1 // SPDX-License-Identifier: GPL-2.0-only
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/string.h>
23 #include <linux/sched.h>
24 #include <linux/wait.h>
25 #include <linux/delay.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/module.h>
28 #include <linux/notifier.h>
29 #include <linux/device.h>
30 #include <linux/log2.h>
32 #include "card_base.h"
33 #include "card_ddcb.h"
35 MODULE_AUTHOR("Frank Haverkamp <haver@linux.vnet.ibm.com>");
36 MODULE_AUTHOR("Michael Ruettger <michael@ibmra.de>");
37 MODULE_AUTHOR("Joerg-Stephan Vogt <jsvogt@de.ibm.com>");
38 MODULE_AUTHOR("Michael Jung <mijung@gmx.net>");
40 MODULE_DESCRIPTION("GenWQE Card");
41 MODULE_VERSION(DRV_VERSION);
42 MODULE_LICENSE("GPL");
44 static char genwqe_driver_name[] = GENWQE_DEVNAME;
45 static struct class *class_genwqe;
46 static struct dentry *debugfs_genwqe;
47 static struct genwqe_dev *genwqe_devices[GENWQE_CARD_NO_MAX];
49 /* PCI structure for identifying device by PCI vendor and device ID */
50 static const struct pci_device_id genwqe_device_table[] = {
51 { .vendor = PCI_VENDOR_ID_IBM,
52 .device = PCI_DEVICE_GENWQE,
53 .subvendor = PCI_SUBVENDOR_ID_IBM,
54 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
55 .class = (PCI_CLASSCODE_GENWQE5 << 8),
59 /* Initial SR-IOV bring-up image */
60 { .vendor = PCI_VENDOR_ID_IBM,
61 .device = PCI_DEVICE_GENWQE,
62 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
63 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
64 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
68 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
69 .device = 0x0000, /* VF Device ID */
70 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
71 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_SRIOV,
72 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
77 { .vendor = PCI_VENDOR_ID_IBM,
78 .device = PCI_DEVICE_GENWQE,
79 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
80 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
81 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
85 { .vendor = PCI_VENDOR_ID_IBM, /* VF Vendor ID */
86 .device = 0x0000, /* VF Device ID */
87 .subvendor = PCI_SUBVENDOR_ID_IBM_SRIOV,
88 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5,
89 .class = (PCI_CLASSCODE_GENWQE5_SRIOV << 8),
93 /* Even one more ... */
94 { .vendor = PCI_VENDOR_ID_IBM,
95 .device = PCI_DEVICE_GENWQE,
96 .subvendor = PCI_SUBVENDOR_ID_IBM,
97 .subdevice = PCI_SUBSYSTEM_ID_GENWQE5_NEW,
98 .class = (PCI_CLASSCODE_GENWQE5 << 8),
102 { 0, } /* 0 terminated list. */
105 MODULE_DEVICE_TABLE(pci, genwqe_device_table);
108 * genwqe_dev_alloc() - Create and prepare a new card descriptor
110 * Return: Pointer to card descriptor, or ERR_PTR(err) on error
112 static struct genwqe_dev *genwqe_dev_alloc(void)
114 unsigned int i = 0, j;
115 struct genwqe_dev *cd;
117 for (i = 0; i < GENWQE_CARD_NO_MAX; i++) {
118 if (genwqe_devices[i] == NULL)
121 if (i >= GENWQE_CARD_NO_MAX)
122 return ERR_PTR(-ENODEV);
124 cd = kzalloc(sizeof(struct genwqe_dev), GFP_KERNEL);
126 return ERR_PTR(-ENOMEM);
129 cd->class_genwqe = class_genwqe;
130 cd->debugfs_genwqe = debugfs_genwqe;
133 * This comes from kernel config option and can be overritten via
136 cd->use_platform_recovery = CONFIG_GENWQE_PLATFORM_ERROR_RECOVERY;
138 init_waitqueue_head(&cd->queue_waitq);
140 spin_lock_init(&cd->file_lock);
141 INIT_LIST_HEAD(&cd->file_list);
143 cd->card_state = GENWQE_CARD_UNUSED;
144 spin_lock_init(&cd->print_lock);
146 cd->ddcb_software_timeout = GENWQE_DDCB_SOFTWARE_TIMEOUT;
147 cd->kill_timeout = GENWQE_KILL_TIMEOUT;
149 for (j = 0; j < GENWQE_MAX_VFS; j++)
150 cd->vf_jobtimeout_msec[j] = GENWQE_VF_JOBTIMEOUT_MSEC;
152 genwqe_devices[i] = cd;
156 static void genwqe_dev_free(struct genwqe_dev *cd)
161 genwqe_devices[cd->card_idx] = NULL;
166 * genwqe_bus_reset() - Card recovery
167 * @cd: GenWQE device information
