net/rds/ib_rdma.c

   1 /*
   2  * Copyright (c) 2006 Oracle.  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/kernel.h>
  34 #include <linux/slab.h>
  35 #include <linux/rculist.h>
  36 #include <linux/llist.h>
  37
  38 #include "rds.h"
  39 #include "ib.h"
  40
  41 static DEFINE_PER_CPU(unsigned long, clean_list_grace);
  42 #define CLEAN_LIST_BUSY_BIT 0
  43
  44 /*
  45  * This is stored as mr->r_trans_private.
  46  */
  47 struct rds_ib_mr {
  48         struct rds_ib_device    *device;
  49         struct rds_ib_mr_pool   *pool;
  50         struct ib_fmr           *fmr;
  51
  52         struct llist_node       llnode;
  53
  54         /* unmap_list is for freeing */
  55         struct list_head        unmap_list;
  56         unsigned int            remap_count;
  57
  58         struct scatterlist      *sg;
  59         unsigned int            sg_len;
  60         u64                     *dma;
  61         int                     sg_dma_len;
  62 };
  63
  64 /*
  65  * Our own little FMR pool
  66  */
  67 struct rds_ib_mr_pool {
  68         struct mutex            flush_lock;             /* serialize fmr invalidate */
  69         struct delayed_work     flush_worker;           /* flush worker */
  70
  71         atomic_t                item_count;             /* total # of MRs */
  72         atomic_t                dirty_count;            /* # dirty of MRs */
  73
  74         struct llist_head       drop_list;              /* MRs that have reached their max_maps limit */
  75         struct llist_head       free_list;              /* unused MRs */
  76         struct llist_head       clean_list;             /* global unused & unamapped MRs */
  77         wait_queue_head_t       flush_wait;
  78
  79         atomic_t                free_pinned;            /* memory pinned by free MRs */
  80         unsigned long           max_items;
  81         unsigned long           max_items_soft;
  82         unsigned long           max_free_pinned;
  83         struct ib_fmr_attr      fmr_attr;
  84 };
  85
  86 static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool, int free_all, struct rds_ib_mr **);
  87 static void rds_ib_teardown_mr(struct rds_ib_mr *ibmr);
  88 static void rds_ib_mr_pool_flush_worker(struct work_struct *work);
  89
  90 static struct rds_ib_device *rds_ib_get_device(__be32 ipaddr)
  91 {
  92         struct rds_ib_device *rds_ibdev;
  93         struct rds_ib_ipaddr *i_ipaddr;
  94
  95         rcu_read_lock();
  96         list_for_each_entry_rcu(rds_ibdev, &rds_ib_devices, list) {
  97                 list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
  98                         if (i_ipaddr->ipaddr == ipaddr) {
  99                                 atomic_inc(&rds_ibdev->refcount);
 100                                 rcu_read_unlock();
 101                                 return rds_ibdev;
 102                         }
 103                 }
 104         }
 105         rcu_read_unlock();
 106
 107         return NULL;
 108 }
 109
 110 static int rds_ib_add_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
 111 {
 112         struct rds_ib_ipaddr *i_ipaddr;
 113
 114         i_ipaddr = kmalloc(sizeof *i_ipaddr, GFP_KERNEL);
 115         if (!i_ipaddr)
 116                 return -ENOMEM;
 117
 118         i_ipaddr->ipaddr = ipaddr;
 119
 120         spin_lock_irq(&rds_ibdev->spinlock);
 121         list_add_tail_rcu(&i_ipaddr->list, &rds_ibdev->ipaddr_list);
 122         spin_unlock_irq(&rds_ibdev->spinlock);
 123
 124         return 0;
 125 }
 126
 127 static void rds_ib_remove_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
 128 {
 129         struct rds_ib_ipaddr *i_ipaddr;
 130         struct rds_ib_ipaddr *to_free = NULL;
 131
 132
 133         spin_lock_irq(&rds_ibdev->spinlock);
