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 <asm-generic/kmap_types.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"
41 MLX5_PAGES_CANT_GIVE = 0,
49 MLX5_POST_INIT_PAGES = 3
52 struct mlx5_pages_req {
53 struct mlx5_core_dev *dev;
56 struct work_struct work;
60 struct rb_node rb_node;
64 unsigned long bitmask;
65 struct list_head list;
69 struct mlx5_query_pages_inbox {
70 struct mlx5_inbox_hdr hdr;
74 struct mlx5_query_pages_outbox {
75 struct mlx5_outbox_hdr hdr;
81 struct mlx5_manage_pages_inbox {
82 struct mlx5_inbox_hdr hdr;
89 struct mlx5_manage_pages_outbox {
90 struct mlx5_outbox_hdr hdr;
97 MAX_RECLAIM_TIME_MSECS = 5000,
101 MLX5_MAX_RECLAIM_TIME_MILI = 5000,
102 MLX5_NUM_4K_IN_PAGE = PAGE_SIZE / MLX5_ADAPTER_PAGE_SIZE,
105 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u16 func_id)
107 struct rb_root *root = &dev->priv.page_root;
108 struct rb_node **new = &root->rb_node;
109 struct rb_node *parent = NULL;
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;
125 nfp = kzalloc(sizeof(*nfp), GFP_KERNEL);
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);
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);
143 static struct fw_page *find_fw_page(struct mlx5_core_dev *dev, u64 addr)
145 struct rb_root *root = &dev->priv.page_root;
146 struct rb_node *tmp = root->rb_node;
147 struct fw_page *result = NULL;
151 tfp = rb_entry(tmp, struct fw_page, rb_node);
152 if (tfp->addr < addr) {
154 } else if (tfp->addr > addr) {
165 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
166 s32 *npages, int boot)
168 struct mlx5_query_pages_inbox in;
169 struct mlx5_query_pages_outbox out;
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);
177 err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
182 return mlx5_cmd_status_to_err(&out.hdr);
184 *npages = be32_to_cpu(out.num_pages);
185 *func_id = be16_to_cpu(out.func_id);
190 static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr)
195 if (list_empty(&dev->priv.free_list))
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");
204 clear_bit(n, &fp->bitmask);
209 *addr = fp->addr + n * MLX5_ADAPTER_PAGE_SIZE;
214 #define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
216 static void free_4k(struct mlx5_core_dev *dev, u64 addr)
221 fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
223 mlx5_core_warn(dev, "page not found\n");
227 n = (addr & ~MLX5_U64_4K_PAGE_MASK) >> MLX5_ADAPTER_PAGE_SHIFT;
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);
238 } else if (fwp->free_count == 1) {
239 list_add(&fwp->list, &dev->priv.free_list);
243 static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
248 int nid = dev_to_node(&dev->pdev->dev);
250 page = alloc_pages_node(nid, GFP_HIGHUSER, 0);
252 mlx5_core_warn(dev, "failed to allocate page\n");
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");
262 err = insert_page(dev, addr, page, func_id);
264 mlx5_core_err(dev, "failed to track allocated page\n");
271 dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
278 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
281 struct mlx5_manage_pages_inbox *in;
282 struct mlx5_manage_pages_outbox out;
283 struct mlx5_manage_pages_inbox *nin;
289 inlen = sizeof(*in) + npages * sizeof(in->pas[0]);
290 in = mlx5_vzalloc(inlen);
292 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
295 memset(&out, 0, sizeof(out));
297 for (i = 0; i < npages; i++) {
299 err = alloc_4k(dev, &addr);
302 err = alloc_system_page(dev, func_id);
308 in->pas[i] = cpu_to_be64(addr);
311 in->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
312 in->hdr.opmod = cpu_to_be16(MLX5_PAGES_GIVE);
313 in->func_id = cpu_to_be16(func_id);
314 in->num_entries = cpu_to_be32(npages);
315 err = mlx5_cmd_exec(dev, in, inlen, &out, sizeof(out));
317 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n",
318 func_id, npages, err);
321 dev->priv.fw_pages += npages;
