Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
[linux/fpc-iii.git] / drivers / misc / genwqe / card_base.c
blob74d51c9bb85870df4ace01ac4d4173531aae5984
1 /**
2 * IBM Accelerator Family 'GenWQE'
4 * (C) Copyright IBM Corp. 2013
6 * Author: Frank Haverkamp <haver@linux.vnet.ibm.com>
7 * Author: Joerg-Stephan Vogt <jsvogt@de.ibm.com>
8 * Author: Michael Jung <mijung@de.ibm.com>
9 * Author: Michael Ruettger <michael@ibmra.de>
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.
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.
22 * Module initialization and PCIe setup. Card health monitoring and
23 * recovery functionality. Character device creation and deletion are
24 * controlled from here.
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 #include <linux/genwqe/genwqe_card.h>
43 #include "card_base.h"
44 #include "card_ddcb.h"
46 MODULE_AUTHOR("Frank Haverkamp <haver@linux.vnet.ibm.com>");
47 MODULE_AUTHOR("Michael Ruettger <michael@ibmra.de>");
48 MODULE_AUTHOR("Joerg-Stephan Vogt <jsvogt@de.ibm.com>");
49 MODULE_AUTHOR("Michal Jung <mijung@de.ibm.com>");
51 MODULE_DESCRIPTION("GenWQE Card");
52 MODULE_VERSION(DRV_VERS_STRING);
53 MODULE_LICENSE("GPL");
55 static char genwqe_driver_name[] = GENWQE_DEVNAME;
56 static struct class *class_genwqe;
57 static struct dentry *debugfs_genwqe;
58 static struct genwqe_dev *genwqe_devices[GENWQE_CARD_NO_MAX];
60 /* PCI structure for identifying device by PCI vendor and device ID */
61 static DEFINE_PCI_DEVICE_TABLE(genwqe_device_table) = {
62 { .vendor = PCI_VENDOR_ID_IBM,
63 .device = PCI_DEVICE_GENWQE,
64 .subvendor = PCI_SUBVENDOR_ID_IBM,
65 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
66 .class = (PCI_CLASSCODE_GENWQE5 << 8),
67 .class_mask = ~0,
68 .driver_data = 0 },
70 /* Initial SR-IOV bring-up image */
71 { .vendor = PCI_VENDOR_ID_IBM,
72 .device = PCI_DEVICE_GENWQE,
73 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
74 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
75 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
76 .class_mask = ~0,
77 .driver_data = 0 },
79 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
80 .device = 0x0000, /* VF Device ID */
81 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
82 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
83 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
84 .class_mask = ~0,
85 .driver_data = 0 },
87 /* Fixed up image */
88 { .vendor = PCI_VENDOR_ID_IBM,
89 .device = PCI_DEVICE_GENWQE,
90 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
91 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
92 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
93 .class_mask = ~0,
94 .driver_data = 0 },
96 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
97 .device = 0x0000, /* VF Device ID */
98 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
99 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
100 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
101 .class_mask = ~0,
102 .driver_data = 0 },
104 /* Even one more ... */
105 { .vendor = PCI_VENDOR_ID_IBM,
106 .device = PCI_DEVICE_GENWQE,
107 .subvendor = PCI_SUBVENDOR_ID_IBM,
108 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_NEW,
109 .class = (PCI_CLASSCODE_GENWQE5 << 8),
110 .class_mask = ~0,
111 .driver_data = 0 },
113 { 0, } /* 0 terminated list. */
116 MODULE_DEVICE_TABLE(pci, genwqe_device_table);
119 * genwqe_dev_alloc() - Create and prepare a new card descriptor
121 * Return: Pointer to card descriptor, or ERR_PTR(err) on error
123 static struct genwqe_dev *genwqe_dev_alloc(void)
125 unsigned int i = 0, j;
126 struct genwqe_dev *cd;
128 for (i = 0; i < GENWQE_CARD_NO_MAX; i++) {
129 if (genwqe_devices[i] == NULL)
130 break;
132 if (i >= GENWQE_CARD_NO_MAX)
133 return ERR_PTR(-ENODEV);
135 cd = kzalloc(sizeof(struct genwqe_dev), GFP_KERNEL);
136 if (!cd)
137 return ERR_PTR(-ENOMEM);
139 cd->card_idx = i;
140 cd->class_genwqe = class_genwqe;
141 cd->debugfs_genwqe = debugfs_genwqe;
143 init_waitqueue_head(&cd->queue_waitq);
145 spin_lock_init(&cd->file_lock);
146 INIT_LIST_HEAD(&cd->file_list);
148 cd->card_state = GENWQE_CARD_UNUSED;
149 spin_lock_init(&cd->print_lock);
151 cd->ddcb_software_timeout = genwqe_ddcb_software_timeout;
152 cd->kill_timeout = genwqe_kill_timeout;
154 for (j = 0; j < GENWQE_MAX_VFS; j++)
155 cd->vf_jobtimeout_msec[j] = genwqe_vf_jobtimeout_msec;
157 genwqe_devices[i] = cd;
158 return cd;
161 static void genwqe_dev_free(struct genwqe_dev *cd)
163 if (!cd)
164 return;
166 genwqe_devices[cd->card_idx] = NULL;
167 kfree(cd);
171 * genwqe_bus_reset() - Card recovery
173 * pci_reset_function() will recover the device and ensure that the
174 * registers are accessible again when it completes with success. If
175 * not, the card will stay dead and registers will be unaccessible
176 * still.