 134         list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
 135                 if (i_ipaddr->ipaddr == ipaddr) {
 136                         list_del_rcu(&i_ipaddr->list);
 137                         to_free = i_ipaddr;
 138                         break;
 139                 }
 140         }
 141         spin_unlock_irq(&rds_ibdev->spinlock);
 142
 143         if (to_free) {
 144                 synchronize_rcu();
 145                 kfree(to_free);
 146         }
 147 }
 148
 149 int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
 150 {
 151         struct rds_ib_device *rds_ibdev_old;
 152
 153         rds_ibdev_old = rds_ib_get_device(ipaddr);
 154         if (rds_ibdev_old) {
 155                 rds_ib_remove_ipaddr(rds_ibdev_old, ipaddr);
 156                 rds_ib_dev_put(rds_ibdev_old);
 157         }
 158
 159         return rds_ib_add_ipaddr(rds_ibdev, ipaddr);
 160 }
 161
 162 void rds_ib_add_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
 163 {
 164         struct rds_ib_connection *ic = conn->c_transport_data;
 165
 166         /* conn was previously on the nodev_conns_list */
 167         spin_lock_irq(&ib_nodev_conns_lock);
 168         BUG_ON(list_empty(&ib_nodev_conns));
 169         BUG_ON(list_empty(&ic->ib_node));
 170         list_del(&ic->ib_node);
 171
 172         spin_lock(&rds_ibdev->spinlock);
 173         list_add_tail(&ic->ib_node, &rds_ibdev->conn_list);
 174         spin_unlock(&rds_ibdev->spinlock);
 175         spin_unlock_irq(&ib_nodev_conns_lock);
 176
 177         ic->rds_ibdev = rds_ibdev;
 178         atomic_inc(&rds_ibdev->refcount);
 179 }
 180
 181 void rds_ib_remove_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
 182 {
 183         struct rds_ib_connection *ic = conn->c_transport_data;
 184
 185         /* place conn on nodev_conns_list */
 186         spin_lock(&ib_nodev_conns_lock);
 187
 188         spin_lock_irq(&rds_ibdev->spinlock);
 189         BUG_ON(list_empty(&ic->ib_node));
 190         list_del(&ic->ib_node);
 191         spin_unlock_irq(&rds_ibdev->spinlock);
 192
 193         list_add_tail(&ic->ib_node, &ib_nodev_conns);
 194
 195         spin_unlock(&ib_nodev_conns_lock);
 196
 197         ic->rds_ibdev = NULL;
 198         rds_ib_dev_put(rds_ibdev);
 199 }
 200
 201 void rds_ib_destroy_nodev_conns(void)
 202 {
 203         struct rds_ib_connection *ic, *_ic;
 204         LIST_HEAD(tmp_list);
 205
 206         /* avoid calling conn_destroy with irqs off */
 207         spin_lock_irq(&ib_nodev_conns_lock);
 208         list_splice(&ib_nodev_conns, &tmp_list);
 209         spin_unlock_irq(&ib_nodev_conns_lock);
 210
 211         list_for_each_entry_safe(ic, _ic, &tmp_list, ib_node)
 212                 rds_conn_destroy(ic->conn);
 213 }
 214
 215 struct rds_ib_mr_pool *rds_ib_create_mr_pool(struct rds_ib_device *rds_ibdev)
 216 {
 217         struct rds_ib_mr_pool *pool;
 218
 219         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 220         if (!pool)
 221                 return ERR_PTR(-ENOMEM);
 222
 223         init_llist_head(&pool->free_list);
 224         init_llist_head(&pool->drop_list);
 225         init_llist_head(&pool->clean_list);
 226         mutex_init(&pool->flush_lock);
 227         init_waitqueue_head(&pool->flush_wait);
 228         INIT_DELAYED_WORK(&pool->flush_worker, rds_ib_mr_pool_flush_worker);
 229
 230         pool->fmr_attr.max_pages = fmr_message_size;
 231         pool->fmr_attr.max_maps = rds_ibdev->fmr_max_remaps;
 232         pool->fmr_attr.page_shift = PAGE_SHIFT;
 233         pool->max_free_pinned = rds_ibdev->max_fmrs * fmr_message_size / 4;
 234
 235         /* We never allow more than max_items MRs to be allocated.
 236          * When we exceed more than max_items_soft, we start freeing
 237          * items more aggressively.
