blob: c926bf4541df3fe0a5d8a66e66305411403ccced [file] [log] [blame]
Ralph Campbellf9315512010-05-23 21:44:54 -07001/*
2 * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33#include <linux/pci.h>
34#include <linux/io.h>
35#include <linux/delay.h>
36#include <linux/vmalloc.h>
37#include <linux/aer.h>
38
39#include "qib.h"
40
41/*
42 * This file contains PCIe utility routines that are common to the
43 * various QLogic InfiniPath adapters
44 */
45
46/*
47 * Code to adjust PCIe capabilities.
48 * To minimize the change footprint, we call it
49 * from qib_pcie_params, which every chip-specific
50 * file calls, even though this violates some
51 * expectations of harmlessness.
52 */
53static int qib_tune_pcie_caps(struct qib_devdata *);
54static int qib_tune_pcie_coalesce(struct qib_devdata *);
55
56/*
57 * Do all the common PCIe setup and initialization.
58 * devdata is not yet allocated, and is not allocated until after this
59 * routine returns success. Therefore qib_dev_err() can't be used for error
60 * printing.
61 */
62int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
63{
64 int ret;
65
66 ret = pci_enable_device(pdev);
67 if (ret) {
68 /*
69 * This can happen (in theory) iff:
70 * We did a chip reset, and then failed to reprogram the
71 * BAR, or the chip reset due to an internal error. We then
72 * unloaded the driver and reloaded it.
73 *
74 * Both reset cases set the BAR back to initial state. For
75 * the latter case, the AER sticky error bit at offset 0x718
76 * should be set, but the Linux kernel doesn't yet know
77 * about that, it appears. If the original BAR was retained
78 * in the kernel data structures, this may be OK.
79 */
80 qib_early_err(&pdev->dev, "pci enable failed: error %d\n",
81 -ret);
82 goto done;
83 }
84
85 ret = pci_request_regions(pdev, QIB_DRV_NAME);
86 if (ret) {
87 qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret);
88 goto bail;
89 }
90
91 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
92 if (ret) {
93 /*
94 * If the 64 bit setup fails, try 32 bit. Some systems
95 * do not setup 64 bit maps on systems with 2GB or less
96 * memory installed.
97 */
98 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
99 if (ret) {
100 qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret);
101 goto bail;
102 }
103 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
104 } else
105 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
106 if (ret)
107 qib_early_err(&pdev->dev,
108 "Unable to set DMA consistent mask: %d\n", ret);
109
110 pci_set_master(pdev);
111 ret = pci_enable_pcie_error_reporting(pdev);
112 if (ret)
113 qib_early_err(&pdev->dev,
114 "Unable to enable pcie error reporting: %d\n",
115 ret);
116 goto done;
117
118bail:
119 pci_disable_device(pdev);
120 pci_release_regions(pdev);
121done:
122 return ret;
123}
124
125/*
126 * Do remaining PCIe setup, once dd is allocated, and save away
127 * fields required to re-initialize after a chip reset, or for
128 * various other purposes
129 */
130int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev,
131 const struct pci_device_id *ent)
132{
133 unsigned long len;
134 resource_size_t addr;
135
136 dd->pcidev = pdev;
137 pci_set_drvdata(pdev, dd);
138
139 addr = pci_resource_start(pdev, 0);
140 len = pci_resource_len(pdev, 0);
141
142#if defined(__powerpc__)
143 /* There isn't a generic way to specify writethrough mappings */
144 dd->kregbase = __ioremap(addr, len, _PAGE_NO_CACHE | _PAGE_WRITETHRU);
145#else
146 dd->kregbase = ioremap_nocache(addr, len);
147#endif
148
149 if (!dd->kregbase)
150 return -ENOMEM;
151
152 dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len);
153 dd->physaddr = addr; /* used for io_remap, etc. */
154
155 /*
156 * Save BARs to rewrite after device reset. Save all 64 bits of
157 * BAR, just in case.
158 */
159 dd->pcibar0 = addr;
160 dd->pcibar1 = addr >> 32;
161 dd->deviceid = ent->device; /* save for later use */
162 dd->vendorid = ent->vendor;
163
164 return 0;
165}
166
167/*
168 * Do PCIe cleanup, after chip-specific cleanup, etc. Just prior
169 * to releasing the dd memory.
