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#include <linux/mlx4/qp.h>
34#include <linux/mlx4/srq.h>
35
36#include "mlx4_ib.h"
37#include "user.h"
38
39static void *get_wqe(struct mlx4_ib_srq *srq, int n)
40{
41 int offset = n << srq->msrq.wqe_shift;
42
43 if (srq->buf.nbufs == 1)
44 return srq->buf.u.direct.buf + offset;
45 else
46 return srq->buf.u.page_list[offset >> PAGE_SHIFT].buf +
47 (offset & (PAGE_SIZE - 1));
48}
49
50static void mlx4_ib_srq_event(struct mlx4_srq *srq, enum mlx4_event type)
51{
52 struct ib_event event;
53 struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
54
55 if (ibsrq->event_handler) {
56 event.device = ibsrq->device;
57 event.element.srq = ibsrq;
58 switch (type) {
59 case MLX4_EVENT_TYPE_SRQ_LIMIT:
60 event.event = IB_EVENT_SRQ_LIMIT_REACHED;
61 break;
62 case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
63 event.event = IB_EVENT_SRQ_ERR;
64 break;
65 default:
66 printk(KERN_WARNING "mlx4_ib: Unexpected event type %d "
67 "on SRQ %06x\n", type, srq->srqn);
68 return;
69 }
70
71 ibsrq->event_handler(&event, ibsrq->srq_context);
72 }
73}
74
75struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
76 struct ib_srq_init_attr *init_attr,
77 struct ib_udata *udata)
78{
79 struct mlx4_ib_dev *dev = to_mdev(pd->device);
80 struct mlx4_ib_srq *srq;
81 struct mlx4_wqe_srq_next_seg *next;
82 int desc_size;
83 int buf_size;
84 int err;
85 int i;
86
87
88 if (init_attr->attr.max_wr >= dev->dev->caps.max_srq_wqes ||
89 init_attr->attr.max_sge > dev->dev->caps.max_srq_sge)
90 return ERR_PTR(-EINVAL);
91
92 srq = kmalloc(sizeof *srq, GFP_KERNEL);
93 if (!srq)
94 return ERR_PTR(-ENOMEM);
95
96 mutex_init(&srq->mutex);
97 spin_lock_init(&srq->lock);
98 srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
99 srq->msrq.max_gs = init_attr->attr.max_sge;
100
101 desc_size = max(32UL,
102 roundup_pow_of_two(sizeof (struct mlx4_wqe_srq_next_seg) +
103 srq->msrq.max_gs *
104 sizeof (struct mlx4_wqe_data_seg)));
105 srq->msrq.wqe_shift = ilog2(desc_size);
106
107 buf_size = srq->msrq.max * desc_size;
108
109 if (pd->uobject) {
110 struct mlx4_ib_create_srq ucmd;
111
112 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
113 err = -EFAULT;
114 goto err_srq;
115 }
116
117 srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
118 buf_size, 0);
119 if (IS_ERR(srq->umem)) {
120 err = PTR_ERR(srq->umem);
121 goto err_srq;
122 }
123
124 err = mlx4_mtt_init(dev->dev, ib_umem_page_count(srq->umem),
125 ilog2(srq->umem->page_size), &srq->mtt);
126 if (err)
127 goto err_buf;
128
129 err = mlx4_ib_umem_write_mtt(dev, &srq->mtt, srq->umem);
130 if (err)
131 goto err_mtt;
132
133 err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
134 ucmd.db_addr, &srq->db);
135 if (err)
136 goto err_mtt;
137 } else {
138 err = mlx4_ib_db_alloc(dev, &srq->db, 0);
139 if (err)
140 goto err_srq;
141
142 *srq->db.db = 0;
143
144 if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
145 err = -ENOMEM;
146 goto err_db;
147 }
148
149 srq->head = 0;
150 srq->tail = srq->msrq.max - 1;
151 srq->wqe_ctr = 0;
152
153 for (i = 0; i < srq->msrq.max; ++i) {
154 next = get_wqe(srq, i);
155 next->next_wqe_index =
156 cpu_to_be16((i + 1) & (srq->msrq.max - 1));
157 }
158
159 err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift,
160 &srq->mtt);
161 if (err)
162 goto err_buf;
163
164 err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf);
165 if (err)
166 goto err_mtt;
167
168 srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
169 if (!srq->wrid) {
170 err = -ENOMEM;
171 goto err_mtt;
172 }
173 }
174
175 err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt,
176 srq->db.dma, &srq->msrq);
177 if (err)
178 goto err_wrid;
179
180 srq->msrq.event = mlx4_ib_srq_event;
181
182 if (pd->uobject)
183 if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
184 err = -EFAULT;
185 goto err_wrid;
186 }
187