 238          * Make sure that max_items > max_items_soft > max_items / 2
 239          */
 240         pool->max_items_soft = rds_ibdev->max_fmrs * 3 / 4;
 241         pool->max_items = rds_ibdev->max_fmrs;
 242
 243         return pool;
 244 }
 245
 246 void rds_ib_get_mr_info(struct rds_ib_device *rds_ibdev, struct rds_info_rdma_connection *iinfo)
 247 {
 248         struct rds_ib_mr_pool *pool = rds_ibdev->mr_pool;
 249
 250         iinfo->rdma_mr_max = pool->max_items;
 251         iinfo->rdma_mr_size = pool->fmr_attr.max_pages;
 252 }
 253
 254 void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *pool)
 255 {
 256         cancel_delayed_work_sync(&pool->flush_worker);
 257         rds_ib_flush_mr_pool(pool, 1, NULL);
 258         WARN_ON(atomic_read(&pool->item_count));
 259         WARN_ON(atomic_read(&pool->free_pinned));
 260         kfree(pool);
 261 }
 262
 263 static inline struct rds_ib_mr *rds_ib_reuse_fmr(struct rds_ib_mr_pool *pool)
 264 {
 265         struct rds_ib_mr *ibmr = NULL;
 266         struct llist_node *ret;
 267         unsigned long *flag;
 268
 269         preempt_disable();
 270         flag = &__get_cpu_var(clean_list_grace);
 271         set_bit(CLEAN_LIST_BUSY_BIT, flag);
 272         ret = llist_del_first(&pool->clean_list);
 273         if (ret)
 274                 ibmr = llist_entry(ret, struct rds_ib_mr, llnode);
 275
 276         clear_bit(CLEAN_LIST_BUSY_BIT, flag);
 277         preempt_enable();
 278         return ibmr;
 279 }
 280
 281 static inline void wait_clean_list_grace(void)
 282 {
 283         int cpu;
 284         unsigned long *flag;
 285
 286         for_each_online_cpu(cpu) {
 287                 flag = &per_cpu(clean_list_grace, cpu);
 288                 while (test_bit(CLEAN_LIST_BUSY_BIT, flag))
 289                         cpu_relax();
 290         }
 291 }
 292
 293 static struct rds_ib_mr *rds_ib_alloc_fmr(struct rds_ib_device *rds_ibdev)
 294 {
 295         struct rds_ib_mr_pool *pool = rds_ibdev->mr_pool;
 296         struct rds_ib_mr *ibmr = NULL;
 297         int err = 0, iter = 0;
 298
 299         if (atomic_read(&pool->dirty_count) >= pool->max_items / 10)
 300                 schedule_delayed_work(&pool->flush_worker, 10);
 301
 302         while (1) {
 303                 ibmr = rds_ib_reuse_fmr(pool);
 304                 if (ibmr)
 305                         return ibmr;
 306
 307                 /* No clean MRs - now we have the choice of either
 308                  * allocating a fresh MR up to the limit imposed by the
 309                  * driver, or flush any dirty unused MRs.
 310                  * We try to avoid stalling in the send path if possible,
 311                  * so we allocate as long as we're allowed to.
