Commit | Line | Data |
---|---|---|
cef1cce5 | 1 | /* |
759d5768 | 2 | * Copyright (c) 2006 QLogic, Inc. All rights reserved. |
cef1cce5 BS |
3 | * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. |
4 | * | |
5 | * This software is available to you under a choice of one of two | |
6 | * licenses. You may choose to be licensed under the terms of the GNU | |
7 | * General Public License (GPL) Version 2, available from the file | |
8 | * COPYING in the main directory of this source tree, or the | |
9 | * OpenIB.org BSD license below: | |
10 | * | |
11 | * Redistribution and use in source and binary forms, with or | |
12 | * without modification, are permitted provided that the following | |
13 | * conditions are met: | |
14 | * | |
15 | * - Redistributions of source code must retain the above | |
16 | * copyright notice, this list of conditions and the following | |
17 | * disclaimer. | |
18 | * | |
19 | * - Redistributions in binary form must reproduce the above | |
20 | * copyright notice, this list of conditions and the following | |
21 | * disclaimer in the documentation and/or other materials | |
22 | * provided with the distribution. | |
23 | * | |
24 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
25 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
26 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
27 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
28 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
29 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
30 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
31 | * SOFTWARE. | |
32 | */ | |
33 | ||
34 | #include <linux/err.h> | |
35 | #include <linux/vmalloc.h> | |
36 | ||
37 | #include "ipath_verbs.h" | |
38 | ||
39 | /** | |
40 | * ipath_cq_enter - add a new entry to the completion queue | |
41 | * @cq: completion queue | |
42 | * @entry: work completion entry to add | |
43 | * @sig: true if @entry is a solicitated entry | |
44 | * | |
373d9915 | 45 | * This may be called with qp->s_lock held. |
cef1cce5 BS |
46 | */ |
47 | void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited) | |
48 | { | |
7a26c474 | 49 | struct ipath_cq_wc *wc; |
cef1cce5 | 50 | unsigned long flags; |
373d9915 | 51 | u32 head; |
cef1cce5 BS |
52 | u32 next; |
53 | ||
54 | spin_lock_irqsave(&cq->lock, flags); | |
55 | ||
373d9915 RC |
56 | /* |
57 | * Note that the head pointer might be writable by user processes. | |
58 | * Take care to verify it is a sane value. | |
59 | */ | |
7a26c474 | 60 | wc = cq->queue; |
373d9915 RC |
61 | head = wc->head; |
62 | if (head >= (unsigned) cq->ibcq.cqe) { | |
63 | head = cq->ibcq.cqe; | |
cef1cce5 | 64 | next = 0; |
373d9915 RC |
65 | } else |
66 | next = head + 1; | |
67 | if (unlikely(next == wc->tail)) { | |
cef1cce5 BS |
68 | spin_unlock_irqrestore(&cq->lock, flags); |
69 | if (cq->ibcq.event_handler) { | |
70 | struct ib_event ev; | |
71 | ||
72 | ev.device = cq->ibcq.device; | |
73 | ev.element.cq = &cq->ibcq; | |
74 | ev.event = IB_EVENT_CQ_ERR; | |
75 | cq->ibcq.event_handler(&ev, cq->ibcq.cq_context); | |
76 | } | |
77 | return; | |
78 | } | |
373d9915 RC |
79 | wc->queue[head] = *entry; |
80 | wc->head = next; | |
cef1cce5 BS |
81 | |
82 | if (cq->notify == IB_CQ_NEXT_COMP || | |
83 | (cq->notify == IB_CQ_SOLICITED && solicited)) { | |
84 | cq->notify = IB_CQ_NONE; | |
85 | cq->triggered++; | |
86 | /* | |
87 | * This will cause send_complete() to be called in | |
88 | * another thread. | |
89 | */ | |
90 | tasklet_hi_schedule(&cq->comptask); | |
91 | } | |
92 | ||
93 | spin_unlock_irqrestore(&cq->lock, flags); | |
94 | ||
95 | if (entry->status != IB_WC_SUCCESS) | |
96 | to_idev(cq->ibcq.device)->n_wqe_errs++; | |
