xref: /openbmc/linux/drivers/infiniband/hw/hfi1/pcie.c (revision b830f94f)
1 /*
2  * Copyright(c) 2015 - 2019 Intel Corporation.
3  *
4  * This file is provided under a dual BSD/GPLv2 license.  When using or
5  * redistributing this file, you may do so under either license.
6  *
7  * GPL LICENSE SUMMARY
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * BSD LICENSE
19  *
20  * Redistribution and use in source and binary forms, with or without
21  * modification, are permitted provided that the following conditions
22  * are met:
23  *
24  *  - Redistributions of source code must retain the above copyright
25  *    notice, this list of conditions and the following disclaimer.
26  *  - Redistributions in binary form must reproduce the above copyright
27  *    notice, this list of conditions and the following disclaimer in
28  *    the documentation and/or other materials provided with the
29  *    distribution.
30  *  - Neither the name of Intel Corporation nor the names of its
31  *    contributors may be used to endorse or promote products derived
32  *    from this software without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45  *
46  */
47 
48 #include <linux/pci.h>
49 #include <linux/io.h>
50 #include <linux/delay.h>
51 #include <linux/vmalloc.h>
52 #include <linux/aer.h>
53 #include <linux/module.h>
54 
55 #include "hfi.h"
56 #include "chip_registers.h"
57 #include "aspm.h"
58 
59 /*
60  * This file contains PCIe utility routines.
61  */
62 
63 /*
64  * Do all the common PCIe setup and initialization.
65  */
66 int hfi1_pcie_init(struct hfi1_devdata *dd)
67 {
68 	int ret;
69 	struct pci_dev *pdev = dd->pcidev;
70 
71 	ret = pci_enable_device(pdev);
72 	if (ret) {
73 		/*
74 		 * This can happen (in theory) iff:
75 		 * We did a chip reset, and then failed to reprogram the
76 		 * BAR, or the chip reset due to an internal error.  We then
77 		 * unloaded the driver and reloaded it.
78 		 *
79 		 * Both reset cases set the BAR back to initial state.  For
80 		 * the latter case, the AER sticky error bit at offset 0x718
81 		 * should be set, but the Linux kernel doesn't yet know
82 		 * about that, it appears.  If the original BAR was retained
83 		 * in the kernel data structures, this may be OK.
84 		 */
85 		dd_dev_err(dd, "pci enable failed: error %d\n", -ret);
86 		return ret;
87 	}
88 
89 	ret = pci_request_regions(pdev, DRIVER_NAME);
90 	if (ret) {
91 		dd_dev_err(dd, "pci_request_regions fails: err %d\n", -ret);
92 		goto bail;
93 	}
94 
95 	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
96 	if (ret) {
97 		/*
98 		 * If the 64 bit setup fails, try 32 bit.  Some systems
99 		 * do not setup 64 bit maps on systems with 2GB or less
100 		 * memory installed.
101 		 */
102 		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
103 		if (ret) {
104 			dd_dev_err(dd, "Unable to set DMA mask: %d\n", ret);
105 			goto bail;
106 		}
107 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
108 	} else {
109 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
110 	}
111 	if (ret) {
112 		dd_dev_err(dd, "Unable to set DMA consistent mask: %d\n", ret);
113 		goto bail;
114 	}
115 
116 	pci_set_master(pdev);
117 	(void)pci_enable_pcie_error_reporting(pdev);
118 	return 0;
119 
120 bail:
121 	hfi1_pcie_cleanup(pdev);
122 	return ret;
123 }
124 
125 /*
126  * Clean what was done in hfi1_pcie_init()
127  */
128 void hfi1_pcie_cleanup(struct pci_dev *pdev)
129 {
130 	pci_disable_device(pdev);
131 	/*
132 	 * Release regions should be called after the disable. OK to
133 	 * call if request regions has not been called or failed.
134 	 */
135 	pci_release_regions(pdev);
136 }
137 
138 /*
139  * Do remaining PCIe setup, once dd is allocated, and save away
140  * fields required to re-initialize after a chip reset, or for
141  * various other purposes
142  */
143 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev)
144 {
145 	unsigned long len;
146 	resource_size_t addr;
147 	int ret = 0;
148 	u32 rcv_array_count;
149 
150 	addr = pci_resource_start(pdev, 0);
151 	len = pci_resource_len(pdev, 0);
152 
153 	/*
154 	 * The TXE PIO buffers are at the tail end of the chip space.
155 	 * Cut them off and map them separately.
156 	 */
157 
158 	/* sanity check vs expectations */
159 	if (len != TXE_PIO_SEND + TXE_PIO_SIZE) {
160 		dd_dev_err(dd, "chip PIO range does not match\n");
161 		return -EINVAL;
162 	}
163 
164 	dd->kregbase1 = ioremap_nocache(addr, RCV_ARRAY);
165 	if (!dd->kregbase1) {
166 		dd_dev_err(dd, "UC mapping of kregbase1 failed\n");
167 		return -ENOMEM;
168 	}
169 	dd_dev_info(dd, "UC base1: %p for %x\n", dd->kregbase1, RCV_ARRAY);
170 
171 	/* verify that reads actually work, save revision for reset check */
172 	dd->revision = readq(dd->kregbase1 + CCE_REVISION);
173 	if (dd->revision == ~(u64)0) {
174 		dd_dev_err(dd, "Cannot read chip CSRs\n");
175 		goto nomem;
176 	}
177 
178 	rcv_array_count = readq(dd->kregbase1 + RCV_ARRAY_CNT);
179 	dd_dev_info(dd, "RcvArray count: %u\n", rcv_array_count);
180 	dd->base2_start  = RCV_ARRAY + rcv_array_count * 8;
181 
182 	dd->kregbase2 = ioremap_nocache(
183 		addr + dd->base2_start,
184 		TXE_PIO_SEND - dd->base2_start);
185 	if (!dd->kregbase2) {
186 		dd_dev_err(dd, "UC mapping of kregbase2 failed\n");
187 		goto nomem;
188 	}
189 	dd_dev_info(dd, "UC base2: %p for %x\n", dd->kregbase2,
190 		    TXE_PIO_SEND - dd->base2_start);
191 
192 	dd->piobase = ioremap_wc(addr + TXE_PIO_SEND, TXE_PIO_SIZE);
193 	if (!dd->piobase) {
194 		dd_dev_err(dd, "WC mapping of send buffers failed\n");
195 		goto nomem;
196 	}
197 	dd_dev_info(dd, "WC piobase: %p for %x\n", dd->piobase, TXE_PIO_SIZE);
198 
199 	dd->physaddr = addr;        /* used for io_remap, etc. */
200 
201 	/*
202 	 * Map the chip's RcvArray as write-combining to allow us
203 	 * to write an entire cacheline worth of entries in one shot.
204 	 */
205 	dd->rcvarray_wc = ioremap_wc(addr + RCV_ARRAY,
206 				     rcv_array_count * 8);
207 	if (!dd->rcvarray_wc) {
208 		dd_dev_err(dd, "WC mapping of receive array failed\n");
209 		goto nomem;
210 	}
211 	dd_dev_info(dd, "WC RcvArray: %p for %x\n",
212 		    dd->rcvarray_wc, rcv_array_count * 8);
213 
214 	dd->flags |= HFI1_PRESENT;	/* chip.c CSR routines now work */
215 	return 0;
216 nomem:
217 	ret = -ENOMEM;
218 	hfi1_pcie_ddcleanup(dd);
219 	return ret;
220 }
221 
222 /*
223  * Do PCIe cleanup related to dd, after chip-specific cleanup, etc.  Just prior
224  * to releasing the dd memory.
