2 * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/pci.h>
35 #include <linux/delay.h>
36 #include <linux/vmalloc.h>
37 #include <linux/aer.h>
38 #include <linux/module.h>
43 * This file contains PCIe utility routines that are common to the
44 * various QLogic InfiniPath adapters
48 * Code to adjust PCIe capabilities.
49 * To minimize the change footprint, we call it
50 * from qib_pcie_params, which every chip-specific
51 * file calls, even though this violates some
52 * expectations of harmlessness.
54 static int qib_tune_pcie_caps(struct qib_devdata
*);
55 static int qib_tune_pcie_coalesce(struct qib_devdata
*);
58 * Do all the common PCIe setup and initialization.
59 * devdata is not yet allocated, and is not allocated until after this
60 * routine returns success. Therefore qib_dev_err() can't be used for error
63 int qib_pcie_init(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
67 ret
= pci_enable_device(pdev
);
70 * This can happen (in theory) iff:
71 * We did a chip reset, and then failed to reprogram the
72 * BAR, or the chip reset due to an internal error. We then
73 * unloaded the driver and reloaded it.
75 * Both reset cases set the BAR back to initial state. For
76 * the latter case, the AER sticky error bit at offset 0x718
77 * should be set, but the Linux kernel doesn't yet know
78 * about that, it appears. If the original BAR was retained
79 * in the kernel data structures, this may be OK.
81 qib_early_err(&pdev
->dev
, "pci enable failed: error %d\n",
86 ret
= pci_request_regions(pdev
, QIB_DRV_NAME
);
88 qib_devinfo(pdev
, "pci_request_regions fails: err %d\n", -ret
);
92 ret
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
95 * If the 64 bit setup fails, try 32 bit. Some systems
96 * do not setup 64 bit maps on systems with 2GB or less
99 ret
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
101 qib_devinfo(pdev
, "Unable to set DMA mask: %d\n", ret
);
104 ret
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
106 ret
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
108 qib_early_err(&pdev
->dev
,
109 "Unable to set DMA consistent mask: %d\n", ret
);
113 pci_set_master(pdev
);
114 ret
= pci_enable_pcie_error_reporting(pdev
);
116 qib_early_err(&pdev
->dev
,
117 "Unable to enable pcie error reporting: %d\n",
124 pci_disable_device(pdev
);
125 pci_release_regions(pdev
);
131 * Do remaining PCIe setup, once dd is allocated, and save away
132 * fields required to re-initialize after a chip reset, or for
133 * various other purposes
135 int qib_pcie_ddinit(struct qib_devdata
*dd
, struct pci_dev
*pdev
,
136 const struct pci_device_id
*ent
)
139 resource_size_t addr
;
142 pci_set_drvdata(pdev
, dd
);
144 addr
= pci_resource_start(pdev
, 0);
145 len
= pci_resource_len(pdev
, 0);
147 #if defined(__powerpc__)
148 /* There isn't a generic way to specify writethrough mappings */
149 dd
->kregbase
= __ioremap(addr
, len
, _PAGE_NO_CACHE
| _PAGE_WRITETHRU
);
151 dd
->kregbase
= ioremap_nocache(addr
, len
);
157 dd
->kregend
= (u64 __iomem
*)((void __iomem
*) dd
->kregbase
+ len
);
158 dd
->physaddr
= addr
; /* used for io_remap, etc. */
161 * Save BARs to rewrite after device reset. Save all 64 bits of
165 dd
->pcibar1
= addr
>> 32;
166 dd
->deviceid
= ent
->device
; /* save for later use */
167 dd
->vendorid
= ent
->vendor
;
173 * Do PCIe cleanup, after chip-specific cleanup, etc. Just prior
174 * to releasing the dd memory.
