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 hfi1_ahg_info
{
188 struct hfi1_pio_header
{
190 struct hfi1_ib_header hdr
;
194 * hfi1 specific data structures that will be hidden from rvt after the queue
195 * pair is made common
197 struct hfi1_qp_priv
{
198 struct hfi1_ahg_info
*s_ahg
; /* ahg info for next header */
199 struct sdma_engine
*s_sde
; /* current sde */
200 struct send_context
*s_sendcontext
; /* current sendcontext */
201 u8 s_sc
; /* SC[0..4] for next packet */
202 u8 r_adefered
; /* number of acks defered */
203 struct iowait s_iowait
;
204 struct timer_list s_rnr_timer
;
205 struct rvt_qp
*owner
;
209 * This structure is used to hold commonly lookedup and computed values during
210 * the send engine progress.
212 struct hfi1_pkt_state
{
213 struct hfi1_ibdev
*dev
;
214 struct hfi1_ibport
*ibp
;
215 struct hfi1_pportdata
*ppd
;
216 struct verbs_txreq
*s_txreq
;
220 #define HFI1_PSN_CREDIT 16
222 struct hfi1_opcode_stats
{
223 u64 n_packets
; /* number of packets */
224 u64 n_bytes
; /* total number of bytes */
227 struct hfi1_opcode_stats_perctx
{
228 struct hfi1_opcode_stats stats
[256];
231 static inline void inc_opstats(
233 struct hfi1_opcode_stats
*stats
)
235 #ifdef CONFIG_DEBUG_FS
236 stats
->n_bytes
+= tlen
;
242 struct rvt_qp __rcu
*qp
[2];
243 struct rvt_ibport rvp
;
245 __be64 guids
[HFI1_GUIDS_PER_PORT
- 1]; /* writable GUIDs */
247 /* the first 16 entries are sl_to_vl for !OPA */
253 struct rvt_dev_info rdi
; /* Must be first */
255 /* QP numbers are shared by all IB ports */
256 /* protect wait lists */
257 seqlock_t iowait_lock
;
258 struct list_head txwait
; /* list for wait verbs_txreq */
259 struct list_head memwait
; /* list for wait kernel memory */
260 struct list_head txreq_free
;
261 struct kmem_cache
*verbs_txreq_cache
;
262 struct timer_list mem_timer
;
269 #ifdef CONFIG_DEBUG_FS
270 /* per HFI debugfs */
271 struct dentry
*hfi1_ibdev_dbg
;
272 /* per HFI symlinks to above */
273 struct dentry
*hfi1_ibdev_link
;
277 static inline struct hfi1_ibdev
*to_idev(struct ib_device
*ibdev
)
279 struct rvt_dev_info
*rdi
;
281 rdi
= container_of(ibdev
, struct rvt_dev_info
, ibdev
);
282 return container_of(rdi
, struct hfi1_ibdev
, rdi
);
285 static inline struct rvt_qp
*iowait_to_qp(struct iowait
*s_iowait
)
287 struct hfi1_qp_priv
*priv
;
289 priv
= container_of(s_iowait
, struct hfi1_qp_priv
, s_iowait
);
294 * Send if not busy or waiting for I/O and either
295 * a RC response is pending or we can process send work requests.
297 static inline int hfi1_send_ok(struct rvt_qp
*qp
)
299 return !(qp
->s_flags
& (RVT_S_BUSY
| RVT_S_ANY_WAIT_IO
)) &&
300 (qp
->s_hdrwords
|| (qp
->s_flags
& RVT_S_RESP_PENDING
) ||
301 !(qp
->s_flags
& RVT_S_ANY_WAIT_SEND
));
305 * This must be called with s_lock held.
