linux/arch/powerpc/kernel/vio.c
<<
>>
Prefs
   1/*
   2 * IBM PowerPC Virtual I/O Infrastructure Support.
   3 *
   4 *    Copyright (c) 2003,2008 IBM Corp.
   5 *     Dave Engebretsen engebret@us.ibm.com
   6 *     Santiago Leon santil@us.ibm.com
   7 *     Hollis Blanchard <hollisb@us.ibm.com>
   8 *     Stephen Rothwell
   9 *     Robert Jennings <rcjenn@us.ibm.com>
  10 *
  11 *      This program is free software; you can redistribute it and/or
  12 *      modify it under the terms of the GNU General Public License
  13 *      as published by the Free Software Foundation; either version
  14 *      2 of the License, or (at your option) any later version.
  15 */
  16
  17#include <linux/types.h>
  18#include <linux/device.h>
  19#include <linux/init.h>
  20#include <linux/console.h>
  21#include <linux/module.h>
  22#include <linux/mm.h>
  23#include <linux/dma-mapping.h>
  24#include <linux/kobject.h>
  25
  26#include <asm/iommu.h>
  27#include <asm/dma.h>
  28#include <asm/vio.h>
  29#include <asm/prom.h>
  30#include <asm/firmware.h>
  31#include <asm/tce.h>
  32#include <asm/abs_addr.h>
  33#include <asm/page.h>
  34#include <asm/hvcall.h>
  35#include <asm/iseries/vio.h>
  36#include <asm/iseries/hv_types.h>
  37#include <asm/iseries/hv_lp_config.h>
  38#include <asm/iseries/hv_call_xm.h>
  39#include <asm/iseries/iommu.h>
  40
  41static struct bus_type vio_bus_type;
  42
  43static struct vio_dev vio_bus_device  = { /* fake "parent" device */
  44        .name = vio_bus_device.dev.bus_id,
  45        .type = "",
  46        .dev.bus_id = "vio",
  47        .dev.bus = &vio_bus_type,
  48};
  49
  50#ifdef CONFIG_PPC_SMLPAR
  51/**
  52 * vio_cmo_pool - A pool of IO memory for CMO use
  53 *
  54 * @size: The size of the pool in bytes
  55 * @free: The amount of free memory in the pool
  56 */
  57struct vio_cmo_pool {
  58        size_t size;
  59        size_t free;
  60};
  61
  62/* How many ms to delay queued balance work */
  63#define VIO_CMO_BALANCE_DELAY 100
  64
  65/* Portion out IO memory to CMO devices by this chunk size */
  66#define VIO_CMO_BALANCE_CHUNK 131072
  67
  68/**
  69 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
  70 *
  71 * @vio_dev: struct vio_dev pointer
  72 * @list: pointer to other devices on bus that are being tracked
  73 */
  74struct vio_cmo_dev_entry {
  75        struct vio_dev *viodev;
  76        struct list_head list;
  77};
  78
  79/**
  80 * vio_cmo - VIO bus accounting structure for CMO entitlement
  81 *
  82 * @lock: spinlock for entire structure
  83 * @balance_q: work queue for balancing system entitlement
  84 * @device_list: list of CMO-enabled devices requiring entitlement
  85 * @entitled: total system entitlement in bytes
  86 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
  87 * @excess: pool of excess entitlement not needed for device reserves or spare
  88 * @spare: IO memory for device hotplug functionality
  89 * @min: minimum necessary for system operation
  90 * @desired: desired memory for system operation
  91 * @curr: bytes currently allocated
  92 * @high: high water mark for IO data usage
  93 */
  94struct vio_cmo {
  95        spinlock_t lock;
  96        struct delayed_work balance_q;
  97        struct list_head device_list;
  98        size_t entitled;
  99        struct vio_cmo_pool reserve;
 100        struct vio_cmo_pool excess;
 101        size_t spare;
 102        size_t min;
 103        size_t desired;
 104        size_t curr;
 105        size_t high;
 106} vio_cmo;
 107
 108/**
 109 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
 110 */
 111static int vio_cmo_num_OF_devs(void)
 112{
 113        struct device_node *node_vroot;
 114        int count = 0;
 115
 116        /*
 117         * Count the number of vdevice entries with an
 118         * ibm,my-dma-window OF property
 119         */
 120        node_vroot = of_find_node_by_name(NULL, "vdevice");
 121        if (node_vroot) {
 122                struct device_node *of_node;
 123                struct property *prop;
 124
 125                for_each_child_of_node(node_vroot, of_node) {
 126                        prop = of_find_property(of_node, "ibm,my-dma-window",
 127                                               NULL);
 128                        if (prop)
 129                                count++;
 130                }
 131        }
 132        of_node_put(node_vroot);
 133        return count;
 134}
 135
 136/**
 137 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
 138 *
 139 * @viodev: VIO device requesting IO memory
 140 * @size: size of allocation requested
 141 *
 142 * Allocations come from memory reserved for the devices and any excess
 143 * IO memory available to all devices.  The spare pool used to service
 144 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
 145 * made available.
 146 *
 147 * Return codes:
 148 *  0 for successful allocation and -ENOMEM for a failure
 149 */
 150static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
 151{
 152        unsigned long flags;
 153        size_t reserve_free = 0;
 154        size_t excess_free = 0;
 155        int ret = -ENOMEM;
 156
 157        spin_lock_irqsave(&vio_cmo.lock, flags);
 158
 159        /* Determine the amount of free entitlement available in reserve */
 160        if (viodev->cmo.entitled > viodev->cmo.allocated)
 161                reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
 162
 163        /* If spare is not fulfilled, the excess pool can not be used. */
 164        if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
 165                excess_free = vio_cmo.excess.free;
 166
 167        /* The request can be satisfied */
 168        if ((reserve_free + excess_free) >= size) {
 169                vio_cmo.curr += size;
 170                if (vio_cmo.curr > vio_cmo.high)
 171                        vio_cmo.high = vio_cmo.curr;
 172                viodev->cmo.allocated += size;
 173                size -= min(reserve_free, size);
 174                vio_cmo.excess.free -= size;
 175                ret = 0;
 176        }
 177
 178        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 179        return ret;
 180}
 181
 182/**
 183 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
 184 * @viodev: VIO device freeing IO memory
 185 * @size: size of deallocation
 186 *
 187 * IO memory is freed by the device back to the correct memory pools.
