• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/bitfield.h>
49 #include <linux/pci.h>
50 #include <linux/io.h>
51 #include <linux/delay.h>
52 #include <linux/vmalloc.h>
53 #include <linux/aer.h>
54 #include <linux/module.h>
55 
56 #include "hfi.h"
57 #include "chip_registers.h"
58 #include "aspm.h"
59 
60 /*
61  * This file contains PCIe utility routines.
62  */
63 
64 /*
65  * Do all the common PCIe setup and initialization.
66  */
hfi1_pcie_init(struct hfi1_devdata * dd)67 int hfi1_pcie_init(struct hfi1_devdata *dd)
68 {
69 	int ret;
70 	struct pci_dev *pdev = dd->pcidev;
71 
72 	ret = pci_enable_device(pdev);
73 	if (ret) {
74 		/*
75 		 * This can happen (in theory) iff:
76 		 * We did a chip reset, and then failed to reprogram the
77 		 * BAR, or the chip reset due to an internal error.  We then
78 		 * unloaded the driver and reloaded it.
79 		 *
80 		 * Both reset cases set the BAR back to initial state.  For
81 		 * the latter case, the AER sticky error bit at offset 0x718
82 		 * should be set, but the Linux kernel doesn't yet know
83 		 * about that, it appears.  If the original BAR was retained
84 		 * in the kernel data structures, this may be OK.
85 		 */
86 		dd_dev_err(dd, "pci enable failed: error %d\n", -ret);
87 		return ret;
88 	}
89 
90 	ret = pci_request_regions(pdev, DRIVER_NAME);
91 	if (ret) {
92 		dd_dev_err(dd, "pci_request_regions fails: err %d\n", -ret);
93 		goto bail;
94 	}
95 
96 	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
97 	if (ret) {
98 		/*
99 		 * If the 64 bit setup fails, try 32 bit.  Some systems
100 		 * do not setup 64 bit maps on systems with 2GB or less
101 		 * memory installed.
102 		 */
103 		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
104 		if (ret) {
105 			dd_dev_err(dd, "Unable to set DMA mask: %d\n", ret);
106 			goto bail;
107 		}
108 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
109 	} else {
110 		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
111 	}
112 	if (ret) {
113 		dd_dev_err(dd, "Unable to set DMA consistent mask: %d\n", ret);
114 		goto bail;
115 	}
116 
117 	pci_set_master(pdev);
118 	(void)pci_enable_pcie_error_reporting(pdev);
119 	return 0;
120 
121 bail:
122 	hfi1_pcie_cleanup(pdev);
123 	return ret;
124 }
125 
126 /*
127  * Clean what was done in hfi1_pcie_init()
128  */
hfi1_pcie_cleanup(struct pci_dev * pdev)129 void hfi1_pcie_cleanup(struct pci_dev *pdev)
130 {
131 	pci_disable_device(pdev);
132 	/*
133 	 * Release regions should be called after the disable. OK to
134 	 * call if request regions has not been called or failed.
135 	 */
136 	pci_release_regions(pdev);
137 }
138 
139 /*
140  * Do remaining PCIe setup, once dd is allocated, and save away
141  * fields required to re-initialize after a chip reset, or for
142  * various other purposes
143  */
hfi1_pcie_ddinit(struct hfi1_devdata * dd,struct pci_dev * pdev)144 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev)
145 {
146 	unsigned long len;
147 	resource_size_t addr;
148 	int ret = 0;
149 	u32 rcv_array_count;
150 
151 	addr = pci_resource_start(pdev, 0);
152 	len = pci_resource_len(pdev, 0);
153 
154 	/*
155 	 * The TXE PIO buffers are at the tail end of the chip space.
156 	 * Cut them off and map them separately.
157 	 */
158 
159 	/* sanity check vs expectations */
160 	if (len != TXE_PIO_SEND + TXE_PIO_SIZE) {
161 		dd_dev_err(dd, "chip PIO range does not match\n");
162 		return -EINVAL;
163 	}
164 
165 	dd->kregbase1 = ioremap(addr, RCV_ARRAY);
166 	if (!dd->kregbase1) {
167 		dd_dev_err(dd, "UC mapping of kregbase1 failed\n");
168 		return -ENOMEM;
169 	}
170 	dd_dev_info(dd, "UC base1: %p for %x\n", dd->kregbase1, RCV_ARRAY);
171 
172 	/* verify that reads actually work, save revision for reset check */
173 	dd->revision = readq(dd->kregbase1 + CCE_REVISION);
174 	if (dd->revision == ~(u64)0) {
175 		dd_dev_err(dd, "Cannot read chip CSRs\n");
176 		goto nomem;
177 	}
178 
179 	rcv_array_count = readq(dd->kregbase1 + RCV_ARRAY_CNT);
180 	dd_dev_info(dd, "RcvArray count: %u\n", rcv_array_count);
181 	dd->base2_start  = RCV_ARRAY + rcv_array_count * 8;
182 
183 	dd->kregbase2 = ioremap(
184 		addr + dd->base2_start,
185 		TXE_PIO_SEND - dd->base2_start);
186 	if (!dd->kregbase2) {
187 		dd_dev_err(dd, "UC mapping of kregbase2 failed\n");
188 		goto nomem;
189 	}
190 	dd_dev_info(dd, "UC base2: %p for %x\n", dd->kregbase2,
191 		    TXE_PIO_SEND - dd->base2_start);
192 
193 	dd->piobase = ioremap_wc(addr + TXE_PIO_SEND, TXE_PIO_SIZE);
194 	if (!dd->piobase) {
195 		dd_dev_err(dd, "WC mapping of send buffers failed\n");
196 		goto nomem;
197 	}
198 	dd_dev_info(dd, "WC piobase: %p for %x\n", dd->piobase, TXE_PIO_SIZE);
199 
200 	dd->physaddr = addr;        /* used for io_remap, etc. */
201 
202 	/*
203 	 * Map the chip's RcvArray as write-combining to allow us
204 	 * to write an entire cacheline worth of entries in one shot.
205 	 */
206 	dd->rcvarray_wc = ioremap_wc(addr + RCV_ARRAY,
207 				     rcv_array_count * 8);
208 	if (!dd->rcvarray_wc) {
209 		dd_dev_err(dd, "WC mapping of receive array failed\n");
210 		goto nomem;
211 	}
212 	dd_dev_info(dd, "WC RcvArray: %p for %x\n",
213 		    dd->rcvarray_wc, rcv_array_count * 8);
214 
215 	dd->flags |= HFI1_PRESENT;	/* chip.c CSR routines now work */
216 	return 0;
217 nomem:
218 	ret = -ENOMEM;
219 	hfi1_pcie_ddcleanup(dd);
220 	return ret;
221 }
222 
223 /*
224  * Do PCIe cleanup related to dd, after chip-specific cleanup, etc.  Just prior
225  * to releasing the dd memory.
226  * Void because all of the core pcie cleanup functions are void.
