virpci.c 62.6 KB
Newer Older
1
/*
2 3
 * virpci.c: helper APIs for managing host PCI devices
 *
E
Eric Blake 已提交
4
 * Copyright (C) 2009-2013 Red Hat, Inc.
5 6 7 8 9 10 11 12 13 14 15 16
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with this library.  If not, see
O
Osier Yang 已提交
18
 * <http://www.gnu.org/licenses/>.
19 20 21 22 23 24 25
 *
 * Authors:
 *     Mark McLoughlin <markmc@redhat.com>
 */

#include <config.h>

26
#include "virpci.h"
27 28 29 30 31 32 33 34 35 36

#include <dirent.h>
#include <fcntl.h>
#include <inttypes.h>
#include <limits.h>
#include <stdio.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
37
#include <stdlib.h>
38

39
#include "virlog.h"
40
#include "viralloc.h"
41
#include "vircommand.h"
42
#include "virerror.h"
E
Eric Blake 已提交
43
#include "virfile.h"
44 45 46 47 48

#define PCI_SYSFS "/sys/bus/pci/"
#define PCI_ID_LEN 10   /* "XXXX XXXX" */
#define PCI_ADDR_LEN 13 /* "XXXX:XX:XX.X" */

49 50 51 52
#define SRIOV_FOUND 0
#define SRIOV_NOT_FOUND 1
#define SRIOV_ERROR -1

53
struct _virPCIDevice {
54 55 56 57 58 59 60
    unsigned      domain;
    unsigned      bus;
    unsigned      slot;
    unsigned      function;

    char          name[PCI_ADDR_LEN]; /* domain:bus:slot.function */
    char          id[PCI_ID_LEN];     /* product vendor */
E
Eric Blake 已提交
61
    char          *path;
62
    const char    *used_by;           /* The domain which uses the device */
63 64 65 66 67

    unsigned      pcie_cap_pos;
    unsigned      pci_pm_cap_pos;
    unsigned      has_flr : 1;
    unsigned      has_pm_reset : 1;
68
    unsigned      managed : 1;
69 70 71 72 73

    /* used by reattach function */
    unsigned      unbind_from_stub : 1;
    unsigned      remove_slot : 1;
    unsigned      reprobe : 1;
74 75
};

76
struct _virPCIDeviceList {
77 78
    virObjectLockable parent;

79
    unsigned count;
80
    virPCIDevicePtr *devs;
81 82 83
};


84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
/* For virReportOOMError()  and virReportSystemError() */
#define VIR_FROM_THIS VIR_FROM_NONE

/* Specifications referenced in comments:
 *  PCI30  - PCI Local Bus Specification 3.0
 *  PCIe20 - PCI Express Base Specification 2.0
 *  BR12   - PCI-to-PCI Bridge Architecture Specification 1.2
 *  PM12   - PCI Bus Power Management Interface Specification 1.2
 *  ECN_AF - Advanced Capabilities for Conventional PCI ECN
 */

/* Type 0 config space header length; PCI30 Section 6.1 Configuration Space Organization */
#define PCI_CONF_LEN            0x100
#define PCI_CONF_HEADER_LEN     0x40

/* PCI30 6.2.1 */
#define PCI_HEADER_TYPE         0x0e    /* Header type */
101 102 103
#define PCI_HEADER_TYPE_BRIDGE 0x1
#define PCI_HEADER_TYPE_MASK   0x7f
#define PCI_HEADER_TYPE_MULTI  0x80
104 105 106 107 108 109 110 111 112

/* PCI30 6.2.1  Device Identification */
#define PCI_CLASS_DEVICE        0x0a    /* Device class */

/* Class Code for bridge; PCI30 D.7  Base Class 06h */
#define PCI_CLASS_BRIDGE_PCI    0x0604

/* PCI30 6.2.3  Device Status */
#define PCI_STATUS              0x06    /* 16 bits */
113
#define PCI_STATUS_CAP_LIST    0x10    /* Support Capability List */
114 115 116 117 118 119 120 121 122 123 124 125 126

/* PCI30 6.7  Capabilities List */
#define PCI_CAPABILITY_LIST     0x34    /* Offset of first capability list entry */

/* PM12 3.2.1  Capability Identifier */
#define PCI_CAP_ID_PM           0x01    /* Power Management */
/* PCI30 H Capability IDs */
#define PCI_CAP_ID_EXP          0x10    /* PCI Express */
/* ECN_AF 6.x.1.1  Capability ID for AF */
#define PCI_CAP_ID_AF           0x13    /* Advanced Features */

/* PCIe20 7.8.3  Device Capabilities Register (Offset 04h) */
#define PCI_EXP_DEVCAP          0x4     /* Device capabilities */
127
#define PCI_EXP_DEVCAP_FLR     (1<<28) /* Function Level Reset */
128 129 130 131 132 133 134

/* Header type 1 BR12 3.2 PCI-to-PCI Bridge Configuration Space Header Format */
#define PCI_PRIMARY_BUS         0x18    /* BR12 3.2.5.2 Primary bus number */
#define PCI_SECONDARY_BUS       0x19    /* BR12 3.2.5.3 Secondary bus number */
#define PCI_SUBORDINATE_BUS     0x1a    /* BR12 3.2.5.4 Highest bus number behind the bridge */
#define PCI_BRIDGE_CONTROL      0x3e
/* BR12 3.2.5.18  Bridge Control Register */
135
#define PCI_BRIDGE_CTL_RESET   0x40    /* Secondary bus reset */
136 137 138

/* PM12 3.2.4  Power Management Control/Status (Offset = 4) */
#define PCI_PM_CTRL                4    /* PM control and status register */
139 140 141 142
#define PCI_PM_CTRL_STATE_MASK    0x3  /* Current power state (D0 to D3) */
#define PCI_PM_CTRL_STATE_D0      0x0  /* D0 state */
#define PCI_PM_CTRL_STATE_D3hot   0x3  /* D3 state */
#define PCI_PM_CTRL_NO_SOFT_RESET 0x8  /* No reset for D3hot->D0 */
143 144 145

/* ECN_AF 6.x.1  Advanced Features Capability Structure */
#define PCI_AF_CAP              0x3     /* Advanced features capabilities */
146
#define PCI_AF_CAP_FLR         0x2     /* Function Level Reset */
147

J
Jiri Denemark 已提交
148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
#define PCI_EXP_FLAGS           0x2
#define PCI_EXP_FLAGS_TYPE      0x00f0
#define PCI_EXP_TYPE_DOWNSTREAM 0x6

#define PCI_EXT_CAP_BASE          0x100
#define PCI_EXT_CAP_LIMIT         0x1000
#define PCI_EXT_CAP_ID_MASK       0x0000ffff
#define PCI_EXT_CAP_OFFSET_SHIFT  20
#define PCI_EXT_CAP_OFFSET_MASK   0x00000ffc

#define PCI_EXT_CAP_ID_ACS      0x000d
#define PCI_EXT_ACS_CTRL        0x06

#define PCI_EXT_CAP_ACS_SV      0x01
#define PCI_EXT_CAP_ACS_RR      0x04
#define PCI_EXT_CAP_ACS_CR      0x08
#define PCI_EXT_CAP_ACS_UF      0x10
165 166 167
#define PCI_EXT_CAP_ACS_ENABLED (PCI_EXT_CAP_ACS_SV |   \
                                 PCI_EXT_CAP_ACS_RR |   \
                                 PCI_EXT_CAP_ACS_CR |   \
J
Jiri Denemark 已提交
168 169
                                 PCI_EXT_CAP_ACS_UF)

170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
static virClassPtr virPCIDeviceListClass;

static void virPCIDeviceListDispose(void *obj);

static int virPCIOnceInit(void)
{
    if (!(virPCIDeviceListClass = virClassNew(virClassForObjectLockable(),
                                              "virPCIDeviceList",
                                              sizeof(virPCIDeviceList),
                                              virPCIDeviceListDispose)))
        return -1;

    return 0;
}

VIR_ONCE_GLOBAL_INIT(virPCI)

187
static int
188
virPCIDeviceConfigOpen(virPCIDevicePtr dev, bool fatal)
189 190 191 192
{
    int fd;

    fd = open(dev->path, O_RDWR);
193

194
    if (fd < 0) {
195 196 197 198 199 200 201 202 203
        if (fatal) {
            virReportSystemError(errno,
                                 _("Failed to open config space file '%s'"),
                                 dev->path);
        } else {
            char ebuf[1024];
            VIR_WARN("Failed to open config space file '%s': %s",
                     dev->path, virStrerror(errno, ebuf, sizeof(ebuf)));
        }
204 205
        return -1;
    }
206

207
    VIR_DEBUG("%s %s: opened %s", dev->id, dev->name, dev->path);
208
    return fd;
209 210
}

211
static void
212
virPCIDeviceConfigClose(virPCIDevicePtr dev, int cfgfd)
213
{
214 215 216 217 218
    if (VIR_CLOSE(cfgfd) < 0) {
        char ebuf[1024];
        VIR_WARN("Failed to close config space file '%s': %s",
                 dev->path, virStrerror(errno, ebuf, sizeof(ebuf)));
    }
219 220
}

221

222
static int
223 224 225 226 227
virPCIDeviceRead(virPCIDevicePtr dev,
                 int cfgfd,
                 unsigned pos,
                 uint8_t *buf,
                 unsigned buflen)
228 229 230
{
    memset(buf, 0, buflen);

