2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/highmem.h>
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/delay.h>
37 #include <linux/mlx5/driver.h>
38 #include "mlx5_core.h"
42 MLX5_PAGES_CANT_GIVE = 0,
47 struct mlx5_pages_req {
48 struct mlx5_core_dev *dev;
52 struct work_struct work;
57 struct rb_node rb_node;
61 unsigned long bitmask;
62 struct list_head list;
67 MAX_RECLAIM_TIME_MSECS = 5000,
68 MAX_RECLAIM_VFS_PAGES_TIME_MSECS = 2 * 1000 * 60,
72 MLX5_MAX_RECLAIM_TIME_MILI = 5000,
73 MLX5_NUM_4K_IN_PAGE = PAGE_SIZE / MLX5_ADAPTER_PAGE_SIZE,
76 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u16 func_id)
78 struct rb_root *root = &dev->priv.page_root;
79 struct rb_node **new = &root->rb_node;
80 struct rb_node *parent = NULL;
87 tfp = rb_entry(parent, struct fw_page, rb_node);
89 new = &parent->rb_left;
90 else if (tfp->addr > addr)
91 new = &parent->rb_right;
96 nfp = kzalloc(sizeof(*nfp), GFP_KERNEL);
102 nfp->func_id = func_id;
103 nfp->free_count = MLX5_NUM_4K_IN_PAGE;
104 for (i = 0; i < MLX5_NUM_4K_IN_PAGE; i++)
105 set_bit(i, &nfp->bitmask);
107 rb_link_node(&nfp->rb_node, parent, new);
108 rb_insert_color(&nfp->rb_node, root);
109 list_add(&nfp->list, &dev->priv.free_list);
114 static struct fw_page *find_fw_page(struct mlx5_core_dev *dev, u64 addr)
116 struct rb_root *root = &dev->priv.page_root;
117 struct rb_node *tmp = root->rb_node;
118 struct fw_page *result = NULL;
122 tfp = rb_entry(tmp, struct fw_page, rb_node);
123 if (tfp->addr < addr) {
125 } else if (tfp->addr > addr) {
136 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
137 s32 *npages, int boot)
139 u32 out[MLX5_ST_SZ_DW(query_pages_out)] = {};
140 u32 in[MLX5_ST_SZ_DW(query_pages_in)] = {};
143 MLX5_SET(query_pages_in, in, opcode, MLX5_CMD_OP_QUERY_PAGES);
144 MLX5_SET(query_pages_in, in, op_mod, boot ?
145 MLX5_QUERY_PAGES_IN_OP_MOD_BOOT_PAGES :
146 MLX5_QUERY_PAGES_IN_OP_MOD_INIT_PAGES);
147 MLX5_SET(query_pages_in, in, embedded_cpu_function, mlx5_core_is_ecpf(dev));
149 err = mlx5_cmd_exec_inout(dev, query_pages, in, out);
153 *npages = MLX5_GET(query_pages_out, out, num_pages);
154 *func_id = MLX5_GET(query_pages_out, out, function_id);
159 static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr, u16 func_id)
161 struct fw_page *fp = NULL;
162 struct fw_page *iter;
165 list_for_each_entry(iter, &dev->priv.free_list, list) {
166 if (iter->func_id != func_id)
171 if (list_empty(&dev->priv.free_list) || !fp)
174 n = find_first_bit(&fp->bitmask, 8 * sizeof(fp->bitmask));
175 if (n >= MLX5_NUM_4K_IN_PAGE) {
176 mlx5_core_warn(dev, "alloc 4k bug\n");
179 clear_bit(n, &fp->bitmask);
184 *addr = fp->addr + n * MLX5_ADAPTER_PAGE_SIZE;
189 #define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
191 static void free_fwp(struct mlx5_core_dev *dev, struct fw_page *fwp,
194 rb_erase(&fwp->rb_node, &dev->priv.page_root);
196 list_del(&fwp->list);
197 dma_unmap_page(dev->device, fwp->addr & MLX5_U64_4K_PAGE_MASK,
198 PAGE_SIZE, DMA_BIDIRECTIONAL);
199 __free_page(fwp->page);
203 static void free_4k(struct mlx5_core_dev *dev, u64 addr)
208 fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
210 mlx5_core_warn_rl(dev, "page not found\n");
213 n = (addr & ~MLX5_U64_4K_PAGE_MASK) >> MLX5_ADAPTER_PAGE_SHIFT;
215 set_bit(n, &fwp->bitmask);
216 if (fwp->free_count == MLX5_NUM_4K_IN_PAGE)
217 free_fwp(dev, fwp, fwp->free_count != 1);
218 else if (fwp->free_count == 1)
219 list_add(&fwp->list, &dev->priv.free_list);
222 static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
224 struct device *device = dev->device;
225 int nid = dev_to_node(device);
231 page = alloc_pages_node(nid, GFP_HIGHUSER, 0);
233 mlx5_core_warn(dev, "failed to allocate page\n");
237 addr = dma_map_page(device, page, 0, PAGE_SIZE, DMA_BIDIRECTIONAL);
238 if (dma_mapping_error(device, addr)) {
239 mlx5_core_warn(dev, "failed dma mapping page\n");
244 /* Firmware doesn't support page with physical address 0 */
250 err = insert_page(dev, addr, page, func_id);
252 mlx5_core_err(dev, "failed to track allocated page\n");
253 dma_unmap_page(device, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
261 dma_unmap_page(device, zero_addr, PAGE_SIZE,
267 static void page_notify_fail(struct mlx5_core_dev *dev, u16 func_id,
270 u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
273 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