175 * void because none of the core pcie cleanup returns are void
177 void qib_pcie_ddcleanup(struct qib_devdata
*dd
)
179 u64 __iomem
*base
= (void __iomem
*) dd
->kregbase
;
184 iounmap(dd
->piobase
);
186 iounmap(dd
->userbase
);
188 iounmap(dd
->piovl15base
);
190 pci_disable_device(dd
->pcidev
);
191 pci_release_regions(dd
->pcidev
);
193 pci_set_drvdata(dd
->pcidev
, NULL
);
196 static void qib_msix_setup(struct qib_devdata
*dd
, int pos
, u32
*msixcnt
,
197 struct msix_entry
*msix_entry
)
203 pci_read_config_word(dd
->pcidev
, pos
+ PCI_MSIX_FLAGS
, &msix_flags
);
204 tabsize
= 1 + (msix_flags
& PCI_MSIX_FLAGS_QSIZE
);
205 if (tabsize
> *msixcnt
)
207 ret
= pci_enable_msix(dd
->pcidev
, msix_entry
, tabsize
);
210 ret
= pci_enable_msix(dd
->pcidev
, msix_entry
, tabsize
);
213 qib_dev_err(dd
, "pci_enable_msix %d vectors failed: %d, "
214 "falling back to INTx\n", tabsize
, ret
);
220 qib_enable_intx(dd
->pcidev
);
225 * We save the msi lo and hi values, so we can restore them after
226 * chip reset (the kernel PCI infrastructure doesn't yet handle that
229 static int qib_msi_setup(struct qib_devdata
*dd
, int pos
)
231 struct pci_dev
*pdev
= dd
->pcidev
;
235 ret
= pci_enable_msi(pdev
);
237 qib_dev_err(dd
, "pci_enable_msi failed: %d, "
238 "interrupts may not work\n", ret
);
239 /* continue even if it fails, we may still be OK... */
241 pci_read_config_dword(pdev
, pos
+ PCI_MSI_ADDRESS_LO
,
243 pci_read_config_dword(pdev
, pos
+ PCI_MSI_ADDRESS_HI
,
245 pci_read_config_word(pdev
, pos
+ PCI_MSI_FLAGS
, &control
);
246 /* now save the data (vector) info */
247 pci_read_config_word(pdev
, pos
+ ((control
& PCI_MSI_FLAGS_64BIT
)
253 int qib_pcie_params(struct qib_devdata
*dd
, u32 minw
, u32
*nent
,
254 struct msix_entry
*entry
)
257 int pos
= 0, pose
, ret
= 1;
259 pose
= pci_pcie_cap(dd
->pcidev
);
261 qib_dev_err(dd
, "Can't find PCI Express capability!\n");
262 /* set up something... */
264 dd
->lbus_speed
= 2500; /* Gen1, 2.5GHz */
268 pos
= pci_find_capability(dd
->pcidev
, PCI_CAP_ID_MSIX
);
269 if (nent
&& *nent
&& pos
) {
270 qib_msix_setup(dd
, pos
, nent
, entry
);
271 ret
= 0; /* did it, either MSIx or INTx */
273 pos
= pci_find_capability(dd
->pcidev
, PCI_CAP_ID_MSI
);
275 ret
= qib_msi_setup(dd
, pos
);
277 qib_dev_err(dd
, "No PCI MSI or MSIx capability!\n");
280 qib_enable_intx(dd
->pcidev
);
282 pci_read_config_word(dd
->pcidev
, pose
+ PCI_EXP_LNKSTA
, &linkstat
);
284 * speed is bits 0-3, linkwidth is bits 4-8
285 * no defines for them in headers
287 speed
= linkstat
& 0xf;
290 dd
->lbus_width
= linkstat
;
294 dd
->lbus_speed
= 2500; /* Gen1, 2.5GHz */
297 dd
->lbus_speed
= 5000; /* Gen1, 5GHz */
299 default: /* not defined, assume gen1 */
300 dd
->lbus_speed
= 2500;
305 * Check against expected pcie width and complain if "wrong"
306 * on first initialization, not afterwards (i.e., reset).
