2 * Copyright(c) 2015 - 2017 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/ib_hdrs.h>
64 #include <rdma/rdma_vt.h>
65 #include <rdma/rdmavt_qp.h>
66 #include <rdma/rdmavt_cq.h>
69 struct hfi1_pportdata;
75 #define HFI1_MAX_RDMA_ATOMIC 16
78 * Increment this value if any changes that break userspace ABI
79 * compatibility are made.
81 #define HFI1_UVERBS_ABI_VERSION 2
83 /* IB Performance Manager status values */
84 #define IB_PMA_SAMPLE_STATUS_DONE 0x00
85 #define IB_PMA_SAMPLE_STATUS_STARTED 0x01
86 #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
88 /* Mandatory IB performance counter select values. */
89 #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
90 #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
91 #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
92 #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
93 #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
95 #define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0)
97 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
98 #define OPA_BTH_MIG_REQ BIT(31)
100 #define RC_OP(x) IB_OPCODE_RC_##x
101 #define UC_OP(x) IB_OPCODE_UC_##x
103 /* flags passed by hfi1_ib_rcv() */
105 HFI1_HAS_GRH = (1 << 0),
108 struct hfi1_16b_header {
113 struct ib_other_headers oth;
115 struct ib_other_headers oth;
119 struct hfi1_opa_header {
121 struct ib_header ibh; /* 9B header */
122 struct hfi1_16b_header opah; /* 16B header */
124 u8 hdr_type; /* 9B or 16B */
127 struct hfi1_ahg_info {
134 struct hfi1_sdma_header {
136 struct hfi1_opa_header hdr;
140 * hfi1 specific data structures that will be hidden from rvt after the queue
141 * pair is made common
143 struct hfi1_qp_priv {
144 struct hfi1_ahg_info *s_ahg; /* ahg info for next header */
145 struct sdma_engine *s_sde; /* current sde */
146 struct send_context *s_sendcontext; /* current sendcontext */
147 u8 s_sc; /* SC[0..4] for next packet */
148 struct iowait s_iowait;
149 struct rvt_qp *owner;
150 u8 hdr_type; /* 9B or 16B */
154 * This structure is used to hold commonly lookedup and computed values during
155 * the send engine progress.
157 struct hfi1_pkt_state {
158 struct hfi1_ibdev *dev;
159 struct hfi1_ibport *ibp;
160 struct hfi1_pportdata *ppd;
161 struct verbs_txreq *s_txreq;
163 unsigned long timeout;
164 unsigned long timeout_int;
170 #define HFI1_PSN_CREDIT 16
172 struct hfi1_opcode_stats {
173 u64 n_packets; /* number of packets */
174 u64 n_bytes; /* total number of bytes */
177 struct hfi1_opcode_stats_perctx {
178 struct hfi1_opcode_stats stats[256];
181 static inline void inc_opstats(
183 struct hfi1_opcode_stats *stats)
185 #ifdef CONFIG_DEBUG_FS
186 stats->n_bytes += tlen;
192 struct rvt_qp __rcu *qp[2];
193 struct rvt_ibport rvp;
195 /* the first 16 entries are sl_to_vl for !OPA */
201 struct rvt_dev_info rdi; /* Must be first */
203 /* QP numbers are shared by all IB ports */
204 /* protect txwait list */
205 seqlock_t txwait_lock ____cacheline_aligned_in_smp;
206 struct list_head txwait; /* list for wait verbs_txreq */
207 struct list_head memwait; /* list for wait kernel memory */
208 struct kmem_cache *verbs_txreq_cache;
212 /* protect iowait lists */
213 seqlock_t iowait_lock ____cacheline_aligned_in_smp;
216 struct timer_list mem_timer;
218 #ifdef CONFIG_DEBUG_FS
219 /* per HFI debugfs */
220 struct dentry *hfi1_ibdev_dbg;
221 /* per HFI symlinks to above */
222 struct dentry *hfi1_ibdev_link;
223 #ifdef CONFIG_FAULT_INJECTION
224 struct fault_opcode *fault_opcode;
225 struct fault_packet *fault_packet;
226 bool fault_suppress_err;
231 static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
233 struct rvt_dev_info *rdi;
235 rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
236 return container_of(rdi, struct hfi1_ibdev, rdi);
239 static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
241 struct hfi1_qp_priv *priv;
243 priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
248 * Send if not busy or waiting for I/O and either
249 * a RC response is pending or we can process send work requests.
251 static inline int hfi1_send_ok(struct rvt_qp *qp)
253 return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
254 (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
255 !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
259 * This must be called with s_lock held.
