1 /**
2 * IBM Accelerator Family 'GenWQE'
3 *
4 * (C) Copyright IBM Corp. 2013
5 *
6 * Author: Frank Haverkamp <haver@linux.vnet.ibm.com>
7 * Author: Joerg-Stephan Vogt <jsvogt@de.ibm.com>
8 * Author: Michael Jung <mijung@gmx.net>
9 * Author: Michael Ruettger <michael@ibmra.de>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License (version 2 only)
13 * as published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 */
20
21 /*
22 * Module initialization and PCIe setup. Card health monitoring and
23 * recovery functionality. Character device creation and deletion are
24 * controlled from here.
25 */
26
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/pci.h>
30 #include <linux/err.h>
31 #include <linux/aer.h>
32 #include <linux/string.h>
33 #include <linux/sched.h>
34 #include <linux/wait.h>
35 #include <linux/delay.h>
36 #include <linux/dma-mapping.h>
37 #include <linux/module.h>
38 #include <linux/notifier.h>
39 #include <linux/device.h>
40 #include <linux/log2.h>
41
42 #include "card_base.h"
43 #include "card_ddcb.h"
44
45 MODULE_AUTHOR("Frank Haverkamp <haver@linux.vnet.ibm.com>");
46 MODULE_AUTHOR("Michael Ruettger <michael@ibmra.de>");
47 MODULE_AUTHOR("Joerg-Stephan Vogt <jsvogt@de.ibm.com>");
48 MODULE_AUTHOR("Michael Jung <mijung@gmx.net>");
49
50 MODULE_DESCRIPTION("GenWQE Card");
51 MODULE_VERSION(DRV_VERSION);
52 MODULE_LICENSE("GPL");
53
54 static char genwqe_driver_name[] = GENWQE_DEVNAME;
55 static struct class *class_genwqe;
56 static struct dentry *debugfs_genwqe;
57 static struct genwqe_dev *genwqe_devices[GENWQE_CARD_NO_MAX];
58
59 /* PCI structure for identifying device by PCI vendor and device ID */
60 static const struct pci_device_id genwqe_device_table[] = {
61 { .vendor = PCI_VENDOR_ID_IBM,
62 .device = PCI_DEVICE_GENWQE,
63 .subvendor = PCI_SUBVENDOR_ID_IBM,
64 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
65 .class = (PCI_CLASSCODE_GENWQE5 << 8),
66 .class_mask = ~0,
67 .driver_data = 0 },
68
69 /* Initial SR-IOV bring-up image */
70 { .vendor = PCI_VENDOR_ID_IBM,
71 .device = PCI_DEVICE_GENWQE,
72 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
73 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
74 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
75 .class_mask = ~0,
76 .driver_data = 0 },
77
78 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
79 .device = 0x0000, /* VF Device ID */
80 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
81 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
82 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
83 .class_mask = ~0,
84 .driver_data = 0 },
85
86 /* Fixed up image */
87 { .vendor = PCI_VENDOR_ID_IBM,
88 .device = PCI_DEVICE_GENWQE,
89 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
90 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
91 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
92 .class_mask = ~0,
93 .driver_data = 0 },
94
95 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
96 .device = 0x0000, /* VF Device ID */
97 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
98 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
99 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
100 .class_mask = ~0,
101 .driver_data = 0 },
102
103 /* Even one more ... */
104 { .vendor = PCI_VENDOR_ID_IBM,
105 .device = PCI_DEVICE_GENWQE,
106 .subvendor = PCI_SUBVENDOR_ID_IBM,
107 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_NEW,
108 .class = (PCI_CLASSCODE_GENWQE5 << 8),
109 .class_mask = ~0,
110 .driver_data = 0 },
111
112 { 0, } /* 0 terminated list. */
113 };
114
115 MODULE_DEVICE_TABLE(pci, genwqe_device_table);
116
117 /**
118 * genwqe_dev_alloc() - Create and prepare a new card descriptor
119 *
120 * Return: Pointer to card descriptor, or ERR_PTR(err) on error
121 */
genwqe_dev_alloc(void)122 static struct genwqe_dev *genwqe_dev_alloc(void)
123 {
124 unsigned int i = 0, j;
125 struct genwqe_dev *cd;
126
127 for (i = 0; i < GENWQE_CARD_NO_MAX; i++) {
128 if (genwqe_devices[i] == NULL)
129 break;
130 }
131 if (i >= GENWQE_CARD_NO_MAX)
132 return ERR_PTR(-ENODEV);
133
134 cd = kzalloc(sizeof(struct genwqe_dev), GFP_KERNEL);
135 if (!cd)
136 return ERR_PTR(-ENOMEM);
137
138 cd->card_idx = i;
139 cd->class_genwqe = class_genwqe;
140 cd->debugfs_genwqe = debugfs_genwqe;
141
142 /*
143 * This comes from kernel config option and can be overritten via
144 * debugfs.
145 */
146 cd->use_platform_recovery = CONFIG_GENWQE_PLATFORM_ERROR_RECOVERY;
147
148 init_waitqueue_head(&cd->queue_waitq);
149
150 spin_lock_init(&cd->file_lock);
151 INIT_LIST_HEAD(&cd->file_list);
152
153 cd->card_state = GENWQE_CARD_UNUSED;
154 spin_lock_init(&cd->print_lock);
155
156 cd->ddcb_software_timeout = GENWQE_DDCB_SOFTWARE_TIMEOUT;
157 cd->kill_timeout = GENWQE_KILL_TIMEOUT;
158
159 for (j = 0; j < GENWQE_MAX_VFS; j++)
160 cd->vf_jobtimeout_msec[j] = GENWQE_VF_JOBTIMEOUT_MSEC;
161
162 genwqe_devices[i] = cd;
163 return cd;
164 }
165
genwqe_dev_free(struct genwqe_dev * cd)166 static void genwqe_dev_free(struct genwqe_dev *cd)
167 {
168 if (!cd)
169 return;
170
171 genwqe_devices[cd->card_idx] = NULL;
172 kfree(cd);
173 }
174
175 /**
176 * genwqe_bus_reset() - Card recovery
177 *
178 * pci_reset_function() will recover the device and ensure that the
179 * registers are accessible again when it completes with success. If
180 * not, the card will stay dead and registers will be unaccessible
181 * still.
