1 /*
2 * Intel MIC Platform Software Stack (MPSS)
3 *
4 * Copyright(c) 2014 Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2, as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
14 *
15 * Intel SCIF driver.
16 *
17 */
18 #include "../bus/scif_bus.h"
19 #include "scif_peer_bus.h"
20 #include "scif_main.h"
21 #include "scif_nodeqp.h"
22 #include "scif_map.h"
23
24 /*
25 ************************************************************************
26 * SCIF node Queue Pair (QP) setup flow:
27 *
28 * 1) SCIF driver gets probed with a scif_hw_dev via the scif_hw_bus
29 * 2) scif_setup_qp(..) allocates the local qp and calls
30 * scif_setup_qp_connect(..) which allocates and maps the local
31 * buffer for the inbound QP
32 * 3) The local node updates the device page with the DMA address of the QP
33 * 4) A delayed work is scheduled (qp_dwork) which periodically reads if
34 * the peer node has updated its QP DMA address
35 * 5) Once a valid non zero address is found in the QP DMA address field
36 * in the device page, the local node maps the remote node's QP,
37 * updates its outbound QP and sends a SCIF_INIT message to the peer
38 * 6) The SCIF_INIT message is received by the peer node QP interrupt bottom
39 * half handler by calling scif_init(..)
40 * 7) scif_init(..) registers a new SCIF peer node by calling
41 * scif_peer_register_device(..) which signifies the addition of a new
42 * SCIF node
43 * 8) On the mgmt node, P2P network setup/teardown is initiated if all the
44 * remote nodes are online via scif_p2p_setup(..)
45 * 9) For P2P setup, the host maps the remote nodes' aperture and memory
46 * bars and sends a SCIF_NODE_ADD message to both nodes
47 * 10) As part of scif_nodeadd, both nodes set up their local inbound
48 * QPs and send a SCIF_NODE_ADD_ACK to the mgmt node
49 * 11) As part of scif_node_add_ack(..) the mgmt node forwards the
50 * SCIF_NODE_ADD_ACK to the remote nodes
51 * 12) As part of scif_node_add_ack(..) the remote nodes update their
52 * outbound QPs, make sure they can access memory on the remote node
53 * and then add a new SCIF peer node by calling
54 * scif_peer_register_device(..) which signifies the addition of a new
55 * SCIF node.
56 * 13) The SCIF network is now established across all nodes.
57 *
58 ************************************************************************
59 * SCIF node QP teardown flow (initiated by non mgmt node):
60 *
61 * 1) SCIF driver gets a remove callback with a scif_hw_dev via the scif_hw_bus
62 * 2) The device page QP DMA address field is updated with 0x0
63 * 3) A non mgmt node now cleans up all local data structures and sends a
64 * SCIF_EXIT message to the peer and waits for a SCIF_EXIT_ACK
65 * 4) As part of scif_exit(..) handling scif_disconnect_node(..) is called
66 * 5) scif_disconnect_node(..) sends a SCIF_NODE_REMOVE message to all the
67 * peers and waits for a SCIF_NODE_REMOVE_ACK
68 * 6) As part of scif_node_remove(..) a remote node unregisters the peer
69 * node from the SCIF network and sends a SCIF_NODE_REMOVE_ACK
70 * 7) When the mgmt node has received all the SCIF_NODE_REMOVE_ACKs
71 * it sends itself a node remove message whose handling cleans up local
72 * data structures and unregisters the peer node from the SCIF network
73 * 8) The mgmt node sends a SCIF_EXIT_ACK
74 * 9) Upon receipt of the SCIF_EXIT_ACK the node initiating the teardown
75 * completes the SCIF remove routine
76 * 10) The SCIF network is now torn down for the node initiating the
77 * teardown sequence
78 *
79 ************************************************************************
80 * SCIF node QP teardown flow (initiated by mgmt node):
81 *
82 * 1) SCIF driver gets a remove callback with a scif_hw_dev via the scif_hw_bus
83 * 2) The device page QP DMA address field is updated with 0x0
84 * 3) The mgmt node calls scif_disconnect_node(..)
85 * 4) scif_disconnect_node(..) sends a SCIF_NODE_REMOVE message to all the peers
86 * and waits for a SCIF_NODE_REMOVE_ACK
87 * 5) As part of scif_node_remove(..) a remote node unregisters the peer
88 * node from the SCIF network and sends a SCIF_NODE_REMOVE_ACK
89 * 6) When the mgmt node has received all the SCIF_NODE_REMOVE_ACKs
90 * it unregisters the peer node from the SCIF network
91 * 7) The mgmt node sends a SCIF_EXIT message and waits for a SCIF_EXIT_ACK.
92 * 8) A non mgmt node upon receipt of a SCIF_EXIT message calls scif_stop(..)
93 * which would clean up local data structures for all SCIF nodes and
94 * then send a SCIF_EXIT_ACK back to the mgmt node
95 * 9) Upon receipt of the SCIF_EXIT_ACK the the mgmt node sends itself a node
96 * remove message whose handling cleans up local data structures and
97 * destroys any P2P mappings.
98 * 10) The SCIF hardware device for which a remove callback was received is now
99 * disconnected from the SCIF network.
100 */
101 /*
102 * Initializes "local" data structures for the QP. Allocates the QP
103 * ring buffer (rb) and initializes the "in bound" queue.
104 */
scif_setup_qp_connect(struct scif_qp * qp,dma_addr_t * qp_offset,int local_size,struct scif_dev * scifdev)105 int scif_setup_qp_connect(struct scif_qp *qp, dma_addr_t *qp_offset,
106 int local_size, struct scif_dev *scifdev)
107 {
108 void *local_q = qp->inbound_q.rb_base;
109 int err = 0;
110 u32 tmp_rd = 0;
111
112 spin_lock_init(&qp->send_lock);
113 spin_lock_init(&qp->recv_lock);
114
115 /* Allocate rb only if not already allocated */
116 if (!local_q) {
117 local_q = kzalloc(local_size, GFP_KERNEL);
118 if (!local_q) {
119 err = -ENOMEM;
120 return err;
121 }
122 }
123
124 err = scif_map_single(&qp->local_buf, local_q, scifdev, local_size);
125 if (err)
126 goto kfree;
127 /*
128 * To setup the inbound_q, the buffer lives locally, the read pointer
129 * is remote and the write pointer is local.
130 */
131 scif_rb_init(&qp->inbound_q,
132 &tmp_rd,
133 &qp->local_write,
134 local_q, get_count_order(local_size));
135 /*
136 * The read pointer is NULL initially and it is unsafe to use the ring
137 * buffer til this changes!