 188 * The spare pool is replenished first from either memory pool, then
 189 * the reserve pool is used to reduce device entitlement, the excess
 190 * pool is used to increase the reserve pool toward the desired entitlement
 191 * target, and then the remaining memory is returned to the pools.
 192 *
 193 */
 194static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
 195{
 196        unsigned long flags;
 197        size_t spare_needed = 0;
 198        size_t excess_freed = 0;
 199        size_t reserve_freed = size;
 200        size_t tmp;
 201        int balance = 0;
 202
 203        spin_lock_irqsave(&vio_cmo.lock, flags);
 204        vio_cmo.curr -= size;
 205
 206        /* Amount of memory freed from the excess pool */
 207        if (viodev->cmo.allocated > viodev->cmo.entitled) {
 208                excess_freed = min(reserve_freed, (viodev->cmo.allocated -
 209                                                   viodev->cmo.entitled));
 210                reserve_freed -= excess_freed;
 211        }
 212
 213        /* Remove allocation from device */
 214        viodev->cmo.allocated -= (reserve_freed + excess_freed);
 215
 216        /* Spare is a subset of the reserve pool, replenish it first. */
 217        spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
 218
 219        /*
 220         * Replenish the spare in the reserve pool from the excess pool.
 221         * This moves entitlement into the reserve pool.
 222         */
 223        if (spare_needed && excess_freed) {
 224                tmp = min(excess_freed, spare_needed);
 225                vio_cmo.excess.size -= tmp;
 226                vio_cmo.reserve.size += tmp;
 227                vio_cmo.spare += tmp;
 228                excess_freed -= tmp;
 229                spare_needed -= tmp;
 230                balance = 1;
 231        }
 232
 233        /*
 234         * Replenish the spare in the reserve pool from the reserve pool.
 235         * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
 236         * if needed, and gives it to the spare pool. The amount of used
 237         * memory in this pool does not change.
 238         */
 239        if (spare_needed && reserve_freed) {
 240                tmp = min(spare_needed, min(reserve_freed,
 241                                            (viodev->cmo.entitled -
 242                                             VIO_CMO_MIN_ENT)));
 243
 244                vio_cmo.spare += tmp;
 245                viodev->cmo.entitled -= tmp;
 246                reserve_freed -= tmp;
 247                spare_needed -= tmp;
 248                balance = 1;
 249        }
 250
 251        /*
 252         * Increase the reserve pool until the desired allocation is met.
 253         * Move an allocation freed from the excess pool into the reserve
 254         * pool and schedule a balance operation.
 255         */
 256        if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
 257                tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
 258
 259                vio_cmo.excess.size -= tmp;
 260                vio_cmo.reserve.size += tmp;
 261                excess_freed -= tmp;
 262                balance = 1;
 263        }
 264
 265        /* Return memory from the excess pool to that pool */
 266        if (excess_freed)
 267                vio_cmo.excess.free += excess_freed;
 268
 269        if (balance)
 270                schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
 271        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 272}
 273
 274/**
 275 * vio_cmo_entitlement_update - Manage system entitlement changes
 276 *
 277 * @new_entitlement: new system entitlement to attempt to accommodate
 278 *
 279 * Increases in entitlement will be used to fulfill the spare entitlement
 280 * and the rest is given to the excess pool.  Decreases, if they are
 281 * possible, come from the excess pool and from unused device entitlement
 282 *
 283 * Returns: 0 on success, -ENOMEM when change can not be made
 284 */
 285int vio_cmo_entitlement_update(size_t new_entitlement)
 286{
 287        struct vio_dev *viodev;
 288        struct vio_cmo_dev_entry *dev_ent;
 289        unsigned long flags;
 290        size_t avail, delta, tmp;
 291
 292        spin_lock_irqsave(&vio_cmo.lock, flags);
 293
 294        /* Entitlement increases */
 295        if (new_entitlement > vio_cmo.entitled) {
 296                delta = new_entitlement - vio_cmo.entitled;
 297
 298                /* Fulfill spare allocation */
 299                if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
 300                        tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
 301                        vio_cmo.spare += tmp;
 302                        vio_cmo.reserve.size += tmp;
 303                        delta -= tmp;
 304                }
 305
 306                /* Remaining new allocation goes to the excess pool */
 307                vio_cmo.entitled += delta;
 308                vio_cmo.excess.size += delta;
 309                vio_cmo.excess.free += delta;
 310
 311                goto out;
 312        }
 313
 314        /* Entitlement decreases */
 315        delta = vio_cmo.entitled - new_entitlement;
 316        avail = vio_cmo.excess.free;
 317
 318        /*
 319         * Need to check how much unused entitlement each device can
 320         * sacrifice to fulfill entitlement change.
 321         */
 322        list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
 323                if (avail >= delta)
 324                        break;
 325
 326                viodev = dev_ent->viodev;
 327                if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
 328                    (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
 329                                avail += viodev->cmo.entitled -
 330                                         max_t(size_t, viodev->cmo.allocated,
 331                                               VIO_CMO_MIN_ENT);
 332        }
 333
 334        if (delta <= avail) {
 335                vio_cmo.entitled -= delta;
 336
 337                /* Take entitlement from the excess pool first */
 338                tmp = min(vio_cmo.excess.free, delta);
 339                vio_cmo.excess.size -= tmp;
 340                vio_cmo.excess.free -= tmp;
 341                delta -= tmp;
 342
 343                /*
 344                 * Remove all but VIO_CMO_MIN_ENT bytes from devices
 345                 * until entitlement change is served
 346                 */
 347                list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
 348                        if (!delta)
 349                                break;
 350
 351                        viodev = dev_ent->viodev;
 352                        tmp = 0;
 353                        if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
 354                            (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
 355                                tmp = viodev->cmo.entitled -
 356                                      max_t(size_t, viodev->cmo.allocated,
 357                                            VIO_CMO_MIN_ENT);
 358                        viodev->cmo.entitled -= min(tmp, delta);
 359                        delta -= min(tmp, delta);
 360                }
 361        } else {
 362                spin_unlock_irqrestore(&vio_cmo.lock, flags);
 363                return -ENOMEM;
 364        }
 365
 366out:
 367        schedule_delayed_work(&vio_cmo.balance_q, 0);
 368        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 369        return 0;
 370}
 371
 372/**
 373 * vio_cmo_balance - Balance entitlement among devices
 374 *
 375 * @work: work queue structure for this operation
 376 *
 377 * Any system entitlement above the minimum needed for devices, or
 378 * already allocated to devices, can be distributed to the devices.
