drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c

   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/delay.h>
  37 #include <linux/mlx5/driver.h>
  38 #include "mlx5_core.h"
  39 #include "lib/eq.h"
  40
  41 enum {
  42         MLX5_PAGES_CANT_GIVE    = 0,
  43         MLX5_PAGES_GIVE         = 1,
  44         MLX5_PAGES_TAKE         = 2
  45 };
  46
  47 struct mlx5_pages_req {
  48         struct mlx5_core_dev *dev;
  49         u16     func_id;
  50         u8      ec_function;
  51         s32     npages;
  52         struct work_struct work;
  53         u8      release_all;
  54 };
  55
  56 struct fw_page {
  57         struct rb_node          rb_node;
  58         u64                     addr;
  59         struct page            *page;
  60         u16                     func_id;
  61         unsigned long           bitmask;
  62         struct list_head        list;
  63         unsigned                free_count;
  64 };
  65
  66 enum {
  67         MAX_RECLAIM_TIME_MSECS  = 5000,
  68         MAX_RECLAIM_VFS_PAGES_TIME_MSECS = 2 * 1000 * 60,
  69 };
  70
  71 enum {
  72         MLX5_MAX_RECLAIM_TIME_MILI      = 5000,
  73         MLX5_NUM_4K_IN_PAGE             = PAGE_SIZE / MLX5_ADAPTER_PAGE_SIZE,
  74 };
  75
  76 static int insert_page(struct mlx5_core_dev *dev, u64 addr, struct page *page, u16 func_id)
  77 {
  78         struct rb_root *root = &dev->priv.page_root;
  79         struct rb_node **new = &root->rb_node;
  80         struct rb_node *parent = NULL;
  81         struct fw_page *nfp;
  82         struct fw_page *tfp;
  83         int i;
  84
  85         while (*new) {
  86                 parent = *new;
  87                 tfp = rb_entry(parent, struct fw_page, rb_node);
  88                 if (tfp->addr < addr)
  89                         new = &parent->rb_left;
  90                 else if (tfp->addr > addr)
  91                         new = &parent->rb_right;
  92                 else
  93                         return -EEXIST;
  94         }
  95
  96         nfp = kzalloc(sizeof(*nfp), GFP_KERNEL);
  97         if (!nfp)
  98                 return -ENOMEM;
  99
 100         nfp->addr = addr;
 101         nfp->page = page;
 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);
 106
 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);
 110
 111         return 0;
 112 }
 113
 114 static struct fw_page *find_fw_page(struct mlx5_core_dev *dev, u64 addr)
 115 {
 116         struct rb_root *root = &dev->priv.page_root;
 117         struct rb_node *tmp = root->rb_node;
 118         struct fw_page *result = NULL;
 119         struct fw_page *tfp;
 120
 121         while (tmp) {
 122                 tfp = rb_entry(tmp, struct fw_page, rb_node);
 123                 if (tfp->addr < addr) {
 124                         tmp = tmp->rb_left;
 125                 } else if (tfp->addr > addr) {
 126                         tmp = tmp->rb_right;
 127                 } else {
 128                         result = tfp;
 129                         break;
 130                 }
 131         }
 132
 133         return result;
 134 }
 135
 136 static int mlx5_cmd_query_pages(struct mlx5_core_dev *dev, u16 *func_id,
 137                                 s32 *npages, int boot)
 138 {
 139         u32 out[MLX5_ST_SZ_DW(query_pages_out)] = {};
 140         u32 in[MLX5_ST_SZ_DW(query_pages_in)] = {};
 141         int err;
 142
 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));
 148
 149         err = mlx5_cmd_exec_inout(dev, query_pages, in, out);
 150         if (err)
 151                 return err;
 152
 153         *npages = MLX5_GET(query_pages_out, out, num_pages);
 154         *func_id = MLX5_GET(query_pages_out, out, function_id);
 155
 156         return err;
 157 }
 158
 159 static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr, u16 func_id)
 160 {
 161         struct fw_page *fp = NULL;
 162         struct fw_page *iter;
 163         unsigned n;
 164
 165         list_for_each_entry(iter, &dev->priv.free_list, list) {
 166                 if (iter->func_id != func_id)
 167                         continue;
 168                 fp = iter;
 169         }
 170
 171         if (list_empty(&dev->priv.free_list) || !fp)
 172                 return -ENOMEM;
 173
 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");
 177                 return -ENOENT;
 178         }
 179         clear_bit(n, &fp->bitmask);
 180         fp->free_count--;
 181         if (!fp->free_count)
 182                 list_del(&fp->list);
 183
 184         *addr = fp->addr + n * MLX5_ADAPTER_PAGE_SIZE;
 185
 186         return 0;
 187 }
 188
 189 #define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
 190
 191 static void free_fwp(struct mlx5_core_dev *dev, struct fw_page *fwp,