97 | } | |
98 | ||
99 | /** | |
100 | * ipath_poll_cq - poll for work completion entries | |
101 | * @ibcq: the completion queue to poll | |
102 | * @num_entries: the maximum number of entries to return | |
103 | * @entry: pointer to array where work completions are placed | |
104 | * | |
105 | * Returns the number of completion entries polled. | |
106 | * | |
107 | * This may be called from interrupt context. Also called by ib_poll_cq() | |
108 | * in the generic verbs code. | |
109 | */ | |
110 | int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry) | |
111 | { | |
112 | struct ipath_cq *cq = to_icq(ibcq); | |
7a26c474 | 113 | struct ipath_cq_wc *wc; |
cef1cce5 BS |
114 | unsigned long flags; |
115 | int npolled; | |
7a26c474 | 116 | u32 tail; |
cef1cce5 BS |
117 | |
118 | spin_lock_irqsave(&cq->lock, flags); | |
119 | ||
7a26c474 BS |
120 | wc = cq->queue; |
121 | tail = wc->tail; | |
122 | if (tail > (u32) cq->ibcq.cqe) | |
123 | tail = (u32) cq->ibcq.cqe; | |
cef1cce5 | 124 | for (npolled = 0; npolled < num_entries; ++npolled, ++entry) { |
7a26c474 | 125 | if (tail == wc->head) |
cef1cce5 | 126 | break; |
7a26c474 BS |
127 | *entry = wc->queue[tail]; |
128 | if (tail >= cq->ibcq.cqe) | |
129 | tail = 0; | |
cef1cce5 | 130 | else |
7a26c474 | 131 | tail++; |
cef1cce5 | 132 | } |
7a26c474 | 133 | wc->tail = tail; |
cef1cce5 BS |
134 | |
135 | spin_unlock_irqrestore(&cq->lock, flags); | |
136 | ||
137 | return npolled; | |
138 | } | |
139 | ||
140 | static void send_complete(unsigned long data) | |
141 | { | |
142 | struct ipath_cq *cq = (struct ipath_cq *)data; | |
143 | ||
144 | /* | |
145 | * The completion handler will most likely rearm the notification | |
146 | * and poll for all pending entries. If a new completion entry | |
147 | * is added while we are in this routine, tasklet_hi_schedule() | |
148 | * won't call us again until we return so we check triggered to | |
149 | * see if we need to call the handler again. | |
150 | */ | |
151 | for (;;) { | |
152 | u8 triggered = cq->triggered; | |
153 | ||
154 | cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context); | |
155 | ||
156 | if (cq->triggered == triggered) | |
157 | return; | |
158 | } | |
159 | } | |
160 | ||
161 | /** | |
162 | * ipath_create_cq - create a completion queue | |
163 | * @ibdev: the device this completion queue is attached to | |
164 | * @entries: the minimum size of the completion queue | |
165 | * @context: unused by the InfiniPath driver | |
166 | * @udata: unused by the InfiniPath driver | |
167 | * | |
168 | * Returns a pointer to the completion queue or negative errno values | |
169 | * for failure. | |
170 | * | |
171 | * Called by ib_create_cq() in the generic verbs code. | |
172 | */ | |
173 | struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries, | |
174 | struct ib_ucontext *context, | |
175 | struct ib_udata *udata) | |
176 | { | |
fe62546a | 177 | struct ipath_ibdev *dev = to_idev(ibdev); |
cef1cce5 | 178 | struct ipath_cq *cq; |
373d9915 | 179 | struct ipath_cq_wc *wc; |
cef1cce5 BS |
180 | struct ib_cq *ret; |
181 | ||
eae33d47 | 182 | if (entries < 1 || entries > ib_ipath_max_cqes) { |
fe62546a | 183 | ret = ERR_PTR(-EINVAL); |
373d9915 | 184 | goto done; |
fe62546a BS |
185 | } |
186 | ||
373d9915 | 187 | /* Allocate the completion queue structure. */ |
cef1cce5 BS |
188 | cq = kmalloc(sizeof(*cq), GFP_KERNEL); |
189 | if (!cq) { | |
190 | ret = ERR_PTR(-ENOMEM); | |
373d9915 | 191 | goto done; |
cef1cce5 BS |
192 | } |
193 | ||
194 | /* | |
373d9915 RC |
195 | * Allocate the completion queue entries and head/tail pointers. |
196 | * This is allocated separately so that it can be resized and | |