 312                  *
 313                  * We're fussy with enforcing the FMR limit, though. If the driver
 314                  * tells us we can't use more than N fmrs, we shouldn't start
 315                  * arguing with it */
 316                 if (atomic_inc_return(&pool->item_count) <= pool->max_items)
 317                         break;
 318
 319                 atomic_dec(&pool->item_count);
 320
 321                 if (++iter > 2) {
 322                         rds_ib_stats_inc(s_ib_rdma_mr_pool_depleted);
 323                         return ERR_PTR(-EAGAIN);
 324                 }
 325
 326                 /* We do have some empty MRs. Flush them out. */
 327                 rds_ib_stats_inc(s_ib_rdma_mr_pool_wait);
 328                 rds_ib_flush_mr_pool(pool, 0, &ibmr);
 329                 if (ibmr)
 330                         return ibmr;
 331         }
 332
 333         ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL, rdsibdev_to_node(rds_ibdev));
 334         if (!ibmr) {
 335                 err = -ENOMEM;
 336                 goto out_no_cigar;
 337         }
 338
 339         memset(ibmr, 0, sizeof(*ibmr));
 340
 341         ibmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
 342                         (IB_ACCESS_LOCAL_WRITE |
 343                          IB_ACCESS_REMOTE_READ |
 344                          IB_ACCESS_REMOTE_WRITE|
 345                          IB_ACCESS_REMOTE_ATOMIC),
 346                         &pool->fmr_attr);
 347         if (IS_ERR(ibmr->fmr)) {
 348                 err = PTR_ERR(ibmr->fmr);
 349                 ibmr->fmr = NULL;
 350                 printk(KERN_WARNING "RDS/IB: ib_alloc_fmr failed (err=%d)\n", err);
 351                 goto out_no_cigar;
 352         }
 353
 354         rds_ib_stats_inc(s_ib_rdma_mr_alloc);
 355         return ibmr;
 356
 357 out_no_cigar:
 358         if (ibmr) {
 359                 if (ibmr->fmr)
 360                         ib_dealloc_fmr(ibmr->fmr);
 361                 kfree(ibmr);
 362         }
 363         atomic_dec(&pool->item_count);
 364         return ERR_PTR(err);
 365 }
 366
 367 static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr,
 368                struct scatterlist *sg, unsigned int nents)
 369 {
 370         struct ib_device *dev = rds_ibdev->dev;
 371         struct scatterlist *scat = sg;
 372         u64 io_addr = 0;
 373         u64 *dma_pages;
 374         u32 len;
 375         int page_cnt, sg_dma_len;
 376         int i, j;
 377         int ret;
 378
 379         sg_dma_len = ib_dma_map_sg(dev, sg, nents,
 380                                  DMA_BIDIRECTIONAL);
 381         if (unlikely(!sg_dma_len)) {
 382                 printk(KERN_WARNING "RDS/IB: dma_map_sg failed!\n");
 383                 return -EBUSY;
 384         }
 385
 386         len = 0;
 387         page_cnt = 0;
 388
 389         for (i = 0; i < sg_dma_len; ++i) {
 390                 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
 391                 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
 392
 393                 if (dma_addr & ~PAGE_MASK) {
 394                         if (i > 0)
 395                                 return -EINVAL;
 396                         else
 397                                 ++page_cnt;
 398                 }
 399                 if ((dma_addr + dma_len) & ~PAGE_MASK) {
 400                         if (i < sg_dma_len - 1)
 401                                 return -EINVAL;
 402                         else
 403                                 ++page_cnt;
 404                 }
 405
 406                 len += dma_len;
 407         }
 408
 409         page_cnt += len >> PAGE_SHIFT;
 410         if (page_cnt > fmr_message_size)
 411                 return -EINVAL;
 412
 413         dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
 414                                  rdsibdev_to_node(rds_ibdev));
 415         if (!dma_pages)
 416                 return -ENOMEM;
 417
 418         page_cnt = 0;
 419         for (i = 0; i < sg_dma_len; ++i) {
 420                 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
 421                 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
 422
 423                 for (j = 0; j < dma_len; j += PAGE_SIZE)
 424                         dma_pages[page_cnt++] =
 425                                 (dma_addr & PAGE_MASK) + j;
 426         }
 427
 428         ret = ib_map_phys_fmr(ibmr->fmr,
 429                                    dma_pages, page_cnt, io_addr);
 430         if (ret)
 431                 goto out;
 432
 433         /* Success - we successfully remapped the MR, so we can