227  */
hfi1_pcie_ddcleanup(struct hfi1_devdata * dd)228 void hfi1_pcie_ddcleanup(struct hfi1_devdata *dd)
229 {
230 	dd->flags &= ~HFI1_PRESENT;
231 	if (dd->kregbase1)
232 		iounmap(dd->kregbase1);
233 	dd->kregbase1 = NULL;
234 	if (dd->kregbase2)
235 		iounmap(dd->kregbase2);
236 	dd->kregbase2 = NULL;
237 	if (dd->rcvarray_wc)
238 		iounmap(dd->rcvarray_wc);
239 	dd->rcvarray_wc = NULL;
240 	if (dd->piobase)
241 		iounmap(dd->piobase);
242 	dd->piobase = NULL;
243 }
244 
245 /* return the PCIe link speed from the given link status */
extract_speed(u16 linkstat)246 static u32 extract_speed(u16 linkstat)
247 {
248 	u32 speed;
249 
250 	switch (linkstat & PCI_EXP_LNKSTA_CLS) {
251 	default: /* not defined, assume Gen1 */
252 	case PCI_EXP_LNKSTA_CLS_2_5GB:
253 		speed = 2500; /* Gen 1, 2.5GHz */
254 		break;
255 	case PCI_EXP_LNKSTA_CLS_5_0GB:
256 		speed = 5000; /* Gen 2, 5GHz */
257 		break;
258 	case PCI_EXP_LNKSTA_CLS_8_0GB:
259 		speed = 8000; /* Gen 3, 8GHz */
260 		break;
261 	}
262 	return speed;
263 }
264 
265 /* read the link status and set dd->{lbus_width,lbus_speed,lbus_info} */
update_lbus_info(struct hfi1_devdata * dd)266 static void update_lbus_info(struct hfi1_devdata *dd)
267 {
268 	u16 linkstat;
269 	int ret;
270 
271 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat);
272 	if (ret) {
273 		dd_dev_err(dd, "Unable to read from PCI config\n");
274 		return;
275 	}
276 
277 	dd->lbus_width = FIELD_GET(PCI_EXP_LNKSTA_NLW, linkstat);
278 	dd->lbus_speed = extract_speed(linkstat);
279 	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
280 		 "PCIe,%uMHz,x%u", dd->lbus_speed, dd->lbus_width);
281 }
282 
283 /*
284  * Read in the current PCIe link width and speed.  Find if the link is
285  * Gen3 capable.
286  */
pcie_speeds(struct hfi1_devdata * dd)287 int pcie_speeds(struct hfi1_devdata *dd)
288 {
289 	u32 linkcap;
290 	struct pci_dev *parent = dd->pcidev->bus->self;
291 	int ret;
292 
293 	if (!pci_is_pcie(dd->pcidev)) {
294 		dd_dev_err(dd, "Can't find PCI Express capability!\n");
295 		return -EINVAL;
296 	}
297 
298 	/* find if our max speed is Gen3 and parent supports Gen3 speeds */
299 	dd->link_gen3_capable = 1;
300 
301 	ret = pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &linkcap);
302 	if (ret) {
303 		dd_dev_err(dd, "Unable to read from PCI config\n");
304 		return pcibios_err_to_errno(ret);
305 	}
306 
307 	if ((linkcap & PCI_EXP_LNKCAP_SLS) != PCI_EXP_LNKCAP_SLS_8_0GB) {
308 		dd_dev_info(dd,
309 			    "This HFI is not Gen3 capable, max speed 0x%x, need 0x3\n",
310 			    linkcap & PCI_EXP_LNKCAP_SLS);
311 		dd->link_gen3_capable = 0;
312 	}
313 
314 	/*
315 	 * bus->max_bus_speed is set from the bridge's linkcap Max Link Speed
316 	 */
317 	if (parent &&
318 	    (dd->pcidev->bus->max_bus_speed == PCIE_SPEED_2_5GT ||
319 	     dd->pcidev->bus->max_bus_speed == PCIE_SPEED_5_0GT)) {
320 		dd_dev_info(dd, "Parent PCIe bridge does not support Gen3\n");
321 		dd->link_gen3_capable = 0;
322 	}
323 
324 	/* obtain the link width and current speed */
325 	update_lbus_info(dd);
326 
327 	dd_dev_info(dd, "%s\n", dd->lbus_info);
328 
329 	return 0;
330 }
331 
332 /**
333  * Restore command and BARs after a reset has wiped them out
334  *
335  * Returns 0 on success, otherwise a negative error value
336  */
restore_pci_variables(struct hfi1_devdata * dd)337 int restore_pci_variables(struct hfi1_devdata *dd)
338 {
339 	int ret;
340 
341 	ret = pci_write_config_word(dd->pcidev, PCI_COMMAND, dd->pci_command);
342 	if (ret)
343 		goto error;
344 
345 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
346 				     dd->pcibar0);
347 	if (ret)
348 		goto error;
349 
350 	ret = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
351 				     dd->pcibar1);
352 	if (ret)
353 		goto error;
354 
355 	ret = pci_write_config_dword(dd->pcidev, PCI_ROM_ADDRESS, dd->pci_rom);
356 	if (ret)
357 		goto error;
358 
359 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL,
360 					 dd->pcie_devctl);
361 	if (ret)
362 		goto error;
363 
364 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL,
365 					 dd->pcie_lnkctl);
366 	if (ret)
367 		goto error;
368 
369 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL2,
370 					 dd->pcie_devctl2);
371 	if (ret)
372 		goto error;
373 
374 	ret = pci_write_config_dword(dd->pcidev, PCI_CFG_MSIX0, dd->pci_msix0);
375 	if (ret)
376 		goto error;
377 
378 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
379 		ret = pci_write_config_dword(dd->pcidev, PCIE_CFG_TPH2,
380 					     dd->pci_tph2);
381 		if (ret)
382 			goto error;
383 	}
384 	return 0;
385 
386 error:
387 	dd_dev_err(dd, "Unable to write to PCI config\n");
388 	return pcibios_err_to_errno(ret);
389 }
390 
391 /**
392  * Save BARs and command to rewrite after device reset
393  *
394  * Returns 0 on success, otherwise a negative error value
395  */
save_pci_variables(struct hfi1_devdata * dd)396 int save_pci_variables(struct hfi1_devdata *dd)
397 {
398 	int ret;
399 
400 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
401 				    &dd->pcibar0);
402 	if (ret)
403 		goto error;
404 
405 	ret = pci_read_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
406 				    &dd->pcibar1);
407 	if (ret)
408 		goto error;
409 
410 	ret = pci_read_config_dword(dd->pcidev, PCI_ROM_ADDRESS, &dd->pci_rom);
411 	if (ret)
412 		goto error;
413 
414 	ret = pci_read_config_word(dd->pcidev, PCI_COMMAND, &dd->pci_command);
415 	if (ret)
416 		goto error;
417 
418 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL,
419 					&dd->pcie_devctl);
420 	if (ret)
421 		goto error;
422 
423 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL,
424 					&dd->pcie_lnkctl);
425 	if (ret)
426 		goto error;
427 
428 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL2,
429 					&dd->pcie_devctl2);
430 	if (ret)
431 		goto error;
432 
433 	ret = pci_read_config_dword(dd->pcidev, PCI_CFG_MSIX0, &dd->pci_msix0);
434 	if (ret)
435 		goto error;
436 
437 	if (pci_find_ext_capability(dd->pcidev, PCI_EXT_CAP_ID_TPH)) {
438 		ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_TPH2,
439 					    &dd->pci_tph2);
440 		if (ret)
441 			goto error;
442 	}
443 	return 0;
444 
445 error:
446 	dd_dev_err(dd, "Unable to read from PCI config\n");
447 	return pcibios_err_to_errno(ret);
448 }
449 
450 /*
451  * BIOS may not set PCIe bus-utilization parameters for best performance.
452  * Check and optionally adjust them to maximize our throughput.
453  */
454 static int hfi1_pcie_caps;
455 module_param_named(pcie_caps, hfi1_pcie_caps, int, 0444);
456 MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
457 
458 /**
459  * tune_pcie_caps() - Code to adjust PCIe capabilities.
