treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / misc / genwqe / card_base.c
blob1dc6c7c5cbce929d40118cf1a0fb7213963744f6
1 // SPDX-License-Identifier: GPL-2.0-only
2 /**
3 * IBM Accelerator Family 'GenWQE'
5 * (C) Copyright IBM Corp. 2013
7 * Author: Frank Haverkamp <haver@linux.vnet.ibm.com>
8 * Author: Joerg-Stephan Vogt <jsvogt@de.ibm.com>
9 * Author: Michael Jung <mijung@gmx.net>
10 * Author: Michael Ruettger <michael@ibmra.de>
14 * Module initialization and PCIe setup. Card health monitoring and
15 * recovery functionality. Character device creation and deletion are
16 * controlled from here.
19 #include <linux/types.h>
20 #include <linux/pci.h>
21 #include <linux/err.h>
22 #include <linux/aer.h>
23 #include <linux/string.h>
24 #include <linux/sched.h>
25 #include <linux/wait.h>
26 #include <linux/delay.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/module.h>
29 #include <linux/notifier.h>
30 #include <linux/device.h>
31 #include <linux/log2.h>
33 #include "card_base.h"
34 #include "card_ddcb.h"
36 MODULE_AUTHOR("Frank Haverkamp <haver@linux.vnet.ibm.com>");
37 MODULE_AUTHOR("Michael Ruettger <michael@ibmra.de>");
38 MODULE_AUTHOR("Joerg-Stephan Vogt <jsvogt@de.ibm.com>");
39 MODULE_AUTHOR("Michael Jung <mijung@gmx.net>");
41 MODULE_DESCRIPTION("GenWQE Card");
42 MODULE_VERSION(DRV_VERSION);
43 MODULE_LICENSE("GPL");
45 static char genwqe_driver_name[] = GENWQE_DEVNAME;
46 static struct class *class_genwqe;
47 static struct dentry *debugfs_genwqe;
48 static struct genwqe_dev *genwqe_devices[GENWQE_CARD_NO_MAX];
50 /* PCI structure for identifying device by PCI vendor and device ID */
51 static const struct pci_device_id genwqe_device_table[] = {
52 { .vendor = PCI_VENDOR_ID_IBM,
53 .device = PCI_DEVICE_GENWQE,
54 .subvendor = PCI_SUBVENDOR_ID_IBM,
55 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
56 .class = (PCI_CLASSCODE_GENWQE5 << 8),
57 .class_mask = ~0,
58 .driver_data = 0 },
60 /* Initial SR-IOV bring-up image */
61 { .vendor = PCI_VENDOR_ID_IBM,
62 .device = PCI_DEVICE_GENWQE,
63 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
64 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
65 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
66 .class_mask = ~0,
67 .driver_data = 0 },
69 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
70 .device = 0x0000, /* VF Device ID */
71 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
72 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
73 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
74 .class_mask = ~0,
75 .driver_data = 0 },
77 /* Fixed up image */
78 { .vendor = PCI_VENDOR_ID_IBM,
79 .device = PCI_DEVICE_GENWQE,
80 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
81 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
82 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
83 .class_mask = ~0,
84 .driver_data = 0 },
86 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
87 .device = 0x0000, /* VF Device ID */
88 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
89 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
90 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
91 .class_mask = ~0,
92 .driver_data = 0 },
94 /* Even one more ... */
95 { .vendor = PCI_VENDOR_ID_IBM,
96 .device = PCI_DEVICE_GENWQE,
97 .subvendor = PCI_SUBVENDOR_ID_IBM,
98 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_NEW,
99 .class = (PCI_CLASSCODE_GENWQE5 << 8),
100 .class_mask = ~0,
101 .driver_data = 0 },
103 { 0, } /* 0 terminated list. */
106 MODULE_DEVICE_TABLE(pci, genwqe_device_table);
109 * genwqe_dev_alloc() - Create and prepare a new card descriptor
111 * Return: Pointer to card descriptor, or ERR_PTR(err) on error
113 static struct genwqe_dev *genwqe_dev_alloc(void)
115 unsigned int i = 0, j;
116 struct genwqe_dev *cd;
118 for (i = 0; i < GENWQE_CARD_NO_MAX; i++) {
119 if (genwqe_devices[i] == NULL)
120 break;
122 if (i >= GENWQE_CARD_NO_MAX)
123 return ERR_PTR(-ENODEV);
125 cd = kzalloc(sizeof(struct genwqe_dev), GFP_KERNEL);
126 if (!cd)
127 return ERR_PTR(-ENOMEM);
129 cd->card_idx = i;
130 cd->class_genwqe = class_genwqe;
131 cd->debugfs_genwqe = debugfs_genwqe;
134 * This comes from kernel config option and can be overritten via
135 * debugfs.
137 cd->use_platform_recovery = CONFIG_GENWQE_PLATFORM_ERROR_RECOVERY;
139 init_waitqueue_head(&cd->queue_waitq);
141 spin_lock_init(&cd->file_lock);
142 INIT_LIST_HEAD(&cd->file_list);
144 cd->card_state = GENWQE_CARD_UNUSED;
145 spin_lock_init(&cd->print_lock);
147 cd->ddcb_software_timeout = GENWQE_DDCB_SOFTWARE_TIMEOUT;
148 cd->kill_timeout = GENWQE_KILL_TIMEOUT;
150 for (j = 0; j < GENWQE_MAX_VFS; j++)
151 cd->vf_jobtimeout_msec[j] = GENWQE_VF_JOBTIMEOUT_MSEC;
153 genwqe_devices[i] = cd;
154 return cd;
157 static void genwqe_dev_free(struct genwqe_dev *cd)
159 if (!cd)
160 return;
162 genwqe_devices[cd->card_idx] = NULL;
163 kfree(cd);
167 * genwqe_bus_reset() - Card recovery
169 * pci_reset_function() will recover the device and ensure that the
170 * registers are accessible again when it completes with success. If
171 * not, the card will stay dead and registers will be unaccessible
172 * still.
