qib_pcie.c 19.0 KB
Newer Older
1 2 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
/*
 * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/pci.h>
#include <linux/io.h>
#include <linux/delay.h>
#include <linux/vmalloc.h>
#include <linux/aer.h>
38
#include <linux/module.h>
39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

#include "qib.h"

/*
 * This file contains PCIe utility routines that are common to the
 * various QLogic InfiniPath adapters
 */

/*
 * Code to adjust PCIe capabilities.
 * To minimize the change footprint, we call it
 * from qib_pcie_params, which every chip-specific
 * file calls, even though this violates some
 * expectations of harmlessness.
 */
54 55
static void qib_tune_pcie_caps(struct qib_devdata *);
static void qib_tune_pcie_coalesce(struct qib_devdata *);
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106

/*
 * Do all the common PCIe setup and initialization.
 * devdata is not yet allocated, and is not allocated until after this
 * routine returns success.  Therefore qib_dev_err() can't be used for error
 * printing.
 */
int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
{
	int ret;

	ret = pci_enable_device(pdev);
	if (ret) {
		/*
		 * This can happen (in theory) iff:
		 * We did a chip reset, and then failed to reprogram the
		 * BAR, or the chip reset due to an internal error.  We then
		 * unloaded the driver and reloaded it.
		 *
		 * Both reset cases set the BAR back to initial state.  For
		 * the latter case, the AER sticky error bit at offset 0x718
		 * should be set, but the Linux kernel doesn't yet know
		 * about that, it appears.  If the original BAR was retained
		 * in the kernel data structures, this may be OK.
		 */
		qib_early_err(&pdev->dev, "pci enable failed: error %d\n",
			      -ret);
		goto done;
	}

	ret = pci_request_regions(pdev, QIB_DRV_NAME);
	if (ret) {
		qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret);
		goto bail;
	}

	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
	if (ret) {
		/*
		 * If the 64 bit setup fails, try 32 bit.  Some systems
		 * do not setup 64 bit maps on systems with 2GB or less
		 * memory installed.
		 */
		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
		if (ret) {
			qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret);
			goto bail;
		}
		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
	} else
		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
107
	if (ret) {
108 109
		qib_early_err(&pdev->dev,
			      "Unable to set DMA consistent mask: %d\n", ret);
110 111
		goto bail;
	}
112 113 114

	pci_set_master(pdev);
	ret = pci_enable_pcie_error_reporting(pdev);
115
	if (ret) {
116 117 118
		qib_early_err(&pdev->dev,
			      "Unable to enable pcie error reporting: %d\n",
			      ret);
119 120
		ret = 0;
	}
121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
	goto done;

bail:
	pci_disable_device(pdev);
	pci_release_regions(pdev);
done:
	return ret;
}

/*
 * Do remaining PCIe setup, once dd is allocated, and save away
 * fields required to re-initialize after a chip reset, or for
 * various other purposes
 */
int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev,
		    const struct pci_device_id *ent)
{
	unsigned long len;
	resource_size_t addr;

	dd->pcidev = pdev;
	pci_set_drvdata(pdev, dd);

	addr = pci_resource_start(pdev, 0);
	len = pci_resource_len(pdev, 0);

#if defined(__powerpc__)
	/* There isn't a generic way to specify writethrough mappings */
	dd->kregbase = __ioremap(addr, len, _PAGE_NO_CACHE | _PAGE_WRITETHRU);
#else
	dd->kregbase = ioremap_nocache(addr, len);
#endif

	if (!dd->kregbase)
		return -ENOMEM;

	dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len);
	dd->physaddr = addr;        /* used for io_remap, etc. */

	/*
	 * Save BARs to rewrite after device reset.  Save all 64 bits of
	 * BAR, just in case.
	 */
	dd->pcibar0 = addr;
	dd->pcibar1 = addr >> 32;
	dd->deviceid = ent->device; /* save for later use */
	dd->vendorid = ent->vendor;

	return 0;
}

/*
 * Do PCIe cleanup, after chip-specific cleanup, etc.  Just prior
 * to releasing the dd memory.
 * void because none of the core pcie cleanup returns are void
 */
void qib_pcie_ddcleanup(struct qib_devdata *dd)
{
	u64 __iomem *base = (void __iomem *) dd->kregbase;

	dd->kregbase = NULL;
	iounmap(base);
	if (dd->piobase)
		iounmap(dd->piobase);
	if (dd->userbase)
		iounmap(dd->userbase);
187 188
	if (dd->piovl15base)
		iounmap(dd->piovl15base);
189 190 191 192 193 194 195 196

	pci_disable_device(dd->pcidev);
	pci_release_regions(dd->pcidev);

	pci_set_drvdata(dd->pcidev, NULL);
}

static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
197
			   struct qib_msix_entry *qib_msix_entry)
198 199
{
	int ret;
200
	int nvec = *msixcnt;
201 202 203
	struct msix_entry *msix_entry;
	int i;

