xref: /openbmc/linux/drivers/infiniband/hw/hfi1/pcie.c (revision 6c33a6f4)
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(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(
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 &&
323 	    (dd->pcidev->bus->max_bus_speed == PCIE_SPEED_2_5GT ||
324 	     dd->pcidev->bus->max_bus_speed == PCIE_SPEED_5_0GT)) {
325 		dd_dev_info(dd, "Parent PCIe bridge does not support Gen3\n");
326 		dd->link_gen3_capable = 0;
327 	}
328 
329 	/* obtain the link width and current speed */
330 	update_lbus_info(dd);
331 
332 	dd_dev_info(dd, "%s\n", dd->lbus_info);
333 
334 	return 0;
335 }
336 
337 /* restore command and BARs after a reset has wiped them out */
338 int restore_pci_variables(struct hfi1_devdata *dd)
339 {
340 	int ret = 0;
341 
342 	ret = pci_write_config_word(dd->pcidev, PCI_COMMAND, dd->pci_command);
343 	if (ret)
344 		goto error;
345 
346 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
347 				     dd->pcibar0);
348 	if (ret)
349 		goto error;
350 
351 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
352 				     dd->pcibar1);
353 	if (ret)
354 		goto error;
355 
356 	ret = pci_write_config_dword(dd->pcidev, PCI_ROM_ADDRESS, dd->pci_rom);
357 	if (ret)
358 		goto error;
359 
360 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL,
361 					 dd->pcie_devctl);
362 	if (ret)
363 		goto error;
364 
365 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL,
366 					 dd->pcie_lnkctl);
367 	if (ret)
368 		goto error;
369 
370 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL2,
371 					 dd->pcie_devctl2);
372 	if (ret)
373 		goto error;
374 
375 	ret = pci_write_config_dword(dd->pcidev, PCI_CFG_MSIX0, dd->pci_msix0);
376 	if (ret)
377 		goto error;
378 
379 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
380 		ret = pci_write_config_dword(dd->pcidev, PCIE_CFG_TPH2,
381 					     dd->pci_tph2);
382 		if (ret)
383 			goto error;
384 	}
385 	return 0;
386 
387 error:
388 	dd_dev_err(dd, "Unable to write to PCI config\n");
389 	return ret;
390 }
391 
392 /* Save BARs and command to rewrite after device reset */
393 int save_pci_variables(struct hfi1_devdata *dd)
394 {
395 	int ret = 0;
396 
397 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
398 				    &dd->pcibar0);
399 	if (ret)
400 		goto error;
401 
402 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
403 				    &dd->pcibar1);
404 	if (ret)
405 		goto error;
406 
407 	ret = pci_read_config_dword(dd->pcidev, PCI_ROM_ADDRESS, &dd->pci_rom);
408 	if (ret)
409 		goto error;
410 
411 	ret = pci_read_config_word(dd->pcidev, PCI_COMMAND, &dd->pci_command);
412 	if (ret)
413 		goto error;
414 
415 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL,
416 					&dd->pcie_devctl);
417 	if (ret)
418 		goto error;
419 
420 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL,
421 					&dd->pcie_lnkctl);
422 	if (ret)
423 		goto error;
424 
425 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL2,
426 					&dd->pcie_devctl2);
427 	if (ret)
428 		goto error;
429 
430 	ret = pci_read_config_dword(dd->pcidev, PCI_CFG_MSIX0, &dd->pci_msix0);
431 	if (ret)
432 		goto error;
433 
434 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
435 		ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_TPH2,
436 					    &dd->pci_tph2);
437 		if (ret)
438 			goto error;
439 	}
440 	return 0;
441 
442 error:
443 	dd_dev_err(dd, "Unable to read from PCI config\n");
444 	return ret;
445 }
446 
447 /*
448  * BIOS may not set PCIe bus-utilization parameters for best performance.
449  * Check and optionally adjust them to maximize our throughput.
450  */
451 static int hfi1_pcie_caps;
452 module_param_named(pcie_caps, hfi1_pcie_caps, int, 0444);
453 MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
454 
455 /**
456  * tune_pcie_caps() - Code to adjust PCIe capabilities.
457  * @dd: Valid device data structure
458  *
459  */
460 void tune_pcie_caps(struct hfi1_devdata *dd)
461 {
462 	struct pci_dev *parent;
463 	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
464 	u16 rc_mrrs, ep_mrrs, max_mrrs, ectl;
465 	int ret;
466 
467 	/*
468 	 * Turn on extended tags in DevCtl in case the BIOS has turned it off
469 	 * to improve WFR SDMA bandwidth
470 	 */
471 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL, &ectl);
472 	if ((!ret) && !(ectl & PCI_EXP_DEVCTL_EXT_TAG)) {
473 		dd_dev_info(dd, "Enabling PCIe extended tags\n");
474 		ectl |= PCI_EXP_DEVCTL_EXT_TAG;
475 		ret = pcie_capability_write_word(dd->pcidev,
476 						 PCI_EXP_DEVCTL, ectl);
477 		if (ret)
478 			dd_dev_info(dd, "Unable to write to PCI config\n");
479 	}
480 	/* Find out supported and configured values for parent (root) */
481 	parent = dd->pcidev->bus->self;
482 	/*
483 	 * The driver cannot perform the tuning if it does not have
484 	 * access to the upstream component.
