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 <linux/xarray.h>
39 #include "mlx5_core.h"
43 MLX5_PAGES_CANT_GIVE = 0,
48 struct mlx5_pages_req {
49 struct mlx5_core_dev *dev;
53 struct work_struct work;
58 struct rb_node rb_node;
62 unsigned long bitmask;
63 struct list_head list;
64 unsigned int free_count;
68 MAX_RECLAIM_TIME_MSECS = 5000,
69 MAX_RECLAIM_VFS_PAGES_TIME_MSECS = 2 * 1000 * 60,
73 MLX5_MAX_RECLAIM_TIME_MILI = 5000,
74 MLX5_NUM_4K_IN_PAGE = PAGE_SIZE / MLX5_ADAPTER_PAGE_SIZE,
77 static u32 get_function(u16 func_id, bool ec_function)
79 return (u32)func_id | (ec_function << 16);
82 static struct rb_root *page_root_per_function(struct mlx5_core_dev *dev, u32 function)
87 root = xa_load(&dev->priv.page_root_xa, function);
91 root = kzalloc(sizeof(*root), GFP_KERNEL);
93 return ERR_PTR(-ENOMEM);
95 err = xa_insert(&dev->priv.page_root_xa, function, root, GFP_KERNEL);
106 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u32 function)
108 struct rb_node *parent = NULL;
109 struct rb_root *root;
110 struct rb_node **new;
115 root = page_root_per_function(dev, function);
117 return PTR_ERR(root);
119 new = &root->rb_node;
123 tfp = rb_entry(parent, struct fw_page, rb_node);
124 if (tfp->addr < addr)
125 new = &parent->rb_left;
126 else if (tfp->addr > addr)
127 new = &parent->rb_right;
132 nfp = kzalloc(sizeof(*nfp), GFP_KERNEL);
138 nfp->function = function;
139 nfp->free_count = MLX5_NUM_4K_IN_PAGE;
140 for (i = 0; i < MLX5_NUM_4K_IN_PAGE; i++)
141 set_bit(i, &nfp->bitmask);
143 rb_link_node(&nfp->rb_node, parent, new);
144 rb_insert_color(&nfp->rb_node, root);
145 list_add(&nfp->list, &dev->priv.free_list);
150 static struct fw_page *find_fw_page(struct mlx5_core_dev *dev, u64 addr,
153 struct fw_page *result = NULL;
154 struct rb_root *root;
158 root = xa_load(&dev->priv.page_root_xa, function);
159 if (WARN_ON_ONCE(!root))
165 tfp = rb_entry(tmp, struct fw_page, rb_node);
166 if (tfp->addr < addr) {
168 } else if (tfp->addr > addr) {
179 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
180 s32 *npages, int boot)
182 u32 out[MLX5_ST_SZ_DW(query_pages_out)] = {};
183 u32 in[MLX5_ST_SZ_DW(query_pages_in)] = {};
186 MLX5_SET(query_pages_in, in, opcode, MLX5_CMD_OP_QUERY_PAGES);
187 MLX5_SET(query_pages_in, in, op_mod, boot ?
188 MLX5_QUERY_PAGES_IN_OP_MOD_BOOT_PAGES :
189 MLX5_QUERY_PAGES_IN_OP_MOD_INIT_PAGES);
190 MLX5_SET(query_pages_in, in, embedded_cpu_function, mlx5_core_is_ecpf(dev));
192 err = mlx5_cmd_exec_inout(dev, query_pages, in, out);
196 *npages = MLX5_GET(query_pages_out, out, num_pages);
197 *func_id = MLX5_GET(query_pages_out, out, function_id);
202 static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr, u32 function)
204 struct fw_page *fp = NULL;
205 struct fw_page *iter;
208 list_for_each_entry(iter, &dev->priv.free_list, list) {
209 if (iter->function != function)
214 if (list_empty(&dev->priv.free_list) || !fp)
217 n = find_first_bit(&fp->bitmask, 8 * sizeof(fp->bitmask));
218 if (n >= MLX5_NUM_4K_IN_PAGE) {
219 mlx5_core_warn(dev, "alloc 4k bug\n");
222 clear_bit(n, &fp->bitmask);
227 *addr = fp->addr + n * MLX5_ADAPTER_PAGE_SIZE;
232 #define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
234 static void free_fwp(struct mlx5_core_dev *dev, struct fw_page *fwp,
237 struct rb_root *root;
239 root = xa_load(&dev->priv.page_root_xa, fwp->function);
240 if (WARN_ON_ONCE(!root))
243 rb_erase(&fwp->rb_node, root);
245 list_del(&fwp->list);
246 dma_unmap_page(mlx5_core_dma_dev(dev), fwp->addr & MLX5_U64_4K_PAGE_MASK,
247 PAGE_SIZE, DMA_BIDIRECTIONAL);
248 __free_page(fwp->page);
252 static void free_4k(struct mlx5_core_dev *dev, u64 addr, u32 function)
