xref: /openbmc/linux/drivers/infiniband/hw/hfi1/pcie.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
1 /*
2  * Copyright(c) 2015 - 2017 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 /* link speed vector for Gen3 speed - not in Linux headers */
60 #define GEN1_SPEED_VECTOR 0x1
61 #define GEN2_SPEED_VECTOR 0x2
62 #define GEN3_SPEED_VECTOR 0x3
63 
64 /*
65  * This file contains PCIe utility routines.
66  */
67 
68 /*
69  * Code to adjust PCIe capabilities.
70  */
71 static void tune_pcie_caps(struct hfi1_devdata *);
72 
73 /*
74  * Do all the common PCIe setup and initialization.
75  * devdata is not yet allocated, and is not allocated until after this
76  * routine returns success.  Therefore dd_dev_err() can't be used for error
77  * printing.
78  */
79 int hfi1_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
80 {
81 	int ret;
82 
83 	ret = pci_enable_device(pdev);
84 	if (ret) {
85 		/*
86 		 * This can happen (in theory) iff:
87 		 * We did a chip reset, and then failed to reprogram the
88 		 * BAR, or the chip reset due to an internal error.  We then
89 		 * unloaded the driver and reloaded it.
90 		 *
91 		 * Both reset cases set the BAR back to initial state.  For
92 		 * the latter case, the AER sticky error bit at offset 0x718
93 		 * should be set, but the Linux kernel doesn't yet know
94 		 * about that, it appears.  If the original BAR was retained
95 		 * in the kernel data structures, this may be OK.
96 		 */
97 		hfi1_early_err(&pdev->dev, "pci enable failed: error %d\n",
98 			       -ret);
99 		goto done;
100 	}
101 
102 	ret = pci_request_regions(pdev, DRIVER_NAME);
103 	if (ret) {
104 		hfi1_early_err(&pdev->dev,
105 			       "pci_request_regions fails: err %d\n", -ret);
106 		goto bail;
107 	}
108 
109 	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
110 	if (ret) {
111 		/*
112 		 * If the 64 bit setup fails, try 32 bit.  Some systems
113 		 * do not setup 64 bit maps on systems with 2GB or less
114 		 * memory installed.
115 		 */
116 		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
117 		if (ret) {
118 			hfi1_early_err(&pdev->dev,
119 				       "Unable to set DMA mask: %d\n", ret);
120 			goto bail;
121 		}
122 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
123 	} else {
124 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
125 	}
126 	if (ret) {
127 		hfi1_early_err(&pdev->dev,
128 			       "Unable to set DMA consistent mask: %d\n", ret);
129 		goto bail;
130 	}
131 
132 	pci_set_master(pdev);
133 	(void)pci_enable_pcie_error_reporting(pdev);
134 	goto done;
135 
136 bail:
137 	hfi1_pcie_cleanup(pdev);
138 done:
139 	return ret;
140 }
141 
142 /*
143  * Clean what was done in hfi1_pcie_init()
144  */
145 void hfi1_pcie_cleanup(struct pci_dev *pdev)
146 {
147 	pci_disable_device(pdev);
148 	/*
149 	 * Release regions should be called after the disable. OK to
150 	 * call if request regions has not been called or failed.
151 	 */
152 	pci_release_regions(pdev);
153 }
154 
155 /*
156  * Do remaining PCIe setup, once dd is allocated, and save away
157  * fields required to re-initialize after a chip reset, or for
158  * various other purposes
159  */
160 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev)
161 {
162 	unsigned long len;
163 	resource_size_t addr;
164 	int ret = 0;
165 
166 	addr = pci_resource_start(pdev, 0);
167 	len = pci_resource_len(pdev, 0);
168 
169 	/*
170 	 * The TXE PIO buffers are at the tail end of the chip space.
171 	 * Cut them off and map them separately.
172 	 */
173 
174 	/* sanity check vs expectations */
175 	if (len != TXE_PIO_SEND + TXE_PIO_SIZE) {
176 		dd_dev_err(dd, "chip PIO range does not match\n");
177 		return -EINVAL;
178 	}
179 
180 	dd->kregbase1 = ioremap_nocache(addr, RCV_ARRAY);
181 	if (!dd->kregbase1) {
182 		dd_dev_err(dd, "UC mapping of kregbase1 failed\n");
183 		return -ENOMEM;
184 	}
185 	dd_dev_info(dd, "UC base1: %p for %x\n", dd->kregbase1, RCV_ARRAY);
186 	dd->chip_rcv_array_count = readq(dd->kregbase1 + RCV_ARRAY_CNT);
187 	dd_dev_info(dd, "RcvArray count: %u\n", dd->chip_rcv_array_count);
188 	dd->base2_start  = RCV_ARRAY + dd->chip_rcv_array_count * 8;
189 
190 	dd->kregbase2 = ioremap_nocache(
191 		addr + dd->base2_start,
192 		TXE_PIO_SEND - dd->base2_start);
193 	if (!dd->kregbase2) {
194 		dd_dev_err(dd, "UC mapping of kregbase2 failed\n");
195 		goto nomem;
196 	}
197 	dd_dev_info(dd, "UC base2: %p for %x\n", dd->kregbase2,
198 		    TXE_PIO_SEND - dd->base2_start);
199 
200 	dd->piobase = ioremap_wc(addr + TXE_PIO_SEND, TXE_PIO_SIZE);
201 	if (!dd->piobase) {
202 		dd_dev_err(dd, "WC mapping of send buffers failed\n");
203 		goto nomem;
204 	}
205 	dd_dev_info(dd, "WC piobase: %p\n for %x", dd->piobase, TXE_PIO_SIZE);
206 
207 	dd->physaddr = addr;        /* used for io_remap, etc. */
208 
209 	/*
210 	 * Map the chip's RcvArray as write-combining to allow us
211 	 * to write an entire cacheline worth of entries in one shot.
212 	 */
213 	dd->rcvarray_wc = ioremap_wc(addr + RCV_ARRAY,
214 				     dd->chip_rcv_array_count * 8);
215 	if (!dd->rcvarray_wc) {
216 		dd_dev_err(dd, "WC mapping of receive array failed\n");
217 		goto nomem;
218 	}
219 	dd_dev_info(dd, "WC RcvArray: %p for %x\n",
220 		    dd->rcvarray_wc, dd->chip_rcv_array_count * 8);
221 
222 	dd->flags |= HFI1_PRESENT;	/* chip.c CSR routines now work */
223 	return 0;
224 nomem:
225 	ret = -ENOMEM;
226 	hfi1_pcie_ddcleanup(dd);
227 	return ret;
228 }
229 
230 /*
231  * Do PCIe cleanup related to dd, after chip-specific cleanup, etc.  Just prior
232  * to releasing the dd memory.
233  * Void because all of the core pcie cleanup functions are void.
