1 /*
2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/highmem.h>
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/mlx5/driver.h>
37 #include <linux/mlx5/cmd.h>
38 #include "mlx5_core.h"
39
40 enum {
41 MLX5_PAGES_CANT_GIVE = 0,
42 MLX5_PAGES_GIVE = 1,
43 MLX5_PAGES_TAKE = 2
44 };
45
46 enum {
47 MLX5_BOOT_PAGES = 1,
48 MLX5_INIT_PAGES = 2,
49 MLX5_POST_INIT_PAGES = 3
50 };
51
52 struct mlx5_pages_req {
53 struct mlx5_core_dev *dev;
54 u16 func_id;
55 s32 npages;
56 struct work_struct work;
57 };
58
59 struct fw_page {
60 struct rb_node rb_node;
61 u64 addr;
62 struct page *page;
63 u16 func_id;
64 unsigned long bitmask;
65 struct list_head list;
66 unsigned free_count;
67 };
68
69 struct mlx5_query_pages_inbox {
70 struct mlx5_inbox_hdr hdr;
71 u8 rsvd[8];
72 };
73
74 struct mlx5_query_pages_outbox {
75 struct mlx5_outbox_hdr hdr;
76 __be16 rsvd;
77 __be16 func_id;
78 __be32 num_pages;
79 };
80
81 struct mlx5_manage_pages_inbox {
82 struct mlx5_inbox_hdr hdr;
83 __be16 rsvd;
84 __be16 func_id;
85 __be32 num_entries;
86 __be64 pas[0];
87 };
88
89 struct mlx5_manage_pages_outbox {
90 struct mlx5_outbox_hdr hdr;
91 __be32 num_entries;
92 u8 rsvd[4];
93 __be64 pas[0];
94 };
95
96 enum {
97 MAX_RECLAIM_TIME_MSECS = 5000,
98 };
99
100 enum {
101 MLX5_MAX_RECLAIM_TIME_MILI = 5000,
102 MLX5_NUM_4K_IN_PAGE = PAGE_SIZE / MLX5_ADAPTER_PAGE_SIZE,
103 };
104
insert_page(struct mlx5_core_dev * dev,u64 addr,struct page * page,u16 func_id)105 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u16 func_id)
106 {
107 struct rb_root *root = &dev->priv.page_root;
108 struct rb_node **new = &root->rb_node;
109 struct rb_node *parent = NULL;
110 struct fw_page *nfp;
111 struct fw_page *tfp;
112 int i;
113
114 while (*new) {
115 parent = *new;
116 tfp = rb_entry(parent, struct fw_page, rb_node);
117 if (tfp->addr < addr)
118 new = &parent->rb_left;
119 else if (tfp->addr > addr)
120 new = &parent->rb_right;
121 else
122 return -EEXIST;
123 }
124
125 nfp = kzalloc(sizeof(*nfp), GFP_KERNEL);
126 if (!nfp)
127 return -ENOMEM;
128
129 nfp->addr = addr;
130 nfp->page = page;
131 nfp->func_id = func_id;
132 nfp->free_count = MLX5_NUM_4K_IN_PAGE;
133 for (i = 0; i < MLX5_NUM_4K_IN_PAGE; i++)
134 set_bit(i, &nfp->bitmask);
135
136 rb_link_node(&nfp->rb_node, parent, new);
137 rb_insert_color(&nfp->rb_node, root);
138 list_add(&nfp->list, &dev->priv.free_list);
139
140 return 0;
141 }
142
find_fw_page(struct mlx5_core_dev * dev,u64 addr)143 static struct fw_page *find_fw_page(struct mlx5_core_dev *dev, u64 addr)
144 {
145 struct rb_root *root = &dev->priv.page_root;
146 struct rb_node *tmp = root->rb_node;
147 struct fw_page *result = NULL;
148 struct fw_page *tfp;
149
150 while (tmp) {
151 tfp = rb_entry(tmp, struct fw_page, rb_node);
152 if (tfp->addr < addr) {
153 tmp = tmp->rb_left;
154 } else if (tfp->addr > addr) {
155 tmp = tmp->rb_right;
156 } else {
157 result = tfp;
158 break;
159 }
160 }
161
162 return result;
163 }
164
mlx5_cmd_query_pages(struct mlx5_core_dev * dev,u16 * func_id,s32 * npages,int boot)165 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
166 s32 *npages, int boot)
167 {
168 struct mlx5_query_pages_inbox in;
169 struct mlx5_query_pages_outbox out;
170 int err;
171
172 memset(&in, 0, sizeof(in));
173 memset(&out, 0, sizeof(out));
174 in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_QUERY_PAGES);