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
174 static int genwqe_bus_reset(struct genwqe_dev *cd)
177 struct pci_dev *pci_dev = cd->pci_dev;
180 if (cd->err_inject & GENWQE_INJECT_BUS_RESET_FAILURE)
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);
197 dev_err(&pci_dev->dev,
198 "[%s] err: failed reset func (rc %d)\n", __func__, 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);
213 dev_err(&pci_dev->dev,
214 "[%s] err: request bars failed (%d)\n", __func__, rc);
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__);
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
265 * @cd: GenWQE device information
267 * Bitstreams older than 2013-02-17 have a bug where fatal GFIRs must
268 * be ignored. This is e.g. true for the bitstream we gave to the card
269 * manufacturer, but also for some old bitstreams we released to our
272 int genwqe_recovery_on_fatal_gfir_required(struct genwqe_dev *cd)
274 return (cd->slu_unitcfg & 0xFFFF0ull) >= 0x32170ull;
277 int genwqe_flash_readback_fails(struct genwqe_dev *cd)
279 return (cd->slu_unitcfg & 0xFFFF0ull) < 0x32170ull;
283 * genwqe_T_psec() - Calculate PF/VF timeout register content
284 * @cd: GenWQE device information
286 * Note: From a design perspective it turned out to be a bad idea to
287 * use codes here to specifiy the frequency/speed values. An old
288 * driver cannot understand new codes and is therefore always a
289 * problem. Better is to measure out the value or put the
290 * speed/frequency directly into a register which is always a valid
291 * value for old as well as for new software.
294 static int genwqe_T_psec(struct genwqe_dev *cd)
296 u16 speed; /* 1/f -> 250, 200, 166, 175 */
297 static const int T[] = { 4000, 5000, 6000, 5714 };
299 speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
300 if (speed >= ARRAY_SIZE(T))
301 return -1; /* illegal value */
307 * genwqe_setup_pf_jtimer() - Setup PF hardware timeouts for DDCB execution
308 * @cd: GenWQE device information
310 * Do this _after_ card_reset() is called. Otherwise the values will
311 * vanish. The settings need to be done when the queues are inactive.
313 * The max. timeout value is 2^(10+x) * T (6ns for 166MHz) * 15/16.
314 * The min. timeout value is 2^(10+x) * T (6ns for 166MHz) * 14/16.
316 static bool genwqe_setup_pf_jtimer(struct genwqe_dev *cd)
318 u32 T = genwqe_T_psec(cd);
321 if (GENWQE_PF_JOBTIMEOUT_MSEC == 0)
324 /* PF: large value needed, flash update 2sec per block */
325 x = ilog2(GENWQE_PF_JOBTIMEOUT_MSEC *
326 16000000000uL/(T * 15)) - 10;
328 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
329 0xff00 | (x & 0xff), 0);
334 * genwqe_setup_vf_jtimer() - Setup VF hardware timeouts for DDCB execution
335 * @cd: GenWQE device information
337 static bool genwqe_setup_vf_jtimer(struct genwqe_dev *cd)
339 struct pci_dev *pci_dev = cd->pci_dev;
341 u32 T = genwqe_T_psec(cd);
345 totalvfs = pci_sriov_get_totalvfs(pci_dev);
349 for (vf = 0; vf < totalvfs; vf++) {
351 if (cd->vf_jobtimeout_msec[vf] == 0)
354 x = ilog2(cd->vf_jobtimeout_msec[vf] *
355 16000000000uL/(T * 15)) - 10;
357 genwqe_write_vreg(cd, IO_SLC_VF_APPJOB_TIMEOUT,
358 0xff00 | (x & 0xff), vf + 1);
363 static int genwqe_ffdc_buffs_alloc(struct genwqe_dev *cd)
365 unsigned int type, e = 0;
367 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
369 case GENWQE_DBG_UNIT0:
370 e = genwqe_ffdc_buff_size(cd, 0);
372 case GENWQE_DBG_UNIT1:
373 e = genwqe_ffdc_buff_size(cd, 1);
375 case GENWQE_DBG_UNIT2:
376 e = genwqe_ffdc_buff_size(cd, 2);
378 case GENWQE_DBG_REGS:
379 e = GENWQE_FFDC_REGS;
383 /* currently support only the debug units mentioned here */
384 cd->ffdc[type].entries = e;
385 cd->ffdc[type].regs =
386 kmalloc_array(e, sizeof(struct genwqe_reg),
389 * regs == NULL is ok, the using code treats this as no regs,
390 * Printing warning is ok in this case.