170 * void because none of the core pcie cleanup returns are void
171 */
172void qib_pcie_ddcleanup(struct qib_devdata *dd)
173{
174 u64 __iomem *base = (void __iomem *) dd->kregbase;
175
176 dd->kregbase = NULL;
177 iounmap(base);
178 if (dd->piobase)
179 iounmap(dd->piobase);
180 if (dd->userbase)
181 iounmap(dd->userbase);
182
183 pci_disable_device(dd->pcidev);
184 pci_release_regions(dd->pcidev);
185
186 pci_set_drvdata(dd->pcidev, NULL);
187}
188
189static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
190 struct msix_entry *msix_entry)
191{
192 int ret;
193 u32 tabsize = 0;
194 u16 msix_flags;
195
196 pci_read_config_word(dd->pcidev, pos + PCI_MSIX_FLAGS, &msix_flags);
197 tabsize = 1 + (msix_flags & PCI_MSIX_FLAGS_QSIZE);
198 if (tabsize > *msixcnt)
199 tabsize = *msixcnt;
200 ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
201 if (ret > 0) {
202 tabsize = ret;
203 ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
204 }
205 if (ret) {
206 qib_dev_err(dd, "pci_enable_msix %d vectors failed: %d, "
207 "falling back to INTx\n", tabsize, ret);
208 tabsize = 0;
209 }
210 *msixcnt = tabsize;
211
212 if (ret)
213 qib_enable_intx(dd->pcidev);
214
215}
216
217/**
218 * We save the msi lo and hi values, so we can restore them after
219 * chip reset (the kernel PCI infrastructure doesn't yet handle that
220 * correctly.
221 */
222static int qib_msi_setup(struct qib_devdata *dd, int pos)
223{
224 struct pci_dev *pdev = dd->pcidev;
225 u16 control;
226 int ret;
227
228 ret = pci_enable_msi(pdev);
229 if (ret)
230 qib_dev_err(dd, "pci_enable_msi failed: %d, "
231 "interrupts may not work\n", ret);
232 /* continue even if it fails, we may still be OK... */
233
234 pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
235 &dd->msi_lo);
236 pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
237 &dd->msi_hi);
238 pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control);
239 /* now save the data (vector) info */
240 pci_read_config_word(pdev, pos + ((control & PCI_MSI_FLAGS_64BIT)
241 ? 12 : 8),
242 &dd->msi_data);
243 return ret;
244}
245
246int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent,
247 struct msix_entry *entry)
248{
249 u16 linkstat, speed;
250 int pos = 0, pose, ret = 1;
251
252 pose = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
253 if (!pose) {
254 qib_dev_err(dd, "Can't find PCI Express capability!\n");
255 /* set up something... */
256 dd->lbus_width = 1;
257 dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
258 goto bail;
259 }
260
261 pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSIX);
262 if (nent && *nent && pos) {
263 qib_msix_setup(dd, pos, nent, entry);
264 ret = 0; /* did it, either MSIx or INTx */
265 } else {
266 pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
267 if (pos)
268 ret = qib_msi_setup(dd, pos);
269 else
270 qib_dev_err(dd, "No PCI MSI or MSIx capability!\n");
271 }
272 if (!pos)
273 qib_enable_intx(dd->pcidev);
274
275 pci_read_config_word(dd->pcidev, pose + PCI_EXP_LNKSTA, &linkstat);
276 /*
277 * speed is bits 0-3, linkwidth is bits 4-8
278 * no defines for them in headers
279 */
280 speed = linkstat & 0xf;
281 linkstat >>= 4;
282 linkstat &= 0x1f;
283 dd->lbus_width = linkstat;
284
285 switch (speed) {
286 case 1:
287 dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
288 break;
289 case 2:
290 dd->lbus_speed = 5000; /* Gen1, 5GHz */
291 break;
292 default: /* not defined, assume gen1 */
293 dd->lbus_speed = 2500;
294 break;
295 }
296
297 /*
298 * Check against expected pcie width and complain if "wrong"
299 * on first initialization, not afterwards (i.e., reset).
