1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
|
// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
/*
* Copyright(c) 2018 Intel Corporation.
*
*/
#include "hfi.h"
#include "trace.h"
#include "qp.h"
#include "opfn.h"
#define IB_BTHE_E BIT(IB_BTHE_E_SHIFT)
#define OPFN_CODE(code) BIT((code) - 1)
#define OPFN_MASK(code) OPFN_CODE(STL_VERBS_EXTD_##code)
struct hfi1_opfn_type {
bool (*request)(struct rvt_qp *qp, u64 *data);
bool (*response)(struct rvt_qp *qp, u64 *data);
bool (*reply)(struct rvt_qp *qp, u64 data);
void (*error)(struct rvt_qp *qp);
};
static struct hfi1_opfn_type hfi1_opfn_handlers[STL_VERBS_EXTD_MAX] = {
[STL_VERBS_EXTD_TID_RDMA] = {
.request = tid_rdma_conn_req,
.response = tid_rdma_conn_resp,
.reply = tid_rdma_conn_reply,
.error = tid_rdma_conn_error,
},
};
static struct workqueue_struct *opfn_wq;
static void opfn_schedule_conn_request(struct rvt_qp *qp);
static bool hfi1_opfn_extended(u32 bth1)
{
return !!(bth1 & IB_BTHE_E);
}
static void opfn_conn_request(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
struct ib_atomic_wr wr;
u16 mask, capcode;
struct hfi1_opfn_type *extd;
u64 data;
unsigned long flags;
int ret = 0;
trace_hfi1_opfn_state_conn_request(qp);
spin_lock_irqsave(&priv->opfn.lock, flags);
/*
* Exit if the extended bit is not set, or if nothing is requested, or
* if we have completed all requests, or if a previous request is in
* progress
*/
if (!priv->opfn.extended || !priv->opfn.requested ||
priv->opfn.requested == priv->opfn.completed || priv->opfn.curr)
goto done;
mask = priv->opfn.requested & ~priv->opfn.completed;
capcode = ilog2(mask & ~(mask - 1)) + 1;
if (capcode >= STL_VERBS_EXTD_MAX) {
priv->opfn.completed |= OPFN_CODE(capcode);
goto done;
}
extd = &hfi1_opfn_handlers[capcode];
if (!extd || !extd->request || !extd->request(qp, &data)) {
/*
* Either there is no handler for this capability or the request
* packet could not be generated. Either way, mark it as done so
* we don't keep attempting to complete it.
*/
priv->opfn.completed |= OPFN_CODE(capcode);
goto done;
}
trace_hfi1_opfn_data_conn_request(qp, capcode, data);
data = (data & ~0xf) | capcode;
memset(&wr, 0, sizeof(wr));
wr.wr.opcode = IB_WR_OPFN;
wr.remote_addr = HFI1_VERBS_E_ATOMIC_VADDR;
wr.compare_add = data;
priv->opfn.curr = capcode; /* A new request is now in progress */
/* Drop opfn.lock before calling ib_post_send() */
spin_unlock_irqrestore(&priv->opfn.lock, flags);
ret = ib_post_send(&qp->ibqp, &wr.wr, NULL);
if (ret)
goto err;
trace_hfi1_opfn_state_conn_request(qp);
return;
err:
trace_hfi1_msg_opfn_conn_request(qp, "ib_ost_send failed: ret = ",
(u64)ret);
spin_lock_irqsave(&priv->opfn.lock, flags);
/*
* In case of an unexpected error return from ib_post_send
* clear opfn.curr and reschedule to try again
*/
priv->opfn.curr = STL_VERBS_EXTD_NONE;
opfn_schedule_conn_request(qp);
done:
spin_unlock_irqrestore(&priv->opfn.lock, flags);
}
void opfn_send_conn_request(struct work_struct *work)
{
struct hfi1_opfn_data *od;
struct hfi1_qp_priv *qpriv;
od = container_of(work, struct hfi1_opfn_data, opfn_work);
qpriv = container_of(od, struct hfi1_qp_priv, opfn);
opfn_conn_request(qpriv->owner);
}
/*
* When QP s_lock is held in the caller, the OPFN request must be scheduled
* to a different workqueue to avoid double locking QP s_lock in call to