 379 * The list of devices is iterated through to recalculate the desired
 380 * entitlement level and to determine how much entitlement above the
 381 * minimum entitlement is allocated to devices.
 382 *
 383 * Small chunks of the available entitlement are given to devices until
 384 * their requirements are fulfilled or there is no entitlement left to give.
 385 * Upon completion sizes of the reserve and excess pools are calculated.
 386 *
 387 * The system minimum entitlement level is also recalculated here.
 388 * Entitlement will be reserved for devices even after vio_bus_remove to
 389 * accommodate reloading the driver.  The OF tree is walked to count the
 390 * number of devices present and this will remove entitlement for devices
 391 * that have actually left the system after having vio_bus_remove called.
 392 */
 393static void vio_cmo_balance(struct work_struct *work)
 394{
 395        struct vio_cmo *cmo;
 396        struct vio_dev *viodev;
 397        struct vio_cmo_dev_entry *dev_ent;
 398        unsigned long flags;
 399        size_t avail = 0, level, chunk, need;
 400        int devcount = 0, fulfilled;
 401
 402        cmo = container_of(work, struct vio_cmo, balance_q.work);
 403
 404        spin_lock_irqsave(&vio_cmo.lock, flags);
 405
 406        /* Calculate minimum entitlement and fulfill spare */
 407        cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
 408        BUG_ON(cmo->min > cmo->entitled);
 409        cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
 410        cmo->min += cmo->spare;
 411        cmo->desired = cmo->min;
 412
 413        /*
 414         * Determine how much entitlement is available and reset device
 415         * entitlements
 416         */
 417        avail = cmo->entitled - cmo->spare;
 418        list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
 419                viodev = dev_ent->viodev;
 420                devcount++;
 421                viodev->cmo.entitled = VIO_CMO_MIN_ENT;
 422                cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
 423                avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
 424        }
 425
 426        /*
 427         * Having provided each device with the minimum entitlement, loop
 428         * over the devices portioning out the remaining entitlement
 429         * until there is nothing left.
 430         */
 431        level = VIO_CMO_MIN_ENT;
 432        while (avail) {
 433                fulfilled = 0;
 434                list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
 435                        viodev = dev_ent->viodev;
 436
 437                        if (viodev->cmo.desired <= level) {
 438                                fulfilled++;
 439                                continue;
 440                        }
 441
 442                        /*
 443                         * Give the device up to VIO_CMO_BALANCE_CHUNK
 444                         * bytes of entitlement, but do not exceed the
 445                         * desired level of entitlement for the device.
 446                         */
 447                        chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
 448                        chunk = min(chunk, (viodev->cmo.desired -
 449                                            viodev->cmo.entitled));
 450                        viodev->cmo.entitled += chunk;
 451
 452                        /*
 453                         * If the memory for this entitlement increase was
 454                         * already allocated to the device it does not come
 455                         * from the available pool being portioned out.
 456                         */
 457                        need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
 458                               max(viodev->cmo.allocated, level);
 459                        avail -= need;
 460
 461                }
 462                if (fulfilled == devcount)
 463                        break;
 464                level += VIO_CMO_BALANCE_CHUNK;
 465        }
 466
 467        /* Calculate new reserve and excess pool sizes */
 468        cmo->reserve.size = cmo->min;
 469        cmo->excess.free = 0;
 470        cmo->excess.size = 0;
 471        need = 0;
 472        list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
 473                viodev = dev_ent->viodev;
 474                /* Calculated reserve size above the minimum entitlement */
 475                if (viodev->cmo.entitled)
 476                        cmo->reserve.size += (viodev->cmo.entitled -
 477                                              VIO_CMO_MIN_ENT);
 478                /* Calculated used excess entitlement */
 479                if (viodev->cmo.allocated > viodev->cmo.entitled)
 480                        need += viodev->cmo.allocated - viodev->cmo.entitled;
 481        }
 482        cmo->excess.size = cmo->entitled - cmo->reserve.size;
 483        cmo->excess.free = cmo->excess.size - need;
 484
 485        cancel_delayed_work(container_of(work, struct delayed_work, work));
 486        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 487}
 488
 489static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
 490                                          dma_addr_t *dma_handle, gfp_t flag)
 491{
 492        struct vio_dev *viodev = to_vio_dev(dev);
 493        void *ret;
 494
 495        if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
 496                atomic_inc(&viodev->cmo.allocs_failed);
 497                return NULL;
 498        }
 499
 500        ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag);
 501        if (unlikely(ret == NULL)) {
 502                vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
 503                atomic_inc(&viodev->cmo.allocs_failed);
 504        }
 505
 506        return ret;
 507}
 508
 509static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
 510                                        void *vaddr, dma_addr_t dma_handle)
 511{
 512        struct vio_dev *viodev = to_vio_dev(dev);
 513
 514        dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle);
 515
 516        vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
 517}
 518
 519static dma_addr_t vio_dma_iommu_map_single(struct device *dev, void *vaddr,
 520                                           size_t size,
 521                                           enum dma_data_direction direction,
 522                                           struct dma_attrs *attrs)
 523{
 524        struct vio_dev *viodev = to_vio_dev(dev);
 525        dma_addr_t ret = DMA_ERROR_CODE;
 526
 527        if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
 528                atomic_inc(&viodev->cmo.allocs_failed);
 529                return ret;
 530        }
 531
 532        ret = dma_iommu_ops.map_single(dev, vaddr, size, direction, attrs);
 533        if (unlikely(dma_mapping_error(dev, ret))) {