174 static int genwqe_bus_reset(struct genwqe_dev *cd)
176 int rc = 0;
177 struct pci_dev *pci_dev = cd->pci_dev;
178 void __iomem *mmio;
180 if (cd->err_inject & GENWQE_INJECT_BUS_RESET_FAILURE)
181 return -EIO;
183 mmio = cd->mmio;
184 cd->mmio = NULL;
185 pci_iounmap(pci_dev, mmio);
187 pci_release_mem_regions(pci_dev);
190 * Firmware/BIOS might change memory mapping during bus reset.
191 * Settings like enable bus-mastering, ... are backuped and
192 * restored by the pci_reset_function().
194 dev_dbg(&pci_dev->dev, "[%s] pci_reset function ...\n", __func__);
195 rc = pci_reset_function(pci_dev);
196 if (rc) {
197 dev_err(&pci_dev->dev,
198 "[%s] err: failed reset func (rc %d)\n", __func__, rc);
199 return rc;
201 dev_dbg(&pci_dev->dev, "[%s] done with rc=%d\n", __func__, rc);
204 * Here is the right spot to clear the register read
205 * failure. pci_bus_reset() does this job in real systems.
207 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
208 GENWQE_INJECT_GFIR_FATAL |
209 GENWQE_INJECT_GFIR_INFO);
211 rc = pci_request_mem_regions(pci_dev, genwqe_driver_name);
212 if (rc) {
213 dev_err(&pci_dev->dev,
214 "[%s] err: request bars failed (%d)\n", __func__, rc);
215 return -EIO;
218 cd->mmio = pci_iomap(pci_dev, 0, 0);
219 if (cd->mmio == NULL) {
220 dev_err(&pci_dev->dev,
221 "[%s] err: mapping BAR0 failed\n", __func__);
222 return -ENOMEM;
224 return 0;
228 * Hardware circumvention section. Certain bitstreams in our test-lab
229 * had different kinds of problems. Here is where we adjust those
230 * bitstreams to function will with this version of our device driver.
232 * Thise circumventions are applied to the physical function only.
233 * The magical numbers below are identifying development/manufacturing
234 * versions of the bitstream used on the card.
236 * Turn off error reporting for old/manufacturing images.
239 bool genwqe_need_err_masking(struct genwqe_dev *cd)
241 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
244 static void genwqe_tweak_hardware(struct genwqe_dev *cd)
246 struct pci_dev *pci_dev = cd->pci_dev;
248 /* Mask FIRs for development images */
249 if (((cd->slu_unitcfg & 0xFFFF0ull) >= 0x32000ull) &&
250 ((cd->slu_unitcfg & 0xFFFF0ull) <= 0x33250ull)) {
251 dev_warn(&pci_dev->dev,
252 "FIRs masked due to bitstream %016llx.%016llx\n",
253 cd->slu_unitcfg, cd->app_unitcfg);
255 __genwqe_writeq(cd, IO_APP_SEC_LEM_DEBUG_OVR,
256 0xFFFFFFFFFFFFFFFFull);
258 __genwqe_writeq(cd, IO_APP_ERR_ACT_MASK,
259 0x0000000000000000ull);
264 * genwqe_recovery_on_fatal_gfir_required() - Version depended actions
266 * Bitstreams older than 2013-02-17 have a bug where fatal GFIRs must
267 * be ignored. This is e.g. true for the bitstream we gave to the card
268 * manufacturer, but also for some old bitstreams we released to our
269 * test-lab.
271 int genwqe_recovery_on_fatal_gfir_required(struct genwqe_dev *cd)
273 return (cd->slu_unitcfg & 0xFFFF0ull) >= 0x32170ull;
276 int genwqe_flash_readback_fails(struct genwqe_dev *cd)
278 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
282 * genwqe_T_psec() - Calculate PF/VF timeout register content
284 * Note: From a design perspective it turned out to be a bad idea to
285 * use codes here to specifiy the frequency/speed values. An old
286 * driver cannot understand new codes and is therefore always a
287 * problem. Better is to measure out the value or put the
288 * speed/frequency directly into a register which is always a valid
289 * value for old as well as for new software.
291 /* T = 1/f */
292 static int genwqe_T_psec(struct genwqe_dev *cd)
294 u16 speed; /* 1/f -> 250, 200, 166, 175 */
295 static const int T[] = { 4000, 5000, 6000, 5714 };
297 speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
298 if (speed >= ARRAY_SIZE(T))
299 return -1; /* illegal value */
301 return T[speed];
305 * genwqe_setup_pf_jtimer() - Setup PF hardware timeouts for DDCB execution
307 * Do this _after_ card_reset() is called. Otherwise the values will
308 * vanish. The settings need to be done when the queues are inactive.
310 * The max. timeout value is 2^(10+x) * T (6ns for 166MHz) * 15/16.