396 static void genwqe_ffdc_buffs_free(struct genwqe_dev *cd)
400 for (type = 0; type < GENWQE_DBG_UNITS; type++) {
401 kfree(cd->ffdc[type].regs);
402 cd->ffdc[type].regs = NULL;
406 static int genwqe_read_ids(struct genwqe_dev *cd)
410 struct pci_dev *pci_dev = cd->pci_dev;
412 cd->slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
413 if (cd->slu_unitcfg == IO_ILLEGAL_VALUE) {
414 dev_err(&pci_dev->dev,
415 "err: SLUID=%016llx\n", cd->slu_unitcfg);
420 slu_id = genwqe_get_slu_id(cd);
421 if (slu_id < GENWQE_SLU_ARCH_REQ || slu_id == 0xff) {
422 dev_err(&pci_dev->dev,
423 "err: incompatible SLU Architecture %u\n", slu_id);
428 cd->app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
429 if (cd->app_unitcfg == IO_ILLEGAL_VALUE) {
430 dev_err(&pci_dev->dev,
431 "err: APPID=%016llx\n", cd->app_unitcfg);
435 genwqe_read_app_id(cd, cd->app_name, sizeof(cd->app_name));
438 * Is access to all registers possible? If we are a VF the
439 * answer is obvious. If we run fully virtualized, we need to
440 * check if we can access all registers. If we do not have
441 * full access we will cause an UR and some informational FIRs
442 * in the PF, but that should not harm.
444 if (pci_dev->is_virtfn)
445 cd->is_privileged = 0;
447 cd->is_privileged = (__genwqe_readq(cd, IO_SLU_BITSTREAM)
448 != IO_ILLEGAL_VALUE);
454 static int genwqe_start(struct genwqe_dev *cd)
457 struct pci_dev *pci_dev = cd->pci_dev;
459 err = genwqe_read_ids(cd);
463 if (genwqe_is_privileged(cd)) {
464 /* do this after the tweaks. alloc fail is acceptable */
465 genwqe_ffdc_buffs_alloc(cd);
466 genwqe_stop_traps(cd);
468 /* Collect registers e.g. FIRs, UNITIDs, traces ... */
469 genwqe_read_ffdc_regs(cd, cd->ffdc[GENWQE_DBG_REGS].regs,
470 cd->ffdc[GENWQE_DBG_REGS].entries, 0);
472 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT0,
473 cd->ffdc[GENWQE_DBG_UNIT0].regs,
474 cd->ffdc[GENWQE_DBG_UNIT0].entries);
476 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT1,
477 cd->ffdc[GENWQE_DBG_UNIT1].regs,
478 cd->ffdc[GENWQE_DBG_UNIT1].entries);
480 genwqe_ffdc_buff_read(cd, GENWQE_DBG_UNIT2,
481 cd->ffdc[GENWQE_DBG_UNIT2].regs,
482 cd->ffdc[GENWQE_DBG_UNIT2].entries);
484 genwqe_start_traps(cd);
486 if (cd->card_state == GENWQE_CARD_FATAL_ERROR) {
487 dev_warn(&pci_dev->dev,
488 "[%s] chip reload/recovery!\n", __func__);
491 * Stealth Mode: Reload chip on either hot
494 cd->softreset = 0x7Cull;
495 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
498 err = genwqe_bus_reset(cd);
500 dev_err(&pci_dev->dev,
501 "[%s] err: bus reset failed!\n",
507 * Re-read the IDs because
508 * it could happen that the bitstream load
511 err = genwqe_read_ids(cd);
517 err = genwqe_setup_service_layer(cd); /* does a reset to the card */
519 dev_err(&pci_dev->dev,
520 "[%s] err: could not setup servicelayer!\n", __func__);
525 if (genwqe_is_privileged(cd)) { /* code is running _after_ reset */
526 genwqe_tweak_hardware(cd);
528 genwqe_setup_pf_jtimer(cd);
529 genwqe_setup_vf_jtimer(cd);
532 err = genwqe_device_create(cd);
534 dev_err(&pci_dev->dev,
535 "err: chdev init failed! (err=%d)\n", err);
536 goto out_release_service_layer;
540 out_release_service_layer:
541 genwqe_release_service_layer(cd);
543 if (genwqe_is_privileged(cd))
544 genwqe_ffdc_buffs_free(cd);
549 * genwqe_stop() - Stop card operation
550 * @cd: GenWQE device information
553 * As long as genwqe_thread runs we might access registers during
554 * error data capture. Same is with the genwqe_health_thread.
555 * When genwqe_bus_reset() fails this function might called two times:
556 * first by the genwqe_health_thread() and later by genwqe_remove() to
557 * unbind the device. We must be able to survive that.