300 */
301 if (minw && linkstat < minw)
302 qib_dev_err(dd,
303 "PCIe width %u (x%u HCA), performance reduced\n",
304 linkstat, minw);
305
306 qib_tune_pcie_caps(dd);
307
308 qib_tune_pcie_coalesce(dd);
309
310bail:
311 /* fill in string, even on errors */
312 snprintf(dd->lbus_info, sizeof(dd->lbus_info),
313 "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width);
314 return ret;
315}
316
317/*
318 * Setup pcie interrupt stuff again after a reset. I'd like to just call
319 * pci_enable_msi() again for msi, but when I do that,
320 * the MSI enable bit doesn't get set in the command word, and
321 * we switch to to a different interrupt vector, which is confusing,
322 * so I instead just do it all inline. Perhaps somehow can tie this
323 * into the PCIe hotplug support at some point
324 */
325int qib_reinit_intr(struct qib_devdata *dd)
326{
327 int pos;
328 u16 control;
329 int ret = 0;
330
331 /* If we aren't using MSI, don't restore it */
332 if (!dd->msi_lo)
333 goto bail;
334
335 pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
336 if (!pos) {
337 qib_dev_err(dd, "Can't find MSI capability, "
338 "can't restore MSI settings\n");
339 ret = 0;
340 /* nothing special for MSIx, just MSI */
341 goto bail;
342 }
343 pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
344 dd->msi_lo);
345 pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
346 dd->msi_hi);
347 pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
348 if (!(control & PCI_MSI_FLAGS_ENABLE)) {
349 control |= PCI_MSI_FLAGS_ENABLE;
350 pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
351 control);
352 }
353 /* now rewrite the data (vector) info */
354 pci_write_config_word(dd->pcidev, pos +
355 ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
356 dd->msi_data);
357 ret = 1;
358bail:
359 if (!ret && (dd->flags & QIB_HAS_INTX)) {
360 qib_enable_intx(dd->pcidev);
361 ret = 1;
362 }
363
364 /* and now set the pci master bit again */
365 pci_set_master(dd->pcidev);
366
367 return ret;
368}
369
370/*
371 * Disable msi interrupt if enabled, and clear msi_lo.
372 * This is used primarily for the fallback to INTx, but
373 * is also used in reinit after reset, and during cleanup.
374 */
375void qib_nomsi(struct qib_devdata *dd)
376{
377 dd->msi_lo = 0;
378 pci_disable_msi(dd->pcidev);
379}
380
381/*
382 * Same as qib_nosmi, but for MSIx.
383 */
384void qib_nomsix(struct qib_devdata *dd)
385{
386 pci_disable_msix(dd->pcidev);
387}
388
389/*
390 * Similar to pci_intx(pdev, 1), except that we make sure
391 * msi(x) is off.
392 */
393void qib_enable_intx(struct pci_dev *pdev)
394{
395 u16 cw, new;
396 int pos;
397
398 /* first, turn on INTx */
399 pci_read_config_word(pdev, PCI_COMMAND, &cw);
400 new = cw & ~PCI_COMMAND_INTX_DISABLE;
401 if (new != cw)
402 pci_write_config_word(pdev, PCI_COMMAND, new);
403
404 pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
405 if (pos) {
406 /* then turn off MSI */
407 pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
408 new = cw & ~PCI_MSI_FLAGS_ENABLE;
409 if (new != cw)
410 pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
411 }
412 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
413 if (pos) {
414 /* then turn off MSIx */
415 pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &cw);
416 new = cw & ~PCI_MSIX_FLAGS_ENABLE;
417 if (new != cw)
418 pci_write_config_word(pdev, pos + PCI_MSIX_FLAGS, new);
419 }
420}
421
422/*
423 * These two routines are helper routines for the device reset code
424 * to move all the pcie code out of the chip-specific driver code.
425 */
426void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline)
427{
428 pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd);
429 pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
430 pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
431}
432
433void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
434{
435 int r;
436 r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
437 dd->pcibar0);
438 if (r)
439 qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
440 r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
441 dd->pcibar1);
442 if (r)
443 qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
444 /* now re-enable memory access, and restore cosmetic settings */
445 pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd);
446 pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
447 pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
448 r = pci_enable_device(dd->pcidev);
449 if (r)
450 qib_dev_err(dd, "pci_enable_device failed after "
451 "reset: %d\n", r);
452}
453
454/* code to adjust PCIe capabilities. */
455
456static int fld2val(int wd, int mask)
457{
458 int lsbmask;
459
460 if (!mask)
461 return 0;
462 wd &= mask;
463 lsbmask = mask ^ (mask & (mask - 1));
464 wd /= lsbmask;
465 return wd;
466}
467
468static int val2fld(int wd, int mask)
469{
470 int lsbmask;
471
472 if (!mask)
473 return 0;
474 lsbmask = mask ^ (mask & (mask - 1));
475 wd *= lsbmask;
476 return wd;
477}
478
479static int qib_pcie_coalesce;
480module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
481MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");
482
483/*
484 * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
485 * chipsets. This is known to be unsafe for some revisions of some
486 * of these chipsets, with some BIOS settings, and enabling it on those
487 * systems may result in the system crashing, and/or data corruption.