204 205 206 207 208 209
	ret = pci_msix_vec_count(dd->pcidev);
	if (ret < 0)
		goto do_intx;

	nvec = min(nvec, ret);

210 211 212
	/* We can't pass qib_msix_entry array to qib_msix_setup
	 * so use a dummy msix_entry array and copy the allocated
	 * irq back to the qib_msix_entry array. */
213
	msix_entry = kcalloc(nvec, sizeof(*msix_entry), GFP_KERNEL);
214
	if (!msix_entry)
215
		goto do_intx;
216 217

	for (i = 0; i < nvec; i++)
218
		msix_entry[i] = qib_msix_entry[i].msix;
219

220 221 222 223 224 225 226
	ret = pci_enable_msix_range(dd->pcidev, msix_entry, 1, nvec);
	if (ret < 0)
		goto free_msix_entry;
	else
		nvec = ret;

	for (i = 0; i < nvec; i++)
227
		qib_msix_entry[i].msix = msix_entry[i];
228

229
	kfree(msix_entry);
230 231
	*msixcnt = nvec;
	return;
232

233 234
free_msix_entry:
	kfree(msix_entry);
235

236
do_intx:
237 238 239 240
	qib_dev_err(
		dd,
		"pci_enable_msix_range %d vectors failed: %d, falling back to INTx\n",
		nvec, ret);
241 242
	*msixcnt = 0;
	qib_enable_intx(dd->pcidev);
243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
}

/**
 * We save the msi lo and hi values, so we can restore them after
 * chip reset (the kernel PCI infrastructure doesn't yet handle that
 * correctly.
 */
static int qib_msi_setup(struct qib_devdata *dd, int pos)
{
	struct pci_dev *pdev = dd->pcidev;
	u16 control;
	int ret;

	ret = pci_enable_msi(pdev);
	if (ret)
M
Mike Marciniszyn 已提交
258 259 260
		qib_dev_err(dd,
			"pci_enable_msi failed: %d, interrupts may not work\n",
			ret);
261 262 263 264 265 266 267 268 269 270 271 272 273 274 275
	/* continue even if it fails, we may still be OK... */

	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
			      &dd->msi_lo);
	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
			      &dd->msi_hi);
	pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control);
	/* now save the data (vector) info */
	pci_read_config_word(pdev, pos + ((control & PCI_MSI_FLAGS_64BIT)
				    ? 12 : 8),
			     &dd->msi_data);
	return ret;
}

int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent,
276
		    struct qib_msix_entry *entry)
277 278
{
	u16 linkstat, speed;
279
	int pos = 0, ret = 1;
280

281
	if (!pci_is_pcie(dd->pcidev)) {
282 283 284 285 286 287 288
		qib_dev_err(dd, "Can't find PCI Express capability!\n");
		/* set up something... */
		dd->lbus_width = 1;
		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
		goto bail;
	}

289
	pos = dd->pcidev->msix_cap;
290 291 292 293
	if (nent && *nent && pos) {
		qib_msix_setup(dd, pos, nent, entry);
		ret = 0; /* did it, either MSIx or INTx */
	} else {
294
		pos = dd->pcidev->msi_cap;
295 296 297 298 299 300 301 302
		if (pos)
			ret = qib_msi_setup(dd, pos);
		else
			qib_dev_err(dd, "No PCI MSI or MSIx capability!\n");
	}
	if (!pos)
		qib_enable_intx(dd->pcidev);

303
	pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat);
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
	/*
	 * speed is bits 0-3, linkwidth is bits 4-8
	 * no defines for them in headers
	 */
	speed = linkstat & 0xf;
	linkstat >>= 4;
	linkstat &= 0x1f;
	dd->lbus_width = linkstat;

	switch (speed) {
	case 1:
		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
		break;
	case 2:
		dd->lbus_speed = 5000; /* Gen1, 5GHz */
		break;
	default: /* not defined, assume gen1 */
		dd->lbus_speed = 2500;
		break;
	}