311 * The min. timeout value is 2^(10+x) * T (6ns for 166MHz) * 14/16.
313 static bool genwqe_setup_pf_jtimer(struct genwqe_dev *cd)
315 u32 T = genwqe_T_psec(cd);
316 u64 x;
318 if (GENWQE_PF_JOBTIMEOUT_MSEC == 0)
319 return false;
321 /* PF: large value needed, flash update 2sec per block */
322 x = ilog2(GENWQE_PF_JOBTIMEOUT_MSEC *
323 16000000000uL/(T * 15)) - 10;
325 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
326 0xff00 | (x & 0xff), 0);
327 return true;
331 * genwqe_setup_vf_jtimer() - Setup VF hardware timeouts for DDCB execution
333 static bool genwqe_setup_vf_jtimer(struct genwqe_dev *cd)
335 struct pci_dev *pci_dev = cd->pci_dev;
336 unsigned int vf;
337 u32 T = genwqe_T_psec(cd);
338 u64 x;
339 int totalvfs;
341 totalvfs = pci_sriov_get_totalvfs(pci_dev);
342 if (totalvfs <= 0)
343 return false;
345 for (vf = 0; vf < totalvfs; 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 =
382 kmalloc_array(e, sizeof(struct genwqe_reg),
383 GFP_KERNEL);
385 * regs == NULL is ok, the using code treats this as no regs,
386 * Printing warning is ok in this case.
389 return 0;
392 static void genwqe_ffdc_buffs_free(struct genwqe_dev *cd)
394 unsigned int type;
396 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
397 kfree(cd->ffdc[type].regs);
398 cd->ffdc[type].regs = NULL;
402 static int genwqe_read_ids(struct genwqe_dev *cd)
404 int err = 0;
405 int slu_id;
406 struct pci_dev *pci_dev = cd->pci_dev;
408 cd->slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
409 if (cd->slu_unitcfg == IO_ILLEGAL_VALUE) {
410 dev_err(&pci_dev->dev,
411 "err: SLUID=%016llx\n", cd->slu_unitcfg);
412 err = -EIO;
413 goto out_err;
416 slu_id = genwqe_get_slu_id(cd);
417 if (slu_id < GENWQE_SLU_ARCH_REQ || slu_id == 0xff) {
418 dev_err(&pci_dev->dev,
419 "err: incompatible SLU Architecture %u\n", slu_id);
420 err = -ENOENT;
421 goto out_err;
424 cd->app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
425 if (cd->app_unitcfg == IO_ILLEGAL_VALUE) {
426 dev_err(&pci_dev->dev,
427 "err: APPID=%016llx\n", cd->app_unitcfg);
428 err = -EIO;
429 goto out_err;
431 genwqe_read_app_id(cd, cd->app_name, sizeof(cd->app_name));
434 * Is access to all registers possible? If we are a VF the
435 * answer is obvious. If we run fully virtualized, we need to
436 * check if we can access all registers. If we do not have
437 * full access we will cause an UR and some informational FIRs
438 * in the PF, but that should not harm.
440 if (pci_dev->is_virtfn)
441 cd->is_privileged = 0;
442 else
443 cd->is_privileged = (__genwqe_readq(cd, IO_SLU_BITSTREAM)
444 != IO_ILLEGAL_VALUE);
446 out_err:
447 return err;
450 static int genwqe_start(struct genwqe_dev *cd)
452 int err;
453 struct pci_dev *pci_dev = cd->pci_dev;
455 err = genwqe_read_ids(cd);
456 if (err)
457 return err;
459 if (genwqe_is_privileged(cd)) {
460 /* do this after the tweaks. alloc fail is acceptable */
461 genwqe_ffdc_buffs_alloc(cd);
462 genwqe_stop_traps(cd);
464 /* Collect registers e.g. FIRs, UNITIDs, traces ... */
465 genwqe_read_ffdc_regs(cd, cd->ffdc[GENWQE_DBG_REGS].regs,
466 cd->ffdc[GENWQE_DBG_REGS].entries, 0);
468 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT0,
469 cd->ffdc[GENWQE_DBG_UNIT0].regs,
470 cd->ffdc[GENWQE_DBG_UNIT0].entries);
472 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT1,
473 cd->ffdc[GENWQE_DBG_UNIT1].regs,
474 cd->ffdc[GENWQE_DBG_UNIT1].entries);
476 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT2,
477 cd->ffdc[GENWQE_DBG_UNIT2].regs,
478 cd->ffdc[GENWQE_DBG_UNIT2].entries);
480 genwqe_start_traps(cd);
482 if (cd->card_state == GENWQE_CARD_FATAL_ERROR) {
483 dev_warn(&pci_dev->dev,
484 "[%s] chip reload/recovery!\n", __func__);
487 * Stealth Mode: Reload chip on either hot
488 * reset or PERST.
490 cd->softreset = 0x7Cull;
491 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
492 cd->softreset);
494 err = genwqe_bus_reset(cd);
495 if (err != 0) {
496 dev_err(&pci_dev->dev,
497 "[%s] err: bus reset failed!\n",
498 __func__);
499 goto out;
503 * Re-read the IDs because
504 * it could happen that the bitstream load
505 * failed!