485 	 */
486 	if (!parent) {
487 		dd_dev_info(dd, "Parent not found\n");
488 		return;
489 	}
490 	if (!pci_is_root_bus(parent->bus)) {
491 		dd_dev_info(dd, "Parent not root\n");
492 		return;
493 	}
494 	if (!pci_is_pcie(parent)) {
495 		dd_dev_info(dd, "Parent is not PCI Express capable\n");
496 		return;
497 	}
498 	if (!pci_is_pcie(dd->pcidev)) {
499 		dd_dev_info(dd, "PCI device is not PCI Express capable\n");
500 		return;
501 	}
502 	rc_mpss = parent->pcie_mpss;
503 	rc_mps = ffs(pcie_get_mps(parent)) - 8;
504 	/* Find out supported and configured values for endpoint (us) */
505 	ep_mpss = dd->pcidev->pcie_mpss;
506 	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
507 
508 	/* Find max payload supported by root, endpoint */
509 	if (rc_mpss > ep_mpss)
510 		rc_mpss = ep_mpss;
511 
512 	/* If Supported greater than limit in module param, limit it */
513 	if (rc_mpss > (hfi1_pcie_caps & 7))
514 		rc_mpss = hfi1_pcie_caps & 7;
515 	/* If less than (allowed, supported), bump root payload */
516 	if (rc_mpss > rc_mps) {
517 		rc_mps = rc_mpss;
518 		pcie_set_mps(parent, 128 << rc_mps);
519 	}
520 	/* If less than (allowed, supported), bump endpoint payload */
521 	if (rc_mpss > ep_mps) {
522 		ep_mps = rc_mpss;
523 		pcie_set_mps(dd->pcidev, 128 << ep_mps);
524 	}
525 
526 	/*
527 	 * Now the Read Request size.
528 	 * No field for max supported, but PCIe spec limits it to 4096,
529 	 * which is code '5' (log2(4096) - 7)
530 	 */
531 	max_mrrs = 5;
532 	if (max_mrrs > ((hfi1_pcie_caps >> 4) & 7))
533 		max_mrrs = (hfi1_pcie_caps >> 4) & 7;
534 
535 	max_mrrs = 128 << max_mrrs;
536 	rc_mrrs = pcie_get_readrq(parent);
537 	ep_mrrs = pcie_get_readrq(dd->pcidev);
538 
539 	if (max_mrrs > rc_mrrs) {
540 		rc_mrrs = max_mrrs;
541 		pcie_set_readrq(parent, rc_mrrs);
542 	}
543 	if (max_mrrs > ep_mrrs) {
544 		ep_mrrs = max_mrrs;
545 		pcie_set_readrq(dd->pcidev, ep_mrrs);
546 	}
547 }
548 
549 /* End of PCIe capability tuning */
550 
551 /*
552  * From here through hfi1_pci_err_handler definition is invoked via
553  * PCI error infrastructure, registered via pci
554  */
555 static pci_ers_result_t
556 pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
557 {
558 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
559 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
560 
561 	switch (state) {
562 	case pci_channel_io_normal:
563 		dd_dev_info(dd, "State Normal, ignoring\n");
564 		break;
565 
566 	case pci_channel_io_frozen:
567 		dd_dev_info(dd, "State Frozen, requesting reset\n");
568 		pci_disable_device(pdev);
569 		ret = PCI_ERS_RESULT_NEED_RESET;
570 		break;
571 
572 	case pci_channel_io_perm_failure:
573 		if (dd) {
574 			dd_dev_info(dd, "State Permanent Failure, disabling\n");
575 			/* no more register accesses! */
576 			dd->flags &= ~HFI1_PRESENT;
577 			hfi1_disable_after_error(dd);
578 		}
579 		 /* else early, or other problem */
580 		ret =  PCI_ERS_RESULT_DISCONNECT;
581 		break;
582 
583 	default: /* shouldn't happen */
584 		dd_dev_info(dd, "HFI1 PCI errors detected (state %d)\n",
585 			    state);
586 		break;
587 	}
588 	return ret;
589 }
590 
591 static pci_ers_result_t
592 pci_mmio_enabled(struct pci_dev *pdev)
593 {
594 	u64 words = 0U;
595 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
596 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
597 
598 	if (dd && dd->pport) {
599 		words = read_port_cntr(dd->pport, C_RX_WORDS, CNTR_INVALID_VL);
600 		if (words == ~0ULL)
601 			ret = PCI_ERS_RESULT_NEED_RESET;
602 		dd_dev_info(dd,
603 			    "HFI1 mmio_enabled function called, read wordscntr %llx, returning %d\n",
604 			    words, ret);
605 	}
606 	return  ret;
607 }
608 
609 static pci_ers_result_t
610 pci_slot_reset(struct pci_dev *pdev)
611 {
612 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
613 
614 	dd_dev_info(dd, "HFI1 slot_reset function called, ignored\n");
615 	return PCI_ERS_RESULT_CAN_RECOVER;
616 }
617 
618 static void
619 pci_resume(struct pci_dev *pdev)
620 {
621 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
622 
623 	dd_dev_info(dd, "HFI1 resume function called\n");
624 	/*
625 	 * Running jobs will fail, since it's asynchronous
626 	 * unlike sysfs-requested reset.   Better than
627 	 * doing nothing.
628 	 */
629 	hfi1_init(dd, 1); /* same as re-init after reset */
630 }
631 
632 const struct pci_error_handlers hfi1_pci_err_handler = {
633 	.error_detected = pci_error_detected,
634 	.mmio_enabled = pci_mmio_enabled,
635 	.slot_reset = pci_slot_reset,
636 	.resume = pci_resume,
637 };
638 
639 /*============================================================================*/
640 /* PCIe Gen3 support */
641 
642 /*
643  * This code is separated out because it is expected to be removed in the
644  * final shipping product.  If not, then it will be revisited and items
645  * will be moved to more standard locations.