225  * Void because all of the core pcie cleanup functions are void.
226  */
227 void hfi1_pcie_ddcleanup(struct hfi1_devdata *dd)
228 {
229 	dd->flags &= ~HFI1_PRESENT;
230 	if (dd->kregbase1)
231 		iounmap(dd->kregbase1);
232 	dd->kregbase1 = NULL;
233 	if (dd->kregbase2)
234 		iounmap(dd->kregbase2);
235 	dd->kregbase2 = NULL;
236 	if (dd->rcvarray_wc)
237 		iounmap(dd->rcvarray_wc);
238 	dd->rcvarray_wc = NULL;
239 	if (dd->piobase)
240 		iounmap(dd->piobase);
241 	dd->piobase = NULL;
242 }
243 
244 /* return the PCIe link speed from the given link status */
245 static u32 extract_speed(u16 linkstat)
246 {
247 	u32 speed;
248 
249 	switch (linkstat & PCI_EXP_LNKSTA_CLS) {
250 	default: /* not defined, assume Gen1 */
251 	case PCI_EXP_LNKSTA_CLS_2_5GB:
252 		speed = 2500; /* Gen 1, 2.5GHz */
253 		break;
254 	case PCI_EXP_LNKSTA_CLS_5_0GB:
255 		speed = 5000; /* Gen 2, 5GHz */
256 		break;
257 	case PCI_EXP_LNKSTA_CLS_8_0GB:
258 		speed = 8000; /* Gen 3, 8GHz */
259 		break;
260 	}
261 	return speed;
262 }
263 
264 /* return the PCIe link speed from the given link status */
265 static u32 extract_width(u16 linkstat)
266 {
267 	return (linkstat & PCI_EXP_LNKSTA_NLW) >> PCI_EXP_LNKSTA_NLW_SHIFT;
268 }
269 
270 /* read the link status and set dd->{lbus_width,lbus_speed,lbus_info} */
271 static void update_lbus_info(struct hfi1_devdata *dd)
272 {
273 	u16 linkstat;
274 	int ret;
275 
276 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat);
277 	if (ret) {
278 		dd_dev_err(dd, "Unable to read from PCI config\n");
279 		return;
280 	}
281 
282 	dd->lbus_width = extract_width(linkstat);
283 	dd->lbus_speed = extract_speed(linkstat);
284 	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
285 		 "PCIe,%uMHz,x%u", dd->lbus_speed, dd->lbus_width);
286 }
287 
288 /*
289  * Read in the current PCIe link width and speed.  Find if the link is
290  * Gen3 capable.
291  */
292 int pcie_speeds(struct hfi1_devdata *dd)
293 {
294 	u32 linkcap;
295 	struct pci_dev *parent = dd->pcidev->bus->self;
296 	int ret;
297 
298 	if (!pci_is_pcie(dd->pcidev)) {
299 		dd_dev_err(dd, "Can't find PCI Express capability!\n");
300 		return -EINVAL;
301 	}
302 
303 	/* find if our max speed is Gen3 and parent supports Gen3 speeds */
304 	dd->link_gen3_capable = 1;
305 
306 	ret = pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &linkcap);
307 	if (ret) {
308 		dd_dev_err(dd, "Unable to read from PCI config\n");
309 		return ret;
310 	}
311 
312 	if ((linkcap & PCI_EXP_LNKCAP_SLS) != PCI_EXP_LNKCAP_SLS_8_0GB) {
313 		dd_dev_info(dd,
314 			    "This HFI is not Gen3 capable, max speed 0x%x, need 0x3\n",
315 			    linkcap & PCI_EXP_LNKCAP_SLS);
316 		dd->link_gen3_capable = 0;
317 	}
318 
319 	/*
320 	 * bus->max_bus_speed is set from the bridge's linkcap Max Link Speed
321 	 */
322 	if (parent && dd->pcidev->bus->max_bus_speed != PCIE_SPEED_8_0GT) {
323 		dd_dev_info(dd, "Parent PCIe bridge does not support Gen3\n");
324 		dd->link_gen3_capable = 0;
325 	}
326 
327 	/* obtain the link width and current speed */
328 	update_lbus_info(dd);
329 
330 	dd_dev_info(dd, "%s\n", dd->lbus_info);
331 
332 	return 0;
333 }
334 
335 /* restore command and BARs after a reset has wiped them out */
336 int restore_pci_variables(struct hfi1_devdata *dd)
337 {
338 	int ret = 0;
339 
340 	ret = pci_write_config_word(dd->pcidev, PCI_COMMAND, dd->pci_command);
341 	if (ret)
342 		goto error;
343 
344 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
345 				     dd->pcibar0);
346 	if (ret)
347 		goto error;
348 
349 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
350 				     dd->pcibar1);
351 	if (ret)
352 		goto error;
353 
354 	ret = pci_write_config_dword(dd->pcidev, PCI_ROM_ADDRESS, dd->pci_rom);
355 	if (ret)
356 		goto error;
357 
358 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL,
359 					 dd->pcie_devctl);
360 	if (ret)
361 		goto error;
362 
363 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL,
364 					 dd->pcie_lnkctl);
365 	if (ret)
366 		goto error;
367 
368 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL2,
369 					 dd->pcie_devctl2);
370 	if (ret)
371 		goto error;
372 
373 	ret = pci_write_config_dword(dd->pcidev, PCI_CFG_MSIX0, dd->pci_msix0);
374 	if (ret)
375 		goto error;
376 
377 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
378 		ret = pci_write_config_dword(dd->pcidev, PCIE_CFG_TPH2,
379 					     dd->pci_tph2);
380 		if (ret)
381 			goto error;
382 	}
383 	return 0;
384 
385 error:
386 	dd_dev_err(dd, "Unable to write to PCI config\n");
387 	return ret;
388 }
389 
390 /* Save BARs and command to rewrite after device reset */
391 int save_pci_variables(struct hfi1_devdata *dd)
392 {
393 	int ret = 0;
394 
395 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
396 				    &dd->pcibar0);
397 	if (ret)
398 		goto error;
399 
400 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
401 				    &dd->pcibar1);
402 	if (ret)
403 		goto error;
404 
405 	ret = pci_read_config_dword(dd->pcidev, PCI_ROM_ADDRESS, &dd->pci_rom);
406 	if (ret)
407 		goto error;
408 
409 	ret = pci_read_config_word(dd->pcidev, PCI_COMMAND, &dd->pci_command);
410 	if (ret)
411 		goto error;
412 
413 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL,
414 					&dd->pcie_devctl);
415 	if (ret)
416 		goto error;
417 
418 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL,
419 					&dd->pcie_lnkctl);
420 	if (ret)
421 		goto error;
422 
423 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL2,
424 					&dd->pcie_devctl2);
425 	if (ret)
426 		goto error;
427 
428 	ret = pci_read_config_dword(dd->pcidev, PCI_CFG_MSIX0, &dd->pci_msix0);
429 	if (ret)
430 		goto error;
431 
432 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
433 		ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_TPH2,
434 					    &dd->pci_tph2);
435 		if (ret)
436 			goto error;
437 	}
438 	return 0;
439 
440 error:
441 	dd_dev_err(dd, "Unable to read from PCI config\n");
442 	return ret;
443 }
444 
445 /*
446  * BIOS may not set PCIe bus-utilization parameters for best performance.