182 */
genwqe_bus_reset(struct genwqe_dev * cd)183 static int genwqe_bus_reset(struct genwqe_dev *cd)
184 {
185 int rc = 0;
186 struct pci_dev *pci_dev = cd->pci_dev;
187 void __iomem *mmio;
188
189 if (cd->err_inject & GENWQE_INJECT_BUS_RESET_FAILURE)
190 return -EIO;
191
192 mmio = cd->mmio;
193 cd->mmio = NULL;
194 pci_iounmap(pci_dev, mmio);
195
196 pci_release_mem_regions(pci_dev);
197
198 /*
199 * Firmware/BIOS might change memory mapping during bus reset.
200 * Settings like enable bus-mastering, ... are backuped and
201 * restored by the pci_reset_function().
202 */
203 dev_dbg(&pci_dev->dev, "[%s] pci_reset function ...\n", __func__);
204 rc = pci_reset_function(pci_dev);
205 if (rc) {
206 dev_err(&pci_dev->dev,
207 "[%s] err: failed reset func (rc %d)\n", __func__, rc);
208 return rc;
209 }
210 dev_dbg(&pci_dev->dev, "[%s] done with rc=%d\n", __func__, rc);
211
212 /*
213 * Here is the right spot to clear the register read
214 * failure. pci_bus_reset() does this job in real systems.
215 */
216 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
217 GENWQE_INJECT_GFIR_FATAL |
218 GENWQE_INJECT_GFIR_INFO);
219
220 rc = pci_request_mem_regions(pci_dev, genwqe_driver_name);
221 if (rc) {
222 dev_err(&pci_dev->dev,
223 "[%s] err: request bars failed (%d)\n", __func__, rc);
224 return -EIO;
225 }
226
227 cd->mmio = pci_iomap(pci_dev, 0, 0);
228 if (cd->mmio == NULL) {
229 dev_err(&pci_dev->dev,
230 "[%s] err: mapping BAR0 failed\n", __func__);
231 return -ENOMEM;
232 }
233 return 0;
234 }
235
236 /*
237 * Hardware circumvention section. Certain bitstreams in our test-lab
238 * had different kinds of problems. Here is where we adjust those
239 * bitstreams to function will with this version of our device driver.
240 *
241 * Thise circumventions are applied to the physical function only.
242 * The magical numbers below are identifying development/manufacturing
243 * versions of the bitstream used on the card.
244 *
245 * Turn off error reporting for old/manufacturing images.
246 */
247
genwqe_need_err_masking(struct genwqe_dev * cd)248 bool genwqe_need_err_masking(struct genwqe_dev *cd)
249 {
250 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
251 }
252
genwqe_tweak_hardware(struct genwqe_dev * cd)253 static void genwqe_tweak_hardware(struct genwqe_dev *cd)
254 {
255 struct pci_dev *pci_dev = cd->pci_dev;
256
257 /* Mask FIRs for development images */
258 if (((cd->slu_unitcfg & 0xFFFF0ull) >= 0x32000ull) &&
259 ((cd->slu_unitcfg & 0xFFFF0ull) <= 0x33250ull)) {
260 dev_warn(&pci_dev->dev,
261 "FIRs masked due to bitstream %016llx.%016llx\n",
262 cd->slu_unitcfg, cd->app_unitcfg);
263
264 __genwqe_writeq(cd, IO_APP_SEC_LEM_DEBUG_OVR,
265 0xFFFFFFFFFFFFFFFFull);
266
267 __genwqe_writeq(cd, IO_APP_ERR_ACT_MASK,
268 0x0000000000000000ull);
269 }
270 }
271
272 /**
273 * genwqe_recovery_on_fatal_gfir_required() - Version depended actions
274 *
275 * Bitstreams older than 2013-02-17 have a bug where fatal GFIRs must
276 * be ignored. This is e.g. true for the bitstream we gave to the card
277 * manufacturer, but also for some old bitstreams we released to our
278 * test-lab.
279 */
genwqe_recovery_on_fatal_gfir_required(struct genwqe_dev * cd)280 int genwqe_recovery_on_fatal_gfir_required(struct genwqe_dev *cd)
281 {
282 return (cd->slu_unitcfg & 0xFFFF0ull) >= 0x32170ull;
283 }
284
genwqe_flash_readback_fails(struct genwqe_dev * cd)285 int genwqe_flash_readback_fails(struct genwqe_dev *cd)
286 {
287 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
288 }
289
290 /**
291 * genwqe_T_psec() - Calculate PF/VF timeout register content
292 *
293 * Note: From a design perspective it turned out to be a bad idea to
294 * use codes here to specifiy the frequency/speed values. An old
295 * driver cannot understand new codes and is therefore always a
296 * problem. Better is to measure out the value or put the
297 * speed/frequency directly into a register which is always a valid
298 * value for old as well as for new software.
299 */
300 /* T = 1/f */
genwqe_T_psec(struct genwqe_dev * cd)301 static int genwqe_T_psec(struct genwqe_dev *cd)
302 {
303 u16 speed; /* 1/f -> 250, 200, 166, 175 */
304 static const int T[] = { 4000, 5000, 6000, 5714 };
305
306 speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
307 if (speed >= ARRAY_SIZE(T))
308 return -1; /* illegal value */
309
310 return T[speed];
311 }
312
313 /**
314 * genwqe_setup_pf_jtimer() - Setup PF hardware timeouts for DDCB execution
315 *
316 * Do this _after_ card_reset() is called. Otherwise the values will
317 * vanish. The settings need to be done when the queues are inactive.
318 *
319 * The max. timeout value is 2^(10+x) * T (6ns for 166MHz) * 15/16.
320 * The min. timeout value is 2^(10+x) * T (6ns for 166MHz) * 14/16.