323 if (out.hdr.status) {
324 err = mlx5_cmd_status_to_err(&out.hdr);
326 mlx5_core_warn(dev, "func_id 0x%x, npages %d, status %d\n",
327 func_id, npages, out.hdr.status);
332 mlx5_core_dbg(dev, "err %d\n", err);
338 nin = kzalloc(sizeof(*nin), GFP_KERNEL);
340 mlx5_core_warn(dev, "allocation failed\n");
343 memset(&out, 0, sizeof(out));
344 nin->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
345 nin->hdr.opmod = cpu_to_be16(MLX5_PAGES_CANT_GIVE);
346 if (mlx5_cmd_exec(dev, nin, sizeof(*nin), &out, sizeof(out)))
347 mlx5_core_warn(dev, "page notify failed\n");
352 for (i--; i >= 0; i--)
353 free_4k(dev, be64_to_cpu(in->pas[i]));
359 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
362 struct mlx5_manage_pages_inbox in;
363 struct mlx5_manage_pages_outbox *out;
373 memset(&in, 0, sizeof(in));
374 outlen = sizeof(*out) + npages * sizeof(out->pas[0]);
375 out = mlx5_vzalloc(outlen);
379 in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
380 in.hdr.opmod = cpu_to_be16(MLX5_PAGES_TAKE);
381 in.func_id = cpu_to_be16(func_id);
382 in.num_entries = cpu_to_be32(npages);
383 mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
384 err = mlx5_cmd_exec(dev, &in, sizeof(in), out, outlen);
386 mlx5_core_err(dev, "failed reclaiming pages\n");
389 dev->priv.fw_pages -= npages;
391 if (out->hdr.status) {
392 err = mlx5_cmd_status_to_err(&out->hdr);
396 num_claimed = be32_to_cpu(out->num_entries);
398 *nclaimed = num_claimed;
400 for (i = 0; i < num_claimed; i++) {
401 addr = be64_to_cpu(out->pas[i]);
410 static void pages_work_handler(struct work_struct *work)
412 struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
413 struct mlx5_core_dev *dev = req->dev;
417 err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL);
418 else if (req->npages > 0)
419 err = give_pages(dev, req->func_id, req->npages, 1);
422 mlx5_core_warn(dev, "%s fail %d\n",
423 req->npages < 0 ? "reclaim" : "give", err);
428 void mlx5_core_req_pages_handler(struct mlx5_core_dev *dev, u16 func_id,
431 struct mlx5_pages_req *req;
433 req = kzalloc(sizeof(*req), GFP_ATOMIC);
435 mlx5_core_warn(dev, "failed to allocate pages request\n");
440 req->func_id = func_id;
441 req->npages = npages;
442 INIT_WORK(&req->work, pages_work_handler);
443 queue_work(dev->priv.pg_wq, &req->work);
446 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
448 u16 uninitialized_var(func_id);
449 s32 uninitialized_var(npages);
452 err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
456 mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
457 npages, boot ? "boot" : "init", func_id);
459 return give_pages(dev, func_id, npages, 0);
463 MLX5_BLKS_FOR_RECLAIM_PAGES = 12
466 static int optimal_reclaimed_pages(void)
468 struct mlx5_cmd_prot_block *block;
469 struct mlx5_cmd_layout *lay;
472 ret = (sizeof(lay->out) + MLX5_BLKS_FOR_RECLAIM_PAGES * sizeof(block->data) -
473 sizeof(struct mlx5_manage_pages_outbox)) /
474 FIELD_SIZEOF(struct mlx5_manage_pages_outbox, pas[0]);
479 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
481 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
488 p = rb_first(&dev->priv.page_root);
490 fwp = rb_entry(p, struct fw_page, rb_node);
491 err = reclaim_pages(dev, fwp->func_id,
492 optimal_reclaimed_pages(),
495 mlx5_core_warn(dev, "failed reclaiming pages (%d)\n",
500 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
502 if (time_after(jiffies, end)) {
503 mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
511 void mlx5_pagealloc_init(struct mlx5_core_dev *dev)
513 dev->priv.page_root = RB_ROOT;
514 INIT_LIST_HEAD(&dev->priv.free_list);
517 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
522 int mlx5_pagealloc_start(struct mlx5_core_dev *dev)
524 dev->priv.pg_wq = create_singlethread_workqueue("mlx5_page_allocator");
525 if (!dev->priv.pg_wq)
531 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
533 destroy_workqueue(dev->priv.pg_wq);