138 */
139 qp->inbound_q.read_ptr = NULL;
140 err = scif_map_single(qp_offset, qp,
141 scifdev, sizeof(struct scif_qp));
142 if (err)
143 goto unmap;
144 qp->local_qp = *qp_offset;
145 return err;
146 unmap:
147 scif_unmap_single(qp->local_buf, scifdev, local_size);
148 qp->local_buf = 0;
149 kfree:
150 kfree(local_q);
151 return err;
152 }
153
154 /* When the other side has already done it's allocation, this is called */
scif_setup_qp_accept(struct scif_qp * qp,dma_addr_t * qp_offset,dma_addr_t phys,int local_size,struct scif_dev * scifdev)155 int scif_setup_qp_accept(struct scif_qp *qp, dma_addr_t *qp_offset,
156 dma_addr_t phys, int local_size,
157 struct scif_dev *scifdev)
158 {
159 void *local_q;
160 void *remote_q;
161 struct scif_qp *remote_qp;
162 int remote_size;
163 int err = 0;
164
165 spin_lock_init(&qp->send_lock);
166 spin_lock_init(&qp->recv_lock);
167 /* Start by figuring out where we need to point */
168 remote_qp = scif_ioremap(phys, sizeof(struct scif_qp), scifdev);
169 if (!remote_qp)
170 return -EIO;
171 qp->remote_qp = remote_qp;
172 if (qp->remote_qp->magic != SCIFEP_MAGIC) {
173 err = -EIO;
174 goto iounmap;
175 }
176 qp->remote_buf = remote_qp->local_buf;
177 remote_size = qp->remote_qp->inbound_q.size;
178 remote_q = scif_ioremap(qp->remote_buf, remote_size, scifdev);
179 if (!remote_q) {
180 err = -EIO;
181 goto iounmap;
182 }
183 qp->remote_qp->local_write = 0;
184 /*
185 * To setup the outbound_q, the buffer lives in remote memory,
186 * the read pointer is local, the write pointer is remote
187 */
188 scif_rb_init(&qp->outbound_q,
189 &qp->local_read,
190 &qp->remote_qp->local_write,
191 remote_q,
192 get_count_order(remote_size));
193 local_q = kzalloc(local_size, GFP_KERNEL);
194 if (!local_q) {
195 err = -ENOMEM;
196 goto iounmap_1;
197 }
198 err = scif_map_single(&qp->local_buf, local_q, scifdev, local_size);
199 if (err)
200 goto kfree;
201 qp->remote_qp->local_read = 0;
202 /*
203 * To setup the inbound_q, the buffer lives locally, the read pointer
204 * is remote and the write pointer is local
205 */
206 scif_rb_init(&qp->inbound_q,
207 &qp->remote_qp->local_read,
208 &qp->local_write,
209 local_q, get_count_order(local_size));
210 err = scif_map_single(qp_offset, qp, scifdev,
211 sizeof(struct scif_qp));
212 if (err)
213 goto unmap;
214 qp->local_qp = *qp_offset;
215 return err;
216 unmap:
217 scif_unmap_single(qp->local_buf, scifdev, local_size);
218 qp->local_buf = 0;
219 kfree:
220 kfree(local_q);
221 iounmap_1:
222 scif_iounmap(remote_q, remote_size, scifdev);
223 qp->outbound_q.rb_base = NULL;
224 iounmap:
225 scif_iounmap(qp->remote_qp, sizeof(struct scif_qp), scifdev);
226 qp->remote_qp = NULL;
227 return err;
228 }
229
scif_setup_qp_connect_response(struct scif_dev * scifdev,struct scif_qp * qp,u64 payload)230 int scif_setup_qp_connect_response(struct scif_dev *scifdev,
231 struct scif_qp *qp, u64 payload)
232 {
233 int err = 0;
234 void *r_buf;
235 int remote_size;
236 phys_addr_t tmp_phys;
237
238 qp->remote_qp = scif_ioremap(payload, sizeof(struct scif_qp), scifdev);
239
240 if (!qp->remote_qp) {
241 err = -ENOMEM;
242 goto error;
243 }
244
245 if (qp->remote_qp->magic != SCIFEP_MAGIC) {
246 dev_err(&scifdev->sdev->dev,
247 "SCIFEP_MAGIC mismatch between self %d remote %d\n",
248 scif_dev[scif_info.nodeid].node, scifdev->node);
249 err = -ENODEV;
250 goto error;
251 }
252
253 tmp_phys = qp->remote_qp->local_buf;
254 remote_size = qp->remote_qp->inbound_q.size;
255 r_buf = scif_ioremap(tmp_phys, remote_size, scifdev);
256
257 if (!r_buf)
258 return -EIO;
259
260 qp->local_read = 0;
261 scif_rb_init(&qp->outbound_q,
262 &qp->local_read,
263 &qp->remote_qp->local_write,
264 r_buf,
265 get_count_order(remote_size));
266 /*
267 * Because the node QP may already be processing an INIT message, set
268 * the read pointer so the cached read offset isn't lost
269 */
270 qp->remote_qp->local_read = qp->inbound_q.current_read_offset;
271 /*
272 * resetup the inbound_q now that we know where the
273 * inbound_read really is.
274 */
275 scif_rb_init(&qp->inbound_q,
276 &qp->remote_qp->local_read,
277 &qp->local_write,
278 qp->inbound_q.rb_base,
279 get_count_order(qp->inbound_q.size));
280 error:
281 return err;
282 }
283
284 static __always_inline void
scif_send_msg_intr(struct scif_dev * scifdev)285 scif_send_msg_intr(struct scif_dev *scifdev)
286 {
287 struct scif_hw_dev *sdev = scifdev->sdev;
288
289 if (scifdev_is_p2p(scifdev))
290 sdev->hw_ops->send_p2p_intr(sdev, scifdev->rdb, &scifdev->mmio);
291 else
292 sdev->hw_ops->send_intr(sdev, scifdev->rdb);
293 }
294
scif_qp_response(phys_addr_t phys,struct scif_dev * scifdev)295 int scif_qp_response(phys_addr_t phys, struct scif_dev *scifdev)
296 {
297 int err = 0;
298 struct scifmsg msg;
299
300 err = scif_setup_qp_connect_response(scifdev, scifdev->qpairs, phys);
301 if (!err) {
302 /*
303 * Now that everything is setup and mapped, we're ready
304 * to tell the peer about our queue's location
305 */
306 msg.uop = SCIF_INIT;
307 msg.dst.node = scifdev->node;
308 err = scif_nodeqp_send(scifdev, &msg);
309 }
310 return err;
311 }
312
scif_send_exit(struct scif_dev * scifdev)313 void scif_send_exit(struct scif_dev *scifdev)
314 {
315 struct scifmsg msg;
316 int ret;
317
318 scifdev->exit = OP_IN_PROGRESS;
319 msg.uop = SCIF_EXIT;
320 msg.src.node = scif_info.nodeid;
321 msg.dst.node = scifdev->node;
322 ret = scif_nodeqp_send(scifdev, &msg);
323 if (ret)
324 goto done;
325 /* Wait for a SCIF_EXIT_ACK message */
326 wait_event_timeout(scif_info.exitwq, scifdev->exit == OP_COMPLETED,
327 SCIF_NODE_ALIVE_TIMEOUT);
328 done:
329 scifdev->exit = OP_IDLE;
330 }
331
scif_setup_qp(struct scif_dev * scifdev)332 int scif_setup_qp(struct scif_dev *scifdev)
333 {
334 int err = 0;
335 int local_size;
336 struct scif_qp *qp;
337
338 local_size = SCIF_NODE_QP_SIZE;
339
340 qp = kzalloc(sizeof(*qp), GFP_KERNEL);
341 if (!qp) {
342 err = -ENOMEM;
343 return err;
344 }
345 qp->magic = SCIFEP_MAGIC;
346 scifdev->qpairs = qp;
347 err = scif_setup_qp_connect(qp, &scifdev->qp_dma_addr,
348 local_size, scifdev);
349 if (err)
350 goto free_qp;
351 /*
352 * We're as setup as we can be. The inbound_q is setup, w/o a usable
353 * outbound q. When we get a message, the read_ptr will be updated,
354 * and we will pull the message.