559 * This function must be robust enough to be called twice.
561 static int genwqe_stop(struct genwqe_dev *cd)
563 genwqe_finish_queue(cd); /* no register access */
564 genwqe_device_remove(cd); /* device removed, procs killed */
565 genwqe_release_service_layer(cd); /* here genwqe_thread is stopped */
567 if (genwqe_is_privileged(cd)) {
568 pci_disable_sriov(cd->pci_dev); /* access pci config space */
569 genwqe_ffdc_buffs_free(cd);
576 * genwqe_recover_card() - Try to recover the card if it is possible
577 * @cd: GenWQE device information
578 * @fatal_err: Indicate whether to attempt soft reset
580 * If fatal_err is set no register access is possible anymore. It is
581 * likely that genwqe_start fails in that situation. Proper error
582 * handling is required in this case.
584 * genwqe_bus_reset() will cause the pci code to call genwqe_remove()
585 * and later genwqe_probe() for all virtual functions.
587 static int genwqe_recover_card(struct genwqe_dev *cd, int fatal_err)
590 struct pci_dev *pci_dev = cd->pci_dev;
595 * Make sure chip is not reloaded to maintain FFDC. Write SLU
596 * Reset Register, CPLDReset field to 0.
599 cd->softreset = 0x70ull;
600 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, cd->softreset);
603 rc = genwqe_bus_reset(cd);
605 dev_err(&pci_dev->dev,
606 "[%s] err: card recovery impossible!\n", __func__);
610 rc = genwqe_start(cd);
612 dev_err(&pci_dev->dev,
613 "[%s] err: failed to launch device!\n", __func__);
619 static int genwqe_health_check_cond(struct genwqe_dev *cd, u64 *gfir)
621 *gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
622 return (*gfir & GFIR_ERR_TRIGGER) &&
623 genwqe_recovery_on_fatal_gfir_required(cd);
627 * genwqe_fir_checking() - Check the fault isolation registers of the card
628 * @cd: GenWQE device information
630 * If this code works ok, can be tried out with help of the genwqe_poke tool:
631 * sudo ./tools/genwqe_poke 0x8 0xfefefefefef
633 * Now the relevant FIRs/sFIRs should be printed out and the driver should
634 * invoke recovery (devices are removed and readded).
636 static u64 genwqe_fir_checking(struct genwqe_dev *cd)
638 int j, iterations = 0;
639 u64 mask, fir, fec, uid, gfir, gfir_masked, sfir, sfec;
640 u32 fir_addr, fir_clr_addr, fec_addr, sfir_addr, sfec_addr;
641 struct pci_dev *pci_dev = cd->pci_dev;
645 if (iterations > 16) {
646 dev_err(&pci_dev->dev, "* exit looping after %d times\n",
651 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
653 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n",
654 IO_SLC_CFGREG_GFIR, gfir);
655 if (gfir == IO_ILLEGAL_VALUE)
659 * Avoid printing when to GFIR bit is on prevents contignous
660 * printout e.g. for the following bug:
661 * FIR set without a 2ndary FIR/FIR cannot be cleared
662 * Comment out the following if to get the prints:
667 gfir_masked = gfir & GFIR_ERR_TRIGGER; /* fatal errors */
669 for (uid = 0; uid < GENWQE_MAX_UNITS; uid++) { /* 0..2 in zEDC */
671 /* read the primary FIR (pfir) */
672 fir_addr = (uid << 24) + 0x08;
673 fir = __genwqe_readq(cd, fir_addr);
675 continue; /* no error in this unit */
677 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fir_addr, fir);
678 if (fir == IO_ILLEGAL_VALUE)
681 /* read primary FEC */
682 fec_addr = (uid << 24) + 0x18;
683 fec = __genwqe_readq(cd, fec_addr);
685 dev_err(&pci_dev->dev, "* 0x%08x 0x%016llx\n", fec_addr, fec);
686 if (fec == IO_ILLEGAL_VALUE)
689 for (j = 0, mask = 1ULL; j < 64; j++, mask <<= 1) {
691 /* secondary fir empty, skip it */
692 if ((fir & mask) == 0x0)
695 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
696 sfir = __genwqe_readq(cd, sfir_addr);
698 if (sfir == IO_ILLEGAL_VALUE)
700 dev_err(&pci_dev->dev,
701 "* 0x%08x 0x%016llx\n", sfir_addr, sfir);
703 sfec_addr = (uid << 24) + 0x300 + 0x08 * j;
704 sfec = __genwqe_readq(cd, sfec_addr);
706 if (sfec == IO_ILLEGAL_VALUE)
708 dev_err(&pci_dev->dev,
709 "* 0x%08x 0x%016llx\n", sfec_addr, sfec);
711 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
712 if (gfir == IO_ILLEGAL_VALUE)
715 /* gfir turned on during routine! get out and
717 if ((gfir_masked == 0x0) &&
718 (gfir & GFIR_ERR_TRIGGER)) {
722 /* do not clear if we entered with a fatal gfir */
723 if (gfir_masked == 0x0) {
725 /* NEW clear by mask the logged bits */
726 sfir_addr = (uid << 24) + 0x100 + 0x08 * j;
727 __genwqe_writeq(cd, sfir_addr, sfir);
729 dev_dbg(&pci_dev->dev,
730 "[HM] Clearing 2ndary FIR 0x%08x with 0x%016llx\n",
734 * note, these cannot be error-Firs
735 * since gfir_masked is 0 after sfir
736 * was read. Also, it is safe to do
737 * this write if sfir=0. Still need to
738 * clear the primary. This just means
739 * there is no secondary FIR.