234  */
235 void hfi1_pcie_ddcleanup(struct hfi1_devdata *dd)
236 {
237 	dd->flags &= ~HFI1_PRESENT;
238 	if (dd->kregbase1)
239 		iounmap(dd->kregbase1);
240 	dd->kregbase1 = NULL;
241 	if (dd->kregbase2)
242 		iounmap(dd->kregbase2);
243 	dd->kregbase2 = NULL;
244 	if (dd->rcvarray_wc)
245 		iounmap(dd->rcvarray_wc);
246 	dd->rcvarray_wc = NULL;
247 	if (dd->piobase)
248 		iounmap(dd->piobase);
249 	dd->piobase = NULL;
250 }
251 
252 /* return the PCIe link speed from the given link status */
253 static u32 extract_speed(u16 linkstat)
254 {
255 	u32 speed;
256 
257 	switch (linkstat & PCI_EXP_LNKSTA_CLS) {
258 	default: /* not defined, assume Gen1 */
259 	case PCI_EXP_LNKSTA_CLS_2_5GB:
260 		speed = 2500; /* Gen 1, 2.5GHz */
261 		break;
262 	case PCI_EXP_LNKSTA_CLS_5_0GB:
263 		speed = 5000; /* Gen 2, 5GHz */
264 		break;
265 	case GEN3_SPEED_VECTOR:
266 		speed = 8000; /* Gen 3, 8GHz */
267 		break;
268 	}
269 	return speed;
270 }
271 
272 /* return the PCIe link speed from the given link status */
273 static u32 extract_width(u16 linkstat)
274 {
275 	return (linkstat & PCI_EXP_LNKSTA_NLW) >> PCI_EXP_LNKSTA_NLW_SHIFT;
276 }
277 
278 /* read the link status and set dd->{lbus_width,lbus_speed,lbus_info} */
279 static void update_lbus_info(struct hfi1_devdata *dd)
280 {
281 	u16 linkstat;
282 	int ret;
283 
284 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat);
285 	if (ret) {
286 		dd_dev_err(dd, "Unable to read from PCI config\n");
287 		return;
288 	}
289 
290 	dd->lbus_width = extract_width(linkstat);
291 	dd->lbus_speed = extract_speed(linkstat);
292 	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
293 		 "PCIe,%uMHz,x%u", dd->lbus_speed, dd->lbus_width);
294 }
295 
296 /*
297  * Read in the current PCIe link width and speed.  Find if the link is
298  * Gen3 capable.
299  */
300 int pcie_speeds(struct hfi1_devdata *dd)
301 {
302 	u32 linkcap;
303 	struct pci_dev *parent = dd->pcidev->bus->self;
304 	int ret;
305 
306 	if (!pci_is_pcie(dd->pcidev)) {
307 		dd_dev_err(dd, "Can't find PCI Express capability!\n");
308 		return -EINVAL;
309 	}
310 
311 	/* find if our max speed is Gen3 and parent supports Gen3 speeds */
312 	dd->link_gen3_capable = 1;
313 
314 	ret = pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &linkcap);
315 	if (ret) {
316 		dd_dev_err(dd, "Unable to read from PCI config\n");
317 		return ret;
318 	}
319 
320 	if ((linkcap & PCI_EXP_LNKCAP_SLS) != GEN3_SPEED_VECTOR) {
321 		dd_dev_info(dd,
322 			    "This HFI is not Gen3 capable, max speed 0x%x, need 0x3\n",
323 			    linkcap & PCI_EXP_LNKCAP_SLS);
324 		dd->link_gen3_capable = 0;
325 	}
326 
327 	/*
328 	 * bus->max_bus_speed is set from the bridge's linkcap Max Link Speed
329 	 */
330 	if (parent && dd->pcidev->bus->max_bus_speed != PCIE_SPEED_8_0GT) {
331 		dd_dev_info(dd, "Parent PCIe bridge does not support Gen3\n");
332 		dd->link_gen3_capable = 0;
333 	}
334 
335 	/* obtain the link width and current speed */
336 	update_lbus_info(dd);
337 
338 	dd_dev_info(dd, "%s\n", dd->lbus_info);
339 
340 	return 0;
341 }
342 
343 /*
344  * Returns:
345  *	- actual number of interrupts allocated or
346  *	- 0 if fell back to INTx.
347  *      - error
348  */
349 int request_msix(struct hfi1_devdata *dd, u32 msireq)
350 {
351 	int nvec;
352 
353 	nvec = pci_alloc_irq_vectors(dd->pcidev, 1, msireq,
354 				     PCI_IRQ_MSIX | PCI_IRQ_LEGACY);
355 	if (nvec < 0) {
356 		dd_dev_err(dd, "pci_alloc_irq_vectors() failed: %d\n", nvec);
357 		return nvec;
358 	}
359 
360 	tune_pcie_caps(dd);
361 
362 	/* check for legacy IRQ */
363 	if (nvec == 1 && !dd->pcidev->msix_enabled)
364 		return 0;
365 
366 	return nvec;
367 }
368 
369 /* restore command and BARs after a reset has wiped them out */
370 int restore_pci_variables(struct hfi1_devdata *dd)
371 {
372 	int ret = 0;
373 
374 	ret = pci_write_config_word(dd->pcidev, PCI_COMMAND, dd->pci_command);
375 	if (ret)
376 		goto error;
377 
378 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
379 				     dd->pcibar0);
380 	if (ret)
381 		goto error;
382 
383 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
384 				     dd->pcibar1);
385 	if (ret)
386 		goto error;
387 
388 	ret = pci_write_config_dword(dd->pcidev, PCI_ROM_ADDRESS, dd->pci_rom);
389 	if (ret)
390 		goto error;
391 
392 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL,
393 					 dd->pcie_devctl);
394 	if (ret)
395 		goto error;
396 
397 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL,
398 					 dd->pcie_lnkctl);
399 	if (ret)
400 		goto error;
401 
402 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL2,
403 					 dd->pcie_devctl2);
404 	if (ret)
405 		goto error;
406 
407 	ret = pci_write_config_dword(dd->pcidev, PCI_CFG_MSIX0, dd->pci_msix0);
408 	if (ret)
409 		goto error;
410 
411 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
412 		ret = pci_write_config_dword(dd->pcidev, PCIE_CFG_TPH2,
413 					     dd->pci_tph2);
414 		if (ret)
415 			goto error;
416 	}
417 	return 0;
418 
419 error:
420 	dd_dev_err(dd, "Unable to write to PCI config\n");
421 	return ret;
422 }
423 
424 /* Save BARs and command to rewrite after device reset */
425 int save_pci_variables(struct hfi1_devdata *dd)
426 {
427 	int ret = 0;
428 
429 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
430 				    &dd->pcibar0);
431 	if (ret)
432 		goto error;
433 
434 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
435 				    &dd->pcibar1);
436 	if (ret)
437 		goto error;
438 
439 	ret = pci_read_config_dword(dd->pcidev, PCI_ROM_ADDRESS, &dd->pci_rom);
440 	if (ret)
441 		goto error;
442 
443 	ret = pci_read_config_word(dd->pcidev, PCI_COMMAND, &dd->pci_command);
444 	if (ret)
445 		goto error;
446 
447 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL,
448 					&dd->pcie_devctl);
449 	if (ret)
450 		goto error;
451 
452 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL,
453 					&dd->pcie_lnkctl);
454 	if (ret)
455 		goto error;
456 
457 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL2,
458 					&dd->pcie_devctl2);
459 	if (ret)
460 		goto error;
461 
462 	ret = pci_read_config_dword(dd->pcidev, PCI_CFG_MSIX0, &dd->pci_msix0);
463 	if (ret)
464 		goto error;
465 
466 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
467 		ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_TPH2,
468 					    &dd->pci_tph2);
469 		if (ret)
470 			goto error;
471 	}
472 	return 0;
473 
474 error:
475 	dd_dev_err(dd, "Unable to read from PCI config\n");
476 	return ret;
477 }
478 
479 /*
480  * BIOS may not set PCIe bus-utilization parameters for best performance.
