virpci.c 65.3 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 47 48 49

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

50 51 52 53
#define SRIOV_FOUND 0
#define SRIOV_NOT_FOUND 1
#define SRIOV_ERROR -1

54
struct _virPCIDevice {
55 56 57 58
    unsigned int  domain;
    unsigned int  bus;
    unsigned int  slot;
    unsigned int  function;
59 60 61

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

65 66
    unsigned int  pcie_cap_pos;
    unsigned int  pci_pm_cap_pos;
67 68
    bool          has_flr;
    bool          has_pm_reset;
69
    bool          managed;
70
    const char   *stubDriver;
71 72

    /* used by reattach function */
73 74 75
    bool          unbind_from_stub;
    bool          remove_slot;
    bool          reprobe;
76 77
};

78
struct _virPCIDeviceList {
79 80
    virObjectLockable parent;

81
    unsigned int count;
82
    virPCIDevicePtr *devs;
83 84 85
};


86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102
/* 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 */
103 104 105
#define PCI_HEADER_TYPE_BRIDGE 0x1
#define PCI_HEADER_TYPE_MASK   0x7f
#define PCI_HEADER_TYPE_MULTI  0x80
106 107 108 109 110 111 112 113 114

/* 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 */
115
#define PCI_STATUS_CAP_LIST    0x10    /* Support Capability List */
116 117 118 119 120 121 122 123 124 125 126 127 128

/* 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 */
129
#define PCI_EXP_DEVCAP_FLR     (1<<28) /* Function Level Reset */
130 131 132 133 134 135 136

/* 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 */
137
#define PCI_BRIDGE_CTL_RESET   0x40    /* Secondary bus reset */
138 139 140

/* PM12 3.2.4  Power Management Control/Status (Offset = 4) */
#define PCI_PM_CTRL                4    /* PM control and status register */
141 142 143 144
#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 */
145 146 147

/* ECN_AF 6.x.1  Advanced Features Capability Structure */
#define PCI_AF_CAP              0x3     /* Advanced features capabilities */
148
#define PCI_AF_CAP_FLR         0x2     /* Function Level Reset */
149

J
Jiri Denemark 已提交
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
#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
167 168 169
#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
170 171
                                 PCI_EXT_CAP_ACS_UF)

172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188
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)

189
static int
190
virPCIDeviceConfigOpen(virPCIDevicePtr dev, bool fatal)
191 192 193 194
{
    int fd;

    fd = open(dev->path, O_RDWR);
195

196
    if (fd < 0) {
197 198 199 200 201 202 203 204 205
        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)));
        }
206 207
        return -1;
    }
208

209
    VIR_DEBUG("%s %s: opened %s", dev->id, dev->name, dev->path);
210
    return fd;
211 212
}

213
static void
214
virPCIDeviceConfigClose(virPCIDevicePtr dev, int cfgfd)
215
{
216 217 218 219 220
    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)));
    }
221 222
}

223

224
static int
225 226
virPCIDeviceRead(virPCIDevicePtr dev,
                 int cfgfd,
227
                 unsigned int pos,
228
                 uint8_t *buf,
229
                 unsigned int buflen)
230 231 232
{
    memset(buf, 0, buflen);

233 234
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        saferead(cfgfd, buf, buflen) != buflen) {
235
        char ebuf[1024];
236
        VIR_WARN("Failed to read from '%s' : %s", dev->path,
237 238 239 240 241 242 243
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static uint8_t
244
virPCIDeviceRead8(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
245 246
{
    uint8_t buf;
247
    virPCIDeviceRead(dev, cfgfd, pos, &buf, sizeof(buf));
248 249 250 251
    return buf;
}

static uint16_t
252
virPCIDeviceRead16(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
253 254
{
    uint8_t buf[2];
255
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
256 257 258 259
    return (buf[0] << 0) | (buf[1] << 8);
}

static uint32_t
260
virPCIDeviceRead32(virPCIDevicePtr dev, int cfgfd, unsigned int pos)
261 262
{
    uint8_t buf[4];
263
    virPCIDeviceRead(dev, cfgfd, pos, &buf[0], sizeof(buf));
264 265 266 267
    return (buf[0] << 0) | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
}

static int
268 269
virPCIDeviceWrite(virPCIDevicePtr dev,
                  int cfgfd,
270
                  unsigned int pos,
271
                  uint8_t *buf,
272
                  unsigned int buflen)
273
{
274 275
    if (lseek(cfgfd, pos, SEEK_SET) != pos ||
        safewrite(cfgfd, buf, buflen) != buflen) {
276
        char ebuf[1024];
277
        VIR_WARN("Failed to write to '%s' : %s", dev->path,
278 279 280 281 282 283 284
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static void
285
virPCIDeviceWrite16(virPCIDevicePtr dev, int cfgfd, unsigned int pos, uint16_t val)
286 287
{
    uint8_t buf[2] = { (val >> 0), (val >> 8) };
288
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
289 290 291
}

static void
292
virPCIDeviceWrite32(virPCIDevicePtr dev, int cfgfd, unsigned int pos, uint32_t val)
293
{
294
    uint8_t buf[4] = { (val >> 0), (val >> 8), (val >> 16), (val >> 24) };
295
    virPCIDeviceWrite(dev, cfgfd, pos, &buf[0], sizeof(buf));
296 297
}

298
typedef int (*virPCIDeviceIterPredicate)(virPCIDevicePtr , virPCIDevicePtr , void *);
299 300 301 302 303 304 305

/* 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
306 307 308 309
virPCIDeviceIterDevices(virPCIDeviceIterPredicate predicate,
                        virPCIDevicePtr dev,
                        virPCIDevicePtr *matched,
                        void *data)
310 311 312
{
    DIR *dir;
    struct dirent *entry;
313
    int ret = 0;
314
    int rc;
315 316 317 318 319 320 321

    *matched = NULL;

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

    dir = opendir(PCI_SYSFS "devices");
    if (!dir) {
322
        VIR_WARN("Failed to open " PCI_SYSFS "devices");
323 324 325 326
        return -1;
    }

    while ((entry = readdir(dir))) {
327
        unsigned int domain, bus, slot, function;
328
        virPCIDevicePtr check;
329
        char *tmp;
330 331 332 333 334

