2 * Copyright(c) 2015, 2016 Intel Corporation.
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
51 #include <linux/types.h>
52 #include <linux/seqlock.h>
53 #include <linux/kernel.h>
54 #include <linux/interrupt.h>
55 #include <linux/kref.h>
56 #include <linux/workqueue.h>
57 #include <linux/kthread.h>
58 #include <linux/completion.h>
59 #include <linux/slab.h>
60 #include <rdma/ib_pack.h>
61 #include <rdma/ib_user_verbs.h>
62 #include <rdma/ib_mad.h>
63 #include <rdma/rdma_vt.h>
64 #include <rdma/rdmavt_qp.h>
65 #include <rdma/rdmavt_cq.h>
68 struct hfi1_pportdata;
74 #define HFI1_MAX_RDMA_ATOMIC 16
75 #define HFI1_GUIDS_PER_PORT 5
78 * Increment this value if any changes that break userspace ABI
79 * compatibility are made.
81 #define HFI1_UVERBS_ABI_VERSION 2
83 #define IB_SEQ_NAK (3 << 29)
85 /* AETH NAK opcode values */
86 #define IB_RNR_NAK 0x20
87 #define IB_NAK_PSN_ERROR 0x60
88 #define IB_NAK_INVALID_REQUEST 0x61
89 #define IB_NAK_REMOTE_ACCESS_ERROR 0x62
90 #define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63
91 #define IB_NAK_INVALID_RD_REQUEST 0x64
93 /* IB Performance Manager status values */
94 #define IB_PMA_SAMPLE_STATUS_DONE 0x00
95 #define IB_PMA_SAMPLE_STATUS_STARTED 0x01
96 #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
98 /* Mandatory IB performance counter select values. */
99 #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
100 #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
101 #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
102 #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
103 #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
105 #define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0)
107 #define IB_BTH_REQ_ACK BIT(31)
108 #define IB_BTH_SOLICITED BIT(23)
109 #define IB_BTH_MIG_REQ BIT(22)
111 #define IB_GRH_VERSION 6
112 #define IB_GRH_VERSION_MASK 0xF
113 #define IB_GRH_VERSION_SHIFT 28
114 #define IB_GRH_TCLASS_MASK 0xFF
115 #define IB_GRH_TCLASS_SHIFT 20
116 #define IB_GRH_FLOW_MASK 0xFFFFF
117 #define IB_GRH_FLOW_SHIFT 0
118 #define IB_GRH_NEXT_HDR 0x1B
120 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
122 /* flags passed by hfi1_ib_rcv() */
124 HFI1_HAS_GRH = (1 << 0),
133 struct ib_atomic_eth {
134 __be32 vaddr[2]; /* unaligned so access as 2 32-bit words */
151 __be32 atomic_ack_eth[2];
156 struct ib_atomic_eth atomic_eth;
159 struct hfi1_other_headers {
165 * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
166 * long (72 w/ imm_data). Only the first 56 bytes of the IB header
167 * will be in the eager header buffer. The remaining 12 or 16 bytes
168 * are in the data buffer.
170 struct hfi1_ib_header {
175 struct hfi1_other_headers oth;
177 struct hfi1_other_headers oth;
181 struct ahg_ib_header {
182 struct sdma_engine *sde;
187 struct hfi1_ib_header ibh;
190 struct hfi1_pio_header {
192 struct hfi1_ib_header hdr;
196 * hfi1 specific data structures that will be hidden from rvt after the queue
197 * pair is made common
199 struct hfi1_qp_priv {
200 struct ahg_ib_header *s_hdr; /* next header to send */
201 struct sdma_engine *s_sde; /* current sde */
202 struct send_context *s_sendcontext; /* current sendcontext */
203 u8 s_sc; /* SC[0..4] for next packet */
204 u8 r_adefered; /* number of acks defered */
205 struct iowait s_iowait;
206 struct timer_list s_rnr_timer;
207 struct rvt_qp *owner;
211 * This structure is used to hold commonly lookedup and computed values during
212 * the send engine progress.