447  * Check and optionally adjust them to maximize our throughput.
448  */
449 static int hfi1_pcie_caps;
450 module_param_named(pcie_caps, hfi1_pcie_caps, int, 0444);
451 MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
452 
453 /**
454  * tune_pcie_caps() - Code to adjust PCIe capabilities.
455  * @dd: Valid device data structure
456  *
457  */
458 void tune_pcie_caps(struct hfi1_devdata *dd)
459 {
460 	struct pci_dev *parent;
461 	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
462 	u16 rc_mrrs, ep_mrrs, max_mrrs, ectl;
463 	int ret;
464 
465 	/*
466 	 * Turn on extended tags in DevCtl in case the BIOS has turned it off
467 	 * to improve WFR SDMA bandwidth
468 	 */
469 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL, &ectl);
470 	if ((!ret) && !(ectl & PCI_EXP_DEVCTL_EXT_TAG)) {
471 		dd_dev_info(dd, "Enabling PCIe extended tags\n");
472 		ectl |= PCI_EXP_DEVCTL_EXT_TAG;
473 		ret = pcie_capability_write_word(dd->pcidev,
474 						 PCI_EXP_DEVCTL, ectl);
475 		if (ret)
476 			dd_dev_info(dd, "Unable to write to PCI config\n");
477 	}
478 	/* Find out supported and configured values for parent (root) */
479 	parent = dd->pcidev->bus->self;
480 	/*
481 	 * The driver cannot perform the tuning if it does not have
482 	 * access to the upstream component.
483 	 */
484 	if (!parent) {
485 		dd_dev_info(dd, "Parent not found\n");
486 		return;
487 	}
488 	if (!pci_is_root_bus(parent->bus)) {
489 		dd_dev_info(dd, "Parent not root\n");
490 		return;
491 	}
492 	if (!pci_is_pcie(parent)) {
493 		dd_dev_info(dd, "Parent is not PCI Express capable\n");
494 		return;
495 	}
496 	if (!pci_is_pcie(dd->pcidev)) {
497 		dd_dev_info(dd, "PCI device is not PCI Express capable\n");
498 		return;
499 	}
500 	rc_mpss = parent->pcie_mpss;
501 	rc_mps = ffs(pcie_get_mps(parent)) - 8;
502 	/* Find out supported and configured values for endpoint (us) */
503 	ep_mpss = dd->pcidev->pcie_mpss;
504 	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
505 
506 	/* Find max payload supported by root, endpoint */
507 	if (rc_mpss > ep_mpss)
508 		rc_mpss = ep_mpss;
509 
510 	/* If Supported greater than limit in module param, limit it */
511 	if (rc_mpss > (hfi1_pcie_caps & 7))
512 		rc_mpss = hfi1_pcie_caps & 7;
513 	/* If less than (allowed, supported), bump root payload */
514 	if (rc_mpss > rc_mps) {
515 		rc_mps = rc_mpss;
516 		pcie_set_mps(parent, 128 << rc_mps);
517 	}
518 	/* If less than (allowed, supported), bump endpoint payload */
519 	if (rc_mpss > ep_mps) {
520 		ep_mps = rc_mpss;
521 		pcie_set_mps(dd->pcidev, 128 << ep_mps);
522 	}
523 
524 	/*
525 	 * Now the Read Request size.
526 	 * No field for max supported, but PCIe spec limits it to 4096,
527 	 * which is code '5' (log2(4096) - 7)
528 	 */
529 	max_mrrs = 5;
530 	if (max_mrrs > ((hfi1_pcie_caps >> 4) & 7))
531 		max_mrrs = (hfi1_pcie_caps >> 4) & 7;
532 
533 	max_mrrs = 128 << max_mrrs;
534 	rc_mrrs = pcie_get_readrq(parent);
535 	ep_mrrs = pcie_get_readrq(dd->pcidev);
536 
537 	if (max_mrrs > rc_mrrs) {
538 		rc_mrrs = max_mrrs;
539 		pcie_set_readrq(parent, rc_mrrs);
540 	}
541 	if (max_mrrs > ep_mrrs) {
542 		ep_mrrs = max_mrrs;
543 		pcie_set_readrq(dd->pcidev, ep_mrrs);
544 	}
545 }
546 
547 /* End of PCIe capability tuning */
548 
549 /*
550  * From here through hfi1_pci_err_handler definition is invoked via
551  * PCI error infrastructure, registered via pci
552  */
553 static pci_ers_result_t
554 pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
555 {
556 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
557 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
558 
559 	switch (state) {
560 	case pci_channel_io_normal:
561 		dd_dev_info(dd, "State Normal, ignoring\n");
562 		break;
563 
564 	case pci_channel_io_frozen:
565 		dd_dev_info(dd, "State Frozen, requesting reset\n");
566 		pci_disable_device(pdev);
567 		ret = PCI_ERS_RESULT_NEED_RESET;
568 		break;
569 
570 	case pci_channel_io_perm_failure:
571 		if (dd) {
572 			dd_dev_info(dd, "State Permanent Failure, disabling\n");
573 			/* no more register accesses! */
574 			dd->flags &= ~HFI1_PRESENT;
575 			hfi1_disable_after_error(dd);
576 		}
577 		 /* else early, or other problem */
578 		ret =  PCI_ERS_RESULT_DISCONNECT;
579 		break;
580 
581 	default: /* shouldn't happen */
582 		dd_dev_info(dd, "HFI1 PCI errors detected (state %d)\n",
583 			    state);
584 		break;
585 	}
586 	return ret;
587 }
588 
589 static pci_ers_result_t
590 pci_mmio_enabled(struct pci_dev *pdev)
591 {
592 	u64 words = 0U;
593 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
594 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
595 
596 	if (dd && dd->pport) {
597 		words = read_port_cntr(dd->pport, C_RX_WORDS, CNTR_INVALID_VL);
598 		if (words == ~0ULL)
599 			ret = PCI_ERS_RESULT_NEED_RESET;
600 		dd_dev_info(dd,
601 			    "HFI1 mmio_enabled function called, read wordscntr %llx, returning %d\n",
602 			    words, ret);
603 	}
604 	return  ret;
605 }
606 
607 static pci_ers_result_t
608 pci_slot_reset(struct pci_dev *pdev)
609 {
610 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
611 
612 	dd_dev_info(dd, "HFI1 slot_reset function called, ignored\n");
613 	return PCI_ERS_RESULT_CAN_RECOVER;
614 }
615 
616 static void
617 pci_resume(struct pci_dev *pdev)
618 {
619 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
620 
621 	dd_dev_info(dd, "HFI1 resume function called\n");
622 	/*
623 	 * Running jobs will fail, since it's asynchronous
624 	 * unlike sysfs-requested reset.   Better than
625 	 * doing nothing.
