pci.c 46.3 KB
Newer Older
1
/*
E
Eric Blake 已提交
2
 * Copyright (C) 2009-2011 Red Hat, Inc.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
 *
 * 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
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307  USA
 *
 * Authors:
 *     Mark McLoughlin <markmc@redhat.com>
 */

#include <config.h>

#include "pci.h"

#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>

#include "logging.h"
#include "memory.h"
38
#include "command.h"
39
#include "virterror_internal.h"
40
#include "files.h"
41

42 43
/* avoid compilation breakage on some systems */
#ifndef MODPROBE
44
# define MODPROBE "modprobe"
45 46
#endif

47 48 49 50 51 52 53 54 55 56 57 58
#define PCI_SYSFS "/sys/bus/pci/"
#define PCI_ID_LEN 10   /* "XXXX XXXX" */
#define PCI_ADDR_LEN 13 /* "XXXX:XX:XX.X" */

struct _pciDevice {
    unsigned      domain;
    unsigned      bus;
    unsigned      slot;
    unsigned      function;

    char          name[PCI_ADDR_LEN]; /* domain:bus:slot.function */
    char          id[PCI_ID_LEN];     /* product vendor */
E
Eric Blake 已提交
59
    char          *path;
60 61 62 63 64 65 66
    int           fd;

    unsigned      initted;
    unsigned      pcie_cap_pos;
    unsigned      pci_pm_cap_pos;
    unsigned      has_flr : 1;
    unsigned      has_pm_reset : 1;
67
    unsigned      managed : 1;
68 69 70 71 72

    /* used by reattach function */
    unsigned      unbind_from_stub : 1;
    unsigned      remove_slot : 1;
    unsigned      reprobe : 1;
73 74
};

75 76 77 78 79 80
struct _pciDeviceList {
    unsigned count;
    pciDevice **devs;
};


81 82 83
/* For virReportOOMError()  and virReportSystemError() */
#define VIR_FROM_THIS VIR_FROM_NONE

84
#define pciReportError(code, ...)                              \
85
    virReportErrorHelper(VIR_FROM_NONE, code, __FILE__,        \
86
                         __FUNCTION__, __LINE__, __VA_ARGS__)
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101

/* 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 */
102 103 104
#define PCI_HEADER_TYPE_BRIDGE 0x1
#define PCI_HEADER_TYPE_MASK   0x7f
#define PCI_HEADER_TYPE_MULTI  0x80
105 106 107 108 109 110 111 112 113

/* 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 */
114
#define PCI_STATUS_CAP_LIST    0x10    /* Support Capability List */
115 116 117 118 119 120 121 122 123 124 125 126 127

/* 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 */
128
#define PCI_EXP_DEVCAP_FLR     (1<<28) /* Function Level Reset */
129 130 131 132 133 134 135

/* 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 */
136
#define PCI_BRIDGE_CTL_RESET   0x40    /* Secondary bus reset */
137 138 139

/* PM12 3.2.4  Power Management Control/Status (Offset = 4) */
#define PCI_PM_CTRL                4    /* PM control and status register */
140 141 142 143
#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 */
144 145 146

/* ECN_AF 6.x.1  Advanced Features Capability Structure */
#define PCI_AF_CAP              0x3     /* Advanced features capabilities */
147
#define PCI_AF_CAP_FLR         0x2     /* Function Level Reset */
148

J
Jiri Denemark 已提交
149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
#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
#define PCI_EXT_CAP_ACS_ENABLED (PCI_EXT_CAP_ACS_SV | \
                                 PCI_EXT_CAP_ACS_RR | \
                                 PCI_EXT_CAP_ACS_CR | \
                                 PCI_EXT_CAP_ACS_UF)

171 172 173 174 175 176 177 178 179 180 181
static int
pciOpenConfig(pciDevice *dev)
{
    int fd;

    if (dev->fd > 0)
        return 0;

    fd = open(dev->path, O_RDWR);
    if (fd < 0) {
        char ebuf[1024];
182
        VIR_WARN("Failed to open config space file '%s': %s",
183 184 185 186 187 188 189 190
                 dev->path, virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    VIR_DEBUG("%s %s: opened %s", dev->id, dev->name, dev->path);
    dev->fd = fd;
    return 0;
}

191 192 193 194 195 196
static void
pciCloseConfig(pciDevice *dev)
{
    if (!dev)
        return;

197
    VIR_FORCE_CLOSE(dev->fd);
198 199
}

200 201 202 203 204 205 206 207
static int
pciRead(pciDevice *dev, unsigned pos, uint8_t *buf, unsigned buflen)
{
    memset(buf, 0, buflen);

    if (pciOpenConfig(dev) < 0)
        return -1;

208 209
    if (lseek(dev->fd, pos, SEEK_SET) != pos ||
        saferead(dev->fd, buf, buflen) != buflen) {
210
        char ebuf[1024];
211
        VIR_WARN("Failed to read from '%s' : %s", dev->path,
212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static uint8_t
pciRead8(pciDevice *dev, unsigned pos)
{
    uint8_t buf;
    pciRead(dev, pos, &buf, sizeof(buf));
    return buf;
}

static uint16_t
pciRead16(pciDevice *dev, unsigned pos)
{
    uint8_t buf[2];
    pciRead(dev, pos, &buf[0], sizeof(buf));
    return (buf[0] << 0) | (buf[1] << 8);
}

static uint32_t
pciRead32(pciDevice *dev, unsigned pos)
{
    uint8_t buf[4];
    pciRead(dev, pos, &buf[0], sizeof(buf));
    return (buf[0] << 0) | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
}

static int
pciWrite(pciDevice *dev, unsigned pos, uint8_t *buf, unsigned buflen)
{
    if (pciOpenConfig(dev) < 0)
        return -1;

