virpci.c 65.2 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 "dirname.h"
40
#include "virlog.h"
41
#include "viralloc.h"
42
#include "vircommand.h"
43
#include "virerror.h"
E
Eric Blake 已提交
44
#include "virfile.h"
45 46
#include "virstring.h"
#include "virutil.h"
47 48 49 50 51

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

52 53 54 55
#define SRIOV_FOUND 0
#define SRIOV_NOT_FOUND 1
#define SRIOV_ERROR -1

56
struct _virPCIDevice {
57 58 59 60
    unsigned int  domain;
    unsigned int  bus;
    unsigned int  slot;
    unsigned int  function;
61 62 63

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

67 68
    unsigned int  pcie_cap_pos;
    unsigned int  pci_pm_cap_pos;
69 70
    bool          has_flr;
    bool          has_pm_reset;
71
    bool          managed;
72
    const char   *stubDriver;
73 74

    /* used by reattach function */
75 76 77
    bool          unbind_from_stub;
    bool          remove_slot;
    bool          reprobe;
78 79
};

80
struct _virPCIDeviceList {
81 82
    virObjectLockable parent;

83
    unsigned int count;
84
    virPCIDevicePtr *devs;
85 86 87
};


88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104
/* 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 */
105 106 107
#define PCI_HEADER_TYPE_BRIDGE 0x1
#define PCI_HEADER_TYPE_MASK   0x7f
#define PCI_HEADER_TYPE_MULTI  0x80
108 109 110 111 112 113 114 115 116

/* 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 */
117
#define PCI_STATUS_CAP_LIST    0x10    /* Support Capability List */
118 119 120 121 122 123 124 125 126 127 128 129 130

/* 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 */
131
#define PCI_EXP_DEVCAP_FLR     (1<<28) /* Function Level Reset */
132 133 134 135 136 137 138

/* 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 */
139
#define PCI_BRIDGE_CTL_RESET   0x40    /* Secondary bus reset */
140 141 142

/* PM12 3.2.4  Power Management Control/Status (Offset = 4) */
#define PCI_PM_CTRL                4    /* PM control and status register */
143 144 145 146
#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 */
147 148 149

/* ECN_AF 6.x.1  Advanced Features Capability Structure */
#define PCI_AF_CAP              0x3     /* Advanced features capabilities */
150
#define PCI_AF_CAP_FLR         0x2     /* Function Level Reset */
151

J
Jiri Denemark 已提交
152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168
#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
169 170 171
#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
172 173
                                 PCI_EXT_CAP_ACS_UF)

174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190
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)

191
static int
192
virPCIDeviceConfigOpen(virPCIDevicePtr dev, bool fatal)
193 194 195 196
{
    int fd;

    fd = open(dev->path, O_RDWR);
197

198
    if (fd < 0) {
199 200 201 202 203 204 205 206 207
        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)));
        }
208 209
        return -1;
    }
210

211
    VIR_DEBUG("%s %s: opened %s", dev->id, dev->name, dev->path);
212
    return fd;
213 214
}

215
static void
216
virPCIDeviceConfigClose(virPCIDevicePtr dev, int cfgfd)
217
{
218 219 220 221 222
    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)));
    }
223 224
}

225

226
static int
227 228
virPCIDeviceRead(virPCIDevicePtr dev,
                 int cfgfd,
229
                 unsigned int pos,
230
                 uint8_t *buf,
231
                 unsigned int buflen)
232 233 234
{
    memset(buf, 0, buflen);

235 236
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        saferead(cfgfd, buf, buflen) != buflen) {
237
        char ebuf[1024];
238
        VIR_WARN("Failed to read from '%s' : %s", dev->path,
239 240 241 242 243 244 245
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static uint8_t
246
virPCIDeviceRead8(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
247 248
{
    uint8_t buf;
249
    virPCIDeviceRead(dev, cfgfd, pos, &buf, sizeof(buf));
250 251 252 253
    return buf;
}

static uint16_t
254
virPCIDeviceRead16(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
255 256
{
    uint8_t buf[2];
257
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
258 259 260 261
    return (buf[0] << 0) | (buf[1] << 8);
}

static uint32_t
262
virPCIDeviceRead32(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
263 264
{
    uint8_t buf[4];
265
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
266 267 268 269
    return (buf[0] << 0) | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
}

static int
270 271
virPCIDeviceWrite(virPCIDevicePtr dev,
                  int cfgfd,
272
                  unsigned int pos,
273
                  uint8_t *buf,
274
                  unsigned int buflen)
275
{
276 277
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        safewrite(cfgfd, buf, buflen) != buflen) {
278
        char ebuf[1024];
279
        VIR_WARN("Failed to write to '%s' : %s", dev->path,
280 281 282 283 284 285 286
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static void
287
virPCIDeviceWrite16(virPCIDevicePtr dev, int cfgfd, unsigned int pos, uint16_t val)
288 289
{
    uint8_t buf[2] = { (val >> 0), (val >> 8) };
290
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
291 292 293
}

static void
294
virPCIDeviceWrite32(virPCIDevicePtr dev, int cfgfd, unsigned int pos, uint32_t val)
295
{
296
    uint8_t buf[4] = { (val >> 0), (val >> 8), (val >> 16), (val >> 24) };
297
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
298 299
}

300
typedef int (*virPCIDeviceIterPredicate)(virPCIDevicePtr , virPCIDevicePtr , void *);
301 302 303 304 305 306 307

/* 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
308 309 310 311
virPCIDeviceIterDevices(virPCIDeviceIterPredicate predicate,
                        virPCIDevicePtr dev,
                        virPCIDevicePtr *matched,
                        void *data)
312 313 314
{
    DIR *dir;
    struct dirent *entry;
315
    int ret = 0;
316
    int rc;
317 318 319 320 321 322 323

    *matched = NULL;

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

    dir = opendir(PCI_SYSFS "devices");
    if (!dir) {
324
        VIR_WARN("Failed to open " PCI_SYSFS "devices");
325 326 327 328
        return -1;
    }

    while ((entry = readdir(dir))) {
329
        unsigned int domain, bus, slot, function;
330
        virPCIDevicePtr check;
331
        char *tmp;
332 333 334 335 336

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

337 338 339 340 341 342 343 344 345
        /* 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) {
346 347 348 349
            VIR_WARN("Unusual entry in " PCI_SYSFS "devices: %s", entry->d_name);
            continue;
        }

350
        check = virPCIDeviceNew(domain, bus, slot, function);
351
        if (!check) {
352 353 354
            ret = -1;
            break;
        }
355

356 357 358
        rc = predicate(dev, check, data);
        if (rc < 0) {
            /* the predicate returned an error, bail */
359
            virPCIDeviceFree(check);
360 361 362 363
            ret = -1;
            break;
        }
        else if (rc == 1) {
364 365
            VIR_DEBUG("%s %s: iter matched on %s", dev->id, dev->name, check->name);
            *matched = check;
366
            ret = 1;
367 368
            break;
        }
369

370
        virPCIDeviceFree(check);
371 372
    }
    closedir(dir);
373
    return ret;
374 375 376
}

static uint8_t
377 378 379
virPCIDeviceFindCapabilityOffset(virPCIDevicePtr dev,
                                 int cfgfd,
                                 unsigned int capability)
380 381 382 383
{
    uint16_t status;
    uint8_t pos;

384
    status = virPCIDeviceRead16(dev, cfgfd, PCI_STATUS);
385 386 387
    if (!(status & PCI_STATUS_CAP_LIST))
        return 0;

388
    pos = virPCIDeviceRead8(dev, cfgfd, PCI_CAPABILITY_LIST);
389 390 391 392 393 394 395 396 397