321 */
genwqe_setup_pf_jtimer(struct genwqe_dev * cd)322 static bool genwqe_setup_pf_jtimer(struct genwqe_dev *cd)
323 {
324 u32 T = genwqe_T_psec(cd);
325 u64 x;
326
327 if (GENWQE_PF_JOBTIMEOUT_MSEC == 0)
328 return false;
329
330 /* PF: large value needed, flash update 2sec per block */
331 x = ilog2(GENWQE_PF_JOBTIMEOUT_MSEC *
332 16000000000uL/(T * 15)) - 10;
333
334 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
335 0xff00 | (x & 0xff), 0);
336 return true;
337 }
338
339 /**
340 * genwqe_setup_vf_jtimer() - Setup VF hardware timeouts for DDCB execution
341 */
genwqe_setup_vf_jtimer(struct genwqe_dev * cd)342 static bool genwqe_setup_vf_jtimer(struct genwqe_dev *cd)
343 {
344 struct pci_dev *pci_dev = cd->pci_dev;
345 unsigned int vf;
346 u32 T = genwqe_T_psec(cd);
347 u64 x;
348 int totalvfs;
349
350 totalvfs = pci_sriov_get_totalvfs(pci_dev);
351 if (totalvfs <= 0)
352 return false;
353
354 for (vf = 0; vf < totalvfs; vf++) {
355
356 if (cd->vf_jobtimeout_msec[vf] == 0)
357 continue;
358
359 x = ilog2(cd->vf_jobtimeout_msec[vf] *
360 16000000000uL/(T * 15)) - 10;
361
362 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
363 0xff00 | (x & 0xff), vf + 1);
364 }
365 return true;
366 }
367
genwqe_ffdc_buffs_alloc(struct genwqe_dev * cd)368 static int genwqe_ffdc_buffs_alloc(struct genwqe_dev *cd)
369 {
370 unsigned int type, e = 0;
371
372 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
373 switch (type) {
374 case GENWQE_DBG_UNIT0:
375 e = genwqe_ffdc_buff_size(cd, 0);
376 break;
377 case GENWQE_DBG_UNIT1:
378 e = genwqe_ffdc_buff_size(cd, 1);
379 break;
380 case GENWQE_DBG_UNIT2:
381 e = genwqe_ffdc_buff_size(cd, 2);
382 break;
383 case GENWQE_DBG_REGS:
384 e = GENWQE_FFDC_REGS;
385 break;
386 }
387
388 /* currently support only the debug units mentioned here */
389 cd->ffdc[type].entries = e;
390 cd->ffdc[type].regs =
391 kmalloc_array(e, sizeof(struct genwqe_reg),
392 GFP_KERNEL);
393 /*
394 * regs == NULL is ok, the using code treats this as no regs,
395 * Printing warning is ok in this case.
396 */
397 }
398 return 0;
399 }
400
genwqe_ffdc_buffs_free(struct genwqe_dev * cd)401 static void genwqe_ffdc_buffs_free(struct genwqe_dev *cd)
402 {
403 unsigned int type;
404
405 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
406 kfree(cd->ffdc[type].regs);
407 cd->ffdc[type].regs = NULL;
408 }
409 }
410
genwqe_read_ids(struct genwqe_dev * cd)411 static int genwqe_read_ids(struct genwqe_dev *cd)
412 {
413 int err = 0;
414 int slu_id;
415 struct pci_dev *pci_dev = cd->pci_dev;
416
417 cd->slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
418 if (cd->slu_unitcfg == IO_ILLEGAL_VALUE) {
419 dev_err(&pci_dev->dev,
420 "err: SLUID=%016llx\n", cd->slu_unitcfg);
421 err = -EIO;
422 goto out_err;
423 }
424
425 slu_id = genwqe_get_slu_id(cd);
426 if (slu_id < GENWQE_SLU_ARCH_REQ || slu_id == 0xff) {
427 dev_err(&pci_dev->dev,
428 "err: incompatible SLU Architecture %u\n", slu_id);
429 err = -ENOENT;
430 goto out_err;
431 }
432
433 cd->app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
434 if (cd->app_unitcfg == IO_ILLEGAL_VALUE) {
435 dev_err(&pci_dev->dev,
436 "err: APPID=%016llx\n", cd->app_unitcfg);
437 err = -EIO;
438 goto out_err;
439 }
440 genwqe_read_app_id(cd, cd->app_name, sizeof(cd->app_name));
441
442 /*
443 * Is access to all registers possible? If we are a VF the
444 * answer is obvious. If we run fully virtualized, we need to
445 * check if we can access all registers. If we do not have
446 * full access we will cause an UR and some informational FIRs
447 * in the PF, but that should not harm.
448 */
449 if (pci_dev->is_virtfn)
450 cd->is_privileged = 0;
451 else
452 cd->is_privileged = (__genwqe_readq(cd, IO_SLU_BITSTREAM)
453 != IO_ILLEGAL_VALUE);
454
455 out_err:
456 return err;
457 }
458
genwqe_start(struct genwqe_dev * cd)459 static int genwqe_start(struct genwqe_dev *cd)
460 {
461 int err;
462 struct pci_dev *pci_dev = cd->pci_dev;
463
464 err = genwqe_read_ids(cd);
465 if (err)
466 return err;
467
468 if (genwqe_is_privileged(cd)) {
469 /* do this after the tweaks. alloc fail is acceptable */
470 genwqe_ffdc_buffs_alloc(cd);
471 genwqe_stop_traps(cd);
472
473 /* Collect registers e.g. FIRs, UNITIDs, traces ... */
474 genwqe_read_ffdc_regs(cd, cd->ffdc[GENWQE_DBG_REGS].regs,
475 cd->ffdc[GENWQE_DBG_REGS].entries, 0);
476
477 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT0,
478 cd->ffdc[GENWQE_DBG_UNIT0].regs,
479 cd->ffdc[GENWQE_DBG_UNIT0].entries);
480
481 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT1,
482 cd->ffdc[GENWQE_DBG_UNIT1].regs,
483 cd->ffdc[GENWQE_DBG_UNIT1].entries);
484
485 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT2,
486 cd->ffdc[GENWQE_DBG_UNIT2].regs,
487 cd->ffdc[GENWQE_DBG_UNIT2].entries);
488
489 genwqe_start_traps(cd);
490
491 if (cd->card_state == GENWQE_CARD_FATAL_ERROR) {
492 dev_warn(&pci_dev->dev,
493 "[%s] chip reload/recovery!\n", __func__);
494
495 /*
496 * Stealth Mode: Reload chip on either hot
497 * reset or PERST.
498 */
499 cd->softreset = 0x7Cull;
500 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
501 cd->softreset);
502
503 err = genwqe_bus_reset(cd);
504 if (err != 0) {
505 dev_err(&pci_dev->dev,
506 "[%s] err: bus reset failed!\n",
507 __func__);
508 goto out;
509 }
510
511 /*
512 * Re-read the IDs because
513 * it could happen that the bitstream load
514 * failed!