355 */
356 return err;
357 free_qp:
358 kfree(scifdev->qpairs);
359 scifdev->qpairs = NULL;
360 return err;
361 }
362
scif_p2p_freesg(struct scatterlist * sg)363 static void scif_p2p_freesg(struct scatterlist *sg)
364 {
365 kfree(sg);
366 }
367
368 static struct scatterlist *
scif_p2p_setsg(phys_addr_t pa,int page_size,int page_cnt)369 scif_p2p_setsg(phys_addr_t pa, int page_size, int page_cnt)
370 {
371 struct scatterlist *sg;
372 struct page *page;
373 int i;
374
375 sg = kcalloc(page_cnt, sizeof(struct scatterlist), GFP_KERNEL);
376 if (!sg)
377 return NULL;
378 sg_init_table(sg, page_cnt);
379 for (i = 0; i < page_cnt; i++) {
380 page = pfn_to_page(pa >> PAGE_SHIFT);
381 sg_set_page(&sg[i], page, page_size, 0);
382 pa += page_size;
383 }
384 return sg;
385 }
386
387 /* Init p2p mappings required to access peerdev from scifdev */
388 static struct scif_p2p_info *
scif_init_p2p_info(struct scif_dev * scifdev,struct scif_dev * peerdev)389 scif_init_p2p_info(struct scif_dev *scifdev, struct scif_dev *peerdev)
390 {
391 struct scif_p2p_info *p2p;
392 int num_mmio_pages, num_aper_pages, sg_page_shift, err, num_aper_chunks;
393 struct scif_hw_dev *psdev = peerdev->sdev;
394 struct scif_hw_dev *sdev = scifdev->sdev;
395
396 num_mmio_pages = psdev->mmio->len >> PAGE_SHIFT;
397 num_aper_pages = psdev->aper->len >> PAGE_SHIFT;
398
399 p2p = kzalloc(sizeof(*p2p), GFP_KERNEL);
400 if (!p2p)
401 return NULL;
402 p2p->ppi_sg[SCIF_PPI_MMIO] = scif_p2p_setsg(psdev->mmio->pa,
403 PAGE_SIZE, num_mmio_pages);
404 if (!p2p->ppi_sg[SCIF_PPI_MMIO])
405 goto free_p2p;
406 p2p->sg_nentries[SCIF_PPI_MMIO] = num_mmio_pages;
407 sg_page_shift = get_order(min(psdev->aper->len, (u64)(1 << 30)));
408 num_aper_chunks = num_aper_pages >> (sg_page_shift - PAGE_SHIFT);
409 p2p->ppi_sg[SCIF_PPI_APER] = scif_p2p_setsg(psdev->aper->pa,
410 1 << sg_page_shift,
411 num_aper_chunks);
412 p2p->sg_nentries[SCIF_PPI_APER] = num_aper_chunks;
413 err = dma_map_sg(&sdev->dev, p2p->ppi_sg[SCIF_PPI_MMIO],
414 num_mmio_pages, PCI_DMA_BIDIRECTIONAL);
415 if (err != num_mmio_pages)
416 goto scif_p2p_free;
417 err = dma_map_sg(&sdev->dev, p2p->ppi_sg[SCIF_PPI_APER],
418 num_aper_chunks, PCI_DMA_BIDIRECTIONAL);
419 if (err != num_aper_chunks)
420 goto dma_unmap;
421 p2p->ppi_da[SCIF_PPI_MMIO] = sg_dma_address(p2p->ppi_sg[SCIF_PPI_MMIO]);
422 p2p->ppi_da[SCIF_PPI_APER] = sg_dma_address(p2p->ppi_sg[SCIF_PPI_APER]);
423 p2p->ppi_len[SCIF_PPI_MMIO] = num_mmio_pages;
424 p2p->ppi_len[SCIF_PPI_APER] = num_aper_pages;
425 p2p->ppi_peer_id = peerdev->node;
426 return p2p;
427 dma_unmap:
428 dma_unmap_sg(&sdev->dev, p2p->ppi_sg[SCIF_PPI_MMIO],
429 p2p->sg_nentries[SCIF_PPI_MMIO], DMA_BIDIRECTIONAL);
430 scif_p2p_free:
431 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_MMIO]);
432 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_APER]);
433 free_p2p:
434 kfree(p2p);
435 return NULL;
436 }
437
438 /* Uninitialize and release resources from a p2p mapping */
scif_deinit_p2p_info(struct scif_dev * scifdev,struct scif_p2p_info * p2p)439 static void scif_deinit_p2p_info(struct scif_dev *scifdev,
440 struct scif_p2p_info *p2p)
441 {
442 struct scif_hw_dev *sdev = scifdev->sdev;
443
444 dma_unmap_sg(&sdev->dev, p2p->ppi_sg[SCIF_PPI_MMIO],
445 p2p->sg_nentries[SCIF_PPI_MMIO], DMA_BIDIRECTIONAL);
446 dma_unmap_sg(&sdev->dev, p2p->ppi_sg[SCIF_PPI_APER],
447 p2p->sg_nentries[SCIF_PPI_APER], DMA_BIDIRECTIONAL);
448 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_MMIO]);
449 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_APER]);
450 kfree(p2p);
451 }
452
453 /**
454 * scif_node_connect: Respond to SCIF_NODE_CONNECT interrupt message
455 * @dst: Destination node
456 *
457 * Connect the src and dst node by setting up the p2p connection
458 * between them. Management node here acts like a proxy.