460  * @dd: Valid device data structure
461  *
462  */
tune_pcie_caps(struct hfi1_devdata * dd)463 void tune_pcie_caps(struct hfi1_devdata *dd)
464 {
465 	struct pci_dev *parent;
466 	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
467 	u16 rc_mrrs, ep_mrrs, max_mrrs, ectl;
468 	int ret;
469 
470 	/*
471 	 * Turn on extended tags in DevCtl in case the BIOS has turned it off
472 	 * to improve WFR SDMA bandwidth
473 	 */
474 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL, &ectl);
475 	if ((!ret) && !(ectl & PCI_EXP_DEVCTL_EXT_TAG)) {
476 		dd_dev_info(dd, "Enabling PCIe extended tags\n");
477 		ectl |= PCI_EXP_DEVCTL_EXT_TAG;
478 		ret = pcie_capability_write_word(dd->pcidev,
479 						 PCI_EXP_DEVCTL, ectl);
480 		if (ret)
481 			dd_dev_info(dd, "Unable to write to PCI config\n");
482 	}
483 	/* Find out supported and configured values for parent (root) */
484 	parent = dd->pcidev->bus->self;
485 	/*
486 	 * The driver cannot perform the tuning if it does not have
487 	 * access to the upstream component.
488 	 */
489 	if (!parent) {
490 		dd_dev_info(dd, "Parent not found\n");
491 		return;
492 	}
493 	if (!pci_is_root_bus(parent->bus)) {
494 		dd_dev_info(dd, "Parent not root\n");
495 		return;
496 	}
497 	if (!pci_is_pcie(parent)) {
498 		dd_dev_info(dd, "Parent is not PCI Express capable\n");
499 		return;
500 	}
501 	if (!pci_is_pcie(dd->pcidev)) {
502 		dd_dev_info(dd, "PCI device is not PCI Express capable\n");
503 		return;
504 	}
505 	rc_mpss = parent->pcie_mpss;
506 	rc_mps = ffs(pcie_get_mps(parent)) - 8;
507 	/* Find out supported and configured values for endpoint (us) */
508 	ep_mpss = dd->pcidev->pcie_mpss;
509 	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
510 
511 	/* Find max payload supported by root, endpoint */
512 	if (rc_mpss > ep_mpss)
513 		rc_mpss = ep_mpss;
514 
515 	/* If Supported greater than limit in module param, limit it */
516 	if (rc_mpss > (hfi1_pcie_caps & 7))
517 		rc_mpss = hfi1_pcie_caps & 7;
518 	/* If less than (allowed, supported), bump root payload */
519 	if (rc_mpss > rc_mps) {
520 		rc_mps = rc_mpss;
521 		pcie_set_mps(parent, 128 << rc_mps);
522 	}
523 	/* If less than (allowed, supported), bump endpoint payload */
524 	if (rc_mpss > ep_mps) {
525 		ep_mps = rc_mpss;
526 		pcie_set_mps(dd->pcidev, 128 << ep_mps);
527 	}
528 
529 	/*
530 	 * Now the Read Request size.
531 	 * No field for max supported, but PCIe spec limits it to 4096,
532 	 * which is code '5' (log2(4096) - 7)
533 	 */
534 	max_mrrs = 5;
535 	if (max_mrrs > ((hfi1_pcie_caps >> 4) & 7))
536 		max_mrrs = (hfi1_pcie_caps >> 4) & 7;
537 
538 	max_mrrs = 128 << max_mrrs;
539 	rc_mrrs = pcie_get_readrq(parent);
540 	ep_mrrs = pcie_get_readrq(dd->pcidev);
541 
542 	if (max_mrrs > rc_mrrs) {
543 		rc_mrrs = max_mrrs;
544 		pcie_set_readrq(parent, rc_mrrs);
545 	}
546 	if (max_mrrs > ep_mrrs) {
547 		ep_mrrs = max_mrrs;
548 		pcie_set_readrq(dd->pcidev, ep_mrrs);
549 	}
550 }
551 
552 /* End of PCIe capability tuning */
553 
554 /*
555  * From here through hfi1_pci_err_handler definition is invoked via
556  * PCI error infrastructure, registered via pci
557  */
558 static pci_ers_result_t
pci_error_detected(struct pci_dev * pdev,pci_channel_state_t state)559 pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
560 {
561 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
562 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
563 
564 	switch (state) {
565 	case pci_channel_io_normal:
566 		dd_dev_info(dd, "State Normal, ignoring\n");
567 		break;
568 
569 	case pci_channel_io_frozen:
570 		dd_dev_info(dd, "State Frozen, requesting reset\n");
571 		pci_disable_device(pdev);
572 		ret = PCI_ERS_RESULT_NEED_RESET;
573 		break;
574 
575 	case pci_channel_io_perm_failure:
576 		if (dd) {
577 			dd_dev_info(dd, "State Permanent Failure, disabling\n");
578 			/* no more register accesses! */
579 			dd->flags &= ~HFI1_PRESENT;
580 			hfi1_disable_after_error(dd);
581 		}
582 		 /* else early, or other problem */
583 		ret =  PCI_ERS_RESULT_DISCONNECT;
584 		break;
585 
586 	default: /* shouldn't happen */
587 		dd_dev_info(dd, "HFI1 PCI errors detected (state %d)\n",
588 			    state);
589 		break;
590 	}
591 	return ret;
592 }
593 
594 static pci_ers_result_t
pci_mmio_enabled(struct pci_dev * pdev)595 pci_mmio_enabled(struct pci_dev *pdev)
596 {
597 	u64 words = 0U;
598 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
599 	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
600 
601 	if (dd && dd->pport) {
602 		words = read_port_cntr(dd->pport, C_RX_WORDS, CNTR_INVALID_VL);
603 		if (words == ~0ULL)
604 			ret = PCI_ERS_RESULT_NEED_RESET;
605 		dd_dev_info(dd,
606 			    "HFI1 mmio_enabled function called, read wordscntr %llx, returning %d\n",
607 			    words, ret);
608 	}
609 	return  ret;
610 }
611 
612 static pci_ers_result_t
pci_slot_reset(struct pci_dev * pdev)613 pci_slot_reset(struct pci_dev *pdev)
614 {
615 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
616 
617 	dd_dev_info(dd, "HFI1 slot_reset function called, ignored\n");
618 	return PCI_ERS_RESULT_CAN_RECOVER;
619 }
620 
621 static void
pci_resume(struct pci_dev * pdev)622 pci_resume(struct pci_dev *pdev)
623 {
624 	struct hfi1_devdata *dd = pci_get_drvdata(pdev);
625 
626 	dd_dev_info(dd, "HFI1 resume function called\n");
627 	/*
628 	 * Running jobs will fail, since it's asynchronous
629 	 * unlike sysfs-requested reset.   Better than
630 	 * doing nothing.
631 	 */
632 	hfi1_init(dd, 1); /* same as re-init after reset */
633 }
634 
635 const struct pci_error_handlers hfi1_pci_err_handler = {
636 	.error_detected = pci_error_detected,
637 	.mmio_enabled = pci_mmio_enabled,
638 	.slot_reset = pci_slot_reset,
639 	.resume = pci_resume,
640 };
641 
642 /*============================================================================*/
643 /* PCIe Gen3 support */
644 
645 /*
646  * This code is separated out because it is expected to be removed in the
647  * final shipping product.  If not, then it will be revisited and items
648  * will be moved to more standard locations.