261 void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
262 u32 qp1, u32 qp2, u16 lid1, u16 lid2);
263 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num);
264 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
265 void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
266 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
267 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
268 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
269 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
270 u16 *out_mad_pkey_index);
273 * The PSN_MASK and PSN_SHIFT allow for
274 * 1) comparing two PSNs
275 * 2) returning the PSN with any upper bits masked
276 * 3) returning the difference between to PSNs
278 * The number of significant bits in the PSN must
279 * necessarily be at least one bit less than
280 * the container holding the PSN.
282 #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
283 #define PSN_MASK 0xFFFFFF
286 #define PSN_MASK 0x7FFFFFFF
289 #define PSN_MODIFY_MASK 0xFFFFFF
293 * Returns an integer <, ==, or > than zero.
295 static inline int cmp_psn(u32 a, u32 b)
297 return (((int)a) - ((int)b)) << PSN_SHIFT;
303 static inline u32 mask_psn(u32 a)
309 * Return delta between two PSNs
311 static inline u32 delta_psn(u32 a, u32 b)
313 return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
317 void hfi1_put_txreq(struct verbs_txreq *tx);
319 int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
321 void hfi1_copy_sge(struct rvt_sge_state *ss, void *data, u32 length,
322 bool release, bool copy_last);
324 void hfi1_cnp_rcv(struct hfi1_packet *packet);
326 void hfi1_uc_rcv(struct hfi1_packet *packet);
328 void hfi1_rc_rcv(struct hfi1_packet *packet);
331 struct hfi1_ctxtdata *rcd,
332 struct hfi1_packet *packet,
335 u8 ah_to_sc(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
337 struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u16 dlid);
339 void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah);
341 void hfi1_ud_rcv(struct hfi1_packet *packet);
343 int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
345 int hfi1_rvt_get_rwqe(struct rvt_qp *qp, int wr_id_only);
347 void hfi1_migrate_qp(struct rvt_qp *qp);
349 int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
350 int attr_mask, struct ib_udata *udata);
352 void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
353 int attr_mask, struct ib_udata *udata);
354 void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait);
355 int hfi1_check_send_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe);
357 extern const u32 rc_only_opcode;
358 extern const u32 uc_only_opcode;
360 int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct hfi1_packet *packet);
362 u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
363 const struct ib_global_route *grh, u32 hwords, u32 nwords);
365 void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
366 u32 bth0, u32 bth2, int middle,
367 struct hfi1_pkt_state *ps);
369 void _hfi1_do_send(struct work_struct *work);
371 void hfi1_do_send_from_rvt(struct rvt_qp *qp);
373 void hfi1_do_send(struct rvt_qp *qp, bool in_thread);
375 void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
376 enum ib_wc_status status);
378 void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn);
380 int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
382 int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
384 int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
386 int hfi1_register_ib_device(struct hfi1_devdata *);
388 void hfi1_unregister_ib_device(struct hfi1_devdata *);
390 void hfi1_ib_rcv(struct hfi1_packet *packet);
392 void hfi1_16B_rcv(struct hfi1_packet *packet);
394 unsigned hfi1_get_npkeys(struct hfi1_devdata *);
396 int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
399 int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
402 int hfi1_wss_init(void);
403 void hfi1_wss_exit(void);
405 /* platform specific: return the lowest level cache (llc) size, in KiB */
406 static inline int wss_llc_size(void)
408 /* assume that the boot CPU value is universal for all CPUs */
409 return boot_cpu_data.x86_cache_size;
412 /* platform specific: cacheless copy */
413 static inline void cacheless_memcpy(void *dst, void *src, size_t n)
416 * Use the only available X64 cacheless copy. Add a __user cast
417 * to quiet sparse. The src agument is already in the kernel so
418 * there are no security issues. The extra fault recovery machinery
421 __copy_user_nocache(dst, (void __user *)src, n, 0);
424 extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
426 extern const u8 hdr_len_by_opcode[];
428 extern const int ib_rvt_state_ops[];
430 extern __be64 ib_hfi1_sys_image_guid; /* in network order */
432 extern unsigned int hfi1_max_cqes;
434 extern unsigned int hfi1_max_cqs;
436 extern unsigned int hfi1_max_qp_wrs;
438 extern unsigned int hfi1_max_qps;
440 extern unsigned int hfi1_max_sges;
442 extern unsigned int hfi1_max_mcast_grps;
444 extern unsigned int hfi1_max_mcast_qp_attached;
446 extern unsigned int hfi1_max_srqs;
448 extern unsigned int hfi1_max_srq_sges;
450 extern unsigned int hfi1_max_srq_wrs;
452 extern unsigned short piothreshold;
454 extern const u32 ib_hfi1_rnr_table[];
456 #endif /* HFI1_VERBS_H */