175 in.hdr.opmod = boot ? cpu_to_be16(MLX5_BOOT_PAGES) : cpu_to_be16(MLX5_INIT_PAGES);
176
177 err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
178 if (err)
179 return err;
180
181 if (out.hdr.status)
182 return mlx5_cmd_status_to_err(&out.hdr);
183
184 *npages = be32_to_cpu(out.num_pages);
185 *func_id = be16_to_cpu(out.func_id);
186
187 return err;
188 }
189
alloc_4k(struct mlx5_core_dev * dev,u64 * addr)190 static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr)
191 {
192 struct fw_page *fp;
193 unsigned n;
194
195 if (list_empty(&dev->priv.free_list))
196 return -ENOMEM;
197
198 fp = list_entry(dev->priv.free_list.next, struct fw_page, list);
199 n = find_first_bit(&fp->bitmask, 8 * sizeof(fp->bitmask));
200 if (n >= MLX5_NUM_4K_IN_PAGE) {
201 mlx5_core_warn(dev, "alloc 4k bug\n");
202 return -ENOENT;
203 }
204 clear_bit(n, &fp->bitmask);
205 fp->free_count--;
206 if (!fp->free_count)
207 list_del(&fp->list);
208
209 *addr = fp->addr + n * MLX5_ADAPTER_PAGE_SIZE;
210
211 return 0;
212 }
213
214 #define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
215
free_4k(struct mlx5_core_dev * dev,u64 addr)216 static void free_4k(struct mlx5_core_dev *dev, u64 addr)
217 {
218 struct fw_page *fwp;
219 int n;
220
221 fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
222 if (!fwp) {
223 mlx5_core_warn(dev, "page not found\n");
224 return;
225 }
226
227 n = (addr & ~MLX5_U64_4K_PAGE_MASK) >> MLX5_ADAPTER_PAGE_SHIFT;
228 fwp->free_count++;
229 set_bit(n, &fwp->bitmask);
230 if (fwp->free_count == MLX5_NUM_4K_IN_PAGE) {
231 rb_erase(&fwp->rb_node, &dev->priv.page_root);
232 if (fwp->free_count != 1)
233 list_del(&fwp->list);
234 dma_unmap_page(&dev->pdev->dev, addr & MLX5_U64_4K_PAGE_MASK,
235 PAGE_SIZE, DMA_BIDIRECTIONAL);
236 __free_page(fwp->page);
237 kfree(fwp);
238 } else if (fwp->free_count == 1) {
239 list_add(&fwp->list, &dev->priv.free_list);
240 }
241 }
242
alloc_system_page(struct mlx5_core_dev * dev,u16 func_id)243 static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
244 {
245 struct page *page;
246 u64 addr;
247 int err;
248 int nid = dev_to_node(&dev->pdev->dev);
249
250 page = alloc_pages_node(nid, GFP_HIGHUSER, 0);
251 if (!page) {
252 mlx5_core_warn(dev, "failed to allocate page\n");
253 return -ENOMEM;
254 }
255 addr = dma_map_page(&dev->pdev->dev, page, 0,
256 PAGE_SIZE, DMA_BIDIRECTIONAL);
257 if (dma_mapping_error(&dev->pdev->dev, addr)) {
258 mlx5_core_warn(dev, "failed dma mapping page\n");
259 err = -ENOMEM;
260 goto out_alloc;
261 }
262 err = insert_page(dev, addr, page, func_id);
263 if (err) {
264 mlx5_core_err(dev, "failed to track allocated page\n");
265 goto out_mapping;
266 }
267
268 return 0;
269
270 out_mapping:
271 dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
272
273 out_alloc:
274 __free_page(page);
275
276 return err;
277 }
278
page_notify_fail(struct mlx5_core_dev * dev,u16 func_id)279 static void page_notify_fail(struct mlx5_core_dev *dev, u16 func_id)
280 {
281 struct mlx5_manage_pages_inbox *in;
282 struct mlx5_manage_pages_outbox out;
283 int err;
284
285 in = kzalloc(sizeof(*in), GFP_KERNEL);
286 if (!in)
287 return;
288
289 memset(&out, 0, sizeof(out));
290 in->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
291 in->hdr.opmod = cpu_to_be16(MLX5_PAGES_CANT_GIVE);
292 in->func_id = cpu_to_be16(func_id);
293 err = mlx5_cmd_exec(dev, in, sizeof(*in), &out, sizeof(out));
294 if (!err)
295 err = mlx5_cmd_status_to_err(&out.hdr);