481  * Check and optionally adjust them to maximize our throughput.
482  */
483 static int hfi1_pcie_caps;
484 module_param_named(pcie_caps, hfi1_pcie_caps, int, S_IRUGO);
485 MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
486 
487 uint aspm_mode = ASPM_MODE_DISABLED;
488 module_param_named(aspm, aspm_mode, uint, S_IRUGO);
489 MODULE_PARM_DESC(aspm, "PCIe ASPM: 0: disable, 1: enable, 2: dynamic");
490 
491 static void tune_pcie_caps(struct hfi1_devdata *dd)
492 {
493 	struct pci_dev *parent;
494 	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
495 	u16 rc_mrrs, ep_mrrs, max_mrrs, ectl;
496 	int ret;
497 
498 	/*
499 	 * Turn on extended tags in DevCtl in case the BIOS has turned it off
500 	 * to improve WFR SDMA bandwidth
501 	 */
502 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL, &ectl);
503 	if ((!ret) && !(ectl & PCI_EXP_DEVCTL_EXT_TAG)) {
504 		dd_dev_info(dd, "Enabling PCIe extended tags\n");
505 		ectl |= PCI_EXP_DEVCTL_EXT_TAG;
506 		ret = pcie_capability_write_word(dd->pcidev,
507 						 PCI_EXP_DEVCTL, ectl);
508 		if (ret)
509 			dd_dev_info(dd, "Unable to write to PCI config\n");
510 	}
511 	/* Find out supported and configured values for parent (root) */
512 	parent = dd->pcidev->bus->self;
513 	/*
514 	 * The driver cannot perform the tuning if it does not have
515 	 * access to the upstream component.
516 	 */
517 	if (!parent) {
518 		dd_dev_info(dd, "Parent not found\n");
519 		return;
520 	}
521 	if (!pci_is_root_bus(parent->bus)) {
522 		dd_dev_info(dd, "Parent not root\n");
523 		return;
524 	}
525 	if (!pci_is_pcie(parent)) {
526 		dd_dev_info(dd, "Parent is not PCI Express capable\n");
527 		return;
528 	}
529 	if (!pci_is_pcie(dd->pcidev)) {
530 		dd_dev_info(dd, "PCI device is not PCI Express capable\n");
531 		return;
532 	}
533 	rc_mpss = parent->pcie_mpss;
534 	rc_mps = ffs(pcie_get_mps(parent)) - 8;
535 	/* Find out supported and configured values for endpoint (us) */
536 	ep_mpss = dd->pcidev->pcie_mpss;
537 	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
538 
539 	/* Find max payload supported by root, endpoint */
540 	if (rc_mpss > ep_mpss)
541 		rc_mpss = ep_mpss;
542 
543 	/* If Supported greater than limit in module param, limit it */
544 	if (rc_mpss > (hfi1_pcie_caps & 7))
545 		rc_mpss = hfi1_pcie_caps & 7;
546 	/* If less than (allowed, supported), bump root payload */
547 	if (rc_mpss > rc_mps) {
548 		rc_mps = rc_mpss;
549 		pcie_set_mps(parent, 128 << rc_mps);
550 	}
551 	/* If less than (allowed, supported), bump endpoint payload */
552 	if (rc_mpss > ep_mps) {
553 		ep_mps = rc_mpss;
554 		pcie_set_mps(dd->pcidev, 128 << ep_mps);
555 	}
556 
557 	/*
558 	 * Now the Read Request size.
559 	 * No field for max supported, but PCIe spec limits it to 4096,
560 	 * which is code '5' (log2(4096) - 7)
561 	 */
562 	max_mrrs = 5;
563 	if (max_mrrs > ((hfi1_pcie_caps >> 4) & 7))
564 		max_mrrs = (hfi1_pcie_caps >> 4) & 7;
565 
566 	max_mrrs = 128 << max_mrrs;
567 	rc_mrrs = pcie_get_readrq(parent);
568 	ep_mrrs = pcie_get_readrq(dd->pcidev);
569 
570 	if (max_mrrs > rc_mrrs) {
571 		rc_mrrs = max_mrrs;
572 		pcie_set_readrq(parent, rc_mrrs);
573 	}
574 	if (max_mrrs > ep_mrrs) {
575 		ep_mrrs = max_mrrs;
576 		pcie_set_readrq(dd->pcidev, ep_mrrs);
577 	}
578 }
579 
580 /* End of PCIe capability tuning */
581 
582 /*
583  * From here through hfi1_pci_err_handler definition is invoked via
584  * PCI error infrastructure, registered via pci
585  */
586 static pci_ers_result_t
587 pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
588 {
589 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
590 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
591 
592 	switch (state) {
593 	case pci_channel_io_normal:
594 		dd_dev_info(dd, "State Normal, ignoring\n");
595 		break;
596 
597 	case pci_channel_io_frozen:
598 		dd_dev_info(dd, "State Frozen, requesting reset\n");
599 		pci_disable_device(pdev);
600 		ret = PCI_ERS_RESULT_NEED_RESET;
601 		break;
602 
603 	case pci_channel_io_perm_failure:
604 		if (dd) {
605 			dd_dev_info(dd, "State Permanent Failure, disabling\n");
606 			/* no more register accesses! */
607 			dd->flags &= ~HFI1_PRESENT;
608 			hfi1_disable_after_error(dd);
609 		}
610 		 /* else early, or other problem */
611 		ret =  PCI_ERS_RESULT_DISCONNECT;
612 		break;
613 
614 	default: /* shouldn't happen */
615 		dd_dev_info(dd, "HFI1 PCI errors detected (state %d)\n",
616 			    state);
617 		break;
618 	}
619 	return ret;
620 }
621 
622 static pci_ers_result_t
623 pci_mmio_enabled(struct pci_dev *pdev)
624 {
625 	u64 words = 0U;
626 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
627 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
628 
629 	if (dd && dd->pport) {
630 		words = read_port_cntr(dd->pport, C_RX_WORDS, CNTR_INVALID_VL);
631 		if (words == ~0ULL)
632 			ret = PCI_ERS_RESULT_NEED_RESET;
633 		dd_dev_info(dd,
634 			    "HFI1 mmio_enabled function called, read wordscntr %llx, returning %d\n",
635 			    words, ret);
636 	}
637 	return  ret;
638 }
639 
640 static pci_ers_result_t
641 pci_slot_reset(struct pci_dev *pdev)
642 {
643 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
644 
645 	dd_dev_info(dd, "HFI1 slot_reset function called, ignored\n");
646 	return PCI_ERS_RESULT_CAN_RECOVER;
647 }
648 
649 static void
650 pci_resume(struct pci_dev *pdev)
651 {
652 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
653 
654 	dd_dev_info(dd, "HFI1 resume function called\n");
655 	pci_cleanup_aer_uncorrect_error_status(pdev);
656 	/*
657 	 * Running jobs will fail, since it's asynchronous
658 	 * unlike sysfs-requested reset.   Better than
659 	 * doing nothing.
660 	 */
661 	hfi1_init(dd, 1); /* same as re-init after reset */
662 }
663 
664 const struct pci_error_handlers hfi1_pci_err_handler = {
665 	.error_detected = pci_error_detected,
666 	.mmio_enabled = pci_mmio_enabled,
667 	.slot_reset = pci_slot_reset,
668 	.resume = pci_resume,
669 };
670 
671 /*============================================================================*/
672 /* PCIe Gen3 support */
673 
674 /*
675  * This code is separated out because it is expected to be removed in the
676  * final shipping product.  If not, then it will be revisited and items
677  * will be moved to more standard locations.
