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];
155 struct ib_atomic_eth atomic_eth
;
158 struct hfi1_other_headers
{
164 * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
165 * long (72 w/ imm_data). Only the first 56 bytes of the IB header
166 * will be in the eager header buffer. The remaining 12 or 16 bytes
167 * are in the data buffer.
169 struct hfi1_ib_header
{
174 struct hfi1_other_headers oth
;
176 struct hfi1_other_headers oth
;
180 struct ahg_ib_header
{
181 struct sdma_engine
*sde
;
186 struct hfi1_ib_header ibh
;
189 struct hfi1_pio_header
{
191 struct hfi1_ib_header hdr
;
195 * hfi1 specific data structures that will be hidden from rvt after the queue
196 * pair is made common
198 struct hfi1_qp_priv
{
199 struct ahg_ib_header
*s_hdr
; /* next header to send */
200 struct sdma_engine
*s_sde
; /* current sde */
201 struct send_context
*s_sendcontext
; /* current sendcontext */
202 u8 s_sc
; /* SC[0..4] for next packet */
203 u8 r_adefered
; /* number of acks defered */
204 struct iowait s_iowait
;
205 struct timer_list s_rnr_timer
;
206 struct rvt_qp
*owner
;
210 * This structure is used to hold commonly lookedup and computed values during
211 * the send engine progress.
213 struct hfi1_pkt_state
{
214 struct hfi1_ibdev
*dev
;
215 struct hfi1_ibport
*ibp
;
216 struct hfi1_pportdata
*ppd
;
217 struct verbs_txreq
*s_txreq
;
221 #define HFI1_PSN_CREDIT 16
223 struct hfi1_opcode_stats
{
224 u64 n_packets
; /* number of packets */
225 u64 n_bytes
; /* total number of bytes */
228 struct hfi1_opcode_stats_perctx
{
229 struct hfi1_opcode_stats stats
[256];
232 static inline void inc_opstats(
234 struct hfi1_opcode_stats
*stats
)
236 #ifdef CONFIG_DEBUG_FS
237 stats
->n_bytes
+= tlen
;
243 struct rvt_qp __rcu
*qp
[2];
244 struct rvt_ibport rvp
;
246 __be64 guids
[HFI1_GUIDS_PER_PORT
- 1]; /* writable GUIDs */
248 /* the first 16 entries are sl_to_vl for !OPA */
254 struct rvt_dev_info rdi
; /* Must be first */
256 /* QP numbers are shared by all IB ports */
257 /* protect wait lists */
258 seqlock_t iowait_lock
;
259 struct list_head txwait
; /* list for wait verbs_txreq */
260 struct list_head memwait
; /* list for wait kernel memory */
261 struct list_head txreq_free
;
262 struct kmem_cache
*verbs_txreq_cache
;
263 struct timer_list mem_timer
;
270 #ifdef CONFIG_DEBUG_FS
271 /* per HFI debugfs */
272 struct dentry
*hfi1_ibdev_dbg
;
273 /* per HFI symlinks to above */
274 struct dentry
*hfi1_ibdev_link
;
278 static inline struct hfi1_ibdev
*to_idev(struct ib_device
*ibdev
)
280 struct rvt_dev_info
*rdi
;
282 rdi
= container_of(ibdev
, struct rvt_dev_info
, ibdev
);
283 return container_of(rdi
, struct hfi1_ibdev
, rdi
);
286 static inline struct rvt_qp
*iowait_to_qp(struct iowait
*s_iowait
)
288 struct hfi1_qp_priv
*priv
;
290 priv
= container_of(s_iowait
, struct hfi1_qp_priv
, s_iowait
);
295 * Send if not busy or waiting for I/O and either
296 * a RC response is pending or we can process send work requests.
298 static inline int hfi1_send_ok(struct rvt_qp
*qp
)
300 return !(qp
->s_flags
& (RVT_S_BUSY
| RVT_S_ANY_WAIT_IO
)) &&
301 (qp
->s_hdrwords
|| (qp
->s_flags
& RVT_S_RESP_PENDING
) ||
302 !(qp
->s_flags
& RVT_S_ANY_WAIT_SEND
));
306 * This must be called with s_lock held.