649  */
650 
651 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_STS field values */
652 #define DL_STATUS_HFI0 0x1	/* hfi0 firmware download complete */
653 #define DL_STATUS_HFI1 0x2	/* hfi1 firmware download complete */
654 #define DL_STATUS_BOTH 0x3	/* hfi0 and hfi1 firmware download complete */
655 
656 /* ASIC_PCI_SD_HOST_STATUS.FW_DNLD_ERR field values */
657 #define DL_ERR_NONE		0x0	/* no error */
658 #define DL_ERR_SWAP_PARITY	0x1	/* parity error in SerDes interrupt */
659 					/*   or response data */
660 #define DL_ERR_DISABLED	0x2	/* hfi disabled */
661 #define DL_ERR_SECURITY	0x3	/* security check failed */
662 #define DL_ERR_SBUS		0x4	/* SBus status error */
663 #define DL_ERR_XFR_PARITY	0x5	/* parity error during ROM transfer*/
664 
665 /* gasket block secondary bus reset delay */
666 #define SBR_DELAY_US 200000	/* 200ms */
667 
668 static uint pcie_target = 3;
669 module_param(pcie_target, uint, S_IRUGO);
670 MODULE_PARM_DESC(pcie_target, "PCIe target speed (0 skip, 1-3 Gen1-3)");
671 
672 static uint pcie_force;
673 module_param(pcie_force, uint, S_IRUGO);
674 MODULE_PARM_DESC(pcie_force, "Force driver to do a PCIe firmware download even if already at target speed");
675 
676 static uint pcie_retry = 5;
677 module_param(pcie_retry, uint, S_IRUGO);
678 MODULE_PARM_DESC(pcie_retry, "Driver will try this many times to reach requested speed");
679 
680 #define UNSET_PSET 255
681 #define DEFAULT_DISCRETE_PSET 2	/* discrete HFI */
682 #define DEFAULT_MCP_PSET 6	/* MCP HFI */
683 static uint pcie_pset = UNSET_PSET;
684 module_param(pcie_pset, uint, S_IRUGO);
685 MODULE_PARM_DESC(pcie_pset, "PCIe Eq Pset value to use, range is 0-10");
686 
687 static uint pcie_ctle = 3; /* discrete on, integrated on */
688 module_param(pcie_ctle, uint, S_IRUGO);
689 MODULE_PARM_DESC(pcie_ctle, "PCIe static CTLE mode, bit 0 - discrete on/off, bit 1 - integrated on/off");
690 
691 /* equalization columns */
692 #define PREC 0
693 #define ATTN 1
694 #define POST 2
695 
696 /* discrete silicon preliminary equalization values */
697 static const u8 discrete_preliminary_eq[11][3] = {
698 	/* prec   attn   post */
699 	{  0x00,  0x00,  0x12 },	/* p0 */
700 	{  0x00,  0x00,  0x0c },	/* p1 */
701 	{  0x00,  0x00,  0x0f },	/* p2 */
702 	{  0x00,  0x00,  0x09 },	/* p3 */
703 	{  0x00,  0x00,  0x00 },	/* p4 */
704 	{  0x06,  0x00,  0x00 },	/* p5 */
705 	{  0x09,  0x00,  0x00 },	/* p6 */
706 	{  0x06,  0x00,  0x0f },	/* p7 */
707 	{  0x09,  0x00,  0x09 },	/* p8 */
708 	{  0x0c,  0x00,  0x00 },	/* p9 */
709 	{  0x00,  0x00,  0x18 },	/* p10 */
710 };
711 
712 /* integrated silicon preliminary equalization values */
713 static const u8 integrated_preliminary_eq[11][3] = {
714 	/* prec   attn   post */
715 	{  0x00,  0x1e,  0x07 },	/* p0 */
716 	{  0x00,  0x1e,  0x05 },	/* p1 */
717 	{  0x00,  0x1e,  0x06 },	/* p2 */
718 	{  0x00,  0x1e,  0x04 },	/* p3 */
719 	{  0x00,  0x1e,  0x00 },	/* p4 */
720 	{  0x03,  0x1e,  0x00 },	/* p5 */
721 	{  0x04,  0x1e,  0x00 },	/* p6 */
722 	{  0x03,  0x1e,  0x06 },	/* p7 */
723 	{  0x03,  0x1e,  0x04 },	/* p8 */
724 	{  0x05,  0x1e,  0x00 },	/* p9 */
725 	{  0x00,  0x1e,  0x0a },	/* p10 */
726 };
727 
728 static const u8 discrete_ctle_tunings[11][4] = {
729 	/* DC     LF     HF     BW */
730 	{  0x48,  0x0b,  0x04,  0x04 },	/* p0 */
731 	{  0x60,  0x05,  0x0f,  0x0a },	/* p1 */
732 	{  0x50,  0x09,  0x06,  0x06 },	/* p2 */
733 	{  0x68,  0x05,  0x0f,  0x0a },	/* p3 */
734 	{  0x80,  0x05,  0x0f,  0x0a },	/* p4 */
735 	{  0x70,  0x05,  0x0f,  0x0a },	/* p5 */
736 	{  0x68,  0x05,  0x0f,  0x0a },	/* p6 */
737 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
738 	{  0x48,  0x09,  0x06,  0x06 },	/* p8 */
739 	{  0x60,  0x05,  0x0f,  0x0a },	/* p9 */
740 	{  0x38,  0x0f,  0x00,  0x00 },	/* p10 */
741 };
742 
743 static const u8 integrated_ctle_tunings[11][4] = {
744 	/* DC     LF     HF     BW */
745 	{  0x38,  0x0f,  0x00,  0x00 },	/* p0 */
746 	{  0x38,  0x0f,  0x00,  0x00 },	/* p1 */
747 	{  0x38,  0x0f,  0x00,  0x00 },	/* p2 */
748 	{  0x38,  0x0f,  0x00,  0x00 },	/* p3 */
749 	{  0x58,  0x0a,  0x05,  0x05 },	/* p4 */
750 	{  0x48,  0x0a,  0x05,  0x05 },	/* p5 */
751 	{  0x40,  0x0a,  0x05,  0x05 },	/* p6 */
752 	{  0x38,  0x0f,  0x00,  0x00 },	/* p7 */
753 	{  0x38,  0x0f,  0x00,  0x00 },	/* p8 */
754 	{  0x38,  0x09,  0x06,  0x06 },	/* p9 */
755 	{  0x38,  0x0e,  0x01,  0x01 },	/* p10 */
756 };
757 
758 /* helper to format the value to write to hardware */
759 #define eq_value(pre, curr, post) \
760 	((((u32)(pre)) << \
761 			PCIE_CFG_REG_PL102_GEN3_EQ_PRE_CURSOR_PSET_SHIFT) \
762 	| (((u32)(curr)) << PCIE_CFG_REG_PL102_GEN3_EQ_CURSOR_PSET_SHIFT) \
763 	| (((u32)(post)) << \
764 		PCIE_CFG_REG_PL102_GEN3_EQ_POST_CURSOR_PSET_SHIFT))
765 
766 /*
767  * Load the given EQ preset table into the PCIe hardware.