742 /* clear by mask the logged bit. */
743 fir_clr_addr = (uid << 24) + 0x10;
744 __genwqe_writeq(cd, fir_clr_addr, mask);
746 dev_dbg(&pci_dev->dev,
747 "[HM] Clearing primary FIR 0x%08x with 0x%016llx\n",
752 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
753 if (gfir == IO_ILLEGAL_VALUE)
756 if ((gfir_masked == 0x0) && (gfir & GFIR_ERR_TRIGGER)) {
758 * Check once more that it didn't go on after all the
761 dev_dbg(&pci_dev->dev, "ACK! Another FIR! Recursing %d!\n",
768 return IO_ILLEGAL_VALUE;
772 * genwqe_pci_fundamental_reset() - trigger a PCIe fundamental reset on the slot
773 * @pci_dev: PCI device information struct
775 * Note: pci_set_pcie_reset_state() is not implemented on all archs, so this
776 * reset method will not work in all cases.
778 * Return: 0 on success or error code from pci_set_pcie_reset_state()
780 static int genwqe_pci_fundamental_reset(struct pci_dev *pci_dev)
785 * lock pci config space access from userspace,
786 * save state and issue PCIe fundamental reset
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);
792 /* keep PCIe reset asserted for 250ms */
794 pci_set_pcie_reset_state(pci_dev, pcie_deassert_reset);
795 /* Wait for 2s to reload flash and train the link */
798 pci_restore_state(pci_dev);
799 pci_cfg_access_unlock(pci_dev);
804 static int genwqe_platform_recovery(struct genwqe_dev *cd)
806 struct pci_dev *pci_dev = cd->pci_dev;
809 dev_info(&pci_dev->dev,
810 "[%s] resetting card for error recovery\n", __func__);
812 /* Clear out error injection flags */
813 cd->err_inject &= ~(GENWQE_INJECT_HARDWARE_FAILURE |
814 GENWQE_INJECT_GFIR_FATAL |
815 GENWQE_INJECT_GFIR_INFO);
819 /* Try recoverying the card with fundamental reset */
820 rc = genwqe_pci_fundamental_reset(pci_dev);
822 rc = genwqe_start(cd);
824 dev_info(&pci_dev->dev,
825 "[%s] card recovered\n", __func__);
827 dev_err(&pci_dev->dev,
828 "[%s] err: cannot start card services! (err=%d)\n",
831 dev_err(&pci_dev->dev,
832 "[%s] card reset failed\n", __func__);
839 * genwqe_reload_bistream() - reload card bitstream
840 * @cd: GenWQE device information
842 * Set the appropriate register and call fundamental reset to reaload the card
845 * Return: 0 on success, error code otherwise
847 static int genwqe_reload_bistream(struct genwqe_dev *cd)
849 struct pci_dev *pci_dev = cd->pci_dev;
852 dev_info(&pci_dev->dev,
853 "[%s] resetting card for bitstream reload\n",
859 * Cause a CPLD reprogram with the 'next_bitstream'
860 * partition on PCIe hot or fundamental reset
862 __genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET,
863 (cd->softreset & 0xcull) | 0x70ull);
865 rc = genwqe_pci_fundamental_reset(pci_dev);
868 * A fundamental reset failure can be caused
869 * by lack of support on the arch, so we just
870 * log the error and try to start the card
873 dev_err(&pci_dev->dev,
874 "[%s] err: failed to reset card for bitstream reload\n",
878 rc = genwqe_start(cd);
880 dev_err(&pci_dev->dev,
881 "[%s] err: cannot start card services! (err=%d)\n",
885 dev_info(&pci_dev->dev,
886 "[%s] card reloaded\n", __func__);
892 * genwqe_health_thread() - Health checking thread
893 * @data: GenWQE device information
895 * This thread is only started for the PF of the card.