    /* 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) {
398
        uint8_t capid = virPCIDeviceRead8(dev, cfgfd, pos);
399 400 401 402 403 404
        if (capid == capability) {
            VIR_DEBUG("%s %s: found cap 0x%.2x at 0x%.2x",
                      dev->id, dev->name, capability, pos);
            return pos;
        }

405
        pos = virPCIDeviceRead8(dev, cfgfd, pos + 1);
406 407 408 409 410 411 412
    }

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

    return 0;
}

J
Jiri Denemark 已提交
413
static unsigned int
414 415
virPCIDeviceFindExtendedCapabilityOffset(virPCIDevicePtr dev,
                                         int cfgfd,
416
                                         unsigned int capability)
J
Jiri Denemark 已提交
417 418 419 420 421 422 423 424 425 426
{
    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) {
427
        header = virPCIDeviceRead32(dev, cfgfd, pos);
J
Jiri Denemark 已提交
428 429 430 431 432 433 434 435 436 437 438

        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;
}

439 440 441 442
/* 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
443
virPCIDeviceDetectFunctionLevelReset(virPCIDevicePtr dev, int cfgfd)
444
{
M
Mark McLoughlin 已提交
445
    uint32_t caps;
446
    uint8_t pos;
447 448
    char *path;
    int found;
449 450 451 452 453 454 455 456

    /* 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) {
457
        caps = virPCIDeviceRead32(dev, cfgfd, dev->pcie_cap_pos + PCI_EXP_DEVCAP);
458 459 460 461 462 463 464 465 466 467
        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.
     */
468
    pos = virPCIDeviceFindCapabilityOffset(dev, cfgfd, PCI_CAP_ID_AF);
469
    if (pos) {
470
        caps = virPCIDeviceRead16(dev, cfgfd, pos + PCI_AF_CAP);
471 472 473 474 475 476
        if (caps & PCI_AF_CAP_FLR) {
            VIR_DEBUG("%s %s: detected PCI FLR capability", dev->id, dev->name);
            return 1;
        }
    }

477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
    /* 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;
    }

496 497 498 499 500 501 502 503 504
    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.
 */
505
static unsigned int
506
virPCIDeviceDetectPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
507 508 509 510 511
{
    if (dev->pci_pm_cap_pos) {
        uint32_t ctl;

        /* require the NO_SOFT_RESET bit is clear */
512
        ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
513 514 515 516 517 518 519 520 521 522 523
        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;
}

524
/* Any active devices on the same domain/bus ? */
525
static int
526
virPCIDeviceSharesBusWithActive(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
527
{
528
    virPCIDeviceList *inactiveDevs = data;
529

530
    /* Different domain, different bus, or simply identical device */
531 532
    if (dev->domain != check->domain ||
        dev->bus != check->bus ||
533 534
        (dev->slot == check->slot &&
         dev->function == check->function))
535 536
        return 0;

537
    /* same bus, but inactive, i.e. about to be assigned to guest */
538
    if (inactiveDevs && virPCIDeviceListFind(inactiveDevs, check))
539
        return 0;
540

541
    return 1;
542 543
}

544 545 546
static virPCIDevicePtr
virPCIDeviceBusContainsActiveDevices(virPCIDevicePtr dev,
                                     virPCIDeviceList *inactiveDevs)
547
{
548 549 550
    virPCIDevicePtr active = NULL;
    if (virPCIDeviceIterDevices(virPCIDeviceSharesBusWithActive,
                                dev, &active, inactiveDevs) < 0)
551 552 553 554 555
        return NULL;
    return active;
}

/* Is @check the parent of @dev ? */
556
static int
557
virPCIDeviceIsParent(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
558 559 560
{
    uint16_t device_class;
    uint8_t header_type, secondary, subordinate;
561
    virPCIDevicePtr *best = data;
562 563
    int ret = 0;
    int fd;
564

565
    if (dev->domain != check->domain)
566 567
        return 0;

568
    if ((fd = virPCIDeviceConfigOpen(check, false)) < 0)
569 570
        return 0;

571
    /* Is it a bridge? */
572
    device_class = virPCIDeviceRead16(check, fd, PCI_CLASS_DEVICE);
573
    if (device_class != PCI_CLASS_BRIDGE_PCI)
574
        goto cleanup;
575 576

    /* Is it a plane? */
577
    header_type = virPCIDeviceRead8(check, fd, PCI_HEADER_TYPE);
578
    if ((header_type & PCI_HEADER_TYPE_MASK) != PCI_HEADER_TYPE_BRIDGE)
579
        goto cleanup;
580

581 582
    secondary   = virPCIDeviceRead8(check, fd, PCI_SECONDARY_BUS);
    subordinate = virPCIDeviceRead8(check, fd, PCI_SUBORDINATE_BUS);
583

584
    VIR_DEBUG("%s %s: found parent device %s", dev->id, dev->name, check->name);
585

586 587 588
    /* if the secondary bus exactly equals the device's bus, then we found
     * the direct parent.  No further work is necessary
     */
589 590 591 592
    if (dev->bus == secondary) {
        ret = 1;
        goto cleanup;
    }
593 594 595 596 597 598 599

    /* 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) {
600 601
            *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                    check->function);
602 603 604 605 606
            if (*best == NULL) {
                ret = -1;
                goto cleanup;
            }
        } else {
607 608 609 610
            /* 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
             */
611 612 613
            int bestfd;
            uint8_t best_secondary;

614
            if ((bestfd = virPCIDeviceConfigOpen(*best, false)) < 0)
615
                goto cleanup;
616 617
            best_secondary = virPCIDeviceRead8(*best, bestfd, PCI_SECONDARY_BUS);
            virPCIDeviceConfigClose(*best, bestfd);
618 619

            if (secondary > best_secondary) {
620 621 622
                virPCIDeviceFree(*best);
                *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                        check->function);
623 624 625 626
                if (*best == NULL) {
                    ret = -1;
                    goto cleanup;
                }
627 628 629 630
            }
        }
    }

631
cleanup:
632
    virPCIDeviceConfigClose(check, fd);
633
    return ret;
634 635
}

636
static int
637
virPCIDeviceGetParent(virPCIDevicePtr dev, virPCIDevicePtr *parent)
638
{
639
    virPCIDevicePtr best = NULL;
640 641 642
    int ret;

    *parent = NULL;
643
    ret = virPCIDeviceIterDevices(virPCIDeviceIsParent, dev, parent, &best);
644
    if (ret == 1)
645
        virPCIDeviceFree(best);
646 647 648
    else if (ret == 0)
        *parent = best;
    return ret;
649 650 651 652 653 654
}

/* Secondary Bus Reset is our sledgehammer - it resets all
 * devices behind a bus.
 */
static int
655 656 657
virPCIDeviceTrySecondaryBusReset(virPCIDevicePtr dev,
                                 int cfgfd,
                                 virPCIDeviceList *inactiveDevs)
658
{
659
    virPCIDevicePtr parent, conflict;
660 661 662
    uint8_t config_space[PCI_CONF_LEN];
    uint16_t ctl;
    int ret = -1;
663
    int parentfd;
664

665 666 667
    /* Refuse to do a secondary bus reset if there are other
     * devices/functions behind the bus are used by the host
     * or other guests.
668
     */
669
    if ((conflict = virPCIDeviceBusContainsActiveDevices(dev, inactiveDevs))) {
670
        virReportError(VIR_ERR_INTERNAL_ERROR,
671 672
                       _("Active %s devices on bus with %s, not doing bus reset"),
                       conflict->name, dev->name);
673 674 675 676
        return -1;
    }