768  */
load_eq_table(struct hfi1_devdata * dd,const u8 eq[11][3],u8 fs,u8 div)769 static int load_eq_table(struct hfi1_devdata *dd, const u8 eq[11][3], u8 fs,
770 			 u8 div)
771 {
772 	struct pci_dev *pdev = dd->pcidev;
773 	u32 hit_error = 0;
774 	u32 violation;
775 	u32 i;
776 	u8 c_minus1, c0, c_plus1;
777 	int ret;
778 
779 	for (i = 0; i < 11; i++) {
780 		/* set index */
781 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL103, i);
782 		/* write the value */
783 		c_minus1 = eq[i][PREC] / div;
784 		c0 = fs - (eq[i][PREC] / div) - (eq[i][POST] / div);
785 		c_plus1 = eq[i][POST] / div;
786 		pci_write_config_dword(pdev, PCIE_CFG_REG_PL102,
787 				       eq_value(c_minus1, c0, c_plus1));
788 		/* check if these coefficients violate EQ rules */
789 		ret = pci_read_config_dword(dd->pcidev,
790 					    PCIE_CFG_REG_PL105, &violation);
791 		if (ret) {
792 			dd_dev_err(dd, "Unable to read from PCI config\n");
793 			hit_error = 1;
794 			break;
795 		}
796 
797 		if (violation
798 		    & PCIE_CFG_REG_PL105_GEN3_EQ_VIOLATE_COEF_RULES_SMASK){
799 			if (hit_error == 0) {
800 				dd_dev_err(dd,
801 					   "Gen3 EQ Table Coefficient rule violations\n");
802 				dd_dev_err(dd, "         prec   attn   post\n");
803 			}
804 			dd_dev_err(dd, "   p%02d:   %02x     %02x     %02x\n",
805 				   i, (u32)eq[i][0], (u32)eq[i][1],
806 				   (u32)eq[i][2]);
807 			dd_dev_err(dd, "            %02x     %02x     %02x\n",
808 				   (u32)c_minus1, (u32)c0, (u32)c_plus1);
809 			hit_error = 1;
810 		}
811 	}
812 	if (hit_error)
813 		return -EINVAL;
814 	return 0;
815 }
816 
817 /*
818  * Steps to be done after the PCIe firmware is downloaded and
819  * before the SBR for the Pcie Gen3.
820  * The SBus resource is already being held.
821  */
pcie_post_steps(struct hfi1_devdata * dd)822 static void pcie_post_steps(struct hfi1_devdata *dd)
823 {
824 	int i;
825 
826 	set_sbus_fast_mode(dd);
827 	/*
828 	 * Write to the PCIe PCSes to set the G3_LOCKED_NEXT bits to 1.
829 	 * This avoids a spurious framing error that can otherwise be
830 	 * generated by the MAC layer.
831 	 *
832 	 * Use individual addresses since no broadcast is set up.
833 	 */
834 	for (i = 0; i < NUM_PCIE_SERDES; i++) {
835 		sbus_request(dd, pcie_pcs_addrs[dd->hfi1_id][i],
836 			     0x03, WRITE_SBUS_RECEIVER, 0x00022132);
837 	}
838 
839 	clear_sbus_fast_mode(dd);
840 }
841 
842 /*
843  * Trigger a secondary bus reset (SBR) on ourselves using our parent.
844  *
845  * Based on pci_parent_bus_reset() which is not exported by the
846  * kernel core.
847  */
trigger_sbr(struct hfi1_devdata * dd)848 static int trigger_sbr(struct hfi1_devdata *dd)
849 {
850 	struct pci_dev *dev = dd->pcidev;
851 	struct pci_dev *pdev;
852 
853 	/* need a parent */
854 	if (!dev->bus->self) {
855 		dd_dev_err(dd, "%s: no parent device\n", __func__);
856 		return -ENOTTY;
857 	}
858 
859 	/* should not be anyone else on the bus */
860 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
861 		if (pdev != dev) {
862 			dd_dev_err(dd,
863 				   "%s: another device is on the same bus\n",
864 				   __func__);
865 			return -ENOTTY;
866 		}
867 
868 	/*
869 	 * This is an end around to do an SBR during probe time. A new API needs
870 	 * to be implemented to have cleaner interface but this fixes the
871 	 * current brokenness
872 	 */
873 	return pci_bridge_secondary_bus_reset(dev->bus->self);
874 }
875 
876 /*
877  * Write the given gasket interrupt register.
878  */
write_gasket_interrupt(struct hfi1_devdata * dd,int index,u16 code,u16 data)879 static void write_gasket_interrupt(struct hfi1_devdata *dd, int index,
880 				   u16 code, u16 data)
881 {
882 	write_csr(dd, ASIC_PCIE_SD_INTRPT_LIST + (index * 8),
883 		  (((u64)code << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_CODE_SHIFT) |
884 		   ((u64)data << ASIC_PCIE_SD_INTRPT_LIST_INTRPT_DATA_SHIFT)));
885 }
886 
887 /*
888  * Tell the gasket logic how to react to the reset.
889  */
arm_gasket_logic(struct hfi1_devdata * dd)890 static void arm_gasket_logic(struct hfi1_devdata *dd)
891 {
892 	u64 reg;
893 
894 	reg = (((u64)1 << dd->hfi1_id) <<
895 	       ASIC_PCIE_SD_HOST_CMD_INTRPT_CMD_SHIFT) |
896 	      ((u64)pcie_serdes_broadcast[dd->hfi1_id] <<
897 	       ASIC_PCIE_SD_HOST_CMD_SBUS_RCVR_ADDR_SHIFT |
898 	       ASIC_PCIE_SD_HOST_CMD_SBR_MODE_SMASK |
899 	       ((u64)SBR_DELAY_US & ASIC_PCIE_SD_HOST_CMD_TIMER_MASK) <<
900 	       ASIC_PCIE_SD_HOST_CMD_TIMER_SHIFT);
901 	write_csr(dd, ASIC_PCIE_SD_HOST_CMD, reg);
902 	/* read back to push the write */
903 	read_csr(dd, ASIC_PCIE_SD_HOST_CMD);
904 }
905 
906 /*
907  * CCE_PCIE_CTRL long name helpers
908  * We redefine these shorter macros to use in the code while leaving
909  * chip_registers.h to be autogenerated from the hardware spec.
910  */
911 #define LANE_BUNDLE_MASK              CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_MASK
912 #define LANE_BUNDLE_SHIFT             CCE_PCIE_CTRL_PCIE_LANE_BUNDLE_SHIFT
913 #define LANE_DELAY_MASK               CCE_PCIE_CTRL_PCIE_LANE_DELAY_MASK
914 #define LANE_DELAY_SHIFT              CCE_PCIE_CTRL_PCIE_LANE_DELAY_SHIFT
915 #define MARGIN_OVERWRITE_ENABLE_SHIFT CCE_PCIE_CTRL_XMT_MARGIN_OVERWRITE_ENABLE_SHIFT
916 #define MARGIN_SHIFT                  CCE_PCIE_CTRL_XMT_MARGIN_SHIFT
917 #define MARGIN_G1_G2_OVERWRITE_MASK   CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_MASK
918 #define MARGIN_G1_G2_OVERWRITE_SHIFT  CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_OVERWRITE_ENABLE_SHIFT
919 #define MARGIN_GEN1_GEN2_MASK         CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_MASK
920 #define MARGIN_GEN1_GEN2_SHIFT        CCE_PCIE_CTRL_XMT_MARGIN_GEN1_GEN2_SHIFT
921 
922  /*
923   * Write xmt_margin for full-swing (WFR-B) or half-swing (WFR-C).
924   */
write_xmt_margin(struct hfi1_devdata * dd,const char * fname)925 static void write_xmt_margin(struct hfi1_devdata *dd, const char *fname)
926 {
927 	u64 pcie_ctrl;
928 	u64 xmt_margin;
929 	u64 xmt_margin_oe;
930 	u64 lane_delay;
931 	u64 lane_bundle;
932 
933 	pcie_ctrl = read_csr(dd, CCE_PCIE_CTRL);
934 
935 	/*
936 	 * For Discrete, use full-swing.
937 	 *  - PCIe TX defaults to full-swing.