	/*
	 * Check against expected pcie width and complain if "wrong"
	 * on first initialization, not afterwards (i.e., reset).
	 */
	if (minw && linkstat < minw)
		qib_dev_err(dd,
			    "PCIe width %u (x%u HCA), performance reduced\n",
			    linkstat, minw);

	qib_tune_pcie_caps(dd);

	qib_tune_pcie_coalesce(dd);

bail:
	/* fill in string, even on errors */
	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
		 "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width);
	return ret;
}

/*
 * Setup pcie interrupt stuff again after a reset.  I'd like to just call
 * pci_enable_msi() again for msi, but when I do that,
 * the MSI enable bit doesn't get set in the command word, and
 * we switch to to a different interrupt vector, which is confusing,
 * so I instead just do it all inline.  Perhaps somehow can tie this
 * into the PCIe hotplug support at some point
 */
int qib_reinit_intr(struct qib_devdata *dd)
{
	int pos;
	u16 control;
	int ret = 0;

	/* If we aren't using MSI, don't restore it */
	if (!dd->msi_lo)
		goto bail;

363
	pos = dd->pcidev->msi_cap;
364
	if (!pos) {
M
Mike Marciniszyn 已提交
365 366
		qib_dev_err(dd,
			"Can't find MSI capability, can't restore MSI settings\n");
367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431
		ret = 0;
		/* nothing special for MSIx, just MSI */
		goto bail;
	}
	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
			       dd->msi_lo);
	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
			       dd->msi_hi);
	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
		control |= PCI_MSI_FLAGS_ENABLE;
		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
				      control);
	}
	/* now rewrite the data (vector) info */
	pci_write_config_word(dd->pcidev, pos +
			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
			      dd->msi_data);
	ret = 1;
bail:
	if (!ret && (dd->flags & QIB_HAS_INTX)) {
		qib_enable_intx(dd->pcidev);
		ret = 1;
	}

	/* and now set the pci master bit again */
	pci_set_master(dd->pcidev);

	return ret;
}

/*
 * Disable msi interrupt if enabled, and clear msi_lo.
 * This is used primarily for the fallback to INTx, but
 * is also used in reinit after reset, and during cleanup.
 */
void qib_nomsi(struct qib_devdata *dd)
{
	dd->msi_lo = 0;
	pci_disable_msi(dd->pcidev);
}

/*
 * Same as qib_nosmi, but for MSIx.
 */
void qib_nomsix(struct qib_devdata *dd)
{
	pci_disable_msix(dd->pcidev);
}

/*
 * Similar to pci_intx(pdev, 1), except that we make sure
 * msi(x) is off.
 */
void qib_enable_intx(struct pci_dev *pdev)
{
	u16 cw, new;
	int pos;

	/* first, turn on INTx */
	pci_read_config_word(pdev, PCI_COMMAND, &cw);
	new = cw & ~PCI_COMMAND_INTX_DISABLE;
	if (new != cw)
		pci_write_config_word(pdev, PCI_COMMAND, new);

432
	pos = pdev->msi_cap;
433 434 435 436 437 438 439
	if (pos) {
		/* then turn off MSI */
		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
		new = cw & ~PCI_MSI_FLAGS_ENABLE;
		if (new != cw)
			pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
	}
440
	pos = pdev->msix_cap;
441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
	if (pos) {
		/* then turn off MSIx */
		pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &cw);
		new = cw & ~PCI_MSIX_FLAGS_ENABLE;
		if (new != cw)
			pci_write_config_word(pdev, pos + PCI_MSIX_FLAGS, new);
	}
}

/*
 * These two routines are helper routines for the device reset code
 * to move all the pcie code out of the chip-specific driver code.
 */
void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline)
{
	pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd);
	pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
	pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
}

void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
{
	int r;
464

465 466 467 468 469 470 471 472 473 474 475 476 477 478
	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
				   dd->pcibar0);
	if (r)
		qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
				   dd->pcibar1);
	if (r)
		qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
	/* now re-enable memory access, and restore cosmetic settings */
	pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd);
	pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
	r = pci_enable_device(dd->pcidev);
	if (r)
M
Mike Marciniszyn 已提交
479 480
		qib_dev_err(dd,
			"pci_enable_device failed after reset: %d\n", r);
481 482 483 484 485 486 487 488 489 490 491 492 493
}


static int qib_pcie_coalesce;
module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");