308 if (minw
&& linkstat
< minw
)
310 "PCIe width %u (x%u HCA), performance reduced\n",
313 qib_tune_pcie_caps(dd
);
315 qib_tune_pcie_coalesce(dd
);
318 /* fill in string, even on errors */
319 snprintf(dd
->lbus_info
, sizeof(dd
->lbus_info
),
320 "PCIe,%uMHz,x%u\n", dd
->lbus_speed
, dd
->lbus_width
);
325 * Setup pcie interrupt stuff again after a reset. I'd like to just call
326 * pci_enable_msi() again for msi, but when I do that,
327 * the MSI enable bit doesn't get set in the command word, and
328 * we switch to to a different interrupt vector, which is confusing,
329 * so I instead just do it all inline. Perhaps somehow can tie this
330 * into the PCIe hotplug support at some point
332 int qib_reinit_intr(struct qib_devdata
*dd
)
338 /* If we aren't using MSI, don't restore it */
342 pos
= pci_find_capability(dd
->pcidev
, PCI_CAP_ID_MSI
);
344 qib_dev_err(dd
, "Can't find MSI capability, "
345 "can't restore MSI settings\n");
347 /* nothing special for MSIx, just MSI */
350 pci_write_config_dword(dd
->pcidev
, pos
+ PCI_MSI_ADDRESS_LO
,
352 pci_write_config_dword(dd
->pcidev
, pos
+ PCI_MSI_ADDRESS_HI
,
354 pci_read_config_word(dd
->pcidev
, pos
+ PCI_MSI_FLAGS
, &control
);
355 if (!(control
& PCI_MSI_FLAGS_ENABLE
)) {
356 control
|= PCI_MSI_FLAGS_ENABLE
;
357 pci_write_config_word(dd
->pcidev
, pos
+ PCI_MSI_FLAGS
,
360 /* now rewrite the data (vector) info */
361 pci_write_config_word(dd
->pcidev
, pos
+
362 ((control
& PCI_MSI_FLAGS_64BIT
) ? 12 : 8),
366 if (!ret
&& (dd
->flags
& QIB_HAS_INTX
)) {
367 qib_enable_intx(dd
->pcidev
);
371 /* and now set the pci master bit again */
372 pci_set_master(dd
->pcidev
);
378 * Disable msi interrupt if enabled, and clear msi_lo.
379 * This is used primarily for the fallback to INTx, but
380 * is also used in reinit after reset, and during cleanup.
382 void qib_nomsi(struct qib_devdata
*dd
)
385 pci_disable_msi(dd
->pcidev
);
389 * Same as qib_nosmi, but for MSIx.
391 void qib_nomsix(struct qib_devdata
*dd
)
393 pci_disable_msix(dd
->pcidev
);
397 * Similar to pci_intx(pdev, 1), except that we make sure
400 void qib_enable_intx(struct pci_dev
*pdev
)
405 /* first, turn on INTx */
406 pci_read_config_word(pdev
, PCI_COMMAND
, &cw
);
407 new = cw
& ~PCI_COMMAND_INTX_DISABLE
;
409 pci_write_config_word(pdev
, PCI_COMMAND
, new);
411 pos
= pci_find_capability(pdev
, PCI_CAP_ID_MSI
);
413 /* then turn off MSI */
414 pci_read_config_word(pdev
, pos
+ PCI_MSI_FLAGS
, &cw
);
415 new = cw
& ~PCI_MSI_FLAGS_ENABLE
;
417 pci_write_config_word(pdev
, pos
+ PCI_MSI_FLAGS
, new);
419 pos
= pci_find_capability(pdev
, PCI_CAP_ID_MSIX
);
421 /* then turn off MSIx */
422 pci_read_config_word(pdev
, pos
+ PCI_MSIX_FLAGS
, &cw
);
423 new = cw
& ~PCI_MSIX_FLAGS_ENABLE
;
425 pci_write_config_word(pdev
, pos
+ PCI_MSIX_FLAGS
, new);
430 * These two routines are helper routines for the device reset code
431 * to move all the pcie code out of the chip-specific driver code.