938 	 *    Leave this register as default.
939 	 * For Integrated, use half-swing
940 	 *  - Copy xmt_margin and xmt_margin_oe
941 	 *    from Gen1/Gen2 to Gen3.
942 	 */
943 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL1) { /* integrated */
944 		/* extract initial fields */
945 		xmt_margin = (pcie_ctrl >> MARGIN_GEN1_GEN2_SHIFT)
946 			      & MARGIN_GEN1_GEN2_MASK;
947 		xmt_margin_oe = (pcie_ctrl >> MARGIN_G1_G2_OVERWRITE_SHIFT)
948 				 & MARGIN_G1_G2_OVERWRITE_MASK;
949 		lane_delay = (pcie_ctrl >> LANE_DELAY_SHIFT) & LANE_DELAY_MASK;
950 		lane_bundle = (pcie_ctrl >> LANE_BUNDLE_SHIFT)
951 			       & LANE_BUNDLE_MASK;
952 
953 		/*
954 		 * For A0, EFUSE values are not set.  Override with the
955 		 * correct values.
956 		 */
957 		if (is_ax(dd)) {
958 			/*
959 			 * xmt_margin and OverwiteEnabel should be the
960 			 * same for Gen1/Gen2 and Gen3
961 			 */
962 			xmt_margin = 0x5;
963 			xmt_margin_oe = 0x1;
964 			lane_delay = 0xF; /* Delay 240ns. */
965 			lane_bundle = 0x0; /* Set to 1 lane. */
966 		}
967 
968 		/* overwrite existing values */
969 		pcie_ctrl = (xmt_margin << MARGIN_GEN1_GEN2_SHIFT)
970 			| (xmt_margin_oe << MARGIN_G1_G2_OVERWRITE_SHIFT)
971 			| (xmt_margin << MARGIN_SHIFT)
972 			| (xmt_margin_oe << MARGIN_OVERWRITE_ENABLE_SHIFT)
973 			| (lane_delay << LANE_DELAY_SHIFT)
974 			| (lane_bundle << LANE_BUNDLE_SHIFT);
975 
976 		write_csr(dd, CCE_PCIE_CTRL, pcie_ctrl);
977 	}
978 
979 	dd_dev_dbg(dd, "%s: program XMT margin, CcePcieCtrl 0x%llx\n",
980 		   fname, pcie_ctrl);
981 }
982 
983 /*
984  * Do all the steps needed to transition the PCIe link to Gen3 speed.
985  */
do_pcie_gen3_transition(struct hfi1_devdata * dd)986 int do_pcie_gen3_transition(struct hfi1_devdata *dd)
987 {
988 	struct pci_dev *parent = dd->pcidev->bus->self;
989 	u64 fw_ctrl;
990 	u64 reg, therm;
991 	u32 reg32, fs, lf;
992 	u32 status, err;
993 	int ret;
994 	int do_retry, retry_count = 0;
995 	int intnum = 0;
996 	uint default_pset;
997 	uint pset = pcie_pset;
998 	u16 target_vector, target_speed;
999 	u16 lnkctl2, vendor;
1000 	u8 div;
1001 	const u8 (*eq)[3];
1002 	const u8 (*ctle_tunings)[4];
1003 	uint static_ctle_mode;
1004 	int return_error = 0;
1005 	u32 target_width;
1006 
1007 	/* PCIe Gen3 is for the ASIC only */
1008 	if (dd->icode != ICODE_RTL_SILICON)
1009 		return 0;
1010 
1011 	if (pcie_target == 1) {			/* target Gen1 */
1012 		target_vector = PCI_EXP_LNKCTL2_TLS_2_5GT;
1013 		target_speed = 2500;
1014 	} else if (pcie_target == 2) {		/* target Gen2 */
1015 		target_vector = PCI_EXP_LNKCTL2_TLS_5_0GT;
1016 		target_speed = 5000;
1017 	} else if (pcie_target == 3) {		/* target Gen3 */
1018 		target_vector = PCI_EXP_LNKCTL2_TLS_8_0GT;
1019 		target_speed = 8000;
1020 	} else {
1021 		/* off or invalid target - skip */
1022 		dd_dev_info(dd, "%s: Skipping PCIe transition\n", __func__);
1023 		return 0;
1024 	}
1025 
1026 	/* if already at target speed, done (unless forced) */
1027 	if (dd->lbus_speed == target_speed) {
1028 		dd_dev_info(dd, "%s: PCIe already at gen%d, %s\n", __func__,
1029 			    pcie_target,
1030 			    pcie_force ? "re-doing anyway" : "skipping");
1031 		if (!pcie_force)
1032 			return 0;
1033 	}
1034 
1035 	/*
1036 	 * The driver cannot do the transition if it has no access to the
1037 	 * upstream component
1038 	 */
1039 	if (!parent) {
1040 		dd_dev_info(dd, "%s: No upstream, Can't do gen3 transition\n",
1041 			    __func__);
1042 		return 0;
1043 	}
1044 
1045 	/* Previous Gen1/Gen2 bus width */
1046 	target_width = dd->lbus_width;
1047 
1048 	/*
1049 	 * Do the Gen3 transition.  Steps are those of the PCIe Gen3
1050 	 * recipe.
1051 	 */
1052 
1053 	/* step 1: pcie link working in gen1/gen2 */
1054 
1055 	/* step 2: if either side is not capable of Gen3, done */
1056 	if (pcie_target == 3 && !dd->link_gen3_capable) {
1057 		dd_dev_err(dd, "The PCIe link is not Gen3 capable\n");
1058 		ret = -ENOSYS;
1059 		goto done_no_mutex;
1060 	}
1061 
1062 	/* hold the SBus resource across the firmware download and SBR */
1063 	ret = acquire_chip_resource(dd, CR_SBUS, SBUS_TIMEOUT);
1064 	if (ret) {
1065 		dd_dev_err(dd, "%s: unable to acquire SBus resource\n",
1066 			   __func__);
1067 		return ret;
1068 	}
1069 
1070 	/* make sure thermal polling is not causing interrupts */
1071 	therm = read_csr(dd, ASIC_CFG_THERM_POLL_EN);
1072 	if (therm) {
1073 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x0);
1074 		msleep(100);
1075 		dd_dev_info(dd, "%s: Disabled therm polling\n",
1076 			    __func__);
1077 	}
1078 
1079 retry:
1080 	/* the SBus download will reset the spico for thermal */
1081 
1082 	/* step 3: download SBus Master firmware */
1083 	/* step 4: download PCIe Gen3 SerDes firmware */
1084 	dd_dev_info(dd, "%s: downloading firmware\n", __func__);
1085 	ret = load_pcie_firmware(dd);
1086 	if (ret) {
1087 		/* do not proceed if the firmware cannot be downloaded */
1088 		return_error = 1;
1089 		goto done;
1090 	}
1091 
1092 	/* step 5: set up device parameter settings */
1093 	dd_dev_info(dd, "%s: setting PCIe registers\n", __func__);
1094 
1095 	/*
1096 	 * PcieCfgSpcie1 - Link Control 3
1097 	 * Leave at reset value.  No need to set PerfEq - link equalization
1098 	 * will be performed automatically after the SBR when the target
1099 	 * speed is 8GT/s.
1100 	 */
1101 
1102 	/* clear all 16 per-lane error bits (PCIe: Lane Error Status) */
1103 	pci_write_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, 0xffff);
1104 
1105 	/* step 5a: Set Synopsys Port Logic registers */
1106 
1107 	/*
1108 	 * PcieCfgRegPl2 - Port Force Link
1109 	 *
1110 	 * Set the low power field to 0x10 to avoid unnecessary power
1111 	 * management messages.  All other fields are zero.