    /* Find the parent bus */
677
    if (virPCIDeviceGetParent(dev, &parent) < 0)
678
        return -1;
679
    if (!parent) {
680
        virReportError(VIR_ERR_INTERNAL_ERROR,
681 682
                       _("Failed to find parent device for %s"),
                       dev->name);
683 684
        return -1;
    }
685
    if ((parentfd = virPCIDeviceConfigOpen(parent, true)) < 0)
686
        goto out;
687 688 689 690 691 692 693

    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
     */
694
    if (virPCIDeviceRead(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
695
        virReportError(VIR_ERR_INTERNAL_ERROR,
696
                       _("Failed to read PCI config space for %s"),
697
                       dev->name);
698 699 700 701 702 703
        goto out;
    }

    /* Read the control register, set the reset flag, wait 200ms,
     * unset the reset flag and wait 200ms.
     */
704
    ctl = virPCIDeviceRead16(dev, cfgfd, PCI_BRIDGE_CONTROL);
705

706 707
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL,
                        ctl | PCI_BRIDGE_CTL_RESET);
708 709 710

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

711
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL, ctl);
712 713 714

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

715
    if (virPCIDeviceWrite(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
716
        virReportError(VIR_ERR_INTERNAL_ERROR,
717 718 719 720
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        goto out;
    }
721
    ret = 0;
722

723
out:
724 725
    virPCIDeviceConfigClose(parent, parentfd);
    virPCIDeviceFree(parent);
726 727 728 729 730 731 732 733
    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
734
virPCIDeviceTryPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
735 736 737 738 739 740 741 742
{
    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. */
743
    if (virPCIDeviceRead(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
744
        virReportError(VIR_ERR_INTERNAL_ERROR,
745
                       _("Failed to read PCI config space for %s"),
746
                       dev->name);
747 748 749 750 751
        return -1;
    }

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

752
    ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
753 754
    ctl &= ~PCI_PM_CTRL_STATE_MASK;

755 756
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D3hot);
757 758 759

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

760 761
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D0);
762 763 764

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

765
    if (virPCIDeviceWrite(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
766
        virReportError(VIR_ERR_INTERNAL_ERROR,
767 768 769 770
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        return -1;
    }
771 772 773 774 775

    return 0;
}

static int
776
virPCIDeviceInit(virPCIDevicePtr dev, int cfgfd)
777
{
778 779
    int flr;

780 781 782
    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);
783
    if (flr < 0)
784
        return flr;
785 786
    dev->has_flr        = !!flr;
    dev->has_pm_reset   = !!virPCIDeviceDetectPowerManagementReset(dev, cfgfd);
787

788 789 790 791
    return 0;
}

int
792 793 794
virPCIDeviceReset(virPCIDevicePtr dev,
                  virPCIDeviceList *activeDevs,
                  virPCIDeviceList *inactiveDevs)
795 796
{
    int ret = -1;
797
    int fd;
798

799
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
800
        virReportError(VIR_ERR_INTERNAL_ERROR,
801 802 803 804
                       _("Not resetting active device %s"), dev->name);
        return -1;
    }

805
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
806 807
        return -1;

808
    if (virPCIDeviceInit(dev, fd) < 0)
809 810
        goto cleanup;

811 812 813
    /* KVM will perform FLR when starting and stopping
     * a guest, so there is no need for us to do it here.
     */
814 815 816 817
    if (dev->has_flr) {
        ret = 0;
        goto cleanup;
    }
818

819 820 821 822 823
    /* 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)
824
        ret = virPCIDeviceTryPowerManagementReset(dev, fd);
825

826
    /* Bus reset is not an option with the root bus */
827
    if (ret < 0 && dev->bus != 0)
828
        ret = virPCIDeviceTrySecondaryBusReset(dev, fd, inactiveDevs);
829

830 831
    if (ret < 0) {
        virErrorPtr err = virGetLastError();
832
        virReportError(VIR_ERR_INTERNAL_ERROR,
833 834 835 836 837
                       _("Unable to reset PCI device %s: %s"),
                       dev->name,
                       err ? err->message : _("no FLR, PM reset or bus reset available"));
    }

838
cleanup:
839
    virPCIDeviceConfigClose(dev, fd);
840 841 842
    return ret;
}

843

844
static int
845
virPCIDriverDir(char **buffer, const char *driver)
846
{
847 848 849 850 851 852 853 854
    VIR_FREE(*buffer);

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

    return 0;
855 856
}

857
static int
858
virPCIDriverFile(char **buffer, const char *driver, const char *file)
859
{
860 861 862 863 864 865 866 867
    VIR_FREE(*buffer);

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

    return 0;
868
}
869

870
static int
871
virPCIFile(char **buffer, const char *device, const char *file)
872
{
873 874 875 876 877 878 879 880
    VIR_FREE(*buffer);

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

    return 0;
881
}
882

883
static int
884
virPCIProbeStubDriver(const char *driver)
885
{
886
    char *drvpath = NULL;
887
    bool probed = false;
888 889

recheck:
890
    if (virPCIDriverDir(&drvpath, driver) == 0 && virFileExists(drvpath)) {
891
        /* driver already loaded, return */
892
        VIR_FREE(drvpath);
893
        return 0;
894 895 896
    }

    VIR_FREE(drvpath);
897 898

    if (!probed) {
899
        const char *const probecmd[] = { MODPROBE, driver, NULL };
900
        probed = true;
901
        if (virRun(probecmd, NULL) < 0) {
902
            char ebuf[1024];
903
            VIR_WARN("failed to load driver %s: %s", driver,
904
                     virStrerror(errno, ebuf, sizeof(ebuf)));
905
            return -1;
906
        }
907 908

        goto recheck;
909 910
    }

911
    return -1;
912 913
}

914 915 916 917 918 919 920
static const char *virPCIKnownStubs[] = {
    "pciback",  /* used by xen */
    "pci-stub", /* used by kvm legacy passthrough */
    "vfio-pci", /* used by VFIO device assignment */
    NULL
};

921
static int
922
virPCIDeviceUnbindFromStub(virPCIDevicePtr dev)
923 924 925 926
{
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;
927 928 929
    char *driver = NULL;
    const char **stubTest;
    bool isStub = false;
930

931 932 933 934
    /* If the device is currently bound to one of the "well known"
     * stub drivers, then unbind it, otherwise ignore it.
     */
    if (virPCIFile(&path, dev->name, "driver") < 0)
E
Eric Blake 已提交
935
        goto cleanup;
936 937 938 939 940 941 942 943 944 945 946 947 948 949
    /* path = "/sys/bus/pci/dddd:bb:ss.ff/driver" */
    if (virFileIsLink(path) != 1) {
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Invalid device %s driver file %s is not a symlink"),
                       dev->name, path);
        goto cleanup;
    }
    if (virFileResolveLink(path, &drvdir) < 0) {
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Unable to resolve device %s driver symlink %s"),
                       dev->name, path);
        goto cleanup;
    }
    /* drvdir = "/sys/bus/pci/drivers/${drivername}" */
950
    if (VIR_STRDUP(driver, last_component(drvdir)) < 0)
951
        goto cleanup;
E
Eric Blake 已提交
952

953 954 955
    if (!dev->unbind_from_stub)
        goto remove_slot;

956 957 958 959 960 961 962 963 964 965
    /* If the device isn't bound to a known stub, skip the unbind. */
    for (stubTest = virPCIKnownStubs; *stubTest != NULL; stubTest++) {
        if (STREQ(driver, *stubTest)) {
            isStub = true;
            VIR_DEBUG("Found stub driver %s", *stubTest);
            break;
        }
    }
    if (!isStub)
        goto remove_slot;
966