197 | * also mapped into user space. | |
198 | * We need to use vmalloc() in order to support mmap and large | |
199 | * numbers of entries. | |
cef1cce5 | 200 | */ |
373d9915 | 201 | wc = vmalloc_user(sizeof(*wc) + sizeof(struct ib_wc) * entries); |
cef1cce5 | 202 | if (!wc) { |
cef1cce5 | 203 | ret = ERR_PTR(-ENOMEM); |
373d9915 | 204 | goto bail_cq; |
cef1cce5 | 205 | } |
373d9915 RC |
206 | |
207 | /* | |
208 | * Return the address of the WC as the offset to mmap. | |
209 | * See ipath_mmap() for details. | |
210 | */ | |
211 | if (udata && udata->outlen >= sizeof(__u64)) { | |
212 | struct ipath_mmap_info *ip; | |
213 | __u64 offset = (__u64) wc; | |
214 | int err; | |
215 | ||
216 | err = ib_copy_to_udata(udata, &offset, sizeof(offset)); | |
217 | if (err) { | |
218 | ret = ERR_PTR(err); | |
219 | goto bail_wc; | |
220 | } | |
221 | ||
222 | /* Allocate info for ipath_mmap(). */ | |
223 | ip = kmalloc(sizeof(*ip), GFP_KERNEL); | |
224 | if (!ip) { | |
225 | ret = ERR_PTR(-ENOMEM); | |
226 | goto bail_wc; | |
227 | } | |
228 | cq->ip = ip; | |
229 | ip->context = context; | |
230 | ip->obj = wc; | |
231 | kref_init(&ip->ref); | |
232 | ip->mmap_cnt = 0; | |
233 | ip->size = PAGE_ALIGN(sizeof(*wc) + | |
234 | sizeof(struct ib_wc) * entries); | |
235 | spin_lock_irq(&dev->pending_lock); | |
236 | ip->next = dev->pending_mmaps; | |
237 | dev->pending_mmaps = ip; | |
238 | spin_unlock_irq(&dev->pending_lock); | |
239 | } else | |
240 | cq->ip = NULL; | |
241 | ||
aa4eaed7 BS |
242 | spin_lock(&dev->n_cqs_lock); |
243 | if (dev->n_cqs_allocated == ib_ipath_max_cqs) { | |
244 | spin_unlock(&dev->n_cqs_lock); | |
245 | ret = ERR_PTR(-ENOMEM); | |
246 | goto bail_wc; | |
247 | } | |
248 | ||
249 | dev->n_cqs_allocated++; | |
250 | spin_unlock(&dev->n_cqs_lock); | |
251 | ||
cef1cce5 BS |
252 | /* |
253 | * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe. | |
254 | * The number of entries should be >= the number requested or return | |
255 | * an error. | |
256 | */ | |
257 | cq->ibcq.cqe = entries; | |
258 | cq->notify = IB_CQ_NONE; | |
259 | cq->triggered = 0; | |
260 | spin_lock_init(&cq->lock); | |
261 | tasklet_init(&cq->comptask, send_complete, (unsigned long)cq); | |
373d9915 RC |
262 | wc->head = 0; |
263 | wc->tail = 0; | |
cef1cce5 BS |
264 | cq->queue = wc; |
265 | ||
266 | ret = &cq->ibcq; | |
267 | ||
373d9915 | 268 | goto done; |
fe62546a | 269 | |
373d9915 RC |
270 | bail_wc: |
271 | vfree(wc); | |
272 | ||
273 | bail_cq: | |
274 | kfree(cq); | |
275 | ||
276 | done: | |
cef1cce5 BS |
277 | return ret; |
278 | } | |
279 | ||
280 | /** | |
281 | * ipath_destroy_cq - destroy a completion queue | |
282 | * @ibcq: the completion queue to destroy. | |
283 | * | |
284 | * Returns 0 for success. | |
285 | * | |
286 | * Called by ib_destroy_cq() in the generic verbs code. | |
287 | */ | |
288 | int ipath_destroy_cq(struct ib_cq *ibcq) | |
289 | { | |
fe62546a | 290 | struct ipath_ibdev *dev = to_idev(ibcq->device); |
cef1cce5 BS |
291 | struct ipath_cq *cq = to_icq(ibcq); |
292 | ||
293 | tasklet_kill(&cq->comptask); | |
aa4eaed7 | 294 | spin_lock(&dev->n_cqs_lock); |
fe62546a | 295 | dev->n_cqs_allocated--; |
aa4eaed7 | 296 | spin_unlock(&dev->n_cqs_lock); |
373d9915 RC |
297 | if (cq->ip) |
298 | kref_put(&cq->ip->ref, ipath_release_mmap_info); | |
299 | else | |
300 | vfree(cq->queue); | |
cef1cce5 BS |
301 | kfree(cq); |
302 | ||
303 | return 0; | |
304 | } | |
305 | ||
306 | /** | |
307 | * ipath_req_notify_cq - change the notification type for a completion queue | |
308 | * @ibcq: the completion queue | |
309 | * @notify: the type of notification to request | |
310 | * | |
311 | * Returns 0 for success. | |
312 | * | |