 434          * safely tear down the old mapping. */
 435         rds_ib_teardown_mr(ibmr);
 436
 437         ibmr->sg = scat;
 438         ibmr->sg_len = nents;
 439         ibmr->sg_dma_len = sg_dma_len;
 440         ibmr->remap_count++;
 441
 442         rds_ib_stats_inc(s_ib_rdma_mr_used);
 443         ret = 0;
 444
 445 out:
 446         kfree(dma_pages);
 447
 448         return ret;
 449 }
 450
 451 void rds_ib_sync_mr(void *trans_private, int direction)
 452 {
 453         struct rds_ib_mr *ibmr = trans_private;
 454         struct rds_ib_device *rds_ibdev = ibmr->device;
 455
 456         switch (direction) {
 457         case DMA_FROM_DEVICE:
 458                 ib_dma_sync_sg_for_cpu(rds_ibdev->dev, ibmr->sg,
 459                         ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
 460                 break;
 461         case DMA_TO_DEVICE:
 462                 ib_dma_sync_sg_for_device(rds_ibdev->dev, ibmr->sg,
 463                         ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
 464                 break;
 465         }
 466 }
 467
 468 static void __rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
 469 {
 470         struct rds_ib_device *rds_ibdev = ibmr->device;
 471
 472         if (ibmr->sg_dma_len) {
 473                 ib_dma_unmap_sg(rds_ibdev->dev,
 474                                 ibmr->sg, ibmr->sg_len,
 475                                 DMA_BIDIRECTIONAL);
 476                 ibmr->sg_dma_len = 0;
 477         }
 478
 479         /* Release the s/g list */
 480         if (ibmr->sg_len) {
 481                 unsigned int i;
 482
 483                 for (i = 0; i < ibmr->sg_len; ++i) {
 484                         struct page *page = sg_page(&ibmr->sg[i]);
 485
 486                         /* FIXME we need a way to tell a r/w MR
 487                          * from a r/o MR */
 488                         BUG_ON(irqs_disabled());
 489                         set_page_dirty(page);
 490                         put_page(page);
 491                 }
 492                 kfree(ibmr->sg);
 493
 494                 ibmr->sg = NULL;
 495                 ibmr->sg_len = 0;
 496         }
 497 }
 498
 499 static void rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
 500 {
 501         unsigned int pinned = ibmr->sg_len;
 502
 503         __rds_ib_teardown_mr(ibmr);
 504         if (pinned) {
 505                 struct rds_ib_device *rds_ibdev = ibmr->device;
 506                 struct rds_ib_mr_pool *pool = rds_ibdev->mr_pool;
 507
 508                 atomic_sub(pinned, &pool->free_pinned);
 509         }
 510 }
 511
 512 static inline unsigned int rds_ib_flush_goal(struct rds_ib_mr_pool *pool, int free_all)
 513 {
 514         unsigned int item_count;
 515
 516         item_count = atomic_read(&pool->item_count);
 517         if (free_all)
 518                 return item_count;
 519
 520         return 0;
 521 }
 522
 523 /*
 524  * given an llist of mrs, put them all into the list_head for more processing
 525  */
 526 static void llist_append_to_list(struct llist_head *llist, struct list_head *list)
 527 {
 528         struct rds_ib_mr *ibmr;
 529         struct llist_node *node;
 530         struct llist_node *next;
 531
 532         node = llist_del_all(llist);
 533         while (node) {
 534                 next = node->next;
 535                 ibmr = llist_entry(node, struct rds_ib_mr, llnode);
 536                 list_add_tail(&ibmr->unmap_list, list);
 537                 node = next;
 538         }
 539 }
 540
 541 /*
 542  * this takes a list head of mrs and turns it into linked llist nodes
 543  * of clusters.  Each cluster has linked llist nodes of
 544  * MR_CLUSTER_SIZE mrs that are ready for reuse.
 545  */
 546 static void list_to_llist_nodes(struct rds_ib_mr_pool *pool,
 547                                 struct list_head *list,
 548                                 struct llist_node **nodes_head,
 549                                 struct llist_node **nodes_tail)
 550 {
 551         struct rds_ib_mr *ibmr;
 552         struct llist_node *cur = NULL;
 553         struct llist_node **next = nodes_head;
 554
 555         list_for_each_entry(ibmr, list, unmap_list) {
 556                 cur = &ibmr->llnode;
 557                 *next = cur;
 558                 next = &cur->next;
 559         }
 560         *next = NULL;
 561         *nodes_tail = cur;
 562 }
 563
 564 /*
 565  * Flush our pool of MRs.
 566  * At a minimum, all currently unused MRs are unmapped.
 567  * If the number of MRs allocated exceeds the limit, we also try
 568  * to free as many MRs as needed to get back to this limit.