897 * This thread monitors the health of the card. A critical situation
898 * is when we read registers which contain -1 (IO_ILLEGAL_VALUE). In
899 * this case we need to be recovered from outside. Writing to
900 * registers will very likely not work either.
902 * This thread must only exit if kthread_should_stop() becomes true.
904 * Condition for the health-thread to trigger:
905 * a) when a kthread_stop() request comes in or
906 * b) a critical GFIR occured
908 * Informational GFIRs are checked and potentially printed in
909 * GENWQE_HEALTH_CHECK_INTERVAL seconds.
911 static int genwqe_health_thread(void *data)
913 int rc, should_stop = 0;
914 struct genwqe_dev *cd = data;
915 struct pci_dev *pci_dev = cd->pci_dev;
916 u64 gfir, gfir_masked, slu_unitcfg, app_unitcfg;
919 while (!kthread_should_stop()) {
920 rc = wait_event_interruptible_timeout(cd->health_waitq,
921 (genwqe_health_check_cond(cd, &gfir) ||
922 (should_stop = kthread_should_stop())),
923 GENWQE_HEALTH_CHECK_INTERVAL * HZ);
928 if (gfir == IO_ILLEGAL_VALUE) {
929 dev_err(&pci_dev->dev,
930 "[%s] GFIR=%016llx\n", __func__, gfir);
934 slu_unitcfg = __genwqe_readq(cd, IO_SLU_UNITCFG);
935 if (slu_unitcfg == IO_ILLEGAL_VALUE) {
936 dev_err(&pci_dev->dev,
937 "[%s] SLU_UNITCFG=%016llx\n",
938 __func__, slu_unitcfg);
942 app_unitcfg = __genwqe_readq(cd, IO_APP_UNITCFG);
943 if (app_unitcfg == IO_ILLEGAL_VALUE) {
944 dev_err(&pci_dev->dev,
945 "[%s] APP_UNITCFG=%016llx\n",
946 __func__, app_unitcfg);
950 gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
951 if (gfir == IO_ILLEGAL_VALUE) {
952 dev_err(&pci_dev->dev,
953 "[%s] %s: GFIR=%016llx\n", __func__,
954 (gfir & GFIR_ERR_TRIGGER) ? "err" : "info",
959 gfir_masked = genwqe_fir_checking(cd);
960 if (gfir_masked == IO_ILLEGAL_VALUE)
964 * GFIR ErrorTrigger bits set => reset the card!
965 * Never do this for old/manufacturing images!
967 if ((gfir_masked) && !cd->skip_recovery &&
968 genwqe_recovery_on_fatal_gfir_required(cd)) {
970 cd->card_state = GENWQE_CARD_FATAL_ERROR;
972 rc = genwqe_recover_card(cd, 0);
974 /* FIXME Card is unusable and needs unbind! */
979 if (cd->card_state == GENWQE_CARD_RELOAD_BITSTREAM) {
980 /* Userspace requested card bitstream reload */
981 rc = genwqe_reload_bistream(cd);
986 cd->last_gfir = gfir;
993 if (cd->use_platform_recovery) {
995 * Since we use raw accessors, EEH errors won't be detected
996 * by the platform until we do a non-raw MMIO or config space
999 readq(cd->mmio + IO_SLC_CFGREG_GFIR);
1001 /* We do nothing if the card is going over PCI recovery */
1002 if (pci_channel_offline(pci_dev))
1006 * If it's supported by the platform, we try a fundamental reset
1007 * to recover from a fatal error. Otherwise, we continue to wait
1008 * for an external recovery procedure to take care of it.