433 void qib_pcie_getcmd(struct qib_devdata
*dd
, u16
*cmd
, u8
*iline
, u8
*cline
)
435 pci_read_config_word(dd
->pcidev
, PCI_COMMAND
, cmd
);
436 pci_read_config_byte(dd
->pcidev
, PCI_INTERRUPT_LINE
, iline
);
437 pci_read_config_byte(dd
->pcidev
, PCI_CACHE_LINE_SIZE
, cline
);
440 void qib_pcie_reenable(struct qib_devdata
*dd
, u16 cmd
, u8 iline
, u8 cline
)
443 r
= pci_write_config_dword(dd
->pcidev
, PCI_BASE_ADDRESS_0
,
446 qib_dev_err(dd
, "rewrite of BAR0 failed: %d\n", r
);
447 r
= pci_write_config_dword(dd
->pcidev
, PCI_BASE_ADDRESS_1
,
450 qib_dev_err(dd
, "rewrite of BAR1 failed: %d\n", r
);
451 /* now re-enable memory access, and restore cosmetic settings */
452 pci_write_config_word(dd
->pcidev
, PCI_COMMAND
, cmd
);
453 pci_write_config_byte(dd
->pcidev
, PCI_INTERRUPT_LINE
, iline
);
454 pci_write_config_byte(dd
->pcidev
, PCI_CACHE_LINE_SIZE
, cline
);
455 r
= pci_enable_device(dd
->pcidev
);
457 qib_dev_err(dd
, "pci_enable_device failed after "
461 /* code to adjust PCIe capabilities. */
463 static int fld2val(int wd
, int mask
)
470 lsbmask
= mask
^ (mask
& (mask
- 1));
475 static int val2fld(int wd
, int mask
)
481 lsbmask
= mask
^ (mask
& (mask
- 1));
486 static int qib_pcie_coalesce
;
487 module_param_named(pcie_coalesce
, qib_pcie_coalesce
, int, S_IRUGO
);
488 MODULE_PARM_DESC(pcie_coalesce
, "tune PCIe colescing on some Intel chipsets");
491 * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
492 * chipsets. This is known to be unsafe for some revisions of some
493 * of these chipsets, with some BIOS settings, and enabling it on those
494 * systems may result in the system crashing, and/or data corruption.
496 static int qib_tune_pcie_coalesce(struct qib_devdata
*dd
)
499 struct pci_dev
*parent
;
504 if (!qib_pcie_coalesce
)
507 /* Find out supported and configured values for parent (root) */
508 parent
= dd
->pcidev
->bus
->self
;
509 if (parent
->bus
->parent
) {
510 qib_devinfo(dd
->pcidev
, "Parent not root\n");
513 ppos
= pci_pcie_cap(parent
);
516 if (parent
->vendor
!= 0x8086)
520 * - bit 12: Max_rdcmp_Imt_EN: need to set to 1
521 * - bit 11: COALESCE_FORCE: need to set to 0
522 * - bit 10: COALESCE_EN: need to set to 1
523 * (but limitations on some on some chipsets)
525 * On the Intel 5000, 5100, and 7300 chipsets, there is
526 * also: - bit 25:24: COALESCE_MODE, need to set to 0
528 devid
= parent
->device
;
529 if (devid
>= 0x25e2 && devid
<= 0x25fa) {
531 if (parent
->revision
<= 0xb2)
535 mask
= (3U << 24) | (7U << 10);
536 } else if (devid
>= 0x65e2 && devid
<= 0x65fa) {
539 mask
= (3U << 24) | (7U << 10);
540 } else if (devid
>= 0x4021 && devid
<= 0x402e) {
544 } else if (devid
>= 0x3604 && devid
<= 0x360a) {
547 mask
= (3U << 24) | (7U << 10);
549 /* not one of the chipsets that we know about */
552 pci_read_config_dword(parent
, 0x48, &val
);
555 r
= pci_write_config_dword(parent
, 0x48, val
);
560 * BIOS may not set PCIe bus-utilization parameters for best performance.
