte fragment from an ACK frame.
	98	*/
	99	struct rds_ib_ack_state {
	100	u64 ack_next;
	101	u64 ack_recv;
	102	unsigned int ack_required:1;
	103	unsigned int ack_next_valid:1;
	104	unsigned int ack_recv_valid:1;
	105	};
	106
	107
ec16227e AG	108	struct rds_ib_device;
	109
	110	struct rds_ib_connection {
	111
	112	struct list_head ib_node;
	113	struct rds_ib_device *rds_ibdev;
	114	struct rds_connection *conn;
	115
	116	/* alphabet soup, IBTA style */
	117	struct rdma_cm_id *i_cm_id;
	118	struct ib_pd *i_pd;
ec16227e AG	119	struct ib_cq *i_send_cq;
ec16227e AG	120	struct ib_cq *i_recv_cq;
f4f943c9 SS	121	struct ib_wc i_recv_wc[RDS_IB_WC_MAX];
	122
	123	/* interrupt handling */
	124	struct tasklet_struct i_recv_tasklet;
ec16227e AG	125
	126	/* tx */
	127	struct rds_ib_work_ring i_send_ring;
ff3d7d36	128	struct rm_data_op *i_data_op;
ec16227e AG	129	struct rds_header *i_send_hdrs;
	130	u64 i_send_hdrs_dma;
	131	struct rds_ib_send_work *i_sends;
f046011c	132	atomic_t i_signaled_sends;
ec16227e AG	133
	134	/* rx */
	135	struct mutex i_recv_mutex;
	136	struct rds_ib_work_ring i_recv_ring;
	137	struct rds_ib_incoming *i_ibinc;
	138	u32 i_recv_data_rem;
	139	struct rds_header *i_recv_hdrs;
	140	u64 i_recv_hdrs_dma;
	141	struct rds_ib_recv_work *i_recvs;
ec16227e	142	u64 i_ack_recv; /* last ACK received */
33244125 CM	143	struct rds_ib_refill_cache i_cache_incs;
33244125 CM	144	struct rds_ib_refill_cache i_cache_frags;
ec16227e AG	145
	146	/* sending acks */
	147	unsigned long i_ack_flags;
8cbd9606 AG	148	#ifdef KERNEL_HAS_ATOMIC64
	149	atomic64_t i_ack_next; /* next ACK to send */
	150	#else
	151	spinlock_t i_ack_lock; /* protect i_ack_next */
ec16227e	152	u64 i_ack_next; /* next ACK to send */
8cbd9606	153	#endif
ec16227e AG	154	struct rds_header *i_ack;
	155	struct ib_send_wr i_ack_wr;
	156	struct ib_sge i_ack_sge;
	157	u64 i_ack_dma;
	158	unsigned long i_ack_queued;
	159
	160	/* Flow control related information
	161	*
	162	* Our algorithm uses a pair variables that we need to access
	163	* atomically - one for the send credits, and one posted
	164	* recv credits we need to transfer to remote.