 192                      bool in_free_list)
 193 {
 194         rb_erase(&fwp->rb_node, &dev->priv.page_root);
 195         if (in_free_list)
 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);
 200         kfree(fwp);
 201 }
 202
 203 static void free_4k(struct mlx5_core_dev *dev, u64 addr)
 204 {
 205         struct fw_page *fwp;
 206         int n;
 207
 208         fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
 209         if (!fwp) {
 210                 mlx5_core_warn_rl(dev, "page not found\n");
 211                 return;
 212         }
 213         n = (addr & ~MLX5_U64_4K_PAGE_MASK) >> MLX5_ADAPTER_PAGE_SHIFT;
 214         fwp->free_count++;
 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);
 220 }
 221
 222 static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
 223 {
 224         struct device *device = dev->device;
 225         int nid = dev_to_node(device);
 226         struct page *page;
 227         u64 zero_addr = 1;
 228         u64 addr;
 229         int err;
 230
 231         page = alloc_pages_node(nid, GFP_HIGHUSER, 0);
 232         if (!page) {
 233                 mlx5_core_warn(dev, "failed to allocate page\n");
 234                 return -ENOMEM;
 235         }
 236 map:
 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");
 240                 err = -ENOMEM;
 241                 goto err_mapping;
 242         }
 243
 244         /* Firmware doesn't support page with physical address 0 */
 245         if (addr == 0) {
 246                 zero_addr = addr;
 247                 goto map;
 248         }
 249
 250         err = insert_page(dev, addr, page, func_id);
 251         if (err) {
 252                 mlx5_core_err(dev, "failed to track allocated page\n");
 253                 dma_unmap_page(device, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
 254         }
 255
 256 err_mapping:
 257         if (err)
 258                 __free_page(page);
 259
 260         if (zero_addr == 0)
 261                 dma_unmap_page(device, zero_addr, PAGE_SIZE,
 262                                DMA_BIDIRECTIONAL);
 263
 264         return err;
 265 }
 266
 267 static void page_notify_fail(struct mlx5_core_dev *dev, u16 func_id,
 268                              bool ec_function)
 269 {
 270         u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
 271         int err;
 272
 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);
 277
 278         err = mlx5_cmd_exec_in(dev, manage_pages, in);
 279         if (err)
 280                 mlx5_core_warn(dev, "page notify failed func_id(%d) err(%d)\n",
 281                                func_id, err);
 282 }
 283
 284 static int give_pages(struct mlx5_core_dev *dev, u16 func_id, int npages,
 285                       int notify_fail, bool ec_function)
 286 {
 287         u32 out[MLX5_ST_SZ_DW(manage_pages_out)] = {0};
 288         int inlen = MLX5_ST_SZ_BYTES(manage_pages_in);
 289         u64 addr;
 290         int err;
 291         u32 *in;
 292         int i;
 293
 294         inlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_in, pas[0]);
 295         in = kvzalloc(inlen, GFP_KERNEL);
 296         if (!in) {
 297                 err = -ENOMEM;
 298                 mlx5_core_warn(dev, "vzalloc failed %d\n", inlen);
 299                 goto out_free;
 300         }
 301
 302         for (i = 0; i < npages; i++) {
 303 retry:
 304                 err = alloc_4k(dev, &addr, func_id);
 305                 if (err) {
 306                         if (err == -ENOMEM)
 307                                 err = alloc_system_page(dev, func_id);
 308                         if (err)
 309                                 goto out_4k;
 310
 311                         goto retry;
 312                 }
 313                 MLX5_ARRAY_SET64(manage_pages_in, in, pas, i, addr);
 314         }
 315
 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);
 321
 322         err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
 323         if (err) {
 324                 mlx5_core_warn(dev, "func_id 0x%x, npages %d, err %d\n",
 325                                func_id, npages, err);
 326                 goto out_4k;
 327         }
 328
 329         dev->priv.fw_pages += npages;
 330         if (func_id)
 331                 dev->priv.vfs_pages += npages;
 332         else if (mlx5_core_is_ecpf(dev) && !ec_function)
 333                 dev->priv.peer_pf_pages += npages;
 334
 335         mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x, err %d\n",
 336                       npages, ec_function, func_id, err);
 337
 338         kvfree(in);
 339         return 0;
 340
 341 out_4k:
 342         for (i--; i >= 0; i--)