178 static int genwqe_bus_reset(struct genwqe_dev *cd)
180 int bars, rc = 0;
181 struct pci_dev *pci_dev = cd->pci_dev;
182 void __iomem *mmio;
184 if (cd->err_inject & GENWQE_INJECT_BUS_RESET_FAILURE)
185 return -EIO;
187 mmio = cd->mmio;
188 cd->mmio = NULL;
189 pci_iounmap(pci_dev, mmio);
191 bars = pci_select_bars(pci_dev, IORESOURCE_MEM);
192 pci_release_selected_regions(pci_dev, bars);
195 * Firmware/BIOS might change memory mapping during bus reset.
196 * Settings like enable bus-mastering, ... are backuped and
197 * restored by the pci_reset_function().
199 dev_dbg(&pci_dev->dev, "[%s] pci_reset function ...\n", __func__);
200 rc = pci_reset_function(pci_dev);
201 if (rc) {
202 dev_err(&pci_dev->dev,
203 "[%s] err: failed reset func (rc %d)\n", __func__, rc);
204 return rc;
206 dev_dbg(&pci_dev->dev, "[%s] done with rc=%d\n", __func__, rc);
209 * Here is the right spot to clear the register read
210 * failure. pci_bus_reset() does this job in real systems.
212 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
213 GENWQE_INJECT_GFIR_FATAL |
214 GENWQE_INJECT_GFIR_INFO);
216 rc = pci_request_selected_regions(pci_dev, bars, genwqe_driver_name);
217 if (rc) {
218 dev_err(&pci_dev->dev,
219 "[%s] err: request bars failed (%d)\n", __func__, rc);
220 return -EIO;
223 cd->mmio = pci_iomap(pci_dev, 0, 0);
224 if (cd->mmio == NULL) {
225 dev_err(&pci_dev->dev,
226 "[%s] err: mapping BAR0 failed\n", __func__);
227 return -ENOMEM;
229 return 0;
233 * Hardware circumvention section. Certain bitstreams in our test-lab
234 * had different kinds of problems. Here is where we adjust those
235 * bitstreams to function will with this version of our device driver.
237 * Thise circumventions are applied to the physical function only.
238 * The magical numbers below are identifying development/manufacturing
239 * versions of the bitstream used on the card.
241 * Turn off error reporting for old/manufacturing images.
244 bool genwqe_need_err_masking(struct genwqe_dev *cd)
246 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
249 static void genwqe_tweak_hardware(struct genwqe_dev *cd)
251 struct pci_dev *pci_dev = cd->pci_dev;
253 /* Mask FIRs for development images */
254 if (((cd->slu_unitcfg & 0xFFFF0ull) >= 0x32000ull) &&
255 ((cd->slu_unitcfg & 0xFFFF0ull) <= 0x33250ull)) {
256 dev_warn(&pci_dev->dev,
257 "FIRs masked due to bitstream %016llx.%016llx\n",
258 cd->slu_unitcfg, cd->app_unitcfg);
260 __genwqe_writeq(cd, IO_APP_SEC_LEM_DEBUG_OVR,
261 0xFFFFFFFFFFFFFFFFull);
263 __genwqe_writeq(cd, IO_APP_ERR_ACT_MASK,
264 0x0000000000000000ull);
269 * genwqe_recovery_on_fatal_gfir_required() - Version depended actions
271 * Bitstreams older than 2013-02-17 have a bug where fatal GFIRs must
272 * be ignored. This is e.g. true for the bitstream we gave to the card
273 * manufacturer, but also for some old bitstreams we released to our
274 * test-lab.
276 int genwqe_recovery_on_fatal_gfir_required(struct genwqe_dev *cd)
278 return (cd->slu_unitcfg & 0xFFFF0ull) >= 0x32170ull;
281 int genwqe_flash_readback_fails(struct genwqe_dev *cd)
283 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
287 * genwqe_T_psec() - Calculate PF/VF timeout register content
289 * Note: From a design perspective it turned out to be a bad idea to
290 * use codes here to specifiy the frequency/speed values. An old
291 * driver cannot understand new codes and is therefore always a
292 * problem. Better is to measure out the value or put the
293 * speed/frequency directly into a register which is always a valid
294 * value for old as well as for new software.