248 249
    if (lseek(dev->fd, pos, SEEK_SET) != pos ||
        safewrite(dev->fd, buf, buflen) != buflen) {
250
        char ebuf[1024];
251
        VIR_WARN("Failed to write to '%s' : %s", dev->path,
252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
                 virStrerror(errno, ebuf, sizeof(ebuf)));
        return -1;
    }
    return 0;
}

static void
pciWrite16(pciDevice *dev, unsigned pos, uint16_t val)
{
    uint8_t buf[2] = { (val >> 0), (val >> 8) };
    pciWrite(dev, pos, &buf[0], sizeof(buf));
}

static void
pciWrite32(pciDevice *dev, unsigned pos, uint32_t val)
{
    uint8_t buf[4] = { (val >> 0), (val >> 8), (val >> 16), (val >> 14) };
    pciWrite(dev, pos, &buf[0], sizeof(buf));
}

272
typedef int (*pciIterPredicate)(pciDevice *, pciDevice *, void *);
273 274 275 276 277 278 279

/* 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
280
pciIterDevices(pciIterPredicate predicate,
281
               pciDevice *dev,
282 283
               pciDevice **matched,
               void *data)
284 285 286
{
    DIR *dir;
    struct dirent *entry;
287
    int ret = 0;
288
    int rc;
289 290 291 292 293 294 295

    *matched = NULL;

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

    dir = opendir(PCI_SYSFS "devices");
    if (!dir) {
296
        VIR_WARN("Failed to open " PCI_SYSFS "devices");
297 298 299 300
        return -1;
    }

    while ((entry = readdir(dir))) {
301
        unsigned int domain, bus, slot, function;
302
        pciDevice *check;
303
        char *tmp;
304 305 306 307 308

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

309 310 311 312 313 314 315 316 317
        /* 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) {
318 319 320 321
            VIR_WARN("Unusual entry in " PCI_SYSFS "devices: %s", entry->d_name);
            continue;
        }

322
        check = pciGetDevice(domain, bus, slot, function);
323
        if (!check) {
324 325 326
            ret = -1;
            break;
        }
327

328 329 330 331 332 333 334 335
        rc = predicate(dev, check, data);
        if (rc < 0) {
            /* the predicate returned an error, bail */
            pciFreeDevice(check);
            ret = -1;
            break;
        }
        else if (rc == 1) {
336 337
            VIR_DEBUG("%s %s: iter matched on %s", dev->id, dev->name, check->name);
            *matched = check;
338
            ret = 1;
339 340
            break;
        }
341

342
        pciFreeDevice(check);
343 344
    }
    closedir(dir);
345
    return ret;
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
}

static uint8_t
pciFindCapabilityOffset(pciDevice *dev, unsigned capability)
{
    uint16_t status;
    uint8_t pos;

    status = pciRead16(dev, PCI_STATUS);
    if (!(status & PCI_STATUS_CAP_LIST))
        return 0;

    pos = pciRead8(dev, PCI_CAPABILITY_LIST);

    /* 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) {
        uint8_t capid = pciRead8(dev, pos);
        if (capid == capability) {
            VIR_DEBUG("%s %s: found cap 0x%.2x at 0x%.2x",
                      dev->id, dev->name, capability, pos);
            return pos;
        }

        pos = pciRead8(dev, pos + 1);
    }

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

    return 0;
}

J
Jiri Denemark 已提交
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
static unsigned int
pciFindExtendedCapabilityOffset(pciDevice *dev, unsigned capability)
{
    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) {
        header = pciRead32(dev, pos);

        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;
}

407 408 409 410
/* 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
411 412
pciDetectFunctionLevelReset(pciDevice *dev)
{
M
Mark McLoughlin 已提交
413
    uint32_t caps;
414
    uint8_t pos;
415 416
    char *path;
    int found;
417 418 419 420 421 422 423 424

    /* 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) {
M
Mark McLoughlin 已提交
425
        caps = pciRead32(dev, dev->pcie_cap_pos + PCI_EXP_DEVCAP);
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
        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.
     */
    pos = pciFindCapabilityOffset(dev, PCI_CAP_ID_AF);
    if (pos) {
        caps = pciRead16(dev, pos + PCI_AF_CAP);
        if (caps & PCI_AF_CAP_FLR) {
            VIR_DEBUG("%s %s: detected PCI FLR capability", dev->id, dev->name);
            return 1;
        }
    }

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
    /* 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;
    }

464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
    VIR_DEBUG("%s %s: no FLR capability found", dev->id, dev->name);

    return 0;
}

/* Require the device has the PCI Power Management capability
 * and that a D3hot->D0 transition will results in a full
 * internal reset, not just a soft reset.
 */
static unsigned
pciDetectPowerManagementReset(pciDevice *dev)
{
    if (dev->pci_pm_cap_pos) {
        uint32_t ctl;

        /* require the NO_SOFT_RESET bit is clear */
        ctl = pciRead32(dev, dev->pci_pm_cap_pos + PCI_PM_CTRL);
        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;
}

492
/* Any active devices on the same domain/bus ? */
493
static int
494
pciSharesBusWithActive(pciDevice *dev, pciDevice *check, void *data)
495
{
496
    pciDeviceList *inactiveDevs = data;
497