646  */
647 
648 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_STS field values */
649 #define DL_STATUS_HFI0 0x1	/* hfi0 firmware download complete */
650 #define DL_STATUS_HFI1 0x2	/* hfi1 firmware download complete */
651 #define DL_STATUS_BOTH 0x3	/* hfi0 and hfi1 firmware download complete */
652 
653 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_ERR field values */
654 #define DL_ERR_NONE		0x0	/* no error */
655 #define DL_ERR_SWAP_PARITY	0x1	/* parity error in SerDes interrupt */
656 					/*   or response data */
657 #define DL_ERR_DISABLED	0x2	/* hfi disabled */
658 #define DL_ERR_SECURITY	0x3	/* security check failed */
659 #define DL_ERR_SBUS		0x4	/* SBus status error */
660 #define DL_ERR_XFR_PARITY	0x5	/* parity error during ROM transfer*/
661 
662 /* gasket block secondary bus reset delay */
663 #define SBR_DELAY_US 200000	/* 200ms */
664 
665 static uint pcie_target = 3;
666 module_param(pcie_target, uint, S_IRUGO);
667 MODULE_PARM_DESC(pcie_target, "PCIe target speed (0 skip, 1-3 Gen1-3)");
668 
669 static uint pcie_force;
670 module_param(pcie_force, uint, S_IRUGO);
671 MODULE_PARM_DESC(pcie_force, "Force driver to do a PCIe firmware download even if already at target speed");
672 
673 static uint pcie_retry = 5;
674 module_param(pcie_retry, uint, S_IRUGO);
675 MODULE_PARM_DESC(pcie_retry, "Driver will try this many times to reach requested speed");
676 
677 #define UNSET_PSET 255
678 #define DEFAULT_DISCRETE_PSET 2	/* discrete HFI */
679 #define DEFAULT_MCP_PSET 6	/* MCP HFI */
680 static uint pcie_pset = UNSET_PSET;
681 module_param(pcie_pset, uint, S_IRUGO);
682 MODULE_PARM_DESC(pcie_pset, "PCIe Eq Pset value to use, range is 0-10");
683 
684 static uint pcie_ctle = 3; /* discrete on, integrated on */
685 module_param(pcie_ctle, uint, S_IRUGO);
686 MODULE_PARM_DESC(pcie_ctle, "PCIe static CTLE mode, bit 0 - discrete on/off, bit 1 - integrated on/off");
687 
688 /* equalization columns */
689 #define PREC 0
690 #define ATTN 1
691 #define POST 2
692 
693 /* discrete silicon preliminary equalization values */
694 static const u8 discrete_preliminary_eq[11][3] = {
695 	/* prec   attn   post */
696 	{  0x00,  0x00,  0x12 },	/* p0 */
697 	{  0x00,  0x00,  0x0c },	/* p1 */
698 	{  0x00,  0x00,  0x0f },	/* p2 */
699 	{  0x00,  0x00,  0x09 },	/* p3 */
700 	{  0x00,  0x00,  0x00 },	/* p4 */
701 	{  0x06,  0x00,  0x00 },	/* p5 */
702 	{  0x09,  0x00,  0x00 },	/* p6 */
703 	{  0x06,  0x00,  0x0f },	/* p7 */
704 	{  0x09,  0x00,  0x09 },	/* p8 */
705 	{  0x0c,  0x00,  0x00 },	/* p9 */
706 	{  0x00,  0x00,  0x18 },	/* p10 */
707 };
708 
709 /* integrated silicon preliminary equalization values */
710 static const u8 integrated_preliminary_eq[11][3] = {
711 	/* prec   attn   post */
712 	{  0x00,  0x1e,  0x07 },	/* p0 */
713 	{  0x00,  0x1e,  0x05 },	/* p1 */
714 	{  0x00,  0x1e,  0x06 },	/* p2 */
715 	{  0x00,  0x1e,  0x04 },	/* p3 */
716 	{  0x00,  0x1e,  0x00 },	/* p4 */
717 	{  0x03,  0x1e,  0x00 },	/* p5 */
718 	{  0x04,  0x1e,  0x00 },	/* p6 */
719 	{  0x03,  0x1e,  0x06 },	/* p7 */
720 	{  0x03,  0x1e,  0x04 },	/* p8 */
721 	{  0x05,  0x1e,  0x00 },	/* p9 */
722 	{  0x00,  0x1e,  0x0a },	/* p10 */
723 };
724 
725 static const u8 discrete_ctle_tunings[11][4] = {
726 	/* DC     LF     HF     BW */
727 	{  0x48,  0x0b,  0x04,  0x04 },	/* p0 */
728 	{  0x60,  0x05,  0x0f,  0x0a },	/* p1 */
729 	{  0x50,  0x09,  0x06,  0x06 },	/* p2 */
730 	{  0x68,  0x05,  0x0f,  0x0a },	/* p3 */
731 	{  0x80,  0x05,  0x0f,  0x0a },	/* p4 */
732 	{  0x70,  0x05,  0x0f,  0x0a },	/* p5 */
733 	{  0x68,  0x05,  0x0f,  0x0a },	/* p6 */
734 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
735 	{  0x48,  0x09,  0x06,  0x06 },	/* p8 */
736 	{  0x60,  0x05,  0x0f,  0x0a },	/* p9 */
737 	{  0x38,  0x0f,  0x00,  0x00 },	/* p10 */
738 };
739 
740 static const u8 integrated_ctle_tunings[11][4] = {
741 	/* DC     LF     HF     BW */
742 	{  0x38,  0x0f,  0x00,  0x00 },	/* p0 */
743 	{  0x38,  0x0f,  0x00,  0x00 },	/* p1 */
744 	{  0x38,  0x0f,  0x00,  0x00 },	/* p2 */
745 	{  0x38,  0x0f,  0x00,  0x00 },	/* p3 */
746 	{  0x58,  0x0a,  0x05,  0x05 },	/* p4 */
747 	{  0x48,  0x0a,  0x05,  0x05 },	/* p5 */
748 	{  0x40,  0x0a,  0x05,  0x05 },	/* p6 */
749 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
750 	{  0x38,  0x0f,  0x00,  0x00 },	/* p8 */
751 	{  0x38,  0x09,  0x06,  0x06 },	/* p9 */
752 	{  0x38,  0x0e,  0x01,  0x01 },	/* p10 */
753 };
754 
755 /* helper to format the value to write to hardware */
756 #define eq_value(pre, curr, post) \
757 	((((u32)(pre)) << \
758 			PCIE_CFG_REG_PL102_GEN3_EQ_PRE_CURSOR_PSET_SHIFT) \
759 	| (((u32)(curr)) << PCIE_CFG_REG_PL102_GEN3_EQ_CURSOR_PSET_SHIFT) \
760 	| (((u32)(post)) << \
761 		PCIE_CFG_REG_PL102_GEN3_EQ_POST_CURSOR_PSET_SHIFT))
762 
763 /*
764  * Load the given EQ preset table into the PCIe hardware.