515 */
516 err = genwqe_read_ids(cd);
517 if (err)
518 goto out;
519 }
520 }
521
522 err = genwqe_setup_service_layer(cd); /* does a reset to the card */
523 if (err != 0) {
524 dev_err(&pci_dev->dev,
525 "[%s] err: could not setup servicelayer!\n", __func__);
526 err = -ENODEV;
527 goto out;
528 }
529
530 if (genwqe_is_privileged(cd)) { /* code is running _after_ reset */
531 genwqe_tweak_hardware(cd);
532
533 genwqe_setup_pf_jtimer(cd);
534 genwqe_setup_vf_jtimer(cd);
535 }
536
537 err = genwqe_device_create(cd);
538 if (err < 0) {
539 dev_err(&pci_dev->dev,
540 "err: chdev init failed! (err=%d)\n", err);
541 goto out_release_service_layer;
542 }
543 return 0;
544
545 out_release_service_layer:
546 genwqe_release_service_layer(cd);
547 out:
548 if (genwqe_is_privileged(cd))
549 genwqe_ffdc_buffs_free(cd);
550 return -EIO;
551 }
552
553 /**
554 * genwqe_stop() - Stop card operation
555 *
556 * Recovery notes:
557 * As long as genwqe_thread runs we might access registers during
558 * error data capture. Same is with the genwqe_health_thread.
559 * When genwqe_bus_reset() fails this function might called two times:
560 * first by the genwqe_health_thread() and later by genwqe_remove() to
561 * unbind the device. We must be able to survive that.
562 *
563 * This function must be robust enough to be called twice.
564 */
genwqe_stop(struct genwqe_dev * cd)565 static int genwqe_stop(struct genwqe_dev *cd)
566 {
567 genwqe_finish_queue(cd); /* no register access */
568 genwqe_device_remove(cd); /* device removed, procs killed */
569 genwqe_release_service_layer(cd); /* here genwqe_thread is stopped */
570
571 if (genwqe_is_privileged(cd)) {
572 pci_disable_sriov(cd->pci_dev); /* access pci config space */
573 genwqe_ffdc_buffs_free(cd);
574 }
575
576 return 0;
577 }
578
579 /**
580 * genwqe_recover_card() - Try to recover the card if it is possible
581 *
582 * If fatal_err is set no register access is possible anymore. It is
583 * likely that genwqe_start fails in that situation. Proper error
584 * handling is required in this case.
585 *
586 * genwqe_bus_reset() will cause the pci code to call genwqe_remove()
587 * and later genwqe_probe() for all virtual functions.
588 */
genwqe_recover_card(struct genwqe_dev * cd,int fatal_err)589 static int genwqe_recover_card(struct genwqe_dev *cd, int fatal_err)
590 {
591 int rc;
592 struct pci_dev *pci_dev = cd->pci_dev;
593
594 genwqe_stop(cd);
595
596 /*
597 * Make sure chip is not reloaded to maintain FFDC. Write SLU
598 * Reset Register, CPLDReset field to 0.
599 */
600 if (!fatal_err) {
601 cd->softreset = 0x70ull;
602 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, cd->softreset);
603 }
604
605 rc = genwqe_bus_reset(cd);
606 if (rc != 0) {
607 dev_err(&pci_dev->dev,
608 "[%s] err: card recovery impossible!\n", __func__);
609 return rc;
610 }
611
612 rc = genwqe_start(cd);
613 if (rc < 0) {
614 dev_err(&pci_dev->dev,
615 "[%s] err: failed to launch device!\n", __func__);
616 return rc;
617 }
618 return 0;
619 }
620
genwqe_health_check_cond(struct genwqe_dev * cd,u64 * gfir)621 static int genwqe_health_check_cond(struct genwqe_dev *cd, u64 *gfir)
622 {
623 *gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
624 return (*gfir & GFIR_ERR_TRIGGER) &&
625 genwqe_recovery_on_fatal_gfir_required(cd);
626 }
627
628 /**
629 * genwqe_fir_checking() - Check the fault isolation registers of the card
630 *
631 * If this code works ok, can be tried out with help of the genwqe_poke tool:
632 * sudo ./tools/genwqe_poke 0x8 0xfefefefefef
633 *
634 * Now the relevant FIRs/sFIRs should be printed out and the driver should
635 * invoke recovery (devices are removed and readded).
636 */
genwqe_fir_checking(struct genwqe_dev * cd)637 static u64 genwqe_fir_checking(struct genwqe_dev *cd)
638 {
639 int j, iterations = 0;
640 u64 mask, fir, fec, uid, gfir, gfir_masked, sfir, sfec;
641 u32 fir_addr, fir_clr_addr, fec_addr, sfir_addr, sfec_addr;
642 struct pci_dev *pci_dev = cd->pci_dev;
643
644 healthMonitor:
645 iterations++;
646 if (iterations > 16) {
647 dev_err(&pci_dev->dev, "* exit looping after %d times\n",
648 iterations);
649 goto fatal_error;
650 }
651
652 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
653 if (gfir != 0x0)
654 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n",
655 IO_SLC_CFGREG_GFIR, gfir);
656 if (gfir == IO_ILLEGAL_VALUE)
657 goto fatal_error;
658
659 /*
660 * Avoid printing when to GFIR bit is on prevents contignous
661 * printout e.g. for the following bug:
662 * FIR set without a 2ndary FIR/FIR cannot be cleared
663 * Comment out the following if to get the prints:
664 */
665 if (gfir == 0)
666 return 0;
667
668 gfir_masked = gfir & GFIR_ERR_TRIGGER; /* fatal errors */
669
670 for (uid = 0; uid < GENWQE_MAX_UNITS; uid++) { /* 0..2 in zEDC */
671
672 /* read the primary FIR (pfir) */
673 fir_addr = (uid << 24) + 0x08;
674 fir = __genwqe_readq(cd, fir_addr);
675 if (fir == 0x0)
676 continue; /* no error in this unit */
677
678 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fir_addr, fir);
679 if (fir == IO_ILLEGAL_VALUE)
680 goto fatal_error;
681
682 /* read primary FEC */
683 fec_addr = (uid << 24) + 0x18;
684 fec = __genwqe_readq(cd, fec_addr);
685
686 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fec_addr, fec);
687 if (fec == IO_ILLEGAL_VALUE)
688 goto fatal_error;
689
690 for (j = 0, mask = 1ULL; j < 64; j++, mask <<= 1) {
691
692 /* secondary fir empty, skip it */
693 if ((fir & mask) == 0x0)
694 continue;
695
696 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
697 sfir = __genwqe_readq(cd, sfir_addr);
698
699 if (sfir == IO_ILLEGAL_VALUE)
700 goto fatal_error;
701 dev_err(&pci_dev->dev,
702 "* 0x%08x 0x%016llx\n", sfir_addr, sfir);
703
704 sfec_addr = (uid << 24) + 0x300 + 0x08 * j;
705 sfec = __genwqe_readq(cd, sfec_addr);
706
707 if (sfec == IO_ILLEGAL_VALUE)
708 goto fatal_error;
709 dev_err(&pci_dev->dev,
710 "* 0x%08x 0x%016llx\n", sfec_addr, sfec);
711
712 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
713 if (gfir == IO_ILLEGAL_VALUE)
714 goto fatal_error;
715
716 /* gfir turned on during routine! get out and
717 start over. */
718 if ((gfir_masked == 0x0) &&
719 (gfir & GFIR_ERR_TRIGGER)) {
720 goto healthMonitor;
721 }
722
723 /* do not clear if we entered with a fatal gfir */
724 if (gfir_masked == 0x0) {
725
726 /* NEW clear by mask the logged bits */
727 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
728 __genwqe_writeq(cd, sfir_addr, sfir);
729
730 dev_dbg(&pci_dev->dev,
731 "[HM] Clearing 2ndary FIR 0x%08x with 0x%016llx\n",
732 sfir_addr, sfir);
733
734 /*
735 * note, these cannot be error-Firs
736 * since gfir_masked is 0 after sfir
737 * was read. Also, it is safe to do
738 * this write if sfir=0. Still need to
739 * clear the primary. This just means
740 * there is no secondary FIR.