498
    /* Different domain, different bus, or simply identical device */
499 500
    if (dev->domain != check->domain ||
        dev->bus != check->bus ||
501 502
        (dev->slot == check->slot &&
         dev->function == check->function))
503 504
        return 0;

505 506
    /* same bus, but inactive, i.e. about to be assigned to guest */
    if (inactiveDevs && pciDeviceListFind(inactiveDevs, check))
507
        return 0;
508

509
    return 1;
510 511
}

512
static pciDevice *
513
pciBusContainsActiveDevices(pciDevice *dev,
514
                            pciDeviceList *inactiveDevs)
515 516
{
    pciDevice *active = NULL;
517
    if (pciIterDevices(pciSharesBusWithActive,
518
                       dev, &active, inactiveDevs) < 0)
519 520 521 522 523
        return NULL;
    return active;
}

/* Is @check the parent of @dev ? */
524
static int
525
pciIsParent(pciDevice *dev, pciDevice *check, void *data)
526 527 528
{
    uint16_t device_class;
    uint8_t header_type, secondary, subordinate;
529
    pciDevice **best = data;
530

531
    if (dev->domain != check->domain)
532 533 534
        return 0;

    /* Is it a bridge? */
535
    device_class = pciRead16(check, PCI_CLASS_DEVICE);
536 537 538 539
    if (device_class != PCI_CLASS_BRIDGE_PCI)
        return 0;

    /* Is it a plane? */
540
    header_type = pciRead8(check, PCI_HEADER_TYPE);
541 542 543
    if ((header_type & PCI_HEADER_TYPE_MASK) != PCI_HEADER_TYPE_BRIDGE)
        return 0;

544 545
    secondary   = pciRead8(check, PCI_SECONDARY_BUS);
    subordinate = pciRead8(check, PCI_SUBORDINATE_BUS);
546

547
    VIR_DEBUG("%s %s: found parent device %s", dev->id, dev->name, check->name);
548

549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581
    /* if the secondary bus exactly equals the device's bus, then we found
     * the direct parent.  No further work is necessary
     */
    if (dev->bus == secondary)
        return 1;

    /* 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) {
            *best = pciGetDevice(check->domain, check->bus, check->slot,
                                 check->function);
            if (*best == NULL)
                return -1;
        }
        else {
            /* 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
             */
            if (secondary > pciRead8(*best, PCI_SECONDARY_BUS)) {
                pciFreeDevice(*best);
                *best = pciGetDevice(check->domain, check->bus, check->slot,
                                     check->function);
                if (*best == NULL)
                    return -1;
            }
        }
    }

    return 0;
582 583
}

584 585
static int
pciGetParentDevice(pciDevice *dev, pciDevice **parent)
586
{
587 588 589 590 591 592 593 594 595 596
    pciDevice *best = NULL;
    int ret;

    *parent = NULL;
    ret = pciIterDevices(pciIsParent, dev, parent, &best);
    if (ret == 1)
        pciFreeDevice(best);
    else if (ret == 0)
        *parent = best;
    return ret;
597 598 599 600 601 602
}

/* Secondary Bus Reset is our sledgehammer - it resets all
 * devices behind a bus.
 */
static int
603
pciTrySecondaryBusReset(pciDevice *dev,
604
                        pciDeviceList *inactiveDevs)
605
{
606
    pciDevice *parent, *conflict;
607 608 609 610 611 612 613 614 615
    uint8_t config_space[PCI_CONF_LEN];
    uint16_t ctl;
    int ret = -1;

    /* For now, we just refuse to do a secondary bus reset
     * if there are other devices/functions behind the bus.
     * In future, we could allow it so long as those devices
     * are not in use by the host or other guests.
     */
616
    if ((conflict = pciBusContainsActiveDevices(dev, inactiveDevs))) {
617
        pciReportError(VIR_ERR_INTERNAL_ERROR,
618 619
                       _("Active %s devices on bus with %s, not doing bus reset"),
                       conflict->name, dev->name);
620 621 622 623
        return -1;
    }

    /* Find the parent bus */
624 625
    if (pciGetParentDevice(dev, &parent) < 0)
        return -1;
626
    if (!parent) {
627
        pciReportError(VIR_ERR_INTERNAL_ERROR,
628 629
                       _("Failed to find parent device for %s"),
                       dev->name);
630 631 632 633 634 635 636 637 638 639
        return -1;
    }

    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
     */
    if (pciRead(dev, 0, config_space, PCI_CONF_LEN) < 0) {
640
        pciReportError(VIR_ERR_INTERNAL_ERROR,
641
                       _("Failed to read PCI config space for %s"),
642
                       dev->name);
643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
        goto out;
    }

    /* Read the control register, set the reset flag, wait 200ms,
     * unset the reset flag and wait 200ms.
     */
    ctl = pciRead16(dev, PCI_BRIDGE_CONTROL);

    pciWrite16(parent, PCI_BRIDGE_CONTROL, ctl | PCI_BRIDGE_CTL_RESET);

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

    pciWrite16(parent, PCI_BRIDGE_CONTROL, ctl);

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

659
    if (pciWrite(dev, 0, config_space, PCI_CONF_LEN) < 0) {
660
        pciReportError(VIR_ERR_INTERNAL_ERROR,
661 662 663 664
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        goto out;
    }
665 666
    ret = 0;
out:
667
    pciFreeDevice(parent);
668 669 670 671 672 673 674 675
    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
676
pciTryPowerManagementReset(pciDevice *dev)
677 678 679 680 681 682 683 684 685
{
    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. */
    if (pciRead(dev, 0, &config_space[0], PCI_CONF_LEN) < 0) {
686
        pciReportError(VIR_ERR_INTERNAL_ERROR,
687
                       _("Failed to read PCI config space for %s"),
688
                       dev->name);
689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704
        return -1;
    }