459 */
scif_node_connect(struct scif_dev * scifdev,int dst)460 static void scif_node_connect(struct scif_dev *scifdev, int dst)
461 {
462 struct scif_dev *dev_j = scifdev;
463 struct scif_dev *dev_i = NULL;
464 struct scif_p2p_info *p2p_ij = NULL; /* bus addr for j from i */
465 struct scif_p2p_info *p2p_ji = NULL; /* bus addr for i from j */
466 struct scif_p2p_info *p2p;
467 struct list_head *pos, *tmp;
468 struct scifmsg msg;
469 int err;
470 u64 tmppayload;
471
472 if (dst < 1 || dst > scif_info.maxid)
473 return;
474
475 dev_i = &scif_dev[dst];
476
477 if (!_scifdev_alive(dev_i))
478 return;
479 /*
480 * If the p2p connection is already setup or in the process of setting
481 * up then just ignore this request. The requested node will get
482 * informed by SCIF_NODE_ADD_ACK or SCIF_NODE_ADD_NACK
483 */
484 if (!list_empty(&dev_i->p2p)) {
485 list_for_each_safe(pos, tmp, &dev_i->p2p) {
486 p2p = list_entry(pos, struct scif_p2p_info, ppi_list);
487 if (p2p->ppi_peer_id == dev_j->node)
488 return;
489 }
490 }
491 p2p_ij = scif_init_p2p_info(dev_i, dev_j);
492 if (!p2p_ij)
493 return;
494 p2p_ji = scif_init_p2p_info(dev_j, dev_i);
495 if (!p2p_ji) {
496 scif_deinit_p2p_info(dev_i, p2p_ij);
497 return;
498 }
499 list_add_tail(&p2p_ij->ppi_list, &dev_i->p2p);
500 list_add_tail(&p2p_ji->ppi_list, &dev_j->p2p);
501
502 /*
503 * Send a SCIF_NODE_ADD to dev_i, pass it its bus address
504 * as seen from dev_j
505 */
506 msg.uop = SCIF_NODE_ADD;
507 msg.src.node = dev_j->node;
508 msg.dst.node = dev_i->node;
509
510 msg.payload[0] = p2p_ji->ppi_da[SCIF_PPI_APER];
511 msg.payload[1] = p2p_ij->ppi_da[SCIF_PPI_MMIO];
512 msg.payload[2] = p2p_ij->ppi_da[SCIF_PPI_APER];
513 msg.payload[3] = p2p_ij->ppi_len[SCIF_PPI_APER] << PAGE_SHIFT;
514
515 err = scif_nodeqp_send(dev_i, &msg);
516 if (err) {
517 dev_err(&scifdev->sdev->dev,
518 "%s %d error %d\n", __func__, __LINE__, err);
519 return;
520 }
521
522 /* Same as above but to dev_j */
523 msg.uop = SCIF_NODE_ADD;
524 msg.src.node = dev_i->node;
525 msg.dst.node = dev_j->node;
526
527 tmppayload = msg.payload[0];
528 msg.payload[0] = msg.payload[2];
529 msg.payload[2] = tmppayload;
530 msg.payload[1] = p2p_ji->ppi_da[SCIF_PPI_MMIO];
531 msg.payload[3] = p2p_ji->ppi_len[SCIF_PPI_APER] << PAGE_SHIFT;
532
533 scif_nodeqp_send(dev_j, &msg);
534 }
535
scif_p2p_setup(void)536 static void scif_p2p_setup(void)
537 {
538 int i, j;
539
540 if (!scif_info.p2p_enable)
541 return;
542
543 for (i = 1; i <= scif_info.maxid; i++)
544 if (!_scifdev_alive(&scif_dev[i]))
545 return;
546
547 for (i = 1; i <= scif_info.maxid; i++) {
548 for (j = 1; j <= scif_info.maxid; j++) {
549 struct scif_dev *scifdev = &scif_dev[i];
550
551 if (i == j)
552 continue;
553 scif_node_connect(scifdev, j);
554 }
555 }
556 }
557
558 static char *message_types[] = {"BAD",
559 "INIT",
560 "EXIT",
561 "SCIF_EXIT_ACK",
562 "SCIF_NODE_ADD",
563 "SCIF_NODE_ADD_ACK",
564 "SCIF_NODE_ADD_NACK",
565 "REMOVE_NODE",
566 "REMOVE_NODE_ACK",
567 "CNCT_REQ",
568 "CNCT_GNT",
569 "CNCT_GNTACK",
570 "CNCT_GNTNACK",
571 "CNCT_REJ",
572 "DISCNCT",
573 "DISCNT_ACK",
574 "CLIENT_SENT",
575 "CLIENT_RCVD",
576 "SCIF_GET_NODE_INFO",
577 "REGISTER",
578 "REGISTER_ACK",
579 "REGISTER_NACK",
580 "UNREGISTER",
581 "UNREGISTER_ACK",
582 "UNREGISTER_NACK",
583 "ALLOC_REQ",
584 "ALLOC_GNT",
585 "ALLOC_REJ",
586 "FREE_PHYS",
587 "FREE_VIRT",
588 "MUNMAP",
589 "MARK",
590 "MARK_ACK",
591 "MARK_NACK",
592 "WAIT",
593 "WAIT_ACK",
594 "WAIT_NACK",
595 "SIGNAL_LOCAL",
596 "SIGNAL_REMOTE",
597 "SIG_ACK",
598 "SIG_NACK"};
599
600 static void
scif_display_message(struct scif_dev * scifdev,struct scifmsg * msg,const char * label)601 scif_display_message(struct scif_dev *scifdev, struct scifmsg *msg,
602 const char *label)
603 {
604 if (!scif_info.en_msg_log)
605 return;
606 if (msg->uop > SCIF_MAX_MSG) {
607 dev_err(&scifdev->sdev->dev,
608 "%s: unknown msg type %d\n", label, msg->uop);
609 return;
610 }
611 dev_info(&scifdev->sdev->dev,
612 "%s: msg type %s, src %d:%d, dest %d:%d payload 0x%llx:0x%llx:0x%llx:0x%llx\n",
613 label, message_types[msg->uop], msg->src.node, msg->src.port,
614 msg->dst.node, msg->dst.port, msg->payload[0], msg->payload[1],
615 msg->payload[2], msg->payload[3]);
616 }
617
_scif_nodeqp_send(struct scif_dev * scifdev,struct scifmsg * msg)618 int _scif_nodeqp_send(struct scif_dev *scifdev, struct scifmsg *msg)
619 {
620 struct scif_qp *qp = scifdev->qpairs;
621 int err = -ENOMEM, loop_cnt = 0;
622
623 scif_display_message(scifdev, msg, "Sent");
624 if (!qp) {
625 err = -EINVAL;
626 goto error;
627 }
628 spin_lock(&qp->send_lock);
629
630 while ((err = scif_rb_write(&qp->outbound_q,
631 msg, sizeof(struct scifmsg)))) {
632 mdelay(1);
633 #define SCIF_NODEQP_SEND_TO_MSEC (3 * 1000)
634 if (loop_cnt++ > (SCIF_NODEQP_SEND_TO_MSEC)) {
635 err = -ENODEV;
636 break;
637 }
638 }
639 if (!err)
640 scif_rb_commit(&qp->outbound_q);
641 spin_unlock(&qp->send_lock);
642 if (!err) {
643 if (scifdev_self(scifdev))
644 /*
645 * For loopback we need to emulate an interrupt by
646 * queuing work for the queue handling real node
647 * Qp interrupts.
648 */
649 queue_work(scifdev->intr_wq, &scifdev->intr_bh);
650 else
651 scif_send_msg_intr(scifdev);
652 }
653 error:
654 if (err)
655 dev_dbg(&scifdev->sdev->dev,
656 "%s %d error %d uop %d\n",
657 __func__, __LINE__, err, msg->uop);
658 return err;
659 }
660
661 /**
662 * scif_nodeqp_send - Send a message on the node queue pair
663 * @scifdev: Scif Device.
664 * @msg: The message to be sent.
665 */
scif_nodeqp_send(struct scif_dev * scifdev,struct scifmsg * msg)666 int scif_nodeqp_send(struct scif_dev *scifdev, struct scifmsg *msg)
667 {
668 int err;
669 struct device *spdev = NULL;
670
671 if (msg->uop > SCIF_EXIT_ACK) {
672 /* Dont send messages once the exit flow has begun */
673 if (OP_IDLE != scifdev->exit)
674 return -ENODEV;
675 spdev = scif_get_peer_dev(scifdev);
676 if (IS_ERR(spdev)) {
677 err = PTR_ERR(spdev);
678 return err;
679 }
680 }
681 err = _scif_nodeqp_send(scifdev, msg);
682 if (msg->uop > SCIF_EXIT_ACK)
683 scif_put_peer_dev(spdev);
684 return err;
685 }
686
687 /*
688 * scif_misc_handler:
689 *
690 * Work queue handler for servicing miscellaneous SCIF tasks.
691 * Examples include:
692 * 1) Remote fence requests.
693 * 2) Destruction of temporary registered windows
694 * created during scif_vreadfrom()/scif_vwriteto().
695 * 3) Cleanup of zombie endpoints.