741 */
742
743 /* clear by mask the logged bit. */
744 fir_clr_addr = (uid << 24) + 0x10;
745 __genwqe_writeq(cd, fir_clr_addr, mask);
746
747 dev_dbg(&pci_dev->dev,
748 "[HM] Clearing primary FIR 0x%08x with 0x%016llx\n",
749 fir_clr_addr, mask);
750 }
751 }
752 }
753 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
754 if (gfir == IO_ILLEGAL_VALUE)
755 goto fatal_error;
756
757 if ((gfir_masked == 0x0) && (gfir & GFIR_ERR_TRIGGER)) {
758 /*
759 * Check once more that it didn't go on after all the
760 * FIRS were cleared.
761 */
762 dev_dbg(&pci_dev->dev, "ACK! Another FIR! Recursing %d!\n",
763 iterations);
764 goto healthMonitor;
765 }
766 return gfir_masked;
767
768 fatal_error:
769 return IO_ILLEGAL_VALUE;
770 }
771
772 /**
773 * genwqe_pci_fundamental_reset() - trigger a PCIe fundamental reset on the slot
774 *
775 * Note: pci_set_pcie_reset_state() is not implemented on all archs, so this
776 * reset method will not work in all cases.
777 *
778 * Return: 0 on success or error code from pci_set_pcie_reset_state()
779 */
genwqe_pci_fundamental_reset(struct pci_dev * pci_dev)780 static int genwqe_pci_fundamental_reset(struct pci_dev *pci_dev)
781 {
782 int rc;
783
784 /*
785 * lock pci config space access from userspace,
786 * save state and issue PCIe fundamental reset
787 */
788 pci_cfg_access_lock(pci_dev);
789 pci_save_state(pci_dev);
790 rc = pci_set_pcie_reset_state(pci_dev, pcie_warm_reset);
791 if (!rc) {
792 /* keep PCIe reset asserted for 250ms */
793 msleep(250);
794 pci_set_pcie_reset_state(pci_dev, pcie_deassert_reset);
795 /* Wait for 2s to reload flash and train the link */
796 msleep(2000);
797 }
798 pci_restore_state(pci_dev);
799 pci_cfg_access_unlock(pci_dev);
800 return rc;
801 }
802
803
genwqe_platform_recovery(struct genwqe_dev * cd)804 static int genwqe_platform_recovery(struct genwqe_dev *cd)
805 {
806 struct pci_dev *pci_dev = cd->pci_dev;
807 int rc;
808
809 dev_info(&pci_dev->dev,
810 "[%s] resetting card for error recovery\n", __func__);
811
812 /* Clear out error injection flags */
813 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
814 GENWQE_INJECT_GFIR_FATAL |
815 GENWQE_INJECT_GFIR_INFO);
816
817 genwqe_stop(cd);
818
819 /* Try recoverying the card with fundamental reset */
820 rc = genwqe_pci_fundamental_reset(pci_dev);
821 if (!rc) {
822 rc = genwqe_start(cd);
823 if (!rc)
824 dev_info(&pci_dev->dev,
825 "[%s] card recovered\n", __func__);
826 else
827 dev_err(&pci_dev->dev,
828 "[%s] err: cannot start card services! (err=%d)\n",
829 __func__, rc);
830 } else {
831 dev_err(&pci_dev->dev,
832 "[%s] card reset failed\n", __func__);
833 }
834
835 return rc;
836 }
837
838 /*
839 * genwqe_reload_bistream() - reload card bitstream
840 *
841 * Set the appropriate register and call fundamental reset to reaload the card
842 * bitstream.
843 *
844 * Return: 0 on success, error code otherwise
845 */
genwqe_reload_bistream(struct genwqe_dev * cd)846 static int genwqe_reload_bistream(struct genwqe_dev *cd)
847 {
848 struct pci_dev *pci_dev = cd->pci_dev;
849 int rc;
850
851 dev_info(&pci_dev->dev,
852 "[%s] resetting card for bitstream reload\n",
853 __func__);
854
855 genwqe_stop(cd);
856
857 /*
858 * Cause a CPLD reprogram with the 'next_bitstream'
859 * partition on PCIe hot or fundamental reset
860 */
861 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
862 (cd->softreset & 0xcull) | 0x70ull);
863
864 rc = genwqe_pci_fundamental_reset(pci_dev);
865 if (rc) {
866 /*
867 * A fundamental reset failure can be caused
868 * by lack of support on the arch, so we just
869 * log the error and try to start the card
870 * again.