296 /* T = 1/f */
297 static int genwqe_T_psec(struct genwqe_dev *cd)
299 u16 speed; /* 1/f -> 250, 200, 166, 175 */
300 static const int T[] = { 4000, 5000, 6000, 5714 };
302 speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
303 if (speed >= ARRAY_SIZE(T))
304 return -1; /* illegal value */
306 return T[speed];
310 * genwqe_setup_pf_jtimer() - Setup PF hardware timeouts for DDCB execution
312 * Do this _after_ card_reset() is called. Otherwise the values will
313 * vanish. The settings need to be done when the queues are inactive.
315 * The max. timeout value is 2^(10+x) * T (6ns for 166MHz) * 15/16.
316 * The min. timeout value is 2^(10+x) * T (6ns for 166MHz) * 14/16.
318 static bool genwqe_setup_pf_jtimer(struct genwqe_dev *cd)
320 u32 T = genwqe_T_psec(cd);
321 u64 x;
323 if (genwqe_pf_jobtimeout_msec == 0)
324 return false;
326 /* PF: large value needed, flash update 2sec per block */
327 x = ilog2(genwqe_pf_jobtimeout_msec *
328 16000000000uL/(T * 15)) - 10;
330 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
331 0xff00 | (x & 0xff), 0);
332 return true;
336 * genwqe_setup_vf_jtimer() - Setup VF hardware timeouts for DDCB execution
338 static bool genwqe_setup_vf_jtimer(struct genwqe_dev *cd)
340 struct pci_dev *pci_dev = cd->pci_dev;
341 unsigned int vf;
342 u32 T = genwqe_T_psec(cd);
343 u64 x;
345 for (vf = 0; vf < pci_sriov_get_totalvfs(pci_dev); vf++) {
347 if (cd->vf_jobtimeout_msec[vf] == 0)
348 continue;
350 x = ilog2(cd->vf_jobtimeout_msec[vf] *
351 16000000000uL/(T * 15)) - 10;
353 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
354 0xff00 | (x & 0xff), vf + 1);
356 return true;
359 static int genwqe_ffdc_buffs_alloc(struct genwqe_dev *cd)
361 unsigned int type, e = 0;
363 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
364 switch (type) {
365 case GENWQE_DBG_UNIT0:
366 e = genwqe_ffdc_buff_size(cd, 0);
367 break;
368 case GENWQE_DBG_UNIT1:
369 e = genwqe_ffdc_buff_size(cd, 1);
370 break;
371 case GENWQE_DBG_UNIT2:
372 e = genwqe_ffdc_buff_size(cd, 2);
373 break;
374 case GENWQE_DBG_REGS:
375 e = GENWQE_FFDC_REGS;
376 break;
379 /* currently support only the debug units mentioned here */
380 cd->ffdc[type].entries = e;
381 cd->ffdc[type].regs = kmalloc(e * sizeof(struct genwqe_reg),
382 GFP_KERNEL);
384 * regs == NULL is ok, the using code treats this as no regs,
385 * Printing warning is ok in this case.
388 return 0;
391 static void genwqe_ffdc_buffs_free(struct genwqe_dev *cd)
393 unsigned int type;
395 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
396 kfree(cd->ffdc[type].regs);
397 cd->ffdc[type].regs = NULL;
401 static int genwqe_read_ids(struct genwqe_dev *cd)
403 int err = 0;
404 int slu_id;
405 struct pci_dev *pci_dev = cd->pci_dev;
407 cd->slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
408 if (cd->slu_unitcfg == IO_ILLEGAL_VALUE) {
409 dev_err(&pci_dev->dev,
410 "err: SLUID=%016llx\n", cd->slu_unitcfg);
411 err = -EIO;
412 goto out_err;
415 slu_id = genwqe_get_slu_id(cd);
416 if (slu_id < GENWQE_SLU_ARCH_REQ || slu_id == 0xff) {
417 dev_err(&pci_dev->dev,
418 "err: incompatible SLU Architecture %u\n", slu_id);
419 err = -ENOENT;
420 goto out_err;
423 cd->app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
424 if (cd->app_unitcfg == IO_ILLEGAL_VALUE) {
425 dev_err(&pci_dev->dev,
426 "err: APPID=%016llx\n", cd->app_unitcfg);
427 err = -EIO;
428 goto out_err;
430 genwqe_read_app_id(cd, cd->app_name, sizeof(cd->app_name));
433 * Is access to all registers possible? If we are a VF the
434 * answer is obvious. If we run fully virtualized, we need to
435 * check if we can access all registers. If we do not have
436 * full access we will cause an UR and some informational FIRs
437 * in the PF, but that should not harm.