696 */
scif_misc_handler(struct work_struct * work)697 void scif_misc_handler(struct work_struct *work)
698 {
699 scif_rma_handle_remote_fences();
700 scif_rma_destroy_windows();
701 scif_rma_destroy_tcw_invalid();
702 scif_cleanup_zombie_epd();
703 }
704
705 /**
706 * scif_init() - Respond to SCIF_INIT interrupt message
707 * @scifdev: Remote SCIF device node
708 * @msg: Interrupt message
709 */
710 static __always_inline void
scif_init(struct scif_dev * scifdev,struct scifmsg * msg)711 scif_init(struct scif_dev *scifdev, struct scifmsg *msg)
712 {
713 /*
714 * Allow the thread waiting for device page updates for the peer QP DMA
715 * address to complete initializing the inbound_q.
716 */
717 flush_delayed_work(&scifdev->qp_dwork);
718
719 scif_peer_register_device(scifdev);
720
721 if (scif_is_mgmt_node()) {
722 mutex_lock(&scif_info.conflock);
723 scif_p2p_setup();
724 mutex_unlock(&scif_info.conflock);
725 }
726 }
727
728 /**
729 * scif_exit() - Respond to SCIF_EXIT interrupt message
730 * @scifdev: Remote SCIF device node
731 * @msg: Interrupt message
732 *
733 * This function stops the SCIF interface for the node which sent
734 * the SCIF_EXIT message and starts waiting for that node to
735 * resetup the queue pair again.
736 */
737 static __always_inline void
scif_exit(struct scif_dev * scifdev,struct scifmsg * unused)738 scif_exit(struct scif_dev *scifdev, struct scifmsg *unused)
739 {
740 scifdev->exit_ack_pending = true;
741 if (scif_is_mgmt_node())
742 scif_disconnect_node(scifdev->node, false);
743 else
744 scif_stop(scifdev);
745 schedule_delayed_work(&scifdev->qp_dwork,
746 msecs_to_jiffies(1000));
747 }
748
749 /**
750 * scif_exitack() - Respond to SCIF_EXIT_ACK interrupt message
751 * @scifdev: Remote SCIF device node
752 * @msg: Interrupt message
753 *
754 */
755 static __always_inline void
scif_exit_ack(struct scif_dev * scifdev,struct scifmsg * unused)756 scif_exit_ack(struct scif_dev *scifdev, struct scifmsg *unused)
757 {
758 scifdev->exit = OP_COMPLETED;
759 wake_up(&scif_info.exitwq);
760 }
761
762 /**
763 * scif_node_add() - Respond to SCIF_NODE_ADD interrupt message
764 * @scifdev: Remote SCIF device node
765 * @msg: Interrupt message
766 *
767 * When the mgmt node driver has finished initializing a MIC node queue pair it
768 * marks the node as online. It then looks for all currently online MIC cards
769 * and send a SCIF_NODE_ADD message to identify the ID of the new card for
770 * peer to peer initialization
771 *
772 * The local node allocates its incoming queue and sends its address in the
773 * SCIF_NODE_ADD_ACK message back to the mgmt node, the mgmt node "reflects"
774 * this message to the new node
775 */
776 static __always_inline void
scif_node_add(struct scif_dev * scifdev,struct scifmsg * msg)777 scif_node_add(struct scif_dev *scifdev, struct scifmsg *msg)
778 {
779 struct scif_dev *newdev;
780 dma_addr_t qp_offset;
781 int qp_connect;
782 struct scif_hw_dev *sdev;
783
784 dev_dbg(&scifdev->sdev->dev,
785 "Scifdev %d:%d received NODE_ADD msg for node %d\n",
786 scifdev->node, msg->dst.node, msg->src.node);
787 dev_dbg(&scifdev->sdev->dev,
788 "Remote address for this node's aperture %llx\n",
789 msg->payload[0]);
790 newdev = &scif_dev[msg->src.node];
791 newdev->node = msg->src.node;
792 newdev->sdev = scif_dev[SCIF_MGMT_NODE].sdev;
793 sdev = newdev->sdev;
794
795 if (scif_setup_intr_wq(newdev)) {
796 dev_err(&scifdev->sdev->dev,
797 "failed to setup interrupts for %d\n", msg->src.node);
798 goto interrupt_setup_error;
799 }
800 newdev->mmio.va = ioremap_nocache(msg->payload[1], sdev->mmio->len);
801 if (!newdev->mmio.va) {
802 dev_err(&scifdev->sdev->dev,
803 "failed to map mmio for %d\n", msg->src.node);
804 goto mmio_map_error;
805 }
806 newdev->qpairs = kzalloc(sizeof(*newdev->qpairs), GFP_KERNEL);
807 if (!newdev->qpairs)
808 goto qp_alloc_error;
809 /*
810 * Set the base address of the remote node's memory since it gets
811 * added to qp_offset
812 */
813 newdev->base_addr = msg->payload[0];
814
815 qp_connect = scif_setup_qp_connect(newdev->qpairs, &qp_offset,
816 SCIF_NODE_QP_SIZE, newdev);
817 if (qp_connect) {
818 dev_err(&scifdev->sdev->dev,
819 "failed to setup qp_connect %d\n", qp_connect);
820 goto qp_connect_error;
821 }
822
823 newdev->db = sdev->hw_ops->next_db(sdev);
824 newdev->cookie = sdev->hw_ops->request_irq(sdev, scif_intr_handler,
825 "SCIF_INTR", newdev,
826 newdev->db);
827 if (IS_ERR(newdev->cookie))
828 goto qp_connect_error;
829 newdev->qpairs->magic = SCIFEP_MAGIC;
830 newdev->qpairs->qp_state = SCIF_QP_OFFLINE;
831
832 msg->uop = SCIF_NODE_ADD_ACK;
833 msg->dst.node = msg->src.node;
834 msg->src.node = scif_info.nodeid;
835 msg->payload[0] = qp_offset;
836 msg->payload[2] = newdev->db;
837 scif_nodeqp_send(&scif_dev[SCIF_MGMT_NODE], msg);
838 return;
839 qp_connect_error:
840 kfree(newdev->qpairs);
841 newdev->qpairs = NULL;
842 qp_alloc_error:
843 iounmap(newdev->mmio.va);
844 newdev->mmio.va = NULL;
845 mmio_map_error:
846 interrupt_setup_error:
847 dev_err(&scifdev->sdev->dev,
848 "node add failed for node %d\n", msg->src.node);
849 msg->uop = SCIF_NODE_ADD_NACK;
850 msg->dst.node = msg->src.node;
851 msg->src.node = scif_info.nodeid;
852 scif_nodeqp_send(&scif_dev[SCIF_MGMT_NODE], msg);
853 }
854
scif_poll_qp_state(struct work_struct * work)855 void scif_poll_qp_state(struct work_struct *work)
856 {
857 #define SCIF_NODE_QP_RETRY 100
858 #define SCIF_NODE_QP_TIMEOUT 100
859 struct scif_dev *peerdev = container_of(work, struct scif_dev,
860 p2p_dwork.work);
861 struct scif_qp *qp = &peerdev->qpairs[0];
862
863 if (qp->qp_state != SCIF_QP_ONLINE ||
864 qp->remote_qp->qp_state != SCIF_QP_ONLINE) {
865 if (peerdev->p2p_retry++ == SCIF_NODE_QP_RETRY) {
866 dev_err(&peerdev->sdev->dev,
867 "Warning: QP check timeout with state %d\n",
868 qp->qp_state);
869 goto timeout;
870 }
871 schedule_delayed_work(&peerdev->p2p_dwork,
872 msecs_to_jiffies(SCIF_NODE_QP_TIMEOUT));
873 return;
874 }
875 return;
876 timeout:
877 dev_err(&peerdev->sdev->dev,
878 "%s %d remote node %d offline, state = 0x%x\n",
879 __func__, __LINE__, peerdev->node, qp->qp_state);
880 qp->remote_qp->qp_state = SCIF_QP_OFFLINE;
881 scif_peer_unregister_device(peerdev);
882 scif_cleanup_scifdev(peerdev);
883 }
884
885 /**
886 * scif_node_add_ack() - Respond to SCIF_NODE_ADD_ACK interrupt message
887 * @scifdev: Remote SCIF device node
888 * @msg: Interrupt message
889 *
890 * After a MIC node receives the SCIF_NODE_ADD_ACK message it send this
891 * message to the mgmt node to confirm the sequence is finished.