678  */
679 
680 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_STS field values */
681 #define DL_STATUS_HFI0 0x1	/* hfi0 firmware download complete */
682 #define DL_STATUS_HFI1 0x2	/* hfi1 firmware download complete */
683 #define DL_STATUS_BOTH 0x3	/* hfi0 and hfi1 firmware download complete */
684 
685 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_ERR field values */
686 #define DL_ERR_NONE		0x0	/* no error */
687 #define DL_ERR_SWAP_PARITY	0x1	/* parity error in SerDes interrupt */
688 					/*   or response data */
689 #define DL_ERR_DISABLED	0x2	/* hfi disabled */
690 #define DL_ERR_SECURITY	0x3	/* security check failed */
691 #define DL_ERR_SBUS		0x4	/* SBus status error */
692 #define DL_ERR_XFR_PARITY	0x5	/* parity error during ROM transfer*/
693 
694 /* gasket block secondary bus reset delay */
695 #define SBR_DELAY_US 200000	/* 200ms */
696 
697 /* mask for PCIe capability register lnkctl2 target link speed */
698 #define LNKCTL2_TARGET_LINK_SPEED_MASK 0xf
699 
700 static uint pcie_target = 3;
701 module_param(pcie_target, uint, S_IRUGO);
702 MODULE_PARM_DESC(pcie_target, "PCIe target speed (0 skip, 1-3 Gen1-3)");
703 
704 static uint pcie_force;
705 module_param(pcie_force, uint, S_IRUGO);
706 MODULE_PARM_DESC(pcie_force, "Force driver to do a PCIe firmware download even if already at target speed");
707 
708 static uint pcie_retry = 5;
709 module_param(pcie_retry, uint, S_IRUGO);
710 MODULE_PARM_DESC(pcie_retry, "Driver will try this many times to reach requested speed");
711 
712 #define UNSET_PSET 255
713 #define DEFAULT_DISCRETE_PSET 2	/* discrete HFI */
714 #define DEFAULT_MCP_PSET 6	/* MCP HFI */
715 static uint pcie_pset = UNSET_PSET;
716 module_param(pcie_pset, uint, S_IRUGO);
717 MODULE_PARM_DESC(pcie_pset, "PCIe Eq Pset value to use, range is 0-10");
718 
719 static uint pcie_ctle = 3; /* discrete on, integrated on */
720 module_param(pcie_ctle, uint, S_IRUGO);
721 MODULE_PARM_DESC(pcie_ctle, "PCIe static CTLE mode, bit 0 - discrete on/off, bit 1 - integrated on/off");
722 
723 /* equalization columns */
724 #define PREC 0
725 #define ATTN 1
726 #define POST 2
727 
728 /* discrete silicon preliminary equalization values */
729 static const u8 discrete_preliminary_eq[11][3] = {
730 	/* prec   attn   post */
731 	{  0x00,  0x00,  0x12 },	/* p0 */
732 	{  0x00,  0x00,  0x0c },	/* p1 */
733 	{  0x00,  0x00,  0x0f },	/* p2 */
734 	{  0x00,  0x00,  0x09 },	/* p3 */
735 	{  0x00,  0x00,  0x00 },	/* p4 */
736 	{  0x06,  0x00,  0x00 },	/* p5 */
737 	{  0x09,  0x00,  0x00 },	/* p6 */
738 	{  0x06,  0x00,  0x0f },	/* p7 */
739 	{  0x09,  0x00,  0x09 },	/* p8 */
740 	{  0x0c,  0x00,  0x00 },	/* p9 */
741 	{  0x00,  0x00,  0x18 },	/* p10 */
742 };
743 
744 /* integrated silicon preliminary equalization values */
745 static const u8 integrated_preliminary_eq[11][3] = {
746 	/* prec   attn   post */
747 	{  0x00,  0x1e,  0x07 },	/* p0 */
748 	{  0x00,  0x1e,  0x05 },	/* p1 */
749 	{  0x00,  0x1e,  0x06 },	/* p2 */
750 	{  0x00,  0x1e,  0x04 },	/* p3 */
751 	{  0x00,  0x1e,  0x00 },	/* p4 */
752 	{  0x03,  0x1e,  0x00 },	/* p5 */
753 	{  0x04,  0x1e,  0x00 },	/* p6 */
754 	{  0x03,  0x1e,  0x06 },	/* p7 */
755 	{  0x03,  0x1e,  0x04 },	/* p8 */
756 	{  0x05,  0x1e,  0x00 },	/* p9 */
757 	{  0x00,  0x1e,  0x0a },	/* p10 */
758 };
759 
760 static const u8 discrete_ctle_tunings[11][4] = {
761 	/* DC     LF     HF     BW */
762 	{  0x48,  0x0b,  0x04,  0x04 },	/* p0 */
763 	{  0x60,  0x05,  0x0f,  0x0a },	/* p1 */
764 	{  0x50,  0x09,  0x06,  0x06 },	/* p2 */
765 	{  0x68,  0x05,  0x0f,  0x0a },	/* p3 */
766 	{  0x80,  0x05,  0x0f,  0x0a },	/* p4 */
767 	{  0x70,  0x05,  0x0f,  0x0a },	/* p5 */
768 	{  0x68,  0x05,  0x0f,  0x0a },	/* p6 */
769 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
770 	{  0x48,  0x09,  0x06,  0x06 },	/* p8 */
771 	{  0x60,  0x05,  0x0f,  0x0a },	/* p9 */
772 	{  0x38,  0x0f,  0x00,  0x00 },	/* p10 */
773 };
774 
775 static const u8 integrated_ctle_tunings[11][4] = {
776 	/* DC     LF     HF     BW */
777 	{  0x38,  0x0f,  0x00,  0x00 },	/* p0 */
778 	{  0x38,  0x0f,  0x00,  0x00 },	/* p1 */
779 	{  0x38,  0x0f,  0x00,  0x00 },	/* p2 */
780 	{  0x38,  0x0f,  0x00,  0x00 },	/* p3 */
781 	{  0x58,  0x0a,  0x05,  0x05 },	/* p4 */
782 	{  0x48,  0x0a,  0x05,  0x05 },	/* p5 */
783 	{  0x40,  0x0a,  0x05,  0x05 },	/* p6 */
784 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
785 	{  0x38,  0x0f,  0x00,  0x00 },	/* p8 */
786 	{  0x38,  0x09,  0x06,  0x06 },	/* p9 */
787 	{  0x38,  0x0e,  0x01,  0x01 },	/* p10 */
788 };
789 
790 /* helper to format the value to write to hardware */
791 #define eq_value(pre, curr, post) \
792 	((((u32)(pre)) << \
793 			PCIE_CFG_REG_PL102_GEN3_EQ_PRE_CURSOR_PSET_SHIFT) \
794 	| (((u32)(curr)) << PCIE_CFG_REG_PL102_GEN3_EQ_CURSOR_PSET_SHIFT) \
795 	| (((u32)(post)) << \
796 		PCIE_CFG_REG_PL102_GEN3_EQ_POST_CURSOR_PSET_SHIFT))
797 
798 /*
799  * Load the given EQ preset table into the PCIe hardware.