488 */
489static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
490{
491 int r;
492 struct pci_dev *parent;
493 int ppos;
494 u16 devid;
495 u32 mask, bits, val;
496
497 if (!qib_pcie_coalesce)
498 return 0;
499
500 /* Find out supported and configured values for parent (root) */
501 parent = dd->pcidev->bus->self;
502 if (parent->bus->parent) {
503 qib_devinfo(dd->pcidev, "Parent not root\n");
504 return 1;
505 }
506 ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
507 if (!ppos)
508 return 1;
509 if (parent->vendor != 0x8086)
510 return 1;
511
512 /*
513 * - bit 12: Max_rdcmp_Imt_EN: need to set to 1
514 * - bit 11: COALESCE_FORCE: need to set to 0
515 * - bit 10: COALESCE_EN: need to set to 1
516 * (but limitations on some on some chipsets)
517 *
518 * On the Intel 5000, 5100, and 7300 chipsets, there is
519 * also: - bit 25:24: COALESCE_MODE, need to set to 0
520 */
521 devid = parent->device;
522 if (devid >= 0x25e2 && devid <= 0x25fa) {
523 u8 rev;
524
525 /* 5000 P/V/X/Z */
526 pci_read_config_byte(parent, PCI_REVISION_ID, &rev);
527 if (rev <= 0xb2)
528 bits = 1U << 10;
529 else
530 bits = 7U << 10;
531 mask = (3U << 24) | (7U << 10);
532 } else if (devid >= 0x65e2 && devid <= 0x65fa) {
533 /* 5100 */
534 bits = 1U << 10;
535 mask = (3U << 24) | (7U << 10);
536 } else if (devid >= 0x4021 && devid <= 0x402e) {
537 /* 5400 */
538 bits = 7U << 10;
539 mask = 7U << 10;
540 } else if (devid >= 0x3604 && devid <= 0x360a) {
541 /* 7300 */
542 bits = 7U << 10;
543 mask = (3U << 24) | (7U << 10);
544 } else {
545 /* not one of the chipsets that we know about */
546 return 1;
547 }
548 pci_read_config_dword(parent, 0x48, &val);
549 val &= ~mask;
550 val |= bits;
551 r = pci_write_config_dword(parent, 0x48, val);
552 return 0;
553}
554
555/*
556 * BIOS may not set PCIe bus-utilization parameters for best performance.
557 * Check and optionally adjust them to maximize our throughput.
558 */
559static int qib_pcie_caps;
560module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
561MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (4lsb), ReadReq (D4..7)");
562
563static int qib_tune_pcie_caps(struct qib_devdata *dd)
564{
565 int ret = 1; /* Assume the worst */
566 struct pci_dev *parent;
567 int ppos, epos;
568 u16 pcaps, pctl, ecaps, ectl;
569 int rc_sup, ep_sup;
570 int rc_cur, ep_cur;
571
572 /* Find out supported and configured values for parent (root) */
573 parent = dd->pcidev->bus->self;
574 if (parent->bus->parent) {
575 qib_devinfo(dd->pcidev, "Parent not root\n");
576 goto bail;
577 }
578 ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
579 if (ppos) {
580 pci_read_config_word(parent, ppos + PCI_EXP_DEVCAP, &pcaps);
581 pci_read_config_word(parent, ppos + PCI_EXP_DEVCTL, &pctl);
582 } else
583 goto bail;
584 /* Find out supported and configured values for endpoint (us) */
585 epos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
586 if (epos) {
587 pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCAP, &ecaps);
588 pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, &ectl);
589 } else
590 goto bail;
591 ret = 0;
592 /* Find max payload supported by root, endpoint */
593 rc_sup = fld2val(pcaps, PCI_EXP_DEVCAP_PAYLOAD);
594 ep_sup = fld2val(ecaps, PCI_EXP_DEVCAP_PAYLOAD);
595 if (rc_sup > ep_sup)
596 rc_sup = ep_sup;
597
598 rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_PAYLOAD);
599 ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_PAYLOAD);
600
601 /* If Supported greater than limit in module param, limit it */
602 if (rc_sup > (qib_pcie_caps & 7))
603 rc_sup = qib_pcie_caps & 7;
604 /* If less than (allowed, supported), bump root payload */
605 if (rc_sup > rc_cur) {
606 rc_cur = rc_sup;
607 pctl = (pctl & ~PCI_EXP_DEVCTL_PAYLOAD) |
608 val2fld(rc_cur, PCI_EXP_DEVCTL_PAYLOAD);
609 pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
610 }
611 /* If less than (allowed, supported), bump endpoint payload */
612 if (rc_sup > ep_cur) {
613 ep_cur = rc_sup;
614 ectl = (ectl & ~PCI_EXP_DEVCTL_PAYLOAD) |
615 val2fld(ep_cur, PCI_EXP_DEVCTL_PAYLOAD);
616 pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
617 }
618
619 /*
620 * Now the Read Request size.