257 fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK, function);
259 mlx5_core_warn_rl(dev, "page not found\n");
262 n = (addr & ~MLX5_U64_4K_PAGE_MASK) >> MLX5_ADAPTER_PAGE_SHIFT;
264 set_bit(n, &fwp->bitmask);
265 if (fwp->free_count == MLX5_NUM_4K_IN_PAGE)
266 free_fwp(dev, fwp, fwp->free_count != 1);
267 else if (fwp->free_count == 1)
268 list_add(&fwp->list, &dev->priv.free_list);
271 static int alloc_system_page(struct mlx5_core_dev *dev, u32 function)
273 struct device *device = mlx5_core_dma_dev(dev);
274 int nid = dev_to_node(device);
280 page = alloc_pages_node(nid, GFP_HIGHUSER, 0);
282 mlx5_core_warn(dev, "failed to allocate page\n");
286 addr = dma_map_page(device, page, 0, PAGE_SIZE, DMA_BIDIRECTIONAL);
287 if (dma_mapping_error(device, addr)) {
288 mlx5_core_warn(dev, "failed dma mapping page\n");
293 /* Firmware doesn't support page with physical address 0 */
299 err = insert_page(dev, addr, page, function);
301 mlx5_core_err(dev, "failed to track allocated page\n");
302 dma_unmap_page(device, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
310 dma_unmap_page(device, zero_addr, PAGE_SIZE,
316 static void page_notify_fail(struct mlx5_core_dev *dev, u16 func_id,
319 u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
322 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
323 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_CANT_GIVE);
324 MLX5_SET(manage_pages_in, in, function_id, func_id);
325 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
327 err = mlx5_cmd_exec_in(dev, manage_pages, in);
329 mlx5_core_warn(dev, "page notify failed func_id(%d) err(%d)\n",
333 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
334 int notify_fail, bool ec_function)
336 u32 function = get_function(func_id, ec_function);
337 u32 out[MLX5_ST_SZ_DW(manage_pages_out)] = {0};
338 int inlen = MLX5_ST_SZ_BYTES(manage_pages_in);
344 inlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_in, pas[0]);
345 in = kvzalloc(inlen, GFP_KERNEL);
348 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
352 for (i = 0; i < npages; i++) {
354 err = alloc_4k(dev, &addr, function);
357 err = alloc_system_page(dev, function);
363 MLX5_ARRAY_SET64(manage_pages_in, in, pas, i, addr);
366 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
367 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_GIVE);
368 MLX5_SET(manage_pages_in, in, function_id, func_id);
369 MLX5_SET(manage_pages_in, in, input_num_entries, npages);
370 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
372 err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
374 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n",
375 func_id, npages, err);
379 dev->priv.fw_pages += npages;
381 dev->priv.vfs_pages += npages;
382 else if (mlx5_core_is_ecpf(dev) && !ec_function)
383 dev->priv.host_pf_pages += npages;
385 mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x, err %d\n",
386 npages, ec_function, func_id, err);
392 for (i--; i >= 0; i--)
393 free_4k(dev, MLX5_GET64(manage_pages_in, in, pas[i]), function);
397 page_notify_fail(dev, func_id, ec_function);
401 static void release_all_pages(struct mlx5_core_dev *dev, u16 func_id,
404 u32 function = get_function(func_id, ec_function);
405 struct rb_root *root;
409 root = xa_load(&dev->priv.page_root_xa, function);
410 if (WARN_ON_ONCE(!root))
415 struct fw_page *fwp = rb_entry(p, struct fw_page, rb_node);
418 npages += (MLX5_NUM_4K_IN_PAGE - fwp->free_count);
419 free_fwp(dev, fwp, fwp->free_count);
422 dev->priv.fw_pages -= npages;
424 dev->priv.vfs_pages -= npages;
425 else if (mlx5_core_is_ecpf(dev) && !ec_function)
426 dev->priv.host_pf_pages -= npages;
428 mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x\n",