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

335 336 337 338 339 340 341 342 343
        /* 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) {
344 345 346 347
            VIR_WARN("Unusual entry in " PCI_SYSFS "devices: %s", entry->d_name);
            continue;
        }

348
        check = virPCIDeviceNew(domain, bus, slot, function);
349
        if (!check) {
350 351 352
            ret = -1;
            break;
        }
353

354 355 356
        rc = predicate(dev, check, data);
        if (rc < 0) {
            /* the predicate returned an error, bail */
357
            virPCIDeviceFree(check);
358 359 360 361
            ret = -1;
            break;
        }
        else if (rc == 1) {
362 363
            VIR_DEBUG("%s %s: iter matched on %s", dev->id, dev->name, check->name);
            *matched = check;
364
            ret = 1;
365 366
            break;
        }
367

368
        virPCIDeviceFree(check);
369 370
    }
    closedir(dir);
371
    return ret;
372 373 374
}

static uint8_t
375 376 377
virPCIDeviceFindCapabilityOffset(virPCIDevicePtr dev,
                                 int cfgfd,
                                 unsigned int capability)
378 379 380 381
{
    uint16_t status;
    uint8_t pos;

382
    status = virPCIDeviceRead16(dev, cfgfd, PCI_STATUS);
383 384 385
    if (!(status & PCI_STATUS_CAP_LIST))
        return 0;

386
    pos = virPCIDeviceRead8(dev, cfgfd, PCI_CAPABILITY_LIST);
387 388 389 390 391 392 393 394 395

    /* 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) {
396
        uint8_t capid = virPCIDeviceRead8(dev, cfgfd, pos);
397 398 399 400 401 402
        if (capid == capability) {
            VIR_DEBUG("%s %s: found cap 0x%.2x at 0x%.2x",
                      dev->id, dev->name, capability, pos);
            return pos;
        }

403
        pos = virPCIDeviceRead8(dev, cfgfd, pos + 1);
404 405 406 407 408 409 410
    }

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

    return 0;
}

J
Jiri Denemark 已提交
411
static unsigned int
412 413
virPCIDeviceFindExtendedCapabilityOffset(virPCIDevicePtr dev,
                                         int cfgfd,
414
                                         unsigned int capability)
J
Jiri Denemark 已提交
415 416 417 418 419 420 421 422 423 424
{
    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) {
425
        header = virPCIDeviceRead32(dev, cfgfd, pos);
J
Jiri Denemark 已提交
426 427 428 429 430 431 432 433 434 435 436

        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;
}

437 438 439 440
/* 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
441
virPCIDeviceDetectFunctionLevelReset(virPCIDevicePtr dev, int cfgfd)
442
{
M
Mark McLoughlin 已提交
443
    uint32_t caps;
444
    uint8_t pos;
445 446
    char *path;
    int found;
447 448 449 450 451 452 453 454

    /* 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) {
455
        caps = virPCIDeviceRead32(dev, cfgfd, dev->pcie_cap_pos + PCI_EXP_DEVCAP);
456 457 458 459 460 461 462 463 464 465
        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.
     */
466
    pos = virPCIDeviceFindCapabilityOffset(dev, cfgfd, PCI_CAP_ID_AF);
467
    if (pos) {
468
        caps = virPCIDeviceRead16(dev, cfgfd, pos + PCI_AF_CAP);
469 470 471 472 473 474
        if (caps & PCI_AF_CAP_FLR) {
            VIR_DEBUG("%s %s: detected PCI FLR capability", dev->id, dev->name);
            return 1;
        }
    }

475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
    /* 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;
    }

494 495 496 497 498 499 500 501 502
    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.
 */
503
static unsigned int
504
virPCIDeviceDetectPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
505 506 507 508 509
{
    if (dev->pci_pm_cap_pos) {
        uint32_t ctl;

        /* require the NO_SOFT_RESET bit is clear */
510
        ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
511 512 513 514 515 516 517 518 519 520 521
        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;
}

522
/* Any active devices on the same domain/bus ? */
523
static int
524
virPCIDeviceSharesBusWithActive(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
525
{
526
    virPCIDeviceList *inactiveDevs = data;
527

528
    /* Different domain, different bus, or simply identical device */
529 530
    if (dev->domain != check->domain ||
        dev->bus != check->bus ||
531 532
        (dev->slot == check->slot &&
         dev->function == check->function))
533 534
        return 0;

535
    /* same bus, but inactive, i.e. about to be assigned to guest */
536
    if (inactiveDevs && virPCIDeviceListFind(inactiveDevs, check))
537
        return 0;
538

539
    return 1;
540 541
}

542 543 544
static virPCIDevicePtr
virPCIDeviceBusContainsActiveDevices(virPCIDevicePtr dev,
                                     virPCIDeviceList *inactiveDevs)
545
{
546 547 548
    virPCIDevicePtr active = NULL;
    if (virPCIDeviceIterDevices(virPCIDeviceSharesBusWithActive,
                                dev, &active, inactiveDevs) < 0)
549 550 551 552 553
        return NULL;
    return active;
}

/* Is @check the parent of @dev ? */
554
static int
555
virPCIDeviceIsParent(virPCIDevicePtr dev, virPCIDevicePtr check, void *data)
556 557 558
{
    uint16_t device_class;
    uint8_t header_type, secondary, subordinate;
559
    virPCIDevicePtr *best = data;
560 561
    int ret = 0;
    int fd;
562

563
    if (dev->domain != check->domain)
564 565
        return 0;

566
    if ((fd = virPCIDeviceConfigOpen(check, false)) < 0)
567 568
        return 0;

569
    /* Is it a bridge? */
570
    device_class = virPCIDeviceRead16(check, fd, PCI_CLASS_DEVICE);
571
    if (device_class != PCI_CLASS_BRIDGE_PCI)
572
        goto cleanup;
573 574

    /* Is it a plane? */
575
    header_type = virPCIDeviceRead8(check, fd, PCI_HEADER_TYPE);
576
    if ((header_type & PCI_HEADER_TYPE_MASK) != PCI_HEADER_TYPE_BRIDGE)
577
        goto cleanup;
578

579 580
    secondary   = virPCIDeviceRead8(check, fd, PCI_SECONDARY_BUS);
    subordinate = virPCIDeviceRead8(check, fd, PCI_SUBORDINATE_BUS);
581