507 err = genwqe_read_ids(cd);
508 if (err)
509 goto out;
513 err = genwqe_setup_service_layer(cd); /* does a reset to the card */
514 if (err != 0) {
515 dev_err(&pci_dev->dev,
516 "[%s] err: could not setup servicelayer!\n", __func__);
517 err = -ENODEV;
518 goto out;
521 if (genwqe_is_privileged(cd)) { /* code is running _after_ reset */
522 genwqe_tweak_hardware(cd);
524 genwqe_setup_pf_jtimer(cd);
525 genwqe_setup_vf_jtimer(cd);
528 err = genwqe_device_create(cd);
529 if (err < 0) {
530 dev_err(&pci_dev->dev,
531 "err: chdev init failed! (err=%d)\n", err);
532 goto out_release_service_layer;
534 return 0;
536 out_release_service_layer:
537 genwqe_release_service_layer(cd);
538 out:
539 if (genwqe_is_privileged(cd))
540 genwqe_ffdc_buffs_free(cd);
541 return -EIO;
545 * genwqe_stop() - Stop card operation
547 * Recovery notes:
548 * As long as genwqe_thread runs we might access registers during
549 * error data capture. Same is with the genwqe_health_thread.
550 * When genwqe_bus_reset() fails this function might called two times:
551 * first by the genwqe_health_thread() and later by genwqe_remove() to
552 * unbind the device. We must be able to survive that.
554 * This function must be robust enough to be called twice.
556 static int genwqe_stop(struct genwqe_dev *cd)
558 genwqe_finish_queue(cd); /* no register access */
559 genwqe_device_remove(cd); /* device removed, procs killed */
560 genwqe_release_service_layer(cd); /* here genwqe_thread is stopped */
562 if (genwqe_is_privileged(cd)) {
563 pci_disable_sriov(cd->pci_dev); /* access pci config space */
564 genwqe_ffdc_buffs_free(cd);
567 return 0;
571 * genwqe_recover_card() - Try to recover the card if it is possible
573 * If fatal_err is set no register access is possible anymore. It is
574 * likely that genwqe_start fails in that situation. Proper error
575 * handling is required in this case.
577 * genwqe_bus_reset() will cause the pci code to call genwqe_remove()
578 * and later genwqe_probe() for all virtual functions.
580 static int genwqe_recover_card(struct genwqe_dev *cd, int fatal_err)
582 int rc;
583 struct pci_dev *pci_dev = cd->pci_dev;
585 genwqe_stop(cd);
588 * Make sure chip is not reloaded to maintain FFDC. Write SLU
589 * Reset Register, CPLDReset field to 0.
591 if (!fatal_err) {
592 cd->softreset = 0x70ull;
593 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, cd->softreset);
596 rc = genwqe_bus_reset(cd);
597 if (rc != 0) {
598 dev_err(&pci_dev->dev,
599 "[%s] err: card recovery impossible!\n", __func__);
600 return rc;
603 rc = genwqe_start(cd);
604 if (rc < 0) {
605 dev_err(&pci_dev->dev,
606 "[%s] err: failed to launch device!\n", __func__);
607 return rc;
609 return 0;
612 static int genwqe_health_check_cond(struct genwqe_dev *cd, u64 *gfir)
614 *gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
615 return (*gfir & GFIR_ERR_TRIGGER) &&
616 genwqe_recovery_on_fatal_gfir_required(cd);
620 * genwqe_fir_checking() - Check the fault isolation registers of the card
622 * If this code works ok, can be tried out with help of the genwqe_poke tool:
623 * sudo ./tools/genwqe_poke 0x8 0xfefefefefef
625 * Now the relevant FIRs/sFIRs should be printed out and the driver should
626 * invoke recovery (devices are removed and readded).
628 static u64 genwqe_fir_checking(struct genwqe_dev *cd)
630 int j, iterations = 0;
631 u64 mask, fir, fec, uid, gfir, gfir_masked, sfir, sfec;
632 u32 fir_addr, fir_clr_addr, fec_addr, sfir_addr, sfec_addr;
633 struct pci_dev *pci_dev = cd->pci_dev;
635 healthMonitor:
636 iterations++;
637 if (iterations > 16) {
638 dev_err(&pci_dev->dev, "* exit looping after %d times\n",
639 iterations);
640 goto fatal_error;
643 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
644 if (gfir != 0x0)
645 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n",
646 IO_SLC_CFGREG_GFIR, gfir);
647 if (gfir == IO_ILLEGAL_VALUE)
648 goto fatal_error;
651 * Avoid printing when to GFIR bit is on prevents contignous
652 * printout e.g. for the following bug:
653 * FIR set without a 2ndary FIR/FIR cannot be cleared
654 * Comment out the following if to get the prints:
656 if (gfir == 0)
657 return 0;
659 gfir_masked = gfir & GFIR_ERR_TRIGGER; /* fatal errors */
661 for (uid = 0; uid < GENWQE_MAX_UNITS; uid++) { /* 0..2 in zEDC */
663 /* read the primary FIR (pfir) */
664 fir_addr = (uid << 24) + 0x08;
665 fir = __genwqe_readq(cd, fir_addr);
666 if (fir == 0x0)
667 continue; /* no error in this unit */
669 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fir_addr, fir);
670 if (fir == IO_ILLEGAL_VALUE)
671 goto fatal_error;
673 /* read primary FEC */
674 fec_addr = (uid << 24) + 0x18;
675 fec = __genwqe_readq(cd, fec_addr);
677 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fec_addr, fec);