 343                 free_4k(dev, MLX5_GET64(manage_pages_in, in, pas[i]));
 344 out_free:
 345         kvfree(in);
 346         if (notify_fail)
 347                 page_notify_fail(dev, func_id, ec_function);
 348         return err;
 349 }
 350
 351 static void release_all_pages(struct mlx5_core_dev *dev, u32 func_id,
 352                               bool ec_function)
 353 {
 354         struct rb_node *p;
 355         int npages = 0;
 356
 357         p = rb_first(&dev->priv.page_root);
 358         while (p) {
 359                 struct fw_page *fwp = rb_entry(p, struct fw_page, rb_node);
 360
 361                 p = rb_next(p);
 362                 if (fwp->func_id != func_id)
 363                         continue;
 364                 npages += (MLX5_NUM_4K_IN_PAGE - fwp->free_count);
 365                 free_fwp(dev, fwp, fwp->free_count);
 366         }
 367
 368         dev->priv.fw_pages -= npages;
 369         if (func_id)
 370                 dev->priv.vfs_pages -= npages;
 371         else if (mlx5_core_is_ecpf(dev) && !ec_function)
 372                 dev->priv.peer_pf_pages -= npages;
 373
 374         mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x\n",
 375                       npages, ec_function, func_id);
 376 }
 377
 378 static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
 379                              u32 *in, int in_size, u32 *out, int out_size)
 380 {
 381         struct fw_page *fwp;
 382         struct rb_node *p;
 383         u32 func_id;
 384         u32 npages;
 385         u32 i = 0;
 386
 387         if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
 388                 return mlx5_cmd_exec(dev, in, in_size, out, out_size);
 389
 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);
 393
 394         p = rb_first(&dev->priv.page_root);
 395         while (p && i < npages) {
 396                 fwp = rb_entry(p, struct fw_page, rb_node);
 397                 p = rb_next(p);
 398                 if (fwp->func_id != func_id)
 399                         continue;
 400
 401                 MLX5_ARRAY_SET64(manage_pages_out, out, pas, i, fwp->addr);
 402                 i++;
 403         }
 404
 405         MLX5_SET(manage_pages_out, out, output_num_entries, i);
 406         return 0;
 407 }
 408
 409 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
 410                          int *nclaimed, bool ec_function)
 411 {
 412         int outlen = MLX5_ST_SZ_BYTES(manage_pages_out);
 413         u32 in[MLX5_ST_SZ_DW(manage_pages_in)] = {};
 414         int num_claimed;
 415         u32 *out;
 416         int err;
 417         int i;
 418
 419         if (nclaimed)
 420                 *nclaimed = 0;
 421
 422         outlen += npages * MLX5_FLD_SZ_BYTES(manage_pages_out, pas[0]);
 423         out = kvzalloc(outlen, GFP_KERNEL);
 424         if (!out)
 425                 return -ENOMEM;
 426
 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);
 432
 433         mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
 434         err = reclaim_pages_cmd(dev, in, sizeof(in), out, outlen);
 435         if (err) {
 436                 mlx5_core_err(dev, "failed reclaiming pages: err %d\n", err);
 437                 goto out_free;
 438         }
 439
 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);
 444                 err = -EINVAL;
 445                 goto out_free;
 446         }
 447
 448         for (i = 0; i < num_claimed; i++)
 449                 free_4k(dev, MLX5_GET64(manage_pages_out, out, pas[i]));
 450
 451         if (nclaimed)
 452                 *nclaimed = num_claimed;
 453
 454         dev->priv.fw_pages -= num_claimed;
 455         if (func_id)
 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;
 459
 460 out_free:
 461         kvfree(out);
 462         return err;
 463 }
 464
 465 static void pages_work_handler(struct work_struct *work)
 466 {
 467         struct mlx5_pages_req *req = container_of(work, struct mlx5_pages_req, work);
 468         struct mlx5_core_dev *dev = req->dev;
 469         int err = 0;
 470
 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,
 475                                     req->ec_function);
 476         else if (req->npages > 0)
 477                 err = give_pages(dev, req->func_id, req->npages, 1, req->ec_function);
 478
 479         if (err)
 480                 mlx5_core_warn(dev, "%s fail %d\n",
 481                                req->npages < 0 ? "reclaim" : "give", err);
 482
 483         kfree(req);
 484 }
 485
 486 enum {
 487         EC_FUNCTION_MASK = 0x8000,
 488         RELEASE_ALL_PAGES_MASK = 0x4000,
 489 };
 490
 491 static int req_pages_handler(struct notifier_block *nb,
 492                              unsigned long type, void *data)
 493 {