582
    VIR_DEBUG("%s %s: found parent device %s", dev->id, dev->name, check->name);
583

584 585 586
    /* if the secondary bus exactly equals the device's bus, then we found
     * the direct parent.  No further work is necessary
     */
587 588 589 590
    if (dev->bus == secondary) {
        ret = 1;
        goto cleanup;
    }
591 592 593 594 595 596 597

    /* 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) {
598 599
            *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                    check->function);
600 601 602 603 604
            if (*best == NULL) {
                ret = -1;
                goto cleanup;
            }
        } else {
605 606 607 608
            /* 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
             */
609 610 611
            int bestfd;
            uint8_t best_secondary;

612
            if ((bestfd = virPCIDeviceConfigOpen(*best, false)) < 0)
613
                goto cleanup;
614 615
            best_secondary = virPCIDeviceRead8(*best, bestfd, PCI_SECONDARY_BUS);
            virPCIDeviceConfigClose(*best, bestfd);
616 617

            if (secondary > best_secondary) {
618 619 620
                virPCIDeviceFree(*best);
                *best = virPCIDeviceNew(check->domain, check->bus, check->slot,
                                        check->function);
621 622 623 624
                if (*best == NULL) {
                    ret = -1;
                    goto cleanup;
                }
625 626 627 628
            }
        }
    }

629
cleanup:
630
    virPCIDeviceConfigClose(check, fd);
631
    return ret;
632 633
}

634
static int
635
virPCIDeviceGetParent(virPCIDevicePtr dev, virPCIDevicePtr *parent)
636
{
637
    virPCIDevicePtr best = NULL;
638 639 640
    int ret;

    *parent = NULL;
641
    ret = virPCIDeviceIterDevices(virPCIDeviceIsParent, dev, parent, &best);
642
    if (ret == 1)
643
        virPCIDeviceFree(best);
644 645 646
    else if (ret == 0)
        *parent = best;
    return ret;
647 648 649 650 651 652
}

/* Secondary Bus Reset is our sledgehammer - it resets all
 * devices behind a bus.
 */
static int
653 654 655
virPCIDeviceTrySecondaryBusReset(virPCIDevicePtr dev,
                                 int cfgfd,
                                 virPCIDeviceList *inactiveDevs)
656
{
657
    virPCIDevicePtr parent, conflict;
658 659 660
    uint8_t config_space[PCI_CONF_LEN];
    uint16_t ctl;
    int ret = -1;
661
    int parentfd;
662

663 664 665
    /* Refuse to do a secondary bus reset if there are other
     * devices/functions behind the bus are used by the host
     * or other guests.
666
     */
667
    if ((conflict = virPCIDeviceBusContainsActiveDevices(dev, inactiveDevs))) {
668
        virReportError(VIR_ERR_INTERNAL_ERROR,
669 670
                       _("Active %s devices on bus with %s, not doing bus reset"),
                       conflict->name, dev->name);
671 672 673 674
        return -1;
    }

    /* Find the parent bus */
675
    if (virPCIDeviceGetParent(dev, &parent) < 0)
676
        return -1;
677
    if (!parent) {
678
        virReportError(VIR_ERR_INTERNAL_ERROR,
679 680
                       _("Failed to find parent device for %s"),
                       dev->name);
681 682
        return -1;
    }
683
    if ((parentfd = virPCIDeviceConfigOpen(parent, true)) < 0)
684
        goto out;
685 686 687 688 689 690 691

    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
     */
692
    if (virPCIDeviceRead(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
693
        virReportError(VIR_ERR_INTERNAL_ERROR,
694
                       _("Failed to read PCI config space for %s"),
695
                       dev->name);
696 697 698 699 700 701
        goto out;
    }

    /* Read the control register, set the reset flag, wait 200ms,
     * unset the reset flag and wait 200ms.
     */
702
    ctl = virPCIDeviceRead16(dev, cfgfd, PCI_BRIDGE_CONTROL);
703

704 705
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL,
                        ctl | PCI_BRIDGE_CTL_RESET);
706 707 708

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

709
    virPCIDeviceWrite16(parent, parentfd, PCI_BRIDGE_CONTROL, ctl);
710 711 712

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

713
    if (virPCIDeviceWrite(dev, cfgfd, 0, config_space, PCI_CONF_LEN) < 0) {
714
        virReportError(VIR_ERR_INTERNAL_ERROR,
715 716 717 718
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        goto out;
    }
719
    ret = 0;
720

721
out:
722 723
    virPCIDeviceConfigClose(parent, parentfd);
    virPCIDeviceFree(parent);
724 725 726 727 728 729 730 731
    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
732
virPCIDeviceTryPowerManagementReset(virPCIDevicePtr dev, int cfgfd)
733 734 735 736 737 738 739 740
{
    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. */
741
    if (virPCIDeviceRead(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
742
        virReportError(VIR_ERR_INTERNAL_ERROR,
743
                       _("Failed to read PCI config space for %s"),
744
                       dev->name);
745 746 747 748 749
        return -1;
    }

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

750
    ctl = virPCIDeviceRead32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL);
751 752
    ctl &= ~PCI_PM_CTRL_STATE_MASK;

753 754
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D3hot);
755 756 757

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

758 759
    virPCIDeviceWrite32(dev, cfgfd, dev->pci_pm_cap_pos + PCI_PM_CTRL,
                        ctl | PCI_PM_CTRL_STATE_D0);
760 761 762

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

763
    if (virPCIDeviceWrite(dev, cfgfd, 0, &config_space[0], PCI_CONF_LEN) < 0) {
764
        virReportError(VIR_ERR_INTERNAL_ERROR,
765 766 767 768
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        return -1;
    }
769 770 771 772 773

    return 0;
}

static int
774
virPCIDeviceInit(virPCIDevicePtr dev, int cfgfd)
775
{
776 777
    int flr;

778 779 780
    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);
781
    if (flr < 0)
782
        return flr;
783 784
    dev->has_flr        = !!flr;
    dev->has_pm_reset   = !!virPCIDeviceDetectPowerManagementReset(dev, cfgfd);
785