892 *
893 */
894 static __always_inline void
scif_node_add_ack(struct scif_dev * scifdev,struct scifmsg * msg)895 scif_node_add_ack(struct scif_dev *scifdev, struct scifmsg *msg)
896 {
897 struct scif_dev *peerdev;
898 struct scif_qp *qp;
899 struct scif_dev *dst_dev = &scif_dev[msg->dst.node];
900
901 dev_dbg(&scifdev->sdev->dev,
902 "Scifdev %d received SCIF_NODE_ADD_ACK msg src %d dst %d\n",
903 scifdev->node, msg->src.node, msg->dst.node);
904 dev_dbg(&scifdev->sdev->dev,
905 "payload %llx %llx %llx %llx\n", msg->payload[0],
906 msg->payload[1], msg->payload[2], msg->payload[3]);
907 if (scif_is_mgmt_node()) {
908 /*
909 * the lock serializes with scif_qp_response_ack. The mgmt node
910 * is forwarding the NODE_ADD_ACK message from src to dst we
911 * need to make sure that the dst has already received a
912 * NODE_ADD for src and setup its end of the qp to dst
913 */
914 mutex_lock(&scif_info.conflock);
915 msg->payload[1] = scif_info.maxid;
916 scif_nodeqp_send(dst_dev, msg);
917 mutex_unlock(&scif_info.conflock);
918 return;
919 }
920 peerdev = &scif_dev[msg->src.node];
921 peerdev->sdev = scif_dev[SCIF_MGMT_NODE].sdev;
922 peerdev->node = msg->src.node;
923
924 qp = &peerdev->qpairs[0];
925
926 if ((scif_setup_qp_connect_response(peerdev, &peerdev->qpairs[0],
927 msg->payload[0])))
928 goto local_error;
929 peerdev->rdb = msg->payload[2];
930 qp->remote_qp->qp_state = SCIF_QP_ONLINE;
931
932 scif_peer_register_device(peerdev);
933
934 schedule_delayed_work(&peerdev->p2p_dwork, 0);
935 return;
936 local_error:
937 scif_cleanup_scifdev(peerdev);
938 }
939
940 /**
941 * scif_node_add_nack: Respond to SCIF_NODE_ADD_NACK interrupt message
942 * @msg: Interrupt message
943 *
944 * SCIF_NODE_ADD failed, so inform the waiting wq.
945 */
946 static __always_inline void
scif_node_add_nack(struct scif_dev * scifdev,struct scifmsg * msg)947 scif_node_add_nack(struct scif_dev *scifdev, struct scifmsg *msg)
948 {
949 if (scif_is_mgmt_node()) {
950 struct scif_dev *dst_dev = &scif_dev[msg->dst.node];
951
952 dev_dbg(&scifdev->sdev->dev,
953 "SCIF_NODE_ADD_NACK received from %d\n", scifdev->node);
954 scif_nodeqp_send(dst_dev, msg);
955 }
956 }
957
958 /*
959 * scif_node_remove: Handle SCIF_NODE_REMOVE message
960 * @msg: Interrupt message
961 *
962 * Handle node removal.
963 */
964 static __always_inline void
scif_node_remove(struct scif_dev * scifdev,struct scifmsg * msg)965 scif_node_remove(struct scif_dev *scifdev, struct scifmsg *msg)
966 {
967 int node = msg->payload[0];
968 struct scif_dev *scdev = &scif_dev[node];
969
970 scdev->node_remove_ack_pending = true;
971 scif_handle_remove_node(node);
972 }
973
974 /*
975 * scif_node_remove_ack: Handle SCIF_NODE_REMOVE_ACK message
976 * @msg: Interrupt message
977 *
978 * The peer has acked a SCIF_NODE_REMOVE message.
979 */
980 static __always_inline void
scif_node_remove_ack(struct scif_dev * scifdev,struct scifmsg * msg)981 scif_node_remove_ack(struct scif_dev *scifdev, struct scifmsg *msg)
982 {
983 struct scif_dev *sdev = &scif_dev[msg->payload[0]];
984
985 atomic_inc(&sdev->disconn_rescnt);
986 wake_up(&sdev->disconn_wq);
987 }
988
989 /**
990 * scif_get_node_info: Respond to SCIF_GET_NODE_INFO interrupt message
991 * @msg: Interrupt message
992 *
993 * Retrieve node info i.e maxid and total from the mgmt node.
994 */
995 static __always_inline void
scif_get_node_info_resp(struct scif_dev * scifdev,struct scifmsg * msg)996 scif_get_node_info_resp(struct scif_dev *scifdev, struct scifmsg *msg)
997 {
998 if (scif_is_mgmt_node()) {
999 swap(msg->dst.node, msg->src.node);
1000 mutex_lock(&scif_info.conflock);
1001 msg->payload[1] = scif_info.maxid;
1002 msg->payload[2] = scif_info.total;
1003 mutex_unlock(&scif_info.conflock);
1004 scif_nodeqp_send(scifdev, msg);
1005 } else {
1006 struct completion *node_info =
1007 (struct completion *)msg->payload[3];
1008
1009 mutex_lock(&scif_info.conflock);
1010 scif_info.maxid = msg->payload[1];
1011 scif_info.total = msg->payload[2];
1012 complete_all(node_info);
1013 mutex_unlock(&scif_info.conflock);
1014 }
1015 }
1016
1017 static void
scif_msg_unknown(struct scif_dev * scifdev,struct scifmsg * msg)1018 scif_msg_unknown(struct scif_dev *scifdev, struct scifmsg *msg)
1019 {
1020 /* Bogus Node Qp Message? */
1021 dev_err(&scifdev->sdev->dev,
1022 "Unknown message 0x%xn scifdev->node 0x%x\n",
1023 msg->uop, scifdev->node);
1024 }
1025
1026 static void (*scif_intr_func[SCIF_MAX_MSG + 1])
1027 (struct scif_dev *, struct scifmsg *msg) = {
1028 scif_msg_unknown, /* Error */
1029 scif_init, /* SCIF_INIT */
1030 scif_exit, /* SCIF_EXIT */
1031 scif_exit_ack, /* SCIF_EXIT_ACK */
1032 scif_node_add, /* SCIF_NODE_ADD */
1033 scif_node_add_ack, /* SCIF_NODE_ADD_ACK */
1034 scif_node_add_nack, /* SCIF_NODE_ADD_NACK */
1035 scif_node_remove, /* SCIF_NODE_REMOVE */
1036 scif_node_remove_ack, /* SCIF_NODE_REMOVE_ACK */
1037 scif_cnctreq, /* SCIF_CNCT_REQ */
1038 scif_cnctgnt, /* SCIF_CNCT_GNT */
1039 scif_cnctgnt_ack, /* SCIF_CNCT_GNTACK */
1040 scif_cnctgnt_nack, /* SCIF_CNCT_GNTNACK */
1041 scif_cnctrej, /* SCIF_CNCT_REJ */
1042 scif_discnct, /* SCIF_DISCNCT */
1043 scif_discnt_ack, /* SCIF_DISCNT_ACK */
1044 scif_clientsend, /* SCIF_CLIENT_SENT */
1045 scif_clientrcvd, /* SCIF_CLIENT_RCVD */
1046 scif_get_node_info_resp,/* SCIF_GET_NODE_INFO */
1047 scif_recv_reg, /* SCIF_REGISTER */
1048 scif_recv_reg_ack, /* SCIF_REGISTER_ACK */
1049 scif_recv_reg_nack, /* SCIF_REGISTER_NACK */