1010 rc = genwqe_platform_recovery(cd);
1012 goto health_thread_begin;
1015 dev_err(&pci_dev->dev,
1016 "[%s] card unusable. Please trigger unbind!\n", __func__);
1018 /* Bring down logical devices to inform user space via udev remove. */
1019 cd->card_state = GENWQE_CARD_FATAL_ERROR;
1022 /* genwqe_bus_reset failed(). Now wait for genwqe_remove(). */
1023 while (!kthread_should_stop())
1029 static int genwqe_health_check_start(struct genwqe_dev *cd)
1033 if (GENWQE_HEALTH_CHECK_INTERVAL <= 0)
1034 return 0; /* valid for disabling the service */
1036 /* moved before request_irq() */
1037 /* init_waitqueue_head(&cd->health_waitq); */
1039 cd->health_thread = kthread_run(genwqe_health_thread, cd,
1040 GENWQE_DEVNAME "%d_health",
1042 if (IS_ERR(cd->health_thread)) {
1043 rc = PTR_ERR(cd->health_thread);
1044 cd->health_thread = NULL;
1050 static int genwqe_health_thread_running(struct genwqe_dev *cd)
1052 return cd->health_thread != NULL;
1055 static int genwqe_health_check_stop(struct genwqe_dev *cd)
1057 if (!genwqe_health_thread_running(cd))
1060 kthread_stop(cd->health_thread);
1061 cd->health_thread = NULL;
1066 * genwqe_pci_setup() - Allocate PCIe related resources for our card
1067 * @cd: GenWQE device information
1069 static int genwqe_pci_setup(struct genwqe_dev *cd)
1072 struct pci_dev *pci_dev = cd->pci_dev;
1074 err = pci_enable_device_mem(pci_dev);
1076 dev_err(&pci_dev->dev,
1077 "err: failed to enable pci memory (err=%d)\n", err);
1081 /* Reserve PCI I/O and memory resources */
1082 err = pci_request_mem_regions(pci_dev, genwqe_driver_name);
1084 dev_err(&pci_dev->dev,
1085 "[%s] err: request bars failed (%d)\n", __func__, err);
1087 goto err_disable_device;
1090 /* check for 64-bit DMA address supported (DAC) */
1091 /* check for 32-bit DMA address supported (SAC) */
1092 if (dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64)) &&
1093 dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(32))) {
1094 dev_err(&pci_dev->dev,
1095 "err: neither DMA32 nor DMA64 supported\n");
1097 goto out_release_resources;
1100 pci_set_master(pci_dev);
1102 /* EEH recovery requires PCIe fundamental reset */
1103 pci_dev->needs_freset = 1;
1105 /* request complete BAR-0 space (length = 0) */
1106 cd->mmio_len = pci_resource_len(pci_dev, 0);
1107 cd->mmio = pci_iomap(pci_dev, 0, 0);
1108 if (cd->mmio == NULL) {
1109 dev_err(&pci_dev->dev,
1110 "[%s] err: mapping BAR0 failed\n", __func__);
1112 goto out_release_resources;
1115 cd->num_vfs = pci_sriov_get_totalvfs(pci_dev);
1116 if (cd->num_vfs < 0)
1119 err = genwqe_read_ids(cd);
1126 pci_iounmap(pci_dev, cd->mmio);
1127 out_release_resources:
1128 pci_release_mem_regions(pci_dev);
1130 pci_disable_device(pci_dev);
1136 * genwqe_pci_remove() - Free PCIe related resources for our card
1137 * @cd: GenWQE device information
1139 static void genwqe_pci_remove(struct genwqe_dev *cd)
1141 struct pci_dev *pci_dev = cd->pci_dev;
1144 pci_iounmap(pci_dev, cd->mmio);
1146 pci_release_mem_regions(pci_dev);
1147 pci_disable_device(pci_dev);
1151 * genwqe_probe() - Device initialization
1152 * @pci_dev: PCI device information struct
1153 * @id: PCI device ID
1155 * Callable for multiple cards. This function is called on bind.
1157 * Return: 0 if succeeded, < 0 when failed
1159 static int genwqe_probe(struct pci_dev *pci_dev,
1160 const struct pci_device_id *id)
1163 struct genwqe_dev *cd;
1165 genwqe_init_crc32();
1167 cd = genwqe_dev_alloc();
1169 dev_err(&pci_dev->dev, "err: could not alloc mem (err=%d)!\n",
1174 dev_set_drvdata(&pci_dev->dev, cd);
1175 cd->pci_dev = pci_dev;
1177 err = genwqe_pci_setup(cd);
1179 dev_err(&pci_dev->dev,
1180 "err: problems with PCI setup (err=%d)\n", err);
1184 err = genwqe_start(cd);
1186 dev_err(&pci_dev->dev,
1187 "err: cannot start card services! (err=%d)\n", err);
1188 goto out_pci_remove;
1191 if (genwqe_is_privileged(cd)) {
1192 err = genwqe_health_check_start(cd);
1194 dev_err(&pci_dev->dev,
1195 "err: cannot start health checking! (err=%d)\n",
1197 goto out_stop_services;
1205 genwqe_pci_remove(cd);
1207 genwqe_dev_free(cd);
1212 * genwqe_remove() - Called when device is removed (hot-plugable)
1213 * @pci_dev: PCI device information struct
1215 * Or when driver is unloaded respecitively when unbind is done.
1217 static void genwqe_remove(struct pci_dev *pci_dev)
1219 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1221 genwqe_health_check_stop(cd);
1224 * genwqe_stop() must survive if it is called twice
1225 * sequentially. This happens when the health thread calls it
1226 * and fails on genwqe_bus_reset().