439 if (pci_dev->is_virtfn)
440 cd->is_privileged = 0;
441 else
442 cd->is_privileged = (__genwqe_readq(cd, IO_SLU_BITSTREAM)
443 != IO_ILLEGAL_VALUE);
445 out_err:
446 return err;
449 static int genwqe_start(struct genwqe_dev *cd)
451 int err;
452 struct pci_dev *pci_dev = cd->pci_dev;
454 err = genwqe_read_ids(cd);
455 if (err)
456 return err;
458 if (genwqe_is_privileged(cd)) {
459 /* do this after the tweaks. alloc fail is acceptable */
460 genwqe_ffdc_buffs_alloc(cd);
461 genwqe_stop_traps(cd);
463 /* Collect registers e.g. FIRs, UNITIDs, traces ... */
464 genwqe_read_ffdc_regs(cd, cd->ffdc[GENWQE_DBG_REGS].regs,
465 cd->ffdc[GENWQE_DBG_REGS].entries, 0);
467 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT0,
468 cd->ffdc[GENWQE_DBG_UNIT0].regs,
469 cd->ffdc[GENWQE_DBG_UNIT0].entries);
471 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT1,
472 cd->ffdc[GENWQE_DBG_UNIT1].regs,
473 cd->ffdc[GENWQE_DBG_UNIT1].entries);
475 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT2,
476 cd->ffdc[GENWQE_DBG_UNIT2].regs,
477 cd->ffdc[GENWQE_DBG_UNIT2].entries);
479 genwqe_start_traps(cd);
481 if (cd->card_state == GENWQE_CARD_FATAL_ERROR) {
482 dev_warn(&pci_dev->dev,
483 "[%s] chip reload/recovery!\n", __func__);
486 * Stealth Mode: Reload chip on either hot
487 * reset or PERST.
489 cd->softreset = 0x7Cull;
490 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
491 cd->softreset);
493 err = genwqe_bus_reset(cd);
494 if (err != 0) {
495 dev_err(&pci_dev->dev,
496 "[%s] err: bus reset failed!\n",
497 __func__);
498 goto out;
502 * Re-read the IDs because
503 * it could happen that the bitstream load
504 * failed!
506 err = genwqe_read_ids(cd);
507 if (err)
508 goto out;
512 err = genwqe_setup_service_layer(cd); /* does a reset to the card */
513 if (err != 0) {
514 dev_err(&pci_dev->dev,
515 "[%s] err: could not setup servicelayer!\n", __func__);
516 err = -ENODEV;
517 goto out;
520 if (genwqe_is_privileged(cd)) { /* code is running _after_ reset */
521 genwqe_tweak_hardware(cd);
523 genwqe_setup_pf_jtimer(cd);
524 genwqe_setup_vf_jtimer(cd);
527 err = genwqe_device_create(cd);
528 if (err < 0) {
529 dev_err(&pci_dev->dev,
530 "err: chdev init failed! (err=%d)\n", err);
531 goto out_release_service_layer;
533 return 0;
535 out_release_service_layer:
536 genwqe_release_service_layer(cd);
537 out:
538 if (genwqe_is_privileged(cd))
539 genwqe_ffdc_buffs_free(cd);
540 return -EIO;
544 * genwqe_stop() - Stop card operation
546 * Recovery notes:
547 * As long as genwqe_thread runs we might access registers during
548 * error data capture. Same is with the genwqe_health_thread.
549 * When genwqe_bus_reset() fails this function might called two times:
550 * first by the genwqe_health_thread() and later by genwqe_remove() to
551 * unbind the device. We must be able to survive that.
553 * This function must be robust enough to be called twice.
555 static int genwqe_stop(struct genwqe_dev *cd)
557 genwqe_finish_queue(cd); /* no register access */
558 genwqe_device_remove(cd); /* device removed, procs killed */
559 genwqe_release_service_layer(cd); /* here genwqe_thread is stopped */
561 if (genwqe_is_privileged(cd)) {
562 pci_disable_sriov(cd->pci_dev); /* access pci config space */
563 genwqe_ffdc_buffs_free(cd);
566 return 0;
570 * genwqe_recover_card() - Try to recover the card if it is possible
572 * If fatal_err is set no register access is possible anymore. It is
573 * likely that genwqe_start fails in that situation. Proper error
574 * handling is required in this case.
576 * genwqe_bus_reset() will cause the pci code to call genwqe_remove()
577 * and later genwqe_probe() for all virtual functions.
579 static int genwqe_recover_card(struct genwqe_dev *cd, int fatal_err)
581 int rc;
582 struct pci_dev *pci_dev = cd->pci_dev;
584 genwqe_stop(cd);
587 * Make sure chip is not reloaded to maintain FFDC. Write SLU
588 * Reset Register, CPLDReset field to 0.