786 787 788 789
    return 0;
}

int
790 791 792
virPCIDeviceReset(virPCIDevicePtr dev,
                  virPCIDeviceList *activeDevs,
                  virPCIDeviceList *inactiveDevs)
793 794
{
    int ret = -1;
795
    int fd;
796

797
    if (activeDevs && virPCIDeviceListFind(activeDevs, dev)) {
798
        virReportError(VIR_ERR_INTERNAL_ERROR,
799 800 801 802
                       _("Not resetting active device %s"), dev->name);
        return -1;
    }

803
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
804 805
        return -1;

806
    if (virPCIDeviceInit(dev, fd) < 0)
807 808
        goto cleanup;

809 810 811
    /* KVM will perform FLR when starting and stopping
     * a guest, so there is no need for us to do it here.
     */
812 813 814 815
    if (dev->has_flr) {
        ret = 0;
        goto cleanup;
    }
816

817 818 819 820 821
    /* 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)
822
        ret = virPCIDeviceTryPowerManagementReset(dev, fd);
823

824
    /* Bus reset is not an option with the root bus */
825
    if (ret < 0 && dev->bus != 0)
826
        ret = virPCIDeviceTrySecondaryBusReset(dev, fd, inactiveDevs);
827

828 829
    if (ret < 0) {
        virErrorPtr err = virGetLastError();
830
        virReportError(VIR_ERR_INTERNAL_ERROR,
831 832 833 834 835
                       _("Unable to reset PCI device %s: %s"),
                       dev->name,
                       err ? err->message : _("no FLR, PM reset or bus reset available"));
    }

836
cleanup:
837
    virPCIDeviceConfigClose(dev, fd);
838 839 840
    return ret;
}

841

842
static int
843
virPCIDriverDir(char **buffer, const char *driver)
844
{
845 846 847 848 849 850 851 852
    VIR_FREE(*buffer);

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

    return 0;
853 854
}

855
static int
856
virPCIDriverFile(char **buffer, const char *driver, const char *file)
857
{
858 859 860 861 862 863 864 865
    VIR_FREE(*buffer);

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

    return 0;
866
}
867

868
static int
869
virPCIFile(char **buffer, const char *device, const char *file)
870
{
871 872 873 874 875 876 877 878
    VIR_FREE(*buffer);

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

    return 0;
879
}
880

881
static int
882
virPCIProbeStubDriver(const char *driver)
883
{
884
    char *drvpath = NULL;
885 886 887
    int probed = 0;

recheck:
888
    if (virPCIDriverDir(&drvpath, driver) == 0 && virFileExists(drvpath)) {
889
        /* driver already loaded, return */
890
        VIR_FREE(drvpath);
891
        return 0;
892 893 894
    }

    VIR_FREE(drvpath);
895 896

    if (!probed) {
897
        const char *const probecmd[] = { MODPROBE, driver, NULL };
898
        probed = 1;
899
        if (virRun(probecmd, NULL) < 0) {
900
            char ebuf[1024];
901
            VIR_WARN("failed to load driver %s: %s", driver,
902
                     virStrerror(errno, ebuf, sizeof(ebuf)));
903
            return -1;
904
        }
905 906

        goto recheck;
907 908
    }

909
    return -1;
910 911
}

912 913 914 915 916 917 918
static const char *virPCIKnownStubs[] = {
    "pciback",  /* used by xen */
    "pci-stub", /* used by kvm legacy passthrough */
    "vfio-pci", /* used by VFIO device assignment */
    NULL
};

919
static int
920
virPCIDeviceUnbindFromStub(virPCIDevicePtr dev)
921 922 923 924
{
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;
925 926 927
    char *driver = NULL;
    const char **stubTest;
    bool isStub = false;
928

929 930 931 932
    /* 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
933
        goto cleanup;
934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951
    /* 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}" */
    if (!(driver = strdup(last_component(drvdir)))) {
        virReportOOMError();
        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 1280 1281 1282 1283 1284 1285 1286 1287 1288
    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
         */
1289
        VIR_DEBUG("Failed to open /proc/iomem, trying to continue anyway");
1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
        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))) {
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 1320 1321
                ret = 1;
                break;
            }
        }
        else {
            in_matching_device = 0;

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 1341 1342 1343 1344 1345
                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
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 1393 1394 1395 1396 1397 1398 1399 1400
    if (dev != NULL) {
        if ((*pciConfigAddr = strdup(dev->name)) == NULL) {
            virReportOOMError();
            goto cleanup;
        }
        ret = 0;
    }

cleanup:
1401
    virPCIDeviceFree(dev);
1402 1403 1404
    return ret;
}

1405
virPCIDevicePtr
1406 1407 1408 1409
virPCIDeviceNew(unsigned int domain,
                unsigned int bus,
                unsigned int slot,
                unsigned int function)
1410
{
1411
    virPCIDevicePtr dev;
E
Eric Blake 已提交
1412 1413
    char *vendor = NULL;
    char *product = NULL;
1414 1415

    if (VIR_ALLOC(dev) < 0) {
1416
        virReportOOMError();
1417 1418 1419 1420 1421 1422 1423 1424
        return NULL;
    }

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

E
Eric Blake 已提交
1425 1426 1427
    if (snprintf(dev->name, sizeof(dev->name), "%.4x:%.2x:%.2x.%.1x",
                 dev->domain, dev->bus, dev->slot,
                 dev->function) >= sizeof(dev->name)) {
1428
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1429 1430
                       _("dev->name buffer overflow: %.4x:%.2x:%.2x.%.1x"),
                       dev->domain, dev->bus, dev->slot, dev->function);
E
Eric Blake 已提交
1431
        goto error;
E
Eric Blake 已提交
1432 1433 1434 1435
    }
    if (virAsprintf(&dev->path, PCI_SYSFS "devices/%s/config",
                    dev->name) < 0) {
        virReportOOMError();
E
Eric Blake 已提交
1436
        goto error;
E
Eric Blake 已提交
1437
    }
1438

1439 1440 1441 1442
    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
1443
        goto error;
1444 1445
    }

1446 1447
    vendor  = virPCIDeviceReadID(dev, "vendor");
    product = virPCIDeviceReadID(dev, "device");
1448 1449

    if (!vendor || !product) {
1450
        virReportError(VIR_ERR_INTERNAL_ERROR,
1451 1452
                       _("Failed to read product/vendor ID for %s"),
                       dev->name);
E
Eric Blake 已提交
1453
        goto error;
1454 1455 1456
    }