231 232
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        saferead(cfgfd, buf, buflen) != buflen) {
233
        char ebuf[1024];
234
        VIR_WARN("Failed to read from '%s' : %s", dev->path,
235 236 237 238 239 240 241
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static uint8_t
242
virPCIDeviceRead8(virPCIDevicePtr dev, int cfgfd, unsigned pos)
243 244
{
    uint8_t buf;
245
    virPCIDeviceRead(dev, cfgfd, pos, &buf, sizeof(buf));
246 247 248 249
    return buf;
}

static uint16_t
250
virPCIDeviceRead16(virPCIDevicePtr dev, int cfgfd, unsigned pos)
251 252
{
    uint8_t buf[2];
253
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
254 255 256 257
    return (buf[0] << 0) | (buf[1] << 8);
}

static uint32_t
258
virPCIDeviceRead32(virPCIDevicePtr dev, int cfgfd, unsigned pos)
259 260
{
    uint8_t buf[4];
261
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
262 263 264 265
    return (buf[0] << 0) | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
}

static int
266 267 268 269 270
virPCIDeviceWrite(virPCIDevicePtr dev,
                  int cfgfd,
                  unsigned pos,
                  uint8_t *buf,
                  unsigned buflen)
271
{
272 273
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        safewrite(cfgfd, buf, buflen) != buflen) {
274
        char ebuf[1024];
275
        VIR_WARN("Failed to write to '%s' : %s", dev->path,
276 277 278 279 280 281 282
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static void
283
virPCIDeviceWrite16(virPCIDevicePtr dev, int cfgfd, unsigned pos, uint16_t val)
284 285
{
    uint8_t buf[2] = { (val >> 0), (val >> 8) };
286
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
287 288 289
}

static void
290
virPCIDeviceWrite32(virPCIDevicePtr dev, int cfgfd, unsigned pos, uint32_t val)
291
{
292
    uint8_t buf[4] = { (val >> 0), (val >> 8), (val >> 16), (val >> 24) };
293
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
294 295
}

296
typedef int (*virPCIDeviceIterPredicate)(virPCIDevicePtr , virPCIDevicePtr , void *);
297 298 299 300 301 302 303

/* Iterate over available PCI devices calling @predicate
 * to compare each one to @dev.
 * Return -1 on error since we don't want to assume it is
 * safe to reset if there is an error.
 */
static int
304 305 306 307
virPCIDeviceIterDevices(virPCIDeviceIterPredicate predicate,
                        virPCIDevicePtr dev,
                        virPCIDevicePtr *matched,
                        void *data)
308 309 310
{
    DIR *dir;
    struct dirent *entry;
311
    int ret = 0;
312
    int rc;
313 314 315 316 317 318 319

    *matched = NULL;

    VIR_DEBUG("%s %s: iterating over " PCI_SYSFS "devices", dev->id, dev->name);

    dir = opendir(PCI_SYSFS "devices");
    if (!dir) {
320
        VIR_WARN("Failed to open " PCI_SYSFS "devices");
321 322 323 324
        return -1;
    }

    while ((entry = readdir(dir))) {
325
        unsigned int domain, bus, slot, function;
326
        virPCIDevicePtr check;
327
        char *tmp;
328 329 330 331 332

        /* Ignore '.' and '..' */
        if (entry->d_name[0] == '.')
            continue;

333 334 335 336 337 338 339 340 341
        /* expected format: <domain>:<bus>:<slot>.<function> */
        if (/* domain */
            virStrToLong_ui(entry->d_name, &tmp, 16, &domain) < 0 || *tmp != ':' ||
            /* bus */
            virStrToLong_ui(tmp + 1, &tmp, 16, &bus) < 0 || *tmp != ':' ||
            /* slot */
            virStrToLong_ui(tmp + 1, &tmp, 16, &slot) < 0 || *tmp != '.' ||
            /* function */
            virStrToLong_ui(tmp + 1, NULL, 16, &function) < 0) {
342 343 344 345
            VIR_WARN("Unusual entry in " PCI_SYSFS "devices: %s", entry->d_name);
            continue;
        }

346
        check = virPCIDeviceNew(domain, bus, slot, function);
347
        if (!check) {
348 349 350
            ret = -1;
            break;
        }
351

352 353 354
        rc = predicate(dev, check, data);
        if (rc < 0) {
            /* the predicate returned an error, bail */
355
            virPCIDeviceFree(check);
356 357 358 359
            ret = -1;
            break;
        }
        else if (rc == 1) {
360 361
            VIR_DEBUG("%s %s: iter matched on %s", dev->id, dev->name, check->name);
            *matched = check;
362
            ret = 1;
363 364
            break;
        }
365

366
        virPCIDeviceFree(check);
367 368
    }
    closedir(dir);
369
    return ret;
370 371 372
}

static uint8_t
373
virPCIDeviceFindCapabilityOffset(virPCIDevicePtr dev, int cfgfd, unsigned capability)
374 375 376 377
{
    uint16_t status;
    uint8_t pos;

378
    status = virPCIDeviceRead16(dev, cfgfd, PCI_STATUS);
379 380 381
    if (!(status & PCI_STATUS_CAP_LIST))
        return 0;

382
    pos = virPCIDeviceRead8(dev, cfgfd, PCI_CAPABILITY_LIST);
383 384 385 386 387 388 389 390 391

    /* Zero indicates last capability, capabilities can't
     * be in the config space header and 0xff is returned
     * by the kernel if we don't have access to this region
     *
     * Note: we're not handling loops or extended
     * capabilities here.
     */
    while (pos >= PCI_CONF_HEADER_LEN && pos != 0xff) {
392
        uint8_t capid = virPCIDeviceRead8(dev, cfgfd, pos);
393 394 395 396 397 398
        if (capid == capability) {
            VIR_DEBUG("%s %s: found cap 0x%.2x at 0x%.2x",
                      dev->id, dev->name, capability, pos);
            return pos;
        }

399
        pos = virPCIDeviceRead8(dev, cfgfd, pos + 1);
400 401 402 403 404 405 406
    }

    VIR_DEBUG("%s %s: failed to find cap 0x%.2x", dev->id, dev->name, capability);

    return 0;
}

J
Jiri Denemark 已提交
407
static unsigned int
408 409 410
virPCIDeviceFindExtendedCapabilityOffset(virPCIDevicePtr dev,
                                         int cfgfd,
                                         unsigned capability)
J
Jiri Denemark 已提交
411 412 413 414 415 416 417 418 419 420
{
    int ttl;
    unsigned int pos;
    uint32_t header;

    /* minimum 8 bytes per capability */
    ttl = (PCI_EXT_CAP_LIMIT - PCI_EXT_CAP_BASE) / 8;
    pos = PCI_EXT_CAP_BASE;

    while (ttl > 0 && pos >= PCI_EXT_CAP_BASE) {
421
        header = virPCIDeviceRead32(dev, cfgfd, pos);
J
Jiri Denemark 已提交
422 423 424 425 426 427 428 429 430 431 432

        if ((header & PCI_EXT_CAP_ID_MASK) == capability)
            return pos;

        pos = (header >> PCI_EXT_CAP_OFFSET_SHIFT) & PCI_EXT_CAP_OFFSET_MASK;
        ttl--;
    }

    return 0;
}

433 434 435 436
/* detects whether this device has FLR.  Returns 0 if the device does
 * not have FLR, 1 if it does, and -1 on error
 */
static int
437
virPCIDeviceDetectFunctionLevelReset(virPCIDevicePtr dev, int cfgfd)
438
{
M
Mark McLoughlin 已提交
439
    uint32_t caps;
440
    uint8_t pos;
441 442
    char *path;
    int found;
443 444 445 446 447 448 449 450

    /* The PCIe Function Level Reset capability allows
     * individual device functions to be reset without
     * affecting any other functions on the device or
     * any other devices on the bus. This is only common
     * on SR-IOV NICs at the moment.
     */
    if (dev->pcie_cap_pos) {
451
        caps = virPCIDeviceRead32(dev, cfgfd, dev->pcie_cap_pos + PCI_EXP_DEVCAP);
452 453 454 455 456 457 458 459 460 461
        if (caps & PCI_EXP_DEVCAP_FLR) {
            VIR_DEBUG("%s %s: detected PCIe FLR capability", dev->id, dev->name);
            return 1;
        }
    }

    /* The PCI AF Function Level Reset capability is
     * the same thing, except for conventional PCI
     * devices. This is not common yet.
     */
462
    pos = virPCIDeviceFindCapabilityOffset(dev, cfgfd, PCI_CAP_ID_AF);
463
    if (pos) {
464
        caps = virPCIDeviceRead16(dev, cfgfd, pos + PCI_AF_CAP);
465 466 467 468 469 470
        if (caps & PCI_AF_CAP_FLR) {
            VIR_DEBUG("%s %s: detected PCI FLR capability", dev->id, dev->name);
            return 1;
        }
    }

471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
    /* there are some buggy devices that do support FLR, but forget to
     * advertise that fact in their capabilities.  However, FLR is *required*
     * to be present for virtual functions (VFs), so if we see that this
     * device is a VF, we just assume FLR works
     */

    if (virAsprintf(&path, PCI_SYSFS "devices/%s/physfn", dev->name) < 0) {
        virReportOOMError();
        return -1;
    }

    found = virFileExists(path);
    VIR_FREE(path);
    if (found) {
        VIR_DEBUG("%s %s: buggy device didn't advertise FLR, but is a VF; forcing flr on",
                  dev->id, dev->name);
        return 1;
    }

490 491 492 493 494 495 496 497 498 499
    VIR_DEBUG("%s %s: no FLR capability found", dev->id, dev->name);

    return 0;
}

/* Require the device has the PCI Power Management capability
 * and that a D3hot->D0 transition will results in a full
 * internal reset, not just a soft reset.
 */
static unsigned
500
virPCIDeviceDetectPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
501 502 503 504 505
{
    if (dev->pci_pm_cap_pos) {
        uint32_t ctl;

        /* require the NO_SOFT_RESET bit is clear */
506
        ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
507 508 509 510 511 512 513 514 515 516 517
        if (!(ctl & PCI_PM_CTRL_NO_SOFT_RESET)) {
            VIR_DEBUG("%s %s: detected PM reset capability", dev->id, dev->name);
            return 1;
        }
    }

    VIR_DEBUG("%s %s: no PM reset capability found", dev->id, dev->name);

    return 0;
}

518
/* Any active devices on the same domain/bus ? */
519
static int
520
virPCIDeviceSharesBusWithActive(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
521
{
522
    virPCIDeviceList *inactiveDevs = data;
523