274 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_CANT_GIVE);
275 MLX5_SET(manage_pages_in, in, function_id, func_id);
276 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
278 err = mlx5_cmd_exec_in(dev, manage_pages, in);
280 mlx5_core_warn(dev, "page notify failed func_id(%d) err(%d)\n",
284 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
285 int notify_fail, bool ec_function)
287 u32 out[MLX5_ST_SZ_DW(manage_pages_out)] = {0};
288 int inlen = MLX5_ST_SZ_BYTES(manage_pages_in);
294 inlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_in, pas[0]);
295 in = kvzalloc(inlen, GFP_KERNEL);
298 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
302 for (i = 0; i < npages; i++) {
304 err = alloc_4k(dev, &addr, func_id);
307 err = alloc_system_page(dev, func_id);
313 MLX5_ARRAY_SET64(manage_pages_in, in, pas, i, addr);
316 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
317 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_GIVE);
318 MLX5_SET(manage_pages_in, in, function_id, func_id);
319 MLX5_SET(manage_pages_in, in, input_num_entries, npages);
320 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
322 err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
324 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n",
325 func_id, npages, err);
329 dev->priv.fw_pages += npages;
331 dev->priv.vfs_pages += npages;
332 else if (mlx5_core_is_ecpf(dev) && !ec_function)
333 dev->priv.peer_pf_pages += npages;
335 mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x, err %d\n",
336 npages, ec_function, func_id, err);
342 for (i--; i >= 0; i--)
343 free_4k(dev, MLX5_GET64(manage_pages_in, in, pas[i]));
347 page_notify_fail(dev, func_id, ec_function);
351 static void release_all_pages(struct mlx5_core_dev *dev, u32 func_id,
357 p = rb_first(&dev->priv.page_root);
359 struct fw_page *fwp = rb_entry(p, struct fw_page, rb_node);
362 if (fwp->func_id != func_id)
364 npages += (MLX5_NUM_4K_IN_PAGE - fwp->free_count);
365 free_fwp(dev, fwp, fwp->free_count);
368 dev->priv.fw_pages -= npages;
370 dev->priv.vfs_pages -= npages;
371 else if (mlx5_core_is_ecpf(dev) && !ec_function)
372 dev->priv.peer_pf_pages -= npages;
374 mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x\n",
375 npages, ec_function, func_id);
378 static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
379 u32 *in, int in_size, u32 *out, int out_size)
387 if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
388 return mlx5_cmd_exec(dev, in, in_size, out, out_size);
390 /* No hard feelings, we want our pages back! */
391 npages = MLX5_GET(manage_pages_in, in, input_num_entries);
392 func_id = MLX5_GET(manage_pages_in, in, function_id);
394 p = rb_first(&dev->priv.page_root);
395 while (p && i < npages) {
396 fwp = rb_entry(p, struct fw_page, rb_node);
398 if (fwp->func_id != func_id)
401 MLX5_ARRAY_SET64(manage_pages_out, out, pas, i, fwp->addr);
405 MLX5_SET(manage_pages_out, out, output_num_entries, i);
409 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
410 int *nclaimed, bool ec_function)
412 int outlen = MLX5_ST_SZ_BYTES(manage_pages_out);
413 u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
422 outlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_out, pas[0]);
423 out = kvzalloc(outlen, GFP_KERNEL);
427 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
428 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_TAKE);
429 MLX5_SET(manage_pages_in, in, function_id, func_id);
430 MLX5_SET(manage_pages_in, in, input_num_entries, npages);
431 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
433 mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
434 err = reclaim_pages_cmd(dev, in, sizeof(in), out, outlen);
436 mlx5_core_err(dev, "failed reclaiming pages: err %d\n", err);
440 num_claimed = MLX5_GET(manage_pages_out, out, output_num_entries);
441 if (num_claimed > npages) {
442 mlx5_core_warn(dev, "fw returned %d, driver asked %d => corruption\n",
443 num_claimed, npages);
448 for (i = 0; i < num_claimed; i++)
449 free_4k(dev, MLX5_GET64(manage_pages_out, out, pas[i]));
452 *nclaimed = num_claimed;
454 dev->priv.fw_pages -= num_claimed;
456 dev->priv.vfs_pages -= num_claimed;
457 else if (mlx5_core_is_ecpf(dev) && !ec_function)
458 dev->priv.peer_pf_pages -= num_claimed;
465 static void pages_work_handler(struct work_struct *work)
467 struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
468 struct mlx5_core_dev *dev = req->dev;