590 if (!fatal_err) {
591 cd->softreset = 0x70ull;
592 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, cd->softreset);
595 rc = genwqe_bus_reset(cd);
596 if (rc != 0) {
597 dev_err(&pci_dev->dev,
598 "[%s] err: card recovery impossible!\n", __func__);
599 return rc;
602 rc = genwqe_start(cd);
603 if (rc < 0) {
604 dev_err(&pci_dev->dev,
605 "[%s] err: failed to launch device!\n", __func__);
606 return rc;
608 return 0;
611 static int genwqe_health_check_cond(struct genwqe_dev *cd, u64 *gfir)
613 *gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
614 return (*gfir & GFIR_ERR_TRIGGER) &&
615 genwqe_recovery_on_fatal_gfir_required(cd);
619 * genwqe_fir_checking() - Check the fault isolation registers of the card
621 * If this code works ok, can be tried out with help of the genwqe_poke tool:
622 * sudo ./tools/genwqe_poke 0x8 0xfefefefefef
624 * Now the relevant FIRs/sFIRs should be printed out and the driver should
625 * invoke recovery (devices are removed and readded).
627 static u64 genwqe_fir_checking(struct genwqe_dev *cd)
629 int j, iterations = 0;
630 u64 mask, fir, fec, uid, gfir, gfir_masked, sfir, sfec;
631 u32 fir_addr, fir_clr_addr, fec_addr, sfir_addr, sfec_addr;
632 struct pci_dev *pci_dev = cd->pci_dev;
634 healthMonitor:
635 iterations++;
636 if (iterations > 16) {
637 dev_err(&pci_dev->dev, "* exit looping after %d times\n",
638 iterations);
639 goto fatal_error;
642 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
643 if (gfir != 0x0)
644 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n",
645 IO_SLC_CFGREG_GFIR, gfir);
646 if (gfir == IO_ILLEGAL_VALUE)
647 goto fatal_error;
650 * Avoid printing when to GFIR bit is on prevents contignous
651 * printout e.g. for the following bug:
652 * FIR set without a 2ndary FIR/FIR cannot be cleared
653 * Comment out the following if to get the prints:
655 if (gfir == 0)
656 return 0;
658 gfir_masked = gfir & GFIR_ERR_TRIGGER; /* fatal errors */
660 for (uid = 0; uid < GENWQE_MAX_UNITS; uid++) { /* 0..2 in zEDC */
662 /* read the primary FIR (pfir) */
663 fir_addr = (uid << 24) + 0x08;
664 fir = __genwqe_readq(cd, fir_addr);
665 if (fir == 0x0)
666 continue; /* no error in this unit */
668 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fir_addr, fir);
669 if (fir == IO_ILLEGAL_VALUE)
670 goto fatal_error;
672 /* read primary FEC */
673 fec_addr = (uid << 24) + 0x18;
674 fec = __genwqe_readq(cd, fec_addr);
676 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fec_addr, fec);
677 if (fec == IO_ILLEGAL_VALUE)
678 goto fatal_error;
680 for (j = 0, mask = 1ULL; j < 64; j++, mask <<= 1) {
682 /* secondary fir empty, skip it */
683 if ((fir & mask) == 0x0)
684 continue;
686 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
687 sfir = __genwqe_readq(cd, sfir_addr);
689 if (sfir == IO_ILLEGAL_VALUE)
690 goto fatal_error;
691 dev_err(&pci_dev->dev,
692 "* 0x%08x 0x%016llx\n", sfir_addr, sfir);
694 sfec_addr = (uid << 24) + 0x300 + 0x08 * j;
695 sfec = __genwqe_readq(cd, sfec_addr);
697 if (sfec == IO_ILLEGAL_VALUE)
698 goto fatal_error;
699 dev_err(&pci_dev->dev,
700 "* 0x%08x 0x%016llx\n", sfec_addr, sfec);
702 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
703 if (gfir == IO_ILLEGAL_VALUE)
704 goto fatal_error;
706 /* gfir turned on during routine! get out and
707 start over. */
708 if ((gfir_masked == 0x0) &&
709 (gfir & GFIR_ERR_TRIGGER)) {
710 goto healthMonitor;
713 /* do not clear if we entered with a fatal gfir */
714 if (gfir_masked == 0x0) {
716 /* NEW clear by mask the logged bits */
717 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
718 __genwqe_writeq(cd, sfir_addr, sfir);
720 dev_dbg(&pci_dev->dev,
721 "[HM] Clearing 2ndary FIR 0x%08x "
722 "with 0x%016llx\n", sfir_addr, sfir);
725 * note, these cannot be error-Firs
726 * since gfir_masked is 0 after sfir
727 * was read. Also, it is safe to do
728 * this write if sfir=0. Still need to
729 * clear the primary. This just means
730 * there is no secondary FIR.
