Add queue pair data structure as well as supporting structures to rdmavt.
Reviewed-by: Ira Weiny <ira.we...@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marcinis...@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessan...@intel.com>
---
Changes since v1:
Rename rvt_sge.m => rvt_sge.cur_map
Rename rvt_sge.n => rvt_sge.cur
drivers/infiniband/sw/rdmavt/qp.h | 5 -
include/rdma/rdma_vt.h | 233 +++++++++++++++++++++++++++++++++++++
2 files changed, 233 insertions(+), 5 deletions(-)
diff --git a/drivers/infiniband/sw/rdmavt/qp.h
b/drivers/infiniband/sw/rdmavt/qp.h
index 10bc636..9c2999d 100644
--- a/drivers/infiniband/sw/rdmavt/qp.h
+++ b/drivers/infiniband/sw/rdmavt/qp.h
@@ -50,11 +50,6 @@
#include <rdma/rdma_vt.h>
-struct rvt_qp {
- struct ib_qp *ibqp;
- /* Other stuff */
-};
-
struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata);
diff --git a/include/rdma/rdma_vt.h b/include/rdma/rdma_vt.h
index f232e39..9baa7f0 100644
--- a/include/rdma/rdma_vt.h
+++ b/include/rdma/rdma_vt.h
@@ -106,6 +106,239 @@ struct rvt_lkey_table {
/* End Memmory Region */
/*
+ * Things needed for the Queue Pair definition. Like the MR stuff above the
+ * following should probably get moved to qp.h once drivers stop trying to make
+ * and manipulate thier own QPs. For the few instnaces where a driver may need
+ * to look into a queue pair there should be a pointer to a driver priavte data
+ * structure that they can look at.
+ */
+
+/*
+ * These keep track of the copy progress within a memory region.
+ * Used by the verbs layer.
+ */
+struct rvt_sge {
+ struct rvt_mregion *mr;
+ void *vaddr; /* kernel virtual address of segment */
+ u32 sge_length; /* length of the SGE */
+ u32 length; /* remaining length of the segment */
+ u16 m; /* current index: mr->map[m] */
+ u16 n; /* current index: mr->map[m]->segs[n] */
+};
+
+/*
+ * Send work request queue entry.
+ * The size of the sg_list is determined when the QP is created and stored
+ * in qp->s_max_sge.
+ */
+struct rvt_swqe {
+ union {
+ struct ib_send_wr wr; /* don't use wr.sg_list */
+ struct ib_ud_wr ud_wr;
+ struct ib_reg_wr reg_wr;
+ struct ib_rdma_wr rdma_wr;
+ struct ib_atomic_wr atomic_wr;
+ };
+ u32 psn; /* first packet sequence number */
+ u32 lpsn; /* last packet sequence number */
+ u32 ssn; /* send sequence number */
+ u32 length; /* total length of data in sg_list */
+ struct rvt_sge sg_list[0];
+};
+
+/*
+ * Receive work request queue entry.
+ * The size of the sg_list is determined when the QP (or SRQ) is created
+ * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
+ */
+struct rvt_rwqe {
+ u64 wr_id;
+ u8 num_sge;
+ struct ib_sge sg_list[0];
+};
+
+/*
+ * This structure is used to contain the head pointer, tail pointer,
+ * and receive work queue entries as a single memory allocation so
+ * it can be mmap'ed into user space.
+ * Note that the wq array elements are variable size so you can't
+ * just index into the array to get the N'th element;
+ * use get_rwqe_ptr() instead.
+ */
+struct rvt_rwq {
+ u32 head; /* new work requests posted to the head */
+ u32 tail; /* receives pull requests from here. */
+ struct rvt_rwqe wq[0];
+};
+
+struct rvt_rq {
+ struct rvt_rwq *wq;
+ u32 size; /* size of RWQE array */
+ u8 max_sge;
+ /* protect changes in this struct */
+ spinlock_t lock ____cacheline_aligned_in_smp;
+};
+
+/*
+ * This structure is used by rvt_mmap() to validate an offset
+ * when an mmap() request is made. The vm_area_struct then uses
+ * this as its vm_private_data.
+ */
+struct rvt_mmap_info {
+ struct list_head pending_mmaps;
+ struct ib_ucontext *context;
+ void *obj;
+ __u64 offset;
+ struct kref ref;
+ unsigned size;
+};
+
+#define RVT_MAX_RDMA_ATOMIC 16
+
+/*
+ * This structure holds the information that the send tasklet needs
+ * to send a RDMA read response or atomic operation.
+ */
+struct rvt_ack_entry {
+ u8 opcode;
+ u8 sent;
+ u32 psn;
+ u32 lpsn;
+ union {
+ struct rvt_sge rdma_sge;
+ u64 atomic_data;
+ };
+};
+
+struct rvt_sge_state {
+ struct rvt_sge *sg_list; /* next SGE to be used if any */
+ struct rvt_sge sge; /* progress state for the current SGE */
+ u32 total_len;
+ u8 num_sge;
+};
+
+/*
+ * Variables prefixed with s_ are for the requester (sender).
+ * Variables prefixed with r_ are for the responder (receiver).
+ * Variables prefixed with ack_ are for responder replies.
+ *
+ * Common variables are protected by both r_rq.lock and s_lock in that order
+ * which only happens in modify_qp() or changing the QP 'state'.