621 * No field for max supported, but PCIe spec limits it to 4096,
622 * which is code '5' (log2(4096) - 7)
623 */
624 rc_sup = 5;
625 if (rc_sup > ((qib_pcie_caps >> 4) & 7))
626 rc_sup = (qib_pcie_caps >> 4) & 7;
627 rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_READRQ);
628 ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_READRQ);
629
630 if (rc_sup > rc_cur) {
631 rc_cur = rc_sup;
632 pctl = (pctl & ~PCI_EXP_DEVCTL_READRQ) |
633 val2fld(rc_cur, PCI_EXP_DEVCTL_READRQ);
634 pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
635 }
636 if (rc_sup > ep_cur) {
637 ep_cur = rc_sup;
638 ectl = (ectl & ~PCI_EXP_DEVCTL_READRQ) |
639 val2fld(ep_cur, PCI_EXP_DEVCTL_READRQ);
640 pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
641 }
642bail:
643 return ret;
644}
645/* End of PCIe capability tuning */
646
647/*
648 * From here through qib_pci_err_handler definition is invoked via
649 * PCI error infrastructure, registered via pci
650 */
651static pci_ers_result_t
652qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
653{
654 struct qib_devdata *dd = pci_get_drvdata(pdev);
655 pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
656
657 switch (state) {
658 case pci_channel_io_normal:
659 qib_devinfo(pdev, "State Normal, ignoring\n");
660 break;
661
662 case pci_channel_io_frozen:
663 qib_devinfo(pdev, "State Frozen, requesting reset\n");
664 pci_disable_device(pdev);
665 ret = PCI_ERS_RESULT_NEED_RESET;
666 break;
667
668 case pci_channel_io_perm_failure:
669 qib_devinfo(pdev, "State Permanent Failure, disabling\n");
670 if (dd) {
671 /* no more register accesses! */
672 dd->flags &= ~QIB_PRESENT;
673 qib_disable_after_error(dd);
674 }
675 /* else early, or other problem */
676 ret = PCI_ERS_RESULT_DISCONNECT;
677 break;
678
679 default: /* shouldn't happen */
680 qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n",
681 state);
682 break;
683 }
684 return ret;
685}
686
687static pci_ers_result_t
688qib_pci_mmio_enabled(struct pci_dev *pdev)
689{
690 u64 words = 0U;
691 struct qib_devdata *dd = pci_get_drvdata(pdev);
692 pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
693
694 if (dd && dd->pport) {
695 words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV);
696 if (words == ~0ULL)
697 ret = PCI_ERS_RESULT_NEED_RESET;
698 }
699 qib_devinfo(pdev, "QIB mmio_enabled function called, "
700 "read wordscntr %Lx, returning %d\n", words, ret);
701 return ret;
702}
703
704static pci_ers_result_t
705qib_pci_slot_reset(struct pci_dev *pdev)
706{
707 qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
708 return PCI_ERS_RESULT_CAN_RECOVER;
709}
710
711static pci_ers_result_t
712qib_pci_link_reset(struct pci_dev *pdev)
713{
714 qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
715 return PCI_ERS_RESULT_CAN_RECOVER;
716}
717
718static void
719qib_pci_resume(struct pci_dev *pdev)
720{
721 struct qib_devdata *dd = pci_get_drvdata(pdev);
722 qib_devinfo(pdev, "QIB resume function called\n");
723 pci_cleanup_aer_uncorrect_error_status(pdev);
724 /*
725 * Running jobs will fail, since it's asynchronous
726 * unlike sysfs-requested reset. Better than
727 * doing nothing.
728 */
729 qib_init(dd, 1); /* same as re-init after reset */
730}
731
732struct pci_error_handlers qib_pci_err_handler = {
733 .error_detected = qib_pci_error_detected,
734 .mmio_enabled = qib_pci_mmio_enabled,
735 .link_reset = qib_pci_link_reset,
736 .slot_reset = qib_pci_slot_reset,
737 .resume = qib_pci_resume,
738};