524
    /* Different domain, different bus, or simply identical device */
525 526
    if (dev->domain != check->domain ||
        dev->bus != check->bus ||
527 528
        (dev->slot == check->slot &&
         dev->function == check->function))
529 530
        return 0;

531
    /* same bus, but inactive, i.e. about to be assigned to guest */
532
    if (inactiveDevs && virPCIDeviceListFind(inactiveDevs, check))
533
        return 0;
534

535
    return 1;
536 537
}

538 539 540
static virPCIDevicePtr
virPCIDeviceBusContainsActiveDevices(virPCIDevicePtr dev,
                                     virPCIDeviceList *inactiveDevs)
541
{
542 543 544
    virPCIDevicePtr active = NULL;
    if (virPCIDeviceIterDevices(virPCIDeviceSharesBusWithActive,
                                dev, &active, inactiveDevs) < 0)
545 546 547 548 549
        return NULL;
    return active;
}

/* Is @check the parent of @dev ? */
550
static int
551
virPCIDeviceIsParent(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
552 553 554
{
    uint16_t device_class;
    uint8_t header_type, secondary, subordinate;
555
    virPCIDevicePtr *best = data;
556 557
    int ret = 0;
    int fd;
558

559
    if (dev->domain != check->domain)
560 561
        return 0;

562
    if ((fd = virPCIDeviceConfigOpen(check, false)) < 0)
563 564
        return 0;

565
    /* Is it a bridge? */
566
    device_class = virPCIDeviceRead16(check, fd, PCI_CLASS_DEVICE);
567
    if (device_class != PCI_CLASS_BRIDGE_PCI)
568
        goto cleanup;
569 570

    /* Is it a plane? */
571
    header_type = virPCIDeviceRead8(check, fd, PCI_HEADER_TYPE);
572
    if ((header_type & PCI_HEADER_TYPE_MASK) != PCI_HEADER_TYPE_BRIDGE)
573
        goto cleanup;
574

575 576
    secondary   = virPCIDeviceRead8(check, fd, PCI_SECONDARY_BUS);
    subordinate = virPCIDeviceRead8(check, fd, PCI_SUBORDINATE_BUS);
577

578
    VIR_DEBUG("%s %s: found parent device %s", dev->id, dev->name, check->name);
579

580 581 582
    /* if the secondary bus exactly equals the device's bus, then we found
     * the direct parent.  No further work is necessary
     */
583 584 585 586
    if (dev->bus == secondary) {
        ret = 1;
        goto cleanup;
    }
587 588 589 590 591 592 593

    /* otherwise, SRIOV allows VFs to be on different busses then their PFs.
     * In this case, what we need to do is look for the "best" match; i.e.
     * the most restrictive match that still satisfies all of the conditions.
     */
    if (dev->bus > secondary && dev->bus <= subordinate) {
        if (*best == NULL) {
594 595
            *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                    check->function);
596 597 598 599 600
            if (*best == NULL) {
                ret = -1;
                goto cleanup;
            }
        } else {
601 602 603 604
            /* OK, we had already recorded a previous "best" match for the
             * parent.  See if the current device is more restrictive than the
             * best, and if so, make it the new best
             */
605 606 607
            int bestfd;
            uint8_t best_secondary;

608
            if ((bestfd = virPCIDeviceConfigOpen(*best, false)) < 0)
609
                goto cleanup;
610 611
            best_secondary = virPCIDeviceRead8(*best, bestfd, PCI_SECONDARY_BUS);
            virPCIDeviceConfigClose(*best, bestfd);
612 613

            if (secondary > best_secondary) {
614 615 616
                virPCIDeviceFree(*best);
                *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                        check->function);
617 618 619 620
                if (*best == NULL) {
                    ret = -1;
                    goto cleanup;
                }
621 622 623 624
            }
        }
    }

625
cleanup:
626
    virPCIDeviceConfigClose(check, fd);
627
    return ret;
628 629
}

630
static int
631
virPCIDeviceGetParent(virPCIDevicePtr dev, virPCIDevicePtr *parent)
632
{
633
    virPCIDevicePtr best = NULL;
634 635 636
    int ret;

    *parent = NULL;
637
    ret = virPCIDeviceIterDevices(virPCIDeviceIsParent, dev, parent, &best);
638
    if (ret == 1)
639
        virPCIDeviceFree(best);
640 641 642
    else if (ret == 0)
        *parent = best;
    return ret;
643 644 645 646 647 648
}

/* Secondary Bus Reset is our sledgehammer - it resets all
 * devices behind a bus.
 */
static int
649 650 651
virPCIDeviceTrySecondaryBusReset(virPCIDevicePtr dev,
                                 int cfgfd,
                                 virPCIDeviceList *inactiveDevs)
652
{
653
    virPCIDevicePtr parent, conflict;
654 655 656
    uint8_t config_space[PCI_CONF_LEN];
    uint16_t ctl;
    int ret = -1;
657
    int parentfd;
658

659 660 661
    /* Refuse to do a secondary bus reset if there are other
     * devices/functions behind the bus are used by the host
     * or other guests.
662
     */
663
    if ((conflict = virPCIDeviceBusContainsActiveDevices(dev, inactiveDevs))) {
664
        virReportError(VIR_ERR_INTERNAL_ERROR,
665 666
                       _("Active %s devices on bus with %s, not doing bus reset"),
                       conflict->name, dev->name);
667 668 669 670
        return -1;
    }

    /* Find the parent bus */
671
    if (virPCIDeviceGetParent(dev, &parent) < 0)
672
        return -1;
673
    if (!parent) {
674
        virReportError(VIR_ERR_INTERNAL_ERROR,
675 676
                       _("Failed to find parent device for %s"),
                       dev->name);
677 678
        return -1;
    }
679
    if ((parentfd = virPCIDeviceConfigOpen(parent, true)) < 0)
680
        goto out;
681 682 683 684 685 686 687

    VIR_DEBUG("%s %s: doing a secondary bus reset", dev->id, dev->name);

    /* Save and restore the device's config space; we only do this
     * for the supplied device since we refuse to do a reset if there
     * are multiple devices/functions
     */
688
    if (virPCIDeviceRead(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
689
        virReportError(VIR_ERR_INTERNAL_ERROR,
690
                       _("Failed to read PCI config space for %s"),
691
                       dev->name);
692 693 694 695 696 697
        goto out;
    }

    /* Read the control register, set the reset flag, wait 200ms,
     * unset the reset flag and wait 200ms.
     */
698
    ctl = virPCIDeviceRead16(dev, cfgfd, PCI_BRIDGE_CONTROL);
699

700 701
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL,
                        ctl | PCI_BRIDGE_CTL_RESET);
702 703 704

    usleep(200 * 1000); /* sleep 200ms */

705
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL, ctl);
706 707 708

    usleep(200 * 1000); /* sleep 200ms */

709
    if (virPCIDeviceWrite(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
710
        virReportError(VIR_ERR_INTERNAL_ERROR,
711 712 713 714
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        goto out;
    }
715
    ret = 0;
716

717
out:
718 719
    virPCIDeviceConfigClose(parent, parentfd);
    virPCIDeviceFree(parent);
720 721 722 723 724 725 726 727
    return ret;
}

/* Power management reset attempts to reset a device using a
 * D-state transition from D3hot to D0. Note, in detect_pm_reset()
 * above we require the device supports a full internal reset.
 */
static int
728
virPCIDeviceTryPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
729 730 731 732 733 734 735 736
{
    uint8_t config_space[PCI_CONF_LEN];
    uint32_t ctl;

    if (!dev->pci_pm_cap_pos)
        return -1;

    /* Save and restore the device's config space. */
737
    if (virPCIDeviceRead(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
738
        virReportError(VIR_ERR_INTERNAL_ERROR,
739
                       _("Failed to read PCI config space for %s"),
740
                       dev->name);
741 742 743 744 745
        return -1;
    }

    VIR_DEBUG("%s %s: doing a power management reset", dev->id, dev->name);

746
    ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
747 748
    ctl &= ~PCI_PM_CTRL_STATE_MASK;

749 750
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D3hot);
751 752 753

    usleep(10 * 1000); /* sleep 10ms */

754 755
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D0);
756 757 758

    usleep(10 * 1000); /* sleep 10ms */

759
    if (virPCIDeviceWrite(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
760
        virReportError(VIR_ERR_INTERNAL_ERROR,
761 762 763 764
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        return -1;
    }
765 766 767 768 769

    return 0;
}

static int
770
virPCIDeviceInit(virPCIDevicePtr dev, int cfgfd)
771
{
772 773
    int flr;

774 775 776
    dev->pcie_cap_pos   = virPCIDeviceFindCapabilityOffset(dev, cfgfd, PCI_CAP_ID_EXP);
    dev->pci_pm_cap_pos = virPCIDeviceFindCapabilityOffset(dev, cfgfd, PCI_CAP_ID_PM);
    flr = virPCIDeviceDetectFunctionLevelReset(dev, cfgfd);
777
    if (flr < 0)
778 779
        return flr;
    dev->has_flr        = flr;
780
    dev->has_pm_reset   = virPCIDeviceDetectPowerManagementReset(dev, cfgfd);
781

782 783 784 785
    return 0;
}

int
786 787 788
virPCIDeviceReset(virPCIDevicePtr dev,
                  virPCIDeviceList *activeDevs,
                  virPCIDeviceList *inactiveDevs)
789 790
{
    int ret = -1;
791
    int fd;
792

793
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
794
        virReportError(VIR_ERR_INTERNAL_ERROR,
795 796 797 798
                       _("Not resetting active device %s"), dev->name);
        return -1;
    }

799
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
800 801
        return -1;

802
    if (virPCIDeviceInit(dev, fd) < 0)
803 804
        goto cleanup;