626 	 */
627 	hfi1_init(dd, 1); /* same as re-init after reset */
628 }
629 
630 const struct pci_error_handlers hfi1_pci_err_handler = {
631 	.error_detected = pci_error_detected,
632 	.mmio_enabled = pci_mmio_enabled,
633 	.slot_reset = pci_slot_reset,
634 	.resume = pci_resume,
635 };
636 
637 /*============================================================================*/
638 /* PCIe Gen3 support */
639 
640 /*
641  * This code is separated out because it is expected to be removed in the
642  * final shipping product.  If not, then it will be revisited and items
643  * will be moved to more standard locations.
644  */
645 
646 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_STS field values */
647 #define DL_STATUS_HFI0 0x1	/* hfi0 firmware download complete */
648 #define DL_STATUS_HFI1 0x2	/* hfi1 firmware download complete */
649 #define DL_STATUS_BOTH 0x3	/* hfi0 and hfi1 firmware download complete */
650 
651 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_ERR field values */
652 #define DL_ERR_NONE		0x0	/* no error */
653 #define DL_ERR_SWAP_PARITY	0x1	/* parity error in SerDes interrupt */
654 					/*   or response data */
655 #define DL_ERR_DISABLED	0x2	/* hfi disabled */
656 #define DL_ERR_SECURITY	0x3	/* security check failed */
657 #define DL_ERR_SBUS		0x4	/* SBus status error */
658 #define DL_ERR_XFR_PARITY	0x5	/* parity error during ROM transfer*/
659 
660 /* gasket block secondary bus reset delay */
661 #define SBR_DELAY_US 200000	/* 200ms */
662 
663 static uint pcie_target = 3;
664 module_param(pcie_target, uint, S_IRUGO);
665 MODULE_PARM_DESC(pcie_target, "PCIe target speed (0 skip, 1-3 Gen1-3)");
666 
667 static uint pcie_force;
668 module_param(pcie_force, uint, S_IRUGO);
669 MODULE_PARM_DESC(pcie_force, "Force driver to do a PCIe firmware download even if already at target speed");
670 
671 static uint pcie_retry = 5;
672 module_param(pcie_retry, uint, S_IRUGO);
673 MODULE_PARM_DESC(pcie_retry, "Driver will try this many times to reach requested speed");
674 
675 #define UNSET_PSET 255
676 #define DEFAULT_DISCRETE_PSET 2	/* discrete HFI */
677 #define DEFAULT_MCP_PSET 6	/* MCP HFI */
678 static uint pcie_pset = UNSET_PSET;
679 module_param(pcie_pset, uint, S_IRUGO);
680 MODULE_PARM_DESC(pcie_pset, "PCIe Eq Pset value to use, range is 0-10");
681 
682 static uint pcie_ctle = 3; /* discrete on, integrated on */
683 module_param(pcie_ctle, uint, S_IRUGO);
684 MODULE_PARM_DESC(pcie_ctle, "PCIe static CTLE mode, bit 0 - discrete on/off, bit 1 - integrated on/off");
685 
686 /* equalization columns */
687 #define PREC 0
688 #define ATTN 1
689 #define POST 2
690 
691 /* discrete silicon preliminary equalization values */
692 static const u8 discrete_preliminary_eq[11][3] = {
693 	/* prec   attn   post */
694 	{  0x00,  0x00,  0x12 },	/* p0 */
695 	{  0x00,  0x00,  0x0c },	/* p1 */
696 	{  0x00,  0x00,  0x0f },	/* p2 */
697 	{  0x00,  0x00,  0x09 },	/* p3 */
698 	{  0x00,  0x00,  0x00 },	/* p4 */
699 	{  0x06,  0x00,  0x00 },	/* p5 */
700 	{  0x09,  0x00,  0x00 },	/* p6 */
701 	{  0x06,  0x00,  0x0f },	/* p7 */
702 	{  0x09,  0x00,  0x09 },	/* p8 */
703 	{  0x0c,  0x00,  0x00 },	/* p9 */
704 	{  0x00,  0x00,  0x18 },	/* p10 */
705 };
706 
707 /* integrated silicon preliminary equalization values */
708 static const u8 integrated_preliminary_eq[11][3] = {
709 	/* prec   attn   post */
710 	{  0x00,  0x1e,  0x07 },	/* p0 */
711 	{  0x00,  0x1e,  0x05 },	/* p1 */
712 	{  0x00,  0x1e,  0x06 },	/* p2 */
713 	{  0x00,  0x1e,  0x04 },	/* p3 */
714 	{  0x00,  0x1e,  0x00 },	/* p4 */
715 	{  0x03,  0x1e,  0x00 },	/* p5 */
716 	{  0x04,  0x1e,  0x00 },	/* p6 */
717 	{  0x03,  0x1e,  0x06 },	/* p7 */
718 	{  0x03,  0x1e,  0x04 },	/* p8 */
719 	{  0x05,  0x1e,  0x00 },	/* p9 */
720 	{  0x00,  0x1e,  0x0a },	/* p10 */
721 };
722 
723 static const u8 discrete_ctle_tunings[11][4] = {
724 	/* DC     LF     HF     BW */
725 	{  0x48,  0x0b,  0x04,  0x04 },	/* p0 */
726 	{  0x60,  0x05,  0x0f,  0x0a },	/* p1 */
727 	{  0x50,  0x09,  0x06,  0x06 },	/* p2 */
728 	{  0x68,  0x05,  0x0f,  0x0a },	/* p3 */
729 	{  0x80,  0x05,  0x0f,  0x0a },	/* p4 */
730 	{  0x70,  0x05,  0x0f,  0x0a },	/* p5 */
731 	{  0x68,  0x05,  0x0f,  0x0a },	/* p6 */
732 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
733 	{  0x48,  0x09,  0x06,  0x06 },	/* p8 */
734 	{  0x60,  0x05,  0x0f,  0x0a },	/* p9 */
735 	{  0x38,  0x0f,  0x00,  0x00 },	/* p10 */
736 };
737 
738 static const u8 integrated_ctle_tunings[11][4] = {
739 	/* DC     LF     HF     BW */
740 	{  0x38,  0x0f,  0x00,  0x00 },	/* p0 */
741 	{  0x38,  0x0f,  0x00,  0x00 },	/* p1 */
742 	{  0x38,  0x0f,  0x00,  0x00 },	/* p2 */
743 	{  0x38,  0x0f,  0x00,  0x00 },	/* p3 */
744 	{  0x58,  0x0a,  0x05,  0x05 },	/* p4 */
745 	{  0x48,  0x0a,  0x05,  0x05 },	/* p5 */
746 	{  0x40,  0x0a,  0x05,  0x05 },	/* p6 */
747 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
748 	{  0x38,  0x0f,  0x00,  0x00 },	/* p8 */
749 	{  0x38,  0x09,  0x06,  0x06 },	/* p9 */
750 	{  0x38,  0x0e,  0x01,  0x01 },	/* p10 */
751 };
752 
753 /* helper to format the value to write to hardware */
754 #define eq_value(pre, curr, post) \
755 	((((u32)(pre)) << \
756 			PCIE_CFG_REG_PL102_GEN3_EQ_PRE_CURSOR_PSET_SHIFT) \
757 	| (((u32)(curr)) << PCIE_CFG_REG_PL102_GEN3_EQ_CURSOR_PSET_SHIFT) \
758 	| (((u32)(post)) << \
759 		PCIE_CFG_REG_PL102_GEN3_EQ_POST_CURSOR_PSET_SHIFT))
760 
761 /*
762  * Load the given EQ preset table into the PCIe hardware.