765  */
766 static int load_eq_table(struct hfi1_devdata *dd, const u8 eq[11][3], u8 fs,
767 			 u8 div)
768 {
769 	struct pci_dev *pdev = dd->pcidev;
770 	u32 hit_error = 0;
771 	u32 violation;
772 	u32 i;
773 	u8 c_minus1, c0, c_plus1;
774 	int ret;
775 
776 	for (i = 0; i < 11; i++) {
777 		/* set index */
778 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL103, i);
779 		/* write the value */
780 		c_minus1 = eq[i][PREC] / div;
781 		c0 = fs - (eq[i][PREC] / div) - (eq[i][POST] / div);
782 		c_plus1 = eq[i][POST] / div;
783 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL102,
784 				       eq_value(c_minus1, c0, c_plus1));
785 		/* check if these coefficients violate EQ rules */
786 		ret = pci_read_config_dword(dd->pcidev,
787 					    PCIE_CFG_REG_PL105, &violation);
788 		if (ret) {
789 			dd_dev_err(dd, "Unable to read from PCI config\n");
790 			hit_error = 1;
791 			break;
792 		}
793 
794 		if (violation
795 		    & PCIE_CFG_REG_PL105_GEN3_EQ_VIOLATE_COEF_RULES_SMASK){
796 			if (hit_error == 0) {
797 				dd_dev_err(dd,
798 					   "Gen3 EQ Table Coefficient rule violations\n");
799 				dd_dev_err(dd, "         prec   attn   post\n");
800 			}
801 			dd_dev_err(dd, "   p%02d:   %02x     %02x     %02x\n",
802 				   i, (u32)eq[i][0], (u32)eq[i][1],
803 				   (u32)eq[i][2]);
804 			dd_dev_err(dd, "            %02x     %02x     %02x\n",
805 				   (u32)c_minus1, (u32)c0, (u32)c_plus1);
806 			hit_error = 1;
807 		}
808 	}
809 	if (hit_error)
810 		return -EINVAL;
811 	return 0;
812 }
813 
814 /*
815  * Steps to be done after the PCIe firmware is downloaded and
816  * before the SBR for the Pcie Gen3.
817  * The SBus resource is already being held.
818  */
819 static void pcie_post_steps(struct hfi1_devdata *dd)
820 {
821 	int i;
822 
823 	set_sbus_fast_mode(dd);
824 	/*
825 	 * Write to the PCIe PCSes to set the G3_LOCKED_NEXT bits to 1.
826 	 * This avoids a spurious framing error that can otherwise be
827 	 * generated by the MAC layer.
828 	 *
829 	 * Use individual addresses since no broadcast is set up.
830 	 */
831 	for (i = 0; i < NUM_PCIE_SERDES; i++) {
832 		sbus_request(dd, pcie_pcs_addrs[dd->hfi1_id][i],
833 			     0x03, WRITE_SBUS_RECEIVER, 0x00022132);
834 	}
835 
836 	clear_sbus_fast_mode(dd);
837 }
838 
839 /*
840  * Trigger a secondary bus reset (SBR) on ourselves using our parent.
841  *
842  * Based on pci_parent_bus_reset() which is not exported by the
843  * kernel core.
844  */
845 static int trigger_sbr(struct hfi1_devdata *dd)
846 {
847 	struct pci_dev *dev = dd->pcidev;
848 	struct pci_dev *pdev;
849 
850 	/* need a parent */
851 	if (!dev->bus->self) {
852 		dd_dev_err(dd, "%s: no parent device\n", __func__);
853 		return -ENOTTY;
854 	}
855 
856 	/* should not be anyone else on the bus */
857 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
858 		if (pdev != dev) {
859 			dd_dev_err(dd,
860 				   "%s: another device is on the same bus\n",
861 				   __func__);
862 			return -ENOTTY;
863 		}
864 
865 	/*
866 	 * This is an end around to do an SBR during probe time. A new API needs
867 	 * to be implemented to have cleaner interface but this fixes the
868 	 * current brokenness
869 	 */
870 	return pci_bridge_secondary_bus_reset(dev->bus->self);
871 }
872 
873 /*
874  * Write the given gasket interrupt register.
875  */
876 static void write_gasket_interrupt(struct hfi1_devdata *dd, int index,
877 				   u16 code, u16 data)
878 {
879 	write_csr(dd, ASIC_PCIE_SD_INTRPT_LIST + (index * 8),
880 		  (((u64)code << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_CODE_SHIFT) |
881 		   ((u64)data << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_DATA_SHIFT)));
882 }
883 
884 /*
885  * Tell the gasket logic how to react to the reset.
886  */
887 static void arm_gasket_logic(struct hfi1_devdata *dd)
888 {
889 	u64 reg;
890 
891 	reg = (((u64)1 << dd->hfi1_id) <<
892 	       ASIC_PCIE_SD_HOST_CMD_INTRPT_CMD_SHIFT) |
893 	      ((u64)pcie_serdes_broadcast[dd->hfi1_id] <<
894 	       ASIC_PCIE_SD_HOST_CMD_SBUS_RCVR_ADDR_SHIFT |
895 	       ASIC_PCIE_SD_HOST_CMD_SBR_MODE_SMASK |
896 	       ((u64)SBR_DELAY_US & ASIC_PCIE_SD_HOST_CMD_TIMER_MASK) <<
897 	       ASIC_PCIE_SD_HOST_CMD_TIMER_SHIFT);
898 	write_csr(dd, ASIC_PCIE_SD_HOST_CMD, reg);
899 	/* read back to push the write */
900 	read_csr(dd, ASIC_PCIE_SD_HOST_CMD);
901 }
902 
903 /*
904  * CCE_PCIE_CTRL long name helpers
905  * We redefine these shorter macros to use in the code while leaving
906  * chip_registers.h to be autogenerated from the hardware spec.