561 * Check and optionally adjust them to maximize our throughput.
563 static int qib_pcie_caps
;
564 module_param_named(pcie_caps
, qib_pcie_caps
, int, S_IRUGO
);
565 MODULE_PARM_DESC(pcie_caps
, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
567 static int qib_tune_pcie_caps(struct qib_devdata
*dd
)
569 int ret
= 1; /* Assume the worst */
570 struct pci_dev
*parent
;
572 u16 pcaps
, pctl
, ecaps
, ectl
;
576 /* Find out supported and configured values for parent (root) */
577 parent
= dd
->pcidev
->bus
->self
;
578 if (parent
->bus
->parent
) {
579 qib_devinfo(dd
->pcidev
, "Parent not root\n");
582 ppos
= pci_pcie_cap(parent
);
584 pci_read_config_word(parent
, ppos
+ PCI_EXP_DEVCAP
, &pcaps
);
585 pci_read_config_word(parent
, ppos
+ PCI_EXP_DEVCTL
, &pctl
);
588 /* Find out supported and configured values for endpoint (us) */
589 epos
= pci_pcie_cap(dd
->pcidev
);
591 pci_read_config_word(dd
->pcidev
, epos
+ PCI_EXP_DEVCAP
, &ecaps
);
592 pci_read_config_word(dd
->pcidev
, epos
+ PCI_EXP_DEVCTL
, &ectl
);
596 /* Find max payload supported by root, endpoint */
597 rc_sup
= fld2val(pcaps
, PCI_EXP_DEVCAP_PAYLOAD
);
598 ep_sup
= fld2val(ecaps
, PCI_EXP_DEVCAP_PAYLOAD
);
602 rc_cur
= fld2val(pctl
, PCI_EXP_DEVCTL_PAYLOAD
);
603 ep_cur
= fld2val(ectl
, PCI_EXP_DEVCTL_PAYLOAD
);
605 /* If Supported greater than limit in module param, limit it */
606 if (rc_sup
> (qib_pcie_caps
& 7))
607 rc_sup
= qib_pcie_caps
& 7;
608 /* If less than (allowed, supported), bump root payload */
609 if (rc_sup
> rc_cur
) {
611 pctl
= (pctl
& ~PCI_EXP_DEVCTL_PAYLOAD
) |
612 val2fld(rc_cur
, PCI_EXP_DEVCTL_PAYLOAD
);
613 pci_write_config_word(parent
, ppos
+ PCI_EXP_DEVCTL
, pctl
);
615 /* If less than (allowed, supported), bump endpoint payload */
616 if (rc_sup
> ep_cur
) {
618 ectl
= (ectl
& ~PCI_EXP_DEVCTL_PAYLOAD
) |
619 val2fld(ep_cur
, PCI_EXP_DEVCTL_PAYLOAD
);
620 pci_write_config_word(dd
->pcidev
, epos
+ PCI_EXP_DEVCTL
, ectl
);
624 * Now the Read Request size.