	165	* Rather than protect them using a slow spinlock, we put both into
	166	* a single atomic_t and update it using cmpxchg
	167	*/
	168	atomic_t i_credits;
	169
	170	/* Protocol version specific information */
	171	unsigned int i_flowctl:1; /* enable/disable flow ctl */
	172
	173	/* Batched completions */
	174	unsigned int i_unsignaled_wrs;
ec16227e AG	175	};
	176
	177	/* This assumes that atomic_t is at least 32 bits */
	178	#define IB_GET_SEND_CREDITS(v) ((v) & 0xffff)
	179	#define IB_GET_POST_CREDITS(v) ((v) >> 16)
	180	#define IB_SET_SEND_CREDITS(v) ((v) & 0xffff)
	181	#define IB_SET_POST_CREDITS(v) ((v) << 16)
	182
	183	struct rds_ib_ipaddr {
	184	struct list_head list;
	185	__be32 ipaddr;
59fe4606	186	struct rcu_head rcu;
ec16227e AG	187	};
	188
	189	struct rds_ib_device {
	190	struct list_head list;
	191	struct list_head ipaddr_list;
	192	struct list_head conn_list;
	193	struct ib_device *dev;
	194	struct ib_pd *pd;
ec16227e	195	struct rds_ib_mr_pool *mr_pool;
ec16227e AG	196	unsigned int fmr_max_remaps;
	197	unsigned int max_fmrs;
	198	int max_sge;
	199	unsigned int max_wrs;
40589e74 AG	200	unsigned int max_initiator_depth;
40589e74 AG	201	unsigned int max_responder_resources;
ec16227e	202	spinlock_t spinlock; /* protect the above */
3e0249f9 ZB	203	atomic_t refcount;
3e0249f9 ZB	204	struct work_struct free_work;
ec16227e AG	205	};
ec16227e AG	206
a0c6ffbc	207	#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
e4c52c98 AG	208	#define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)
e4c52c98 AG	209
ec16227e AG	210	/* bits for i_ack_flags */
	211	#define IB_ACK_IN_FLIGHT 0
	212	#define IB_ACK_REQUESTED 1
	213
	214	/* Magic WR_ID for ACKs */
	215	#define RDS_IB_ACK_WR_ID (~(u64) 0)
	216
	217	struct rds_ib_statistics {
	218	uint64_t s_ib_connect_raced;
	219	uint64_t s_ib_listen_closed_stale;
	220	uint64_t s_ib_tx_cq_call;
f4f943c9 SS	221	uint64_t s_ib_evt_handler_call;
f4f943c9 SS	222	uint64_t s_ib_tasklet_call;
ec16227e AG	223	uint64_t s_ib_tx_cq_event;
	224	uint64_t s_ib_tx_ring_full;
	225	uint64_t s_ib_tx_throttle;
	226	uint64_t s_ib_tx_sg_mapping_failure;
	227	uint64_t s_ib_tx_stalled;
	228	uint64_t s_ib_tx_credit_updates;
ec16227e AG	229	uint64_t s_ib_rx_cq_event;
	230	uint64_t s_ib_rx_ring_empty;
	231	uint64_t s_ib_rx_refill_from_cq;
	232	uint64_t s_ib_rx_refill_from_thread;
	233	uint64_t s_ib_rx_alloc_limit;
	234	uint64_t s_ib_rx_credit_updates;
	235	uint64_t s_ib_ack_sent;
	236	uint64_t s_ib_ack_send_failure;
	237	uint64_t s_ib_ack_send_delayed;
	238	uint64_t s_ib_ack_send_piggybacked;
	239	uint64_t s_ib_ack_received;
	240	uint64_t s_ib_rdma_mr_alloc;
	241	uint64_t s_ib_rdma_mr_free;
	242	uint64_t s_ib_rdma_mr_used;
	243	uint64_t s_ib_rdma_mr_pool_flush;
	244	uint64_t s_ib_rdma_mr_pool_wait;
	245	uint64_t s_ib_rdma_mr_pool_depleted;
51e2cba8 AG	246	uint64_t s_ib_atomic_cswp;
51e2cba8 AG	247	uint64_t s_ib_atomic_fadd;
ec16227e AG	248	};
	249
	250	extern struct workqueue_struct *rds_ib_wq;
	251
	252	/*
	253	* Fake ib_dma_sync_sg_for_{cpu,device} as long as ib_verbs.h
	254	* doesn't define it.