907  */
908 #define LANE_BUNDLE_MASK              CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_MASK
909 #define LANE_BUNDLE_SHIFT             CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_SHIFT
910 #define LANE_DELAY_MASK               CCE_PCIE_CTRL_PCIE_LANE_DELAY_MASK
911 #define LANE_DELAY_SHIFT              CCE_PCIE_CTRL_PCIE_LANE_DELAY_SHIFT
912 #define MARGIN_OVERWRITE_ENABLE_SHIFT CCE_PCIE_CTRL_XMT_MARGIN_OVERWRITE_ENABLE_SHIFT
913 #define MARGIN_SHIFT                  CCE_PCIE_CTRL_XMT_MARGIN_SHIFT
914 #define MARGIN_G1_G2_OVERWRITE_MASK   CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_MASK
915 #define MARGIN_G1_G2_OVERWRITE_SHIFT  CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_SHIFT
916 #define MARGIN_GEN1_GEN2_MASK         CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_MASK
917 #define MARGIN_GEN1_GEN2_SHIFT        CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_SHIFT
918 
919  /*
920   * Write xmt_margin for full-swing (WFR-B) or half-swing (WFR-C).
921   */
922 static void write_xmt_margin(struct hfi1_devdata *dd, const char *fname)
923 {
924 	u64 pcie_ctrl;
925 	u64 xmt_margin;
926 	u64 xmt_margin_oe;
927 	u64 lane_delay;
928 	u64 lane_bundle;
929 
930 	pcie_ctrl = read_csr(dd, CCE_PCIE_CTRL);
931 
932 	/*
933 	 * For Discrete, use full-swing.
934 	 *  - PCIe TX defaults to full-swing.
935 	 *    Leave this register as default.
936 	 * For Integrated, use half-swing
937 	 *  - Copy xmt_margin and xmt_margin_oe
938 	 *    from Gen1/Gen2 to Gen3.
939 	 */
940 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL1) { /* integrated */
941 		/* extract initial fields */
942 		xmt_margin = (pcie_ctrl >> MARGIN_GEN1_GEN2_SHIFT)
943 			      & MARGIN_GEN1_GEN2_MASK;
944 		xmt_margin_oe = (pcie_ctrl >> MARGIN_G1_G2_OVERWRITE_SHIFT)
945 				 & MARGIN_G1_G2_OVERWRITE_MASK;
946 		lane_delay = (pcie_ctrl >> LANE_DELAY_SHIFT) & LANE_DELAY_MASK;
947 		lane_bundle = (pcie_ctrl >> LANE_BUNDLE_SHIFT)
948 			       & LANE_BUNDLE_MASK;
949 
950 		/*
951 		 * For A0, EFUSE values are not set.  Override with the
952 		 * correct values.
953 		 */
954 		if (is_ax(dd)) {
955 			/*
956 			 * xmt_margin and OverwiteEnabel should be the
957 			 * same for Gen1/Gen2 and Gen3
958 			 */
959 			xmt_margin = 0x5;
960 			xmt_margin_oe = 0x1;
961 			lane_delay = 0xF; /* Delay 240ns. */
962 			lane_bundle = 0x0; /* Set to 1 lane. */
963 		}
964 
965 		/* overwrite existing values */
966 		pcie_ctrl = (xmt_margin << MARGIN_GEN1_GEN2_SHIFT)
967 			| (xmt_margin_oe << MARGIN_G1_G2_OVERWRITE_SHIFT)
968 			| (xmt_margin << MARGIN_SHIFT)
969 			| (xmt_margin_oe << MARGIN_OVERWRITE_ENABLE_SHIFT)
970 			| (lane_delay << LANE_DELAY_SHIFT)
971 			| (lane_bundle << LANE_BUNDLE_SHIFT);
972 
973 		write_csr(dd, CCE_PCIE_CTRL, pcie_ctrl);
974 	}
975 
976 	dd_dev_dbg(dd, "%s: program XMT margin, CcePcieCtrl 0x%llx\n",
977 		   fname, pcie_ctrl);
978 }
979 
980 /*
981  * Do all the steps needed to transition the PCIe link to Gen3 speed.
982  */
983 int do_pcie_gen3_transition(struct hfi1_devdata *dd)
984 {
985 	struct pci_dev *parent = dd->pcidev->bus->self;
986 	u64 fw_ctrl;
987 	u64 reg, therm;
988 	u32 reg32, fs, lf;
989 	u32 status, err;
990 	int ret;
991 	int do_retry, retry_count = 0;
992 	int intnum = 0;
993 	uint default_pset;
994 	uint pset = pcie_pset;
995 	u16 target_vector, target_speed;
996 	u16 lnkctl2, vendor;
997 	u8 div;
998 	const u8 (*eq)[3];
999 	const u8 (*ctle_tunings)[4];
1000 	uint static_ctle_mode;
1001 	int return_error = 0;
1002 	u32 target_width;
1003 
1004 	/* PCIe Gen3 is for the ASIC only */
1005 	if (dd->icode != ICODE_RTL_SILICON)
1006 		return 0;
1007 
1008 	if (pcie_target == 1) {			/* target Gen1 */
1009 		target_vector = PCI_EXP_LNKCTL2_TLS_2_5GT;
1010 		target_speed = 2500;
1011 	} else if (pcie_target == 2) {		/* target Gen2 */
1012 		target_vector = PCI_EXP_LNKCTL2_TLS_5_0GT;
1013 		target_speed = 5000;
1014 	} else if (pcie_target == 3) {		/* target Gen3 */
1015 		target_vector = PCI_EXP_LNKCTL2_TLS_8_0GT;
1016 		target_speed = 8000;
1017 	} else {
1018 		/* off or invalid target - skip */
1019 		dd_dev_info(dd, "%s: Skipping PCIe transition\n", __func__);
1020 		return 0;
1021 	}
1022 
1023 	/* if already at target speed, done (unless forced) */
1024 	if (dd->lbus_speed == target_speed) {
1025 		dd_dev_info(dd, "%s: PCIe already at gen%d, %s\n", __func__,
1026 			    pcie_target,
1027 			    pcie_force ? "re-doing anyway" : "skipping");
1028 		if (!pcie_force)
1029 			return 0;
1030 	}
1031 
1032 	/*
1033 	 * The driver cannot do the transition if it has no access to the
1034 	 * upstream component
1035 	 */
1036 	if (!parent) {
1037 		dd_dev_info(dd, "%s: No upstream, Can't do gen3 transition\n",
1038 			    __func__);
1039 		return 0;
1040 	}
1041 
1042 	/* Previous Gen1/Gen2 bus width */
1043 	target_width = dd->lbus_width;
1044 
1045 	/*
1046 	 * Do the Gen3 transition.  Steps are those of the PCIe Gen3
1047 	 * recipe.