471 if (req->release_all)
472 release_all_pages(dev, req->func_id, req->ec_function);
473 else if (req->npages < 0)
474 err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL,
476 else if (req->npages > 0)
477 err = give_pages(dev, req->func_id, req->npages, 1, req->ec_function);
480 mlx5_core_warn(dev, "%s fail %d\n",
481 req->npages < 0 ? "reclaim" : "give", err);
487 EC_FUNCTION_MASK = 0x8000,
488 RELEASE_ALL_PAGES_MASK = 0x4000,
491 static int req_pages_handler(struct notifier_block *nb,
492 unsigned long type, void *data)
494 struct mlx5_pages_req *req;
495 struct mlx5_core_dev *dev;
496 struct mlx5_priv *priv;
497 struct mlx5_eqe *eqe;
503 priv = mlx5_nb_cof(nb, struct mlx5_priv, pg_nb);
504 dev = container_of(priv, struct mlx5_core_dev, priv);
507 func_id = be16_to_cpu(eqe->data.req_pages.func_id);
508 npages = be32_to_cpu(eqe->data.req_pages.num_pages);
509 ec_function = be16_to_cpu(eqe->data.req_pages.ec_function) & EC_FUNCTION_MASK;
510 release_all = be16_to_cpu(eqe->data.req_pages.ec_function) &
511 RELEASE_ALL_PAGES_MASK;
512 mlx5_core_dbg(dev, "page request for func 0x%x, npages %d, release_all %d\n",
513 func_id, npages, release_all);
514 req = kzalloc(sizeof(*req), GFP_ATOMIC);
516 mlx5_core_warn(dev, "failed to allocate pages request\n");
521 req->func_id = func_id;
522 req->npages = npages;
523 req->ec_function = ec_function;
524 req->release_all = release_all;
525 INIT_WORK(&req->work, pages_work_handler);
526 queue_work(dev->priv.pg_wq, &req->work);
530 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
532 u16 uninitialized_var(func_id);
533 s32 uninitialized_var(npages);
536 err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
540 mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
541 npages, boot ? "boot" : "init", func_id);
543 return give_pages(dev, func_id, npages, 0, mlx5_core_is_ecpf(dev));
547 MLX5_BLKS_FOR_RECLAIM_PAGES = 12
550 static int optimal_reclaimed_pages(void)
552 struct mlx5_cmd_prot_block *block;
553 struct mlx5_cmd_layout *lay;
556 ret = (sizeof(lay->out) + MLX5_BLKS_FOR_RECLAIM_PAGES * sizeof(block->data) -
557 MLX5_ST_SZ_BYTES(manage_pages_out)) /
558 MLX5_FLD_SZ_BYTES(manage_pages_out, pas[0]);
563 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
565 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
572 p = rb_first(&dev->priv.page_root);
574 fwp = rb_entry(p, struct fw_page, rb_node);
575 err = reclaim_pages(dev, fwp->func_id,
576 optimal_reclaimed_pages(),
577 &nclaimed, mlx5_core_is_ecpf(dev));
580 mlx5_core_warn(dev, "failed reclaiming pages (%d)\n",
585 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
587 if (time_after(jiffies, end)) {
588 mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
593 WARN(dev->priv.fw_pages,
594 "FW pages counter is %d after reclaiming all pages\n",
596 WARN(dev->priv.vfs_pages,
597 "VFs FW pages counter is %d after reclaiming all pages\n",
598 dev->priv.vfs_pages);
599 WARN(dev->priv.peer_pf_pages,
600 "Peer PF FW pages counter is %d after reclaiming all pages\n",
601 dev->priv.peer_pf_pages);
606 int mlx5_pagealloc_init(struct mlx5_core_dev *dev)
608 dev->priv.page_root = RB_ROOT;
609 INIT_LIST_HEAD(&dev->priv.free_list);
610 dev->priv.pg_wq = create_singlethread_workqueue("mlx5_page_allocator");
611 if (!dev->priv.pg_wq)
617 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
619 destroy_workqueue(dev->priv.pg_wq);
622 void mlx5_pagealloc_start(struct mlx5_core_dev *dev)
624 MLX5_NB_INIT(&dev->priv.pg_nb, req_pages_handler, PAGE_REQUEST);
625 mlx5_eq_notifier_register(dev, &dev->priv.pg_nb);
628 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
630 mlx5_eq_notifier_unregister(dev, &dev->priv.pg_nb);
631 flush_workqueue(dev->priv.pg_wq);
634 int mlx5_wait_for_pages(struct mlx5_core_dev *dev, int *pages)
636 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
637 int prev_pages = *pages;
639 /* In case of internal error we will free the pages manually later */
640 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
641 mlx5_core_warn(dev, "Skipping wait for vf pages stage");
645 mlx5_core_dbg(dev, "Waiting for %d pages\n", prev_pages);
647 if (time_after(jiffies, end)) {
648 mlx5_core_warn(dev, "aborting while there are %d pending pages\n", *pages);
651 if (*pages < prev_pages) {
652 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
658 mlx5_core_dbg(dev, "All pages received\n");