678 if (fec == IO_ILLEGAL_VALUE)
679 goto fatal_error;
681 for (j = 0, mask = 1ULL; j < 64; j++, mask <<= 1) {
683 /* secondary fir empty, skip it */
684 if ((fir & mask) == 0x0)
685 continue;
687 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
688 sfir = __genwqe_readq(cd, sfir_addr);
690 if (sfir == IO_ILLEGAL_VALUE)
691 goto fatal_error;
692 dev_err(&pci_dev->dev,
693 "* 0x%08x 0x%016llx\n", sfir_addr, sfir);
695 sfec_addr = (uid << 24) + 0x300 + 0x08 * j;
696 sfec = __genwqe_readq(cd, sfec_addr);
698 if (sfec == IO_ILLEGAL_VALUE)
699 goto fatal_error;
700 dev_err(&pci_dev->dev,
701 "* 0x%08x 0x%016llx\n", sfec_addr, sfec);
703 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
704 if (gfir == IO_ILLEGAL_VALUE)
705 goto fatal_error;
707 /* gfir turned on during routine! get out and
708 start over. */
709 if ((gfir_masked == 0x0) &&
710 (gfir & GFIR_ERR_TRIGGER)) {
711 goto healthMonitor;
714 /* do not clear if we entered with a fatal gfir */
715 if (gfir_masked == 0x0) {
717 /* NEW clear by mask the logged bits */
718 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
719 __genwqe_writeq(cd, sfir_addr, sfir);
721 dev_dbg(&pci_dev->dev,
722 "[HM] Clearing 2ndary FIR 0x%08x with 0x%016llx\n",
723 sfir_addr, sfir);
726 * note, these cannot be error-Firs
727 * since gfir_masked is 0 after sfir
728 * was read. Also, it is safe to do
729 * this write if sfir=0. Still need to
730 * clear the primary. This just means
731 * there is no secondary FIR.
734 /* clear by mask the logged bit. */
735 fir_clr_addr = (uid << 24) + 0x10;
736 __genwqe_writeq(cd, fir_clr_addr, mask);
738 dev_dbg(&pci_dev->dev,
739 "[HM] Clearing primary FIR 0x%08x with 0x%016llx\n",
740 fir_clr_addr, 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_pci_fundamental_reset() - trigger a PCIe fundamental reset on the slot
766 * Note: pci_set_pcie_reset_state() is not implemented on all archs, so this
767 * reset method will not work in all cases.
769 * Return: 0 on success or error code from pci_set_pcie_reset_state()
771 static int genwqe_pci_fundamental_reset(struct pci_dev *pci_dev)
773 int rc;
776 * lock pci config space access from userspace,
777 * save state and issue PCIe fundamental reset
779 pci_cfg_access_lock(pci_dev);
780 pci_save_state(pci_dev);
781 rc = pci_set_pcie_reset_state(pci_dev, pcie_warm_reset);
782 if (!rc) {
783 /* keep PCIe reset asserted for 250ms */
784 msleep(250);
785 pci_set_pcie_reset_state(pci_dev, pcie_deassert_reset);
786 /* Wait for 2s to reload flash and train the link */
787 msleep(2000);
789 pci_restore_state(pci_dev);
790 pci_cfg_access_unlock(pci_dev);
791 return rc;
795 static int genwqe_platform_recovery(struct genwqe_dev *cd)
797 struct pci_dev *pci_dev = cd->pci_dev;
798 int rc;
800 dev_info(&pci_dev->dev,
801 "[%s] resetting card for error recovery\n", __func__);
803 /* Clear out error injection flags */
804 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
805 GENWQE_INJECT_GFIR_FATAL |
806 GENWQE_INJECT_GFIR_INFO);
808 genwqe_stop(cd);
810 /* Try recoverying the card with fundamental reset */
811 rc = genwqe_pci_fundamental_reset(pci_dev);
812 if (!rc) {
813 rc = genwqe_start(cd);
814 if (!rc)
815 dev_info(&pci_dev->dev,
816 "[%s] card recovered\n", __func__);
817 else
818 dev_err(&pci_dev->dev,
819 "[%s] err: cannot start card services! (err=%d)\n",
820 __func__, rc);
821 } else {
822 dev_err(&pci_dev->dev,
823 "[%s] card reset failed\n", __func__);
826 return rc;
830 * genwqe_reload_bistream() - reload card bitstream
832 * Set the appropriate register and call fundamental reset to reaload the card
833 * bitstream.
835 * Return: 0 on success, error code otherwise
837 static int genwqe_reload_bistream(struct genwqe_dev *cd)
839 struct pci_dev *pci_dev = cd->pci_dev;
840 int rc;
842 dev_info(&pci_dev->dev,
843 "[%s] resetting card for bitstream reload\n",
844 __func__);
846 genwqe_stop(cd);
849 * Cause a CPLD reprogram with the 'next_bitstream'
850 * partition on PCIe hot or fundamental reset
852 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
853 (cd->softreset & 0xcull) | 0x70ull);
855 rc = genwqe_pci_fundamental_reset(pci_dev);
856 if (rc) {
858 * A fundamental reset failure can be caused
859 * by lack of support on the arch, so we just
860 * log the error and try to start the card
861 * again.
863 dev_err(&pci_dev->dev,
864 "[%s] err: failed to reset card for bitstream reload\n",
865 __func__);
868 rc = genwqe_start(cd);
869 if (rc) {
870 dev_err(&pci_dev->dev,
871 "[%s] err: cannot start card services! (err=%d)\n",
872 __func__, rc);
873 return rc;
875 dev_info(&pci_dev->dev,
876 "[%s] card reloaded\n", __func__);
877 return 0;
882 * genwqe_health_thread() - Health checking thread
884 * This thread is only started for the PF of the card.