763  */
764 static int load_eq_table(struct hfi1_devdata *dd, const u8 eq[11][3], u8 fs,
765 			 u8 div)
766 {
767 	struct pci_dev *pdev = dd->pcidev;
768 	u32 hit_error = 0;
769 	u32 violation;
770 	u32 i;
771 	u8 c_minus1, c0, c_plus1;
772 	int ret;
773 
774 	for (i = 0; i < 11; i++) {
775 		/* set index */
776 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL103, i);
777 		/* write the value */
778 		c_minus1 = eq[i][PREC] / div;
779 		c0 = fs - (eq[i][PREC] / div) - (eq[i][POST] / div);
780 		c_plus1 = eq[i][POST] / div;
781 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL102,
782 				       eq_value(c_minus1, c0, c_plus1));
783 		/* check if these coefficients violate EQ rules */
784 		ret = pci_read_config_dword(dd->pcidev,
785 					    PCIE_CFG_REG_PL105, &violation);
786 		if (ret) {
787 			dd_dev_err(dd, "Unable to read from PCI config\n");
788 			hit_error = 1;
789 			break;
790 		}
791 
792 		if (violation
793 		    & PCIE_CFG_REG_PL105_GEN3_EQ_VIOLATE_COEF_RULES_SMASK){
794 			if (hit_error == 0) {
795 				dd_dev_err(dd,
796 					   "Gen3 EQ Table Coefficient rule violations\n");
797 				dd_dev_err(dd, "         prec   attn   post\n");
798 			}
799 			dd_dev_err(dd, "   p%02d:   %02x     %02x     %02x\n",
800 				   i, (u32)eq[i][0], (u32)eq[i][1],
801 				   (u32)eq[i][2]);
802 			dd_dev_err(dd, "            %02x     %02x     %02x\n",
803 				   (u32)c_minus1, (u32)c0, (u32)c_plus1);
804 			hit_error = 1;
805 		}
806 	}
807 	if (hit_error)
808 		return -EINVAL;
809 	return 0;
810 }
811 
812 /*
813  * Steps to be done after the PCIe firmware is downloaded and
814  * before the SBR for the Pcie Gen3.
815  * The SBus resource is already being held.
816  */
817 static void pcie_post_steps(struct hfi1_devdata *dd)
818 {
819 	int i;
820 
821 	set_sbus_fast_mode(dd);
822 	/*
823 	 * Write to the PCIe PCSes to set the G3_LOCKED_NEXT bits to 1.
824 	 * This avoids a spurious framing error that can otherwise be
825 	 * generated by the MAC layer.
826 	 *
827 	 * Use individual addresses since no broadcast is set up.
828 	 */
829 	for (i = 0; i < NUM_PCIE_SERDES; i++) {
830 		sbus_request(dd, pcie_pcs_addrs[dd->hfi1_id][i],
831 			     0x03, WRITE_SBUS_RECEIVER, 0x00022132);
832 	}
833 
834 	clear_sbus_fast_mode(dd);
835 }
836 
837 /*
838  * Trigger a secondary bus reset (SBR) on ourselves using our parent.
839  *
840  * Based on pci_parent_bus_reset() which is not exported by the
841  * kernel core.
842  */
843 static int trigger_sbr(struct hfi1_devdata *dd)
844 {
845 	struct pci_dev *dev = dd->pcidev;
846 	struct pci_dev *pdev;
847 
848 	/* need a parent */
849 	if (!dev->bus->self) {
850 		dd_dev_err(dd, "%s: no parent device\n", __func__);
851 		return -ENOTTY;
852 	}
853 
854 	/* should not be anyone else on the bus */
855 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
856 		if (pdev != dev) {
857 			dd_dev_err(dd,
858 				   "%s: another device is on the same bus\n",
859 				   __func__);
860 			return -ENOTTY;
861 		}
862 
863 	/*
864 	 * This is an end around to do an SBR during probe time. A new API needs
865 	 * to be implemented to have cleaner interface but this fixes the
866 	 * current brokenness
867 	 */
868 	return pci_bridge_secondary_bus_reset(dev->bus->self);
869 }
870 
871 /*
872  * Write the given gasket interrupt register.
873  */
874 static void write_gasket_interrupt(struct hfi1_devdata *dd, int index,
875 				   u16 code, u16 data)
876 {
877 	write_csr(dd, ASIC_PCIE_SD_INTRPT_LIST + (index * 8),
878 		  (((u64)code << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_CODE_SHIFT) |
879 		   ((u64)data << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_DATA_SHIFT)));
880 }
881 
882 /*
883  * Tell the gasket logic how to react to the reset.
884  */
885 static void arm_gasket_logic(struct hfi1_devdata *dd)
886 {
887 	u64 reg;
888 
889 	reg = (((u64)1 << dd->hfi1_id) <<
890 	       ASIC_PCIE_SD_HOST_CMD_INTRPT_CMD_SHIFT) |
891 	      ((u64)pcie_serdes_broadcast[dd->hfi1_id] <<
892 	       ASIC_PCIE_SD_HOST_CMD_SBUS_RCVR_ADDR_SHIFT |
893 	       ASIC_PCIE_SD_HOST_CMD_SBR_MODE_SMASK |
894 	       ((u64)SBR_DELAY_US & ASIC_PCIE_SD_HOST_CMD_TIMER_MASK) <<
895 	       ASIC_PCIE_SD_HOST_CMD_TIMER_SHIFT);
896 	write_csr(dd, ASIC_PCIE_SD_HOST_CMD, reg);
897 	/* read back to push the write */
898 	read_csr(dd, ASIC_PCIE_SD_HOST_CMD);
899 }
900 
901 /*
902  * CCE_PCIE_CTRL long name helpers
903  * We redefine these shorter macros to use in the code while leaving
904  * chip_registers.h to be autogenerated from the hardware spec.
905  */
906 #define LANE_BUNDLE_MASK              CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_MASK
907 #define LANE_BUNDLE_SHIFT             CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_SHIFT
908 #define LANE_DELAY_MASK               CCE_PCIE_CTRL_PCIE_LANE_DELAY_MASK
909 #define LANE_DELAY_SHIFT              CCE_PCIE_CTRL_PCIE_LANE_DELAY_SHIFT
910 #define MARGIN_OVERWRITE_ENABLE_SHIFT CCE_PCIE_CTRL_XMT_MARGIN_OVERWRITE_ENABLE_SHIFT
911 #define MARGIN_SHIFT                  CCE_PCIE_CTRL_XMT_MARGIN_SHIFT
912 #define MARGIN_G1_G2_OVERWRITE_MASK   CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_MASK
913 #define MARGIN_G1_G2_OVERWRITE_SHIFT  CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_SHIFT
914 #define MARGIN_GEN1_GEN2_MASK         CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_MASK
915 #define MARGIN_GEN1_GEN2_SHIFT        CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_SHIFT
916 
917  /*
918   * Write xmt_margin for full-swing (WFR-B) or half-swing (WFR-C).
919   */
920 static void write_xmt_margin(struct hfi1_devdata *dd, const char *fname)
921 {
922 	u64 pcie_ctrl;
923 	u64 xmt_margin;
924 	u64 xmt_margin_oe;
925 	u64 lane_delay;
926 	u64 lane_bundle;
927 
928 	pcie_ctrl = read_csr(dd, CCE_PCIE_CTRL);
929 
930 	/*
931 	 * For Discrete, use full-swing.
