2 * Copyright (c) 2013, Mellanox Technologies inc. 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;
66 struct mlx5_query_pages_inbox {
67 struct mlx5_inbox_hdr hdr;
71 struct mlx5_query_pages_outbox {
72 struct mlx5_outbox_hdr hdr;
78 struct mlx5_manage_pages_inbox {
79 struct mlx5_inbox_hdr hdr;
86 struct mlx5_manage_pages_outbox {
87 struct mlx5_outbox_hdr hdr;
94 MAX_RECLAIM_TIME_MSECS = 5000,
97 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u16 func_id)
99 struct rb_root *root = &dev->priv.page_root;
100 struct rb_node **new = &root->rb_node;
101 struct rb_node *parent = NULL;
107 tfp = rb_entry(parent, struct fw_page, rb_node);
108 if (tfp->addr < addr)
109 new = &parent->rb_left;
110 else if (tfp->addr > addr)
111 new = &parent->rb_right;
116 nfp = kmalloc(sizeof(*nfp), GFP_KERNEL);
122 nfp->func_id = func_id;
124 rb_link_node(&nfp->rb_node, parent, new);
125 rb_insert_color(&nfp->rb_node, root);
130 static struct page *remove_page(struct mlx5_core_dev *dev, u64 addr)
132 struct rb_root *root = &dev->priv.page_root;
133 struct rb_node *tmp = root->rb_node;
134 struct page *result = NULL;
138 tfp = rb_entry(tmp, struct fw_page, rb_node);
139 if (tfp->addr < addr) {
141 } else if (tfp->addr > addr) {
144 rb_erase(&tfp->rb_node, root);
154 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
155 s32 *npages, int boot)
157 struct mlx5_query_pages_inbox in;
158 struct mlx5_query_pages_outbox out;
161 memset(&in, 0, sizeof(in));
162 memset(&out, 0, sizeof(out));
163 in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_QUERY_PAGES);
164 in.hdr.opmod = boot ? cpu_to_be16(MLX5_BOOT_PAGES) : cpu_to_be16(MLX5_INIT_PAGES);
166 err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
171 return mlx5_cmd_status_to_err(&out.hdr);
173 *npages = be32_to_cpu(out.num_pages);
174 *func_id = be16_to_cpu(out.func_id);
179 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
182 struct mlx5_manage_pages_inbox *in;
183 struct mlx5_manage_pages_outbox out;
190 inlen = sizeof(*in) + npages * sizeof(in->pas[0]);
191 in = mlx5_vzalloc(inlen);
193 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
196 memset(&out, 0, sizeof(out));
198 for (i = 0; i < npages; i++) {
199 page = alloc_page(GFP_HIGHUSER);
202 mlx5_core_warn(dev, "failed to allocate page\n");
205 addr = dma_map_page(&dev->pdev->dev, page, 0,
206 PAGE_SIZE, DMA_BIDIRECTIONAL);
207 if (dma_mapping_error(&dev->pdev->dev, addr)) {
208 mlx5_core_warn(dev, "failed dma mapping page\n");
213 err = insert_page(dev, addr, page, func_id);
215 mlx5_core_err(dev, "failed to track allocated page\n");
216 dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
221 in->pas[i] = cpu_to_be64(addr);
224 in->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
225 in->hdr.opmod = cpu_to_be16(MLX5_PAGES_GIVE);
226 in->func_id = cpu_to_be16(func_id);
227 in->num_entries = cpu_to_be32(npages);
228 err = mlx5_cmd_exec(dev, in, inlen, &out, sizeof(out));
229 mlx5_core_dbg(dev, "err %d\n", err);
231 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n", func_id, npages, err);
234 dev->priv.fw_pages += npages;
236 if (out.hdr.status) {
237 err = mlx5_cmd_status_to_err(&out.hdr);
239 mlx5_core_warn(dev, "func_id 0x%x, npages %d, status %d\n", func_id, npages, out.hdr.status);
244 mlx5_core_dbg(dev, "err %d\n", err);
250 memset(in, 0, inlen);
251 memset(&out, 0, sizeof(out));
252 in->hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
253 in->hdr.opmod = cpu_to_be16(MLX5_PAGES_CANT_GIVE);
254 if (mlx5_cmd_exec(dev, in, sizeof(*in), &out, sizeof(out)))