886 * This thread monitors the health of the card. A critical situation
887 * is when we read registers which contain -1 (IO_ILLEGAL_VALUE). In
888 * this case we need to be recovered from outside. Writing to
889 * registers will very likely not work either.
891 * This thread must only exit if kthread_should_stop() becomes true.
893 * Condition for the health-thread to trigger:
894 * a) when a kthread_stop() request comes in or
895 * b) a critical GFIR occured
897 * Informational GFIRs are checked and potentially printed in
898 * GENWQE_HEALTH_CHECK_INTERVAL seconds.
900 static int genwqe_health_thread(void *data)
902 int rc, should_stop = 0;
903 struct genwqe_dev *cd = data;
904 struct pci_dev *pci_dev = cd->pci_dev;
905 u64 gfir, gfir_masked, slu_unitcfg, app_unitcfg;
907 health_thread_begin:
908 while (!kthread_should_stop()) {
909 rc = wait_event_interruptible_timeout(cd->health_waitq,
910 (genwqe_health_check_cond(cd, &gfir) ||
911 (should_stop = kthread_should_stop())),
912 GENWQE_HEALTH_CHECK_INTERVAL * HZ);
914 if (should_stop)
915 break;
917 if (gfir == IO_ILLEGAL_VALUE) {
918 dev_err(&pci_dev->dev,
919 "[%s] GFIR=%016llx\n", __func__, gfir);
920 goto fatal_error;
923 slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
924 if (slu_unitcfg == IO_ILLEGAL_VALUE) {
925 dev_err(&pci_dev->dev,
926 "[%s] SLU_UNITCFG=%016llx\n",
927 __func__, slu_unitcfg);
928 goto fatal_error;
931 app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
932 if (app_unitcfg == IO_ILLEGAL_VALUE) {
933 dev_err(&pci_dev->dev,
934 "[%s] APP_UNITCFG=%016llx\n",
935 __func__, app_unitcfg);
936 goto fatal_error;
939 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
940 if (gfir == IO_ILLEGAL_VALUE) {
941 dev_err(&pci_dev->dev,
942 "[%s] %s: GFIR=%016llx\n", __func__,
943 (gfir & GFIR_ERR_TRIGGER) ? "err" : "info",
944 gfir);
945 goto fatal_error;
948 gfir_masked = genwqe_fir_checking(cd);
949 if (gfir_masked == IO_ILLEGAL_VALUE)
950 goto fatal_error;
953 * GFIR ErrorTrigger bits set => reset the card!
954 * Never do this for old/manufacturing images!
956 if ((gfir_masked) && !cd->skip_recovery &&
957 genwqe_recovery_on_fatal_gfir_required(cd)) {
959 cd->card_state = GENWQE_CARD_FATAL_ERROR;
961 rc = genwqe_recover_card(cd, 0);
962 if (rc < 0) {
963 /* FIXME Card is unusable and needs unbind! */
964 goto fatal_error;
968 if (cd->card_state == GENWQE_CARD_RELOAD_BITSTREAM) {
969 /* Userspace requested card bitstream reload */
970 rc = genwqe_reload_bistream(cd);
971 if (rc)
972 goto fatal_error;
975 cd->last_gfir = gfir;
976 cond_resched();
979 return 0;
981 fatal_error:
982 if (cd->use_platform_recovery) {
984 * Since we use raw accessors, EEH errors won't be detected
985 * by the platform until we do a non-raw MMIO or config space
986 * read
988 readq(cd->mmio + IO_SLC_CFGREG_GFIR);
990 /* We do nothing if the card is going over PCI recovery */
991 if (pci_channel_offline(pci_dev))
992 return -EIO;
995 * If it's supported by the platform, we try a fundamental reset
996 * to recover from a fatal error. Otherwise, we continue to wait
997 * for an external recovery procedure to take care of it.