805 806 807
    /* KVM will perform FLR when starting and stopping
     * a guest, so there is no need for us to do it here.
     */
808 809 810 811
    if (dev->has_flr) {
        ret = 0;
        goto cleanup;
    }
812

813 814 815 816 817
    /* If the device supports PCI power management reset,
     * that's the next best thing because it only resets
     * the function, not the whole device.
     */
    if (dev->has_pm_reset)
818
        ret = virPCIDeviceTryPowerManagementReset(dev, fd);
819

820
    /* Bus reset is not an option with the root bus */
821
    if (ret < 0 && dev->bus != 0)
822
        ret = virPCIDeviceTrySecondaryBusReset(dev, fd, inactiveDevs);
823

824 825
    if (ret < 0) {
        virErrorPtr err = virGetLastError();
826
        virReportError(VIR_ERR_INTERNAL_ERROR,
827 828 829 830 831
                       _("Unable to reset PCI device %s: %s"),
                       dev->name,
                       err ? err->message : _("no FLR, PM reset or bus reset available"));
    }

832
cleanup:
833
    virPCIDeviceConfigClose(dev, fd);
834 835 836
    return ret;
}

837

838
static int
839
virPCIDriverDir(char **buffer, const char *driver)
840
{
841 842 843 844 845 846 847 848
    VIR_FREE(*buffer);

    if (virAsprintf(buffer, PCI_SYSFS "drivers/%s", driver) < 0) {
        virReportOOMError();
        return -1;
    }

    return 0;
849 850
}

851
static int
852
virPCIDriverFile(char **buffer, const char *driver, const char *file)
853
{
854 855 856 857 858 859 860 861
    VIR_FREE(*buffer);

    if (virAsprintf(buffer, PCI_SYSFS "drivers/%s/%s", driver, file) < 0) {
        virReportOOMError();
        return -1;
    }

    return 0;
862
}
863

864
static int
865
virPCIFile(char **buffer, const char *device, const char *file)
866
{
867 868 869 870 871 872 873 874
    VIR_FREE(*buffer);

    if (virAsprintf(buffer, PCI_SYSFS "devices/%s/%s", device, file) < 0) {
        virReportOOMError();
        return -1;
    }

    return 0;
875
}
876

877
static int
878
virPCIProbeStubDriver(const char *driver)
879
{
880
    char *drvpath = NULL;
881 882 883
    int probed = 0;

recheck:
884
    if (virPCIDriverDir(&drvpath, driver) == 0 && virFileExists(drvpath)) {
885
        /* driver already loaded, return */
886
        VIR_FREE(drvpath);
887
        return 0;
888 889 890
    }

    VIR_FREE(drvpath);
891 892

    if (!probed) {
893
        const char *const probecmd[] = { MODPROBE, driver, NULL };
894
        probed = 1;
895
        if (virRun(probecmd, NULL) < 0) {
896
            char ebuf[1024];
897
            VIR_WARN("failed to load driver %s: %s", driver,
898
                     virStrerror(errno, ebuf, sizeof(ebuf)));
899
            return -1;
900
        }
901 902

        goto recheck;
903 904
    }

905
    return -1;
906 907
}

908
static int
909
virPCIDeviceUnbindFromStub(virPCIDevicePtr dev, const char *driver)
910 911 912 913 914
{
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;

915
    if (virPCIDriverDir(&drvdir, driver) < 0)
E
Eric Blake 已提交
916 917
        goto cleanup;

918 919 920
    if (!dev->unbind_from_stub)
        goto remove_slot;

921 922
    /* If the device is bound to stub, unbind it.
     */
923
    if (virPCIFile(&path, dev->name, "driver") < 0)
924 925 926
        goto cleanup;

    if (virFileExists(drvdir) && virFileLinkPointsTo(path, drvdir)) {
927
        if (virPCIDriverFile(&path, driver, "unbind") < 0) {
928 929 930 931 932
            goto cleanup;
        }

        if (virFileWriteStr(path, dev->name, 0) < 0) {
            virReportSystemError(errno,
933
                                 _("Failed to unbind PCI device '%s' from %s"),
934 935 936 937
                                 dev->name, driver);
            goto cleanup;
        }
    }
938 939 940 941 942
    dev->unbind_from_stub = 0;

remove_slot:
    if (!dev->remove_slot)
        goto reprobe;
943 944

    /* Xen's pciback.ko wants you to use remove_slot on the specific device */
945
    if (virPCIDriverFile(&path, driver, "remove_slot") < 0) {
946 947 948 949 950
        goto cleanup;
    }

    if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
        virReportSystemError(errno,
951
                             _("Failed to remove slot for PCI device '%s' from %s"),
952 953 954
                             dev->name, driver);
        goto cleanup;
    }
955 956 957 958 959 960 961
    dev->remove_slot = 0;

reprobe:
    if (!dev->reprobe) {
        result = 0;
        goto cleanup;
    }
962 963 964 965 966 967

    /* Trigger a re-probe of the device is not in the stub's dynamic
     * ID table. If the stub is available, but 'remove_id' isn't
     * available, then re-probing would just cause the device to be
     * re-bound to the stub.
     */
968
    if (virPCIDriverFile(&path, driver, "remove_id") < 0) {
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
        goto cleanup;
    }

    if (!virFileExists(drvdir) || virFileExists(path)) {
        if (virFileWriteStr(PCI_SYSFS "drivers_probe", dev->name, 0) < 0) {
            virReportSystemError(errno,
                                 _("Failed to trigger a re-probe for PCI device '%s'"),
                                 dev->name);
            goto cleanup;
        }
    }

    result = 0;

cleanup:
984 985 986 987 988
    /* do not do it again */
    dev->unbind_from_stub = 0;
    dev->remove_slot = 0;
    dev->reprobe = 0;

989 990 991 992 993 994
    VIR_FREE(drvdir);
    VIR_FREE(path);

    return result;
}

995 996

static int
997
virPCIDeviceBindToStub(virPCIDevicePtr dev, const char *driver)
998
{
999 1000 1001
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;
1002 1003 1004
    int reprobe = 0;

    /* check whether the device is already bound to a driver */
1005 1006
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
        goto cleanup;
    }

    if (virFileExists(path)) {
        if (virFileLinkPointsTo(path, drvdir)) {
            /* The device is already bound to pci-stub */
            result = 0;
            goto cleanup;
        }
        reprobe = 1;
    }
1018 1019 1020 1021 1022 1023 1024 1025 1026

    /* Add the PCI device ID to the stub's dynamic ID table;
     * this is needed to allow us to bind the device to the stub.
     * Note: if the device is not currently bound to any driver,
     * stub will immediately be bound to the device. Also, note
     * that if a new device with this ID is hotplugged, or if a probe
     * is triggered for such a device, it will also be immediately
     * bound by the stub.
     */
1027
    if (virPCIDriverFile(&path, driver, "new_id") < 0) {
1028
        goto cleanup;
1029 1030
    }

1031
    if (virFileWriteStr(path, dev->id, 0) < 0) {
1032
        virReportSystemError(errno,
1033 1034
                             _("Failed to add PCI device ID '%s' to %s"),
                             dev->id, driver);
1035
        goto cleanup;
1036 1037
    }

1038 1039 1040
    /* check whether the device is bound to pci-stub when we write dev->id to
     * new_id.
     */
1041 1042
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1043 1044 1045 1046 1047 1048 1049 1050 1051
        goto remove_id;
    }

    if (virFileLinkPointsTo(path, drvdir)) {
        dev->unbind_from_stub = 1;
        dev->remove_slot = 1;
        goto remove_id;
    }

1052 1053 1054 1055 1056
    /* If the device is already bound to a driver, unbind it.
     * Note, this will have rather unpleasant side effects if this
     * PCI device happens to be IDE controller for the disk hosting
     * your root filesystem.
     */
1057
    if (virPCIFile(&path, dev->name, "driver/unbind") < 0) {
1058 1059 1060
        goto cleanup;
    }

1061 1062 1063 1064 1065 1066 1067 1068
    if (virFileExists(path)) {
        if (virFileWriteStr(path, dev->name, 0) < 0) {
            virReportSystemError(errno,
                                 _("Failed to unbind PCI device '%s'"),
                                 dev->name);
            goto cleanup;
        }
        dev->reprobe = reprobe;
1069 1070
    }

1071 1072
    /* If the device isn't already bound to pci-stub, try binding it now.
     */
1073 1074
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1075
        goto remove_id;
1076 1077
    }

1078 1079
    if (!virFileLinkPointsTo(path, drvdir)) {
        /* Xen's pciback.ko wants you to use new_slot first */
1080
        if (virPCIDriverFile(&path, driver, "new_slot") < 0) {
1081
            goto remove_id;
1082 1083
        }

1084
        if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
1085
            virReportSystemError(errno,
1086 1087
                                 _("Failed to add slot for PCI device '%s' to %s"),
                                 dev->name, driver);
1088
            goto remove_id;
1089
        }
1090
        dev->remove_slot = 1;
1091

1092
        if (virPCIDriverFile(&path, driver, "bind") < 0) {
1093
            goto remove_id;
1094 1095
        }

1096
        if (virFileWriteStr(path, dev->name, 0) < 0) {
1097
            virReportSystemError(errno,
1098 1099
                                 _("Failed to bind PCI device '%s' to %s"),
                                 dev->name, driver);
1100
            goto remove_id;
1101
        }
1102
        dev->unbind_from_stub = 1;
1103 1104
    }

1105
remove_id:
1106 1107 1108
    /* If 'remove_id' exists, remove the device id from pci-stub's dynamic
     * ID table so that 'drivers_probe' works below.
     */
1109
    if (virPCIDriverFile(&path, driver, "remove_id") < 0) {
E
Eric Blake 已提交
1110
        /* We do not remove PCI ID from pci-stub, and we cannot reprobe it */
1111 1112 1113 1114 1115
        if (dev->reprobe) {
            VIR_WARN("Could not remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
        dev->reprobe = 0;
1116 1117 1118
        goto cleanup;
    }