932 	 *  - PCIe TX defaults to full-swing.
933 	 *    Leave this register as default.
934 	 * For Integrated, use half-swing
935 	 *  - Copy xmt_margin and xmt_margin_oe
936 	 *    from Gen1/Gen2 to Gen3.
937 	 */
938 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL1) { /* integrated */
939 		/* extract initial fields */
940 		xmt_margin = (pcie_ctrl >> MARGIN_GEN1_GEN2_SHIFT)
941 			      & MARGIN_GEN1_GEN2_MASK;
942 		xmt_margin_oe = (pcie_ctrl >> MARGIN_G1_G2_OVERWRITE_SHIFT)
943 				 & MARGIN_G1_G2_OVERWRITE_MASK;
944 		lane_delay = (pcie_ctrl >> LANE_DELAY_SHIFT) & LANE_DELAY_MASK;
945 		lane_bundle = (pcie_ctrl >> LANE_BUNDLE_SHIFT)
946 			       & LANE_BUNDLE_MASK;
947 
948 		/*
949 		 * For A0, EFUSE values are not set.  Override with the
950 		 * correct values.
951 		 */
952 		if (is_ax(dd)) {
953 			/*
954 			 * xmt_margin and OverwiteEnabel should be the
955 			 * same for Gen1/Gen2 and Gen3
956 			 */
957 			xmt_margin = 0x5;
958 			xmt_margin_oe = 0x1;
959 			lane_delay = 0xF; /* Delay 240ns. */
960 			lane_bundle = 0x0; /* Set to 1 lane. */
961 		}
962 
963 		/* overwrite existing values */
964 		pcie_ctrl = (xmt_margin << MARGIN_GEN1_GEN2_SHIFT)
965 			| (xmt_margin_oe << MARGIN_G1_G2_OVERWRITE_SHIFT)
966 			| (xmt_margin << MARGIN_SHIFT)
967 			| (xmt_margin_oe << MARGIN_OVERWRITE_ENABLE_SHIFT)
968 			| (lane_delay << LANE_DELAY_SHIFT)
969 			| (lane_bundle << LANE_BUNDLE_SHIFT);
970 
971 		write_csr(dd, CCE_PCIE_CTRL, pcie_ctrl);
972 	}
973 
974 	dd_dev_dbg(dd, "%s: program XMT margin, CcePcieCtrl 0x%llx\n",
975 		   fname, pcie_ctrl);
976 }
977 
978 /*
979  * Do all the steps needed to transition the PCIe link to Gen3 speed.
980  */
981 int do_pcie_gen3_transition(struct hfi1_devdata *dd)
982 {
983 	struct pci_dev *parent = dd->pcidev->bus->self;
984 	u64 fw_ctrl;
985 	u64 reg, therm;
986 	u32 reg32, fs, lf;
987 	u32 status, err;
988 	int ret;
989 	int do_retry, retry_count = 0;
990 	int intnum = 0;
991 	uint default_pset;
992 	uint pset = pcie_pset;
993 	u16 target_vector, target_speed;
994 	u16 lnkctl2, vendor;
995 	u8 div;
996 	const u8 (*eq)[3];
997 	const u8 (*ctle_tunings)[4];
998 	uint static_ctle_mode;
999 	int return_error = 0;
1000 	u32 target_width;
1001 
1002 	/* PCIe Gen3 is for the ASIC only */
1003 	if (dd->icode != ICODE_RTL_SILICON)
1004 		return 0;
1005 
1006 	if (pcie_target == 1) {			/* target Gen1 */
1007 		target_vector = PCI_EXP_LNKCTL2_TLS_2_5GT;
1008 		target_speed = 2500;
1009 	} else if (pcie_target == 2) {		/* target Gen2 */
1010 		target_vector = PCI_EXP_LNKCTL2_TLS_5_0GT;
1011 		target_speed = 5000;
1012 	} else if (pcie_target == 3) {		/* target Gen3 */
1013 		target_vector = PCI_EXP_LNKCTL2_TLS_8_0GT;
1014 		target_speed = 8000;
1015 	} else {
1016 		/* off or invalid target - skip */
1017 		dd_dev_info(dd, "%s: Skipping PCIe transition\n", __func__);
1018 		return 0;
1019 	}
1020 
1021 	/* if already at target speed, done (unless forced) */
1022 	if (dd->lbus_speed == target_speed) {
1023 		dd_dev_info(dd, "%s: PCIe already at gen%d, %s\n", __func__,
1024 			    pcie_target,
1025 			    pcie_force ? "re-doing anyway" : "skipping");
1026 		if (!pcie_force)
1027 			return 0;
1028 	}
1029 
1030 	/*
1031 	 * The driver cannot do the transition if it has no access to the
1032 	 * upstream component
1033 	 */
1034 	if (!parent) {
1035 		dd_dev_info(dd, "%s: No upstream, Can't do gen3 transition\n",
1036 			    __func__);
1037 		return 0;
1038 	}
1039 
1040 	/* Previous Gen1/Gen2 bus width */
1041 	target_width = dd->lbus_width;
1042 
1043 	/*
1044 	 * Do the Gen3 transition.  Steps are those of the PCIe Gen3
1045 	 * recipe.
1046 	 */
1047 
1048 	/* step 1: pcie link working in gen1/gen2 */
1049 
1050 	/* step 2: if either side is not capable of Gen3, done */
1051 	if (pcie_target == 3 && !dd->link_gen3_capable) {
1052 		dd_dev_err(dd, "The PCIe link is not Gen3 capable\n");
1053 		ret = -ENOSYS;
1054 		goto done_no_mutex;
1055 	}
1056 
1057 	/* hold the SBus resource across the firmware download and SBR */
1058 	ret = acquire_chip_resource(dd, CR_SBUS, SBUS_TIMEOUT);
1059 	if (ret) {
1060 		dd_dev_err(dd, "%s: unable to acquire SBus resource\n",
1061 			   __func__);
1062 		return ret;
1063 	}
1064 
1065 	/* make sure thermal polling is not causing interrupts */
1066 	therm = read_csr(dd, ASIC_CFG_THERM_POLL_EN);
1067 	if (therm) {
1068 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x0);
1069 		msleep(100);
1070 		dd_dev_info(dd, "%s: Disabled therm polling\n",
1071 			    __func__);
1072 	}
1073 
1074 retry:
1075 	/* the SBus download will reset the spico for thermal */
1076 
1077 	/* step 3: download SBus Master firmware */
1078 	/* step 4: download PCIe Gen3 SerDes firmware */
1079 	dd_dev_info(dd, "%s: downloading firmware\n", __func__);
1080 	ret = load_pcie_firmware(dd);
1081 	if (ret) {
1082 		/* do not proceed if the firmware cannot be downloaded */
1083 		return_error = 1;
1084 		goto done;
1085 	}
1086 
1087 	/* step 5: set up device parameter settings */
1088 	dd_dev_info(dd, "%s: setting PCIe registers\n", __func__);
1089 
1090 	/*
1091 	 * PcieCfgSpcie1 - Link Control 3
1092 	 * Leave at reset value.  No need to set PerfEq - link equalization
1093 	 * will be performed automatically after the SBR when the target
1094 	 * speed is 8GT/s.