800  */
801 static int load_eq_table(struct hfi1_devdata *dd, const u8 eq[11][3], u8 fs,
802 			 u8 div)
803 {
804 	struct pci_dev *pdev = dd->pcidev;
805 	u32 hit_error = 0;
806 	u32 violation;
807 	u32 i;
808 	u8 c_minus1, c0, c_plus1;
809 	int ret;
810 
811 	for (i = 0; i < 11; i++) {
812 		/* set index */
813 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL103, i);
814 		/* write the value */
815 		c_minus1 = eq[i][PREC] / div;
816 		c0 = fs - (eq[i][PREC] / div) - (eq[i][POST] / div);
817 		c_plus1 = eq[i][POST] / div;
818 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL102,
819 				       eq_value(c_minus1, c0, c_plus1));
820 		/* check if these coefficients violate EQ rules */
821 		ret = pci_read_config_dword(dd->pcidev,
822 					    PCIE_CFG_REG_PL105, &violation);
823 		if (ret) {
824 			dd_dev_err(dd, "Unable to read from PCI config\n");
825 			hit_error = 1;
826 			break;
827 		}
828 
829 		if (violation
830 		    & PCIE_CFG_REG_PL105_GEN3_EQ_VIOLATE_COEF_RULES_SMASK){
831 			if (hit_error == 0) {
832 				dd_dev_err(dd,
833 					   "Gen3 EQ Table Coefficient rule violations\n");
834 				dd_dev_err(dd, "         prec   attn   post\n");
835 			}
836 			dd_dev_err(dd, "   p%02d:   %02x     %02x     %02x\n",
837 				   i, (u32)eq[i][0], (u32)eq[i][1],
838 				   (u32)eq[i][2]);
839 			dd_dev_err(dd, "            %02x     %02x     %02x\n",
840 				   (u32)c_minus1, (u32)c0, (u32)c_plus1);
841 			hit_error = 1;
842 		}
843 	}
844 	if (hit_error)
845 		return -EINVAL;
846 	return 0;
847 }
848 
849 /*
850  * Steps to be done after the PCIe firmware is downloaded and
851  * before the SBR for the Pcie Gen3.
852  * The SBus resource is already being held.
853  */
854 static void pcie_post_steps(struct hfi1_devdata *dd)
855 {
856 	int i;
857 
858 	set_sbus_fast_mode(dd);
859 	/*
860 	 * Write to the PCIe PCSes to set the G3_LOCKED_NEXT bits to 1.
861 	 * This avoids a spurious framing error that can otherwise be
862 	 * generated by the MAC layer.
863 	 *
864 	 * Use individual addresses since no broadcast is set up.
865 	 */
866 	for (i = 0; i < NUM_PCIE_SERDES; i++) {
867 		sbus_request(dd, pcie_pcs_addrs[dd->hfi1_id][i],
868 			     0x03, WRITE_SBUS_RECEIVER, 0x00022132);
869 	}
870 
871 	clear_sbus_fast_mode(dd);
872 }
873 
874 /*
875  * Trigger a secondary bus reset (SBR) on ourselves using our parent.
876  *
877  * Based on pci_parent_bus_reset() which is not exported by the
878  * kernel core.
879  */
880 static int trigger_sbr(struct hfi1_devdata *dd)
881 {
882 	struct pci_dev *dev = dd->pcidev;
883 	struct pci_dev *pdev;
884 
885 	/* need a parent */
886 	if (!dev->bus->self) {
887 		dd_dev_err(dd, "%s: no parent device\n", __func__);
888 		return -ENOTTY;
889 	}
890 
891 	/* should not be anyone else on the bus */
892 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
893 		if (pdev != dev) {
894 			dd_dev_err(dd,
895 				   "%s: another device is on the same bus\n",
896 				   __func__);
897 			return -ENOTTY;
898 		}
899 
900 	/*
901 	 * A secondary bus reset (SBR) issues a hot reset to our device.
902 	 * The following routine does a 1s wait after the reset is dropped
903 	 * per PCI Trhfa (recovery time).  PCIe 3.0 section 6.6.1 -
904 	 * Conventional Reset, paragraph 3, line 35 also says that a 1s
905 	 * delay after a reset is required.  Per spec requirements,
906 	 * the link is either working or not after that point.
907 	 */
908 	pci_reset_bridge_secondary_bus(dev->bus->self);
909 
910 	return 0;
911 }
912 
913 /*
914  * Write the given gasket interrupt register.
915  */
916 static void write_gasket_interrupt(struct hfi1_devdata *dd, int index,
917 				   u16 code, u16 data)
918 {
919 	write_csr(dd, ASIC_PCIE_SD_INTRPT_LIST + (index * 8),
920 		  (((u64)code << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_CODE_SHIFT) |
921 		   ((u64)data << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_DATA_SHIFT)));
922 }
923 
924 /*
925  * Tell the gasket logic how to react to the reset.
926  */
927 static void arm_gasket_logic(struct hfi1_devdata *dd)
928 {
929 	u64 reg;
930 
931 	reg = (((u64)1 << dd->hfi1_id) <<
932 	       ASIC_PCIE_SD_HOST_CMD_INTRPT_CMD_SHIFT) |
933 	      ((u64)pcie_serdes_broadcast[dd->hfi1_id] <<
934 	       ASIC_PCIE_SD_HOST_CMD_SBUS_RCVR_ADDR_SHIFT |
935 	       ASIC_PCIE_SD_HOST_CMD_SBR_MODE_SMASK |
936 	       ((u64)SBR_DELAY_US & ASIC_PCIE_SD_HOST_CMD_TIMER_MASK) <<
937 	       ASIC_PCIE_SD_HOST_CMD_TIMER_SHIFT);
938 	write_csr(dd, ASIC_PCIE_SD_HOST_CMD, reg);
939 	/* read back to push the write */
940 	read_csr(dd, ASIC_PCIE_SD_HOST_CMD);
941 }
942 
943 /*
944  * CCE_PCIE_CTRL long name helpers
945  * We redefine these shorter macros to use in the code while leaving
946  * chip_registers.h to be autogenerated from the hardware spec.
947  */
948 #define LANE_BUNDLE_MASK              CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_MASK
949 #define LANE_BUNDLE_SHIFT             CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_SHIFT
950 #define LANE_DELAY_MASK               CCE_PCIE_CTRL_PCIE_LANE_DELAY_MASK
951 #define LANE_DELAY_SHIFT              CCE_PCIE_CTRL_PCIE_LANE_DELAY_SHIFT
952 #define MARGIN_OVERWRITE_ENABLE_SHIFT CCE_PCIE_CTRL_XMT_MARGIN_OVERWRITE_ENABLE_SHIFT
953 #define MARGIN_SHIFT                  CCE_PCIE_CTRL_XMT_MARGIN_SHIFT
954 #define MARGIN_G1_G2_OVERWRITE_MASK   CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_MASK
955 #define MARGIN_G1_G2_OVERWRITE_SHIFT  CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_SHIFT
956 #define MARGIN_GEN1_GEN2_MASK         CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_MASK
957 #define MARGIN_GEN1_GEN2_SHIFT        CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_SHIFT
958 
959  /*
960   * Write xmt_margin for full-swing (WFR-B) or half-swing (WFR-C).
961   */
962 static void write_xmt_margin(struct hfi1_devdata *dd, const char *fname)
963 {
964 	u64 pcie_ctrl;
965 	u64 xmt_margin;
966 	u64 xmt_margin_oe;
967 	u64 lane_delay;
968 	u64 lane_bundle;
969 
970 	pcie_ctrl = read_csr(dd, CCE_PCIE_CTRL);
971 
972 	/*
973 	 * For Discrete, use full-swing.