/*
 * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
 * chipsets.   This is known to be unsafe for some revisions of some
 * of these chipsets, with some BIOS settings, and enabling it on those
 * systems may result in the system crashing, and/or data corruption.
 */
494
static void qib_tune_pcie_coalesce(struct qib_devdata *dd)
495 496 497 498 499 500 501
{
	int r;
	struct pci_dev *parent;
	u16 devid;
	u32 mask, bits, val;

	if (!qib_pcie_coalesce)
502
		return;
503 504 505 506 507

	/* Find out supported and configured values for parent (root) */
	parent = dd->pcidev->bus->self;
	if (parent->bus->parent) {
		qib_devinfo(dd->pcidev, "Parent not root\n");
508
		return;
509
	}
510
	if (!pci_is_pcie(parent))
511
		return;
512
	if (parent->vendor != 0x8086)
513
		return;
514 515 516 517 518 519 520 521 522 523 524 525 526

	/*
	 *  - bit 12: Max_rdcmp_Imt_EN: need to set to 1
	 *  - bit 11: COALESCE_FORCE: need to set to 0
	 *  - bit 10: COALESCE_EN: need to set to 1
	 *  (but limitations on some on some chipsets)
	 *
	 *  On the Intel 5000, 5100, and 7300 chipsets, there is
	 *  also: - bit 25:24: COALESCE_MODE, need to set to 0
	 */
	devid = parent->device;
	if (devid >= 0x25e2 && devid <= 0x25fa) {
		/* 5000 P/V/X/Z */
S
Sergei Shtylyov 已提交
527
		if (parent->revision <= 0xb2)
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
			bits = 1U << 10;
		else
			bits = 7U << 10;
		mask = (3U << 24) | (7U << 10);
	} else if (devid >= 0x65e2 && devid <= 0x65fa) {
		/* 5100 */
		bits = 1U << 10;
		mask = (3U << 24) | (7U << 10);
	} else if (devid >= 0x4021 && devid <= 0x402e) {
		/* 5400 */
		bits = 7U << 10;
		mask = 7U << 10;
	} else if (devid >= 0x3604 && devid <= 0x360a) {
		/* 7300 */
		bits = 7U << 10;
		mask = (3U << 24) | (7U << 10);
	} else {
		/* not one of the chipsets that we know about */
546
		return;
547 548 549 550 551 552 553 554 555 556 557
	}
	pci_read_config_dword(parent, 0x48, &val);
	val &= ~mask;
	val |= bits;
	r = pci_write_config_dword(parent, 0x48, val);
}

/*
 * BIOS may not set PCIe bus-utilization parameters for best performance.
 * Check and optionally adjust them to maximize our throughput.
 */
558
static int qib_pcie_caps;
559
module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
560
MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
561

562
static void qib_tune_pcie_caps(struct qib_devdata *dd)
563 564
{
	struct pci_dev *parent;
565 566
	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
	u16 rc_mrrs, ep_mrrs, max_mrrs;
567 568 569

	/* Find out supported and configured values for parent (root) */
	parent = dd->pcidev->bus->self;
570
	if (!pci_is_root_bus(parent->bus)) {
571
		qib_devinfo(dd->pcidev, "Parent not root\n");
572
		return;
573
	}
574 575

	if (!pci_is_pcie(parent) || !pci_is_pcie(dd->pcidev))
576 577
		return;

578 579
	rc_mpss = parent->pcie_mpss;
	rc_mps = ffs(pcie_get_mps(parent)) - 8;
580
	/* Find out supported and configured values for endpoint (us) */
581 582
	ep_mpss = dd->pcidev->pcie_mpss;
	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
583

584
	/* Find max payload supported by root, endpoint */
585 586
	if (rc_mpss > ep_mpss)
		rc_mpss = ep_mpss;
587 588

	/* If Supported greater than limit in module param, limit it */
589 590
	if (rc_mpss > (qib_pcie_caps & 7))
		rc_mpss = qib_pcie_caps & 7;
591
	/* If less than (allowed, supported), bump root payload */
592 593 594
	if (rc_mpss > rc_mps) {
		rc_mps = rc_mpss;
		pcie_set_mps(parent, 128 << rc_mps);
595 596
	}
	/* If less than (allowed, supported), bump endpoint payload */
597 598 599
	if (rc_mpss > ep_mps) {
		ep_mps = rc_mpss;
		pcie_set_mps(dd->pcidev, 128 << ep_mps);
600 601 602 603 604 605 606
	}