296
297 if (err)
298 mlx5_core_warn(dev, "page notify failed\n");
299
300 kfree(in);
301 }
302
give_pages(struct mlx5_core_dev * dev,u16 func_id,int npages,int notify_fail)303 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
304 int notify_fail)
305 {
306 struct mlx5_manage_pages_inbox *in;
307 struct mlx5_manage_pages_outbox out;
308 int inlen;
309 u64 addr;
310 int err;
311 int i;
312
313 inlen = sizeof(*in) + npages * sizeof(in->pas[0]);
314 in = mlx5_vzalloc(inlen);
315 if (!in) {
316 err = -ENOMEM;
317 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
318 goto out_free;
319 }
320 memset(&out, 0, sizeof(out));
321
322 for (i = 0; i < npages; i++) {
323 retry:
324 err = alloc_4k(dev, &addr);
325 if (err) {
326 if (err == -ENOMEM)
327 err = alloc_system_page(dev, func_id);
328 if (err)
329 goto out_4k;
330
331 goto retry;
332 }
333 in->pas[i] = cpu_to_be64(addr);
334 }
335
336 in->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
337 in->hdr.opmod = cpu_to_be16(MLX5_PAGES_GIVE);
338 in->func_id = cpu_to_be16(func_id);
339 in->num_entries = cpu_to_be32(npages);
340 err = mlx5_cmd_exec(dev, in, inlen, &out, sizeof(out));
341 if (err) {
342 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n",
343 func_id, npages, err);
344 goto out_4k;
345 }
346 dev->priv.fw_pages += npages;
347
348 err = mlx5_cmd_status_to_err(&out.hdr);
349 if (err) {
350 mlx5_core_warn(dev, "func_id 0x%x, npages %d, status %d\n",
351 func_id, npages, out.hdr.status);
352 goto out_4k;
353 }
354
355 mlx5_core_dbg(dev, "err %d\n", err);
356
357 kvfree(in);
358 return 0;
359
360 out_4k:
361 for (i--; i >= 0; i--)
362 free_4k(dev, be64_to_cpu(in->pas[i]));
363 out_free:
364 kvfree(in);
365 if (notify_fail)
366 page_notify_fail(dev, func_id);
367 return err;
368 }
369
reclaim_pages(struct mlx5_core_dev * dev,u32 func_id,int npages,int * nclaimed)370 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
371 int *nclaimed)
372 {
373 struct mlx5_manage_pages_inbox in;
374 struct mlx5_manage_pages_outbox *out;
375 int num_claimed;
376 int outlen;
377 u64 addr;
378 int err;
379 int i;
380
381 if (nclaimed)
382 *nclaimed = 0;
383
384 memset(&in, 0, sizeof(in));
385 outlen = sizeof(*out) + npages * sizeof(out->pas[0]);
386 out = mlx5_vzalloc(outlen);
387 if (!out)
388 return -ENOMEM;
389
390 in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
391 in.hdr.opmod = cpu_to_be16(MLX5_PAGES_TAKE);
392 in.func_id = cpu_to_be16(func_id);
393 in.num_entries = cpu_to_be32(npages);
394 mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
395 err = mlx5_cmd_exec(dev, &in, sizeof(in), out, outlen);
396 if (err) {
397 mlx5_core_err(dev, "failed reclaiming pages\n");
398 goto out_free;
399 }
400 dev->priv.fw_pages -= npages;
401
402 if (out->hdr.status) {
403 err = mlx5_cmd_status_to_err(&out->hdr);
404 goto out_free;
405 }
406
407 num_claimed = be32_to_cpu(out->num_entries);
408 if (nclaimed)
409 *nclaimed = num_claimed;
410
411 for (i = 0; i < num_claimed; i++) {
412 addr = be64_to_cpu(out->pas[i]);
413 free_4k(dev, addr);
414 }
415
416 out_free:
417 kvfree(out);
418 return err;
419 }
420
pages_work_handler(struct work_struct * work)421 static void pages_work_handler(struct work_struct *work)
422 {
423 struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
424 struct mlx5_core_dev *dev = req->dev;
425 int err = 0;
426
427 if (req->npages < 0)