967
    if (virFileExists(drvdir)) {
968
        if (virPCIDriverFile(&path, driver, "unbind") < 0) {
969 970 971 972 973
            goto cleanup;
        }

        if (virFileWriteStr(path, dev->name, 0) < 0) {
            virReportSystemError(errno,
974
                                 _("Failed to unbind PCI device '%s' from %s"),
975 976 977 978
                                 dev->name, driver);
            goto cleanup;
        }
    }
979
    dev->unbind_from_stub = false;
980 981 982 983

remove_slot:
    if (!dev->remove_slot)
        goto reprobe;
984 985

    /* Xen's pciback.ko wants you to use remove_slot on the specific device */
986
    if (virPCIDriverFile(&path, driver, "remove_slot") < 0) {
987 988 989 990 991
        goto cleanup;
    }

    if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
        virReportSystemError(errno,
992
                             _("Failed to remove slot for PCI device '%s' from %s"),
993 994 995
                             dev->name, driver);
        goto cleanup;
    }
996
    dev->remove_slot = false;
997 998 999 1000 1001 1002

reprobe:
    if (!dev->reprobe) {
        result = 0;
        goto cleanup;
    }
1003 1004 1005 1006 1007 1008

    /* 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.
     */
1009
    if (virPCIDriverFile(&path, driver, "remove_id") < 0) {
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
        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:
1025
    /* do not do it again */
1026 1027 1028
    dev->unbind_from_stub = false;
    dev->remove_slot = false;
    dev->reprobe = false;
1029

1030 1031
    VIR_FREE(drvdir);
    VIR_FREE(path);
1032
    VIR_FREE(driver);
1033 1034 1035 1036

    return result;
}

1037 1038

static int
1039
virPCIDeviceBindToStub(virPCIDevicePtr dev, const char *driver)
1040
{
1041 1042 1043
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;
1044
    int reprobe = false;
1045 1046

    /* check whether the device is already bound to a driver */
1047 1048
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1049 1050 1051 1052 1053 1054 1055 1056 1057
        goto cleanup;
    }

    if (virFileExists(path)) {
        if (virFileLinkPointsTo(path, drvdir)) {
            /* The device is already bound to pci-stub */
            result = 0;
            goto cleanup;
        }
1058
        reprobe = true;
1059
    }
1060 1061 1062 1063 1064 1065 1066 1067 1068

    /* 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.
     */
1069
    if (virPCIDriverFile(&path, driver, "new_id") < 0) {
1070
        goto cleanup;
1071 1072
    }

1073
    if (virFileWriteStr(path, dev->id, 0) < 0) {
1074
        virReportSystemError(errno,
1075 1076
                             _("Failed to add PCI device ID '%s' to %s"),
                             dev->id, driver);
1077
        goto cleanup;
1078 1079
    }

1080 1081 1082
    /* check whether the device is bound to pci-stub when we write dev->id to
     * new_id.
     */
1083 1084
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1085 1086 1087 1088
        goto remove_id;
    }

    if (virFileLinkPointsTo(path, drvdir)) {
1089 1090
        dev->unbind_from_stub = true;
        dev->remove_slot = true;
1091 1092 1093
        goto remove_id;
    }

1094 1095 1096 1097 1098
    /* 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.
     */
1099
    if (virPCIFile(&path, dev->name, "driver/unbind") < 0) {
1100 1101 1102
        goto cleanup;
    }

1103 1104 1105 1106 1107 1108 1109 1110
    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;
1111 1112
    }

1113 1114
    /* If the device isn't already bound to pci-stub, try binding it now.
     */
1115 1116
    if (virPCIDriverDir(&drvdir, driver) < 0 ||
        virPCIFile(&path, dev->name, "driver") < 0) {
1117
        goto remove_id;
1118 1119
    }

1120 1121
    if (!virFileLinkPointsTo(path, drvdir)) {
        /* Xen's pciback.ko wants you to use new_slot first */
1122
        if (virPCIDriverFile(&path, driver, "new_slot") < 0) {
1123
            goto remove_id;
1124 1125
        }

1126
        if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
1127
            virReportSystemError(errno,
1128 1129
                                 _("Failed to add slot for PCI device '%s' to %s"),
                                 dev->name, driver);
1130
            goto remove_id;
1131
        }
1132
        dev->remove_slot = true;
1133

1134
        if (virPCIDriverFile(&path, driver, "bind") < 0) {
1135
            goto remove_id;
1136 1137
        }

1138
        if (virFileWriteStr(path, dev->name, 0) < 0) {
1139
            virReportSystemError(errno,
1140 1141
                                 _("Failed to bind PCI device '%s' to %s"),
                                 dev->name, driver);
1142
            goto remove_id;
1143
        }
1144
        dev->unbind_from_stub = true;
1145 1146
    }

1147
remove_id:
1148 1149 1150
    /* If 'remove_id' exists, remove the device id from pci-stub's dynamic
     * ID table so that 'drivers_probe' works below.
     */
1151
    if (virPCIDriverFile(&path, driver, "remove_id") < 0) {
E
Eric Blake 已提交
1152
        /* We do not remove PCI ID from pci-stub, and we cannot reprobe it */
1153 1154 1155 1156
        if (dev->reprobe) {
            VIR_WARN("Could not remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
1157
        dev->reprobe = false;
1158 1159 1160
        goto cleanup;
    }

1161
    if (virFileExists(path) && virFileWriteStr(path, dev->id, 0) < 0) {
1162
        virReportSystemError(errno,
1163 1164
                             _("Failed to remove PCI ID '%s' from %s"),
                             dev->id, driver);
1165

E
Eric Blake 已提交
1166
        /* remove PCI ID from pci-stub failed, and we cannot reprobe it */
1167 1168 1169 1170
        if (dev->reprobe) {
            VIR_WARN("Failed to remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
1171
        dev->reprobe = false;
1172
        goto cleanup;
1173 1174
    }

1175 1176 1177 1178 1179 1180
    result = 0;

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

1181
    if (result < 0) {
1182
        virPCIDeviceUnbindFromStub(dev);
1183 1184
    }

1185
    return result;
1186 1187 1188
}

int
1189 1190 1191 1192
virPCIDeviceDetach(virPCIDevicePtr dev,
                   virPCIDeviceList *activeDevs,
                   virPCIDeviceList *inactiveDevs,
                   const char *driver)
1193
{
1194 1195 1196
    if (!driver && dev->stubDriver)
        driver = dev->stubDriver;

1197
    if (virPCIProbeStubDriver(driver) < 0) {
1198 1199
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Failed to load PCI stub module %s"), driver);
1200 1201 1202
        return -1;
    }

1203
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
1204
        virReportError(VIR_ERR_INTERNAL_ERROR,
1205 1206 1207 1208
                       _("Not detaching active device %s"), dev->name);
        return -1;
    }

1209
    if (virPCIDeviceBindToStub(dev, driver) < 0)
1210 1211 1212
        return -1;

    /* Add the dev into list inactiveDevs */
1213 1214
    if (inactiveDevs && !virPCIDeviceListFind(inactiveDevs, dev)) {
        if (virPCIDeviceListAdd(inactiveDevs, dev) < 0)
1215 1216 1217 1218
            return -1;
    }

    return 0;
1219 1220 1221
}

int
1222 1223
virPCIDeviceReattach(virPCIDevicePtr dev,
                     virPCIDeviceListPtr activeDevs,
1224
                     virPCIDeviceListPtr inactiveDevs)
1225
{
1226
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
1227
        virReportError(VIR_ERR_INTERNAL_ERROR,
1228 1229 1230 1231
                       _("Not reattaching active device %s"), dev->name);
        return -1;
    }

1232
    if (virPCIDeviceUnbindFromStub(dev) < 0)
1233 1234 1235 1236
        return -1;