733 /* clear by mask the logged bit. */
734 fir_clr_addr = (uid << 24) + 0x10;
735 __genwqe_writeq(cd, fir_clr_addr, mask);
737 dev_dbg(&pci_dev->dev,
738 "[HM] Clearing primary FIR 0x%08x "
739 "with 0x%016llx\n", fir_clr_addr,
740 mask);
744 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
745 if (gfir == IO_ILLEGAL_VALUE)
746 goto fatal_error;
748 if ((gfir_masked == 0x0) && (gfir & GFIR_ERR_TRIGGER)) {
750 * Check once more that it didn't go on after all the
751 * FIRS were cleared.
753 dev_dbg(&pci_dev->dev, "ACK! Another FIR! Recursing %d!\n",
754 iterations);
755 goto healthMonitor;
757 return gfir_masked;
759 fatal_error:
760 return IO_ILLEGAL_VALUE;
764 * genwqe_health_thread() - Health checking thread
766 * This thread is only started for the PF of the card.
768 * This thread monitors the health of the card. A critical situation
769 * is when we read registers which contain -1 (IO_ILLEGAL_VALUE). In
770 * this case we need to be recovered from outside. Writing to
771 * registers will very likely not work either.
773 * This thread must only exit if kthread_should_stop() becomes true.
775 * Condition for the health-thread to trigger:
776 * a) when a kthread_stop() request comes in or
777 * b) a critical GFIR occured
779 * Informational GFIRs are checked and potentially printed in
780 * health_check_interval seconds.
782 static int genwqe_health_thread(void *data)
784 int rc, should_stop = 0;
785 struct genwqe_dev *cd = data;
786 struct pci_dev *pci_dev = cd->pci_dev;
787 u64 gfir, gfir_masked, slu_unitcfg, app_unitcfg;
789 while (!kthread_should_stop()) {
790 rc = wait_event_interruptible_timeout(cd->health_waitq,
791 (genwqe_health_check_cond(cd, &gfir) ||
792 (should_stop = kthread_should_stop())),
793 genwqe_health_check_interval * HZ);
795 if (should_stop)
796 break;
798 if (gfir == IO_ILLEGAL_VALUE) {
799 dev_err(&pci_dev->dev,
800 "[%s] GFIR=%016llx\n", __func__, gfir);
801 goto fatal_error;
804 slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
805 if (slu_unitcfg == IO_ILLEGAL_VALUE) {
806 dev_err(&pci_dev->dev,
807 "[%s] SLU_UNITCFG=%016llx\n",
808 __func__, slu_unitcfg);
809 goto fatal_error;
812 app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
813 if (app_unitcfg == IO_ILLEGAL_VALUE) {
814 dev_err(&pci_dev->dev,
815 "[%s] APP_UNITCFG=%016llx\n",
816 __func__, app_unitcfg);
817 goto fatal_error;
820 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
821 if (gfir == IO_ILLEGAL_VALUE) {
822 dev_err(&pci_dev->dev,
823 "[%s] %s: GFIR=%016llx\n", __func__,
824 (gfir & GFIR_ERR_TRIGGER) ? "err" : "info",
825 gfir);
826 goto fatal_error;
829 gfir_masked = genwqe_fir_checking(cd);
830 if (gfir_masked == IO_ILLEGAL_VALUE)
831 goto fatal_error;
834 * GFIR ErrorTrigger bits set => reset the card!
835 * Never do this for old/manufacturing images!
837 if ((gfir_masked) && !cd->skip_recovery &&
838 genwqe_recovery_on_fatal_gfir_required(cd)) {
840 cd->card_state = GENWQE_CARD_FATAL_ERROR;
842 rc = genwqe_recover_card(cd, 0);
843 if (rc < 0) {
844 /* FIXME Card is unusable and needs unbind! */
845 goto fatal_error;
849 cd->last_gfir = gfir;
850 cond_resched();
853 return 0;
855 fatal_error:
856 dev_err(&pci_dev->dev,
857 "[%s] card unusable. Please trigger unbind!\n", __func__);
859 /* Bring down logical devices to inform user space via udev remove. */
860 cd->card_state = GENWQE_CARD_FATAL_ERROR;
861 genwqe_stop(cd);
863 /* genwqe_bus_reset failed(). Now wait for genwqe_remove(). */
864 while (!kthread_should_stop())
865 cond_resched();
867 return -EIO;
870 static int genwqe_health_check_start(struct genwqe_dev *cd)
872 int rc;
874 if (genwqe_health_check_interval <= 0)
875 return 0; /* valid for disabling the service */
877 /* moved before request_irq() */
878 /* init_waitqueue_head(&cd->health_waitq); */
880 cd->health_thread = kthread_run(genwqe_health_thread, cd,
881 GENWQE_DEVNAME "%d_health",
882 cd->card_idx);
883 if (IS_ERR(cd->health_thread)) {
884 rc = PTR_ERR(cd->health_thread);
885 cd->health_thread = NULL;
886 return rc;
888 return 0;
891 static int genwqe_health_thread_running(struct genwqe_dev *cd)
893 return cd->health_thread != NULL;
896 static int genwqe_health_check_stop(struct genwqe_dev *cd)