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

    ctl = pciRead32(dev, dev->pci_pm_cap_pos + PCI_PM_CTRL);
    ctl &= ~PCI_PM_CTRL_STATE_MASK;

    pciWrite32(dev, dev->pci_pm_cap_pos + PCI_PM_CTRL, ctl|PCI_PM_CTRL_STATE_D3hot);

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

    pciWrite32(dev, dev->pci_pm_cap_pos + PCI_PM_CTRL, ctl|PCI_PM_CTRL_STATE_D0);

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

705
    if (pciWrite(dev, 0, &config_space[0], PCI_CONF_LEN) < 0) {
706
        pciReportError(VIR_ERR_INTERNAL_ERROR,
707 708 709 710
                       _("Failed to restore PCI config space for %s"),
                       dev->name);
        return -1;
    }
711 712 713 714 715

    return 0;
}

static int
716
pciInitDevice(pciDevice *dev)
717
{
718 719
    int flr;

720
    if (pciOpenConfig(dev) < 0) {
721
        virReportSystemError(errno,
722 723 724 725 726 727 728
                             _("Failed to open config space file '%s'"),
                             dev->path);
        return -1;
    }

    dev->pcie_cap_pos   = pciFindCapabilityOffset(dev, PCI_CAP_ID_EXP);
    dev->pci_pm_cap_pos = pciFindCapabilityOffset(dev, PCI_CAP_ID_PM);
729
    flr = pciDetectFunctionLevelReset(dev);
730
    if (flr < 0)
731 732
        return flr;
    dev->has_flr        = flr;
733 734 735 736 737 738
    dev->has_pm_reset   = pciDetectPowerManagementReset(dev);
    dev->initted        = 1;
    return 0;
}

int
739
pciResetDevice(pciDevice *dev,
740 741
               pciDeviceList *activeDevs,
               pciDeviceList *inactiveDevs)
742 743 744
{
    int ret = -1;

745
    if (activeDevs && pciDeviceListFind(activeDevs, dev)) {
746
        pciReportError(VIR_ERR_INTERNAL_ERROR,
747 748 749 750
                       _("Not resetting active device %s"), dev->name);
        return -1;
    }

751
    if (!dev->initted && pciInitDevice(dev) < 0)
752 753 754 755 756 757 758 759
        return -1;

    /* KVM will perform FLR when starting and stopping
     * a guest, so there is no need for us to do it here.
     */
    if (dev->has_flr)
        return 0;

760 761 762 763 764
    /* 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)
765
        ret = pciTryPowerManagementReset(dev);
766

767
    /* Bus reset is not an option with the root bus */
768
    if (ret < 0 && dev->bus != 0)
769
        ret = pciTrySecondaryBusReset(dev, inactiveDevs);
770

771 772
    if (ret < 0) {
        virErrorPtr err = virGetLastError();
773
        pciReportError(VIR_ERR_INTERNAL_ERROR,
774 775 776 777 778
                       _("Unable to reset PCI device %s: %s"),
                       dev->name,
                       err ? err->message : _("no FLR, PM reset or bus reset available"));
    }

779 780 781
    return ret;
}

782

783 784
static int
pciDriverDir(char **buffer, const char *driver)
785
{
786 787 788 789 790 791 792 793
    VIR_FREE(*buffer);

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

    return 0;
794 795
}

796 797
static int
pciDriverFile(char **buffer, const char *driver, const char *file)
798
{
799 800 801 802 803 804 805 806
    VIR_FREE(*buffer);

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

    return 0;
807
}
808

809 810
static int
pciDeviceFile(char **buffer, const char *device, const char *file)
811
{
812 813 814 815 816 817 818 819
    VIR_FREE(*buffer);

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

    return 0;
820
}
821 822


823
static const char *
824
pciFindStubDriver(void)
825
{
826
    char *drvpath = NULL;
827 828 829
    int probed = 0;

recheck:
830 831 832 833 834 835
    if (pciDriverDir(&drvpath, "pci-stub") < 0) {
        return NULL;
    }

    if (virFileExists(drvpath)) {
        VIR_FREE(drvpath);
836
        return "pci-stub";
837 838 839 840 841 842 843 844
    }

    if (pciDriverDir(&drvpath, "pciback") < 0) {
        return NULL;
    }

    if (virFileExists(drvpath)) {
        VIR_FREE(drvpath);
845
        return "pciback";
846 847 848
    }

    VIR_FREE(drvpath);
849 850 851 852 853 854 855 856 857 858 859 860

    if (!probed) {
        const char *const stubprobe[] = { MODPROBE, "pci-stub", NULL };
        const char *const backprobe[] = { MODPROBE, "pciback", NULL };

        probed = 1;
        /*
         * Probing for pci-stub will succeed regardless of whether
         * on native or Xen kernels.
         * On Xen though, we want to prefer pciback, so probe
         * for that first, because that will only work on Xen
         */
861 862
        if (virRun(backprobe, NULL) < 0 &&
            virRun(stubprobe, NULL) < 0) {
863
            char ebuf[1024];
864
            VIR_WARN("failed to load pci-stub or pciback drivers: %s",
865
                     virStrerror(errno, ebuf, sizeof ebuf));
866
            return NULL;
867
        }
868 869

        goto recheck;
870 871
    }

872 873 874
    return NULL;
}

875 876 877 878 879 880 881
static int
pciUnbindDeviceFromStub(pciDevice *dev, const char *driver)
{
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;