1050 scif_recv_unreg, /* SCIF_UNREGISTER */
1051 scif_recv_unreg_ack, /* SCIF_UNREGISTER_ACK */
1052 scif_recv_unreg_nack, /* SCIF_UNREGISTER_NACK */
1053 scif_alloc_req, /* SCIF_ALLOC_REQ */
1054 scif_alloc_gnt_rej, /* SCIF_ALLOC_GNT */
1055 scif_alloc_gnt_rej, /* SCIF_ALLOC_REJ */
1056 scif_free_virt, /* SCIF_FREE_VIRT */
1057 scif_recv_munmap, /* SCIF_MUNMAP */
1058 scif_recv_mark, /* SCIF_MARK */
1059 scif_recv_mark_resp, /* SCIF_MARK_ACK */
1060 scif_recv_mark_resp, /* SCIF_MARK_NACK */
1061 scif_recv_wait, /* SCIF_WAIT */
1062 scif_recv_wait_resp, /* SCIF_WAIT_ACK */
1063 scif_recv_wait_resp, /* SCIF_WAIT_NACK */
1064 scif_recv_sig_local, /* SCIF_SIG_LOCAL */
1065 scif_recv_sig_remote, /* SCIF_SIG_REMOTE */
1066 scif_recv_sig_resp, /* SCIF_SIG_ACK */
1067 scif_recv_sig_resp, /* SCIF_SIG_NACK */
1068 };
1069
1070 /**
1071 * scif_nodeqp_msg_handler() - Common handler for node messages
1072 * @scifdev: Remote device to respond to
1073 * @qp: Remote memory pointer
1074 * @msg: The message to be handled.
1075 *
1076 * This routine calls the appropriate routine to handle a Node Qp
1077 * message receipt
1078 */
1079 static int scif_max_msg_id = SCIF_MAX_MSG;
1080
1081 static void
scif_nodeqp_msg_handler(struct scif_dev * scifdev,struct scif_qp * qp,struct scifmsg * msg)1082 scif_nodeqp_msg_handler(struct scif_dev *scifdev,
1083 struct scif_qp *qp, struct scifmsg *msg)
1084 {
1085 scif_display_message(scifdev, msg, "Rcvd");
1086
1087 if (msg->uop > (u32)scif_max_msg_id) {
1088 /* Bogus Node Qp Message? */
1089 dev_err(&scifdev->sdev->dev,
1090 "Unknown message 0x%xn scifdev->node 0x%x\n",
1091 msg->uop, scifdev->node);
1092 return;
1093 }
1094
1095 scif_intr_func[msg->uop](scifdev, msg);
1096 }
1097
1098 /**
1099 * scif_nodeqp_intrhandler() - Interrupt handler for node messages
1100 * @scifdev: Remote device to respond to
1101 * @qp: Remote memory pointer
1102 *
1103 * This routine is triggered by the interrupt mechanism. It reads
1104 * messages from the node queue RB and calls the Node QP Message handling
1105 * routine.
1106 */
scif_nodeqp_intrhandler(struct scif_dev * scifdev,struct scif_qp * qp)1107 void scif_nodeqp_intrhandler(struct scif_dev *scifdev, struct scif_qp *qp)
1108 {
1109 struct scifmsg msg;
1110 int read_size;
1111
1112 do {
1113 read_size = scif_rb_get_next(&qp->inbound_q, &msg, sizeof(msg));
1114 if (!read_size)
1115 break;
1116 scif_nodeqp_msg_handler(scifdev, qp, &msg);
1117 /*
1118 * The node queue pair is unmapped so skip the read pointer
1119 * update after receipt of a SCIF_EXIT_ACK
1120 */
1121 if (SCIF_EXIT_ACK == msg.uop)
1122 break;
1123 scif_rb_update_read_ptr(&qp->inbound_q);
1124 } while (1);
1125 }
1126
1127 /**
1128 * scif_loopb_wq_handler - Loopback Workqueue Handler.
1129 * @work: loop back work
1130 *
1131 * This work queue routine is invoked by the loopback work queue handler.
1132 * It grabs the recv lock, dequeues any available messages from the head
1133 * of the loopback message list, calls the node QP message handler,
1134 * waits for it to return, then frees up this message and dequeues more
1135 * elements of the list if available.
1136 */
scif_loopb_wq_handler(struct work_struct * unused)1137 static void scif_loopb_wq_handler(struct work_struct *unused)
1138 {
1139 struct scif_dev *scifdev = scif_info.loopb_dev;
1140 struct scif_qp *qp = scifdev->qpairs;
1141 struct scif_loopb_msg *msg;
1142
1143 do {
1144 msg = NULL;
1145 spin_lock(&qp->recv_lock);
1146 if (!list_empty(&scif_info.loopb_recv_q)) {
1147 msg = list_first_entry(&scif_info.loopb_recv_q,
1148 struct scif_loopb_msg,
1149 list);
1150 list_del(&msg->list);
1151 }
1152 spin_unlock(&qp->recv_lock);
1153
1154 if (msg) {
1155 scif_nodeqp_msg_handler(scifdev, qp, &msg->msg);
1156 kfree(msg);
1157 }
1158 } while (msg);
1159 }
1160
1161 /**
1162 * scif_loopb_msg_handler() - Workqueue handler for loopback messages.
1163 * @scifdev: SCIF device
1164 * @qp: Queue pair.
1165 *
1166 * This work queue routine is triggered when a loopback message is received.
1167 *
1168 * We need special handling for receiving Node Qp messages on a loopback SCIF
1169 * device via two workqueues for receiving messages.
1170 *
1171 * The reason we need the extra workqueue which is not required with *normal*
1172 * non-loopback SCIF devices is the potential classic deadlock described below:
1173 *
1174 * Thread A tries to send a message on a loopback SCIF device and blocks since
1175 * there is no space in the RB while it has the send_lock held or another
1176 * lock called lock X for example.
1177 *
1178 * Thread B: The Loopback Node QP message receive workqueue receives the message
1179 * and tries to send a message (eg an ACK) to the loopback SCIF device. It tries
1180 * to grab the send lock again or lock X and deadlocks with Thread A. The RB
1181 * cannot be drained any further due to this classic deadlock.
1182 *
1183 * In order to avoid deadlocks as mentioned above we have an extra level of
1184 * indirection achieved by having two workqueues.
1185 * 1) The first workqueue whose handler is scif_loopb_msg_handler reads
1186 * messages from the Node QP RB, adds them to a list and queues work for the
1187 * second workqueue.
1188 *
1189 * 2) The second workqueue whose handler is scif_loopb_wq_handler dequeues
1190 * messages from the list, handles them, frees up the memory and dequeues
1191 * more elements from the list if possible.