898 int rc;
900 if (!genwqe_health_thread_running(cd))
901 return -EIO;
903 rc = kthread_stop(cd->health_thread);
904 cd->health_thread = NULL;
905 return 0;
909 * genwqe_pci_setup() - Allocate PCIe related resources for our card
911 static int genwqe_pci_setup(struct genwqe_dev *cd)
913 int err, bars;
914 struct pci_dev *pci_dev = cd->pci_dev;
916 bars = pci_select_bars(pci_dev, IORESOURCE_MEM);
917 err = pci_enable_device_mem(pci_dev);
918 if (err) {
919 dev_err(&pci_dev->dev,
920 "err: failed to enable pci memory (err=%d)\n", err);
921 goto err_out;
924 /* Reserve PCI I/O and memory resources */
925 err = pci_request_selected_regions(pci_dev, bars, genwqe_driver_name);
926 if (err) {
927 dev_err(&pci_dev->dev,
928 "[%s] err: request bars failed (%d)\n", __func__, err);
929 err = -EIO;
930 goto err_disable_device;
933 /* check for 64-bit DMA address supported (DAC) */
934 if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(64))) {
935 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(64));
936 if (err) {
937 dev_err(&pci_dev->dev,
938 "err: DMA64 consistent mask error\n");
939 err = -EIO;
940 goto out_release_resources;
942 /* check for 32-bit DMA address supported (SAC) */
943 } else if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(32))) {
944 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(32));
945 if (err) {
946 dev_err(&pci_dev->dev,
947 "err: DMA32 consistent mask error\n");
948 err = -EIO;
949 goto out_release_resources;
951 } else {
952 dev_err(&pci_dev->dev,
953 "err: neither DMA32 nor DMA64 supported\n");
954 err = -EIO;
955 goto out_release_resources;
958 pci_set_master(pci_dev);
959 pci_enable_pcie_error_reporting(pci_dev);
961 /* request complete BAR-0 space (length = 0) */
962 cd->mmio_len = pci_resource_len(pci_dev, 0);
963 cd->mmio = pci_iomap(pci_dev, 0, 0);
964 if (cd->mmio == NULL) {
965 dev_err(&pci_dev->dev,
966 "[%s] err: mapping BAR0 failed\n", __func__);
967 err = -ENOMEM;
968 goto out_release_resources;
971 cd->num_vfs = pci_sriov_get_totalvfs(pci_dev);
973 err = genwqe_read_ids(cd);
974 if (err)
975 goto out_iounmap;
977 return 0;
979 out_iounmap:
980 pci_iounmap(pci_dev, cd->mmio);
981 out_release_resources:
982 pci_release_selected_regions(pci_dev, bars);
983 err_disable_device:
984 pci_disable_device(pci_dev);
985 err_out:
986 return err;
990 * genwqe_pci_remove() - Free PCIe related resources for our card
992 static void genwqe_pci_remove(struct genwqe_dev *cd)
994 int bars;
995 struct pci_dev *pci_dev = cd->pci_dev;
997 if (cd->mmio)
998 pci_iounmap(pci_dev, cd->mmio);
1000 bars = pci_select_bars(pci_dev, IORESOURCE_MEM);
1001 pci_release_selected_regions(pci_dev, bars);
1002 pci_disable_device(pci_dev);
1006 * genwqe_probe() - Device initialization
1007 * @pdev: PCI device information struct
1009 * Callable for multiple cards. This function is called on bind.
1011 * Return: 0 if succeeded, < 0 when failed
1013 static int genwqe_probe(struct pci_dev *pci_dev,
1014 const struct pci_device_id *id)
1016 int err;
1017 struct genwqe_dev *cd;
1019 genwqe_init_crc32();
1021 cd = genwqe_dev_alloc();
1022 if (IS_ERR(cd)) {
1023 dev_err(&pci_dev->dev, "err: could not alloc mem (err=%d)!\n",
1024 (int)PTR_ERR(cd));
1025 return PTR_ERR(cd);
1028 dev_set_drvdata(&pci_dev->dev, cd);
1029 cd->pci_dev = pci_dev;
1031 err = genwqe_pci_setup(cd);
1032 if (err < 0) {
1033 dev_err(&pci_dev->dev,
1034 "err: problems with PCI setup (err=%d)\n", err);
1035 goto out_free_dev;
1038 err = genwqe_start(cd);
1039 if (err < 0) {
1040 dev_err(&pci_dev->dev,
1041 "err: cannot start card services! (err=%d)\n", err);
1042 goto out_pci_remove;
1045 if (genwqe_is_privileged(cd)) {
1046 err = genwqe_health_check_start(cd);
1047 if (err < 0) {
1048 dev_err(&pci_dev->dev,
1049 "err: cannot start health checking! "
1050 "(err=%d)\n", err);
1051 goto out_stop_services;
1054 return 0;
1056 out_stop_services:
1057 genwqe_stop(cd);
1058 out_pci_remove:
1059 genwqe_pci_remove(cd);
1060 out_free_dev:
1061 genwqe_dev_free(cd);
1062 return err;
1066 * genwqe_remove() - Called when device is removed (hot-plugable)
1068 * Or when driver is unloaded respecitively when unbind is done.