871 */
872 dev_err(&pci_dev->dev,
873 "[%s] err: failed to reset card for bitstream reload\n",
874 __func__);
875 }
876
877 rc = genwqe_start(cd);
878 if (rc) {
879 dev_err(&pci_dev->dev,
880 "[%s] err: cannot start card services! (err=%d)\n",
881 __func__, rc);
882 return rc;
883 }
884 dev_info(&pci_dev->dev,
885 "[%s] card reloaded\n", __func__);
886 return 0;
887 }
888
889
890 /**
891 * genwqe_health_thread() - Health checking thread
892 *
893 * This thread is only started for the PF of the card.
894 *
895 * This thread monitors the health of the card. A critical situation
896 * is when we read registers which contain -1 (IO_ILLEGAL_VALUE). In
897 * this case we need to be recovered from outside. Writing to
898 * registers will very likely not work either.
899 *
900 * This thread must only exit if kthread_should_stop() becomes true.
901 *
902 * Condition for the health-thread to trigger:
903 * a) when a kthread_stop() request comes in or
904 * b) a critical GFIR occured
905 *
906 * Informational GFIRs are checked and potentially printed in
907 * GENWQE_HEALTH_CHECK_INTERVAL seconds.
908 */
genwqe_health_thread(void * data)909 static int genwqe_health_thread(void *data)
910 {
911 int rc, should_stop = 0;
912 struct genwqe_dev *cd = data;
913 struct pci_dev *pci_dev = cd->pci_dev;
914 u64 gfir, gfir_masked, slu_unitcfg, app_unitcfg;
915
916 health_thread_begin:
917 while (!kthread_should_stop()) {
918 rc = wait_event_interruptible_timeout(cd->health_waitq,
919 (genwqe_health_check_cond(cd, &gfir) ||
920 (should_stop = kthread_should_stop())),
921 GENWQE_HEALTH_CHECK_INTERVAL * HZ);
922
923 if (should_stop)
924 break;
925
926 if (gfir == IO_ILLEGAL_VALUE) {
927 dev_err(&pci_dev->dev,
928 "[%s] GFIR=%016llx\n", __func__, gfir);
929 goto fatal_error;
930 }
931
932 slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
933 if (slu_unitcfg == IO_ILLEGAL_VALUE) {
934 dev_err(&pci_dev->dev,
935 "[%s] SLU_UNITCFG=%016llx\n",
936 __func__, slu_unitcfg);
937 goto fatal_error;
938 }
939
940 app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
941 if (app_unitcfg == IO_ILLEGAL_VALUE) {
942 dev_err(&pci_dev->dev,
943 "[%s] APP_UNITCFG=%016llx\n",
944 __func__, app_unitcfg);
945 goto fatal_error;
946 }
947
948 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
949 if (gfir == IO_ILLEGAL_VALUE) {
950 dev_err(&pci_dev->dev,
951 "[%s] %s: GFIR=%016llx\n", __func__,
952 (gfir & GFIR_ERR_TRIGGER) ? "err" : "info",
953 gfir);
954 goto fatal_error;
955 }
956
957 gfir_masked = genwqe_fir_checking(cd);
958 if (gfir_masked == IO_ILLEGAL_VALUE)
959 goto fatal_error;
960
961 /*
962 * GFIR ErrorTrigger bits set => reset the card!
963 * Never do this for old/manufacturing images!
964 */
965 if ((gfir_masked) && !cd->skip_recovery &&
966 genwqe_recovery_on_fatal_gfir_required(cd)) {
967
968 cd->card_state = GENWQE_CARD_FATAL_ERROR;
969
970 rc = genwqe_recover_card(cd, 0);
971 if (rc < 0) {
972 /* FIXME Card is unusable and needs unbind! */
973 goto fatal_error;
974 }
975 }
976
977 if (cd->card_state == GENWQE_CARD_RELOAD_BITSTREAM) {
978 /* Userspace requested card bitstream reload */
979 rc = genwqe_reload_bistream(cd);
980 if (rc)
981 goto fatal_error;
982 }
983
984 cd->last_gfir = gfir;
985 cond_resched();
986 }
987
988 return 0;
989
990 fatal_error:
991 if (cd->use_platform_recovery) {
992 /*
993 * Since we use raw accessors, EEH errors won't be detected
994 * by the platform until we do a non-raw MMIO or config space
995 * read
996 */
997 readq(cd->mmio + IO_SLC_CFGREG_GFIR);
998
999 /* We do nothing if the card is going over PCI recovery */
1000 if (pci_channel_offline(pci_dev))
1001 return -EIO;
1002
1003 /*
1004 * If it's supported by the platform, we try a fundamental reset
1005 * to recover from a fatal error. Otherwise, we continue to wait
1006 * for an external recovery procedure to take care of it.