308 void hfi1_bad_pqkey(struct hfi1_ibport
*ibp
, __be16 trap_num
, u32 key
, u32 sl
,
309 u32 qp1
, u32 qp2
, u16 lid1
, u16 lid2
);
310 void hfi1_cap_mask_chg(struct rvt_dev_info
*rdi
, u8 port_num
);
311 void hfi1_sys_guid_chg(struct hfi1_ibport
*ibp
);
312 void hfi1_node_desc_chg(struct hfi1_ibport
*ibp
);
313 int hfi1_process_mad(struct ib_device
*ibdev
, int mad_flags
, u8 port
,
314 const struct ib_wc
*in_wc
, const struct ib_grh
*in_grh
,
315 const struct ib_mad_hdr
*in_mad
, size_t in_mad_size
,
316 struct ib_mad_hdr
*out_mad
, size_t *out_mad_size
,
317 u16
*out_mad_pkey_index
);
320 * The PSN_MASK and PSN_SHIFT allow for
321 * 1) comparing two PSNs
322 * 2) returning the PSN with any upper bits masked
323 * 3) returning the difference between to PSNs
325 * The number of significant bits in the PSN must
326 * necessarily be at least one bit less than
327 * the container holding the PSN.
329 #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
330 #define PSN_MASK 0xFFFFFF
333 #define PSN_MASK 0x7FFFFFFF
336 #define PSN_MODIFY_MASK 0xFFFFFF
339 * Compare the lower 24 bits of the msn values.
340 * Returns an integer <, ==, or > than zero.
342 static inline int cmp_msn(u32 a
, u32 b
)
344 return (((int)a
) - ((int)b
)) << 8;
349 * Returns an integer <, ==, or > than zero.
351 static inline int cmp_psn(u32 a
, u32 b
)
353 return (((int)a
) - ((int)b
)) << PSN_SHIFT
;
359 static inline u32
mask_psn(u32 a
)
365 * Return delta between two PSNs
367 static inline u32
delta_psn(u32 a
, u32 b
)
369 return (((int)a
- (int)b
) << PSN_SHIFT
) >> PSN_SHIFT
;
373 void hfi1_put_txreq(struct verbs_txreq
*tx
);
375 int hfi1_verbs_send(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
377 void hfi1_copy_sge(struct rvt_sge_state
*ss
, void *data
, u32 length
,
378 int release
, int copy_last
);
380 void hfi1_skip_sge(struct rvt_sge_state
*ss
, u32 length
, int release
);
382 void hfi1_cnp_rcv(struct hfi1_packet
*packet
);
384 void hfi1_uc_rcv(struct hfi1_packet
*packet
);
386 void hfi1_rc_rcv(struct hfi1_packet
*packet
);
389 struct hfi1_ctxtdata
*rcd
,
390 struct hfi1_ib_header
*hdr
,
394 u8
ah_to_sc(struct ib_device
*ibdev
, struct ib_ah_attr
*ah_attr
);
396 struct ib_ah
*hfi1_create_qp0_ah(struct hfi1_ibport
*ibp
, u16 dlid
);
398 void hfi1_rc_rnr_retry(unsigned long arg
);
399 void hfi1_add_rnr_timer(struct rvt_qp
*qp
, u32 to
);
400 void hfi1_rc_timeout(unsigned long arg
);
401 void hfi1_del_timers_sync(struct rvt_qp
*qp
);
402 void hfi1_stop_rc_timers(struct rvt_qp
*qp
);
404 void hfi1_rc_send_complete(struct rvt_qp
*qp
, struct hfi1_ib_header
*hdr
);
406 void hfi1_rc_error(struct rvt_qp
*qp
, enum ib_wc_status err
);
408 void hfi1_ud_rcv(struct hfi1_packet
*packet
);
410 int hfi1_lookup_pkey_idx(struct hfi1_ibport
*ibp
, u16 pkey
);
412 int hfi1_rvt_get_rwqe(struct rvt_qp
*qp
, int wr_id_only
);
414 void hfi1_migrate_qp(struct rvt_qp
*qp
);
416 int hfi1_check_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
417 int attr_mask
, struct ib_udata
*udata
);
419 void hfi1_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
420 int attr_mask
, struct ib_udata