307 void hfi1_bad_pqkey(struct hfi1_ibport
*ibp
, __be16 trap_num
, u32 key
, u32 sl
,
308 u32 qp1
, u32 qp2
, u16 lid1
, u16 lid2
);
309 void hfi1_cap_mask_chg(struct rvt_dev_info
*rdi
, u8 port_num
);
310 void hfi1_sys_guid_chg(struct hfi1_ibport
*ibp
);
311 void hfi1_node_desc_chg(struct hfi1_ibport
*ibp
);
312 int hfi1_process_mad(struct ib_device
*ibdev
, int mad_flags
, u8 port
,
313 const struct ib_wc
*in_wc
, const struct ib_grh
*in_grh
,
314 const struct ib_mad_hdr
*in_mad
, size_t in_mad_size
,
315 struct ib_mad_hdr
*out_mad
, size_t *out_mad_size
,
316 u16
*out_mad_pkey_index
);
319 * The PSN_MASK and PSN_SHIFT allow for
320 * 1) comparing two PSNs
321 * 2) returning the PSN with any upper bits masked
322 * 3) returning the difference between to PSNs
324 * The number of significant bits in the PSN must
325 * necessarily be at least one bit less than
326 * the container holding the PSN.
328 #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
329 #define PSN_MASK 0xFFFFFF
332 #define PSN_MASK 0x7FFFFFFF
335 #define PSN_MODIFY_MASK 0xFFFFFF
338 * Compare the lower 24 bits of the msn values.
339 * Returns an integer <, ==, or > than zero.
341 static inline int cmp_msn(u32 a
, u32 b
)
343 return (((int)a
) - ((int)b
)) << 8;
348 * Returns an integer <, ==, or > than zero.
350 static inline int cmp_psn(u32 a
, u32 b
)
352 return (((int)a
) - ((int)b
)) << PSN_SHIFT
;
358 static inline u32
mask_psn(u32 a
)
364 * Return delta between two PSNs
366 static inline u32
delta_psn(u32 a
, u32 b
)
368 return (((int)a
- (int)b
) << PSN_SHIFT
) >> PSN_SHIFT
;
372 void hfi1_put_txreq(struct verbs_txreq
*tx
);
374 int hfi1_verbs_send(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
376 void hfi1_copy_sge(struct rvt_sge_state
*ss
, void *data
, u32 length
,
377 int release
, int copy_last
);
379 void hfi1_skip_sge(struct rvt_sge_state
*ss
, u32 length
, int release
);
381 void hfi1_cnp_rcv(struct hfi1_packet
*packet
);
383 void hfi1_uc_rcv(struct hfi1_packet
*packet
);
385 void hfi1_rc_rcv(struct hfi1_packet
*packet
);
388 struct hfi1_ctxtdata
*rcd
,
389 struct hfi1_ib_header
*hdr
,
393 u8
ah_to_sc(struct ib_device
*ibdev
, struct ib_ah_attr
*ah_attr
);
395 struct ib_ah
*hfi1_create_qp0_ah(struct hfi1_ibport
*ibp
, u16 dlid
);
397 void hfi1_rc_rnr_retry(unsigned long arg
);
398 void hfi1_add_rnr_timer(struct rvt_qp
*qp
, u32 to
);
399 void hfi1_rc_timeout(unsigned long arg
);
400 void hfi1_del_timers_sync(struct rvt_qp
*qp
);
401 void hfi1_stop_rc_timers(struct rvt_qp
*qp
);
403 void hfi1_rc_send_complete(struct rvt_qp
*qp
, struct hfi1_ib_header
*hdr
);
405 void hfi1_rc_error(struct rvt_qp
*qp
, enum ib_wc_status err
);
407 void hfi1_ud_rcv(struct hfi1_packet
*packet
);
409 int hfi1_lookup_pkey_idx(struct hfi1_ibport
*ibp
, u16 pkey
);
411 int hfi1_rvt_get_rwqe(struct rvt_qp
*qp
, int wr_id_only
);
413 void hfi1_migrate_qp(struct rvt_qp
*qp
);
415 int hfi1_check_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
416 int attr_mask
, struct ib_udata
*udata
);
418 void hfi1_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
419 int attr_mask
, struct ib_udata