    /* Steal the dev from list inactiveDevs */
    if (inactiveDevs)
1237
        virPCIDeviceListSteal(inactiveDevs, dev);
1238 1239

    return 0;
1240 1241
}

1242 1243 1244 1245 1246
/* 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
E
Eric Blake 已提交
1247 1248
 * still holding on to the bar (denoted by the string in the matcher
 * variable), then we can wait around a bit for that to clear up.
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
 *
 * 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
E
Eric Blake 已提交
1269
 * holding on to the resource.
1270 1271
 */
int
1272
virPCIDeviceWaitForCleanup(virPCIDevicePtr dev, const char *matcher)
1273 1274 1275
{
    FILE *fp;
    char line[160];
1276
    char *tmp;
1277
    unsigned long long start, end;
1278
    unsigned int domain, bus, slot, function;
1279
    bool in_matching_device;
1280 1281 1282 1283 1284 1285 1286 1287 1288
    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
         */
1289
        VIR_DEBUG("Failed to open /proc/iomem, trying to continue anyway");
1290 1291 1292 1293
        return 0;
    }

    ret = 0;
1294
    in_matching_device = false;
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
    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))) {
1306 1307 1308 1309 1310 1311
            /* 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)
1312 1313
                continue;

1314
            if (STRPREFIX(tmp, matcher)) {
1315 1316 1317 1318 1319
                ret = 1;
                break;
            }
        }
        else {
1320
            in_matching_device = false;
1321

1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335
            /* 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')
1336 1337 1338 1339 1340
                continue;

            if (domain != dev->domain || bus != dev->bus || slot != dev->slot ||
                function != dev->function)
                continue;
1341
            in_matching_device = true;
1342 1343 1344 1345
            match_depth = strspn(line, " ");
        }
    }

E
Eric Blake 已提交
1346
    VIR_FORCE_FCLOSE(fp);
1347 1348 1349 1350

    return ret;
}

1351
static char *
1352
virPCIDeviceReadID(virPCIDevicePtr dev, const char *id_name)
1353
{
1354
    char *path = NULL;
1355 1356
    char *id_str;

1357
    if (virPCIFile(&path, dev->name, id_name) < 0) {
1358 1359
        return NULL;
    }
1360 1361

    /* ID string is '0xNNNN\n' ... i.e. 7 bytes */
1362 1363
    if (virFileReadAll(path, 7, &id_str) < 0) {
        VIR_FREE(path);
1364
        return NULL;
1365 1366 1367
    }

    VIR_FREE(path);
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380

    /* 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;
}

1381
int
1382 1383 1384 1385
virPCIGetAddrString(unsigned int domain,
                    unsigned int bus,
                    unsigned int slot,
                    unsigned int function,
1386
                    char **pciConfigAddr)
1387
{
1388
    virPCIDevicePtr dev = NULL;
1389 1390
    int ret = -1;

1391
    dev = virPCIDeviceNew(domain, bus, slot, function);
1392
    if (dev != NULL) {
1393
        if (VIR_STRDUP(*pciConfigAddr, dev->name) < 0)
1394 1395 1396 1397 1398
            goto cleanup;
        ret = 0;
    }

cleanup:
1399
    virPCIDeviceFree(dev);
1400 1401 1402
    return ret;
}

1403
virPCIDevicePtr
1404 1405 1406 1407
virPCIDeviceNew(unsigned int domain,
                unsigned int bus,
                unsigned int slot,
                unsigned int function)
1408
{
1409
    virPCIDevicePtr dev;
E
Eric Blake 已提交
1410 1411
    char *vendor = NULL;
    char *product = NULL;
1412 1413

    if (VIR_ALLOC(dev) < 0) {
1414
        virReportOOMError();
1415 1416 1417 1418 1419 1420 1421 1422
        return NULL;
    }

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

E
Eric Blake 已提交
1423 1424 1425
    if (snprintf(dev->name, sizeof(dev->name), "%.4x:%.2x:%.2x.%.1x",
                 dev->domain, dev->bus, dev->slot,
                 dev->function) >= sizeof(dev->name)) {
1426
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1427 1428
                       _("dev->name buffer overflow: %.4x:%.2x:%.2x.%.1x"),
                       dev->domain, dev->bus, dev->slot, dev->function);
E
Eric Blake 已提交
1429
        goto error;
E
Eric Blake 已提交
1430 1431 1432 1433
    }
    if (virAsprintf(&dev->path, PCI_SYSFS "devices/%s/config",
                    dev->name) < 0) {
        virReportOOMError();
E
Eric Blake 已提交
1434
        goto error;
E
Eric Blake 已提交
1435
    }
1436

1437 1438 1439 1440
    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
1441
        goto error;
1442 1443
    }

1444 1445
    vendor  = virPCIDeviceReadID(dev, "vendor");
    product = virPCIDeviceReadID(dev, "device");
1446 1447

    if (!vendor || !product) {
1448
        virReportError(VIR_ERR_INTERNAL_ERROR,
1449 1450
                       _("Failed to read product/vendor ID for %s"),
                       dev->name);
E
Eric Blake 已提交
1451
        goto error;
1452 1453 1454
    }

    /* strings contain '0x' prefix */
E
Eric Blake 已提交
1455 1456
    if (snprintf(dev->id, sizeof(dev->id), "%s %s", &vendor[2],
                 &product[2]) >= sizeof(dev->id)) {
1457
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1458 1459
                       _("dev->id buffer overflow: %s %s"),
                       &vendor[2], &product[2]);
E
Eric Blake 已提交
1460
        goto error;
E
Eric Blake 已提交
1461
    }
1462 1463 1464

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

E
Eric Blake 已提交
1465 1466 1467
cleanup:
    VIR_FREE(product);
    VIR_FREE(vendor);
1468
    return dev;
E
Eric Blake 已提交
1469 1470

error:
1471
    virPCIDeviceFree(dev);
E
Eric Blake 已提交
1472 1473
    dev = NULL;
    goto cleanup;
1474 1475 1476
}

void
1477
virPCIDeviceFree(virPCIDevicePtr dev)
1478
{
1479 1480
    if (!dev)
        return;
1481
    VIR_DEBUG("%s %s: freeing", dev->id, dev->name);
E
Eric Blake 已提交
1482
    VIR_FREE(dev->path);
1483 1484
    VIR_FREE(dev);
}
1485

1486
const char *
1487
virPCIDeviceGetName(virPCIDevicePtr dev)
1488 1489 1490 1491
{
    return dev->name;
}

1492
void virPCIDeviceSetManaged(virPCIDevicePtr dev, bool managed)
1493
{
1494
    dev->managed = managed;
1495 1496
}

1497 1498
unsigned int
virPCIDeviceGetManaged(virPCIDevicePtr dev)
1499 1500 1501 1502
{
    return dev->managed;
}

1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514
void
virPCIDeviceSetStubDriver(virPCIDevicePtr dev, const char *driver)
{
    dev->stubDriver = driver;
}

const char *
virPCIDeviceGetStubDriver(virPCIDevicePtr dev)
{
    return dev->stubDriver;
}

1515
unsigned int
1516
virPCIDeviceGetUnbindFromStub(virPCIDevicePtr dev)
1517 1518 1519 1520 1521
{
    return dev->unbind_from_stub;
}

void
1522
virPCIDeviceSetUnbindFromStub(virPCIDevicePtr dev, bool unbind)
1523
{
1524
    dev->unbind_from_stub = unbind;
1525 1526
}

1527
unsigned int
1528
virPCIDeviceGetRemoveSlot(virPCIDevicePtr dev)
1529 1530 1531 1532 1533
{
    return dev->remove_slot;
}

void
1534
virPCIDeviceSetRemoveSlot(virPCIDevicePtr dev, bool remove_slot)
1535
{
1536
    dev->remove_slot = remove_slot;
1537 1538
}