1112 	 */
1113 	reg32 = 0x10ul << PCIE_CFG_REG_PL2_LOW_PWR_ENT_CNT_SHIFT;
1114 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL2, reg32);
1115 
1116 	/*
1117 	 * PcieCfgRegPl100 - Gen3 Control
1118 	 *
1119 	 * turn off PcieCfgRegPl100.Gen3ZRxDcNonCompl
1120 	 * turn on PcieCfgRegPl100.EqEieosCnt
1121 	 * Everything else zero.
1122 	 */
1123 	reg32 = PCIE_CFG_REG_PL100_EQ_EIEOS_CNT_SMASK;
1124 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL100, reg32);
1125 
1126 	/*
1127 	 * PcieCfgRegPl101 - Gen3 EQ FS and LF
1128 	 * PcieCfgRegPl102 - Gen3 EQ Presets to Coefficients Mapping
1129 	 * PcieCfgRegPl103 - Gen3 EQ Preset Index
1130 	 * PcieCfgRegPl105 - Gen3 EQ Status
1131 	 *
1132 	 * Give initial EQ settings.
1133 	 */
1134 	if (dd->pcidev->device == PCI_DEVICE_ID_INTEL0) { /* discrete */
1135 		/* 1000mV, FS=24, LF = 8 */
1136 		fs = 24;
1137 		lf = 8;
1138 		div = 3;
1139 		eq = discrete_preliminary_eq;
1140 		default_pset = DEFAULT_DISCRETE_PSET;
1141 		ctle_tunings = discrete_ctle_tunings;
1142 		/* bit 0 - discrete on/off */
1143 		static_ctle_mode = pcie_ctle & 0x1;
1144 	} else {
1145 		/* 400mV, FS=29, LF = 9 */
1146 		fs = 29;
1147 		lf = 9;
1148 		div = 1;
1149 		eq = integrated_preliminary_eq;
1150 		default_pset = DEFAULT_MCP_PSET;
1151 		ctle_tunings = integrated_ctle_tunings;
1152 		/* bit 1 - integrated on/off */
1153 		static_ctle_mode = (pcie_ctle >> 1) & 0x1;
1154 	}
1155 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL101,
1156 			       (fs <<
1157 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_FS_SHIFT) |
1158 			       (lf <<
1159 				PCIE_CFG_REG_PL101_GEN3_EQ_LOCAL_LF_SHIFT));
1160 	ret = load_eq_table(dd, eq, fs, div);
1161 	if (ret)
1162 		goto done;
1163 
1164 	/*
1165 	 * PcieCfgRegPl106 - Gen3 EQ Control
1166 	 *
1167 	 * Set Gen3EqPsetReqVec, leave other fields 0.
1168 	 */
1169 	if (pset == UNSET_PSET)
1170 		pset = default_pset;
1171 	if (pset > 10) {	/* valid range is 0-10, inclusive */
1172 		dd_dev_err(dd, "%s: Invalid Eq Pset %u, setting to %d\n",
1173 			   __func__, pset, default_pset);
1174 		pset = default_pset;
1175 	}
1176 	dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pset);
1177 	pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL106,
1178 			       ((1 << pset) <<
1179 			PCIE_CFG_REG_PL106_GEN3_EQ_PSET_REQ_VEC_SHIFT) |
1180 			PCIE_CFG_REG_PL106_GEN3_EQ_EVAL2MS_DISABLE_SMASK |
1181 			PCIE_CFG_REG_PL106_GEN3_EQ_PHASE23_EXIT_MODE_SMASK);
1182 
1183 	/*
1184 	 * step 5b: Do post firmware download steps via SBus
1185 	 */
1186 	dd_dev_info(dd, "%s: doing pcie post steps\n", __func__);
1187 	pcie_post_steps(dd);
1188 
1189 	/*
1190 	 * step 5c: Program gasket interrupts
1191 	 */
1192 	/* set the Rx Bit Rate to REFCLK ratio */
1193 	write_gasket_interrupt(dd, intnum++, 0x0006, 0x0050);
1194 	/* disable pCal for PCIe Gen3 RX equalization */
1195 	/* select adaptive or static CTLE */
1196 	write_gasket_interrupt(dd, intnum++, 0x0026,
1197 			       0x5b01 | (static_ctle_mode << 3));
1198 	/*
1199 	 * Enable iCal for PCIe Gen3 RX equalization, and set which
1200 	 * evaluation of RX_EQ_EVAL will launch the iCal procedure.
1201 	 */
1202 	write_gasket_interrupt(dd, intnum++, 0x0026, 0x5202);
1203 
1204 	if (static_ctle_mode) {
1205 		/* apply static CTLE tunings */
1206 		u8 pcie_dc, pcie_lf, pcie_hf, pcie_bw;
1207 
1208 		pcie_dc = ctle_tunings[pset][0];
1209 		pcie_lf = ctle_tunings[pset][1];
1210 		pcie_hf = ctle_tunings[pset][2];
1211 		pcie_bw = ctle_tunings[pset][3];
1212 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0200 | pcie_dc);
1213 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0100 | pcie_lf);
1214 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x0000 | pcie_hf);
1215 		write_gasket_interrupt(dd, intnum++, 0x0026, 0x5500 | pcie_bw);
1216 	}
1217 
1218 	/* terminate list */
1219 	write_gasket_interrupt(dd, intnum++, 0x0000, 0x0000);
1220 
1221 	/*
1222 	 * step 5d: program XMT margin
1223 	 */
1224 	write_xmt_margin(dd, __func__);
1225 
1226 	/*
1227 	 * step 5e: disable active state power management (ASPM). It
1228 	 * will be enabled if required later
1229 	 */
1230 	dd_dev_info(dd, "%s: clearing ASPM\n", __func__);
1231 	aspm_hw_disable_l1(dd);
1232 
1233 	/*
1234 	 * step 5f: clear DirectSpeedChange
1235 	 * PcieCfgRegPl67.DirectSpeedChange must be zero to prevent the
1236 	 * change in the speed target from starting before we are ready.
1237 	 * This field defaults to 0 and we are not changing it, so nothing
1238 	 * needs to be done.
1239 	 */
1240 
1241 	/* step 5g: Set target link speed */
1242 	/*
1243 	 * Set target link speed to be target on both device and parent.
1244 	 * On setting the parent: Some system BIOSs "helpfully" set the
1245 	 * parent target speed to Gen2 to match the ASIC's initial speed.
1246 	 * We can set the target Gen3 because we have already checked
1247 	 * that it is Gen3 capable earlier.