	/*
	 * Now the Read Request size.
	 * No field for max supported, but PCIe spec limits it to 4096,
	 * which is code '5' (log2(4096) - 7)
	 */
607 608 609 610 611 612 613 614 615 616 617
	max_mrrs = 5;
	if (max_mrrs > ((qib_pcie_caps >> 4) & 7))
		max_mrrs = (qib_pcie_caps >> 4) & 7;

	max_mrrs = 128 << max_mrrs;
	rc_mrrs = pcie_get_readrq(parent);
	ep_mrrs = pcie_get_readrq(dd->pcidev);

	if (max_mrrs > rc_mrrs) {
		rc_mrrs = max_mrrs;
		pcie_set_readrq(parent, rc_mrrs);
618
	}
619 620 621
	if (max_mrrs > ep_mrrs) {
		ep_mrrs = max_mrrs;
		pcie_set_readrq(dd->pcidev, ep_mrrs);
622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
	}
}
/* End of PCIe capability tuning */

/*
 * From here through qib_pci_err_handler definition is invoked via
 * PCI error infrastructure, registered via pci
 */
static pci_ers_result_t
qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
{
	struct qib_devdata *dd = pci_get_drvdata(pdev);
	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;

	switch (state) {
	case pci_channel_io_normal:
		qib_devinfo(pdev, "State Normal, ignoring\n");
		break;

	case pci_channel_io_frozen:
		qib_devinfo(pdev, "State Frozen, requesting reset\n");
		pci_disable_device(pdev);
		ret = PCI_ERS_RESULT_NEED_RESET;
		break;

	case pci_channel_io_perm_failure:
		qib_devinfo(pdev, "State Permanent Failure, disabling\n");
		if (dd) {
			/* no more register accesses! */
			dd->flags &= ~QIB_PRESENT;
			qib_disable_after_error(dd);
		}
		 /* else early, or other problem */
		ret =  PCI_ERS_RESULT_DISCONNECT;
		break;

	default: /* shouldn't happen */
		qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n",
			state);
		break;
	}
	return ret;
}

static pci_ers_result_t
qib_pci_mmio_enabled(struct pci_dev *pdev)
{
	u64 words = 0U;
	struct qib_devdata *dd = pci_get_drvdata(pdev);
	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;

	if (dd && dd->pport) {
		words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV);
		if (words == ~0ULL)
			ret = PCI_ERS_RESULT_NEED_RESET;
	}
M
Mike Marciniszyn 已提交
678 679 680
	qib_devinfo(pdev,
		"QIB mmio_enabled function called, read wordscntr %Lx, returning %d\n",
		words, ret);
681 682 683 684 685 686
	return  ret;
}

static pci_ers_result_t
qib_pci_slot_reset(struct pci_dev *pdev)
{
687
	qib_devinfo(pdev, "QIB slot_reset function called, ignored\n");
688 689 690 691 692 693 694 695 696 697 698 699 700 701
	return PCI_ERS_RESULT_CAN_RECOVER;
}

static pci_ers_result_t
qib_pci_link_reset(struct pci_dev *pdev)
{
	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
	return PCI_ERS_RESULT_CAN_RECOVER;
}

static void
qib_pci_resume(struct pci_dev *pdev)
{
	struct qib_devdata *dd = pci_get_drvdata(pdev);
702

703 704 705 706 707 708 709 710 711 712
	qib_devinfo(pdev, "QIB resume function called\n");
	pci_cleanup_aer_uncorrect_error_status(pdev);
	/*
	 * Running jobs will fail, since it's asynchronous
	 * unlike sysfs-requested reset.   Better than
	 * doing nothing.
	 */
	qib_init(dd, 1); /* same as re-init after reset */
}

713
const struct pci_error_handlers qib_pci_err_handler = {
714 715 716 717 718 719
	.error_detected = qib_pci_error_detected,
	.mmio_enabled = qib_pci_mmio_enabled,
	.link_reset = qib_pci_link_reset,
	.slot_reset = qib_pci_slot_reset,
	.resume = qib_pci_resume,
};