    /* strings contain '0x' prefix */
E
Eric Blake 已提交
1457 1458
    if (snprintf(dev->id, sizeof(dev->id), "%s %s", &vendor[2],
                 &product[2]) >= sizeof(dev->id)) {
1459
        virReportError(VIR_ERR_INTERNAL_ERROR,
E
Eric Blake 已提交
1460 1461
                       _("dev->id buffer overflow: %s %s"),
                       &vendor[2], &product[2]);
E
Eric Blake 已提交
1462
        goto error;
E
Eric Blake 已提交
1463
    }
1464 1465 1466

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

E
Eric Blake 已提交
1467 1468 1469
cleanup:
    VIR_FREE(product);
    VIR_FREE(vendor);
1470
    return dev;
E
Eric Blake 已提交
1471 1472

error:
1473
    virPCIDeviceFree(dev);
E
Eric Blake 已提交
1474 1475
    dev = NULL;
    goto cleanup;
1476 1477 1478
}

void
1479
virPCIDeviceFree(virPCIDevicePtr dev)
1480
{
1481 1482
    if (!dev)
        return;
1483
    VIR_DEBUG("%s %s: freeing", dev->id, dev->name);
E
Eric Blake 已提交
1484
    VIR_FREE(dev->path);
1485 1486
    VIR_FREE(dev);
}
1487

1488
const char *
1489
virPCIDeviceGetName(virPCIDevicePtr dev)
1490 1491 1492 1493
{
    return dev->name;
}

1494
void virPCIDeviceSetManaged(virPCIDevicePtr dev, bool managed)
1495
{
1496
    dev->managed = managed;
1497 1498
}

1499 1500
unsigned int
virPCIDeviceGetManaged(virPCIDevicePtr dev)
1501 1502 1503 1504
{
    return dev->managed;
}

1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516
void
virPCIDeviceSetStubDriver(virPCIDevicePtr dev, const char *driver)
{
    dev->stubDriver = driver;
}

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

1517
unsigned int
1518
virPCIDeviceGetUnbindFromStub(virPCIDevicePtr dev)
1519 1520 1521 1522 1523
{
    return dev->unbind_from_stub;
}

void
1524
virPCIDeviceSetUnbindFromStub(virPCIDevicePtr dev, bool unbind)
1525
{
1526
    dev->unbind_from_stub = unbind;
1527 1528
}

1529
unsigned int
1530
virPCIDeviceGetRemoveSlot(virPCIDevicePtr dev)
1531 1532 1533 1534 1535
{
    return dev->remove_slot;
}

void
1536
virPCIDeviceSetRemoveSlot(virPCIDevicePtr dev, bool remove_slot)
1537
{
1538
    dev->remove_slot = remove_slot;
1539 1540
}

1541
unsigned int
1542
virPCIDeviceGetReprobe(virPCIDevicePtr dev)
1543 1544 1545 1546 1547
{
    return dev->reprobe;
}

void
1548
virPCIDeviceSetReprobe(virPCIDevicePtr dev, bool reprobe)
1549
{
1550
    dev->reprobe = reprobe;
1551 1552
}

1553
void
1554
virPCIDeviceSetUsedBy(virPCIDevicePtr dev, const char *name)
1555 1556 1557 1558 1559
{
    dev->used_by = name;
}

const char *
1560
virPCIDeviceGetUsedBy(virPCIDevicePtr dev)
1561 1562 1563 1564
{
    return dev->used_by;
}

1565
void virPCIDeviceReattachInit(virPCIDevicePtr pci)
1566
{
1567 1568 1569
    pci->unbind_from_stub = true;
    pci->remove_slot = true;
    pci->reprobe = true;
1570 1571 1572
}


1573 1574
virPCIDeviceListPtr
virPCIDeviceListNew(void)
1575
{
1576
    virPCIDeviceListPtr list;
1577

1578 1579 1580 1581
    if (virPCIInitialize() < 0)
        return NULL;

    if (!(list = virObjectLockableNew(virPCIDeviceListClass)))
1582 1583 1584 1585 1586
        return NULL;

    return list;
}

1587 1588
static void
virPCIDeviceListDispose(void *obj)
1589
{
1590
    virPCIDeviceListPtr list = obj;
1591 1592 1593
    int i;

    for (i = 0; i < list->count; i++) {
1594
        virPCIDeviceFree(list->devs[i]);
1595 1596 1597 1598 1599 1600 1601 1602
        list->devs[i] = NULL;
    }

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

int
1603 1604
virPCIDeviceListAdd(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1605
{
1606
    if (virPCIDeviceListFind(list, dev)) {
1607
        virReportError(VIR_ERR_INTERNAL_ERROR,
1608 1609 1610 1611 1612
                       _("Device %s is already in use"), dev->name);
        return -1;
    }

    if (VIR_REALLOC_N(list->devs, list->count+1) < 0) {
1613
        virReportOOMError();
1614 1615 1616 1617 1618 1619 1620 1621
        return -1;
    }

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

    return 0;
}

1622 1623 1624
virPCIDevicePtr
virPCIDeviceListGet(virPCIDeviceListPtr list,
                    int idx)
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634
{
    if (idx >= list->count)
        return NULL;
    if (idx < 0)
        return NULL;

    return list->devs[idx];
}

int
1635
virPCIDeviceListCount(virPCIDeviceListPtr list)
1636
{
1637 1638 1639
    return list->count;
}

1640 1641 1642
virPCIDevicePtr
virPCIDeviceListStealIndex(virPCIDeviceListPtr list,
                           int idx)
1643
{
1644
    virPCIDevicePtr ret;
1645

1646 1647
    if (idx < 0 || idx >= list->count)
        return NULL;
1648

1649
    ret = list->devs[idx];
1650

1651 1652 1653 1654 1655
    if (idx != --list->count) {
        memmove(&list->devs[idx],
                &list->devs[idx + 1],
                sizeof(*list->devs) * (list->count - idx));
    }
1656

1657 1658
    if (VIR_REALLOC_N(list->devs, list->count) < 0) {
        ; /* not fatal */
1659
    }
1660