999 rc = genwqe_platform_recovery(cd);
1000 if (!rc)
1001 goto health_thread_begin;
1004 dev_err(&pci_dev->dev,
1005 "[%s] card unusable. Please trigger unbind!\n", __func__);
1007 /* Bring down logical devices to inform user space via udev remove. */
1008 cd->card_state = GENWQE_CARD_FATAL_ERROR;
1009 genwqe_stop(cd);
1011 /* genwqe_bus_reset failed(). Now wait for genwqe_remove(). */
1012 while (!kthread_should_stop())
1013 cond_resched();
1015 return -EIO;
1018 static int genwqe_health_check_start(struct genwqe_dev *cd)
1020 int rc;
1022 if (GENWQE_HEALTH_CHECK_INTERVAL <= 0)
1023 return 0; /* valid for disabling the service */
1025 /* moved before request_irq() */
1026 /* init_waitqueue_head(&cd->health_waitq); */
1028 cd->health_thread = kthread_run(genwqe_health_thread, cd,
1029 GENWQE_DEVNAME "%d_health",
1030 cd->card_idx);
1031 if (IS_ERR(cd->health_thread)) {
1032 rc = PTR_ERR(cd->health_thread);
1033 cd->health_thread = NULL;
1034 return rc;
1036 return 0;
1039 static int genwqe_health_thread_running(struct genwqe_dev *cd)
1041 return cd->health_thread != NULL;
1044 static int genwqe_health_check_stop(struct genwqe_dev *cd)
1046 int rc;
1048 if (!genwqe_health_thread_running(cd))
1049 return -EIO;
1051 rc = kthread_stop(cd->health_thread);
1052 cd->health_thread = NULL;
1053 return 0;
1057 * genwqe_pci_setup() - Allocate PCIe related resources for our card
1059 static int genwqe_pci_setup(struct genwqe_dev *cd)
1061 int err;
1062 struct pci_dev *pci_dev = cd->pci_dev;
1064 err = pci_enable_device_mem(pci_dev);
1065 if (err) {
1066 dev_err(&pci_dev->dev,
1067 "err: failed to enable pci memory (err=%d)\n", err);
1068 goto err_out;
1071 /* Reserve PCI I/O and memory resources */
1072 err = pci_request_mem_regions(pci_dev, genwqe_driver_name);
1073 if (err) {
1074 dev_err(&pci_dev->dev,
1075 "[%s] err: request bars failed (%d)\n", __func__, err);
1076 err = -EIO;
1077 goto err_disable_device;
1080 /* check for 64-bit DMA address supported (DAC) */
1081 if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(64))) {
1082 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(64));
1083 if (err) {
1084 dev_err(&pci_dev->dev,
1085 "err: DMA64 consistent mask error\n");
1086 err = -EIO;
1087 goto out_release_resources;
1089 /* check for 32-bit DMA address supported (SAC) */
1090 } else if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(32))) {
1091 err = pci_set_consistent_dma_mask(pci_dev, DMA_BIT_MASK(32));
1092 if (err) {
1093 dev_err(&pci_dev->dev,
1094 "err: DMA32 consistent mask error\n");
1095 err = -EIO;
1096 goto out_release_resources;
1098 } else {
1099 dev_err(&pci_dev->dev,
1100 "err: neither DMA32 nor DMA64 supported\n");
1101 err = -EIO;
1102 goto out_release_resources;
1105 pci_set_master(pci_dev);
1106 pci_enable_pcie_error_reporting(pci_dev);
1108 /* EEH recovery requires PCIe fundamental reset */
1109 pci_dev->needs_freset = 1;
1111 /* request complete BAR-0 space (length = 0) */
1112 cd->mmio_len = pci_resource_len(pci_dev, 0);
1113 cd->mmio = pci_iomap(pci_dev, 0, 0);
1114 if (cd->mmio == NULL) {
1115 dev_err(&pci_dev->dev,
1116 "[%s] err: mapping BAR0 failed\n", __func__);
1117 err = -ENOMEM;
1118 goto out_release_resources;
1121 cd->num_vfs = pci_sriov_get_totalvfs(pci_dev);
1122 if (cd->num_vfs < 0)
1123 cd->num_vfs = 0;
1125 err = genwqe_read_ids(cd);
1126 if (err)
1127 goto out_iounmap;
1129 return 0;
1131 out_iounmap:
1132 pci_iounmap(pci_dev, cd->mmio);
1133 out_release_resources:
1134 pci_release_mem_regions(pci_dev);
1135 err_disable_device:
1136 pci_disable_device(pci_dev);
1137 err_out:
1138 return err;
1142 * genwqe_pci_remove() - Free PCIe related resources for our card
1144 static void genwqe_pci_remove(struct genwqe_dev *cd)
1146 struct pci_dev *pci_dev = cd->pci_dev;
1148 if (cd->mmio)
1149 pci_iounmap(pci_dev, cd->mmio);
1151 pci_release_mem_regions(pci_dev);
1152 pci_disable_device(pci_dev);
1156 * genwqe_probe() - Device initialization
1157 * @pdev: PCI device information struct
1159 * Callable for multiple cards. This function is called on bind.
1161 * Return: 0 if succeeded, < 0 when failed
1163 static int genwqe_probe(struct pci_dev *pci_dev,
1164 const struct pci_device_id *id)
1166 int err;
1167 struct genwqe_dev *cd;
1169 genwqe_init_crc32();
1171 cd = genwqe_dev_alloc();
1172 if (IS_ERR(cd)) {
1173 dev_err(&pci_dev->dev, "err: could not alloc mem (err=%d)!\n",
1174 (int)PTR_ERR(cd));
1175 return PTR_ERR(cd);
1178 dev_set_drvdata(&pci_dev->dev, cd);
1179 cd->pci_dev = pci_dev;
1181 err = genwqe_pci_setup(cd);
1182 if (err < 0) {
1183 dev_err(&pci_dev->dev,
1184 "err: problems with PCI setup (err=%d)\n", err);
1185 goto out_free_dev;
1188 err = genwqe_start(cd);
1189 if (err < 0) {
1190 dev_err(&pci_dev->dev,
1191 "err: cannot start card services! (err=%d)\n", err);
1192 goto out_pci_remove;
1195 if (genwqe_is_privileged(cd)) {
1196 err = genwqe_health_check_start(cd);
1197 if (err < 0) {
1198 dev_err(&pci_dev->dev,
1199 "err: cannot start health checking! (err=%d)\n",
1200 err);
1201 goto out_stop_services;
1204 return 0;
1206 out_stop_services:
1207 genwqe_stop(cd);
1208 out_pci_remove:
1209 genwqe_pci_remove(cd);
1210 out_free_dev:
1211 genwqe_dev_free(cd);
1212 return err;
1216 * genwqe_remove() - Called when device is removed (hot-plugable)
1218 * Or when driver is unloaded respecitively when unbind is done.