 534                vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
 535                atomic_inc(&viodev->cmo.allocs_failed);
 536        }
 537
 538        return ret;
 539}
 540
 541static void vio_dma_iommu_unmap_single(struct device *dev,
 542                dma_addr_t dma_handle, size_t size,
 543                enum dma_data_direction direction,
 544                struct dma_attrs *attrs)
 545{
 546        struct vio_dev *viodev = to_vio_dev(dev);
 547
 548        dma_iommu_ops.unmap_single(dev, dma_handle, size, direction, attrs);
 549
 550        vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
 551}
 552
 553static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
 554                                int nelems, enum dma_data_direction direction,
 555                                struct dma_attrs *attrs)
 556{
 557        struct vio_dev *viodev = to_vio_dev(dev);
 558        struct scatterlist *sgl;
 559        int ret, count = 0;
 560        size_t alloc_size = 0;
 561
 562        for (sgl = sglist; count < nelems; count++, sgl++)
 563                alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE);
 564
 565        if (vio_cmo_alloc(viodev, alloc_size)) {
 566                atomic_inc(&viodev->cmo.allocs_failed);
 567                return 0;
 568        }
 569
 570        ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
 571
 572        if (unlikely(!ret)) {
 573                vio_cmo_dealloc(viodev, alloc_size);
 574                atomic_inc(&viodev->cmo.allocs_failed);
 575        }
 576
 577        for (sgl = sglist, count = 0; count < ret; count++, sgl++)
 578                alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
 579        if (alloc_size)
 580                vio_cmo_dealloc(viodev, alloc_size);
 581
 582        return ret;
 583}
 584
 585static void vio_dma_iommu_unmap_sg(struct device *dev,
 586                struct scatterlist *sglist, int nelems,
 587                enum dma_data_direction direction,
 588                struct dma_attrs *attrs)
 589{
 590        struct vio_dev *viodev = to_vio_dev(dev);
 591        struct scatterlist *sgl;
 592        size_t alloc_size = 0;
 593        int count = 0;
 594
 595        for (sgl = sglist; count < nelems; count++, sgl++)
 596                alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
 597
 598        dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
 599
 600        vio_cmo_dealloc(viodev, alloc_size);
 601}
 602
 603struct dma_mapping_ops vio_dma_mapping_ops = {
 604        .alloc_coherent = vio_dma_iommu_alloc_coherent,
 605        .free_coherent  = vio_dma_iommu_free_coherent,
 606        .map_single     = vio_dma_iommu_map_single,
 607        .unmap_single   = vio_dma_iommu_unmap_single,
 608        .map_sg         = vio_dma_iommu_map_sg,
 609        .unmap_sg       = vio_dma_iommu_unmap_sg,
 610};
 611
 612/**
 613 * vio_cmo_set_dev_desired - Set desired entitlement for a device
 614 *
 615 * @viodev: struct vio_dev for device to alter
 616 * @new_desired: new desired entitlement level in bytes
 617 *
 618 * For use by devices to request a change to their entitlement at runtime or
 619 * through sysfs.  The desired entitlement level is changed and a balancing
 620 * of system resources is scheduled to run in the future.
 621 */
 622void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
 623{
 624        unsigned long flags;
 625        struct vio_cmo_dev_entry *dev_ent;
 626        int found = 0;
 627
 628        if (!firmware_has_feature(FW_FEATURE_CMO))
 629                return;
 630
 631        spin_lock_irqsave(&vio_cmo.lock, flags);
 632        if (desired < VIO_CMO_MIN_ENT)
 633                desired = VIO_CMO_MIN_ENT;
 634
 635        /*
 636         * Changes will not be made for devices not in the device list.
 637         * If it is not in the device list, then no driver is loaded
 638         * for the device and it can not receive entitlement.
 639         */
 640        list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
 641                if (viodev == dev_ent->viodev) {
 642                        found = 1;
 643                        break;
 644                }
 645        if (!found)
 646                return;
 647
 648        /* Increase/decrease in desired device entitlement */
 649        if (desired >= viodev->cmo.desired) {
 650                /* Just bump the bus and device values prior to a balance*/
 651                vio_cmo.desired += desired - viodev->cmo.desired;
 652                viodev->cmo.desired = desired;
 653        } else {
 654                /* Decrease bus and device values for desired entitlement */
 655                vio_cmo.desired -= viodev->cmo.desired - desired;
 656                viodev->cmo.desired = desired;
 657                /*
 658                 * If less entitlement is desired than current entitlement, move
 659                 * any reserve memory in the change region to the excess pool.
 660                 */
 661                if (viodev->cmo.entitled > desired) {
 662                        vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
 663                        vio_cmo.excess.size += viodev->cmo.entitled - desired;
 664                        /*
 665                         * If entitlement moving from the reserve pool to the
 666                         * excess pool is currently unused, add to the excess
 667                         * free counter.
 668                         */
 669                        if (viodev->cmo.allocated < viodev->cmo.entitled)
 670                                vio_cmo.excess.free += viodev->cmo.entitled -
 671                                                       max(viodev->cmo.allocated, desired);
 672                        viodev->cmo.entitled = desired;
 673                }
 674        }
 675        schedule_delayed_work(&vio_cmo.balance_q, 0);
 676        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 677}
 678
 679/**
 680 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
 681 *
 682 * @viodev - Pointer to struct vio_dev for device
 683 *
 684 * Determine the devices IO memory entitlement needs, attempting
 685 * to satisfy the system minimum entitlement at first and scheduling
 686 * a balance operation to take care of the rest at a later time.
 687 *
 688 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
 689 *          -ENOMEM when entitlement is not available for device or
 690 *          device entry.
 691 *
 692 */
 693static int vio_cmo_bus_probe(struct vio_dev *viodev)
 694{
 695        struct vio_cmo_dev_entry *dev_ent;
 696        struct device *dev = &viodev->dev;
 697        struct vio_driver *viodrv = to_vio_driver(dev->driver);
 698        unsigned long flags;
 699        size_t size;
 700
 701        /*
 702         * Check to see that device has a DMA window and configure
 703         * entitlement for the device.