 569  */
 570 static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
 571                                 int free_all, struct rds_ib_mr **ibmr_ret)
 572 {
 573         struct rds_ib_mr *ibmr, *next;
 574         struct llist_node *clean_nodes;
 575         struct llist_node *clean_tail;
 576         LIST_HEAD(unmap_list);
 577         LIST_HEAD(fmr_list);
 578         unsigned long unpinned = 0;
 579         unsigned int nfreed = 0, ncleaned = 0, free_goal;
 580         int ret = 0;
 581
 582         rds_ib_stats_inc(s_ib_rdma_mr_pool_flush);
 583
 584         if (ibmr_ret) {
 585                 DEFINE_WAIT(wait);
 586                 while(!mutex_trylock(&pool->flush_lock)) {
 587                         ibmr = rds_ib_reuse_fmr(pool);
 588                         if (ibmr) {
 589                                 *ibmr_ret = ibmr;
 590                                 finish_wait(&pool->flush_wait, &wait);
 591                                 goto out_nolock;
 592                         }
 593
 594                         prepare_to_wait(&pool->flush_wait, &wait,
 595                                         TASK_UNINTERRUPTIBLE);
 596                         if (llist_empty(&pool->clean_list))
 597                                 schedule();
 598
 599                         ibmr = rds_ib_reuse_fmr(pool);
 600                         if (ibmr) {
 601                                 *ibmr_ret = ibmr;
 602                                 finish_wait(&pool->flush_wait, &wait);
 603                                 goto out_nolock;
 604                         }
 605                 }
 606                 finish_wait(&pool->flush_wait, &wait);
 607         } else
 608                 mutex_lock(&pool->flush_lock);
 609
 610         if (ibmr_ret) {
 611                 ibmr = rds_ib_reuse_fmr(pool);
 612                 if (ibmr) {
 613                         *ibmr_ret = ibmr;
 614                         goto out;
 615                 }
 616         }
 617
 618         /* Get the list of all MRs to be dropped. Ordering matters -
 619          * we want to put drop_list ahead of free_list.
 620          */
 621         llist_append_to_list(&pool->drop_list, &unmap_list);
 622         llist_append_to_list(&pool->free_list, &unmap_list);
 623         if (free_all)
 624                 llist_append_to_list(&pool->clean_list, &unmap_list);
 625
 626         free_goal = rds_ib_flush_goal(pool, free_all);
 627
 628         if (list_empty(&unmap_list))
 629                 goto out;
 630
 631         /* String all ib_mr's onto one list and hand them to ib_unmap_fmr */
 632         list_for_each_entry(ibmr, &unmap_list, unmap_list)
 633                 list_add(&ibmr->fmr->list, &fmr_list);
 634
 635         ret = ib_unmap_fmr(&fmr_list);
 636         if (ret)
 637                 printk(KERN_WARNING "RDS/IB: ib_unmap_fmr failed (err=%d)\n", ret);
 638
 639         /* Now we can destroy the DMA mapping and unpin any pages */
 640         list_for_each_entry_safe(ibmr, next, &unmap_list, unmap_list) {
 641                 unpinned += ibmr->sg_len;
 642                 __rds_ib_teardown_mr(ibmr);
 643                 if (nfreed < free_goal || ibmr->remap_count >= pool->fmr_attr.max_maps) {
 644                         rds_ib_stats_inc(s_ib_rdma_mr_free);
 645                         list_del(&ibmr->unmap_list);
 646                         ib_dealloc_fmr(ibmr->fmr);
 647                         kfree(ibmr);
 648                         nfreed++;
 649                 }
 650                 ncleaned++;
 651         }
 652
 653         if (!list_empty(&unmap_list)) {
 654                 /* we have to make sure that none of the things we're about
 655                  * to put on the clean list would race with other cpus trying
 656                  * to pull items off.  The llist would explode if we managed to
 657                  * remove something from the clean list and then add it back again
 658                  * while another CPU was spinning on that same item in llist_del_first.
 659                  *
 660                  * This is pretty unlikely, but just in case  wait for an llist grace period
 661                  * here before adding anything back into the clean list.