313 | * This may be called from interrupt context. Also called by | |
314 | * ib_req_notify_cq() in the generic verbs code. | |
315 | */ | |
316 | int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) | |
317 | { | |
318 | struct ipath_cq *cq = to_icq(ibcq); | |
319 | unsigned long flags; | |
320 | ||
321 | spin_lock_irqsave(&cq->lock, flags); | |
322 | /* | |
323 | * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow | |
373d9915 | 324 | * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2). |
cef1cce5 BS |
325 | */ |
326 | if (cq->notify != IB_CQ_NEXT_COMP) | |
327 | cq->notify = notify; | |
328 | spin_unlock_irqrestore(&cq->lock, flags); | |
329 | return 0; | |
330 | } | |
331 | ||
7a26c474 BS |
332 | /** |
333 | * ipath_resize_cq - change the size of the CQ | |
334 | * @ibcq: the completion queue | |
335 | * | |
336 | * Returns 0 for success. | |
337 | */ | |
cef1cce5 BS |
338 | int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) |
339 | { | |
340 | struct ipath_cq *cq = to_icq(ibcq); | |
7a26c474 | 341 | struct ipath_cq_wc *old_wc; |
373d9915 RC |
342 | struct ipath_cq_wc *wc; |
343 | u32 head, tail, n; | |
cef1cce5 BS |
344 | int ret; |
345 | ||
eae33d47 BS |
346 | if (cqe < 1 || cqe > ib_ipath_max_cqes) { |
347 | ret = -EINVAL; | |
348 | goto bail; | |
349 | } | |
350 | ||
cef1cce5 BS |
351 | /* |
352 | * Need to use vmalloc() if we want to support large #s of entries. | |
353 | */ | |
373d9915 | 354 | wc = vmalloc_user(sizeof(*wc) + sizeof(struct ib_wc) * cqe); |
cef1cce5 BS |
355 | if (!wc) { |
356 | ret = -ENOMEM; | |
357 | goto bail; | |
358 | } | |
359 | ||
373d9915 RC |
360 | /* |
361 | * Return the address of the WC as the offset to mmap. | |
362 | * See ipath_mmap() for details. | |
363 | */ | |
364 | if (udata && udata->outlen >= sizeof(__u64)) { | |
365 | __u64 offset = (__u64) wc; | |
366 | ||
367 | ret = ib_copy_to_udata(udata, &offset, sizeof(offset)); | |
368 | if (ret) | |
369 | goto bail; | |
370 | } | |
371 | ||
cef1cce5 | 372 | spin_lock_irq(&cq->lock); |
373d9915 RC |
373 | /* |
374 | * Make sure head and tail are sane since they | |
375 | * might be user writable. | |
376 | */ | |
7a26c474 | 377 | old_wc = cq->queue; |
373d9915 RC |
378 | head = old_wc->head; |
379 | if (head > (u32) cq->ibcq.cqe) | |
380 | head = (u32) cq->ibcq.cqe; | |
381 | tail = old_wc->tail; | |
382 | if (tail > (u32) cq->ibcq.cqe) | |
383 | tail = (u32) cq->ibcq.cqe; | |
384 | if (head < tail) | |
385 | n = cq->ibcq.cqe + 1 + head - tail; | |
cef1cce5 | 386 | else |
373d9915 | 387 | n = head - tail; |
cef1cce5 BS |
388 | if (unlikely((u32)cqe < n)) { |
389 | spin_unlock_irq(&cq->lock); | |
390 | vfree(wc); | |
391 | ret = -EOVERFLOW; | |
392 | goto bail; | |
393 | } | |
373d9915 RC |
394 | for (n = 0; tail != head; n++) { |
395 | wc->queue[n] = old_wc->queue[tail]; | |
396 | if (tail == (u32) cq->ibcq.cqe) | |
397 | tail = 0; | |
cef1cce5 | 398 | else |
373d9915 | 399 | tail++; |
cef1cce5 BS |
400 | } |
401 | cq->ibcq.cqe = cqe; | |
373d9915 RC |
402 | wc->head = n; |
403 | wc->tail = 0; | |
cef1cce5 BS |
404 | cq->queue = wc; |
405 | spin_unlock_irq(&cq->lock); | |
406 | ||
407 | vfree(old_wc); | |
408 | ||
373d9915 RC |
409 | if (cq->ip) { |
410 | struct ipath_ibdev *dev = to_idev(ibcq->device); | |
411 | struct ipath_mmap_info *ip = cq->ip; | |
412 | ||
413 | ip->obj = wc; | |
414 | ip->size = PAGE_ALIGN(sizeof(*wc) + | |
415 | sizeof(struct ib_wc) * cqe); | |
416 | spin_lock_irq(&dev->pending_lock); | |
417 | ip->next = dev->pending_mmaps; | |
418 | dev->pending_mmaps = ip; | |
419 | spin_unlock_irq(&dev->pending_lock); | |
420 | } | |
421 | ||
cef1cce5 BS |
422 | ret = 0; |
423 | ||
424 | bail: | |
425 | return ret; | |
426 | } |