1661 1662 1663
    return ret;
}

1664 1665 1666
virPCIDevicePtr
virPCIDeviceListSteal(virPCIDeviceListPtr list,
                      virPCIDevicePtr dev)
1667
{
1668
    return virPCIDeviceListStealIndex(list, virPCIDeviceListFindIndex(list, dev));
1669 1670
}

1671
void
1672 1673
virPCIDeviceListDel(virPCIDeviceListPtr list,
                    virPCIDevicePtr dev)
1674
{
1675
    virPCIDevicePtr ret = virPCIDeviceListSteal(list, dev);
1676
    if (ret)
1677
        virPCIDeviceFree(ret);
1678 1679
}

1680
int
1681
virPCIDeviceListFindIndex(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1682 1683 1684 1685 1686 1687 1688 1689
{
    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)
1690 1691 1692 1693
            return i;
    return -1;
}

1694 1695
virPCIDevicePtr
virPCIDeviceListFind(virPCIDeviceListPtr list, virPCIDevicePtr dev)
1696 1697 1698
{
    int i;

1699
    if ((i = virPCIDeviceListFindIndex(list, dev)) >= 0)
1700 1701 1702
        return list->devs[i];
    else
        return NULL;
1703
}
1704 1705


1706 1707 1708
int virPCIDeviceFileIterate(virPCIDevicePtr dev,
                            virPCIDeviceFileActor actor,
                            void *opaque)
1709 1710 1711 1712 1713 1714 1715 1716 1717
{
    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) {
1718
        virReportOOMError();
1719 1720 1721 1722
        goto cleanup;
    }

    if (!(dir = opendir(pcidir))) {
1723
        virReportSystemError(errno,
1724 1725 1726 1727 1728 1729
                             _("cannot open %s"), pcidir);
        goto cleanup;
    }

    while ((ent = readdir(dir)) != NULL) {
        /* Device assignment requires:
A
Alex Williamson 已提交
1730 1731
         *   $PCIDIR/config, $PCIDIR/resource, $PCIDIR/resourceNNN,
         *   $PCIDIR/rom, $PCIDIR/reset
1732 1733 1734
         */
        if (STREQ(ent->d_name, "config") ||
            STRPREFIX(ent->d_name, "resource") ||
A
Alex Williamson 已提交
1735 1736
            STREQ(ent->d_name, "rom") ||
            STREQ(ent->d_name, "reset")) {
1737
            if (virAsprintf(&file, "%s/%s", pcidir, ent->d_name) < 0) {
1738
                virReportOOMError();
1739 1740
                goto cleanup;
            }
1741
            if ((actor)(dev, file, opaque) < 0)
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756
                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
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 1791 1792
/* 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
1793
static int
1794
virPCIDeviceDownstreamLacksACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1795 1796 1797 1798
{
    uint16_t flags;
    uint16_t ctrl;
    unsigned int pos;
1799 1800
    int fd;
    int ret = 0;
J
Jiri Denemark 已提交
1801

1802
    if ((fd = virPCIDeviceConfigOpen(dev, true)) < 0)
J
Jiri Denemark 已提交
1803 1804
        return -1;

1805
    if (virPCIDeviceInit(dev, fd) < 0) {
1806 1807 1808 1809
        ret = -1;
        goto cleanup;
    }

J
Jiri Denemark 已提交
1810
    pos = dev->pcie_cap_pos;
1811
    if (!pos || virPCIDeviceRead16(dev, fd, PCI_CLASS_DEVICE) != PCI_CLASS_BRIDGE_PCI)
1812
        goto cleanup;
J
Jiri Denemark 已提交
1813

1814
    flags = virPCIDeviceRead16(dev, fd, pos + PCI_EXP_FLAGS);
J
Jiri Denemark 已提交
1815
    if (((flags & PCI_EXP_FLAGS_TYPE) >> 4) != PCI_EXP_TYPE_DOWNSTREAM)
1816
        goto cleanup;
J
Jiri Denemark 已提交
1817

1818
    pos = virPCIDeviceFindExtendedCapabilityOffset(dev, fd, PCI_EXT_CAP_ID_ACS);
J
Jiri Denemark 已提交
1819 1820
    if (!pos) {
        VIR_DEBUG("%s %s: downstream port lacks ACS", dev->id, dev->name);
1821 1822
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1823 1824
    }

1825
    ctrl = virPCIDeviceRead16(dev, fd, pos + PCI_EXT_ACS_CTRL);
J
Jiri Denemark 已提交
1826 1827 1828
    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);
1829 1830
        ret = 1;
        goto cleanup;
J
Jiri Denemark 已提交
1831 1832
    }

1833
cleanup:
1834
    virPCIDeviceConfigClose(dev, fd);
1835
    return ret;
J
Jiri Denemark 已提交
1836 1837 1838
}

static int
1839
virPCIDeviceIsBehindSwitchLackingACS(virPCIDevicePtr dev)
J
Jiri Denemark 已提交
1840
{
1841
    virPCIDevicePtr parent;
J
Jiri Denemark 已提交
1842

1843
    if (virPCIDeviceGetParent(dev, &parent) < 0)
1844
        return -1;
1845 1846 1847 1848 1849 1850 1851 1852
    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 {
1853
            virReportError(VIR_ERR_INTERNAL_ERROR,
1854 1855 1856 1857
                           _("Failed to find parent device for %s"),
                           dev->name);
            return -1;
        }
J
Jiri Denemark 已提交
1858 1859 1860 1861 1862 1863 1864
    }

    /* 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 {
1865
        virPCIDevicePtr tmp;
J
Jiri Denemark 已提交
1866
        int acs;
1867
        int ret;
J
Jiri Denemark 已提交
1868

1869
        acs = virPCIDeviceDownstreamLacksACS(parent);
J
Jiri Denemark 已提交
1870 1871

        if (acs) {
1872
            virPCIDeviceFree(parent);
J
Jiri Denemark 已提交
1873 1874 1875 1876 1877 1878 1879
            if (acs < 0)
                return -1;
            else
                return 1;
        }

        tmp = parent;
1880 1881
        ret = virPCIDeviceGetParent(parent, &parent);
        virPCIDeviceFree(tmp);
1882 1883
        if (ret < 0)
            return -1;
J
Jiri Denemark 已提交
1884 1885 1886 1887 1888
    } while (parent);

    return 0;
}

1889 1890
int virPCIDeviceIsAssignable(virPCIDevicePtr dev,
                             int strict_acs_check)
J
Jiri Denemark 已提交
1891 1892 1893 1894 1895 1896 1897 1898
{
    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.
     */