 662                  */
 663                 wait_clean_list_grace();
 664
 665                 list_to_llist_nodes(pool, &unmap_list, &clean_nodes, &clean_tail);
 666                 if (ibmr_ret)
 667                         *ibmr_ret = llist_entry(clean_nodes, struct rds_ib_mr, llnode);
 668
 669                 /* more than one entry in llist nodes */
 670                 if (clean_nodes->next)
 671                         llist_add_batch(clean_nodes->next, clean_tail, &pool->clean_list);
 672
 673         }
 674
 675         atomic_sub(unpinned, &pool->free_pinned);
 676         atomic_sub(ncleaned, &pool->dirty_count);
 677         atomic_sub(nfreed, &pool->item_count);
 678
 679 out:
 680         mutex_unlock(&pool->flush_lock);
 681         if (waitqueue_active(&pool->flush_wait))
 682                 wake_up(&pool->flush_wait);
 683 out_nolock:
 684         return ret;
 685 }
 686
 687 static void rds_ib_mr_pool_flush_worker(struct work_struct *work)
 688 {
 689         struct rds_ib_mr_pool *pool = container_of(work, struct rds_ib_mr_pool, flush_worker.work);
 690
 691         rds_ib_flush_mr_pool(pool, 0, NULL);
 692 }
 693
 694 void rds_ib_free_mr(void *trans_private, int invalidate)
 695 {
 696         struct rds_ib_mr *ibmr = trans_private;
 697         struct rds_ib_device *rds_ibdev = ibmr->device;
 698         struct rds_ib_mr_pool *pool = rds_ibdev->mr_pool;
 699
 700         rdsdebug("RDS/IB: free_mr nents %u\n", ibmr->sg_len);
 701
 702         /* Return it to the pool's free list */
 703         if (ibmr->remap_count >= pool->fmr_attr.max_maps)
 704                 llist_add(&ibmr->llnode, &pool->drop_list);
 705         else
 706                 llist_add(&ibmr->llnode, &pool->free_list);
 707
 708         atomic_add(ibmr->sg_len, &pool->free_pinned);
 709         atomic_inc(&pool->dirty_count);
 710
 711         /* If we've pinned too many pages, request a flush */
 712         if (atomic_read(&pool->free_pinned) >= pool->max_free_pinned ||
 713             atomic_read(&pool->dirty_count) >= pool->max_items / 10)
 714                 schedule_delayed_work(&pool->flush_worker, 10);
 715
 716         if (invalidate) {
 717                 if (likely(!in_interrupt())) {
 718                         rds_ib_flush_mr_pool(pool, 0, NULL);
 719                 } else {
 720                         /* We get here if the user created a MR marked
 721                          * as use_once and invalidate at the same time. */
 722                         schedule_delayed_work(&pool->flush_worker, 10);
 723                 }
 724         }
 725
 726         rds_ib_dev_put(rds_ibdev);
 727 }
 728
 729 void rds_ib_flush_mrs(void)
 730 {
 731         struct rds_ib_device *rds_ibdev;
 732
 733         down_read(&rds_ib_devices_lock);
 734         list_for_each_entry(rds_ibdev, &rds_ib_devices, list) {
 735                 struct rds_ib_mr_pool *pool = rds_ibdev->mr_pool;
 736
 737                 if (pool)
 738                         rds_ib_flush_mr_pool(pool, 0, NULL);
 739         }
 740         up_read(&rds_ib_devices_lock);
 741 }
 742
 743 void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
 744                     struct rds_sock *rs, u32 *key_ret)
 745 {
 746         struct rds_ib_device *rds_ibdev;
 747         struct rds_ib_mr *ibmr = NULL;
 748         int ret;
 749
 750         rds_ibdev = rds_ib_get_device(rs->rs_bound_addr);
 751         if (!rds_ibdev) {
 752                 ret = -ENODEV;
 753                 goto out;
 754         }
 755
 756         if (!rds_ibdev->mr_pool) {
 757                 ret = -ENODEV;
 758                 goto out;
 759         }
 760
 761         ibmr = rds_ib_alloc_fmr(rds_ibdev);
 762         if (IS_ERR(ibmr))
 763                 return ibmr;
 764
 765         ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
 766         if (ret == 0)
 767                 *key_ret = ibmr->fmr->rkey;
 768         else
 769                 printk(KERN_WARNING "RDS/IB: map_fmr failed (errno=%d)\n", ret);
 770
 771         ibmr->device = rds_ibdev;
 772         rds_ibdev = NULL;
 773
 774  out:
 775         if (ret) {
 776                 if (ibmr)
 777                         rds_ib_free_mr(ibmr, 0);
 778                 ibmr = ERR_PTR(ret);
 779         }
 780         if (rds_ibdev)
 781                 rds_ib_dev_put(rds_ibdev);
 782         return ibmr;
 783 }
 784