1048 	 */
1049 
1050 	/* step 1: pcie link working in gen1/gen2 */
1051 
1052 	/* step 2: if either side is not capable of Gen3, done */
1053 	if (pcie_target == 3 && !dd->link_gen3_capable) {
1054 		dd_dev_err(dd, "The PCIe link is not Gen3 capable\n");
1055 		ret = -ENOSYS;
1056 		goto done_no_mutex;
1057 	}
1058 
1059 	/* hold the SBus resource across the firmware download and SBR */
1060 	ret = acquire_chip_resource(dd, CR_SBUS, SBUS_TIMEOUT);
1061 	if (ret) {
1062 		dd_dev_err(dd, "%s: unable to acquire SBus resource\n",
1063 			   __func__);
1064 		return ret;
1065 	}
1066 
1067 	/* make sure thermal polling is not causing interrupts */
1068 	therm = read_csr(dd, ASIC_CFG_THERM_POLL_EN);
1069 	if (therm) {
1070 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x0);
1071 		msleep(100);
1072 		dd_dev_info(dd, "%s: Disabled therm polling\n",
1073 			    __func__);
1074 	}
1075 
1076 retry:
1077 	/* the SBus download will reset the spico for thermal */
1078 
1079 	/* step 3: download SBus Master firmware */
1080 	/* step 4: download PCIe Gen3 SerDes firmware */
1081 	dd_dev_info(dd, "%s: downloading firmware\n", __func__);
1082 	ret = load_pcie_firmware(dd);
1083 	if (ret) {
1084 		/* do not proceed if the firmware cannot be downloaded */
1085 		return_error = 1;
1086 		goto done;
1087 	}
1088 
1089 	/* step 5: set up device parameter settings */
1090 	dd_dev_info(dd, "%s: setting PCIe registers\n", __func__);
1091 
1092 	/*
1093 	 * PcieCfgSpcie1 - Link Control 3
1094 	 * Leave at reset value.  No need to set PerfEq - link equalization
1095 	 * will be performed automatically after the SBR when the target
1096 	 * speed is 8GT/s.
1097 	 */
1098 
1099 	/* clear all 16 per-lane error bits (PCIe: Lane Error Status) */
1100 	pci_write_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, 0xffff);
1101 
1102 	/* step 5a: Set Synopsys Port Logic registers */
1103 
1104 	/*
1105 	 * PcieCfgRegPl2 - Port Force Link
1106 	 *
1107 	 * Set the low power field to 0x10 to avoid unnecessary power
1108 	 * management messages.  All other fields are zero.
1109 	 */
1110 	reg32 = 0x10ul << PCIE_CFG_REG_PL2_LOW_PWR_ENT_CNT_SHIFT;
1111 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL2, reg32);
1112 
1113 	/*
1114 	 * PcieCfgRegPl100 - Gen3 Control
1115 	 *
1116 	 * turn off PcieCfgRegPl100.Gen3ZRxDcNonCompl
1117 	 * turn on PcieCfgRegPl100.EqEieosCnt
1118 	 * Everything else zero.
1119 	 */
1120 	reg32 = PCIE_CFG_REG_PL100_EQ_EIEOS_CNT_SMASK;
1121 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL100, reg32);
1122 
1123 	/*
1124 	 * PcieCfgRegPl101 - Gen3 EQ FS and LF
1125 	 * PcieCfgRegPl102 - Gen3 EQ Presets to Coefficients Mapping
1126 	 * PcieCfgRegPl103 - Gen3 EQ Preset Index
1127 	 * PcieCfgRegPl105 - Gen3 EQ Status
1128 	 *
1129 	 * Give initial EQ settings.
1130 	 */
1131 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL0) { /* discrete */
1132 		/* 1000mV, FS=24, LF = 8 */
1133 		fs = 24;
1134 		lf = 8;
1135 		div = 3;
1136 		eq = discrete_preliminary_eq;
1137 		default_pset = DEFAULT_DISCRETE_PSET;
1138 		ctle_tunings = discrete_ctle_tunings;
1139 		/* bit 0 - discrete on/off */
1140 		static_ctle_mode = pcie_ctle & 0x1;
1141 	} else {
1142 		/* 400mV, FS=29, LF = 9 */
1143 		fs = 29;
1144 		lf = 9;
1145 		div = 1;
1146 		eq = integrated_preliminary_eq;
1147 		default_pset = DEFAULT_MCP_PSET;
1148 		ctle_tunings = integrated_ctle_tunings;
1149 		/* bit 1 - integrated on/off */
1150 		static_ctle_mode = (pcie_ctle >> 1) & 0x1;
1151 	}
1152 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL101,
1153 			       (fs <<
1154 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_FS_SHIFT) |
1155 			       (lf <<
1156 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_LF_SHIFT));
1157 	ret = load_eq_table(dd, eq, fs, div);
1158 	if (ret)
1159 		goto done;
1160 
1161 	/*
1162 	 * PcieCfgRegPl106 - Gen3 EQ Control
1163 	 *
1164 	 * Set Gen3EqPsetReqVec, leave other fields 0.