1119
    if (virFileExists(path) && virFileWriteStr(path, dev->id, 0) < 0) {
1120
        virReportSystemError(errno,
1121 1122
                             _("Failed to remove PCI ID '%s' from %s"),
                             dev->id, driver);
1123

E
Eric Blake 已提交
1124
        /* remove PCI ID from pci-stub failed, and we cannot reprobe it */
1125 1126 1127 1128 1129
        if (dev->reprobe) {
            VIR_WARN("Failed to remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
        dev->reprobe = 0;
1130
        goto cleanup;
1131 1132
    }

1133 1134 1135 1136 1137 1138
    result = 0;

cleanup:
    VIR_FREE(drvdir);
    VIR_FREE(path);

1139
    if (result < 0) {
1140
        virPCIDeviceUnbindFromStub(dev, driver);
1141 1142
    }

1143
    return result;
1144 1145 1146
}

int
1147 1148 1149 1150
virPCIDeviceDetach(virPCIDevicePtr dev,
                   virPCIDeviceList *activeDevs,
                   virPCIDeviceList *inactiveDevs,
                   const char *driver)
1151
{
1152
    if (virPCIProbeStubDriver(driver) < 0) {
1153 1154
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Failed to load PCI stub module %s"), driver);
1155 1156 1157
        return -1;
    }

1158
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
1159
        virReportError(VIR_ERR_INTERNAL_ERROR,
1160 1161 1162 1163
                       _("Not detaching active device %s"), dev->name);
        return -1;
    }

1164
    if (virPCIDeviceBindToStub(dev, driver) < 0)
1165 1166 1167
        return -1;

    /* Add the dev into list inactiveDevs */
1168 1169
    if (inactiveDevs && !virPCIDeviceListFind(inactiveDevs, dev)) {
        if (virPCIDeviceListAdd(inactiveDevs, dev) < 0)
1170 1171 1172 1173
            return -1;
    }

    return 0;
1174 1175 1176
}

int
1177 1178 1179 1180
virPCIDeviceReattach(virPCIDevicePtr dev,
                     virPCIDeviceListPtr activeDevs,
                     virPCIDeviceListPtr inactiveDevs,
                     const char *driver)
1181
{
1182
    if (virPCIProbeStubDriver(driver) < 0) {
1183 1184
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Failed to load PCI stub module %s"), driver);
1185 1186 1187
        return -1;
    }

1188
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
1189
        virReportError(VIR_ERR_INTERNAL_ERROR,
1190 1191 1192 1193
                       _("Not reattaching active device %s"), dev->name);
        return -1;
    }

1194
    if (virPCIDeviceUnbindFromStub(dev, driver) < 0)
1195 1196 1197 1198
        return -1;

    /* Steal the dev from list inactiveDevs */
    if (inactiveDevs)
1199
        virPCIDeviceListSteal(inactiveDevs, dev);
1200 1201

    return 0;
1202 1203
}

1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233
/* Certain hypervisors (like qemu/kvm) map the PCI bar(s) on
 * the host when doing device passthrough.  This can lead to a race
 * condition where the hypervisor is still cleaning up the device while
 * libvirt is trying to re-attach it to the host device driver.  To avoid
 * this situation, we look through /proc/iomem, and if the hypervisor is
 * still holding onto the bar (denoted by the string in the matcher variable),
 * then we can wait around a bit for that to clear up.
 *
 * A typical /proc/iomem looks like this (snipped for brevity):
 * 00010000-0008efff : System RAM
 * 0008f000-0008ffff : reserved
 * ...
 * 00100000-cc9fcfff : System RAM
 *   00200000-00483d3b : Kernel code
 *   00483d3c-005c88df : Kernel data
 * cc9fd000-ccc71fff : ACPI Non-volatile Storage
 * ...
 * d0200000-d02fffff : PCI Bus #05
 *   d0200000-d021ffff : 0000:05:00.0
 *     d0200000-d021ffff : e1000e
 *   d0220000-d023ffff : 0000:05:00.0
 *     d0220000-d023ffff : e1000e
 * ...
 * f0000000-f0003fff : 0000:00:1b.0
 *   f0000000-f0003fff : kvm_assigned_device
 *
 * Returns 0 if we are clear to continue, and 1 if the hypervisor is still
 * holding onto the resource.
 */
int
1234
virPCIDeviceWaitForCleanup(virPCIDevicePtr dev, const char *matcher)
1235 1236 1237
{
    FILE *fp;
    char line[160];
1238
    char *tmp;
1239
    unsigned long long start, end;
1240
    unsigned int domain, bus, slot, function;
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250
    int in_matching_device;
    int ret;
    size_t match_depth;

    fp = fopen("/proc/iomem", "r");
    if (!fp) {
        /* If we failed to open iomem, we just basically ignore the error.  The
         * unbind might succeed anyway, and besides, it's very likely we have
         * no way to report the error
         */
1251
        VIR_DEBUG("Failed to open /proc/iomem, trying to continue anyway");
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
        return 0;
    }

    ret = 0;
    in_matching_device = 0;
    match_depth = 0;
    while (fgets(line, sizeof(line), fp) != 0) {
        /* the logic here is a bit confusing.  For each line, we look to
         * see if it matches the domain:bus:slot.function we were given.
         * If this line matches the DBSF, then any subsequent lines indented
         * by 2 spaces are the PCI regions for this device.  It's also
         * possible that none of the PCI regions are currently mapped, in
         * which case we have no indented regions.  This code handles all
         * of these situations
         */
        if (in_matching_device && (strspn(line, " ") == (match_depth + 2))) {
1268 1269 1270 1271 1272 1273
            /* expected format: <start>-<end> : <suffix> */
            if (/* start */
                virStrToLong_ull(line, &tmp, 16, &start) < 0 || *tmp != '-' ||
                /* end */
                virStrToLong_ull(tmp + 1, &tmp, 16, &end) < 0 ||
                (tmp = STRSKIP(tmp, " : ")) == NULL)
1274 1275
                continue;

1276
            if (STRPREFIX(tmp, matcher)) {
1277 1278 1279 1280 1281 1282 1283
                ret = 1;
                break;
            }
        }
        else {
            in_matching_device = 0;

1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
            /* expected format: <start>-<end> : <domain>:<bus>:<slot>.<function> */
            if (/* start */
                virStrToLong_ull(line, &tmp, 16, &start) < 0 || *tmp != '-' ||
                /* end */
                virStrToLong_ull(tmp + 1, &tmp, 16, &end) < 0 ||
                (tmp = STRSKIP(tmp, " : ")) == NULL ||
                /* domain */
                virStrToLong_ui(tmp, &tmp, 16, &domain) < 0 || *tmp != ':' ||
                /* bus */
                virStrToLong_ui(tmp + 1, &tmp, 16, &bus) < 0 || *tmp != ':' ||
                /* slot */
                virStrToLong_ui(tmp + 1, &tmp, 16, &slot) < 0 || *tmp != '.' ||
                /* function */
                virStrToLong_ui(tmp + 1, &tmp, 16, &function) < 0 || *tmp != '\n')
1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
                continue;

            if (domain != dev->domain || bus != dev->bus || slot != dev->slot ||
                function != dev->function)
                continue;
            in_matching_device = 1;
            match_depth = strspn(line, " ");
        }
    }

E
Eric Blake 已提交
1308
    VIR_FORCE_FCLOSE(fp);
1309 1310 1311 1312

    return ret;
}

1313
static char *
1314
virPCIDeviceReadID(virPCIDevicePtr dev, const char *id_name)
1315
{
1316
    char *path = NULL;
1317 1318
    char *id_str;

1319
    if (virPCIFile(&path, dev->name, id_name) < 0) {
1320 1321
        return NULL;
    }
1322 1323

    /* ID string is '0xNNNN\n' ... i.e. 7 bytes */
1324 1325
    if (virFileReadAll(path, 7, &id_str) < 0) {
        VIR_FREE(path);
1326
        return NULL;
1327 1328 1329
    }

    VIR_FREE(path);
1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342

    /* Check for 0x suffix */
    if (id_str[0] != '0' || id_str[1] != 'x') {
        VIR_FREE(id_str);
        return NULL;
    }

    /* Chop off the newline; we know the string is 7 bytes */
    id_str[6] = '\0';

    return id_str;
}

1343
int
1344 1345 1346 1347 1348
virPCIGetAddrString(unsigned domain,
                    unsigned bus,
                    unsigned slot,
                    unsigned function,
                    char **pciConfigAddr)
1349
{
1350
    virPCIDevicePtr dev = NULL;
1351 1352
    int ret = -1;

1353
    dev = virPCIDeviceNew(domain, bus, slot, function);
1354 1355 1356 1357 1358 1359 1360 1361 1362
    if (dev != NULL) {
        if ((*pciConfigAddr = strdup(dev->name)) == NULL) {
            virReportOOMError();
            goto cleanup;
        }
        ret = 0;
    }

cleanup:
1363
    virPCIDeviceFree(dev);
1364 1365 1366
    return ret;
}

1367 1368 1369 1370 1371
virPCIDevicePtr
virPCIDeviceNew(unsigned domain,
                unsigned bus,
                unsigned slot,
                unsigned function)
1372
{
1373
    virPCIDevicePtr dev;
E
Eric Blake 已提交
1374 1375
    char *vendor = NULL;
    char *product = NULL;
1376 1377

    if (VIR_ALLOC(dev) < 0) {
1378
        virReportOOMError();
1379 1380 1381 1382 1383 1384 1385 1386
        return NULL;
    }

    dev->domain   = domain;
    dev->bus      = bus;
    dev->slot     = slot;
    dev->function = function;