974 	 *  - PCIe TX defaults to full-swing.
975 	 *    Leave this register as default.
976 	 * For Integrated, use half-swing
977 	 *  - Copy xmt_margin and xmt_margin_oe
978 	 *    from Gen1/Gen2 to Gen3.
979 	 */
980 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL1) { /* integrated */
981 		/* extract initial fields */
982 		xmt_margin = (pcie_ctrl >> MARGIN_GEN1_GEN2_SHIFT)
983 			      & MARGIN_GEN1_GEN2_MASK;
984 		xmt_margin_oe = (pcie_ctrl >> MARGIN_G1_G2_OVERWRITE_SHIFT)
985 				 & MARGIN_G1_G2_OVERWRITE_MASK;
986 		lane_delay = (pcie_ctrl >> LANE_DELAY_SHIFT) & LANE_DELAY_MASK;
987 		lane_bundle = (pcie_ctrl >> LANE_BUNDLE_SHIFT)
988 			       & LANE_BUNDLE_MASK;
989 
990 		/*
991 		 * For A0, EFUSE values are not set.  Override with the
992 		 * correct values.
993 		 */
994 		if (is_ax(dd)) {
995 			/*
996 			 * xmt_margin and OverwiteEnabel should be the
997 			 * same for Gen1/Gen2 and Gen3
998 			 */
999 			xmt_margin = 0x5;
1000 			xmt_margin_oe = 0x1;
1001 			lane_delay = 0xF; /* Delay 240ns. */
1002 			lane_bundle = 0x0; /* Set to 1 lane. */
1003 		}
1004 
1005 		/* overwrite existing values */
1006 		pcie_ctrl = (xmt_margin << MARGIN_GEN1_GEN2_SHIFT)
1007 			| (xmt_margin_oe << MARGIN_G1_G2_OVERWRITE_SHIFT)
1008 			| (xmt_margin << MARGIN_SHIFT)
1009 			| (xmt_margin_oe << MARGIN_OVERWRITE_ENABLE_SHIFT)
1010 			| (lane_delay << LANE_DELAY_SHIFT)
1011 			| (lane_bundle << LANE_BUNDLE_SHIFT);
1012 
1013 		write_csr(dd, CCE_PCIE_CTRL, pcie_ctrl);
1014 	}
1015 
1016 	dd_dev_dbg(dd, "%s: program XMT margin, CcePcieCtrl 0x%llx\n",
1017 		   fname, pcie_ctrl);
1018 }
1019 
1020 /*
1021  * Do all the steps needed to transition the PCIe link to Gen3 speed.
1022  */
1023 int do_pcie_gen3_transition(struct hfi1_devdata *dd)
1024 {
1025 	struct pci_dev *parent = dd->pcidev->bus->self;
1026 	u64 fw_ctrl;
1027 	u64 reg, therm;
1028 	u32 reg32, fs, lf;
1029 	u32 status, err;
1030 	int ret;
1031 	int do_retry, retry_count = 0;
1032 	int intnum = 0;
1033 	uint default_pset;
1034 	uint pset = pcie_pset;
1035 	u16 target_vector, target_speed;
1036 	u16 lnkctl2, vendor;
1037 	u8 div;
1038 	const u8 (*eq)[3];
1039 	const u8 (*ctle_tunings)[4];
1040 	uint static_ctle_mode;
1041 	int return_error = 0;
1042 
1043 	/* PCIe Gen3 is for the ASIC only */
1044 	if (dd->icode != ICODE_RTL_SILICON)
1045 		return 0;
1046 
1047 	if (pcie_target == 1) {			/* target Gen1 */
1048 		target_vector = GEN1_SPEED_VECTOR;
1049 		target_speed = 2500;
1050 	} else if (pcie_target == 2) {		/* target Gen2 */
1051 		target_vector = GEN2_SPEED_VECTOR;
1052 		target_speed = 5000;
1053 	} else if (pcie_target == 3) {		/* target Gen3 */
1054 		target_vector = GEN3_SPEED_VECTOR;
1055 		target_speed = 8000;
1056 	} else {
1057 		/* off or invalid target - skip */
1058 		dd_dev_info(dd, "%s: Skipping PCIe transition\n", __func__);
1059 		return 0;
1060 	}
1061 
1062 	/* if already at target speed, done (unless forced) */
1063 	if (dd->lbus_speed == target_speed) {
1064 		dd_dev_info(dd, "%s: PCIe already at gen%d, %s\n", __func__,
1065 			    pcie_target,
1066 			    pcie_force ? "re-doing anyway" : "skipping");
1067 		if (!pcie_force)
1068 			return 0;
1069 	}
1070 
1071 	/*
1072 	 * The driver cannot do the transition if it has no access to the
1073 	 * upstream component
1074 	 */
1075 	if (!parent) {
1076 		dd_dev_info(dd, "%s: No upstream, Can't do gen3 transition\n",
1077 			    __func__);
1078 		return 0;
1079 	}
1080 
1081 	/*
1082 	 * Do the Gen3 transition.  Steps are those of the PCIe Gen3
1083 	 * recipe.
1084 	 */
1085 
1086 	/* step 1: pcie link working in gen1/gen2 */
1087 
1088 	/* step 2: if either side is not capable of Gen3, done */
1089 	if (pcie_target == 3 && !dd->link_gen3_capable) {
1090 		dd_dev_err(dd, "The PCIe link is not Gen3 capable\n");
1091 		ret = -ENOSYS;
1092 		goto done_no_mutex;
1093 	}
1094 
1095 	/* hold the SBus resource across the firmware download and SBR */
1096 	ret = acquire_chip_resource(dd, CR_SBUS, SBUS_TIMEOUT);
1097 	if (ret) {
1098 		dd_dev_err(dd, "%s: unable to acquire SBus resource\n",
1099 			   __func__);
1100 		return ret;
1101 	}
1102 
1103 	/* make sure thermal polling is not causing interrupts */
1104 	therm = read_csr(dd, ASIC_CFG_THERM_POLL_EN);
1105 	if (therm) {
1106 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x0);
1107 		msleep(100);
1108 		dd_dev_info(dd, "%s: Disabled therm polling\n",
1109 			    __func__);
1110 	}
1111 
1112 retry:
1113 	/* the SBus download will reset the spico for thermal */
1114 
1115 	/* step 3: download SBus Master firmware */
1116 	/* step 4: download PCIe Gen3 SerDes firmware */
1117 	dd_dev_info(dd, "%s: downloading firmware\n", __func__);
1118 	ret = load_pcie_firmware(dd);
1119 	if (ret) {
1120 		/* do not proceed if the firmware cannot be downloaded */
1121 		return_error = 1;
1122 		goto done;
1123 	}
1124 
1125 	/* step 5: set up device parameter settings */
1126 	dd_dev_info(dd, "%s: setting PCIe registers\n", __func__);
1127 
1128 	/*
1129 	 * PcieCfgSpcie1 - Link Control 3
1130 	 * Leave at reset value.  No need to set PerfEq - link equalization
1131 	 * will be performed automatically after the SBR when the target
1132 	 * speed is 8GT/s.