1229 genwqe_pci_remove(cd);
1230 genwqe_dev_free(cd);
1234 * genwqe_err_error_detected() - Error detection callback
1235 * @pci_dev: PCI device information struct
1236 * @state: PCI channel state
1238 * This callback is called by the PCI subsystem whenever a PCI bus
1239 * error is detected.
1241 static pci_ers_result_t genwqe_err_error_detected(struct pci_dev *pci_dev,
1242 pci_channel_state_t state)
1244 struct genwqe_dev *cd;
1246 dev_err(&pci_dev->dev, "[%s] state=%d\n", __func__, state);
1248 cd = dev_get_drvdata(&pci_dev->dev);
1250 return PCI_ERS_RESULT_DISCONNECT;
1253 genwqe_health_check_stop(cd);
1257 * On permanent failure, the PCI code will call device remove
1258 * after the return of this function.
1259 * genwqe_stop() can be called twice.
1261 if (state == pci_channel_io_perm_failure) {
1262 return PCI_ERS_RESULT_DISCONNECT;
1264 genwqe_pci_remove(cd);
1265 return PCI_ERS_RESULT_NEED_RESET;
1269 static pci_ers_result_t genwqe_err_slot_reset(struct pci_dev *pci_dev)
1272 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1274 rc = genwqe_pci_setup(cd);
1276 return PCI_ERS_RESULT_RECOVERED;
1278 dev_err(&pci_dev->dev,
1279 "err: problems with PCI setup (err=%d)\n", rc);
1280 return PCI_ERS_RESULT_DISCONNECT;
1284 static pci_ers_result_t genwqe_err_result_none(struct pci_dev *dev)
1286 return PCI_ERS_RESULT_NONE;
1289 static void genwqe_err_resume(struct pci_dev *pci_dev)
1292 struct genwqe_dev *cd = dev_get_drvdata(&pci_dev->dev);
1294 rc = genwqe_start(cd);
1296 rc = genwqe_health_check_start(cd);
1298 dev_err(&pci_dev->dev,
1299 "err: cannot start health checking! (err=%d)\n",
1302 dev_err(&pci_dev->dev,
1303 "err: cannot start card services! (err=%d)\n", rc);
1307 static int genwqe_sriov_configure(struct pci_dev *dev, int numvfs)
1310 struct genwqe_dev *cd = dev_get_drvdata(&dev->dev);
1313 genwqe_setup_vf_jtimer(cd);
1314 rc = pci_enable_sriov(dev, numvfs);
1320 pci_disable_sriov(dev);
1326 static const struct pci_error_handlers genwqe_err_handler = {
1327 .error_detected = genwqe_err_error_detected,
1328 .mmio_enabled = genwqe_err_result_none,
1329 .slot_reset = genwqe_err_slot_reset,
1330 .resume = genwqe_err_resume,
1333 static struct pci_driver genwqe_driver = {
1334 .name = genwqe_driver_name,
1335 .id_table = genwqe_device_table,
1336 .probe = genwqe_probe,
1337 .remove = genwqe_remove,
1338 .sriov_configure = genwqe_sriov_configure,
1339 .err_handler = &genwqe_err_handler,
1343 * genwqe_devnode() - Set default access mode for genwqe devices.
1344 * @dev: Pointer to device (unused)
1345 * @mode: Carrier to pass-back given mode (permissions)
1347 * Default mode should be rw for everybody. Do not change default
1350 static char *genwqe_devnode(const struct device *dev, umode_t *mode)
1358 * genwqe_init_module() - Driver registration and initialization
1360 static int __init genwqe_init_module(void)
1364 class_genwqe = class_create(GENWQE_DEVNAME);
1365 if (IS_ERR(class_genwqe)) {
1366 pr_err("[%s] create class failed\n", __func__);
1370 class_genwqe->devnode = genwqe_devnode;
1372 debugfs_genwqe = debugfs_create_dir(GENWQE_DEVNAME, NULL);
1374 rc = pci_register_driver(&genwqe_driver);
1376 pr_err("[%s] pci_reg_driver (rc=%d)\n", __func__, rc);
1383 debugfs_remove(debugfs_genwqe);
1384 class_destroy(class_genwqe);
1389 * genwqe_exit_module() - Driver exit
1391 static void __exit genwqe_exit_module(void)
1393 pci_unregister_driver(&genwqe_driver);
1394 debugfs_remove(debugfs_genwqe);
1395 class_destroy(class_genwqe);
1398 module_init(genwqe_init_module);
1399 module_exit(genwqe_exit_module);