E
Eric Blake 已提交
1387 1388 1389
    if (snprintf(dev->name, sizeof(dev->name), "%.4x:%.2x:%.2x.%.1x",
                 dev->domain, dev->bus, dev->slot,
                 dev->function) >= sizeof(dev->name)) {
1390
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1391 1392
                       _("dev->name buffer overflow: %.4x:%.2x:%.2x.%.1x"),
                       dev->domain, dev->bus, dev->slot, dev->function);
E
Eric Blake 已提交
1393
        goto error;
E
Eric Blake 已提交
1394 1395 1396 1397
    }
    if (virAsprintf(&dev->path, PCI_SYSFS "devices/%s/config",
                    dev->name) < 0) {
        virReportOOMError();
E
Eric Blake 已提交
1398
        goto error;
E
Eric Blake 已提交
1399
    }
1400

1401 1402 1403 1404
    if (access(dev->path, F_OK) != 0) {
        virReportSystemError(errno,
                             _("Device %s not found: could not access %s"),
                             dev->name, dev->path);
E
Eric Blake 已提交
1405
        goto error;
1406 1407
    }

1408 1409
    vendor  = virPCIDeviceReadID(dev, "vendor");
    product = virPCIDeviceReadID(dev, "device");
1410 1411

    if (!vendor || !product) {
1412
        virReportError(VIR_ERR_INTERNAL_ERROR,
1413 1414
                       _("Failed to read product/vendor ID for %s"),
                       dev->name);
E
Eric Blake 已提交
1415
        goto error;
1416 1417 1418
    }

    /* strings contain '0x' prefix */
E
Eric Blake 已提交
1419 1420
    if (snprintf(dev->id, sizeof(dev->id), "%s %s", &vendor[2],
                 &product[2]) >= sizeof(dev->id)) {
1421
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1422 1423
                       _("dev->id buffer overflow: %s %s"),
                       &vendor[2], &product[2]);
E
Eric Blake 已提交
1424
        goto error;
E
Eric Blake 已提交
1425
    }
1426 1427 1428

    VIR_DEBUG("%s %s: initialized", dev->id, dev->name);

E
Eric Blake 已提交
1429 1430 1431
cleanup:
    VIR_FREE(product);
    VIR_FREE(vendor);
1432
    return dev;
E
Eric Blake 已提交
1433 1434

error:
1435
    virPCIDeviceFree(dev);
E
Eric Blake 已提交
1436 1437
    dev = NULL;
    goto cleanup;
1438 1439 1440
}

void
1441
virPCIDeviceFree(virPCIDevicePtr dev)
1442
{
1443 1444
    if (!dev)
        return;
1445
    VIR_DEBUG("%s %s: freeing", dev->id, dev->name);
E
Eric Blake 已提交
1446
    VIR_FREE(dev->path);
1447 1448
    VIR_FREE(dev);
}
1449

1450
const char *
1451
virPCIDeviceGetName(virPCIDevicePtr dev)
1452 1453 1454 1455
{
    return dev->name;
}

1456
void virPCIDeviceSetManaged(virPCIDevicePtr dev, unsigned managed)
1457 1458 1459 1460
{
    dev->managed = !!managed;
}

1461
unsigned virPCIDeviceGetManaged(virPCIDevicePtr dev)
1462 1463 1464 1465
{
    return dev->managed;
}

1466
unsigned
1467
virPCIDeviceGetUnbindFromStub(virPCIDevicePtr dev)
1468 1469 1470 1471 1472
{
    return dev->unbind_from_stub;
}

void
1473
virPCIDeviceSetUnbindFromStub(virPCIDevicePtr dev, unsigned unbind)
1474 1475 1476 1477 1478
{
    dev->unbind_from_stub = !!unbind;
}

unsigned
1479
virPCIDeviceGetRemoveSlot(virPCIDevicePtr dev)
1480 1481 1482 1483 1484
{
    return dev->remove_slot;
}

void
1485
virPCIDeviceSetRemoveSlot(virPCIDevicePtr dev, unsigned remove_slot)
1486 1487 1488 1489 1490
{
    dev->remove_slot = !!remove_slot;
}

unsigned
1491
virPCIDeviceGetReprobe(virPCIDevicePtr dev)
1492 1493 1494 1495 1496
{
    return dev->reprobe;
}

void
1497
virPCIDeviceSetReprobe(virPCIDevicePtr dev, unsigned reprobe)
1498 1499 1500 1501
{
    dev->reprobe = !!reprobe;
}

1502
void
1503
virPCIDeviceSetUsedBy(virPCIDevicePtr dev, const char *name)
1504 1505 1506 1507 1508
{
    dev->used_by = name;
}

const char *
1509
virPCIDeviceGetUsedBy(virPCIDevicePtr dev)
1510 1511 1512 1513
{
    return dev->used_by;
}

1514
void virPCIDeviceReattachInit(virPCIDevicePtr pci)
1515 1516 1517 1518 1519 1520 1521
{
    pci->unbind_from_stub = 1;
    pci->remove_slot = 1;
    pci->reprobe = 1;
}


1522 1523
virPCIDeviceListPtr
virPCIDeviceListNew(void)
1524
{
1525
    virPCIDeviceListPtr list;
1526

1527 1528 1529 1530
    if (virPCIInitialize() < 0)
        return NULL;

    if (!(list = virObjectLockableNew(virPCIDeviceListClass)))
1531 1532 1533 1534 1535
        return NULL;

    return list;
}

1536 1537
static void
virPCIDeviceListDispose(void *obj)
1538
{
1539
    virPCIDeviceListPtr list = obj;
1540 1541 1542
    int i;

    for (i = 0; i < list->count; i++) {
1543
        virPCIDeviceFree(list->devs[i]);
1544 1545 1546 1547 1548 1549 1550 1551
        list->devs[i] = NULL;
    }

    list->count = 0;
    VIR_FREE(list->devs);
}

int
1552 1553
virPCIDeviceListAdd(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1554
{
1555
    if (virPCIDeviceListFind(list, dev)) {
1556
        virReportError(VIR_ERR_INTERNAL_ERROR,
1557 1558 1559 1560 1561
                       _("Device %s is already in use"), dev->name);
        return -1;
    }

    if (VIR_REALLOC_N(list->devs, list->count+1) < 0) {
1562
        virReportOOMError();
1563 1564 1565 1566 1567 1568 1569 1570
        return -1;
    }

    list->devs[list->count++] = dev;

    return 0;
}

1571 1572 1573
virPCIDevicePtr
virPCIDeviceListGet(virPCIDeviceListPtr list,
                    int idx)
1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
{
    if (idx >= list->count)
        return NULL;
    if (idx < 0)
        return NULL;

    return list->devs[idx];
}

int
1584
virPCIDeviceListCount(virPCIDeviceListPtr list)
1585
{
1586 1587 1588
    return list->count;
}

1589 1590 1591
virPCIDevicePtr
virPCIDeviceListStealIndex(virPCIDeviceListPtr list,
                           int idx)
1592
{
1593
    virPCIDevicePtr ret;
1594

1595 1596
    if (idx < 0 || idx >= list->count)
        return NULL;
1597

1598
    ret = list->devs[idx];
1599

1600 1601 1602 1603 1604
    if (idx != --list->count) {
        memmove(&list->devs[idx],
                &list->devs[idx + 1],
                sizeof(*list->devs) * (list->count - idx));
    }
1605

1606 1607
    if (VIR_REALLOC_N(list->devs, list->count) < 0) {
        ; /* not fatal */
1608
    }
1609

1610 1611 1612
    return ret;
}

1613 1614 1615
virPCIDevicePtr
virPCIDeviceListSteal(virPCIDeviceListPtr list,
                      virPCIDevicePtr dev)
1616
{
1617
    return virPCIDeviceListStealIndex(list, virPCIDeviceListFindIndex(list, dev));
1618 1619
}

1620
void
1621 1622
virPCIDeviceListDel(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1623
{
1624
    virPCIDevicePtr ret = virPCIDeviceListSteal(list, dev);
1625
    if (ret)
1626
        virPCIDeviceFree(ret);
1627 1628
}

1629
int
1630
virPCIDeviceListFindIndex(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1631 1632 1633 1634 1635 1636 1637 1638
{
    int i;

    for (i = 0; i < list->count; i++)
        if (list->devs[i]->domain   == dev->domain &&
            list->devs[i]->bus      == dev->bus    &&
            list->devs[i]->slot     == dev->slot   &&
            list->devs[i]->function == dev->function)
1639 1640 1641 1642
            return i;
    return -1;
}

1643 1644
virPCIDevicePtr
virPCIDeviceListFind(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1645 1646 1647
{
    int i;

1648
    if ((i = virPCIDeviceListFindIndex(list, dev)) >= 0)
1649 1650 1651
        return list->devs[i];
    else
        return NULL;
1652
}
1653 1654


1655 1656 1657
int virPCIDeviceFileIterate(virPCIDevicePtr dev,
                            virPCIDeviceFileActor actor,
                            void *opaque)
1658 1659 1660 1661 1662 1663 1664 1665 1666
{
    char *pcidir = NULL;
    char *file = NULL;
    DIR *dir = NULL;
    int ret = -1;
    struct dirent *ent;

    if (virAsprintf(&pcidir, "/sys/bus/pci/devices/%04x:%02x:%02x.%x",
                    dev->domain, dev->bus, dev->slot, dev->function) < 0) {
1667
        virReportOOMError();
1668 1669 1670 1671
        goto cleanup;
    }

    if (!(dir = opendir(pcidir))) {
1672
        virReportSystemError(errno,
1673 1674 1675 1676 1677 1678
                             _("cannot open %s"), pcidir);
        goto cleanup;
    }

    while ((ent = readdir(dir)) != NULL) {
        /* Device assignment requires:
A
Alex Williamson 已提交
1679 1680
         *   $PCIDIR/config, $PCIDIR/resource, $PCIDIR/resourceNNN,
         *   $PCIDIR/rom, $PCIDIR/reset
1681 1682 1683
         */
        if (STREQ(ent->d_name, "config") ||
            STRPREFIX(ent->d_name, "resource") ||
A
Alex Williamson 已提交
1684 1685
            STREQ(ent->d_name, "rom") ||
            STREQ(ent->d_name, "reset")) {
1686
            if (virAsprintf(&file, "%s/%s", pcidir, ent->d_name) < 0) {
1687
                virReportOOMError();
1688 1689
                goto cleanup;
            }
1690
            if ((actor)(dev, file, opaque) < 0)
1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
                goto cleanup;