1095 	 */
1096 
1097 	/* clear all 16 per-lane error bits (PCIe: Lane Error Status) */
1098 	pci_write_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, 0xffff);
1099 
1100 	/* step 5a: Set Synopsys Port Logic registers */
1101 
1102 	/*
1103 	 * PcieCfgRegPl2 - Port Force Link
1104 	 *
1105 	 * Set the low power field to 0x10 to avoid unnecessary power
1106 	 * management messages.  All other fields are zero.
1107 	 */
1108 	reg32 = 0x10ul << PCIE_CFG_REG_PL2_LOW_PWR_ENT_CNT_SHIFT;
1109 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL2, reg32);
1110 
1111 	/*
1112 	 * PcieCfgRegPl100 - Gen3 Control
1113 	 *
1114 	 * turn off PcieCfgRegPl100.Gen3ZRxDcNonCompl
1115 	 * turn on PcieCfgRegPl100.EqEieosCnt
1116 	 * Everything else zero.
1117 	 */
1118 	reg32 = PCIE_CFG_REG_PL100_EQ_EIEOS_CNT_SMASK;
1119 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL100, reg32);
1120 
1121 	/*
1122 	 * PcieCfgRegPl101 - Gen3 EQ FS and LF
1123 	 * PcieCfgRegPl102 - Gen3 EQ Presets to Coefficients Mapping
1124 	 * PcieCfgRegPl103 - Gen3 EQ Preset Index
1125 	 * PcieCfgRegPl105 - Gen3 EQ Status
1126 	 *
1127 	 * Give initial EQ settings.
1128 	 */
1129 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL0) { /* discrete */
1130 		/* 1000mV, FS=24, LF = 8 */
1131 		fs = 24;
1132 		lf = 8;
1133 		div = 3;
1134 		eq = discrete_preliminary_eq;
1135 		default_pset = DEFAULT_DISCRETE_PSET;
1136 		ctle_tunings = discrete_ctle_tunings;
1137 		/* bit 0 - discrete on/off */
1138 		static_ctle_mode = pcie_ctle & 0x1;
1139 	} else {
1140 		/* 400mV, FS=29, LF = 9 */
1141 		fs = 29;
1142 		lf = 9;
1143 		div = 1;
1144 		eq = integrated_preliminary_eq;
1145 		default_pset = DEFAULT_MCP_PSET;
1146 		ctle_tunings = integrated_ctle_tunings;
1147 		/* bit 1 - integrated on/off */
1148 		static_ctle_mode = (pcie_ctle >> 1) & 0x1;
1149 	}
1150 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL101,
1151 			       (fs <<
1152 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_FS_SHIFT) |
1153 			       (lf <<
1154 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_LF_SHIFT));
1155 	ret = load_eq_table(dd, eq, fs, div);
1156 	if (ret)
1157 		goto done;
1158 
1159 	/*
1160 	 * PcieCfgRegPl106 - Gen3 EQ Control
1161 	 *
1162 	 * Set Gen3EqPsetReqVec, leave other fields 0.
1163 	 */
1164 	if (pset == UNSET_PSET)
1165 		pset = default_pset;
1166 	if (pset > 10) {	/* valid range is 0-10, inclusive */
1167 		dd_dev_err(dd, "%s: Invalid Eq Pset %u, setting to %d\n",
1168 			   __func__, pset, default_pset);
1169 		pset = default_pset;
1170 	}
1171 	dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pset);
1172 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL106,
1173 			       ((1 << pset) <<
1174 			PCIE_CFG_REG_PL106_GEN3_EQ_PSET_REQ_VEC_SHIFT) |
1175 			PCIE_CFG_REG_PL106_GEN3_EQ_EVAL2MS_DISABLE_SMASK |
1176 			PCIE_CFG_REG_PL106_GEN3_EQ_PHASE23_EXIT_MODE_SMASK);
1177 
1178 	/*
1179 	 * step 5b: Do post firmware download steps via SBus
1180 	 */
1181 	dd_dev_info(dd, "%s: doing pcie post steps\n", __func__);
1182 	pcie_post_steps(dd);
1183 
1184 	/*
1185 	 * step 5c: Program gasket interrupts
1186 	 */
1187 	/* set the Rx Bit Rate to REFCLK ratio */
1188 	write_gasket_interrupt(dd, intnum++, 0x0006, 0x0050);
1189 	/* disable pCal for PCIe Gen3 RX equalization */
1190 	/* select adaptive or static CTLE */
1191 	write_gasket_interrupt(dd, intnum++, 0x0026,
1192 			       0x5b01 | (static_ctle_mode << 3));
1193 	/*
1194 	 * Enable iCal for PCIe Gen3 RX equalization, and set which
1195 	 * evaluation of RX_EQ_EVAL will launch the iCal procedure.
1196 	 */
1197 	write_gasket_interrupt(dd, intnum++, 0x0026, 0x5202);
1198 
1199 	if (static_ctle_mode) {
1200 		/* apply static CTLE tunings */
1201 		u8 pcie_dc, pcie_lf, pcie_hf, pcie_bw;
1202 
1203 		pcie_dc = ctle_tunings[pset][0];
1204 		pcie_lf = ctle_tunings[pset][1];
1205 		pcie_hf = ctle_tunings[pset][2];
1206 		pcie_bw = ctle_tunings[pset][3];
1207 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0200 | pcie_dc);
1208 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0100 | pcie_lf);
1209 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0000 | pcie_hf);
1210 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x5500 | pcie_bw);
1211 	}
1212 
1213 	/* terminate list */
1214 	write_gasket_interrupt(dd, intnum++, 0x0000, 0x0000);
1215 
1216 	/*
1217 	 * step 5d: program XMT margin
1218 	 */
1219 	write_xmt_margin(dd, __func__);
1220 
1221 	/*
1222 	 * step 5e: disable active state power management (ASPM). It
1223 	 * will be enabled if required later
1224 	 */
1225 	dd_dev_info(dd, "%s: clearing ASPM\n", __func__);
1226 	aspm_hw_disable_l1(dd);
1227 
1228 	/*
1229 	 * step 5f: clear DirectSpeedChange
1230 	 * PcieCfgRegPl67.DirectSpeedChange must be zero to prevent the
1231 	 * change in the speed target from starting before we are ready.
1232 	 * This field defaults to 0 and we are not changing it, so nothing
1233 	 * needs to be done.
1234 	 */
1235 
1236 	/* step 5g: Set target link speed */
1237 	/*
1238 	 * Set target link speed to be target on both device and parent.
1239 	 * On setting the parent: Some system BIOSs "helpfully" set the
1240 	 * parent target speed to Gen2 to match the ASIC's initial speed.
1241 	 * We can set the target Gen3 because we have already checked
1242 	 * that it is Gen3 capable earlier.