1539
unsigned int
1540
virPCIDeviceGetReprobe(virPCIDevicePtr dev)
1541 1542 1543 1544 1545
{
    return dev->reprobe;
}

void
1546
virPCIDeviceSetReprobe(virPCIDevicePtr dev, bool reprobe)
1547
{
1548
    dev->reprobe = reprobe;
1549 1550
}

1551
void
1552
virPCIDeviceSetUsedBy(virPCIDevicePtr dev, const char *name)
1553 1554 1555 1556 1557
{
    dev->used_by = name;
}

const char *
1558
virPCIDeviceGetUsedBy(virPCIDevicePtr dev)
1559 1560 1561 1562
{
    return dev->used_by;
}

1563
void virPCIDeviceReattachInit(virPCIDevicePtr pci)
1564
{
1565 1566 1567
    pci->unbind_from_stub = true;
    pci->remove_slot = true;
    pci->reprobe = true;
1568 1569 1570
}


1571 1572
virPCIDeviceListPtr
virPCIDeviceListNew(void)
1573
{
1574
    virPCIDeviceListPtr list;
1575

1576 1577 1578 1579
    if (virPCIInitialize() < 0)
        return NULL;

    if (!(list = virObjectLockableNew(virPCIDeviceListClass)))
1580 1581 1582 1583 1584
        return NULL;

    return list;
}

1585 1586
static void
virPCIDeviceListDispose(void *obj)
1587
{
1588
    virPCIDeviceListPtr list = obj;
1589 1590 1591
    int i;

    for (i = 0; i < list->count; i++) {
1592
        virPCIDeviceFree(list->devs[i]);
1593 1594 1595 1596 1597 1598 1599 1600
        list->devs[i] = NULL;
    }

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

int
1601 1602
virPCIDeviceListAdd(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1603
{
1604
    if (virPCIDeviceListFind(list, dev)) {
1605
        virReportError(VIR_ERR_INTERNAL_ERROR,
1606 1607 1608 1609 1610
                       _("Device %s is already in use"), dev->name);
        return -1;
    }

    if (VIR_REALLOC_N(list->devs, list->count+1) < 0) {
1611
        virReportOOMError();
1612 1613 1614 1615 1616 1617 1618 1619
        return -1;
    }

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

    return 0;
}

1620 1621 1622
virPCIDevicePtr
virPCIDeviceListGet(virPCIDeviceListPtr list,
                    int idx)
1623 1624 1625 1626 1627 1628 1629 1630 1631 1632
{
    if (idx >= list->count)
        return NULL;
    if (idx < 0)
        return NULL;

    return list->devs[idx];
}

int
1633
virPCIDeviceListCount(virPCIDeviceListPtr list)
1634
{
1635 1636 1637
    return list->count;
}

1638 1639 1640
virPCIDevicePtr
virPCIDeviceListStealIndex(virPCIDeviceListPtr list,
                           int idx)
1641
{
1642
    virPCIDevicePtr ret;
1643

1644 1645
    if (idx < 0 || idx >= list->count)
        return NULL;
1646

1647
    ret = list->devs[idx];
1648

1649 1650 1651 1652 1653
    if (idx != --list->count) {
        memmove(&list->devs[idx],
                &list->devs[idx + 1],
                sizeof(*list->devs) * (list->count - idx));
    }
1654

1655 1656
    if (VIR_REALLOC_N(list->devs, list->count) < 0) {
        ; /* not fatal */
1657
    }
1658

1659 1660 1661
    return ret;
}

1662 1663 1664
virPCIDevicePtr
virPCIDeviceListSteal(virPCIDeviceListPtr list,
                      virPCIDevicePtr dev)
1665
{
1666
    return virPCIDeviceListStealIndex(list, virPCIDeviceListFindIndex(list, dev));
1667 1668
}

1669
void
1670 1671
virPCIDeviceListDel(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1672
{
1673
    virPCIDevicePtr ret = virPCIDeviceListSteal(list, dev);
1674
    if (ret)
1675
        virPCIDeviceFree(ret);
1676 1677
}

1678
int
1679
virPCIDeviceListFindIndex(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1680 1681 1682 1683 1684 1685 1686 1687
{
    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)
1688 1689 1690 1691
            return i;
    return -1;
}

1692 1693
virPCIDevicePtr
virPCIDeviceListFind(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1694 1695 1696
{
    int i;

1697
    if ((i = virPCIDeviceListFindIndex(list, dev)) >= 0)
1698 1699 1700
        return list->devs[i];
    else
        return NULL;
1701
}
1702 1703


1704 1705 1706
int virPCIDeviceFileIterate(virPCIDevicePtr dev,
                            virPCIDeviceFileActor actor,
                            void *opaque)
1707 1708 1709 1710 1711 1712 1713 1714 1715
{
    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) {
1716
        virReportOOMError();
1717 1718 1719 1720
        goto cleanup;
    }

    if (!(dir = opendir(pcidir))) {
1721
        virReportSystemError(errno,
1722 1723 1724 1725 1726 1727
                             _("cannot open %s"), pcidir);
        goto cleanup;
    }

    while ((ent = readdir(dir)) != NULL) {
        /* Device assignment requires:
A
Alex Williamson 已提交
1728 1729
         *   $PCIDIR/config, $PCIDIR/resource, $PCIDIR/resourceNNN,
         *   $PCIDIR/rom, $PCIDIR/reset
1730 1731 1732
         */
        if (STREQ(ent->d_name, "config") ||
            STRPREFIX(ent->d_name, "resource") ||
A
Alex Williamson 已提交
1733 1734
            STREQ(ent->d_name, "rom") ||
            STREQ(ent->d_name, "reset")) {
1735
            if (virAsprintf(&file, "%s/%s", pcidir, ent->d_name) < 0) {
1736
                virReportOOMError();
1737 1738
                goto cleanup;
            }
1739
            if ((actor)(dev, file, opaque) < 0)
1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754
                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
1755

1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790
/* virPCIDeviceGetVFIOGroupDev - return the name of the device used to
 * control this PCI device's group (e.g. "/dev/vfio/15")
 */
char *
virPCIDeviceGetVFIOGroupDev(virPCIDevicePtr dev)
{
    char *devPath = NULL;
    char *groupPath = NULL;
    char *groupDev = NULL;

    if (virPCIFile(&devPath, dev->name, "iommu_group") < 0)
        goto cleanup;
    if (virFileIsLink(devPath) != 1) {
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Invalid device %s iommu_group file %s is not a symlink"),
                       dev->name, devPath);
        goto cleanup;
    }
    if (virFileResolveLink(devPath, &groupPath) < 0) {
        virReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Unable to resolve device %s iommu_group symlink %s"),
                       dev->name, devPath);
        goto cleanup;
    }
    if (virAsprintf(&groupDev, "/dev/vfio/%s",
                    last_component(groupPath)) < 0) {
        virReportOOMError();
        goto cleanup;
    }
cleanup:
    VIR_FREE(devPath);
    VIR_FREE(groupPath);
    return groupDev;
}

J
Jiri Denemark 已提交
1791
static int
1792
virPCIDeviceDownstreamLacksACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1793 1794 1795 1796
{
    uint16_t flags;
    uint16_t ctrl;
    unsigned int pos;
1797 1798
    int fd;
    int ret = 0;
J
Jiri Denemark 已提交
1799

1800
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
J
Jiri Denemark 已提交
1801 1802
        return -1;

1803
    if (virPCIDeviceInit(dev, fd) < 0) {
1804 1805 1806 1807
        ret = -1;
        goto cleanup;
    }