* ib_post_send in opfn_conn_request
*/
static void opfn_schedule_conn_request(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
trace_hfi1_opfn_state_sched_conn_request(qp);
queue_work(opfn_wq, &priv->opfn.opfn_work);
}
void opfn_conn_response(struct rvt_qp *qp, struct rvt_ack_entry *e,
struct ib_atomic_eth *ateth)
{
struct hfi1_qp_priv *priv = qp->priv;
u64 data = be64_to_cpu(ateth->compare_data);
struct hfi1_opfn_type *extd;
u8 capcode;
unsigned long flags;
trace_hfi1_opfn_state_conn_response(qp);
capcode = data & 0xf;
trace_hfi1_opfn_data_conn_response(qp, capcode, data);
if (!capcode || capcode >= STL_VERBS_EXTD_MAX)
return;
extd = &hfi1_opfn_handlers[capcode];
if (!extd || !extd->response) {
e->atomic_data = capcode;
return;
}
spin_lock_irqsave(&priv->opfn.lock, flags);
if (priv->opfn.completed & OPFN_CODE(capcode)) {
/*
* We are receiving a request for a feature that has already
* been negotiated. This may mean that the other side has reset
*/
priv->opfn.completed &= ~OPFN_CODE(capcode);
if (extd->error)
extd->error(qp);
}
if (extd->response(qp, &data))
priv->opfn.completed |= OPFN_CODE(capcode);
e->atomic_data = (data & ~0xf) | capcode;
trace_hfi1_opfn_state_conn_response(qp);
spin_unlock_irqrestore(&priv->opfn.lock, flags);
}
void opfn_conn_reply(struct rvt_qp *qp, u64 data)
{
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_opfn_type *extd;
u8 capcode;
unsigned long flags;
trace_hfi1_opfn_state_conn_reply(qp);
capcode = data & 0xf;
trace_hfi1_opfn_data_conn_reply(qp, capcode, data);
if (!capcode || capcode >= STL_VERBS_EXTD_MAX)
return;
spin_lock_irqsave(&priv->opfn.lock, flags);
/*
* Either there is no previous request or the reply is not for the
* current request
*/
if (!priv->opfn.curr || capcode != priv->opfn.curr)
goto done;
extd = &hfi1_opfn_handlers[capcode];
if (!extd || !extd->reply)
goto clear;
if (extd->reply(qp, data))
priv->opfn.completed |= OPFN_CODE(capcode);
clear:
/*
* Clear opfn.curr to indicate that the previous request is no longer in
* progress
*/
priv->opfn.curr = STL_VERBS_EXTD_NONE;
trace_hfi1_opfn_state_conn_reply(qp);
done:
spin_unlock_irqrestore(&priv->opfn.lock, flags);
}
void opfn_conn_error(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_opfn_type *extd = NULL;
unsigned long flags;
u16 capcode;
trace_hfi1_opfn_state_conn_error(qp);
trace_hfi1_msg_opfn_conn_error(qp, "error. qp state ", (u64)qp->state);
/*
* The QP has gone into the Error state. We have to invalidate all
* negotiated feature, including the one in progress (if any). The RC
* QP handling will clean the WQE for the connection request.
*/
spin_lock_irqsave(&priv->opfn.lock, flags);
while (priv->opfn.completed) {
capcode = priv->opfn.completed & ~(priv->opfn.completed - 1);
extd = &hfi1_opfn_handlers[ilog2(capcode) + 1];
if (extd->error)
extd->error(qp);
priv->opfn.completed &= ~OPFN_CODE(capcode);
}
priv->opfn.extended = 0;
priv->opfn.requested = 0;
priv->opfn.curr = STL_VERBS_EXTD_NONE;
spin_unlock_irqrestore(&priv->opfn.lock, flags);
}
void opfn_qp_init(struct rvt_qp *qp, struct ib_qp_attr *attr, int attr_mask)
{
struct ib_qp *ibqp = &qp->ibqp;
struct hfi1_qp_priv *priv = qp->priv;
unsigned long flags;
if (attr_mask & IB_QP_RETRY_CNT)
priv->s_retry = attr->retry_cnt;
spin_lock_irqsave(&priv->opfn.lock, flags);
if (ibqp->
|