 494         struct mlx5_pages_req *req;
 495         struct mlx5_core_dev *dev;
 496         struct mlx5_priv *priv;
 497         struct mlx5_eqe *eqe;
 498         bool ec_function;
 499         bool release_all;
 500         u16 func_id;
 501         s32 npages;
 502
 503         priv = mlx5_nb_cof(nb, struct mlx5_priv, pg_nb);
 504         dev  = container_of(priv, struct mlx5_core_dev, priv);
 505         eqe  = data;
 506
 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);
 515         if (!req) {
 516                 mlx5_core_warn(dev, "failed to allocate pages request\n");
 517                 return NOTIFY_DONE;
 518         }
 519
 520         req->dev = dev;
 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);
 527         return NOTIFY_OK;
 528 }
 529
 530 int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot)
 531 {
 532         u16 uninitialized_var(func_id);
 533         s32 uninitialized_var(npages);
 534         int err;
 535
 536         err = mlx5_cmd_query_pages(dev, &func_id, &npages, boot);
 537         if (err)
 538                 return err;
 539
 540         mlx5_core_dbg(dev, "requested %d %s pages for func_id 0x%x\n",
 541                       npages, boot ? "boot" : "init", func_id);
 542
 543         return give_pages(dev, func_id, npages, 0, mlx5_core_is_ecpf(dev));
 544 }
 545
 546 enum {
 547         MLX5_BLKS_FOR_RECLAIM_PAGES = 12
 548 };
 549
 550 static int optimal_reclaimed_pages(void)
 551 {
 552         struct mlx5_cmd_prot_block *block;
 553         struct mlx5_cmd_layout *lay;
 554         int ret;
 555
 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]);
 559
 560         return ret;
 561 }
 562
 563 int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
 564 {
 565         unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
 566         struct fw_page *fwp;
 567         struct rb_node *p;
 568         int nclaimed = 0;
 569         int err = 0;
 570
 571         do {
 572                 p = rb_first(&dev->priv.page_root);
 573                 if (p) {
 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));
 578
 579                         if (err) {
 580                                 mlx5_core_warn(dev, "failed reclaiming pages (%d)\n",
 581                                                err);
 582                                 return err;
 583                         }
 584                         if (nclaimed)
 585                                 end = jiffies + msecs_to_jiffies(MAX_RECLAIM_TIME_MSECS);
 586                 }
 587                 if (time_after(jiffies, end)) {
 588                         mlx5_core_warn(dev, "FW did not return all pages. giving up...\n");
 589                         break;
 590                 }
 591         } while (p);
 592
 593         WARN(dev->priv.fw_pages,
 594              "FW pages counter is %d after reclaiming all pages\n",
 595              dev->priv.fw_pages);
 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);
 602
 603         return 0;
 604 }
 605
 606 int mlx5_pagealloc_init(struct mlx5_core_dev *dev)
 607 {
 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)
 612                 return -ENOMEM;
 613
 614         return 0;
 615 }
 616
 617 void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev)
 618 {
 619         destroy_workqueue(dev->priv.pg_wq);
 620 }
 621
 622 void mlx5_pagealloc_start(struct mlx5_core_dev *dev)
 623 {
 624         MLX5_NB_INIT(&dev->priv.pg_nb, req_pages_handler, PAGE_REQUEST);
 625         mlx5_eq_notifier_register(dev, &dev->priv.pg_nb);
 626 }
 627
 628 void mlx5_pagealloc_stop(struct mlx5_core_dev *dev)
 629 {
 630         mlx5_eq_notifier_unregister(dev, &dev->priv.pg_nb);
 631         flush_workqueue(dev->priv.pg_wq);
 632 }
 633
 634 int mlx5_wait_for_pages(struct mlx5_core_dev *dev, int *pages)
 635 {
 636         unsigned long end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
 637         int prev_pages = *pages;
 638
 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");
 642                 return 0;
 643         }
 644
 645         mlx5_core_dbg(dev, "Waiting for %d pages\n", prev_pages);
 646         while (*pages) {
 647                 if (time_after(jiffies, end)) {
 648                         mlx5_core_warn(dev, "aborting while there are %d pending pages\n", *pages);
 649                         return -ETIMEDOUT;
 650                 }
 651                 if (*pages < prev_pages) {
 652                         end = jiffies + msecs_to_jiffies(MAX_RECLAIM_VFS_PAGES_TIME_MSECS);
 653                         prev_pages = *pages;
 654                 }
 655                 msleep(50);
 656         }
 657
 658         mlx5_core_dbg(dev, "All pages received\n");
 659         return 0;
 660 }