214 struct hfi1_pkt_state {
215 struct hfi1_ibdev *dev;
216 struct hfi1_ibport *ibp;
217 struct hfi1_pportdata *ppd;
218 struct verbs_txreq *s_txreq;
222 #define HFI1_PSN_CREDIT 16
224 struct hfi1_opcode_stats {
225 u64 n_packets; /* number of packets */
226 u64 n_bytes; /* total number of bytes */
229 struct hfi1_opcode_stats_perctx {
230 struct hfi1_opcode_stats stats[256];
233 static inline void inc_opstats(
235 struct hfi1_opcode_stats *stats)
237 #ifdef CONFIG_DEBUG_FS
238 stats->n_bytes += tlen;
244 struct rvt_qp __rcu *qp[2];
245 struct rvt_ibport rvp;
247 __be64 guids[HFI1_GUIDS_PER_PORT - 1]; /* writable GUIDs */
249 /* the first 16 entries are sl_to_vl for !OPA */
255 struct rvt_dev_info rdi; /* Must be first */
257 /* QP numbers are shared by all IB ports */
258 /* protect wait lists */
259 seqlock_t iowait_lock;
260 struct list_head txwait; /* list for wait verbs_txreq */
261 struct list_head memwait; /* list for wait kernel memory */
262 struct list_head txreq_free;
263 struct kmem_cache *verbs_txreq_cache;
264 struct timer_list mem_timer;
271 #ifdef CONFIG_DEBUG_FS
272 /* per HFI debugfs */
273 struct dentry *hfi1_ibdev_dbg;
274 /* per HFI symlinks to above */
275 struct dentry *hfi1_ibdev_link;
279 static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
281 struct rvt_dev_info *rdi;
283 rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
284 return container_of(rdi, struct hfi1_ibdev, rdi);
287 static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
289 struct hfi1_qp_priv *priv;
291 priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
296 * Send if not busy or waiting for I/O and either
297 * a RC response is pending or we can process send work requests.
299 static inline int hfi1_send_ok(struct rvt_qp *qp)
301 return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
302 (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
303 !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
307 * This must be called with s_lock held.
309 void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
310 u32 qp1, u32 qp2, u16 lid1, u16 lid2);
311 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num);
312 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
313 void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
314 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
315 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
316 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
317 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
318 u16 *out_mad_pkey_index);
321 * The PSN_MASK and PSN_SHIFT allow for
322 * 1) comparing two PSNs
323 * 2) returning the PSN with any upper bits masked
324 * 3) returning the difference between to PSNs
326 * The number of significant bits in the PSN must
327 * necessarily be at least one bit less than
328 * the container holding the PSN.
330 #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
331 #define PSN_MASK 0xFFFFFF
334 #define PSN_MASK 0x7FFFFFFF
337 #define PSN_MODIFY_MASK 0xFFFFFF
340 * Compare the lower 24 bits of the msn values.
341 * Returns an integer <, ==, or > than zero.
343 static inline int cmp_msn(u32 a, u32 b)
345 return (((int)a) - ((int)b)) << 8;
350 * Returns an integer <, ==, or > than zero.