1133 	 */
1134 
1135 	/* clear all 16 per-lane error bits (PCIe: Lane Error Status) */
1136 	pci_write_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, 0xffff);
1137 
1138 	/* step 5a: Set Synopsys Port Logic registers */
1139 
1140 	/*
1141 	 * PcieCfgRegPl2 - Port Force Link
1142 	 *
1143 	 * Set the low power field to 0x10 to avoid unnecessary power
1144 	 * management messages.  All other fields are zero.
1145 	 */
1146 	reg32 = 0x10ul << PCIE_CFG_REG_PL2_LOW_PWR_ENT_CNT_SHIFT;
1147 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL2, reg32);
1148 
1149 	/*
1150 	 * PcieCfgRegPl100 - Gen3 Control
1151 	 *
1152 	 * turn off PcieCfgRegPl100.Gen3ZRxDcNonCompl
1153 	 * turn on PcieCfgRegPl100.EqEieosCnt
1154 	 * Everything else zero.
1155 	 */
1156 	reg32 = PCIE_CFG_REG_PL100_EQ_EIEOS_CNT_SMASK;
1157 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL100, reg32);
1158 
1159 	/*
1160 	 * PcieCfgRegPl101 - Gen3 EQ FS and LF
1161 	 * PcieCfgRegPl102 - Gen3 EQ Presets to Coefficients Mapping
1162 	 * PcieCfgRegPl103 - Gen3 EQ Preset Index
1163 	 * PcieCfgRegPl105 - Gen3 EQ Status
1164 	 *
1165 	 * Give initial EQ settings.
1166 	 */
1167 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL0) { /* discrete */
1168 		/* 1000mV, FS=24, LF = 8 */
1169 		fs = 24;
1170 		lf = 8;
1171 		div = 3;
1172 		eq = discrete_preliminary_eq;
1173 		default_pset = DEFAULT_DISCRETE_PSET;
1174 		ctle_tunings = discrete_ctle_tunings;
1175 		/* bit 0 - discrete on/off */
1176 		static_ctle_mode = pcie_ctle & 0x1;
1177 	} else {
1178 		/* 400mV, FS=29, LF = 9 */
1179 		fs = 29;
1180 		lf = 9;
1181 		div = 1;
1182 		eq = integrated_preliminary_eq;
1183 		default_pset = DEFAULT_MCP_PSET;
1184 		ctle_tunings = integrated_ctle_tunings;
1185 		/* bit 1 - integrated on/off */
1186 		static_ctle_mode = (pcie_ctle >> 1) & 0x1;
1187 	}
1188 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL101,
1189 			       (fs <<
1190 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_FS_SHIFT) |
1191 			       (lf <<
1192 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_LF_SHIFT));
1193 	ret = load_eq_table(dd, eq, fs, div);
1194 	if (ret)
1195 		goto done;
1196 
1197 	/*
1198 	 * PcieCfgRegPl106 - Gen3 EQ Control
1199 	 *
1200 	 * Set Gen3EqPsetReqVec, leave other fields 0.
1201 	 */
1202 	if (pset == UNSET_PSET)
1203 		pset = default_pset;
1204 	if (pset > 10) {	/* valid range is 0-10, inclusive */
1205 		dd_dev_err(dd, "%s: Invalid Eq Pset %u, setting to %d\n",
1206 			   __func__, pset, default_pset);
1207 		pset = default_pset;
1208 	}
1209 	dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pset);
1210 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL106,
1211 			       ((1 << pset) <<
1212 			PCIE_CFG_REG_PL106_GEN3_EQ_PSET_REQ_VEC_SHIFT) |
1213 			PCIE_CFG_REG_PL106_GEN3_EQ_EVAL2MS_DISABLE_SMASK |
1214 			PCIE_CFG_REG_PL106_GEN3_EQ_PHASE23_EXIT_MODE_SMASK);
1215 
1216 	/*
1217 	 * step 5b: Do post firmware download steps via SBus
1218 	 */
1219 	dd_dev_info(dd, "%s: doing pcie post steps\n", __func__);
1220 	pcie_post_steps(dd);
1221 
1222 	/*
1223 	 * step 5c: Program gasket interrupts
1224 	 */
1225 	/* set the Rx Bit Rate to REFCLK ratio */
1226 	write_gasket_interrupt(dd, intnum++, 0x0006, 0x0050);
1227 	/* disable pCal for PCIe Gen3 RX equalization */
1228 	/* select adaptive or static CTLE */
1229 	write_gasket_interrupt(dd, intnum++, 0x0026,
1230 			       0x5b01 | (static_ctle_mode << 3));
1231 	/*
1232 	 * Enable iCal for PCIe Gen3 RX equalization, and set which
1233 	 * evaluation of RX_EQ_EVAL will launch the iCal procedure.
1234 	 */
1235 	write_gasket_interrupt(dd, intnum++, 0x0026, 0x5202);
1236 
1237 	if (static_ctle_mode) {
1238 		/* apply static CTLE tunings */
1239 		u8 pcie_dc, pcie_lf, pcie_hf, pcie_bw;
1240 
1241 		pcie_dc = ctle_tunings[pset][0];
1242 		pcie_lf = ctle_tunings[pset][1];
1243 		pcie_hf = ctle_tunings[pset][2];
1244 		pcie_bw = ctle_tunings[pset][3];
1245 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0200 | pcie_dc);
1246 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0100 | pcie_lf);
1247 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0000 | pcie_hf);
1248 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x5500 | pcie_bw);
1249 	}
1250 
1251 	/* terminate list */
1252 	write_gasket_interrupt(dd, intnum++, 0x0000, 0x0000);
1253 
1254 	/*
1255 	 * step 5d: program XMT margin
1256 	 */
1257 	write_xmt_margin(dd, __func__);
1258 
1259 	/*
1260 	 * step 5e: disable active state power management (ASPM). It
1261 	 * will be enabled if required later
1262 	 */
1263 	dd_dev_info(dd, "%s: clearing ASPM\n", __func__);
1264 	aspm_hw_disable_l1(dd);
1265 
1266 	/*
1267 	 * step 5f: clear DirectSpeedChange
1268 	 * PcieCfgRegPl67.DirectSpeedChange must be zero to prevent the
1269 	 * change in the speed target from starting before we are ready.
1270 	 * This field defaults to 0 and we are not changing it, so nothing
1271 	 * needs to be done.
1272 	 */
1273 
1274 	/* step 5g: Set target link speed */
1275 	/*
1276 	 * Set target link speed to be target on both device and parent.
1277 	 * On setting the parent: Some system BIOSs "helpfully" set the
1278 	 * parent target speed to Gen2 to match the ASIC's initial speed.
1279 	 * We can set the target Gen3 because we have already checked
1280 	 * that it is Gen3 capable earlier.