429 npages, ec_function, func_id);
432 static u32 fwp_fill_manage_pages_out(struct fw_page *fwp, u32 *out, u32 index,
438 for_each_clear_bit(n, &fwp->bitmask, MLX5_NUM_4K_IN_PAGE) {
439 MLX5_ARRAY_SET64(manage_pages_out, out, pas, index + pages_set,
440 fwp->addr + (n * MLX5_ADAPTER_PAGE_SIZE));
450 static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
451 u32 *in, int in_size, u32 *out, int out_size)
453 struct rb_root *root;
461 if (!mlx5_cmd_is_down(dev))
462 return mlx5_cmd_exec(dev, in, in_size, out, out_size);
464 /* No hard feelings, we want our pages back! */
465 npages = MLX5_GET(manage_pages_in, in, input_num_entries);
466 func_id = MLX5_GET(manage_pages_in, in, function_id);
467 ec_function = MLX5_GET(manage_pages_in, in, embedded_cpu_function);
469 root = xa_load(&dev->priv.page_root_xa, get_function(func_id, ec_function));
470 if (WARN_ON_ONCE(!root))
474 while (p && i < npages) {
475 fwp = rb_entry(p, struct fw_page, rb_node);
478 i += fwp_fill_manage_pages_out(fwp, out, i, npages - i);
481 MLX5_SET(manage_pages_out, out, output_num_entries, i);
485 static int reclaim_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
486 int *nclaimed, bool ec_function)
488 u32 function = get_function(func_id, ec_function);
489 int outlen = MLX5_ST_SZ_BYTES(manage_pages_out);
490 u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
499 outlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_out, pas[0]);
500 out = kvzalloc(outlen, GFP_KERNEL);
504 MLX5_SET(manage_pages_in, in, opcode, MLX5_CMD_OP_MANAGE_PAGES);
505 MLX5_SET(manage_pages_in, in, op_mod, MLX5_PAGES_TAKE);
506 MLX5_SET(manage_pages_in, in, function_id, func_id);
507 MLX5_SET(manage_pages_in, in, input_num_entries, npages);
508 MLX5_SET(manage_pages_in, in, embedded_cpu_function, ec_function);
510 mlx5_core_dbg(dev, "func 0x%x, npages %d, outlen %d\n",
511 func_id, npages, outlen);
512 err = reclaim_pages_cmd(dev, in, sizeof(in), out, outlen);
514 mlx5_core_err(dev, "failed reclaiming pages: err %d\n", err);
518 num_claimed = MLX5_GET(manage_pages_out, out, output_num_entries);
519 if (num_claimed > npages) {
520 mlx5_core_warn(dev, "fw returned %d, driver asked %d => corruption\n",
521 num_claimed, npages);
526 for (i = 0; i < num_claimed; i++)
527 free_4k(dev, MLX5_GET64(manage_pages_out, out, pas[i]), function);
530 *nclaimed = num_claimed;
532 dev->priv.fw_pages -= num_claimed;
534 dev->priv.vfs_pages -= num_claimed;
535 else if (mlx5_core_is_ecpf(dev) && !ec_function)
536 dev->priv.host_pf_pages -= num_claimed;
543 static void pages_work_handler(struct work_struct *work)
545 struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
546 struct mlx5_core_dev *dev = req->dev;
549 if (req->release_all)
550 release_all_pages(dev, req->func_id, req->ec_function);
551 else if (req->npages < 0)
552 err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL,
554 else if (req->npages > 0)
555 err = give_pages(dev, req->func_id, req->npages, 1, req->ec_function);
558 mlx5_core_warn(dev, "%s fail %d\n",
559 req->npages < 0 ? "reclaim" : "give", err);
565 EC_FUNCTION_MASK = 0x8000,
566 RELEASE_ALL_PAGES_MASK = 0x4000,
569 static int req_pages_handler(struct notifier_block *nb,
570 unsigned long type, void *data)
572 struct mlx5_pages_req *req;
573 struct mlx5_core_dev *dev;
574 struct mlx5_priv *priv;
575 struct mlx5_eqe *eqe;
581 priv = mlx5_nb_cof(nb, struct mlx5_priv, pg_nb);
582 dev = container_of(priv, struct mlx5_core_dev, priv);
585 func_id = be16_to_cpu(eqe->data.req_pages.func_id);
586 npages = be32_to_cpu(eqe->data.req_pages.num_pages);
587 ec_function = be16_to_cpu(eqe->data.req_pages.ec_function) & EC_FUNCTION_MASK;