*udata
);
422 int hfi1_check_send_wqe(struct rvt_qp
*qp
, struct rvt_swqe
*wqe
);
424 extern const u32 rc_only_opcode
;
425 extern const u32 uc_only_opcode
;
427 static inline u8
get_opcode(struct hfi1_ib_header
*h
)
429 u16 lnh
= be16_to_cpu(h
->lrh
[0]) & 3;
431 if (lnh
== IB_LNH_IBA_LOCAL
)
432 return be32_to_cpu(h
->u
.oth
.bth
[0]) >> 24;
434 return be32_to_cpu(h
->u
.l
.oth
.bth
[0]) >> 24;
437 int hfi1_ruc_check_hdr(struct hfi1_ibport
*ibp
, struct hfi1_ib_header
*hdr
,
438 int has_grh
, struct rvt_qp
*qp
, u32 bth0
);
440 u32
hfi1_make_grh(struct hfi1_ibport
*ibp
, struct ib_grh
*hdr
,
441 struct ib_global_route
*grh
, u32 hwords
, u32 nwords
);
443 void hfi1_make_ruc_header(struct rvt_qp
*qp
, struct hfi1_other_headers
*ohdr
,
444 u32 bth0
, u32 bth2
, int middle
,
445 struct hfi1_pkt_state
*ps
);
447 void _hfi1_do_send(struct work_struct
*work
);
449 void hfi1_do_send(struct rvt_qp
*qp
);
451 void hfi1_send_complete(struct rvt_qp
*qp
, struct rvt_swqe
*wqe
,
452 enum ib_wc_status status
);
454 void hfi1_send_rc_ack(struct hfi1_ctxtdata
*, struct rvt_qp
*qp
, int is_fecn
);
456 int hfi1_make_rc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
458 int hfi1_make_uc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
460 int hfi1_make_ud_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
462 int hfi1_register_ib_device(struct hfi1_devdata
*);
464 void hfi1_unregister_ib_device(struct hfi1_devdata
*);
466 void hfi1_ib_rcv(struct hfi1_packet
*packet
);
468 unsigned hfi1_get_npkeys(struct hfi1_devdata
*);
470 int hfi1_verbs_send_dma(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
473 int hfi1_verbs_send_pio(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
476 int hfi1_wss_init(void);
477 void hfi1_wss_exit(void);
479 /* platform specific: return the lowest level cache (llc) size, in KiB */
480 static inline int wss_llc_size(void)
482 /* assume that the boot CPU value is universal for all CPUs */
483 return boot_cpu_data
.x86_cache_size
;
486 /* platform specific: cacheless copy */
487 static inline void cacheless_memcpy(void *dst
, void *src
, size_t n
)
490 * Use the only available X64 cacheless copy. Add a __user cast
491 * to quiet sparse. The src agument is already in the kernel so
492 * there are no security issues. The extra fault recovery machinery
495 __copy_user_nocache(dst
, (void __user
*)src
, n
, 0);
498 extern const enum ib_wc_opcode ib_hfi1_wc_opcode
[];
500 extern const u8 hdr_len_by_opcode
[];
502 extern const int ib_rvt_state_ops
[];
504 extern __be64 ib_hfi1_sys_image_guid
; /* in network order */
506 extern unsigned int hfi1_max_cqes
;
508 extern unsigned int hfi1_max_cqs
;
510 extern unsigned int hfi1_max_qp_wrs
;
512 extern unsigned int hfi1_max_qps
;
514 extern unsigned int hfi1_max_sges
;
516 extern unsigned int hfi1_max_mcast_grps
;
518 extern unsigned int hfi1_max_mcast_qp_attached
;
520 extern unsigned int hfi1_max_srqs
;
522 extern unsigned int hfi1_max_srq_sges
;
524 extern unsigned int hfi1_max_srq_wrs
;
526 extern unsigned short piothreshold
;
528 extern const u32 ib_hfi1_rnr_table
[];
530 #endif /* HFI1_VERBS_H */