J
Jiri Denemark 已提交
1808
    pos = dev->pcie_cap_pos;
1809
    if (!pos || virPCIDeviceRead16(dev, fd, PCI_CLASS_DEVICE) != PCI_CLASS_BRIDGE_PCI)
1810
        goto cleanup;
J
Jiri Denemark 已提交
1811

1812
    flags = virPCIDeviceRead16(dev, fd, pos + PCI_EXP_FLAGS);
J
Jiri Denemark 已提交
1813
    if (((flags & PCI_EXP_FLAGS_TYPE) >> 4) != PCI_EXP_TYPE_DOWNSTREAM)
1814
        goto cleanup;
J
Jiri Denemark 已提交
1815

1816
    pos = virPCIDeviceFindExtendedCapabilityOffset(dev, fd, PCI_EXT_CAP_ID_ACS);
J
Jiri Denemark 已提交
1817 1818
    if (!pos) {
        VIR_DEBUG("%s %s: downstream port lacks ACS", dev->id, dev->name);
1819 1820
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1821 1822
    }

1823
    ctrl = virPCIDeviceRead16(dev, fd, pos + PCI_EXT_ACS_CTRL);
J
Jiri Denemark 已提交
1824 1825 1826
    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);
1827 1828
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1829 1830
    }

1831
cleanup:
1832
    virPCIDeviceConfigClose(dev, fd);
1833
    return ret;
J
Jiri Denemark 已提交
1834 1835 1836
}

static int
1837
virPCIDeviceIsBehindSwitchLackingACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1838
{
1839
    virPCIDevicePtr parent;
J
Jiri Denemark 已提交
1840

1841
    if (virPCIDeviceGetParent(dev, &parent) < 0)
1842
        return -1;
1843 1844 1845 1846 1847 1848 1849 1850
    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 {
1851
            virReportError(VIR_ERR_INTERNAL_ERROR,
1852 1853 1854 1855
                           _("Failed to find parent device for %s"),
                           dev->name);
            return -1;
        }
J
Jiri Denemark 已提交
1856 1857 1858 1859 1860 1861 1862
    }

    /* 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 {
1863
        virPCIDevicePtr tmp;
J
Jiri Denemark 已提交
1864
        int acs;
1865
        int ret;
J
Jiri Denemark 已提交
1866

1867
        acs = virPCIDeviceDownstreamLacksACS(parent);
J
Jiri Denemark 已提交
1868 1869

        if (acs) {
1870
            virPCIDeviceFree(parent);
J
Jiri Denemark 已提交
1871 1872 1873 1874 1875 1876 1877
            if (acs < 0)
                return -1;
            else
                return 1;
        }

        tmp = parent;
1878 1879
        ret = virPCIDeviceGetParent(parent, &parent);
        virPCIDeviceFree(tmp);
1880 1881
        if (ret < 0)
            return -1;
J
Jiri Denemark 已提交
1882 1883 1884 1885 1886
    } while (parent);

    return 0;
}

1887 1888
int virPCIDeviceIsAssignable(virPCIDevicePtr dev,
                             int strict_acs_check)
J
Jiri Denemark 已提交
1889 1890 1891 1892 1893 1894 1895 1896
{
    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.
     */

1897
    ret = virPCIDeviceIsBehindSwitchLackingACS(dev);
J
Jiri Denemark 已提交
1898 1899 1900 1901 1902 1903 1904 1905
    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 {
1906
            virReportError(VIR_ERR_INTERNAL_ERROR,
J
Jiri Denemark 已提交
1907 1908 1909 1910 1911 1912 1913 1914 1915
                           _("Device %s is behind a switch lacking ACS and "
                             "cannot be assigned"),
                           dev->name);
            return 0;
        }
    }

    return 1;
}
1916

1917 1918
#ifdef __linux__

1919
/*
1920
 * returns true if equal
1921
 */
1922
static bool
1923 1924
virPCIDeviceAddressIsEqual(virPCIDeviceAddressPtr bdf1,
                           virPCIDeviceAddressPtr bdf2)
1925 1926 1927 1928 1929 1930 1931
{
    return ((bdf1->domain == bdf2->domain) &&
            (bdf1->bus == bdf2->bus) &&
            (bdf1->slot == bdf2->slot) &&
            (bdf1->function == bdf2->function));
}

1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950
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
1951 1952
virPCIParseDeviceAddress(char *address,
                         virPCIDeviceAddressPtr bdf)
1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983
{
    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
1984 1985
virPCIGetDeviceAddressFromSysfsLink(const char *device_link,
                                    virPCIDeviceAddressPtr *bdf)
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
{
    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;
    }

2000
    device_path = canonicalize_file_name(device_link);
2001 2002
    if (device_path == NULL) {
        memset(errbuf, '\0', sizeof(errbuf));
2003 2004 2005
        virReportSystemError(errno,
                             _("Failed to resolve device link '%s'"),
                             device_link);
2006 2007 2008
        return ret;
    }

2009
    config_address = last_component(device_path);
2010 2011 2012 2013 2014
    if (VIR_ALLOC(*bdf) != 0) {
        virReportOOMError();
        goto out;
    }

2015
    if (virPCIParseDeviceAddress(config_address, *bdf) != 0) {
2016
        virReportError(VIR_ERR_INTERNAL_ERROR,
2017 2018 2019 2020 2021 2022
                       _("Failed to parse PCI config address '%s'"),
                       config_address);
        VIR_FREE(*bdf);
        goto out;
    }

2023
    VIR_DEBUG("virPCIDeviceAddress %.4x:%.2x:%.2x.%.1x",
2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040
              (*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
2041 2042
virPCIGetPhysicalFunction(const char *vf_sysfs_path,
                          virPCIDeviceAddressPtr *physical_function)
2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053
{
    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 {
2054 2055
        ret = virPCIGetDeviceAddressFromSysfsLink(device_link,
                                                  physical_function);
2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066
    }

    VIR_FREE(device_link);

    return ret;
}

/*
 * Returns virtual functions of a physical function
 */
int
2067 2068 2069
virPCIGetVirtualFunctions(const char *sysfs_path,
                          virPCIDeviceAddressPtr **virtual_functions,
                          unsigned int *num_virtual_functions)
2070 2071
{
    int ret = -1;
2072
    int i;
2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083
    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));
2084 2085 2086
        virReportSystemError(errno,
                             _("Failed to open dir '%s'"),
                             sysfs_path);
2087 2088 2089 2090 2091 2092 2093
        return ret;
    }

    *virtual_functions = NULL;
    *num_virtual_functions = 0;
    while ((entry = readdir(dir))) {
        if (STRPREFIX(entry->d_name, "virtfn")) {
2094
            virPCIDeviceAddress *config_addr = NULL;
2095 2096 2097

            if (virBuildPath(&device_link, sysfs_path, entry->d_name) == -1) {
                virReportOOMError();
2098
                goto error;
2099 2100 2101 2102 2103
            }

            VIR_DEBUG("Number of virtual functions: %d",
                      *num_virtual_functions);

2104
            if (virPCIGetDeviceAddressFromSysfsLink(device_link,
2105
                                                    &config_addr) !=
2106
                SRIOV_FOUND) {
2107 2108
                VIR_WARN("Failed to get SRIOV function from device "
                         "link '%s'", device_link);
2109
                VIR_FREE(device_link);
2110
                continue;
2111
            }
2112

2113 2114
            if (VIR_REALLOC_N(*virtual_functions,
                              *num_virtual_functions + 1) < 0) {
2115 2116 2117 2118 2119 2120 2121
                virReportOOMError();
                VIR_FREE(config_addr);
                goto error;
            }