625 * No field for max supported, but PCIe spec limits it to 4096,
626 * which is code '5' (log2(4096) - 7)
629 if (rc_sup
> ((qib_pcie_caps
>> 4) & 7))
630 rc_sup
= (qib_pcie_caps
>> 4) & 7;
631 rc_cur
= fld2val(pctl
, PCI_EXP_DEVCTL_READRQ
);
632 ep_cur
= fld2val(ectl
, PCI_EXP_DEVCTL_READRQ
);
634 if (rc_sup
> rc_cur
) {
636 pctl
= (pctl
& ~PCI_EXP_DEVCTL_READRQ
) |
637 val2fld(rc_cur
, PCI_EXP_DEVCTL_READRQ
);
638 pci_write_config_word(parent
, ppos
+ PCI_EXP_DEVCTL
, pctl
);
640 if (rc_sup
> ep_cur
) {
642 ectl
= (ectl
& ~PCI_EXP_DEVCTL_READRQ
) |
643 val2fld(ep_cur
, PCI_EXP_DEVCTL_READRQ
);
644 pci_write_config_word(dd
->pcidev
, epos
+ PCI_EXP_DEVCTL
, ectl
);
649 /* End of PCIe capability tuning */
652 * From here through qib_pci_err_handler definition is invoked via
653 * PCI error infrastructure, registered via pci
655 static pci_ers_result_t
656 qib_pci_error_detected(struct pci_dev
*pdev
, pci_channel_state_t state
)
658 struct qib_devdata
*dd
= pci_get_drvdata(pdev
);
659 pci_ers_result_t ret
= PCI_ERS_RESULT_RECOVERED
;
662 case pci_channel_io_normal
:
663 qib_devinfo(pdev
, "State Normal, ignoring\n");
666 case pci_channel_io_frozen
:
667 qib_devinfo(pdev
, "State Frozen, requesting reset\n");
668 pci_disable_device(pdev
);
669 ret
= PCI_ERS_RESULT_NEED_RESET
;
672 case pci_channel_io_perm_failure
:
673 qib_devinfo(pdev
, "State Permanent Failure, disabling\n");
675 /* no more register accesses! */
676 dd
->flags
&= ~QIB_PRESENT
;
677 qib_disable_after_error(dd
);
679 /* else early, or other problem */
680 ret
= PCI_ERS_RESULT_DISCONNECT
;
683 default: /* shouldn't happen */
684 qib_devinfo(pdev
, "QIB PCI errors detected (state %d)\n",
691 static pci_ers_result_t
692 qib_pci_mmio_enabled(struct pci_dev
*pdev
)
695 struct qib_devdata
*dd
= pci_get_drvdata(pdev
);
696 pci_ers_result_t ret
= PCI_ERS_RESULT_RECOVERED
;
698 if (dd
&& dd
->pport
) {
699 words
= dd
->f_portcntr(dd
->pport
, QIBPORTCNTR_WORDRCV
);
701 ret
= PCI_ERS_RESULT_NEED_RESET
;
703 qib_devinfo(pdev
, "QIB mmio_enabled function called, "
704 "read wordscntr %Lx, returning %d\n", words
, ret
);
708 static pci_ers_result_t
709 qib_pci_slot_reset(struct pci_dev
*pdev
)
711 qib_devinfo(pdev
, "QIB link_reset function called, ignored\n");
712 return PCI_ERS_RESULT_CAN_RECOVER
;
715 static pci_ers_result_t
716 qib_pci_link_reset(struct pci_dev
*pdev
)
718 qib_devinfo(pdev
, "QIB link_reset function called, ignored\n");
719 return PCI_ERS_RESULT_CAN_RECOVER
;
723 qib_pci_resume(struct pci_dev
*pdev
)
725 struct qib_devdata
*dd
= pci_get_drvdata(pdev
);
726 qib_devinfo(pdev
, "QIB resume function called\n");
727 pci_cleanup_aer_uncorrect_error_status(pdev
);
729 * Running jobs will fail, since it's asynchronous
730 * unlike sysfs-requested reset. Better than
733 qib_init(dd
, 1); /* same as re-init after reset */
736 struct pci_error_handlers qib_pci_err_handler
= {
737 .error_detected
= qib_pci_error_detected
,
738 .mmio_enabled
= qib_pci_mmio_enabled
,
739 .link_reset
= qib_pci_link_reset
,
740 .slot_reset
= qib_pci_slot_reset
,
741 .resume
= qib_pci_resume
,