1070 static void genwqe_remove(struct pci_dev *pci_dev)
1072 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1074 genwqe_health_check_stop(cd);
1077 * genwqe_stop() must survive if it is called twice
1078 * sequentially. This happens when the health thread calls it
1079 * and fails on genwqe_bus_reset().
1081 genwqe_stop(cd);
1082 genwqe_pci_remove(cd);
1083 genwqe_dev_free(cd);
1087 * genwqe_err_error_detected() - Error detection callback
1089 * This callback is called by the PCI subsystem whenever a PCI bus
1090 * error is detected.
1092 static pci_ers_result_t genwqe_err_error_detected(struct pci_dev *pci_dev,
1093 enum pci_channel_state state)
1095 struct genwqe_dev *cd;
1097 dev_err(&pci_dev->dev, "[%s] state=%d\n", __func__, state);
1099 if (pci_dev == NULL)
1100 return PCI_ERS_RESULT_NEED_RESET;
1102 cd = dev_get_drvdata(&pci_dev->dev);
1103 if (cd == NULL)
1104 return PCI_ERS_RESULT_NEED_RESET;
1106 switch (state) {
1107 case pci_channel_io_normal:
1108 return PCI_ERS_RESULT_CAN_RECOVER;
1109 case pci_channel_io_frozen:
1110 return PCI_ERS_RESULT_NEED_RESET;
1111 case pci_channel_io_perm_failure:
1112 return PCI_ERS_RESULT_DISCONNECT;
1115 return PCI_ERS_RESULT_NEED_RESET;
1118 static pci_ers_result_t genwqe_err_result_none(struct pci_dev *dev)
1120 return PCI_ERS_RESULT_NONE;
1123 static void genwqe_err_resume(struct pci_dev *dev)
1127 static int genwqe_sriov_configure(struct pci_dev *dev, int numvfs)
1129 struct genwqe_dev *cd = dev_get_drvdata(&dev->dev);
1131 if (numvfs > 0) {
1132 genwqe_setup_vf_jtimer(cd);
1133 pci_enable_sriov(dev, numvfs);
1134 return numvfs;
1136 if (numvfs == 0) {
1137 pci_disable_sriov(dev);
1138 return 0;
1140 return 0;
1143 static struct pci_error_handlers genwqe_err_handler = {
1144 .error_detected = genwqe_err_error_detected,
1145 .mmio_enabled = genwqe_err_result_none,
1146 .link_reset = genwqe_err_result_none,
1147 .slot_reset = genwqe_err_result_none,
1148 .resume = genwqe_err_resume,
1151 static struct pci_driver genwqe_driver = {
1152 .name = genwqe_driver_name,
1153 .id_table = genwqe_device_table,
1154 .probe = genwqe_probe,
1155 .remove = genwqe_remove,
1156 .sriov_configure = genwqe_sriov_configure,
1157 .err_handler = &genwqe_err_handler,
1161 * genwqe_init_module() - Driver registration and initialization
1163 static int __init genwqe_init_module(void)
1165 int rc;
1167 class_genwqe = class_create(THIS_MODULE, GENWQE_DEVNAME);
1168 if (IS_ERR(class_genwqe)) {
1169 pr_err("[%s] create class failed\n", __func__);
1170 return -ENOMEM;
1173 debugfs_genwqe = debugfs_create_dir(GENWQE_DEVNAME, NULL);
1174 if (!debugfs_genwqe) {
1175 rc = -ENOMEM;
1176 goto err_out;
1179 rc = pci_register_driver(&genwqe_driver);
1180 if (rc != 0) {
1181 pr_err("[%s] pci_reg_driver (rc=%d)\n", __func__, rc);
1182 goto err_out0;
1185 return rc;
1187 err_out0:
1188 debugfs_remove(debugfs_genwqe);
1189 err_out:
1190 class_destroy(class_genwqe);
1191 return rc;
1195 * genwqe_exit_module() - Driver exit
1197 static void __exit genwqe_exit_module(void)
1199 pci_unregister_driver(&genwqe_driver);
1200 debugfs_remove(debugfs_genwqe);
1201 class_destroy(class_genwqe);
1204 module_init(genwqe_init_module);
1205 module_exit(genwqe_exit_module);