E
Eric Blake 已提交
882 883 884
    if (pciDriverDir(&drvdir, driver) < 0)
        goto cleanup;

885 886 887
    if (!dev->unbind_from_stub)
        goto remove_slot;

888 889
    /* If the device is bound to stub, unbind it.
     */
E
Eric Blake 已提交
890
    if (pciDeviceFile(&path, dev->name, "driver") < 0)
891 892 893 894 895 896 897 898 899
        goto cleanup;

    if (virFileExists(drvdir) && virFileLinkPointsTo(path, drvdir)) {
        if (pciDriverFile(&path, driver, "unbind") < 0) {
            goto cleanup;
        }

        if (virFileWriteStr(path, dev->name, 0) < 0) {
            virReportSystemError(errno,
900
                                 _("Failed to unbind PCI device '%s' from %s"),
901 902 903 904
                                 dev->name, driver);
            goto cleanup;
        }
    }
905 906 907 908 909
    dev->unbind_from_stub = 0;

remove_slot:
    if (!dev->remove_slot)
        goto reprobe;
910 911 912 913 914 915 916 917

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

    if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
        virReportSystemError(errno,
918
                             _("Failed to remove slot for PCI device '%s' from %s"),
919 920 921
                             dev->name, driver);
        goto cleanup;
    }
922 923 924 925 926 927 928
    dev->remove_slot = 0;

reprobe:
    if (!dev->reprobe) {
        result = 0;
        goto cleanup;
    }
929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950

    /* 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.
     */
    if (pciDriverFile(&path, driver, "remove_id") < 0) {
        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:
951 952 953 954 955
    /* do not do it again */
    dev->unbind_from_stub = 0;
    dev->remove_slot = 0;
    dev->reprobe = 0;

956 957 958 959 960 961
    VIR_FREE(drvdir);
    VIR_FREE(path);

    return result;
}

962 963

static int
964
pciBindDeviceToStub(pciDevice *dev, const char *driver)
965
{
966 967 968
    int result = -1;
    char *drvdir = NULL;
    char *path = NULL;
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
    int reprobe = 0;

    /* check whether the device is already bound to a driver */
    if (pciDriverDir(&drvdir, driver) < 0 ||
        pciDeviceFile(&path, dev->name, "driver") < 0) {
        goto cleanup;
    }

    if (virFileExists(path)) {
        if (virFileLinkPointsTo(path, drvdir)) {
            /* The device is already bound to pci-stub */
            result = 0;
            goto cleanup;
        }
        reprobe = 1;
    }
985 986 987 988 989 990 991 992 993

    /* 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.
     */
994
    if (pciDriverFile(&path, driver, "new_id") < 0) {
995
        goto cleanup;
996 997
    }

998
    if (virFileWriteStr(path, dev->id, 0) < 0) {
999
        virReportSystemError(errno,
1000 1001
                             _("Failed to add PCI device ID '%s' to %s"),
                             dev->id, driver);
1002
        goto cleanup;
1003 1004
    }

1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
    /* check whether the device is bound to pci-stub when we write dev->id to
     * new_id.
     */
    if (pciDriverDir(&drvdir, driver) < 0 ||
        pciDeviceFile(&path, dev->name, "driver") < 0) {
        goto remove_id;
    }

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

1019 1020 1021 1022 1023
    /* 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.
     */
1024 1025 1026 1027
    if (pciDeviceFile(&path, dev->name, "driver/unbind") < 0) {
        goto cleanup;
    }

1028 1029 1030 1031 1032 1033 1034 1035
    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;
1036 1037
    }

1038 1039
    /* If the device isn't already bound to pci-stub, try binding it now.
     */
1040 1041
    if (pciDriverDir(&drvdir, driver) < 0 ||
        pciDeviceFile(&path, dev->name, "driver") < 0) {
1042
        goto remove_id;
1043 1044
    }

1045 1046
    if (!virFileLinkPointsTo(path, drvdir)) {
        /* Xen's pciback.ko wants you to use new_slot first */
1047
        if (pciDriverFile(&path, driver, "new_slot") < 0) {
1048
            goto remove_id;
1049 1050
        }

1051
        if (virFileExists(path) && virFileWriteStr(path, dev->name, 0) < 0) {
1052
            virReportSystemError(errno,
1053 1054
                                 _("Failed to add slot for PCI device '%s' to %s"),
                                 dev->name, driver);
1055
            goto remove_id;
1056
        }
1057
        dev->remove_slot = 1;
1058 1059

        if (pciDriverFile(&path, driver, "bind") < 0) {
1060
            goto remove_id;
1061 1062
        }

1063
        if (virFileWriteStr(path, dev->name, 0) < 0) {
1064
            virReportSystemError(errno,
1065 1066
                                 _("Failed to bind PCI device '%s' to %s"),
                                 dev->name, driver);
1067
            goto remove_id;
1068
        }
1069
        dev->unbind_from_stub = 1;
1070 1071
    }

1072
remove_id:
1073 1074 1075
    /* If 'remove_id' exists, remove the device id from pci-stub's dynamic
     * ID table so that 'drivers_probe' works below.
     */
1076
    if (pciDriverFile(&path, driver, "remove_id") < 0) {
E
Eric Blake 已提交
1077
        /* We do not remove PCI ID from pci-stub, and we cannot reprobe it */
1078 1079 1080 1081 1082
        if (dev->reprobe) {
            VIR_WARN("Could not remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
        dev->reprobe = 0;
1083 1084 1085
        goto cleanup;
    }