1192 */
1193 int
scif_loopb_msg_handler(struct scif_dev * scifdev,struct scif_qp * qp)1194 scif_loopb_msg_handler(struct scif_dev *scifdev, struct scif_qp *qp)
1195 {
1196 int read_size;
1197 struct scif_loopb_msg *msg;
1198
1199 do {
1200 msg = kmalloc(sizeof(*msg), GFP_KERNEL);
1201 if (!msg)
1202 return -ENOMEM;
1203 read_size = scif_rb_get_next(&qp->inbound_q, &msg->msg,
1204 sizeof(struct scifmsg));
1205 if (read_size != sizeof(struct scifmsg)) {
1206 kfree(msg);
1207 scif_rb_update_read_ptr(&qp->inbound_q);
1208 break;
1209 }
1210 spin_lock(&qp->recv_lock);
1211 list_add_tail(&msg->list, &scif_info.loopb_recv_q);
1212 spin_unlock(&qp->recv_lock);
1213 queue_work(scif_info.loopb_wq, &scif_info.loopb_work);
1214 scif_rb_update_read_ptr(&qp->inbound_q);
1215 } while (read_size == sizeof(struct scifmsg));
1216 return read_size;
1217 }
1218
1219 /**
1220 * scif_setup_loopback_qp - One time setup work for Loopback Node Qp.
1221 * @scifdev: SCIF device
1222 *
1223 * Sets up the required loopback workqueues, queue pairs and ring buffers
1224 */
scif_setup_loopback_qp(struct scif_dev * scifdev)1225 int scif_setup_loopback_qp(struct scif_dev *scifdev)
1226 {
1227 int err = 0;
1228 void *local_q;
1229 struct scif_qp *qp;
1230
1231 err = scif_setup_intr_wq(scifdev);
1232 if (err)
1233 goto exit;
1234 INIT_LIST_HEAD(&scif_info.loopb_recv_q);
1235 snprintf(scif_info.loopb_wqname, sizeof(scif_info.loopb_wqname),
1236 "SCIF LOOPB %d", scifdev->node);
1237 scif_info.loopb_wq =
1238 alloc_ordered_workqueue(scif_info.loopb_wqname, 0);
1239 if (!scif_info.loopb_wq) {
1240 err = -ENOMEM;
1241 goto destroy_intr;
1242 }
1243 INIT_WORK(&scif_info.loopb_work, scif_loopb_wq_handler);
1244 /* Allocate Self Qpair */
1245 scifdev->qpairs = kzalloc(sizeof(*scifdev->qpairs), GFP_KERNEL);
1246 if (!scifdev->qpairs) {
1247 err = -ENOMEM;
1248 goto destroy_loopb_wq;
1249 }
1250
1251 qp = scifdev->qpairs;
1252 qp->magic = SCIFEP_MAGIC;
1253 spin_lock_init(&qp->send_lock);
1254 spin_lock_init(&qp->recv_lock);
1255
1256 local_q = kzalloc(SCIF_NODE_QP_SIZE, GFP_KERNEL);
1257 if (!local_q) {
1258 err = -ENOMEM;
1259 goto free_qpairs;
1260 }
1261 /*
1262 * For loopback the inbound_q and outbound_q are essentially the same
1263 * since the Node sends a message on the loopback interface to the
1264 * outbound_q which is then received on the inbound_q.
1265 */
1266 scif_rb_init(&qp->outbound_q,
1267 &qp->local_read,
1268 &qp->local_write,
1269 local_q, get_count_order(SCIF_NODE_QP_SIZE));
1270
1271 scif_rb_init(&qp->inbound_q,
1272 &qp->local_read,
1273 &qp->local_write,
1274 local_q, get_count_order(SCIF_NODE_QP_SIZE));
1275 scif_info.nodeid = scifdev->node;
1276
1277 scif_peer_register_device(scifdev);
1278
1279 scif_info.loopb_dev = scifdev;
1280 return err;
1281 free_qpairs:
1282 kfree(scifdev->qpairs);
1283 destroy_loopb_wq:
1284 destroy_workqueue(scif_info.loopb_wq);
1285 destroy_intr:
1286 scif_destroy_intr_wq(scifdev);
1287 exit:
1288 return err;
1289 }
1290
1291 /**
1292 * scif_destroy_loopback_qp - One time uninit work for Loopback Node Qp
1293 * @scifdev: SCIF device
1294 *
1295 * Destroys the workqueues and frees up the Ring Buffer and Queue Pair memory.
1296 */
scif_destroy_loopback_qp(struct scif_dev * scifdev)1297 int scif_destroy_loopback_qp(struct scif_dev *scifdev)
1298 {
1299 scif_peer_unregister_device(scifdev);
1300 destroy_workqueue(scif_info.loopb_wq);
1301 scif_destroy_intr_wq(scifdev);
1302 kfree(scifdev->qpairs->outbound_q.rb_base);
1303 kfree(scifdev->qpairs);
1304 scifdev->sdev = NULL;
1305 scif_info.loopb_dev = NULL;
1306 return 0;
1307 }
1308
scif_destroy_p2p(struct scif_dev * scifdev)1309 void scif_destroy_p2p(struct scif_dev *scifdev)
1310 {
1311 struct scif_dev *peer_dev;
1312 struct scif_p2p_info *p2p;
1313 struct list_head *pos, *tmp;
1314 int bd;
1315
1316 mutex_lock(&scif_info.conflock);
1317 /* Free P2P mappings in the given node for all its peer nodes */
1318 list_for_each_safe(pos, tmp, &scifdev->p2p) {
1319 p2p = list_entry(pos, struct scif_p2p_info, ppi_list);
1320 dma_unmap_sg(&scifdev->sdev->dev, p2p->ppi_sg[SCIF_PPI_MMIO],
1321 p2p->sg_nentries[SCIF_PPI_MMIO],
1322 DMA_BIDIRECTIONAL);
1323 dma_unmap_sg(&scifdev->sdev->dev, p2p->ppi_sg[SCIF_PPI_APER],
1324 p2p->sg_nentries[SCIF_PPI_APER],
1325 DMA_BIDIRECTIONAL);
1326 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_MMIO]);
1327 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_APER]);
1328 list_del(pos);
1329 kfree(p2p);
1330 }
1331
1332 /* Free P2P mapping created in the peer nodes for the given node */
1333 for (bd = SCIF_MGMT_NODE + 1; bd <= scif_info.maxid; bd++) {
1334 peer_dev = &scif_dev[bd];
1335 list_for_each_safe(pos, tmp, &peer_dev->p2p) {
1336 p2p = list_entry(pos, struct scif_p2p_info, ppi_list);
1337 if (p2p->ppi_peer_id == scifdev->node) {
1338 dma_unmap_sg(&peer_dev->sdev->dev,
1339 p2p->ppi_sg[SCIF_PPI_MMIO],
1340 p2p->sg_nentries[SCIF_PPI_MMIO],
1341 DMA_BIDIRECTIONAL);
1342 dma_unmap_sg(&peer_dev->sdev->dev,
1343 p2p->ppi_sg[SCIF_PPI_APER],
1344 p2p->sg_nentries[SCIF_PPI_APER],
1345 DMA_BIDIRECTIONAL);
1346 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_MMIO]);
1347 scif_p2p_freesg(p2p->ppi_sg[SCIF_PPI_APER]);
1348 list_del(pos);
1349 kfree(p2p);
1350 }
1351 }
1352 }
1353 mutex_unlock(&scif_info.conflock);
1354 }
1355