588 release_all = be16_to_cpu(eqe->data.req_pages.ec_function) &
589 RELEASE_ALL_PAGES_MASK;
590 mlx5_core_dbg(dev, "page request for func 0x%x, npages %d, release_all %d\n",
591 func_id, npages, release_all);
592 req = kzalloc(sizeof(*req), GFP_ATOMIC);
594 mlx5_core_warn(dev, "failed to allocate pages request\n");
599 req->func_id = func_id;
600 req->npages = npages;
601 req->ec_function = ec_function;
602 req->release_all = release_all;
603 INIT_WORK(&req->work, pages_work_handler);
604 queue_work(dev->priv.pg_wq, &req->work);
608 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
614 err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
618 mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
619 npages, boot ? "boot" : "init", func_id);
621 return give_pages(dev, func_id, npages, 0, mlx5_core_is_ecpf(dev));
625 MLX5_BLKS_FOR_RECLAIM_PAGES = 12
628 static int optimal_reclaimed_pages(void)
630 struct mlx5_cmd_prot_block *block;
631 struct mlx5_cmd_layout *lay;
634 ret = (sizeof(lay->out) + MLX5_BLKS_FOR_RECLAIM_PAGES * sizeof(block->data) -
635 MLX5_ST_SZ_BYTES(manage_pages_out)) /
636 MLX5_FLD_SZ_BYTES(manage_pages_out, pas[0]);
641 static int mlx5_reclaim_root_pages(struct mlx5_core_dev *dev,
642 struct rb_root *root, u16 func_id)
644 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
646 while (!RB_EMPTY_ROOT(root)) {
650 err = reclaim_pages(dev, func_id, optimal_reclaimed_pages(),
651 &nclaimed, mlx5_core_is_ecpf(dev));
653 mlx5_core_warn(dev, "failed reclaiming pages (%d) for func id 0x%x\n",
659 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
661 if (time_after(jiffies, end)) {
662 mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
670 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
672 struct rb_root *root;
676 xa_for_each(&dev->priv.page_root_xa, id, entry) {
678 mlx5_reclaim_root_pages(dev, root, id);
679 xa_erase(&dev->priv.page_root_xa, id);
683 WARN_ON(!xa_empty(&dev->priv.page_root_xa));
685 WARN(dev->priv.fw_pages,
686 "FW pages counter is %d after reclaiming all pages\n",
688 WARN(dev->priv.vfs_pages,
689 "VFs FW pages counter is %d after reclaiming all pages\n",
690 dev->priv.vfs_pages);
691 WARN(dev->priv.host_pf_pages,
692 "External host PF FW pages counter is %d after reclaiming all pages\n",
693 dev->priv.host_pf_pages);
698 int mlx5_pagealloc_init(struct mlx5_core_dev *dev)
700 INIT_LIST_HEAD(&dev->priv.free_list);
701 dev->priv.pg_wq = create_singlethread_workqueue("mlx5_page_allocator");
702 if (!dev->priv.pg_wq)
705 xa_init(&dev->priv.page_root_xa);
710 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
712 xa_destroy(&dev->priv.page_root_xa);
713 destroy_workqueue(dev->priv.pg_wq);
716 void mlx5_pagealloc_start(struct mlx5_core_dev *dev)
718 MLX5_NB_INIT(&dev->priv.pg_nb, req_pages_handler, PAGE_REQUEST);
719 mlx5_eq_notifier_register(dev, &dev->priv.pg_nb);
722 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
724 mlx5_eq_notifier_unregister(dev, &dev->priv.pg_nb);
725 flush_workqueue(dev->priv.pg_wq);
728 int mlx5_wait_for_pages(struct mlx5_core_dev *dev, int *pages)
730 unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
731 int prev_pages = *pages;
733 /* In case of internal error we will free the pages manually later */
734 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
735 mlx5_core_warn(dev, "Skipping wait for vf pages stage");
739 mlx5_core_dbg(dev, "Waiting for %d pages\n", prev_pages);
741 if (time_after(jiffies, end)) {
742 mlx5_core_warn(dev, "aborting while there are %d pending pages\n", *pages);
745 if (*pages < prev_pages) {
746 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
752 mlx5_core_dbg(dev, "All pages received\n");