	255	*/
	256	static inline void rds_ib_dma_sync_sg_for_cpu(struct ib_device *dev,
d2a9ec64 FF	257	struct scatterlist *sglist,
	258	unsigned int sg_dma_len,
	259	int direction)
ec16227e	260	{
d2a9ec64	261	struct scatterlist *sg;
ec16227e AG	262	unsigned int i;
ec16227e AG	263
d2a9ec64	264	for_each_sg(sglist, sg, sg_dma_len, i) {
ec16227e	265	ib_dma_sync_single_for_cpu(dev,
d2a9ec64 FF	266	ib_sg_dma_address(dev, sg),
d2a9ec64 FF	267	ib_sg_dma_len(dev, sg),
ec16227e AG	268	direction);
	269	}
	270	}
	271	#define ib_dma_sync_sg_for_cpu rds_ib_dma_sync_sg_for_cpu
	272
	273	static inline void rds_ib_dma_sync_sg_for_device(struct ib_device *dev,
d2a9ec64 FF	274	struct scatterlist *sglist,
	275	unsigned int sg_dma_len,
	276	int direction)
ec16227e	277	{
d2a9ec64	278	struct scatterlist *sg;
ec16227e AG	279	unsigned int i;
ec16227e AG	280
d2a9ec64	281	for_each_sg(sglist, sg, sg_dma_len, i) {
ec16227e	282	ib_dma_sync_single_for_device(dev,
d2a9ec64 FF	283	ib_sg_dma_address(dev, sg),
d2a9ec64 FF	284	ib_sg_dma_len(dev, sg),
ec16227e AG	285	direction);
	286	}
	287	}
	288	#define ib_dma_sync_sg_for_device rds_ib_dma_sync_sg_for_device
	289
	290
	291	/* ib.c */
	292	extern struct rds_transport rds_ib_transport;
3e0249f9 ZB	293	struct rds_ib_device rds_ib_get_client_data(struct ib_device device);
3e0249f9 ZB	294	void rds_ib_dev_put(struct rds_ib_device *rds_ibdev);
ec16227e AG	295	extern struct ib_client rds_ib_client;
ec16227e AG	296
ec16227e	297	extern unsigned int fmr_message_size;
3ba23ade	298	extern unsigned int rds_ib_retry_count;
ec16227e AG	299
	300	extern spinlock_t ib_nodev_conns_lock;
	301	extern struct list_head ib_nodev_conns;
	302
	303	/* ib_cm.c */
	304	int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp);
	305	void rds_ib_conn_free(void *arg);
	306	int rds_ib_conn_connect(struct rds_connection *conn);
	307	void rds_ib_conn_shutdown(struct rds_connection *conn);
	308	void rds_ib_state_change(struct sock *sk);
ef87b7ea	309	int rds_ib_listen_init(void);
ec16227e AG	310	void rds_ib_listen_stop(void);
	311	void __rds_ib_conn_error(struct rds_connection conn, const char , ...);
	312	int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
	313	struct rdma_cm_event *event);
	314	int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id);
	315	void rds_ib_cm_connect_complete(struct rds_connection *conn,
	316	struct rdma_cm_event *event);
	317
	318
	319	#define rds_ib_conn_error(conn, fmt...) \
	320	__rds_ib_conn_error(conn, KERN_WARNING "RDS/IB: " fmt)
	321
	322	/* ib_rdma.c */
	323	int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr);
745cbcca AG	324	void rds_ib_add_conn(struct rds_ib_device rds_ibdev, struct rds_connection conn);
745cbcca AG	325	void rds_ib_remove_conn(struct rds_ib_device rds_ibdev, struct rds_connection conn);
8aeb1ba6	326	void rds_ib_destroy_nodev_conns(void);
ec16227e AG	327	struct rds_ib_mr_pool rds_ib_create_mr_pool(struct rds_ib_device );
	328	void rds_ib_get_mr_info(struct rds_ib_device rds_ibdev, struct rds_info_rdma_connection iinfo);
	329	void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *);
	330	void rds_ib_get_mr(struct scatterlist sg, unsigned long nents,
	331	struct rds_sock rs, u32 key_ret);
	332	void rds_ib_sync_mr(void *trans_private, int dir);
	333	void rds_ib_free_mr(void *trans_private, int invalidate);
	334	void rds_ib_flush_mrs(void);
ad1d7dc0	335	int rds_ib_fmr_init(void);
ad1d7dc0	336	void rds_ib_fmr_exit(void);
ec16227e AG	337
ec16227e AG	338	/* ib_recv.c */
ef87b7ea	339	int rds_ib_recv_init(void);