428 err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL);
429 else if (req->npages > 0)
430 err = give_pages(dev, req->func_id, req->npages, 1);
431
432 if (err)
433 mlx5_core_warn(dev, "%s fail %d\n",
434 req->npages < 0 ? "reclaim" : "give", err);
435
436 kfree(req);
437 }
438
mlx5_core_req_pages_handler(struct mlx5_core_dev * dev,u16 func_id,s32 npages)439 void mlx5_core_req_pages_handler(struct mlx5_core_dev *dev, u16 func_id,
440 s32 npages)
441 {
442 struct mlx5_pages_req *req;
443
444 req = kzalloc(sizeof(*req), GFP_ATOMIC);
445 if (!req) {
446 mlx5_core_warn(dev, "failed to allocate pages request\n");
447 return;
448 }
449
450 req->dev = dev;
451 req->func_id = func_id;
452 req->npages = npages;
453 INIT_WORK(&req->work, pages_work_handler);
454 queue_work(dev->priv.pg_wq, &req->work);
455 }
456
mlx5_satisfy_startup_pages(struct mlx5_core_dev * dev,int boot)457 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
458 {
459 u16 uninitialized_var(func_id);
460 s32 uninitialized_var(npages);
461 int err;
462
463 err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
464 if (err)
465 return err;
466
467 mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
468 npages, boot ? "boot" : "init", func_id);
469
470 return give_pages(dev, func_id, npages, 0);
471 }
472
473 enum {
474 MLX5_BLKS_FOR_RECLAIM_PAGES = 12
475 };
476
optimal_reclaimed_pages(void)477 static int optimal_reclaimed_pages(void)
478 {
479 struct mlx5_cmd_prot_block *block;
480 struct mlx5_cmd_layout *lay;
481 int ret;
482
483 ret = (sizeof(lay->out) + MLX5_BLKS_FOR_RECLAIM_PAGES * sizeof(block->data) -
484 sizeof(struct mlx5_manage_pages_outbox)) /
485 FIELD_SIZEOF(struct mlx5_manage_pages_outbox, pas[0]);
486
487 return ret;
488 }
489
mlx5_reclaim_startup_pages(struct mlx5_core_dev * dev)490 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
491 {
492 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
493 struct fw_page *fwp;
494 struct rb_node *p;
495 int nclaimed = 0;
496 int err = 0;
497
498 do {
499 p = rb_first(&dev->priv.page_root);
500 if (p) {
501 fwp = rb_entry(p, struct fw_page, rb_node);
502 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
503 free_4k(dev, fwp->addr);
504 nclaimed = 1;
505 } else {
506 err = reclaim_pages(dev, fwp->func_id,
507 optimal_reclaimed_pages(),
508 &nclaimed);
509 }
510 if (err) {
511 mlx5_core_warn(dev, "failed reclaiming pages (%d)\n",
512 err);
513 return err;
514 }
515 if (nclaimed)
516 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
517 }
518 if (time_after(jiffies, end)) {
519 mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
520 break;
521 }
522 } while (p);
523
524 return 0;
525 }
526
mlx5_pagealloc_init(struct mlx5_core_dev * dev)527 void mlx5_pagealloc_init(struct mlx5_core_dev *dev)
528 {
529 dev->priv.page_root = RB_ROOT;
530 INIT_LIST_HEAD(&dev->priv.free_list);
531 }
532
mlx5_pagealloc_cleanup(struct mlx5_core_dev * dev)533 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
534 {
535 /* nothing */
536 }
537
mlx5_pagealloc_start(struct mlx5_core_dev * dev)538 int mlx5_pagealloc_start(struct mlx5_core_dev *dev)
539 {
540 dev->priv.pg_wq = create_singlethread_workqueue("mlx5_page_allocator");
541 if (!dev->priv.pg_wq)
542 return -ENOMEM;
543
544 return 0;
545 }
546
mlx5_pagealloc_stop(struct mlx5_core_dev * dev)547 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
548 {
549 destroy_workqueue(dev->priv.pg_wq);
550 }
551