1086
    if (virFileExists(path) && virFileWriteStr(path, dev->id, 0) < 0) {
1087
        virReportSystemError(errno,
1088 1089
                             _("Failed to remove PCI ID '%s' from %s"),
                             dev->id, driver);
1090

E
Eric Blake 已提交
1091
        /* remove PCI ID from pci-stub failed, and we cannot reprobe it */
1092 1093 1094 1095 1096
        if (dev->reprobe) {
            VIR_WARN("Failed to remove PCI ID '%s' from %s, and the device "
                     "cannot be probed again.", dev->id, driver);
        }
        dev->reprobe = 0;
1097
        goto cleanup;
1098 1099
    }

1100 1101 1102 1103 1104 1105
    result = 0;

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

1106 1107 1108 1109
    if (result < 0) {
        pciUnbindDeviceFromStub(dev, driver);
    }

1110
    return result;
1111 1112 1113
}

int
1114
pciDettachDevice(pciDevice *dev, pciDeviceList *activeDevs)
1115
{
1116
    const char *driver = pciFindStubDriver();
1117
    if (!driver) {
1118
        pciReportError(VIR_ERR_INTERNAL_ERROR, "%s",
1119 1120 1121 1122
                       _("cannot find any PCI stub module"));
        return -1;
    }

1123 1124 1125 1126 1127 1128
    if (activeDevs && pciDeviceListFind(activeDevs, dev)) {
        pciReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Not detaching active device %s"), dev->name);
        return -1;
    }

1129
    return pciBindDeviceToStub(dev, driver);
1130 1131 1132
}

int
1133
pciReAttachDevice(pciDevice *dev, pciDeviceList *activeDevs)
1134
{
1135
    const char *driver = pciFindStubDriver();
1136
    if (!driver) {
1137
        pciReportError(VIR_ERR_INTERNAL_ERROR, "%s",
1138 1139 1140 1141
                       _("cannot find any PCI stub module"));
        return -1;
    }

1142 1143 1144 1145 1146 1147
    if (activeDevs && pciDeviceListFind(activeDevs, dev)) {
        pciReportError(VIR_ERR_INTERNAL_ERROR,
                       _("Not reattaching active device %s"), dev->name);
        return -1;
    }

1148
    return pciUnbindDeviceFromStub(dev, driver);
1149 1150
}

1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184
/* Certain hypervisors (like qemu/kvm) map the PCI bar(s) on
 * the host when doing device passthrough.  This can lead to a race
 * condition where the hypervisor is still cleaning up the device while
 * libvirt is trying to re-attach it to the host device driver.  To avoid
 * this situation, we look through /proc/iomem, and if the hypervisor is
 * still holding onto the bar (denoted by the string in the matcher variable),
 * then we can wait around a bit for that to clear up.
 *
 * A typical /proc/iomem looks like this (snipped for brevity):
 * 00010000-0008efff : System RAM
 * 0008f000-0008ffff : reserved
 * ...
 * 00100000-cc9fcfff : System RAM
 *   00200000-00483d3b : Kernel code
 *   00483d3c-005c88df : Kernel data
 * cc9fd000-ccc71fff : ACPI Non-volatile Storage
 * ...
 * d0200000-d02fffff : PCI Bus #05
 *   d0200000-d021ffff : 0000:05:00.0
 *     d0200000-d021ffff : e1000e
 *   d0220000-d023ffff : 0000:05:00.0
 *     d0220000-d023ffff : e1000e
 * ...
 * f0000000-f0003fff : 0000:00:1b.0
 *   f0000000-f0003fff : kvm_assigned_device
 *
 * Returns 0 if we are clear to continue, and 1 if the hypervisor is still
 * holding onto the resource.
 */
int
pciWaitForDeviceCleanup(pciDevice *dev, const char *matcher)
{
    FILE *fp;
    char line[160];
1185
    char *tmp;
1186
    unsigned long long start, end;
1187
    unsigned int domain, bus, slot, function;
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
    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
         */
1198
        VIR_DEBUG("Failed to open /proc/iomem, trying to continue anyway");
1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214
        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))) {
1215 1216 1217 1218 1219 1220
            /* 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)
1221 1222
                continue;

1223
            if (STRPREFIX(tmp, matcher)) {
1224 1225 1226 1227 1228 1229 1230
                ret = 1;
                break;
            }
        }
        else {
            in_matching_device = 0;

1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
            /* 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')
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
                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
1255
    VIR_FORCE_FCLOSE(fp);
1256 1257 1258 1259

    return ret;
}

1260 1261 1262
static char *
pciReadDeviceID(pciDevice *dev, const char *id_name)
{
1263
    char *path = NULL;
1264 1265
    char *id_str;

1266 1267 1268
    if (pciDeviceFile(&path, dev->name, id_name) < 0) {
        return NULL;
    }
1269 1270

    /* ID string is '0xNNNN\n' ... i.e. 7 bytes */
1271 1272
    if (virFileReadAll(path, 7, &id_str) < 0) {
        VIR_FREE(path);
1273
        return NULL;
1274 1275 1276
    }