1899
    ret = virPCIDeviceIsBehindSwitchLackingACS(dev);
J
Jiri Denemark 已提交
1900 1901 1902 1903 1904 1905 1906 1907
    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 {
1908
            virReportError(VIR_ERR_INTERNAL_ERROR,
J
Jiri Denemark 已提交
1909 1910 1911 1912 1913 1914 1915 1916 1917
                           _("Device %s is behind a switch lacking ACS and "
                             "cannot be assigned"),
                           dev->name);
            return 0;
        }
    }

    return 1;
}
1918

1919 1920
#ifdef __linux__

1921
/*
1922
 * returns true if equal
1923
 */
1924
static bool
1925 1926
virPCIDeviceAddressIsEqual(virPCIDeviceAddressPtr bdf1,
                           virPCIDeviceAddressPtr bdf2)
1927 1928 1929 1930 1931 1932 1933
{
    return ((bdf1->domain == bdf2->domain) &&
            (bdf1->bus == bdf2->bus) &&
            (bdf1->slot == bdf2->slot) &&
            (bdf1->function == bdf2->function));
}

1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952
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
1953 1954
virPCIParseDeviceAddress(char *address,
                         virPCIDeviceAddressPtr bdf)
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 1984 1985
{
    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
1986 1987
virPCIGetDeviceAddressFromSysfsLink(const char *device_link,
                                    virPCIDeviceAddressPtr *bdf)
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
{
    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;
    }

2002
    device_path = canonicalize_file_name(device_link);
2003 2004
    if (device_path == NULL) {
        memset(errbuf, '\0', sizeof(errbuf));
2005 2006 2007
        virReportSystemError(errno,
                             _("Failed to resolve device link '%s'"),
                             device_link);
2008 2009 2010
        return ret;
    }

2011
    config_address = last_component(device_path);
2012 2013 2014 2015 2016
    if (VIR_ALLOC(*bdf) != 0) {
        virReportOOMError();
        goto out;
    }

2017
    if (virPCIParseDeviceAddress(config_address, *bdf) != 0) {
2018
        virReportError(VIR_ERR_INTERNAL_ERROR,
2019 2020 2021 2022 2023 2024
                       _("Failed to parse PCI config address '%s'"),
                       config_address);
        VIR_FREE(*bdf);
        goto out;
    }

2025
    VIR_DEBUG("virPCIDeviceAddress %.4x:%.2x:%.2x.%.1x",
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042
              (*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
2043 2044
virPCIGetPhysicalFunction(const char *vf_sysfs_path,
                          virPCIDeviceAddressPtr *physical_function)
2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
{
    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 {
2056 2057
        ret = virPCIGetDeviceAddressFromSysfsLink(device_link,
                                                  physical_function);
2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068
    }

    VIR_FREE(device_link);

    return ret;
}

/*
 * Returns virtual functions of a physical function
 */
int
2069 2070 2071
virPCIGetVirtualFunctions(const char *sysfs_path,
                          virPCIDeviceAddressPtr **virtual_functions,
                          unsigned int *num_virtual_functions)
2072 2073
{
    int ret = -1;
2074
    int i;
2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085
    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));
2086 2087 2088
        virReportSystemError(errno,
                             _("Failed to open dir '%s'"),
                             sysfs_path);
2089 2090 2091 2092 2093 2094 2095
        return ret;
    }

    *virtual_functions = NULL;
    *num_virtual_functions = 0;
    while ((entry = readdir(dir))) {
        if (STRPREFIX(entry->d_name, "virtfn")) {
2096
            virPCIDeviceAddress *config_addr = NULL;
2097 2098 2099

            if (virBuildPath(&device_link, sysfs_path, entry->d_name) == -1) {
                virReportOOMError();
2100
                goto error;
2101 2102 2103 2104 2105
            }

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

2106
            if (virPCIGetDeviceAddressFromSysfsLink(device_link,
2107
                                                    &config_addr) !=
2108
                SRIOV_FOUND) {
2109 2110
                VIR_WARN("Failed to get SRIOV function from device "
                         "link '%s'", device_link);
2111
                VIR_FREE(device_link);
2112
                continue;
2113
            }
2114

2115 2116
            if (VIR_REALLOC_N(*virtual_functions,
                              *num_virtual_functions + 1) < 0) {
2117 2118 2119 2120 2121 2122 2123
                virReportOOMError();
                VIR_FREE(config_addr);
                goto error;
            }

            (*virtual_functions)[*num_virtual_functions] = config_addr;
            (*num_virtual_functions)++;
2124 2125 2126 2127 2128 2129
            VIR_FREE(device_link);
        }
    }

    ret = 0;

2130 2131
cleanup:
    VIR_FREE(device_link);
2132 2133 2134
    if (dir)
        closedir(dir);
    return ret;
2135 2136 2137 2138 2139 2140 2141 2142

error:
    if (*virtual_functions) {
        for (i = 0; i < *num_virtual_functions; i++)
            VIR_FREE((*virtual_functions)[i]);
        VIR_FREE(*virtual_functions);
    }
    goto cleanup;
2143
}
2144 2145 2146 2147 2148

/*
 * Returns 1 if vf device is a virtual function, 0 if not, -1 on error
 */
int
2149
virPCIIsVirtualFunction(const char *vf_sysfs_device_link)
2150 2151 2152 2153 2154
{
    char *vf_sysfs_physfn_link = NULL;
    int ret = -1;

    if (virAsprintf(&vf_sysfs_physfn_link, "%s/physfn",
2155
                    vf_sysfs_device_link) < 0) {
2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170
        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
2171 2172 2173
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link,
                              const char *vf_sysfs_device_link,
                              int *vf_index)
2174 2175 2176
{
    int ret = -1, i;
    unsigned int num_virt_fns = 0;
2177 2178
    virPCIDeviceAddressPtr vf_bdf = NULL;
    virPCIDeviceAddressPtr *virt_fns = NULL;
2179