1165 	 */
1166 	if (pset == UNSET_PSET)
1167 		pset = default_pset;
1168 	if (pset > 10) {	/* valid range is 0-10, inclusive */
1169 		dd_dev_err(dd, "%s: Invalid Eq Pset %u, setting to %d\n",
1170 			   __func__, pset, default_pset);
1171 		pset = default_pset;
1172 	}
1173 	dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pset);
1174 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL106,
1175 			       ((1 << pset) <<
1176 			PCIE_CFG_REG_PL106_GEN3_EQ_PSET_REQ_VEC_SHIFT) |
1177 			PCIE_CFG_REG_PL106_GEN3_EQ_EVAL2MS_DISABLE_SMASK |
1178 			PCIE_CFG_REG_PL106_GEN3_EQ_PHASE23_EXIT_MODE_SMASK);
1179 
1180 	/*
1181 	 * step 5b: Do post firmware download steps via SBus
1182 	 */
1183 	dd_dev_info(dd, "%s: doing pcie post steps\n", __func__);
1184 	pcie_post_steps(dd);
1185 
1186 	/*
1187 	 * step 5c: Program gasket interrupts
1188 	 */
1189 	/* set the Rx Bit Rate to REFCLK ratio */
1190 	write_gasket_interrupt(dd, intnum++, 0x0006, 0x0050);
1191 	/* disable pCal for PCIe Gen3 RX equalization */
1192 	/* select adaptive or static CTLE */
1193 	write_gasket_interrupt(dd, intnum++, 0x0026,
1194 			       0x5b01 | (static_ctle_mode << 3));
1195 	/*
1196 	 * Enable iCal for PCIe Gen3 RX equalization, and set which
1197 	 * evaluation of RX_EQ_EVAL will launch the iCal procedure.
1198 	 */
1199 	write_gasket_interrupt(dd, intnum++, 0x0026, 0x5202);
1200 
1201 	if (static_ctle_mode) {
1202 		/* apply static CTLE tunings */
1203 		u8 pcie_dc, pcie_lf, pcie_hf, pcie_bw;
1204 
1205 		pcie_dc = ctle_tunings[pset][0];
1206 		pcie_lf = ctle_tunings[pset][1];
1207 		pcie_hf = ctle_tunings[pset][2];
1208 		pcie_bw = ctle_tunings[pset][3];
1209 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0200 | pcie_dc);
1210 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0100 | pcie_lf);
1211 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0000 | pcie_hf);
1212 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x5500 | pcie_bw);
1213 	}
1214 
1215 	/* terminate list */
1216 	write_gasket_interrupt(dd, intnum++, 0x0000, 0x0000);
1217 
1218 	/*
1219 	 * step 5d: program XMT margin
1220 	 */
1221 	write_xmt_margin(dd, __func__);
1222 
1223 	/*
1224 	 * step 5e: disable active state power management (ASPM). It
1225 	 * will be enabled if required later
1226 	 */
1227 	dd_dev_info(dd, "%s: clearing ASPM\n", __func__);
1228 	aspm_hw_disable_l1(dd);
1229 
1230 	/*
1231 	 * step 5f: clear DirectSpeedChange
1232 	 * PcieCfgRegPl67.DirectSpeedChange must be zero to prevent the
1233 	 * change in the speed target from starting before we are ready.
1234 	 * This field defaults to 0 and we are not changing it, so nothing
1235 	 * needs to be done.
1236 	 */
1237 
1238 	/* step 5g: Set target link speed */
1239 	/*
1240 	 * Set target link speed to be target on both device and parent.
1241 	 * On setting the parent: Some system BIOSs "helpfully" set the
1242 	 * parent target speed to Gen2 to match the ASIC's initial speed.
1243 	 * We can set the target Gen3 because we have already checked
1244 	 * that it is Gen3 capable earlier.
1245 	 */
1246 	dd_dev_info(dd, "%s: setting parent target link speed\n", __func__);
1247 	ret = pcie_capability_read_word(parent, PCI_EXP_LNKCTL2, &lnkctl2);
1248 	if (ret) {
1249 		dd_dev_err(dd, "Unable to read from PCI config\n");
1250 		return_error = 1;
1251 		goto done;
1252 	}
1253 
1254 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1255 		    (u32)lnkctl2);
1256 	/* only write to parent if target is not as high as ours */
1257 	if ((lnkctl2 & PCI_EXP_LNKCTL2_TLS) < target_vector) {
1258 		lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1259 		lnkctl2 |= target_vector;
1260 		dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1261 			    (u32)lnkctl2);
1262 		ret = pcie_capability_write_word(parent,
1263 						 PCI_EXP_LNKCTL2, lnkctl2);
1264 		if (ret) {
1265 			dd_dev_err(dd, "Unable to write to PCI config\n");
1266 			return_error = 1;
1267 			goto done;
1268 		}
1269 	} else {
1270 		dd_dev_info(dd, "%s: ..target speed is OK\n", __func__);
1271 	}
1272 
1273 	dd_dev_info(dd, "%s: setting target link speed\n", __func__);
1274 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL2, &lnkctl2);
1275 	if (ret) {
1276 		dd_dev_err(dd, "Unable to read from PCI config\n");
1277 		return_error = 1;
1278 		goto done;
1279 	}
1280 
1281 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1282 		    (u32)lnkctl2);
1283 	lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1284 	lnkctl2 |= target_vector;
1285 	dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1286 		    (u32)lnkctl2);
1287 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL2, lnkctl2);
1288 	if (ret) {
1289 		dd_dev_err(dd, "Unable to write to PCI config\n");
1290 		return_error = 1;
1291 		goto done;
1292 	}
1293 
1294 	/* step 5h: arm gasket logic */
1295 	/* hold DC in reset across the SBR */
1296 	write_csr(dd, CCE_DC_CTRL, CCE_DC_CTRL_DC_RESET_SMASK);
1297 	(void)read_csr(dd, CCE_DC_CTRL); /* DC reset hold */
1298 	/* save firmware control across the SBR */
1299 	fw_ctrl = read_csr(dd, MISC_CFG_FW_CTRL);
1300 
1301 	dd_dev_info(dd, "%s: arming gasket logic\n", __func__);
1302 	arm_gasket_logic(dd);
1303 
1304 	/*
1305 	 * step 6: quiesce PCIe link
1306 	 * The chip has already been reset, so there will be no traffic
1307 	 * from the chip.  Linux has no easy way to enforce that it will
1308 	 * not try to access the device, so we just need to hope it doesn't
1309 	 * do it while we are doing the reset.