ec16227e AG	340	void rds_ib_recv_exit(void);
ec16227e AG	341	int rds_ib_recv(struct rds_connection *conn);
33244125 CM	342	int rds_ib_recv_alloc_caches(struct rds_ib_connection *ic);
33244125 CM	343	void rds_ib_recv_free_caches(struct rds_ib_connection *ic);
73ce4317	344	void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp);
ec16227e	345	void rds_ib_inc_free(struct rds_incoming *inc);
c310e72c	346	int rds_ib_inc_copy_to_user(struct rds_incoming inc, struct iov_iter to);
f4f943c9 SS	347	void rds_ib_recv_cqe_handler(struct rds_ib_connection ic, struct ib_wc wc,
f4f943c9 SS	348	struct rds_ib_ack_state *state);
d521b63b	349	void rds_ib_recv_tasklet_fn(unsigned long data);
ec16227e AG	350	void rds_ib_recv_init_ring(struct rds_ib_connection *ic);
	351	void rds_ib_recv_clear_ring(struct rds_ib_connection *ic);
	352	void rds_ib_recv_init_ack(struct rds_ib_connection *ic);
	353	void rds_ib_attempt_ack(struct rds_ib_connection *ic);
	354	void rds_ib_ack_send_complete(struct rds_ib_connection *ic);
	355	u64 rds_ib_piggyb_ack(struct rds_ib_connection *ic);
f4f943c9	356	void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq, int ack_required);
ec16227e AG	357
	358	/* ib_ring.c */
	359	void rds_ib_ring_init(struct rds_ib_work_ring *ring, u32 nr);
	360	void rds_ib_ring_resize(struct rds_ib_work_ring *ring, u32 nr);
	361	u32 rds_ib_ring_alloc(struct rds_ib_work_ring ring, u32 val, u32 pos);
	362	void rds_ib_ring_free(struct rds_ib_work_ring *ring, u32 val);
	363	void rds_ib_ring_unalloc(struct rds_ib_work_ring *ring, u32 val);
	364	int rds_ib_ring_empty(struct rds_ib_work_ring *ring);
	365	int rds_ib_ring_low(struct rds_ib_work_ring *ring);
	366	u32 rds_ib_ring_oldest(struct rds_ib_work_ring *ring);
	367	u32 rds_ib_ring_completed(struct rds_ib_work_ring *ring, u32 wr_id, u32 oldest);
	368	extern wait_queue_head_t rds_ib_ring_empty_wait;
	369
	370	/* ib_send.c */
	371	void rds_ib_xmit_complete(struct rds_connection *conn);
	372	int rds_ib_xmit(struct rds_connection conn, struct rds_message rm,
	373	unsigned int hdr_off, unsigned int sg, unsigned int off);
	374	void rds_ib_send_cq_comp_handler(struct ib_cq cq, void context);
	375	void rds_ib_send_init_ring(struct rds_ib_connection *ic);
	376	void rds_ib_send_clear_ring(struct rds_ib_connection *ic);
f8b3aaf2	377	int rds_ib_xmit_rdma(struct rds_connection conn, struct rm_rdma_op op);
ec16227e AG	378	void rds_ib_send_add_credits(struct rds_connection *conn, unsigned int credits);
	379	void rds_ib_advertise_credits(struct rds_connection *conn, unsigned int posted);
	380	int rds_ib_send_grab_credits(struct rds_ib_connection *ic, u32 wanted,
7b70d033	381	u32 *adv_credits, int need_posted, int max_posted);
ff3d7d36	382	int rds_ib_xmit_atomic(struct rds_connection conn, struct rm_atomic_op op);
ec16227e AG	383
	384	/* ib_stats.c */
	385	DECLARE_PER_CPU(struct rds_ib_statistics, rds_ib_stats);
	386	#define rds_ib_stats_inc(member) rds_stats_inc_which(rds_ib_stats, member)
	387	unsigned int rds_ib_stats_info_copy(struct rds_info_iterator *iter,
	388	unsigned int avail);
	389
	390	/* ib_sysctl.c */
ef87b7ea	391	int rds_ib_sysctl_init(void);
ec16227e AG	392	void rds_ib_sysctl_exit(void);
	393	extern unsigned long rds_ib_sysctl_max_send_wr;
	394	extern unsigned long rds_ib_sysctl_max_recv_wr;
	395	extern unsigned long rds_ib_sysctl_max_unsig_wrs;
	396	extern unsigned long rds_ib_sysctl_max_unsig_bytes;
	397	extern unsigned long rds_ib_sysctl_max_recv_allocation;
	398	extern unsigned int rds_ib_sysctl_flow_control;
ec16227e	399
ec16227e	400	#endif