 704         */
 705        if (of_get_property(viodev->dev.archdata.of_node,
 706                            "ibm,my-dma-window", NULL)) {
 707                /* Check that the driver is CMO enabled and get desired DMA */
 708                if (!viodrv->get_desired_dma) {
 709                        dev_err(dev, "%s: device driver does not support CMO\n",
 710                                __func__);
 711                        return -EINVAL;
 712                }
 713
 714                viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
 715                if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
 716                        viodev->cmo.desired = VIO_CMO_MIN_ENT;
 717                size = VIO_CMO_MIN_ENT;
 718
 719                dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
 720                                  GFP_KERNEL);
 721                if (!dev_ent)
 722                        return -ENOMEM;
 723
 724                dev_ent->viodev = viodev;
 725                spin_lock_irqsave(&vio_cmo.lock, flags);
 726                list_add(&dev_ent->list, &vio_cmo.device_list);
 727        } else {
 728                viodev->cmo.desired = 0;
 729                size = 0;
 730                spin_lock_irqsave(&vio_cmo.lock, flags);
 731        }
 732
 733        /*
 734         * If the needs for vio_cmo.min have not changed since they
 735         * were last set, the number of devices in the OF tree has
 736         * been constant and the IO memory for this is already in
 737         * the reserve pool.
 738         */
 739        if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
 740                            VIO_CMO_MIN_ENT)) {
 741                /* Updated desired entitlement if device requires it */
 742                if (size)
 743                        vio_cmo.desired += (viodev->cmo.desired -
 744                                        VIO_CMO_MIN_ENT);
 745        } else {
 746                size_t tmp;
 747
 748                tmp = vio_cmo.spare + vio_cmo.excess.free;
 749                if (tmp < size) {
 750                        dev_err(dev, "%s: insufficient free "
 751                                "entitlement to add device. "
 752                                "Need %lu, have %lu\n", __func__,
 753                                size, (vio_cmo.spare + tmp));
 754                        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 755                        return -ENOMEM;
 756                }
 757
 758                /* Use excess pool first to fulfill request */
 759                tmp = min(size, vio_cmo.excess.free);
 760                vio_cmo.excess.free -= tmp;
 761                vio_cmo.excess.size -= tmp;
 762                vio_cmo.reserve.size += tmp;
 763
 764                /* Use spare if excess pool was insufficient */
 765                vio_cmo.spare -= size - tmp;
 766
 767                /* Update bus accounting */
 768                vio_cmo.min += size;
 769                vio_cmo.desired += viodev->cmo.desired;
 770        }
 771        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 772        return 0;
 773}
 774
 775/**
 776 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
 777 *
 778 * @viodev - Pointer to struct vio_dev for device
 779 *
 780 * Remove the device from the cmo device list.  The minimum entitlement
 781 * will be reserved for the device as long as it is in the system.  The
 782 * rest of the entitlement the device had been allocated will be returned
 783 * to the system.
 784 */
 785static void vio_cmo_bus_remove(struct vio_dev *viodev)
 786{
 787        struct vio_cmo_dev_entry *dev_ent;
 788        unsigned long flags;
 789        size_t tmp;
 790
 791        spin_lock_irqsave(&vio_cmo.lock, flags);
 792        if (viodev->cmo.allocated) {
 793                dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
 794                        "allocated after remove operation.\n",
 795                        __func__, viodev->cmo.allocated);
 796                BUG();
 797        }
 798
 799        /*
 800         * Remove the device from the device list being maintained for
 801         * CMO enabled devices.
 802         */
 803        list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
 804                if (viodev == dev_ent->viodev) {
 805                        list_del(&dev_ent->list);
 806                        kfree(dev_ent);
 807                        break;
 808                }
 809
 810        /*
 811         * Devices may not require any entitlement and they do not need
 812         * to be processed.  Otherwise, return the device's entitlement
 813         * back to the pools.
 814         */
 815        if (viodev->cmo.entitled) {
 816                /*
 817                 * This device has not yet left the OF tree, it's
 818                 * minimum entitlement remains in vio_cmo.min and
 819                 * vio_cmo.desired
 820                 */
 821                vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
 822
 823                /*
 824                 * Save min allocation for device in reserve as long
 825                 * as it exists in OF tree as determined by later
 826                 * balance operation
 827                 */
 828                viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
 829
 830                /* Replenish spare from freed reserve pool */
 831                if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
 832                        tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
 833                                                         vio_cmo.spare));
 834                        vio_cmo.spare += tmp;
 835                        viodev->cmo.entitled -= tmp;
 836                }
 837
 838                /* Remaining reserve goes to excess pool */
 839                vio_cmo.excess.size += viodev->cmo.entitled;
 840                vio_cmo.excess.free += viodev->cmo.entitled;
 841                vio_cmo.reserve.size -= viodev->cmo.entitled;
 842
 843                /*
 844                 * Until the device is removed it will keep a
 845                 * minimum entitlement; this will guarantee that
 846                 * a module unload/load will result in a success.
 847                 */
 848                viodev->cmo.entitled = VIO_CMO_MIN_ENT;
 849                viodev->cmo.desired = VIO_CMO_MIN_ENT;
 850                atomic_set(&viodev->cmo.allocs_failed, 0);
 851        }
 852
 853        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 854}
 855
 856static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
 857{
 858        vio_dma_mapping_ops.dma_supported = dma_iommu_ops.dma_supported;
 859        viodev->dev.archdata.dma_ops = &vio_dma_mapping_ops;
 860}
 861
 862/**
 863 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
 864 *
 865 * Set up the reserve and excess entitlement pools based on available
 866 * system entitlement and the number of devices in the OF tree that
 867 * require entitlement in the reserve pool.
 868 */
 869static void vio_cmo_bus_init(void)
 870{
 871        struct hvcall_mpp_data mpp_data;
 872        int err;
 873
 874        memset(&vio_cmo, 0, sizeof(struct vio_cmo));
 875        spin_lock_init(&vio_cmo.lock);
 876        INIT_LIST_HEAD(&vio_cmo.device_list);
 877        INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
 878
 879        /* Get current system entitlement */
 880        err = h_get_mpp(&mpp_data);
 881
 882        /*
 883         * On failure, continue with entitlement set to 0, will panic()
 884         * later when spare is reserved.