            VIR_FREE(file);
        }
    }

    ret = 0;

cleanup:
    if (dir)
        closedir(dir);
    VIR_FREE(file);
    VIR_FREE(pcidir);
    return ret;
}
J
Jiri Denemark 已提交
1706 1707

static int
1708
virPCIDeviceDownstreamLacksACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1709 1710 1711 1712
{
    uint16_t flags;
    uint16_t ctrl;
    unsigned int pos;
1713 1714
    int fd;
    int ret = 0;
J
Jiri Denemark 已提交
1715

1716
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
J
Jiri Denemark 已提交
1717 1718
        return -1;

1719
    if (virPCIDeviceInit(dev, fd) < 0) {
1720 1721 1722 1723
        ret = -1;
        goto cleanup;
    }

J
Jiri Denemark 已提交
1724
    pos = dev->pcie_cap_pos;
1725
    if (!pos || virPCIDeviceRead16(dev, fd, PCI_CLASS_DEVICE) != PCI_CLASS_BRIDGE_PCI)
1726
        goto cleanup;
J
Jiri Denemark 已提交
1727

1728
    flags = virPCIDeviceRead16(dev, fd, pos + PCI_EXP_FLAGS);
J
Jiri Denemark 已提交
1729
    if (((flags & PCI_EXP_FLAGS_TYPE) >> 4) != PCI_EXP_TYPE_DOWNSTREAM)
1730
        goto cleanup;
J
Jiri Denemark 已提交
1731

1732
    pos = virPCIDeviceFindExtendedCapabilityOffset(dev, fd, PCI_EXT_CAP_ID_ACS);
J
Jiri Denemark 已提交
1733 1734
    if (!pos) {
        VIR_DEBUG("%s %s: downstream port lacks ACS", dev->id, dev->name);
1735 1736
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1737 1738
    }

1739
    ctrl = virPCIDeviceRead16(dev, fd, pos + PCI_EXT_ACS_CTRL);
J
Jiri Denemark 已提交
1740 1741 1742
    if ((ctrl & PCI_EXT_CAP_ACS_ENABLED) != PCI_EXT_CAP_ACS_ENABLED) {
        VIR_DEBUG("%s %s: downstream port has ACS disabled",
                  dev->id, dev->name);
1743 1744
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1745 1746
    }

1747
cleanup:
1748
    virPCIDeviceConfigClose(dev, fd);
1749
    return ret;
J
Jiri Denemark 已提交
1750 1751 1752
}

static int
1753
virPCIDeviceIsBehindSwitchLackingACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1754
{
1755
    virPCIDevicePtr parent;
J
Jiri Denemark 已提交
1756

1757
    if (virPCIDeviceGetParent(dev, &parent) < 0)
1758
        return -1;
1759 1760 1761 1762 1763 1764 1765 1766
    if (!parent) {
        /* if we have no parent, and this is the root bus, ACS doesn't come
         * into play since devices on the root bus can't P2P without going
         * through the root IOMMU.
         */
        if (dev->bus == 0)
            return 0;
        else {
1767
            virReportError(VIR_ERR_INTERNAL_ERROR,
1768 1769 1770 1771
                           _("Failed to find parent device for %s"),
                           dev->name);
            return -1;
        }
J
Jiri Denemark 已提交
1772 1773 1774 1775 1776 1777 1778
    }

    /* XXX we should rather fail when we can't find device's parent and
     * stop the loop when we get to root instead of just stopping when no
     * parent can be found
     */
    do {
1779
        virPCIDevicePtr tmp;
J
Jiri Denemark 已提交
1780
        int acs;
1781
        int ret;
J
Jiri Denemark 已提交
1782

1783
        acs = virPCIDeviceDownstreamLacksACS(parent);
J
Jiri Denemark 已提交
1784 1785

        if (acs) {
1786
            virPCIDeviceFree(parent);
J
Jiri Denemark 已提交
1787 1788 1789 1790 1791 1792 1793
            if (acs < 0)
                return -1;
            else
                return 1;
        }

        tmp = parent;
1794 1795
        ret = virPCIDeviceGetParent(parent, &parent);
        virPCIDeviceFree(tmp);
1796 1797
        if (ret < 0)
            return -1;
J
Jiri Denemark 已提交
1798 1799 1800 1801 1802
    } while (parent);

    return 0;
}

1803 1804
int virPCIDeviceIsAssignable(virPCIDevicePtr dev,
                             int strict_acs_check)
J
Jiri Denemark 已提交
1805 1806 1807 1808 1809 1810 1811 1812
{
    int ret;

    /* XXX This could be a great place to actually check that a non-managed
     * device isn't in use, e.g. by checking that device is either un-bound
     * or bound to a stub driver.
     */

1813
    ret = virPCIDeviceIsBehindSwitchLackingACS(dev);
J
Jiri Denemark 已提交
1814 1815 1816 1817 1818 1819 1820 1821
    if (ret < 0)
        return 0;

    if (ret) {
        if (!strict_acs_check) {
            VIR_DEBUG("%s %s: strict ACS check disabled; device assignment allowed",
                      dev->id, dev->name);
        } else {
1822
            virReportError(VIR_ERR_INTERNAL_ERROR,
J
Jiri Denemark 已提交
1823 1824 1825 1826 1827 1828 1829 1830 1831
                           _("Device %s is behind a switch lacking ACS and "
                             "cannot be assigned"),
                           dev->name);
            return 0;
        }
    }

    return 1;
}
1832

1833 1834
#ifdef __linux__

1835
/*
1836
 * returns true if equal
1837
 */
1838
static bool
1839 1840
virPCIDeviceAddressIsEqual(virPCIDeviceAddressPtr bdf1,
                           virPCIDeviceAddressPtr bdf2)
1841 1842 1843 1844 1845 1846 1847
{
    return ((bdf1->domain == bdf2->domain) &&
            (bdf1->bus == bdf2->bus) &&
            (bdf1->slot == bdf2->slot) &&
            (bdf1->function == bdf2->function));
}

1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866
static int
logStrToLong_ui(char const *s,
                char **end_ptr,
                int base,
                unsigned int *result)
{
    int ret = 0;

    ret = virStrToLong_ui(s, end_ptr, base, result);
    if (ret != 0) {
        VIR_ERROR(_("Failed to convert '%s' to unsigned int"), s);
    } else {
        VIR_DEBUG("Converted '%s' to unsigned int %u", s, *result);
    }

    return ret;
}

static int
1867 1868
virPCIParseDeviceAddress(char *address,
                         virPCIDeviceAddressPtr bdf)
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899
{
    char *p = NULL;
    int ret = -1;

    if ((address == NULL) || (logStrToLong_ui(address, &p, 16,
                                              &bdf->domain) == -1)) {
        goto out;
    }

    if ((p == NULL) || (logStrToLong_ui(p+1, &p, 16,
                                        &bdf->bus) == -1)) {
        goto out;
    }

    if ((p == NULL) || (logStrToLong_ui(p+1, &p, 16,
                                        &bdf->slot) == -1)) {
        goto out;
    }

    if ((p == NULL) || (logStrToLong_ui(p+1, &p, 16,
                                        &bdf->function) == -1)) {
        goto out;
    }

    ret = 0;

out:
    return ret;
}

static int
1900 1901
virPCIGetDeviceAddressFromSysfsLink(const char *device_link,
                                    virPCIDeviceAddressPtr *bdf)
1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915
{
    char *config_address = NULL;
    char *device_path = NULL;
    char errbuf[64];
    int ret = -1;

    VIR_DEBUG("Attempting to resolve device path from device link '%s'",
              device_link);

    if (!virFileExists(device_link)) {
        VIR_DEBUG("sysfs_path '%s' does not exist", device_link);
        return ret;
    }

1916
    device_path = canonicalize_file_name(device_link);
1917 1918
    if (device_path == NULL) {
        memset(errbuf, '\0', sizeof(errbuf));
1919 1920 1921
        virReportSystemError(errno,
                             _("Failed to resolve device link '%s'"),
                             device_link);
1922 1923 1924 1925 1926 1927 1928 1929 1930
        return ret;
    }

    config_address = basename(device_path);
    if (VIR_ALLOC(*bdf) != 0) {
        virReportOOMError();
        goto out;
    }

1931
    if (virPCIParseDeviceAddress(config_address, *bdf) != 0) {
1932
        virReportError(VIR_ERR_INTERNAL_ERROR,
1933 1934 1935 1936 1937 1938
                       _("Failed to parse PCI config address '%s'"),
                       config_address);
        VIR_FREE(*bdf);
        goto out;
    }

1939
    VIR_DEBUG("virPCIDeviceAddress %.4x:%.2x:%.2x.%.1x",
1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956
              (*bdf)->domain,
              (*bdf)->bus,
              (*bdf)->slot,
              (*bdf)->function);

    ret = 0;

out:
    VIR_FREE(device_path);

    return ret;
}

/*
 * Returns Physical function given a virtual function
 */
int
1957 1958
virPCIGetPhysicalFunction(const char *vf_sysfs_path,
                          virPCIDeviceAddressPtr *physical_function)
1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969
{
    int ret = -1;
    char *device_link = NULL;

    VIR_DEBUG("Attempting to get SR IOV physical function for device "
              "with sysfs path '%s'", vf_sysfs_path);

    if (virBuildPath(&device_link, vf_sysfs_path, "physfn") == -1) {
        virReportOOMError();
        return ret;
    } else {
1970 1971
        ret = virPCIGetDeviceAddressFromSysfsLink(device_link,
                                                  physical_function);
1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
    }