2180 2181
    if (virPCIGetDeviceAddressFromSysfsLink(vf_sysfs_device_link,
                                            &vf_bdf) < 0)
2182 2183
        return ret;

2184 2185
    if (virPCIGetVirtualFunctions(pf_sysfs_device_link, &virt_fns,
                                  &num_virt_fns) < 0) {
2186
        virReportError(VIR_ERR_INTERNAL_ERROR,
2187
                       _("Error getting physical function's '%s' "
2188
                         "virtual_functions"), pf_sysfs_device_link);
2189 2190 2191 2192
        goto out;
    }

    for (i = 0; i < num_virt_fns; i++) {
2193 2194 2195 2196 2197
        if (virPCIDeviceAddressIsEqual(vf_bdf, virt_fns[i])) {
            *vf_index = i;
            ret = 0;
            break;
        }
2198 2199 2200 2201 2202 2203
    }

out:

    /* free virtual functions */
    for (i = 0; i < num_virt_fns; i++)
2204
        VIR_FREE(virt_fns[i]);
2205

A
ajia@redhat.com 已提交
2206
    VIR_FREE(virt_fns);
2207 2208 2209 2210 2211
    VIR_FREE(vf_bdf);

    return ret;
}

2212 2213 2214 2215 2216
/*
 * Returns a path to the PCI sysfs file given the BDF of the PCI function
 */

int
2217
virPCIGetSysfsFile(char *virPCIDeviceName, char **pci_sysfs_device_link)
2218 2219
{
    if (virAsprintf(pci_sysfs_device_link, PCI_SYSFS "devices/%s",
2220
                    virPCIDeviceName) < 0) {
2221 2222 2223 2224 2225 2226 2227
        virReportOOMError();
        return -1;
    }

    return 0;
}

R
Roopa Prabhu 已提交
2228
int
2229 2230
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev,
                                char **pci_sysfs_device_link)
R
Roopa Prabhu 已提交
2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241
{
    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;
}

2242 2243 2244 2245
/*
 * Returns the network device name of a pci device
 */
int
2246 2247 2248 2249 2250 2251
virPCIGetNetName(char *device_link_sysfs_path, char **netname)
{
    char *pcidev_sysfs_net_path = NULL;
    int ret = -1;
    DIR *dir = NULL;
    struct dirent *entry = NULL;
2252

2253 2254 2255 2256 2257 2258 2259 2260 2261
    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;
2262

2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277
    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);
2278 2279

out:
2280
    VIR_FREE(pcidev_sysfs_net_path);
2281

2282
    return ret;
2283
}
R
Roopa Prabhu 已提交
2284 2285

int
2286 2287
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path,
                             char **pfname, int *vf_index)
R
Roopa Prabhu 已提交
2288
{
2289
    virPCIDeviceAddressPtr pf_config_address = NULL;
R
Roopa Prabhu 已提交
2290 2291 2292
    char *pf_sysfs_device_path = NULL;
    int ret = -1;

2293
    if (virPCIGetPhysicalFunction(vf_sysfs_device_path, &pf_config_address) < 0)
R
Roopa Prabhu 已提交
2294 2295
        return ret;

2296 2297
    if (virPCIDeviceAddressGetSysfsFile(pf_config_address,
                                        &pf_sysfs_device_path) < 0) {
R
Roopa Prabhu 已提交
2298 2299 2300 2301 2302

        VIR_FREE(pf_config_address);
        return ret;
    }

2303 2304
    if (virPCIGetVirtualFunctionIndex(pf_sysfs_device_path, vf_sysfs_device_path,
                                      vf_index) < 0)
R
Roopa Prabhu 已提交
2305 2306
        goto cleanup;

2307
    ret = virPCIGetNetName(pf_sysfs_device_path, pfname);
R
Roopa Prabhu 已提交
2308 2309 2310 2311 2312 2313 2314 2315

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

    return ret;
}

2316
#else
2317 2318
static const char *unsupported = N_("not supported on non-linux platforms");

2319
int
2320 2321
virPCIGetPhysicalFunction(const char *vf_sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr *physical_function ATTRIBUTE_UNUSED)
2322
{
2323
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2324 2325 2326 2327
    return -1;
}

int
2328 2329 2330
virPCIGetVirtualFunctions(const char *sysfs_path ATTRIBUTE_UNUSED,
                          virPCIDeviceAddressPtr **virtual_functions ATTRIBUTE_UNUSED,
                          unsigned int *num_virtual_functions ATTRIBUTE_UNUSED)
2331
{
2332
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2333 2334
    return -1;
}
2335 2336

int
E
Eric Blake 已提交
2337
virPCIIsVirtualFunction(const char *vf_sysfs_device_link ATTRIBUTE_UNUSED)
2338
{
2339
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2340 2341 2342 2343
    return -1;
}

int
2344 2345 2346
virPCIGetVirtualFunctionIndex(const char *pf_sysfs_device_link ATTRIBUTE_UNUSED,
                              const char *vf_sysfs_device_link ATTRIBUTE_UNUSED,
                              int *vf_index ATTRIBUTE_UNUSED)
2347
{
2348
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2349 2350 2351 2352
    return -1;

}

2353
int
2354 2355
virPCIDeviceAddressGetSysfsFile(virPCIDeviceAddressPtr dev ATTRIBUTE_UNUSED,
                                char **pci_sysfs_device_link ATTRIBUTE_UNUSED)
2356
{
2357
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2358 2359 2360
    return -1;
}

2361
int
2362
virPCIGetNetName(char *device_link_sysfs_path ATTRIBUTE_UNUSED,
2363
                 char **netname ATTRIBUTE_UNUSED)
2364
{
2365
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
2366 2367
    return -1;
}
R
Roopa Prabhu 已提交
2368 2369

int
2370 2371 2372
virPCIGetVirtualFunctionInfo(const char *vf_sysfs_device_path ATTRIBUTE_UNUSED,
                             char **pfname ATTRIBUTE_UNUSED,
                             int *vf_index ATTRIBUTE_UNUSED)
R
Roopa Prabhu 已提交
2373
{
2374
    virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _(unsupported));
R
Roopa Prabhu 已提交
2375 2376
    return -1;
}
2377
#endif /* __linux__ */