 885         */
 886        if (err != H_SUCCESS) {
 887                printk(KERN_ERR "%s: unable to determine system IO "\
 888                       "entitlement. (%d)\n", __func__, err);
 889                vio_cmo.entitled = 0;
 890        } else {
 891                vio_cmo.entitled = mpp_data.entitled_mem;
 892        }
 893
 894        /* Set reservation and check against entitlement */
 895        vio_cmo.spare = VIO_CMO_MIN_ENT;
 896        vio_cmo.reserve.size = vio_cmo.spare;
 897        vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
 898                                 VIO_CMO_MIN_ENT);
 899        if (vio_cmo.reserve.size > vio_cmo.entitled) {
 900                printk(KERN_ERR "%s: insufficient system entitlement\n",
 901                       __func__);
 902                panic("%s: Insufficient system entitlement", __func__);
 903        }
 904
 905        /* Set the remaining accounting variables */
 906        vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
 907        vio_cmo.excess.free = vio_cmo.excess.size;
 908        vio_cmo.min = vio_cmo.reserve.size;
 909        vio_cmo.desired = vio_cmo.reserve.size;
 910}
 911
 912/* sysfs device functions and data structures for CMO */
 913
 914#define viodev_cmo_rd_attr(name)                                        \
 915static ssize_t viodev_cmo_##name##_show(struct device *dev,             \
 916                                        struct device_attribute *attr,  \
 917                                         char *buf)                     \
 918{                                                                       \
 919        return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name);        \
 920}
 921
 922static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
 923                struct device_attribute *attr, char *buf)
 924{
 925        struct vio_dev *viodev = to_vio_dev(dev);
 926        return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
 927}
 928
 929static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
 930                struct device_attribute *attr, const char *buf, size_t count)
 931{
 932        struct vio_dev *viodev = to_vio_dev(dev);
 933        atomic_set(&viodev->cmo.allocs_failed, 0);
 934        return count;
 935}
 936
 937static ssize_t viodev_cmo_desired_set(struct device *dev,
 938                struct device_attribute *attr, const char *buf, size_t count)
 939{
 940        struct vio_dev *viodev = to_vio_dev(dev);
 941        size_t new_desired;
 942        int ret;
 943
 944        ret = strict_strtoul(buf, 10, &new_desired);
 945        if (ret)
 946                return ret;
 947
 948        vio_cmo_set_dev_desired(viodev, new_desired);
 949        return count;
 950}
 951
 952viodev_cmo_rd_attr(desired);
 953viodev_cmo_rd_attr(entitled);
 954viodev_cmo_rd_attr(allocated);
 955
 956static ssize_t name_show(struct device *, struct device_attribute *, char *);
 957static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
 958static struct device_attribute vio_cmo_dev_attrs[] = {
 959        __ATTR_RO(name),
 960        __ATTR_RO(devspec),
 961        __ATTR(cmo_desired,       S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
 962               viodev_cmo_desired_show, viodev_cmo_desired_set),
 963        __ATTR(cmo_entitled,      S_IRUGO, viodev_cmo_entitled_show,      NULL),
 964        __ATTR(cmo_allocated,     S_IRUGO, viodev_cmo_allocated_show,     NULL),
 965        __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
 966               viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
 967        __ATTR_NULL
 968};
 969
 970/* sysfs bus functions and data structures for CMO */
 971
 972#define viobus_cmo_rd_attr(name)                                        \
 973static ssize_t                                                          \
 974viobus_cmo_##name##_show(struct bus_type *bt, char *buf)                \
 975{                                                                       \
 976        return sprintf(buf, "%lu\n", vio_cmo.name);                     \
 977}
 978
 979#define viobus_cmo_pool_rd_attr(name, var)                              \
 980static ssize_t                                                          \
 981viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf)     \
 982{                                                                       \
 983        return sprintf(buf, "%lu\n", vio_cmo.name.var);                 \
 984}
 985
 986static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
 987                                     size_t count)
 988{
 989        unsigned long flags;
 990
 991        spin_lock_irqsave(&vio_cmo.lock, flags);
 992        vio_cmo.high = vio_cmo.curr;
 993        spin_unlock_irqrestore(&vio_cmo.lock, flags);
 994
 995        return count;
 996}
 997
 998viobus_cmo_rd_attr(entitled);
 999viobus_cmo_pool_rd_attr(reserve, size);
1000viobus_cmo_pool_rd_attr(excess, size);
1001viobus_cmo_pool_rd_attr(excess, free);
1002viobus_cmo_rd_attr(spare);
1003viobus_cmo_rd_attr(min);
1004viobus_cmo_rd_attr(desired);
1005viobus_cmo_rd_attr(curr);
1006viobus_cmo_rd_attr(high);
1007
1008static struct bus_attribute vio_cmo_bus_attrs[] = {
1009        __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1010        __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1011        __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1012        __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1013        __ATTR(cmo_spare,   S_IRUGO, viobus_cmo_spare_show,   NULL),
1014        __ATTR(cmo_min,     S_IRUGO, viobus_cmo_min_show,     NULL),
1015        __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1016        __ATTR(cmo_curr,    S_IRUGO, viobus_cmo_curr_show,    NULL),
1017        __ATTR(cmo_high,    S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1018               viobus_cmo_high_show, viobus_cmo_high_reset),
1019        __ATTR_NULL
1020};
1021
1022static void vio_cmo_sysfs_init(void)
1023{
1024        vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1025        vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1026}
1027#else /* CONFIG_PPC_SMLPAR */
1028/* Dummy functions for iSeries platform */
1029int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1030void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1031static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1032static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1033static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1034static void vio_cmo_bus_init(void) {}
1035static void vio_cmo_sysfs_init(void) { }
1036#endif /* CONFIG_PPC_SMLPAR */
1037EXPORT_SYMBOL(vio_cmo_entitlement_update);
1038EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1039
1040static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1041{
1042        const unsigned char *dma_window;
1043        struct iommu_table *tbl;
1044        unsigned long offset, size;
1045
1046        if (firmware_has_feature(FW_FEATURE_ISERIES))
1047                return vio_build_iommu_table_iseries(dev);
1048
1049        dma_window = of_get_property(dev->dev.archdata.of_node,
1050                                  "ibm,my-dma-window", NULL);
1051        if (!dma_window)
1052                return NULL;
1053
1054        tbl = kmalloc(sizeof(*tbl), GFP_KERNEL);
1055
1056        of_parse_dma_window(dev->dev.archdata.of_node, dma_window,
1057                            &tbl->it_index, &offset, &size);
1058
1059        /* TCE table size - measured in tce entries */
1060        tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1061        /* offset for VIO should always be 0 */
1062        tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1063        tbl->it_busno = 0;
1064        tbl->it_type = TCE_VB;
1065
1066        return iommu_init_table(tbl, -1);
1067}
1068
1069/**
1070 * vio_match_device: - Tell if a VIO device has a matching
1071 *                      VIO device id structure.