352 static inline int cmp_psn(u32 a, u32 b)
354 return (((int)a) - ((int)b)) << PSN_SHIFT;
360 static inline u32 mask_psn(u32 a)
366 * Return delta between two PSNs
368 static inline u32 delta_psn(u32 a, u32 b)
370 return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
374 void hfi1_put_txreq(struct verbs_txreq *tx);
376 int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
378 void hfi1_copy_sge(struct rvt_sge_state *ss, void *data, u32 length,
379 int release, int copy_last);
381 void hfi1_skip_sge(struct rvt_sge_state *ss, u32 length, int release);
383 void hfi1_cnp_rcv(struct hfi1_packet *packet);
385 void hfi1_uc_rcv(struct hfi1_packet *packet);
387 void hfi1_rc_rcv(struct hfi1_packet *packet);
390 struct hfi1_ctxtdata *rcd,
391 struct hfi1_ib_header *hdr,
395 u8 ah_to_sc(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
397 struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u16 dlid);
399 void hfi1_rc_rnr_retry(unsigned long arg);
400 void hfi1_add_rnr_timer(struct rvt_qp *qp, u32 to);
401 void hfi1_rc_timeout(unsigned long arg);
402 void hfi1_del_timers_sync(struct rvt_qp *qp);
403 void hfi1_stop_rc_timers(struct rvt_qp *qp);
405 void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_ib_header *hdr);
407 void hfi1_rc_error(struct rvt_qp *qp, enum ib_wc_status err);
409 void hfi1_ud_rcv(struct hfi1_packet *packet);
411 int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
413 int hfi1_rvt_get_rwqe(struct rvt_qp *qp, int wr_id_only);
415 void hfi1_migrate_qp(struct rvt_qp *qp);
417 int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
418 int attr_mask, struct ib_udata *udata);
420 void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
421 int attr_mask, struct ib_udata *udata);
423 int hfi1_check_send_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe);
425 extern const u32 rc_only_opcode;
426 extern const u32 uc_only_opcode;
428 static inline u8 get_opcode(struct hfi1_ib_header *h)
430 u16 lnh = be16_to_cpu(h->lrh[0]) & 3;
432 if (lnh == IB_LNH_IBA_LOCAL)
433 return be32_to_cpu(h->u.oth.bth[0]) >> 24;
435 return be32_to_cpu(h->u.l.oth.bth[0]) >> 24;
438 int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct hfi1_ib_header *hdr,
439 int has_grh, struct rvt_qp *qp, u32 bth0);
441 u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
442 struct ib_global_route *grh, u32 hwords, u32 nwords);
444 void hfi1_make_ruc_header(struct rvt_qp *qp, struct hfi1_other_headers *ohdr,
445 u32 bth0, u32 bth2, int middle,
446 struct hfi1_pkt_state *ps);
448 void _hfi1_do_send(struct work_struct *work);
450 void hfi1_do_send(struct rvt_qp *qp);
452 void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
453 enum ib_wc_status status);
455 void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn);
457 int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
459 int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
461 int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
463 int hfi1_register_ib_device(struct hfi1_devdata *);
465 void hfi1_unregister_ib_device(struct hfi1_devdata *);
467 void hfi1_ib_rcv(struct hfi1_packet *packet);
469 unsigned hfi1_get_npkeys(struct hfi1_devdata *);
471 int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
474 int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
477 int hfi1_wss_init(void);
478 void hfi1_wss_exit(void);
480 /* platform specific: return the lowest level cache (llc) size, in KiB */
481 static inline int wss_llc_size(void)
483 /* assume that the boot CPU value is universal for all CPUs */
484 return boot_cpu_data.x86_cache_size;
487 /* platform specific: cacheless copy */
488 static inline void cacheless_memcpy(void *dst, void *src, size_t n)
491 * Use the only available X64 cacheless copy. Add a __user cast
492 * to quiet sparse. The src agument is already in the kernel so
493 * there are no security issues. The extra fault recovery machinery
496 __copy_user_nocache(dst, (void __user *)src, n, 0);
499 extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
501 extern const u8 hdr_len_by_opcode[];
503 extern const int ib_rvt_state_ops[];
505 extern __be64 ib_hfi1_sys_image_guid; /* in network order */
507 extern unsigned int hfi1_max_cqes;
509 extern unsigned int hfi1_max_cqs;
511 extern unsigned int hfi1_max_qp_wrs;
513 extern unsigned int hfi1_max_qps;
515 extern unsigned int hfi1_max_sges;
517 extern unsigned int hfi1_max_mcast_grps;
519 extern unsigned int hfi1_max_mcast_qp_attached;
521 extern unsigned int hfi1_max_srqs;
523 extern unsigned int hfi1_max_srq_sges;
525 extern unsigned int hfi1_max_srq_wrs;
527 extern unsigned short piothreshold;
529 extern const u32 ib_hfi1_rnr_table[];
531 #endif /* HFI1_VERBS_H */