    VIR_FREE(device_link);

    return ret;
}

/*
 * Returns virtual functions of a physical function
 */
int
1983 1984 1985
virPCIGetVirtualFunctions(const char *sysfs_path,
                          virPCIDeviceAddressPtr **virtual_functions,
                          unsigned int *num_virtual_functions)
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
{
    int ret = -1;
    DIR *dir = NULL;
    struct dirent *entry = NULL;
    char *device_link = NULL;
    char errbuf[64];

    VIR_DEBUG("Attempting to get SR IOV virtual functions for device"
              "with sysfs path '%s'", sysfs_path);

    dir = opendir(sysfs_path);
    if (dir == NULL) {
        memset(errbuf, '\0', sizeof(errbuf));
1999 2000 2001
        virReportSystemError(errno,
                             _("Failed to open dir '%s'"),
                             sysfs_path);
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
        return ret;
    }

    *virtual_functions = NULL;
    *num_virtual_functions = 0;
    while ((entry = readdir(dir))) {
        if (STRPREFIX(entry->d_name, "virtfn")) {

            if (virBuildPath(&device_link, sysfs_path, entry->d_name) == -1) {
                virReportOOMError();
                goto out;
            }

            VIR_DEBUG("Number of virtual functions: %d",
                      *num_virtual_functions);
            if (VIR_REALLOC_N(*virtual_functions,
2018
                              (*num_virtual_functions) + 1) != 0) {
2019 2020 2021 2022 2023
                virReportOOMError();
                VIR_FREE(device_link);
                goto out;
            }

2024 2025
            if (virPCIGetDeviceAddressFromSysfsLink(device_link,
                                                    &((*virtual_functions)[*num_virtual_functions])) !=
2026 2027 2028
                SRIOV_FOUND) {
                /* We should not get back SRIOV_NOT_FOUND in this
                 * case, so if we do, it's an error. */
2029
                virReportError(VIR_ERR_INTERNAL_ERROR,
2030
                               _("Failed to get SR IOV function from device "
2031
                                 "link '%s'"), device_link);
2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048
                VIR_FREE(device_link);
                goto out;
            } else {
                (*num_virtual_functions)++;
            }
            VIR_FREE(device_link);
        }
    }

    ret = 0;

out:
    if (dir)
        closedir(dir);

    return ret;
}
2049 2050 2051 2052 2053

/*
 * Returns 1 if vf device is a virtual function, 0 if not, -1 on error
 */
int
2054
virPCIIsVirtualFunction(const char *vf_sysfs_device_link)
2055 2056 2057 2058 2059
{
    char *vf_sysfs_physfn_link = NULL;
    int ret = -1;

    if (virAsprintf(&vf_sysfs_physfn_link, "%s/physfn",
2060
                    vf_sysfs_device_link) < 0) {
2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075
        virReportOOMError();
        return ret;
    }

    ret = virFileExists(vf_sysfs_physfn_link);

    VIR_FREE(vf_sysfs_physfn_link);

    return ret;
}

/*
 * Returns the sriov virtual function index of vf given its pf
 */
int
2076 2077 2078
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link,
                              const char *vf_sysfs_device_link,
                              int *vf_index)
2079 2080 2081
{
    int ret = -1, i;
    unsigned int num_virt_fns = 0;
2082 2083
    virPCIDeviceAddressPtr vf_bdf = NULL;
    virPCIDeviceAddressPtr *virt_fns = NULL;
2084

2085 2086
    if (virPCIGetDeviceAddressFromSysfsLink(vf_sysfs_device_link,
                                            &vf_bdf) < 0)
2087 2088
        return ret;

2089 2090
    if (virPCIGetVirtualFunctions(pf_sysfs_device_link, &virt_fns,
                                  &num_virt_fns) < 0) {
2091
        virReportError(VIR_ERR_INTERNAL_ERROR,
2092
                       _("Error getting physical function's '%s' "
2093
                         "virtual_functions"), pf_sysfs_device_link);
2094 2095 2096 2097
        goto out;
    }

    for (i = 0; i < num_virt_fns; i++) {
2098 2099 2100 2101 2102
        if (virPCIDeviceAddressIsEqual(vf_bdf, virt_fns[i])) {
            *vf_index = i;
            ret = 0;
            break;
        }
2103 2104 2105 2106 2107 2108
    }

out:

    /* free virtual functions */
    for (i = 0; i < num_virt_fns; i++)
2109
        VIR_FREE(virt_fns[i]);
2110

A
ajia@redhat.com 已提交
2111
    VIR_FREE(virt_fns);
2112 2113 2114 2115 2116
    VIR_FREE(vf_bdf);

    return ret;
}

2117 2118 2119 2120 2121
/*
 * Returns a path to the PCI sysfs file given the BDF of the PCI function
 */

int
2122
virPCIGetSysfsFile(char *virPCIDeviceName, char **pci_sysfs_device_link)
2123 2124
{
    if (virAsprintf(pci_sysfs_device_link, PCI_SYSFS "devices/%s",
2125
                    virPCIDeviceName) < 0) {
2126 2127 2128 2129 2130 2131 2132
        virReportOOMError();
        return -1;
    }

    return 0;
}

R
Roopa Prabhu 已提交
2133
int
2134 2135
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev,
                                char **pci_sysfs_device_link)
R
Roopa Prabhu 已提交
2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146
{
    if (virAsprintf(pci_sysfs_device_link,
                    PCI_SYSFS "devices/%04x:%02x:%02x.%x", dev->domain,
                    dev->bus, dev->slot, dev->function) < 0) {
        virReportOOMError();
        return -1;
    }

    return 0;
}

2147 2148 2149 2150
/*
 * Returns the network device name of a pci device
 */
int
2151 2152 2153 2154 2155 2156
virPCIGetNetName(char *device_link_sysfs_path, char **netname)
{
    char *pcidev_sysfs_net_path = NULL;
    int ret = -1;
    DIR *dir = NULL;
    struct dirent *entry = NULL;
2157

2158 2159 2160 2161 2162 2163 2164 2165 2166
    if (virBuildPath(&pcidev_sysfs_net_path, device_link_sysfs_path,
                     "net") == -1) {
        virReportOOMError();
        return -1;
    }

    dir = opendir(pcidev_sysfs_net_path);
    if (dir == NULL)
        goto out;
2167

2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182
    while ((entry = readdir(dir))) {
        if (STREQ(entry->d_name, ".") ||
            STREQ(entry->d_name, ".."))
            continue;

        /* Assume a single directory entry */
        *netname = strdup(entry->d_name);
        if (!*netname)
            virReportOOMError();
        else
            ret = 0;
        break;
    }

    closedir(dir);
2183 2184

out:
2185
    VIR_FREE(pcidev_sysfs_net_path);
2186

2187
    return ret;
2188
}
R
Roopa Prabhu 已提交
2189 2190

int
2191 2192
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path,
                             char **pfname, int *vf_index)
R
Roopa Prabhu 已提交
2193
{
2194
    virPCIDeviceAddressPtr pf_config_address = NULL;
R
Roopa Prabhu 已提交
2195 2196 2197
    char *pf_sysfs_device_path = NULL;
    int ret = -1;

2198
    if (virPCIGetPhysicalFunction(vf_sysfs_device_path, &pf_config_address) < 0)
R
Roopa Prabhu 已提交
2199 2200
        return ret;

2201 2202
    if (virPCIDeviceAddressGetSysfsFile(pf_config_address,
                                        &pf_sysfs_device_path) < 0) {
R
Roopa Prabhu 已提交
2203 2204 2205 2206 2207

        VIR_FREE(pf_config_address);
        return ret;
    }

2208 2209
    if (virPCIGetVirtualFunctionIndex(pf_sysfs_device_path, vf_sysfs_device_path,
                                      vf_index) < 0)
R
Roopa Prabhu 已提交
2210 2211
        goto cleanup;

2212
    ret = virPCIGetNetName(pf_sysfs_device_path, pfname);
R
Roopa Prabhu 已提交
2213 2214 2215 2216 2217 2218 2219 2220

cleanup:
    VIR_FREE(pf_config_address);
    VIR_FREE(pf_sysfs_device_path);

    return ret;
}

2221
#else
2222 2223
static const char *unsupported = N_("not supported on non-linux platforms");

2224
int
2225 2226
virPCIGetPhysicalFunction(const char *vf_sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr *physical_function ATTRIBUTE_UNUSED)
2227
{
2228
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2229 2230 2231 2232
    return -1;
}

int
2233 2234 2235
virPCIGetVirtualFunctions(const char *sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr **virtual_functions ATTRIBUTE_UNUSED,
                          unsigned int *num_virtual_functions ATTRIBUTE_UNUSED)
2236
{
2237
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2238 2239
    return -1;
}
2240 2241

int
E
Eric Blake 已提交
2242
virPCIIsVirtualFunction(const char *vf_sysfs_device_link ATTRIBUTE_UNUSED)
2243
{
2244
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2245 2246 2247 2248
    return -1;
}

int
2249 2250 2251
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link ATTRIBUTE_UNUSED,
                              const char *vf_sysfs_device_link ATTRIBUTE_UNUSED,
                              int *vf_index ATTRIBUTE_UNUSED)
2252
{
2253
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2254 2255 2256 2257
    return -1;

}

2258
int
2259 2260
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev ATTRIBUTE_UNUSED,
                                char **pci_sysfs_device_link ATTRIBUTE_UNUSED)
2261
{
2262
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2263 2264 2265
    return -1;
}

2266
int
2267
virPCIGetNetName(char *device_link_sysfs_path ATTRIBUTE_UNUSED,
2268
                 char **netname ATTRIBUTE_UNUSED)
2269
{
2270
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2271 2272
    return -1;
}
R
Roopa Prabhu 已提交
2273 2274

int
2275 2276 2277
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path ATTRIBUTE_UNUSED,
                             char **pfname ATTRIBUTE_UNUSED,
                             int *vf_index ATTRIBUTE_UNUSED)
R
Roopa Prabhu 已提交
2278
{
2279
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
R
Roopa Prabhu 已提交
2280 2281
    return -1;
}
2282
#endif /* __linux__ */