1248 	 */
1249 	dd_dev_info(dd, "%s: setting parent target link speed\n", __func__);
1250 	ret = pcie_capability_read_word(parent, PCI_EXP_LNKCTL2, &lnkctl2);
1251 	if (ret) {
1252 		dd_dev_err(dd, "Unable to read from PCI config\n");
1253 		return_error = 1;
1254 		goto done;
1255 	}
1256 
1257 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1258 		    (u32)lnkctl2);
1259 	/* only write to parent if target is not as high as ours */
1260 	if ((lnkctl2 & PCI_EXP_LNKCTL2_TLS) < target_vector) {
1261 		lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1262 		lnkctl2 |= target_vector;
1263 		dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1264 			    (u32)lnkctl2);
1265 		ret = pcie_capability_write_word(parent,
1266 						 PCI_EXP_LNKCTL2, lnkctl2);
1267 		if (ret) {
1268 			dd_dev_err(dd, "Unable to write to PCI config\n");
1269 			return_error = 1;
1270 			goto done;
1271 		}
1272 	} else {
1273 		dd_dev_info(dd, "%s: ..target speed is OK\n", __func__);
1274 	}
1275 
1276 	dd_dev_info(dd, "%s: setting target link speed\n", __func__);
1277 	ret = pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKCTL2, &lnkctl2);
1278 	if (ret) {
1279 		dd_dev_err(dd, "Unable to read from PCI config\n");
1280 		return_error = 1;
1281 		goto done;
1282 	}
1283 
1284 	dd_dev_info(dd, "%s: ..old link control2: 0x%x\n", __func__,
1285 		    (u32)lnkctl2);
1286 	lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
1287 	lnkctl2 |= target_vector;
1288 	dd_dev_info(dd, "%s: ..new link control2: 0x%x\n", __func__,
1289 		    (u32)lnkctl2);
1290 	ret = pcie_capability_write_word(dd->pcidev, PCI_EXP_LNKCTL2, lnkctl2);
1291 	if (ret) {
1292 		dd_dev_err(dd, "Unable to write to PCI config\n");
1293 		return_error = 1;
1294 		goto done;
1295 	}
1296 
1297 	/* step 5h: arm gasket logic */
1298 	/* hold DC in reset across the SBR */
1299 	write_csr(dd, CCE_DC_CTRL, CCE_DC_CTRL_DC_RESET_SMASK);
1300 	(void)read_csr(dd, CCE_DC_CTRL); /* DC reset hold */
1301 	/* save firmware control across the SBR */
1302 	fw_ctrl = read_csr(dd, MISC_CFG_FW_CTRL);
1303 
1304 	dd_dev_info(dd, "%s: arming gasket logic\n", __func__);
1305 	arm_gasket_logic(dd);
1306 
1307 	/*
1308 	 * step 6: quiesce PCIe link
1309 	 * The chip has already been reset, so there will be no traffic
1310 	 * from the chip.  Linux has no easy way to enforce that it will
1311 	 * not try to access the device, so we just need to hope it doesn't
1312 	 * do it while we are doing the reset.
1313 	 */
1314 
1315 	/*
1316 	 * step 7: initiate the secondary bus reset (SBR)
1317 	 * step 8: hardware brings the links back up
1318 	 * step 9: wait for link speed transition to be complete
1319 	 */
1320 	dd_dev_info(dd, "%s: calling trigger_sbr\n", __func__);
1321 	ret = trigger_sbr(dd);
1322 	if (ret)
1323 		goto done;
1324 
1325 	/* step 10: decide what to do next */
1326 
1327 	/* check if we can read PCI space */
1328 	ret = pci_read_config_word(dd->pcidev, PCI_VENDOR_ID, &vendor);
1329 	if (ret) {
1330 		dd_dev_info(dd,
1331 			    "%s: read of VendorID failed after SBR, err %d\n",
1332 			    __func__, ret);
1333 		return_error = 1;
1334 		goto done;
1335 	}
1336 	if (vendor == 0xffff) {
1337 		dd_dev_info(dd, "%s: VendorID is all 1s after SBR\n", __func__);
1338 		return_error = 1;
1339 		ret = -EIO;
1340 		goto done;
1341 	}
1342 
1343 	/* restore PCI space registers we know were reset */
1344 	dd_dev_info(dd, "%s: calling restore_pci_variables\n", __func__);
1345 	ret = restore_pci_variables(dd);
1346 	if (ret) {
1347 		dd_dev_err(dd, "%s: Could not restore PCI variables\n",
1348 			   __func__);
1349 		return_error = 1;
1350 		goto done;
1351 	}
1352 
1353 	/* restore firmware control */
1354 	write_csr(dd, MISC_CFG_FW_CTRL, fw_ctrl);
1355 
1356 	/*
1357 	 * Check the gasket block status.
1358 	 *
1359 	 * This is the first CSR read after the SBR.  If the read returns
1360 	 * all 1s (fails), the link did not make it back.
1361 	 *
1362 	 * Once we're sure we can read and write, clear the DC reset after
1363 	 * the SBR.  Then check for any per-lane errors. Then look over
1364 	 * the status.
1365 	 */
1366 	reg = read_csr(dd, ASIC_PCIE_SD_HOST_STATUS);
1367 	dd_dev_info(dd, "%s: gasket block status: 0x%llx\n", __func__, reg);
1368 	if (reg == ~0ull) {	/* PCIe read failed/timeout */
1369 		dd_dev_err(dd, "SBR failed - unable to read from device\n");
1370 		return_error = 1;
1371 		ret = -ENOSYS;
1372 		goto done;
1373 	}
1374 
1375 	/* clear the DC reset */
1376 	write_csr(dd, CCE_DC_CTRL, 0);
1377 
1378 	/* Set the LED off */
1379 	setextled(dd, 0);
1380 
1381 	/* check for any per-lane errors */
1382 	ret = pci_read_config_dword(dd->pcidev, PCIE_CFG_SPCIE2, &reg32);
1383 	if (ret) {
1384 		dd_dev_err(dd, "Unable to read from PCI config\n");
1385 		return_error = 1;
1386 		goto done;
1387 	}
1388 
1389 	dd_dev_info(dd, "%s: per-lane errors: 0x%x\n", __func__, reg32);
1390 
1391 	/* extract status, look for our HFI */
1392 	status = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_SHIFT)
1393 			& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_STS_MASK;
1394 	if ((status & (1 << dd->hfi1_id)) == 0) {
1395 		dd_dev_err(dd,
1396 			   "%s: gasket status 0x%x, expecting 0x%x\n",
1397 			   __func__, status, 1 << dd->hfi1_id);
1398 		ret = -EIO;
1399 		goto done;
1400 	}
1401 
1402 	/* extract error */
1403 	err = (reg >> ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_SHIFT)
1404 		& ASIC_PCIE_SD_HOST_STATUS_FW_DNLD_ERR_MASK;
1405 	if (err) {
1406 		dd_dev_err(dd, "%s: gasket error %d\n", __func__, err);
1407 		ret = -EIO;
1408 		goto done;
1409 	}
1410 
1411 	/* update our link information cache */
1412 	update_lbus_info(dd);
1413 	dd_dev_info(dd, "%s: new speed and width: %s\n", __func__,
1414 		    dd->lbus_info);
1415 
1416 	if (dd->lbus_speed != target_speed ||
1417 	    dd->lbus_width < target_width) { /* not target */
1418 		/* maybe retry */
1419 		do_retry = retry_count < pcie_retry;
1420 		dd_dev_err(dd, "PCIe link speed or width did not match target%s\n",
1421 			   do_retry ? ", retrying" : "");
1422 		retry_count++;
1423 		if (do_retry) {
1424 			msleep(100); /* allow time to settle */
1425 			goto retry;
1426 		}
1427 		ret = -EIO;
1428 	}
1429 
1430 done:
1431 	if (therm) {
1432 		write_csr(dd, ASIC_CFG_THERM_POLL_EN, 0x1);
1433 		msleep(100);
1434 		dd_dev_info(dd, "%s: Re-enable therm polling\n",
1435 			    __func__);
1436 	}
1437 	release_chip_resource(dd, CR_SBUS);
1438 done_no_mutex:
1439 	/* return no error if it is OK to be at current speed */
1440 	if (ret && !return_error) {
1441 		dd_dev_err(dd, "Proceeding at current speed PCIe speed\n");
1442 		ret = 0;
1443 	}
1444 
1445 	dd_dev_info(dd, "%s: done\n", __func__);
1446 	return ret;
1447 }
1448