1072 * @ids:        array of VIO device id structures to search in
1073 * @dev:        the VIO device structure to match against
1074 *
1075 * Used by a driver to check whether a VIO device present in the
1076 * system is in its list of supported devices. Returns the matching
1077 * vio_device_id structure or NULL if there is no match.
1078 */
1079static const struct vio_device_id *vio_match_device(
1080                const struct vio_device_id *ids, const struct vio_dev *dev)
1081{
1082        while (ids->type[0] != '\0') {
1083                if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1084                    of_device_is_compatible(dev->dev.archdata.of_node,
1085                                         ids->compat))
1086                        return ids;
1087                ids++;
1088        }
1089        return NULL;
1090}
1091
1092/*
1093 * Convert from struct device to struct vio_dev and pass to driver.
1094 * dev->driver has already been set by generic code because vio_bus_match
1095 * succeeded.
1096 */
1097static int vio_bus_probe(struct device *dev)
1098{
1099        struct vio_dev *viodev = to_vio_dev(dev);
1100        struct vio_driver *viodrv = to_vio_driver(dev->driver);
1101        const struct vio_device_id *id;
1102        int error = -ENODEV;
1103
1104        if (!viodrv->probe)
1105                return error;
1106
1107        id = vio_match_device(viodrv->id_table, viodev);
1108        if (id) {
1109                memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1110                if (firmware_has_feature(FW_FEATURE_CMO)) {
1111                        error = vio_cmo_bus_probe(viodev);
1112                        if (error)
1113                                return error;
1114                }
1115                error = viodrv->probe(viodev, id);
1116                if (error && firmware_has_feature(FW_FEATURE_CMO))
1117                        vio_cmo_bus_remove(viodev);
1118        }
1119
1120        return error;
1121}
1122
1123/* convert from struct device to struct vio_dev and pass to driver. */
1124static int vio_bus_remove(struct device *dev)
1125{
1126        struct vio_dev *viodev = to_vio_dev(dev);
1127        struct vio_driver *viodrv = to_vio_driver(dev->driver);
1128        struct device *devptr;
1129        int ret = 1;
1130
1131        /*
1132         * Hold a reference to the device after the remove function is called
1133         * to allow for CMO accounting cleanup for the device.
1134         */
1135        devptr = get_device(dev);
1136
1137        if (viodrv->remove)
1138                ret = viodrv->remove(viodev);
1139
1140        if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1141                vio_cmo_bus_remove(viodev);
1142
1143        put_device(devptr);
1144        return ret;
1145}
1146
1147/**
1148 * vio_register_driver: - Register a new vio driver
1149 * @drv:        The vio_driver structure to be registered.
1150 */
1151int vio_register_driver(struct vio_driver *viodrv)
1152{
1153        printk(KERN_DEBUG "%s: driver %s registering\n", __func__,
1154                viodrv->driver.name);
1155
1156        /* fill in 'struct driver' fields */
1157        viodrv->driver.bus = &vio_bus_type;
1158
1159        return driver_register(&viodrv->driver);
1160}
1161EXPORT_SYMBOL(vio_register_driver);
1162
1163/**
1164 * vio_unregister_driver - Remove registration of vio driver.
1165 * @driver:     The vio_driver struct to be removed form registration
1166 */
1167void vio_unregister_driver(struct vio_driver *viodrv)
1168{
1169        driver_unregister(&viodrv->driver);
1170}
1171EXPORT_SYMBOL(vio_unregister_driver);
1172
1173/* vio_dev refcount hit 0 */
1174static void __devinit vio_dev_release(struct device *dev)
1175{
1176        /* XXX should free TCE table */
1177        of_node_put(dev->archdata.of_node);
1178        kfree(to_vio_dev(dev));
1179}
1180
1181/**
1182 * vio_register_device_node: - Register a new vio device.
1183 * @of_node:    The OF node for this device.
1184 *
1185 * Creates and initializes a vio_dev structure from the data in
1186 * of_node and adds it to the list of virtual devices.
1187 * Returns a pointer to the created vio_dev or NULL if node has
1188 * NULL device_type or compatible fields.
1189 */
1190struct vio_dev *vio_register_device_node(struct device_node *of_node)
1191{
1192        struct vio_dev *viodev;
1193        const unsigned int *unit_address;
1194
1195        /* we need the 'device_type' property, in order to match with drivers */
1196        if (of_node->type == NULL) {
1197                printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
1198                                __func__,
1199                                of_node->name ? of_node->name : "<unknown>");
1200                return NULL;
1201        }
1202
1203        unit_address = of_get_property(of_node, "reg", NULL);
1204        if (unit_address == NULL) {
1205                printk(KERN_WARNING "%s: node %s missing 'reg'\n",
1206                                __func__,
1207                                of_node->name ? of_node->name : "<unknown>");
1208                return NULL;
1209        }
1210
1211        /* allocate a vio_dev for this node */
1212        viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1213        if (viodev == NULL)
1214                return NULL;
1215
1216        viodev->irq = irq_of_parse_and_map(of_node, 0);
1217
1218        snprintf(viodev->dev.bus_id, BUS_ID_SIZE, "%x", *unit_address);
1219        viodev->name = of_node->name;
1220        viodev->type = of_node->type;
1221        viodev->unit_address = *unit_address;
1222        if (firmware_has_feature(FW_FEATURE_ISERIES)) {
1223                unit_address = of_get_property(of_node,
1224                                "linux,unit_address", NULL);
1225                if (unit_address != NULL)
1226                        viodev->unit_address = *unit_address;
1227        }
1228        viodev->dev.archdata.of_node = of_node_get(of_node);
1229
1230        if (firmware_has_feature(FW_FEATURE_CMO))
1231                vio_cmo_set_dma_ops(viodev);
1232        else
1233                viodev->dev.archdata.dma_ops = &dma_iommu_ops;
1234        viodev->dev.archdata.dma_data = vio_build_iommu_table(viodev);
1235        viodev->dev.archdata.numa_node = of_node_to_nid(of_node);
1236
1237        /* init generic 'struct device' fields: */
1238        viodev->dev.parent = &vio_bus_device.dev;