1281 	 */
1282 	dd_dev_info(dd, "%s: setting parent target link speed\n", __func__);
1283 	ret = pcie_capability_read_word(parent, PCI_EXP_LNKCTL2, &lnkctl2);
1284 	if (ret) {
1285 		dd_dev_err(dd, "Unable to read from PCI config\n");
1286 		return_error = 1;
1287 		goto done;
1288 	}
1289 
1290 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1291 		    (u32)lnkctl2);
1292 	/* only write to parent if target is not as high as ours */
1293 	if ((lnkctl2 & LNKCTL2_TARGET_LINK_SPEED_MASK) < target_vector) {
1294 		lnkctl2 &= ~LNKCTL2_TARGET_LINK_SPEED_MASK;
1295 		lnkctl2 |= target_vector;
1296 		dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1297 			    (u32)lnkctl2);
1298 		ret = pcie_capability_write_word(parent,
1299 						 PCI_EXP_LNKCTL2, lnkctl2);
1300 		if (ret) {
1301 			dd_dev_err(dd, "Unable to write to PCI config\n");
1302 			return_error = 1;
1303 			goto done;
1304 		}
1305 	} else {
1306 		dd_dev_info(dd, "%s: ..target speed is OK\n", __func__);
1307 	}
1308 
1309 	dd_dev_info(dd, "%s: setting target link speed\n", __func__);
1310 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL2, &lnkctl2);
1311 	if (ret) {
1312 		dd_dev_err(dd, "Unable to read from PCI config\n");
1313 		return_error = 1;
1314 		goto done;
1315 	}
1316 
1317 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1318 		    (u32)lnkctl2);
1319 	lnkctl2 &= ~LNKCTL2_TARGET_LINK_SPEED_MASK;
1320 	lnkctl2 |= target_vector;
1321 	dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1322 		    (u32)lnkctl2);
1323 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL2, lnkctl2);
1324 	if (ret) {
1325 		dd_dev_err(dd, "Unable to write to PCI config\n");
1326 		return_error = 1;
1327 		goto done;
1328 	}
1329 
1330 	/* step 5h: arm gasket logic */
1331 	/* hold DC in reset across the SBR */
1332 	write_csr(dd, CCE_DC_CTRL, CCE_DC_CTRL_DC_RESET_SMASK);
1333 	(void)read_csr(dd, CCE_DC_CTRL); /* DC reset hold */
1334 	/* save firmware control across the SBR */
1335 	fw_ctrl = read_csr(dd, MISC_CFG_FW_CTRL);
1336 
1337 	dd_dev_info(dd, "%s: arming gasket logic\n", __func__);
1338 	arm_gasket_logic(dd);
1339 
1340 	/*
1341 	 * step 6: quiesce PCIe link
1342 	 * The chip has already been reset, so there will be no traffic
1343 	 * from the chip.  Linux has no easy way to enforce that it will
1344 	 * not try to access the device, so we just need to hope it doesn't
1345 	 * do it while we are doing the reset.
1346 	 */
1347 
1348 	/*
1349 	 * step 7: initiate the secondary bus reset (SBR)
1350 	 * step 8: hardware brings the links back up
1351 	 * step 9: wait for link speed transition to be complete
1352 	 */
1353 	dd_dev_info(dd, "%s: calling trigger_sbr\n", __func__);
1354 	ret = trigger_sbr(dd);
1355 	if (ret)
1356 		goto done;
1357 
1358 	/* step 10: decide what to do next */
1359 
1360 	/* check if we can read PCI space */
1361 	ret = pci_read_config_word(dd->pcidev, PCI_VENDOR_ID, &vendor);
1362 	if (ret) {
1363 		dd_dev_info(dd,
1364 			    "%s: read of VendorID failed after SBR, err %d\n",
1365 			    __func__, ret);
1366 		return_error = 1;
1367 		goto done;
1368 	}
1369 	if (vendor == 0xffff) {
1370 		dd_dev_info(dd, "%s: VendorID is all 1s after SBR\n", __func__);
1371 		return_error = 1;
1372 		ret = -EIO;
1373 		goto done;
1374 	}
1375 
1376 	/* restore PCI space registers we know were reset */
1377 	dd_dev_info(dd, "%s: calling restore_pci_variables\n", __func__);
1378 	ret = restore_pci_variables(dd);
1379 	if (ret) {
1380 		dd_dev_err(dd, "%s: Could not restore PCI variables\n",
1381 			   __func__);
1382 		return_error = 1;
1383 		goto done;
1384 	}
1385 
1386 	/* restore firmware control */
1387 	write_csr(dd, MISC_CFG_FW_CTRL, fw_ctrl);
1388 
1389 	/*
1390 	 * Check the gasket block status.
1391 	 *
1392 	 * This is the first CSR read after the SBR.  If the read returns
1393 	 * all 1s (fails), the link did not make it back.
1394 	 *
1395 	 * Once we're sure we can read and write, clear the DC reset after
1396 	 * the SBR.  Then check for any per-lane errors. Then look over
1397 	 * the status.
1398 	 */
1399 	reg = read_csr(dd, ASIC_PCIE_SD_HOST_STATUS);
1400 	dd_dev_info(dd, "%s: gasket block status: 0x%llx\n", __func__, reg);
1401 	if (reg == ~0ull) {	/* PCIe read failed/timeout */
1402 		dd_dev_err(dd, "SBR failed - unable to read from device\n");
1403 		return_error = 1;
1404 		ret = -ENOSYS;
1405 		goto done;
1406 	}
1407 
1408 	/* clear the DC reset */
1409 	write_csr(dd, CCE_DC_CTRL, 0);
1410 
1411 	/* Set the LED off */
1412 	setextled(dd, 0);
1413 
1414 	/* check for any per-lane errors */
1415 	ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, &reg32);
1416 	if (ret) {
1417 		dd_dev_err(dd, "Unable to read from PCI config\n");
1418 		return_error = 1;
1419 		goto done;
1420 	}
1421 
1422 	dd_dev_info(dd, "%s: per-lane errors: 0x%x\n", __func__, reg32);
1423 
1424 	/* extract status, look for our HFI */
1425 	status = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_SHIFT)
1426 			& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_MASK;
1427 	if ((status & (1 << dd->hfi1_id)) == 0) {
1428 		dd_dev_err(dd,
1429 			   "%s: gasket status 0x%x, expecting 0x%x\n",
1430 			   __func__, status, 1 << dd->hfi1_id);
1431 		ret = -EIO;
1432 		goto done;
1433 	}
1434 
1435 	/* extract error */
1436 	err = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_SHIFT)
1437 		& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_MASK;
1438 	if (err) {
1439 		dd_dev_err(dd, "%s: gasket error %d\n", __func__, err);
1440 		ret = -EIO;
1441 		goto done;
1442 	}
1443 
1444 	/* update our link information cache */
1445 	update_lbus_info(dd);
1446 	dd_dev_info(dd, "%s: new speed and width: %s\n", __func__,
1447 		    dd->lbus_info);
1448 
1449 	if (dd->lbus_speed != target_speed) { /* not target */
1450 		/* maybe retry */
1451 		do_retry = retry_count < pcie_retry;
1452 		dd_dev_err(dd, "PCIe link speed did not switch to Gen%d%s\n",
1453 			   pcie_target, do_retry ? ", retrying" : "");
1454 		retry_count++;
1455 		if (do_retry) {
1456 			msleep(100); /* allow time to settle */
1457 			goto retry;
1458 		}
1459 		ret = -EIO;
1460 	}
1461 
1462 done:
1463 	if (therm) {
1464 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x1);
1465 		msleep(100);
1466 		dd_dev_info(dd, "%s: Re-enable therm polling\n",
1467 			    __func__);
1468 	}
1469 	release_chip_resource(dd, CR_SBUS);
1470 done_no_mutex:
1471 	/* return no error if it is OK to be at current speed */
1472 	if (ret && !return_error) {
1473 		dd_dev_err(dd, "Proceeding at current speed PCIe speed\n");
1474 		ret = 0;
1475 	}
1476 
1477 	dd_dev_info(dd, "%s: done\n", __func__);
1478 	return ret;
1479 }
1480