1310 	 */
1311 
1312 	/*
1313 	 * step 7: initiate the secondary bus reset (SBR)
1314 	 * step 8: hardware brings the links back up
1315 	 * step 9: wait for link speed transition to be complete
1316 	 */
1317 	dd_dev_info(dd, "%s: calling trigger_sbr\n", __func__);
1318 	ret = trigger_sbr(dd);
1319 	if (ret)
1320 		goto done;
1321 
1322 	/* step 10: decide what to do next */
1323 
1324 	/* check if we can read PCI space */
1325 	ret = pci_read_config_word(dd->pcidev, PCI_VENDOR_ID, &vendor);
1326 	if (ret) {
1327 		dd_dev_info(dd,
1328 			    "%s: read of VendorID failed after SBR, err %d\n",
1329 			    __func__, ret);
1330 		return_error = 1;
1331 		goto done;
1332 	}
1333 	if (vendor == 0xffff) {
1334 		dd_dev_info(dd, "%s: VendorID is all 1s after SBR\n", __func__);
1335 		return_error = 1;
1336 		ret = -EIO;
1337 		goto done;
1338 	}
1339 
1340 	/* restore PCI space registers we know were reset */
1341 	dd_dev_info(dd, "%s: calling restore_pci_variables\n", __func__);
1342 	ret = restore_pci_variables(dd);
1343 	if (ret) {
1344 		dd_dev_err(dd, "%s: Could not restore PCI variables\n",
1345 			   __func__);
1346 		return_error = 1;
1347 		goto done;
1348 	}
1349 
1350 	/* restore firmware control */
1351 	write_csr(dd, MISC_CFG_FW_CTRL, fw_ctrl);
1352 
1353 	/*
1354 	 * Check the gasket block status.
1355 	 *
1356 	 * This is the first CSR read after the SBR.  If the read returns
1357 	 * all 1s (fails), the link did not make it back.
1358 	 *
1359 	 * Once we're sure we can read and write, clear the DC reset after
1360 	 * the SBR.  Then check for any per-lane errors. Then look over
1361 	 * the status.
1362 	 */
1363 	reg = read_csr(dd, ASIC_PCIE_SD_HOST_STATUS);
1364 	dd_dev_info(dd, "%s: gasket block status: 0x%llx\n", __func__, reg);
1365 	if (reg == ~0ull) {	/* PCIe read failed/timeout */
1366 		dd_dev_err(dd, "SBR failed - unable to read from device\n");
1367 		return_error = 1;
1368 		ret = -ENOSYS;
1369 		goto done;
1370 	}
1371 
1372 	/* clear the DC reset */
1373 	write_csr(dd, CCE_DC_CTRL, 0);
1374 
1375 	/* Set the LED off */
1376 	setextled(dd, 0);
1377 
1378 	/* check for any per-lane errors */
1379 	ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, &reg32);
1380 	if (ret) {
1381 		dd_dev_err(dd, "Unable to read from PCI config\n");
1382 		return_error = 1;
1383 		goto done;
1384 	}
1385 
1386 	dd_dev_info(dd, "%s: per-lane errors: 0x%x\n", __func__, reg32);
1387 
1388 	/* extract status, look for our HFI */
1389 	status = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_SHIFT)
1390 			& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_MASK;
1391 	if ((status & (1 << dd->hfi1_id)) == 0) {
1392 		dd_dev_err(dd,
1393 			   "%s: gasket status 0x%x, expecting 0x%x\n",
1394 			   __func__, status, 1 << dd->hfi1_id);
1395 		ret = -EIO;
1396 		goto done;
1397 	}
1398 
1399 	/* extract error */
1400 	err = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_SHIFT)
1401 		& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_MASK;
1402 	if (err) {
1403 		dd_dev_err(dd, "%s: gasket error %d\n", __func__, err);
1404 		ret = -EIO;
1405 		goto done;
1406 	}
1407 
1408 	/* update our link information cache */
1409 	update_lbus_info(dd);
1410 	dd_dev_info(dd, "%s: new speed and width: %s\n", __func__,
1411 		    dd->lbus_info);
1412 
1413 	if (dd->lbus_speed != target_speed ||
1414 	    dd->lbus_width < target_width) { /* not target */
1415 		/* maybe retry */
1416 		do_retry = retry_count < pcie_retry;
1417 		dd_dev_err(dd, "PCIe link speed or width did not match target%s\n",
1418 			   do_retry ? ", retrying" : "");
1419 		retry_count++;
1420 		if (do_retry) {
1421 			msleep(100); /* allow time to settle */
1422 			goto retry;
1423 		}
1424 		ret = -EIO;
1425 	}
1426 
1427 done:
1428 	if (therm) {
1429 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x1);
1430 		msleep(100);
1431 		dd_dev_info(dd, "%s: Re-enable therm polling\n",
1432 			    __func__);
1433 	}
1434 	release_chip_resource(dd, CR_SBUS);
1435 done_no_mutex:
1436 	/* return no error if it is OK to be at current speed */
1437 	if (ret && !return_error) {
1438 		dd_dev_err(dd, "Proceeding at current speed PCIe speed\n");
1439 		ret = 0;
1440 	}
1441 
1442 	dd_dev_info(dd, "%s: done\n", __func__);
1443 	return ret;
1444 }
1445