1243 	 */
1244 	dd_dev_info(dd, "%s: setting parent target link speed\n", __func__);
1245 	ret = pcie_capability_read_word(parent, PCI_EXP_LNKCTL2, &lnkctl2);
1246 	if (ret) {
1247 		dd_dev_err(dd, "Unable to read from PCI config\n");
1248 		return_error = 1;
1249 		goto done;
1250 	}
1251 
1252 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1253 		    (u32)lnkctl2);
1254 	/* only write to parent if target is not as high as ours */
1255 	if ((lnkctl2 & PCI_EXP_LNKCTL2_TLS) < target_vector) {
1256 		lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1257 		lnkctl2 |= target_vector;
1258 		dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1259 			    (u32)lnkctl2);
1260 		ret = pcie_capability_write_word(parent,
1261 						 PCI_EXP_LNKCTL2, lnkctl2);
1262 		if (ret) {
1263 			dd_dev_err(dd, "Unable to write to PCI config\n");
1264 			return_error = 1;
1265 			goto done;
1266 		}
1267 	} else {
1268 		dd_dev_info(dd, "%s: ..target speed is OK\n", __func__);
1269 	}
1270 
1271 	dd_dev_info(dd, "%s: setting target link speed\n", __func__);
1272 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL2, &lnkctl2);
1273 	if (ret) {
1274 		dd_dev_err(dd, "Unable to read from PCI config\n");
1275 		return_error = 1;
1276 		goto done;
1277 	}
1278 
1279 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1280 		    (u32)lnkctl2);
1281 	lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1282 	lnkctl2 |= target_vector;
1283 	dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1284 		    (u32)lnkctl2);
1285 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL2, lnkctl2);
1286 	if (ret) {
1287 		dd_dev_err(dd, "Unable to write to PCI config\n");
1288 		return_error = 1;
1289 		goto done;
1290 	}
1291 
1292 	/* step 5h: arm gasket logic */
1293 	/* hold DC in reset across the SBR */
1294 	write_csr(dd, CCE_DC_CTRL, CCE_DC_CTRL_DC_RESET_SMASK);
1295 	(void)read_csr(dd, CCE_DC_CTRL); /* DC reset hold */
1296 	/* save firmware control across the SBR */
1297 	fw_ctrl = read_csr(dd, MISC_CFG_FW_CTRL);
1298 
1299 	dd_dev_info(dd, "%s: arming gasket logic\n", __func__);
1300 	arm_gasket_logic(dd);
1301 
1302 	/*
1303 	 * step 6: quiesce PCIe link
1304 	 * The chip has already been reset, so there will be no traffic
1305 	 * from the chip.  Linux has no easy way to enforce that it will
1306 	 * not try to access the device, so we just need to hope it doesn't
1307 	 * do it while we are doing the reset.
1308 	 */
1309 
1310 	/*
1311 	 * step 7: initiate the secondary bus reset (SBR)
1312 	 * step 8: hardware brings the links back up
1313 	 * step 9: wait for link speed transition to be complete
1314 	 */
1315 	dd_dev_info(dd, "%s: calling trigger_sbr\n", __func__);
1316 	ret = trigger_sbr(dd);
1317 	if (ret)
1318 		goto done;
1319 
1320 	/* step 10: decide what to do next */
1321 
1322 	/* check if we can read PCI space */
1323 	ret = pci_read_config_word(dd->pcidev, PCI_VENDOR_ID, &vendor);
1324 	if (ret) {
1325 		dd_dev_info(dd,
1326 			    "%s: read of VendorID failed after SBR, err %d\n",
1327 			    __func__, ret);
1328 		return_error = 1;
1329 		goto done;
1330 	}
1331 	if (vendor == 0xffff) {
1332 		dd_dev_info(dd, "%s: VendorID is all 1s after SBR\n", __func__);
1333 		return_error = 1;
1334 		ret = -EIO;
1335 		goto done;
1336 	}
1337 
1338 	/* restore PCI space registers we know were reset */
1339 	dd_dev_info(dd, "%s: calling restore_pci_variables\n", __func__);
1340 	ret = restore_pci_variables(dd);
1341 	if (ret) {
1342 		dd_dev_err(dd, "%s: Could not restore PCI variables\n",
1343 			   __func__);
1344 		return_error = 1;
1345 		goto done;
1346 	}
1347 
1348 	/* restore firmware control */
1349 	write_csr(dd, MISC_CFG_FW_CTRL, fw_ctrl);
1350 
1351 	/*
1352 	 * Check the gasket block status.
1353 	 *
1354 	 * This is the first CSR read after the SBR.  If the read returns
1355 	 * all 1s (fails), the link did not make it back.
1356 	 *
1357 	 * Once we're sure we can read and write, clear the DC reset after
1358 	 * the SBR.  Then check for any per-lane errors. Then look over
1359 	 * the status.
1360 	 */
1361 	reg = read_csr(dd, ASIC_PCIE_SD_HOST_STATUS);
1362 	dd_dev_info(dd, "%s: gasket block status: 0x%llx\n", __func__, reg);
1363 	if (reg == ~0ull) {	/* PCIe read failed/timeout */
1364 		dd_dev_err(dd, "SBR failed - unable to read from device\n");
1365 		return_error = 1;
1366 		ret = -ENOSYS;
1367 		goto done;
1368 	}
1369 
1370 	/* clear the DC reset */
1371 	write_csr(dd, CCE_DC_CTRL, 0);
1372 
1373 	/* Set the LED off */
1374 	setextled(dd, 0);
1375 
1376 	/* check for any per-lane errors */
1377 	ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, &reg32);
1378 	if (ret) {
1379 		dd_dev_err(dd, "Unable to read from PCI config\n");
1380 		return_error = 1;
1381 		goto done;
1382 	}
1383 
1384 	dd_dev_info(dd, "%s: per-lane errors: 0x%x\n", __func__, reg32);
1385 
1386 	/* extract status, look for our HFI */
1387 	status = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_SHIFT)
1388 			& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_MASK;
1389 	if ((status & (1 << dd->hfi1_id)) == 0) {
1390 		dd_dev_err(dd,
1391 			   "%s: gasket status 0x%x, expecting 0x%x\n",
1392 			   __func__, status, 1 << dd->hfi1_id);
1393 		ret = -EIO;
1394 		goto done;
1395 	}
1396 
1397 	/* extract error */
1398 	err = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_SHIFT)
1399 		& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_MASK;
1400 	if (err) {
1401 		dd_dev_err(dd, "%s: gasket error %d\n", __func__, err);
1402 		ret = -EIO;
1403 		goto done;
1404 	}
1405 
1406 	/* update our link information cache */
1407 	update_lbus_info(dd);
1408 	dd_dev_info(dd, "%s: new speed and width: %s\n", __func__,
1409 		    dd->lbus_info);
1410 
1411 	if (dd->lbus_speed != target_speed ||
1412 	    dd->lbus_width < target_width) { /* not target */
1413 		/* maybe retry */
1414 		do_retry = retry_count < pcie_retry;
1415 		dd_dev_err(dd, "PCIe link speed or width did not match target%s\n",
1416 			   do_retry ? ", retrying" : "");
1417 		retry_count++;
1418 		if (do_retry) {
1419 			msleep(100); /* allow time to settle */
1420 			goto retry;
1421 		}
1422 		ret = -EIO;
1423 	}
1424 
1425 done:
1426 	if (therm) {
1427 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x1);
1428 		msleep(100);
1429 		dd_dev_info(dd, "%s: Re-enable therm polling\n",
1430 			    __func__);
1431 	}
1432 	release_chip_resource(dd, CR_SBUS);
1433 done_no_mutex:
1434 	/* return no error if it is OK to be at current speed */
1435 	if (ret && !return_error) {
1436 		dd_dev_err(dd, "Proceeding at current speed PCIe speed\n");
1437 		ret = 0;
1438 	}
1439 
1440 	dd_dev_info(dd, "%s: done\n", __func__);
1441 	return ret;
1442 }
1443