+ */
+struct rvt_qp {
+ struct ib_qp ibqp;
+ void *priv; /* Driver private data */
+ /* read mostly fields above and below */
+ struct ib_ah_attr remote_ah_attr;
+ struct ib_ah_attr alt_ah_attr;
+ struct rvt_qp __rcu *next; /* link list for QPN hash table */
+ struct rvt_swqe *s_wq; /* send work queue */
+ struct rvt_mmap_info *ip;
+
+ unsigned long timeout_jiffies; /* computed from timeout */
+
+ enum ib_mtu path_mtu;
+ int srate_mbps; /* s_srate (below) converted to Mbit/s */
+ u32 remote_qpn;
+ u32 pmtu; /* decoded from path_mtu */
+ u32 qkey; /* QKEY for this QP (for UD or RD) */
+ u32 s_size; /* send work queue size */
+ u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */
+ u32 s_ahgpsn; /* set to the psn in the copy of the header */
+
+ u8 state; /* QP state */
+ u8 allowed_ops; /* high order bits of allowed opcodes */
+ u8 qp_access_flags;
+ u8 alt_timeout; /* Alternate path timeout for this QP */
+ u8 timeout; /* Timeout for this QP */
+ u8 s_srate;
+ u8 s_mig_state;
+ u8 port_num;
+ u8 s_pkey_index; /* PKEY index to use */
+ u8 s_alt_pkey_index; /* Alternate path PKEY index to use */
+ u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */
+ u8 s_max_rd_atomic; /* max number of RDMA read/atomic to send */
+ u8 s_retry_cnt; /* number of times to retry */
+ u8 s_rnr_retry_cnt;
+ u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
+ u8 s_max_sge; /* size of s_wq->sg_list */
+ u8 s_draining;
+
+ /* start of read/write fields */
+ atomic_t refcount ____cacheline_aligned_in_smp;
+ wait_queue_head_t wait;
+
+ struct rvt_ack_entry s_ack_queue[RVT_MAX_RDMA_ATOMIC + 1]
+ ____cacheline_aligned_in_smp;
+ struct rvt_sge_state s_rdma_read_sge;
+
+ spinlock_t r_lock ____cacheline_aligned_in_smp; /* used for APM */
+ unsigned long r_aflags;
+ u64 r_wr_id; /* ID for current receive WQE */
+ u32 r_ack_psn; /* PSN for next ACK or atomic ACK */
+ u32 r_len; /* total length of r_sge */
+ u32 r_rcv_len; /* receive data len processed */
+ u32 r_psn; /* expected rcv packet sequence number */
+ u32 r_msn; /* message sequence number */
+
+ u8 r_state; /* opcode of last packet received */
+ u8 r_flags;
+ u8 r_head_ack_queue; /* index into s_ack_queue[] */
+
+ struct list_head rspwait; /* link for waiting to respond */
+
+ struct rvt_sge_state r_sge; /* current receive data */
+ struct rvt_rq r_rq; /* receive work queue */
+
+ spinlock_t s_lock ____cacheline_aligned_in_smp;
+ struct rvt_sge_state *s_cur_sge;
+ u32 s_flags;
+ struct rvt_swqe *s_wqe;
+ struct rvt_sge_state s_sge; /* current send request data */
+ struct rvt_mregion *s_rdma_mr;
+ struct sdma_engine *s_sde; /* current sde */
+ u32 s_cur_size; /* size of send packet in bytes */
+ u32 s_len; /* total length of s_sge */
+ u32 s_rdma_read_len; /* total length of s_rdma_read_sge */
+ u32 s_next_psn; /* PSN for next request */
+ u32 s_last_psn; /* last response PSN processed */
+ u32 s_sending_psn; /* lowest PSN that is being sent */
+ u32 s_sending_hpsn; /* highest PSN that is being sent */
+ u32 s_psn; /* current packet sequence number */
+ u32 s_ack_rdma_psn; /* PSN for sending RDMA read responses */
+ u32 s_ack_psn; /* PSN for acking sends and RDMA writes */
+ u32 s_head; /* new entries added here */
+ u32 s_tail; /* next entry to process */
+ u32 s_cur; /* current work queue entry */
+ u32 s_acked; /* last un-ACK'ed entry */
+ u32 s_last; /* last completed entry */
+ u32 s_ssn; /* SSN of tail entry */
+ u32 s_lsn; /* limit sequence number (credit) */
+ u16 s_hdrwords; /* size of s_hdr in 32 bit words */
+ u16 s_rdma_ack_cnt;
+ s8 s_ahgidx;
+ u8 s_state; /* opcode of last packet sent */
+ u8 s_ack_state; /* opcode of packet to ACK */
+ u8 s_nak_state; /* non-zero if NAK is pending */
+ u8 r_nak_state; /* non-zero if NAK is pending */
+ u8 s_retry; /* requester retry counter */
+ u8 s_rnr_retry; /* requester RNR retry counter */
+ u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */
+ u8 s_tail_ack_queue; /* index into s_ack_queue[] */
+
+ struct rvt_sge_state s_ack_rdma_sge;
+ struct timer_list s_timer;
+
+ /*
+ * This sge list MUST be last. Do not add anything below here.
+ */
+ struct rvt_sge r_sg_list[0] /* verified SGEs */
+ ____cacheline_aligned_in_smp;
+};
+
+/* End QP section */
+
+/*
* Things that are driver specific, module parameters in hfi1 and qib
*/
struct rvt_driver_params {
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