*udata
);
421 int hfi1_check_send_wqe(struct rvt_qp
*qp
, struct rvt_swqe
*wqe
);
423 extern const u32 rc_only_opcode
;
424 extern const u32 uc_only_opcode
;
426 static inline u8
get_opcode(struct hfi1_ib_header
*h
)
428 u16 lnh
= be16_to_cpu(h
->lrh
[0]) & 3;
430 if (lnh
== IB_LNH_IBA_LOCAL
)
431 return be32_to_cpu(h
->u
.oth
.bth
[0]) >> 24;
433 return be32_to_cpu(h
->u
.l
.oth
.bth
[0]) >> 24;
436 int hfi1_ruc_check_hdr(struct hfi1_ibport
*ibp
, struct hfi1_ib_header
*hdr
,
437 int has_grh
, struct rvt_qp
*qp
, u32 bth0
);
439 u32
hfi1_make_grh(struct hfi1_ibport
*ibp
, struct ib_grh
*hdr
,
440 struct ib_global_route
*grh
, u32 hwords
, u32 nwords
);
442 void hfi1_make_ruc_header(struct rvt_qp
*qp
, struct hfi1_other_headers
*ohdr
,
443 u32 bth0
, u32 bth2
, int middle
,
444 struct hfi1_pkt_state
*ps
);
446 void _hfi1_do_send(struct work_struct
*work
);
448 void hfi1_do_send(struct rvt_qp
*qp
);
450 void hfi1_send_complete(struct rvt_qp
*qp
, struct rvt_swqe
*wqe
,
451 enum ib_wc_status status
);
453 void hfi1_send_rc_ack(struct hfi1_ctxtdata
*, struct rvt_qp
*qp
, int is_fecn
);
455 int hfi1_make_rc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
457 int hfi1_make_uc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
459 int hfi1_make_ud_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
461 int hfi1_register_ib_device(struct hfi1_devdata
*);
463 void hfi1_unregister_ib_device(struct hfi1_devdata
*);
465 void hfi1_ib_rcv(struct hfi1_packet
*packet
);
467 unsigned hfi1_get_npkeys(struct hfi1_devdata
*);
469 int hfi1_verbs_send_dma(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
472 int hfi1_verbs_send_pio(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
475 int hfi1_wss_init(void);
476 void hfi1_wss_exit(void);
478 /* platform specific: return the lowest level cache (llc) size, in KiB */
479 static inline int wss_llc_size(void)
481 /* assume that the boot CPU value is universal for all CPUs */
482 return boot_cpu_data
.x86_cache_size
;
485 /* platform specific: cacheless copy */
486 static inline void cacheless_memcpy(void *dst
, void *src
, size_t n
)
489 * Use the only available X64 cacheless copy. Add a __user cast
490 * to quiet sparse. The src agument is already in the kernel so
491 * there are no security issues. The extra fault recovery machinery
494 __copy_user_nocache(dst
, (void __user
*)src
, n
, 0);
497 extern const enum ib_wc_opcode ib_hfi1_wc_opcode
[];
499 extern const u8 hdr_len_by_opcode
[];
501 extern const int ib_rvt_state_ops
[];
503 extern __be64 ib_hfi1_sys_image_guid
; /* in network order */
505 extern unsigned int hfi1_max_cqes
;
507 extern unsigned int hfi1_max_cqs
;
509 extern unsigned int hfi1_max_qp_wrs
;
511 extern unsigned int hfi1_max_qps
;
513 extern unsigned int hfi1_max_sges
;
515 extern unsigned int hfi1_max_mcast_grps
;
517 extern unsigned int hfi1_max_mcast_qp_attached
;
519 extern unsigned int hfi1_max_srqs
;
521 extern unsigned int hfi1_max_srq_sges
;
523 extern unsigned int hfi1_max_srq_wrs
;
525 extern unsigned short piothreshold
;
527 extern const u32 ib_hfi1_rnr_table
[];
529 #endif /* HFI1_VERBS_H */