1007 */
1008 rc = genwqe_platform_recovery(cd);
1009 if (!rc)
1010 goto health_thread_begin;
1011 }
1012
1013 dev_err(&pci_dev->dev,
1014 "[%s] card unusable. Please trigger unbind!\n", __func__);
1015
1016 /* Bring down logical devices to inform user space via udev remove. */
1017 cd->card_state = GENWQE_CARD_FATAL_ERROR;
1018 genwqe_stop(cd);
1019
1020 /* genwqe_bus_reset failed(). Now wait for genwqe_remove(). */
1021 while (!kthread_should_stop())
1022 cond_resched();
1023
1024 return -EIO;
1025 }
1026
genwqe_health_check_start(struct genwqe_dev * cd)1027 static int genwqe_health_check_start(struct genwqe_dev *cd)
1028 {
1029 int rc;
1030
1031 if (GENWQE_HEALTH_CHECK_INTERVAL <= 0)
1032 return 0; /* valid for disabling the service */
1033
1034 /* moved before request_irq() */
1035 /* init_waitqueue_head(&cd->health_waitq); */
1036
1037 cd->health_thread = kthread_run(genwqe_health_thread, cd,
1038 GENWQE_DEVNAME "%d_health",
1039 cd->card_idx);
1040 if (IS_ERR(cd->health_thread)) {
1041 rc = PTR_ERR(cd->health_thread);
1042 cd->health_thread = NULL;
1043 return rc;
1044 }
1045 return 0;
1046 }
1047
genwqe_health_thread_running(struct genwqe_dev * cd)1048 static int genwqe_health_thread_running(struct genwqe_dev *cd)
1049 {
1050 return cd->health_thread != NULL;
1051 }
1052
genwqe_health_check_stop(struct genwqe_dev * cd)1053 static int genwqe_health_check_stop(struct genwqe_dev *cd)
1054 {
1055 int rc;
1056
1057 if (!genwqe_health_thread_running(cd))
1058 return -EIO;
1059
1060 rc = kthread_stop(cd->health_thread);
1061 cd->health_thread = NULL;
1062 return 0;
1063 }
1064
1065 /**
1066 * genwqe_pci_setup() - Allocate PCIe related resources for our card
1067 */
genwqe_pci_setup(struct genwqe_dev * cd)1068 static int genwqe_pci_setup(struct genwqe_dev *cd)
1069 {
1070 int err;
1071 struct pci_dev *pci_dev = cd->pci_dev;
1072
1073 err = pci_enable_device_mem(pci_dev);
1074 if (err) {
1075 dev_err(&pci_dev->dev,
1076 "err: failed to enable pci memory (err=%d)\n", err);
1077 goto err_out;
1078 }
1079
1080 /* Reserve PCI I/O and memory resources */
1081 err = pci_request_mem_regions(pci_dev, genwqe_driver_name);
1082 if (err) {
1083 dev_err(&pci_dev->dev,
1084 "[%s] err: request bars failed (%d)\n", __func__, err);
1085 err = -EIO;
1086 goto err_disable_device;
1087 }
1088
1089 /* check for 64-bit DMA address supported (DAC) */
1090 if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(64))) {
1091 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(64));
1092 if (err) {
1093 dev_err(&pci_dev->dev,
1094 "err: DMA64 consistent mask error\n");
1095 err = -EIO;
1096 goto out_release_resources;
1097 }
1098 /* check for 32-bit DMA address supported (SAC) */
1099 } else if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(32))) {
1100 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(32));
1101 if (err) {
1102 dev_err(&pci_dev->dev,
1103 "err: DMA32 consistent mask error\n");
1104 err = -EIO;
1105 goto out_release_resources;
1106 }
1107 } else {
1108 dev_err(&pci_dev->dev,
1109 "err: neither DMA32 nor DMA64 supported\n");
1110 err = -EIO;
1111 goto out_release_resources;
1112 }
1113
1114 pci_set_master(pci_dev);
1115 pci_enable_pcie_error_reporting(pci_dev);
1116
1117 /* EEH recovery requires PCIe fundamental reset */
1118 pci_dev->needs_freset = 1;
1119
1120 /* request complete BAR-0 space (length = 0) */
1121 cd->mmio_len = pci_resource_len(pci_dev, 0);
1122 cd->mmio = pci_iomap(pci_dev, 0, 0);
1123 if (cd->mmio == NULL) {
1124 dev_err(&pci_dev->dev,
1125 "[%s] err: mapping BAR0 failed\n", __func__);
1126 err = -ENOMEM;
1127 goto out_release_resources;
1128 }
1129
1130 cd->num_vfs = pci_sriov_get_totalvfs(pci_dev);
1131 if (cd->num_vfs < 0)
1132 cd->num_vfs = 0;
1133
1134 err = genwqe_read_ids(cd);
1135 if (err)
1136 goto out_iounmap;
1137
1138 return 0;
1139
1140 out_iounmap:
1141 pci_iounmap(pci_dev, cd->mmio);
1142 out_release_resources:
1143 pci_release_mem_regions(pci_dev);
1144 err_disable_device:
1145 pci_disable_device(pci_dev);
1146 err_out:
1147 return err;
1148 }
1149
1150 /**
1151 * genwqe_pci_remove() - Free PCIe related resources for our card
1152 */
genwqe_pci_remove(struct genwqe_dev * cd)1153 static void genwqe_pci_remove(struct genwqe_dev *cd)
1154 {
1155 struct pci_dev *pci_dev = cd->pci_dev;
1156
1157 if (cd->mmio)
1158 pci_iounmap(pci_dev, cd->mmio);
1159
1160 pci_release_mem_regions(pci_dev);
1161 pci_disable_device(pci_dev);
1162 }
1163
1164 /**
1165 * genwqe_probe() - Device initialization
1166 * @pdev: PCI device information struct
1167 *
1168 * Callable for multiple cards. This function is called on bind.
1169 *
1170 * Return: 0 if succeeded, < 0 when failed
1171 */
genwqe_probe(struct pci_dev * pci_dev,const struct pci_device_id * id)1172 static int genwqe_probe(struct pci_dev *pci_dev,
1173 const struct pci_device_id *id)
1174 {
1175 int err;
1176 struct genwqe_dev *cd;
1177
1178 genwqe_init_crc32();
1179
1180 cd = genwqe_dev_alloc();
1181 if (IS_ERR(cd)) {
1182 dev_err(&pci_dev->dev, "err: could not alloc mem (err=%d)!\n",
1183 (int)PTR_ERR(cd));
1184 return PTR_ERR(cd);
1185 }
1186
1187 dev_set_drvdata(&pci_dev->dev, cd);
1188 cd->pci_dev = pci_dev;
1189
1190 err = genwqe_pci_setup(cd);
1191 if (err < 0) {
1192 dev_err(&pci_dev->dev,
1193 "err: problems with PCI setup (err=%d)\n", err);
1194 goto out_free_dev;
1195 }
1196
1197 err = genwqe_start(cd);
1198 if (err < 0) {
1199 dev_err(&pci_dev->dev,
1200 "err: cannot start card services! (err=%d)\n", err);
1201 goto out_pci_remove;
1202 }
1203
1204 if (genwqe_is_privileged(cd)) {
1205 err = genwqe_health_check_start(cd);
1206 if (err < 0) {
1207 dev_err(&pci_dev->dev,
1208 "err: cannot start health checking! (err=%d)\n",
1209 err);
1210 goto out_stop_services;
1211 }
1212 }
1213 return 0;
1214
1215 out_stop_services:
1216 genwqe_stop(cd);
1217 out_pci_remove:
1218 genwqe_pci_remove(cd);
1219 out_free_dev:
1220 genwqe_dev_free(cd);
1221 return err;
1222 }
1223
1224 /**
1225 * genwqe_remove() - Called when device is removed (hot-plugable)
1226 *
1227 * Or when driver is unloaded respecitively when unbind is done.
1228 */
genwqe_remove(struct pci_dev * pci_dev)1229 static void genwqe_remove(struct pci_dev *pci_dev)
1230 {
1231 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1232
1233 genwqe_health_check_stop(cd);
1234
1235 /*
1236 * genwqe_stop() must survive if it is called twice
1237 * sequentially. This happens when the health thread calls it
1238 * and fails on genwqe_bus_reset().