    VIR_FREE(path);
1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290

    /* 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;
}

pciDevice *
1291
pciGetDevice(unsigned domain,
1292 1293 1294 1295 1296
             unsigned bus,
             unsigned slot,
             unsigned function)
{
    pciDevice *dev;
E
Eric Blake 已提交
1297 1298
    char *vendor = NULL;
    char *product = NULL;
1299 1300

    if (VIR_ALLOC(dev) < 0) {
1301
        virReportOOMError();
1302 1303 1304
        return NULL;
    }

1305
    dev->fd       = -1;
1306 1307 1308 1309 1310
    dev->domain   = domain;
    dev->bus      = bus;
    dev->slot     = slot;
    dev->function = function;

E
Eric Blake 已提交
1311 1312 1313 1314 1315 1316
    if (snprintf(dev->name, sizeof(dev->name), "%.4x:%.2x:%.2x.%.1x",
                 dev->domain, dev->bus, dev->slot,
                 dev->function) >= sizeof(dev->name)) {
        pciReportError(VIR_ERR_INTERNAL_ERROR,
                       _("dev->name buffer overflow: %.4x:%.2x:%.2x.%.1x"),
                       dev->domain, dev->bus, dev->slot, dev->function);
E
Eric Blake 已提交
1317
        goto error;
E
Eric Blake 已提交
1318 1319 1320 1321
    }
    if (virAsprintf(&dev->path, PCI_SYSFS "devices/%s/config",
                    dev->name) < 0) {
        virReportOOMError();
E
Eric Blake 已提交
1322
        goto error;
E
Eric Blake 已提交
1323
    }
1324

1325 1326 1327 1328
    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
1329
        goto error;
1330 1331
    }

1332 1333 1334 1335
    vendor  = pciReadDeviceID(dev, "vendor");
    product = pciReadDeviceID(dev, "device");

    if (!vendor || !product) {
1336
        pciReportError(VIR_ERR_INTERNAL_ERROR,
1337 1338
                       _("Failed to read product/vendor ID for %s"),
                       dev->name);
E
Eric Blake 已提交
1339
        goto error;
1340 1341 1342
    }

    /* strings contain '0x' prefix */
E
Eric Blake 已提交
1343 1344 1345 1346 1347
    if (snprintf(dev->id, sizeof(dev->id), "%s %s", &vendor[2],
                 &product[2]) >= sizeof(dev->id)) {
        pciReportError(VIR_ERR_INTERNAL_ERROR,
                       _("dev->id buffer overflow: %s %s"),
                       &vendor[2], &product[2]);
E
Eric Blake 已提交
1348
        goto error;
E
Eric Blake 已提交
1349
    }
1350 1351 1352

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

E
Eric Blake 已提交
1353 1354 1355
cleanup:
    VIR_FREE(product);
    VIR_FREE(vendor);
1356
    return dev;
E
Eric Blake 已提交
1357 1358 1359 1360 1361

error:
    pciFreeDevice(dev);
    dev = NULL;
    goto cleanup;
1362 1363 1364
}

void
1365
pciFreeDevice(pciDevice *dev)
1366
{
1367 1368
    if (!dev)
        return;
1369
    VIR_DEBUG("%s %s: freeing", dev->id, dev->name);
1370
    pciCloseConfig(dev);
E
Eric Blake 已提交
1371
    VIR_FREE(dev->path);
1372 1373
    VIR_FREE(dev);
}
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384

void pciDeviceSetManaged(pciDevice *dev, unsigned managed)
{
    dev->managed = !!managed;
}

unsigned pciDeviceGetManaged(pciDevice *dev)
{
    return dev->managed;
}

1385 1386 1387 1388 1389 1390 1391 1392
void pciDeviceReAttachInit(pciDevice *pci)
{
    pci->unbind_from_stub = 1;
    pci->remove_slot = 1;
    pci->reprobe = 1;
}


1393
pciDeviceList *
1394
pciDeviceListNew(void)
1395 1396 1397 1398
{
    pciDeviceList *list;

    if (VIR_ALLOC(list) < 0) {
1399
        virReportOOMError();
1400 1401 1402 1403 1404 1405 1406
        return NULL;
    }

    return list;
}

void
1407
pciDeviceListFree(pciDeviceList *list)
1408 1409 1410 1411 1412 1413 1414
{
    int i;

    if (!list)
        return;

    for (i = 0; i < list->count; i++) {
1415
        pciFreeDevice(list->devs[i]);
1416 1417 1418 1419 1420 1421 1422 1423 1424
        list->devs[i] = NULL;
    }

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

int
1425
pciDeviceListAdd(pciDeviceList *list,
1426 1427 1428
                 pciDevice *dev)
{
    if (pciDeviceListFind(list, dev)) {
1429
        pciReportError(VIR_ERR_INTERNAL_ERROR,
1430 1431 1432 1433 1434
                       _("Device %s is already in use"), dev->name);
        return -1;
    }

    if (VIR_REALLOC_N(list->devs, list->count+1) < 0) {
1435
        virReportOOMError();
1436 1437 1438 1439 1440 1441 1442 1443
        return -1;
    }

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

    return 0;
}

1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457
pciDevice *
pciDeviceListGet(pciDeviceList *list,
                 int idx)
{
    if (idx >= list->count)
        return NULL;
    if (idx < 0)
        return NULL;

    return list->devs[idx];
}

int
pciDeviceListCount(pciDeviceList *list)
1458
{
1459 1460 1461 1462
    return list->count;
}

pciDevice *
1463
pciDeviceListSteal(pciDeviceList *list,
1464 1465 1466
                   pciDevice *dev)
{
    pciDevice *ret = NULL;
1467 1468 1469 1470 1471 1472 1473 1474 1475
    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)
            continue;