255 mlx5_core_warn(dev, "\n");
257 for (i--; i >= 0; i--) {
258 addr = be64_to_cpu(in->pas[i]);
259 page = remove_page(dev, addr);
261 mlx5_core_err(dev, "BUG: can't remove page at addr 0x%llx\n",
265 dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
274 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
277 struct mlx5_manage_pages_inbox in;
278 struct mlx5_manage_pages_outbox *out;
289 memset(&in, 0, sizeof(in));
290 outlen = sizeof(*out) + npages * sizeof(out->pas[0]);
291 out = mlx5_vzalloc(outlen);
295 in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_MANAGE_PAGES);
296 in.hdr.opmod = cpu_to_be16(MLX5_PAGES_TAKE);
297 in.func_id = cpu_to_be16(func_id);
298 in.num_entries = cpu_to_be32(npages);
299 mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
300 err = mlx5_cmd_exec(dev, &in, sizeof(in), out, outlen);
302 mlx5_core_err(dev, "failed recliaming pages\n");
305 dev->priv.fw_pages -= npages;
307 if (out->hdr.status) {
308 err = mlx5_cmd_status_to_err(&out->hdr);
312 num_claimed = be32_to_cpu(out->num_entries);
314 *nclaimed = num_claimed;
316 for (i = 0; i < num_claimed; i++) {
317 addr = be64_to_cpu(out->pas[i]);
318 page = remove_page(dev, addr);
320 mlx5_core_warn(dev, "FW reported unknown DMA address 0x%llx\n", addr);
322 dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
332 static void pages_work_handler(struct work_struct *work)
334 struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
335 struct mlx5_core_dev *dev = req->dev;
339 err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL);
340 else if (req->npages > 0)
341 err = give_pages(dev, req->func_id, req->npages, 1);
344 mlx5_core_warn(dev, "%s fail %d\n", req->npages < 0 ?
345 "reclaim" : "give", err);
350 void mlx5_core_req_pages_handler(struct mlx5_core_dev *dev, u16 func_id,
353 struct mlx5_pages_req *req;
355 req = kzalloc(sizeof(*req), GFP_ATOMIC);
357 mlx5_core_warn(dev, "failed to allocate pages request\n");
362 req->func_id = func_id;
363 req->npages = npages;
364 INIT_WORK(&req->work, pages_work_handler);
365 queue_work(dev->priv.pg_wq, &req->work);
368 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
370 u16 uninitialized_var(func_id);
371 s32 uninitialized_var(npages);
374 err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
378 mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
379 npages, boot ? "boot" : "init", func_id);
381 return give_pages(dev, func_id, npages, 0);
384 static int optimal_reclaimed_pages(void)
386 struct mlx5_cmd_prot_block *block;
387 struct mlx5_cmd_layout *lay;
390 ret = (sizeof(lay->in) + sizeof(block->data) -
391 sizeof(struct mlx5_manage_pages_outbox)) / 8;
396 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
398 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
405 p = rb_first(&dev->priv.page_root);
407 fwp = rb_entry(p, struct fw_page, rb_node);
408 err = reclaim_pages(dev, fwp->func_id,
409 optimal_reclaimed_pages(),
412 mlx5_core_warn(dev, "failed reclaiming pages (%d)\n", err);
416 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
418 if (time_after(jiffies, end)) {
419 mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
427 void mlx5_pagealloc_init(struct mlx5_core_dev *dev)
429 dev->priv.page_root = RB_ROOT;
432 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
437 int mlx5_pagealloc_start(struct mlx5_core_dev *dev)
439 dev->priv.pg_wq = create_singlethread_workqueue("mlx5_page_allocator");
440 if (!dev->priv.pg_wq)
446 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
448 destroy_workqueue(dev->priv.pg_wq);