188 init_attr->attr.max_wr = srq->msrq.max - 1;
189
190 return &srq->ibsrq;
191
192err_wrid:
193 if (pd->uobject)
194 mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
195 else
196 kfree(srq->wrid);
197
198err_mtt:
199 mlx4_mtt_cleanup(dev->dev, &srq->mtt);
200
201err_buf:
202 if (pd->uobject)
203 ib_umem_release(srq->umem);
204 else
205 mlx4_buf_free(dev->dev, buf_size, &srq->buf);
206
207err_db:
208 if (!pd->uobject)
209 mlx4_ib_db_free(dev, &srq->db);
210
211err_srq:
212 kfree(srq);
213
214 return ERR_PTR(err);
215}
216
217int mlx4_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
218 enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
219{
220 struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
221 struct mlx4_ib_srq *srq = to_msrq(ibsrq);
222 int ret;
223
224
225 if (attr_mask & IB_SRQ_MAX_WR)
226 return -EINVAL;
227
228 if (attr_mask & IB_SRQ_LIMIT) {
229 if (attr->srq_limit >= srq->msrq.max)
230 return -EINVAL;
231
232 mutex_lock(&srq->mutex);
233 ret = mlx4_srq_arm(dev->dev, &srq->msrq, attr->srq_limit);
234 mutex_unlock(&srq->mutex);
235
236 if (ret)
237 return ret;
238 }
239
240 return 0;
241}
242
243int mlx4_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
244{
245 struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
246 struct mlx4_ib_srq *srq = to_msrq(ibsrq);
247 int ret;
248 int limit_watermark;
249
250 ret = mlx4_srq_query(dev->dev, &srq->msrq, &limit_watermark);
251 if (ret)
252 return ret;
253
254 srq_attr->srq_limit = limit_watermark;
255 srq_attr->max_wr = srq->msrq.max - 1;
256 srq_attr->max_sge = srq->msrq.max_gs;
257
258 return 0;
259}
260
261int mlx4_ib_destroy_srq(struct ib_srq *srq)
262{
263 struct mlx4_ib_dev *dev = to_mdev(srq->device);
264 struct mlx4_ib_srq *msrq = to_msrq(srq);
265
266 mlx4_srq_free(dev->dev, &msrq->msrq);
267 mlx4_mtt_cleanup(dev->dev, &msrq->mtt);
268
269 if (srq->uobject) {
270 mlx4_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
271 ib_umem_release(msrq->umem);
272 } else {
273 kfree(msrq->wrid);
274 mlx4_buf_free(dev->dev, msrq->msrq.max << msrq->msrq.wqe_shift,
275 &msrq->buf);
276 mlx4_ib_db_free(dev, &msrq->db);
277 }
278
279 kfree(msrq);
280
281 return 0;
282}
283
284void mlx4_ib_free_srq_wqe(struct mlx4_ib_srq *srq, int wqe_index)
285{
286 struct mlx4_wqe_srq_next_seg *next;
287
288
289 spin_lock(&srq->lock);
290
291 next = get_wqe(srq, srq->tail);
292 next->next_wqe_index = cpu_to_be16(wqe_index);
293 srq->tail = wqe_index;
294
295 spin_unlock(&srq->lock);
296}
297
298int mlx4_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
299 struct ib_recv_wr **bad_wr)
300{
301 struct mlx4_ib_srq *srq = to_msrq(ibsrq);
302 struct mlx4_wqe_srq_next_seg *next;
303 struct mlx4_wqe_data_seg *scat;
304 unsigned long flags;
305 int err = 0;
306 int nreq;
307 int i;
308
309 spin_lock_irqsave(&srq->lock, flags);
310
311 for (nreq = 0; wr; ++nreq, wr = wr->next) {
312 if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
313 err = -EINVAL;
314 *bad_wr = wr;
315 break;
316 }
317
318 if (unlikely(srq->head == srq->tail)) {
319 err = -ENOMEM;
320 *bad_wr = wr;
321 break;
322 }
323
324 srq->wrid[srq->head] = wr->wr_id;
325
326 next = get_wqe(srq, srq->head);
327 srq->head = be16_to_cpu(next->next_wqe_index);
328 scat = (struct mlx4_wqe_data_seg *) (next + 1);
329
330 for (i = 0; i < wr->num_sge; ++i) {
331 scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
332 scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
333 scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
334 }
335
336 if (i < srq->msrq.max_gs) {
337 scat[i].byte_count = 0;
338 scat[i].lkey = cpu_to_be32(MLX4_INVALID_LKEY);
339 scat[i].addr = 0;
340 }
341 }
342
343 if (likely(nreq)) {
344 srq->wqe_ctr += nreq;
345
346
347
348
349
350 wmb();
351
352 *srq->db.db = cpu_to_be32(srq->wqe_ctr);
353 }
354
355 spin_unlock_irqrestore(&srq->lock, flags);
356
357 return err;
358}
359