1476
        ret = list->devs[i];
1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488

        if (i != --list->count)
            memmove(&list->devs[i],
                    &list->devs[i+1],
                    sizeof(*list->devs) * (list->count-i));

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

        break;
    }
1489 1490 1491 1492
    return ret;
}

void
1493
pciDeviceListDel(pciDeviceList *list,
1494 1495
                 pciDevice *dev)
{
1496
    pciDevice *ret = pciDeviceListSteal(list, dev);
1497
    if (ret)
1498
        pciFreeDevice(ret);
1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513
}

pciDevice *
pciDeviceListFind(pciDeviceList *list, pciDevice *dev)
{
    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)
            return list->devs[i];
    return NULL;
}
1514 1515


1516
int pciDeviceFileIterate(pciDevice *dev,
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527
                         pciDeviceFileActor actor,
                         void *opaque)
{
    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) {
1528
        virReportOOMError();
1529 1530 1531 1532
        goto cleanup;
    }

    if (!(dir = opendir(pcidir))) {
1533
        virReportSystemError(errno,
1534 1535 1536 1537 1538 1539
                             _("cannot open %s"), pcidir);
        goto cleanup;
    }

    while ((ent = readdir(dir)) != NULL) {
        /* Device assignment requires:
A
Alex Williamson 已提交
1540 1541
         *   $PCIDIR/config, $PCIDIR/resource, $PCIDIR/resourceNNN,
         *   $PCIDIR/rom, $PCIDIR/reset
1542 1543 1544
         */
        if (STREQ(ent->d_name, "config") ||
            STRPREFIX(ent->d_name, "resource") ||
A
Alex Williamson 已提交
1545 1546
            STREQ(ent->d_name, "rom") ||
            STREQ(ent->d_name, "reset")) {
1547
            if (virAsprintf(&file, "%s/%s", pcidir, ent->d_name) < 0) {
1548
                virReportOOMError();
1549 1550
                goto cleanup;
            }
1551
            if ((actor)(dev, file, opaque) < 0)
1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
                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
1567 1568

static int
1569
pciDeviceDownstreamLacksACS(pciDevice *dev)
J
Jiri Denemark 已提交
1570 1571 1572 1573 1574
{
    uint16_t flags;
    uint16_t ctrl;
    unsigned int pos;

1575
    if (!dev->initted && pciInitDevice(dev) < 0)
J
Jiri Denemark 已提交
1576 1577 1578
        return -1;

    pos = dev->pcie_cap_pos;
1579
    if (!pos || pciRead16(dev, PCI_CLASS_DEVICE) != PCI_CLASS_BRIDGE_PCI)
J
Jiri Denemark 已提交
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602
        return 0;

    flags = pciRead16(dev, pos + PCI_EXP_FLAGS);
    if (((flags & PCI_EXP_FLAGS_TYPE) >> 4) != PCI_EXP_TYPE_DOWNSTREAM)
        return 0;

    pos = pciFindExtendedCapabilityOffset(dev, PCI_EXT_CAP_ID_ACS);
    if (!pos) {
        VIR_DEBUG("%s %s: downstream port lacks ACS", dev->id, dev->name);
        return 1;
    }

    ctrl = pciRead16(dev, pos + PCI_EXT_ACS_CTRL);
    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);
        return 1;
    }

    return 0;
}

static int
1603
pciDeviceIsBehindSwitchLackingACS(pciDevice *dev)
J
Jiri Denemark 已提交
1604 1605 1606
{
    pciDevice *parent;

1607 1608
    if (pciGetParentDevice(dev, &parent) < 0)
        return -1;
1609 1610 1611 1612 1613 1614 1615 1616
    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 {
1617
            pciReportError(VIR_ERR_INTERNAL_ERROR,
1618 1619 1620 1621
                           _("Failed to find parent device for %s"),
                           dev->name);
            return -1;
        }
J
Jiri Denemark 已提交
1622 1623 1624 1625 1626 1627 1628 1629 1630
    }

    /* 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 {
        pciDevice *tmp;
        int acs;
1631
        int ret;
J
Jiri Denemark 已提交
1632

1633
        acs = pciDeviceDownstreamLacksACS(parent);
J
Jiri Denemark 已提交
1634 1635

        if (acs) {
1636
            pciFreeDevice(parent);
J
Jiri Denemark 已提交
1637 1638 1639 1640 1641 1642 1643
            if (acs < 0)
                return -1;
            else
                return 1;
        }

        tmp = parent;
1644
        ret = pciGetParentDevice(parent, &parent);
1645
        pciFreeDevice(tmp);
1646 1647
        if (ret < 0)
            return -1;
J
Jiri Denemark 已提交
1648 1649 1650 1651 1652
    } while (parent);

    return 0;
}

1653
int pciDeviceIsAssignable(pciDevice *dev,
J
Jiri Denemark 已提交
1654 1655 1656 1657 1658 1659 1660 1661 1662
                          int strict_acs_check)
{
    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.
     */

1663
    ret = pciDeviceIsBehindSwitchLackingACS(dev);
J
Jiri Denemark 已提交
1664 1665 1666 1667 1668 1669 1670 1671
    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 {
1672
            pciReportError(VIR_ERR_INTERNAL_ERROR,
J
Jiri Denemark 已提交
1673 1674 1675 1676 1677 1678 1679 1680 1681
                           _("Device %s is behind a switch lacking ACS and "
                             "cannot be assigned"),
                           dev->name);
            return 0;
        }
    }

    return 1;
}