1239        viodev->dev.bus = &vio_bus_type;
1240        viodev->dev.release = vio_dev_release;
1241
1242        /* register with generic device framework */
1243        if (device_register(&viodev->dev)) {
1244                printk(KERN_ERR "%s: failed to register device %s\n",
1245                                __func__, viodev->dev.bus_id);
1246                /* XXX free TCE table */
1247                kfree(viodev);
1248                return NULL;
1249        }
1250
1251        return viodev;
1252}
1253EXPORT_SYMBOL(vio_register_device_node);
1254
1255/**
1256 * vio_bus_init: - Initialize the virtual IO bus
1257 */
1258static int __init vio_bus_init(void)
1259{
1260        int err;
1261        struct device_node *node_vroot;
1262
1263        if (firmware_has_feature(FW_FEATURE_CMO))
1264                vio_cmo_sysfs_init();
1265
1266        err = bus_register(&vio_bus_type);
1267        if (err) {
1268                printk(KERN_ERR "failed to register VIO bus\n");
1269                return err;
1270        }
1271
1272        /*
1273         * The fake parent of all vio devices, just to give us
1274         * a nice directory
1275         */
1276        err = device_register(&vio_bus_device.dev);
1277        if (err) {
1278                printk(KERN_WARNING "%s: device_register returned %i\n",
1279                                __func__, err);
1280                return err;
1281        }
1282
1283        if (firmware_has_feature(FW_FEATURE_CMO))
1284                vio_cmo_bus_init();
1285
1286        node_vroot = of_find_node_by_name(NULL, "vdevice");
1287        if (node_vroot) {
1288                struct device_node *of_node;
1289
1290                /*
1291                 * Create struct vio_devices for each virtual device in
1292                 * the device tree. Drivers will associate with them later.
1293                 */
1294                for (of_node = node_vroot->child; of_node != NULL;
1295                                of_node = of_node->sibling)
1296                        vio_register_device_node(of_node);
1297                of_node_put(node_vroot);
1298        }
1299
1300        return 0;
1301}
1302__initcall(vio_bus_init);
1303
1304static ssize_t name_show(struct device *dev,
1305                struct device_attribute *attr, char *buf)
1306{
1307        return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1308}
1309
1310static ssize_t devspec_show(struct device *dev,
1311                struct device_attribute *attr, char *buf)
1312{
1313        struct device_node *of_node = dev->archdata.of_node;
1314
1315        return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1316}
1317
1318static struct device_attribute vio_dev_attrs[] = {
1319        __ATTR_RO(name),
1320        __ATTR_RO(devspec),
1321        __ATTR_NULL
1322};
1323
1324void __devinit vio_unregister_device(struct vio_dev *viodev)
1325{
1326        device_unregister(&viodev->dev);
1327}
1328EXPORT_SYMBOL(vio_unregister_device);
1329
1330static int vio_bus_match(struct device *dev, struct device_driver *drv)
1331{
1332        const struct vio_dev *vio_dev = to_vio_dev(dev);
1333        struct vio_driver *vio_drv = to_vio_driver(drv);
1334        const struct vio_device_id *ids = vio_drv->id_table;
1335
1336        return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1337}
1338
1339static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1340{
1341        const struct vio_dev *vio_dev = to_vio_dev(dev);
1342        struct device_node *dn;
1343        const char *cp;
1344
1345        dn = dev->archdata.of_node;
1346        if (!dn)
1347                return -ENODEV;
1348        cp = of_get_property(dn, "compatible", NULL);
1349        if (!cp)
1350                return -ENODEV;
1351
1352        add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1353        return 0;
1354}
1355
1356static struct bus_type vio_bus_type = {
1357        .name = "vio",
1358        .dev_attrs = vio_dev_attrs,
1359        .uevent = vio_hotplug,
1360        .match = vio_bus_match,
1361        .probe = vio_bus_probe,
1362        .remove = vio_bus_remove,
1363};
1364
1365/**
1366 * vio_get_attribute: - get attribute for virtual device
1367 * @vdev:       The vio device to get property.
1368 * @which:      The property/attribute to be extracted.
1369 * @length:     Pointer to length of returned data size (unused if NULL).
1370 *
1371 * Calls prom.c's of_get_property() to return the value of the
1372 * attribute specified by @which
1373*/
1374const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1375{
1376        return of_get_property(vdev->dev.archdata.of_node, which, length);
1377}
1378EXPORT_SYMBOL(vio_get_attribute);
1379
1380#ifdef CONFIG_PPC_PSERIES
1381/* vio_find_name() - internal because only vio.c knows how we formatted the
1382 * kobject name
1383 */
1384static struct vio_dev *vio_find_name(const char *name)
1385{
1386        struct device *found;
1387
1388        found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1389        if (!found)
1390                return NULL;
1391
1392        return to_vio_dev(found);
1393}
1394
1395/**
1396 * vio_find_node - find an already-registered vio_dev
1397 * @vnode: device_node of the virtual device we're looking for
1398 */
1399struct vio_dev *vio_find_node(struct device_node *vnode)
1400{
1401        const uint32_t *unit_address;
1402        char kobj_name[BUS_ID_SIZE];
1403
1404        /* construct the kobject name from the device node */
1405        unit_address = of_get_property(vnode, "reg", NULL);
1406        if (!unit_address)
1407                return NULL;
1408        snprintf(kobj_name, BUS_ID_SIZE, "%x", *unit_address);
1409
1410        return vio_find_name(kobj_name);
1411}
1412EXPORT_SYMBOL(vio_find_node);
1413
1414int vio_enable_interrupts(struct vio_dev *dev)
1415{
1416        int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1417        if (rc != H_SUCCESS)
1418                printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1419        return rc;
1420}
1421EXPORT_SYMBOL(vio_enable_interrupts);
1422
1423int vio_disable_interrupts(struct vio_dev *dev)
1424{
1425        int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1426        if (rc != H_SUCCESS)
1427                printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1428        return rc;
1429}
1430EXPORT_SYMBOL(vio_disable_interrupts);
1431#endif /* CONFIG_PPC_PSERIES */
1432
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.