1239 */
1240 genwqe_stop(cd);
1241 genwqe_pci_remove(cd);
1242 genwqe_dev_free(cd);
1243 }
1244
1245 /*
1246 * genwqe_err_error_detected() - Error detection callback
1247 *
1248 * This callback is called by the PCI subsystem whenever a PCI bus
1249 * error is detected.
1250 */
genwqe_err_error_detected(struct pci_dev * pci_dev,enum pci_channel_state state)1251 static pci_ers_result_t genwqe_err_error_detected(struct pci_dev *pci_dev,
1252 enum pci_channel_state state)
1253 {
1254 struct genwqe_dev *cd;
1255
1256 dev_err(&pci_dev->dev, "[%s] state=%d\n", __func__, state);
1257
1258 cd = dev_get_drvdata(&pci_dev->dev);
1259 if (cd == NULL)
1260 return PCI_ERS_RESULT_DISCONNECT;
1261
1262 /* Stop the card */
1263 genwqe_health_check_stop(cd);
1264 genwqe_stop(cd);
1265
1266 /*
1267 * On permanent failure, the PCI code will call device remove
1268 * after the return of this function.
1269 * genwqe_stop() can be called twice.
1270 */
1271 if (state == pci_channel_io_perm_failure) {
1272 return PCI_ERS_RESULT_DISCONNECT;
1273 } else {
1274 genwqe_pci_remove(cd);
1275 return PCI_ERS_RESULT_NEED_RESET;
1276 }
1277 }
1278
genwqe_err_slot_reset(struct pci_dev * pci_dev)1279 static pci_ers_result_t genwqe_err_slot_reset(struct pci_dev *pci_dev)
1280 {
1281 int rc;
1282 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1283
1284 rc = genwqe_pci_setup(cd);
1285 if (!rc) {
1286 return PCI_ERS_RESULT_RECOVERED;
1287 } else {
1288 dev_err(&pci_dev->dev,
1289 "err: problems with PCI setup (err=%d)\n", rc);
1290 return PCI_ERS_RESULT_DISCONNECT;
1291 }
1292 }
1293
genwqe_err_result_none(struct pci_dev * dev)1294 static pci_ers_result_t genwqe_err_result_none(struct pci_dev *dev)
1295 {
1296 return PCI_ERS_RESULT_NONE;
1297 }
1298
genwqe_err_resume(struct pci_dev * pci_dev)1299 static void genwqe_err_resume(struct pci_dev *pci_dev)
1300 {
1301 int rc;
1302 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1303
1304 rc = genwqe_start(cd);
1305 if (!rc) {
1306 rc = genwqe_health_check_start(cd);
1307 if (rc)
1308 dev_err(&pci_dev->dev,
1309 "err: cannot start health checking! (err=%d)\n",
1310 rc);
1311 } else {
1312 dev_err(&pci_dev->dev,
1313 "err: cannot start card services! (err=%d)\n", rc);
1314 }
1315 }
1316
genwqe_sriov_configure(struct pci_dev * dev,int numvfs)1317 static int genwqe_sriov_configure(struct pci_dev *dev, int numvfs)
1318 {
1319 int rc;
1320 struct genwqe_dev *cd = dev_get_drvdata(&dev->dev);
1321
1322 if (numvfs > 0) {
1323 genwqe_setup_vf_jtimer(cd);
1324 rc = pci_enable_sriov(dev, numvfs);
1325 if (rc < 0)
1326 return rc;
1327 return numvfs;
1328 }
1329 if (numvfs == 0) {
1330 pci_disable_sriov(dev);
1331 return 0;
1332 }
1333 return 0;
1334 }
1335
1336 static struct pci_error_handlers genwqe_err_handler = {
1337 .error_detected = genwqe_err_error_detected,
1338 .mmio_enabled = genwqe_err_result_none,
1339 .slot_reset = genwqe_err_slot_reset,
1340 .resume = genwqe_err_resume,
1341 };
1342
1343 static struct pci_driver genwqe_driver = {
1344 .name = genwqe_driver_name,
1345 .id_table = genwqe_device_table,
1346 .probe = genwqe_probe,
1347 .remove = genwqe_remove,
1348 .sriov_configure = genwqe_sriov_configure,
1349 .err_handler = &genwqe_err_handler,
1350 };
1351
1352 /**
1353 * genwqe_devnode() - Set default access mode for genwqe devices.
1354 *
1355 * Default mode should be rw for everybody. Do not change default
1356 * device name.
1357 */
genwqe_devnode(struct device * dev,umode_t * mode)1358 static char *genwqe_devnode(struct device *dev, umode_t *mode)
1359 {
1360 if (mode)
1361 *mode = 0666;
1362 return NULL;
1363 }
1364
1365 /**
1366 * genwqe_init_module() - Driver registration and initialization
1367 */
genwqe_init_module(void)1368 static int __init genwqe_init_module(void)
1369 {
1370 int rc;
1371
1372 class_genwqe = class_create(THIS_MODULE, GENWQE_DEVNAME);
1373 if (IS_ERR(class_genwqe)) {
1374 pr_err("[%s] create class failed\n", __func__);
1375 return -ENOMEM;
1376 }
1377
1378 class_genwqe->devnode = genwqe_devnode;
1379
1380 debugfs_genwqe = debugfs_create_dir(GENWQE_DEVNAME, NULL);
1381 if (!debugfs_genwqe) {
1382 rc = -ENOMEM;
1383 goto err_out;
1384 }
1385
1386 rc = pci_register_driver(&genwqe_driver);
1387 if (rc != 0) {
1388 pr_err("[%s] pci_reg_driver (rc=%d)\n", __func__, rc);
1389 goto err_out0;
1390 }
1391
1392 return rc;
1393
1394 err_out0:
1395 debugfs_remove(debugfs_genwqe);
1396 err_out:
1397 class_destroy(class_genwqe);
1398 return rc;
1399 }
1400
1401 /**
1402 * genwqe_exit_module() - Driver exit
1403 */
genwqe_exit_module(void)1404 static void __exit genwqe_exit_module(void)
1405 {
1406 pci_unregister_driver(&genwqe_driver);
1407 debugfs_remove(debugfs_genwqe);
1408 class_destroy(class_genwqe);
1409 }
1410
1411 module_init(genwqe_init_module);
1412 module_exit(genwqe_exit_module);
1413