            (*virtual_functions)[*num_virtual_functions] = config_addr;
            (*num_virtual_functions)++;
2122 2123 2124 2125 2126 2127
            VIR_FREE(device_link);
        }
    }

    ret = 0;

2128 2129
cleanup:
    VIR_FREE(device_link);
2130 2131 2132
    if (dir)
        closedir(dir);
    return ret;
2133 2134 2135 2136 2137 2138 2139 2140

error:
    if (*virtual_functions) {
        for (i = 0; i < *num_virtual_functions; i++)
            VIR_FREE((*virtual_functions)[i]);
        VIR_FREE(*virtual_functions);
    }
    goto cleanup;
2141
}
2142 2143 2144 2145 2146

/*
 * Returns 1 if vf device is a virtual function, 0 if not, -1 on error
 */
int
2147
virPCIIsVirtualFunction(const char *vf_sysfs_device_link)
2148 2149 2150 2151 2152
{
    char *vf_sysfs_physfn_link = NULL;
    int ret = -1;

    if (virAsprintf(&vf_sysfs_physfn_link, "%s/physfn",
2153
                    vf_sysfs_device_link) < 0) {
2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168
        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
2169 2170 2171
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link,
                              const char *vf_sysfs_device_link,
                              int *vf_index)
2172 2173 2174
{
    int ret = -1, i;
    unsigned int num_virt_fns = 0;
2175 2176
    virPCIDeviceAddressPtr vf_bdf = NULL;
    virPCIDeviceAddressPtr *virt_fns = NULL;
2177

2178 2179
    if (virPCIGetDeviceAddressFromSysfsLink(vf_sysfs_device_link,
                                            &vf_bdf) < 0)
2180 2181
        return ret;

2182 2183
    if (virPCIGetVirtualFunctions(pf_sysfs_device_link, &virt_fns,
                                  &num_virt_fns) < 0) {
2184
        virReportError(VIR_ERR_INTERNAL_ERROR,
2185
                       _("Error getting physical function's '%s' "
2186
                         "virtual_functions"), pf_sysfs_device_link);
2187 2188 2189 2190
        goto out;
    }

    for (i = 0; i < num_virt_fns; i++) {
2191 2192 2193 2194 2195
        if (virPCIDeviceAddressIsEqual(vf_bdf, virt_fns[i])) {
            *vf_index = i;
            ret = 0;
            break;
        }
2196 2197 2198 2199 2200 2201
    }

out:

    /* free virtual functions */
    for (i = 0; i < num_virt_fns; i++)
2202
        VIR_FREE(virt_fns[i]);
2203

A
ajia@redhat.com 已提交
2204
    VIR_FREE(virt_fns);
2205 2206 2207 2208 2209
    VIR_FREE(vf_bdf);

    return ret;
}

2210 2211 2212 2213 2214
/*
 * Returns a path to the PCI sysfs file given the BDF of the PCI function
 */

int
2215
virPCIGetSysfsFile(char *virPCIDeviceName, char **pci_sysfs_device_link)
2216 2217
{
    if (virAsprintf(pci_sysfs_device_link, PCI_SYSFS "devices/%s",
2218
                    virPCIDeviceName) < 0) {
2219 2220 2221 2222 2223 2224 2225
        virReportOOMError();
        return -1;
    }

    return 0;
}

R
Roopa Prabhu 已提交
2226
int
2227 2228
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev,
                                char **pci_sysfs_device_link)
R
Roopa Prabhu 已提交
2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239
{
    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;
}

2240 2241 2242 2243
/*
 * Returns the network device name of a pci device
 */
int
2244 2245 2246 2247 2248 2249
virPCIGetNetName(char *device_link_sysfs_path, char **netname)
{
    char *pcidev_sysfs_net_path = NULL;
    int ret = -1;
    DIR *dir = NULL;
    struct dirent *entry = NULL;
2250

2251 2252 2253 2254 2255 2256 2257 2258 2259
    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;
2260

2261 2262 2263 2264 2265 2266
    while ((entry = readdir(dir))) {
        if (STREQ(entry->d_name, ".") ||
            STREQ(entry->d_name, ".."))
            continue;

        /* Assume a single directory entry */
2267
        if (VIR_STRDUP(*netname, entry->d_name) > 0)
2268 2269 2270 2271 2272
            ret = 0;
        break;
    }

    closedir(dir);
2273 2274

out:
2275
    VIR_FREE(pcidev_sysfs_net_path);
2276

2277
    return ret;
2278
}
R
Roopa Prabhu 已提交
2279 2280

int
2281 2282
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path,
                             char **pfname, int *vf_index)
R
Roopa Prabhu 已提交
2283
{
2284
    virPCIDeviceAddressPtr pf_config_address = NULL;
R
Roopa Prabhu 已提交
2285 2286 2287
    char *pf_sysfs_device_path = NULL;
    int ret = -1;

2288
    if (virPCIGetPhysicalFunction(vf_sysfs_device_path, &pf_config_address) < 0)
R
Roopa Prabhu 已提交
2289 2290
        return ret;

2291 2292
    if (virPCIDeviceAddressGetSysfsFile(pf_config_address,
                                        &pf_sysfs_device_path) < 0) {
R
Roopa Prabhu 已提交
2293 2294 2295 2296 2297

        VIR_FREE(pf_config_address);
        return ret;
    }

2298 2299
    if (virPCIGetVirtualFunctionIndex(pf_sysfs_device_path, vf_sysfs_device_path,
                                      vf_index) < 0)
R
Roopa Prabhu 已提交
2300 2301
        goto cleanup;

2302
    ret = virPCIGetNetName(pf_sysfs_device_path, pfname);
R
Roopa Prabhu 已提交
2303 2304 2305 2306 2307 2308 2309 2310

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

    return ret;
}

2311
#else
2312 2313
static const char *unsupported = N_("not supported on non-linux platforms");

2314
int
2315 2316
virPCIGetPhysicalFunction(const char *vf_sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr *physical_function ATTRIBUTE_UNUSED)
2317
{
2318
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2319 2320 2321 2322
    return -1;
}

int
2323 2324 2325
virPCIGetVirtualFunctions(const char *sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr **virtual_functions ATTRIBUTE_UNUSED,
                          unsigned int *num_virtual_functions ATTRIBUTE_UNUSED)
2326
{
2327
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2328 2329
    return -1;
}
2330 2331

int
E
Eric Blake 已提交
2332
virPCIIsVirtualFunction(const char *vf_sysfs_device_link ATTRIBUTE_UNUSED)
2333
{
2334
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2335 2336 2337 2338
    return -1;
}

int
2339 2340 2341
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link ATTRIBUTE_UNUSED,
                              const char *vf_sysfs_device_link ATTRIBUTE_UNUSED,
                              int *vf_index ATTRIBUTE_UNUSED)
2342
{
2343
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2344 2345 2346 2347
    return -1;

}

2348
int
2349 2350
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev ATTRIBUTE_UNUSED,
                                char **pci_sysfs_device_link ATTRIBUTE_UNUSED)
2351
{
2352
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2353 2354 2355
    return -1;
}

2356
int
2357
virPCIGetNetName(char *device_link_sysfs_path ATTRIBUTE_UNUSED,
2358
                 char **netname ATTRIBUTE_UNUSED)
2359
{
2360
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2361 2362
    return -1;
}
R
Roopa Prabhu 已提交
2363 2364

int
2365 2366 2367
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path ATTRIBUTE_UNUSED,
                             char **pfname ATTRIBUTE_UNUSED,
                             int *vf_index ATTRIBUTE_UNUSED)
R
Roopa Prabhu 已提交
2368
{
2369
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
R
Roopa Prabhu 已提交
2370 2371
    return -1;
}
2372
#endif /* __linux__ */