1220 static void genwqe_remove(struct pci_dev *pci_dev)
1222 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1224 genwqe_health_check_stop(cd);
1227 * genwqe_stop() must survive if it is called twice
1228 * sequentially. This happens when the health thread calls it
1229 * and fails on genwqe_bus_reset().
1231 genwqe_stop(cd);
1232 genwqe_pci_remove(cd);
1233 genwqe_dev_free(cd);
1237 * genwqe_err_error_detected() - Error detection callback
1239 * This callback is called by the PCI subsystem whenever a PCI bus
1240 * error is detected.
1242 static pci_ers_result_t genwqe_err_error_detected(struct pci_dev *pci_dev,
1243 enum pci_channel_state state)
1245 struct genwqe_dev *cd;
1247 dev_err(&pci_dev->dev, "[%s] state=%d\n", __func__, state);
1249 cd = dev_get_drvdata(&pci_dev->dev);
1250 if (cd == NULL)
1251 return PCI_ERS_RESULT_DISCONNECT;
1253 /* Stop the card */
1254 genwqe_health_check_stop(cd);
1255 genwqe_stop(cd);
1258 * On permanent failure, the PCI code will call device remove
1259 * after the return of this function.
1260 * genwqe_stop() can be called twice.
1262 if (state == pci_channel_io_perm_failure) {
1263 return PCI_ERS_RESULT_DISCONNECT;
1264 } else {
1265 genwqe_pci_remove(cd);
1266 return PCI_ERS_RESULT_NEED_RESET;
1270 static pci_ers_result_t genwqe_err_slot_reset(struct pci_dev *pci_dev)
1272 int rc;
1273 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1275 rc = genwqe_pci_setup(cd);
1276 if (!rc) {
1277 return PCI_ERS_RESULT_RECOVERED;
1278 } else {
1279 dev_err(&pci_dev->dev,
1280 "err: problems with PCI setup (err=%d)\n", rc);
1281 return PCI_ERS_RESULT_DISCONNECT;
1285 static pci_ers_result_t genwqe_err_result_none(struct pci_dev *dev)
1287 return PCI_ERS_RESULT_NONE;
1290 static void genwqe_err_resume(struct pci_dev *pci_dev)
1292 int rc;
1293 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1295 rc = genwqe_start(cd);
1296 if (!rc) {
1297 rc = genwqe_health_check_start(cd);
1298 if (rc)
1299 dev_err(&pci_dev->dev,
1300 "err: cannot start health checking! (err=%d)\n",
1301 rc);
1302 } else {
1303 dev_err(&pci_dev->dev,
1304 "err: cannot start card services! (err=%d)\n", rc);
1308 static int genwqe_sriov_configure(struct pci_dev *dev, int numvfs)
1310 int rc;
1311 struct genwqe_dev *cd = dev_get_drvdata(&dev->dev);
1313 if (numvfs > 0) {
1314 genwqe_setup_vf_jtimer(cd);
1315 rc = pci_enable_sriov(dev, numvfs);
1316 if (rc < 0)
1317 return rc;
1318 return numvfs;
1320 if (numvfs == 0) {
1321 pci_disable_sriov(dev);
1322 return 0;
1324 return 0;
1327 static struct pci_error_handlers genwqe_err_handler = {
1328 .error_detected = genwqe_err_error_detected,
1329 .mmio_enabled = genwqe_err_result_none,
1330 .slot_reset = genwqe_err_slot_reset,
1331 .resume = genwqe_err_resume,
1334 static struct pci_driver genwqe_driver = {
1335 .name = genwqe_driver_name,
1336 .id_table = genwqe_device_table,
1337 .probe = genwqe_probe,
1338 .remove = genwqe_remove,
1339 .sriov_configure = genwqe_sriov_configure,
1340 .err_handler = &genwqe_err_handler,
1344 * genwqe_devnode() - Set default access mode for genwqe devices.
1346 * Default mode should be rw for everybody. Do not change default
1347 * device name.
1349 static char *genwqe_devnode(struct device *dev, umode_t *mode)
1351 if (mode)
1352 *mode = 0666;
1353 return NULL;
1357 * genwqe_init_module() - Driver registration and initialization
1359 static int __init genwqe_init_module(void)
1361 int rc;
1363 class_genwqe = class_create(THIS_MODULE, GENWQE_DEVNAME);
1364 if (IS_ERR(class_genwqe)) {
1365 pr_err("[%s] create class failed\n", __func__);
1366 return -ENOMEM;
1369 class_genwqe->devnode = genwqe_devnode;
1371 debugfs_genwqe = debugfs_create_dir(GENWQE_DEVNAME, NULL);
1373 rc = pci_register_driver(&genwqe_driver);
1374 if (rc != 0) {
1375 pr_err("[%s] pci_reg_driver (rc=%d)\n", __func__, rc);
1376 goto err_out0;
1379 return rc;
1381 err_out0:
1382 debugfs_remove(debugfs_genwqe);
1383 class_destroy(class_genwqe);
1384 return rc;
1388 * genwqe_exit_module() - Driver exit
1390 static void __exit genwqe_exit_module(void)
1392 pci_unregister_driver(&genwqe_driver);
1393 debugfs_remove(debugfs_genwqe);
1394 class_destroy(class_genwqe);
1397 module_init(genwqe_init_module);
1398 module_exit(genwqe_exit_module);