1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * The file intends to implement the platform dependent EEH operations on pseries.
4 * Actually, the pseries platform is built based on RTAS heavily. That means the
5 * pseries platform dependent EEH operations will be built on RTAS calls. The functions
6 * are derived from arch/powerpc/platforms/pseries/eeh.c and necessary cleanup has
7 * been done.
8 *
9 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2011.
10 * Copyright IBM Corporation 2001, 2005, 2006
11 * Copyright Dave Engebretsen & Todd Inglett 2001
12 * Copyright Linas Vepstas 2005, 2006
13 */
14
15#include <linux/atomic.h>
16#include <linux/delay.h>
17#include <linux/export.h>
18#include <linux/init.h>
19#include <linux/list.h>
20#include <linux/of.h>
21#include <linux/pci.h>
22#include <linux/proc_fs.h>
23#include <linux/rbtree.h>
24#include <linux/sched.h>
25#include <linux/seq_file.h>
26#include <linux/spinlock.h>
27#include <linux/crash_dump.h>
28
29#include <asm/eeh.h>
30#include <asm/eeh_event.h>
31#include <asm/io.h>
32#include <asm/machdep.h>
33#include <asm/ppc-pci.h>
34#include <asm/rtas.h>
35
36/* RTAS tokens */
37static int ibm_set_eeh_option;
38static int ibm_set_slot_reset;
39static int ibm_read_slot_reset_state;
40static int ibm_read_slot_reset_state2;
41static int ibm_slot_error_detail;
42static int ibm_get_config_addr_info;
43static int ibm_get_config_addr_info2;
44static int ibm_configure_pe;
45
46static void pseries_eeh_init_edev(struct pci_dn *pdn);
47
48static void pseries_pcibios_bus_add_device(struct pci_dev *pdev)
49{
50 struct pci_dn *pdn = pci_get_pdn(pdev);
51
52 if (eeh_has_flag(EEH_FORCE_DISABLED))
53 return;
54
55 dev_dbg(&pdev->dev, "EEH: Setting up device\n");
56#ifdef CONFIG_PCI_IOV
57 if (pdev->is_virtfn) {
58 pdn->device_id = pdev->device;
59 pdn->vendor_id = pdev->vendor;
60 pdn->class_code = pdev->class;
61 /*
62 * Last allow unfreeze return code used for retrieval
63 * by user space in eeh-sysfs to show the last command
64 * completion from platform.
65 */
66 pdn->last_allow_rc = 0;
67 }
68#endif
69 pseries_eeh_init_edev(pdn);
70#ifdef CONFIG_PCI_IOV
71 if (pdev->is_virtfn) {
72 /*
73 * FIXME: This really should be handled by choosing the right
74 * parent PE in pseries_eeh_init_edev().
75 */
76 struct eeh_pe *physfn_pe = pci_dev_to_eeh_dev(pdev->physfn)->pe;
77 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
78
79 edev->pe_config_addr = (pdn->busno << 16) | (pdn->devfn << 8);
80 eeh_pe_tree_remove(edev); /* Remove as it is adding to bus pe */
81 eeh_pe_tree_insert(edev, physfn_pe); /* Add as VF PE type */
82 }
83#endif
84 eeh_probe_device(pdev);
85}
86
87
88/**
89 * pseries_eeh_get_pe_config_addr - Find the pe_config_addr for a device
90 * @pdn: pci_dn of the input device
91 *
92 * The EEH RTAS calls use a tuple consisting of: (buid_hi, buid_lo,
93 * pe_config_addr) as a handle to a given PE. This function finds the
94 * pe_config_addr based on the device's config addr.
95 *
96 * Keep in mind that the pe_config_addr *might* be numerically identical to the
97 * device's config addr, but the two are conceptually distinct.
98 *
99 * Returns the pe_config_addr, or a negative error code.
100 */
101static int pseries_eeh_get_pe_config_addr(struct pci_dn *pdn)
102{
103 int config_addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
104 struct pci_controller *phb = pdn->phb;
105 int ret, rets[3];
106
107 if (ibm_get_config_addr_info2 != RTAS_UNKNOWN_SERVICE) {
108 /*
109 * First of all, use function 1 to determine if this device is
110 * part of a PE or not. ret[0] being zero indicates it's not.
111 */
112 ret = rtas_call(ibm_get_config_addr_info2, 4, 2, rets,
113 config_addr, BUID_HI(phb->buid),
114 BUID_LO(phb->buid), 1);
115 if (ret || (rets[0] == 0))
116 return -ENOENT;
117
118 /* Retrieve the associated PE config address with function 0 */
119 ret = rtas_call(ibm_get_config_addr_info2, 4, 2, rets,
120 config_addr, BUID_HI(phb->buid),
121 BUID_LO(phb->buid), 0);
122 if (ret) {
123 pr_warn("%s: Failed to get address for PHB#%x-PE#%x\n",
124 __func__, phb->global_number, config_addr);
125 return -ENXIO;
126 }
127
128 return rets[0];
129 }
130
131 if (ibm_get_config_addr_info != RTAS_UNKNOWN_SERVICE) {
132 ret = rtas_call(ibm_get_config_addr_info, 4, 2, rets,
133 config_addr, BUID_HI(phb->buid),
134 BUID_LO(phb->buid), 0);
135 if (ret) {
136 pr_warn("%s: Failed to get address for PHB#%x-PE#%x\n",
137 __func__, phb->global_number, config_addr);
138 return -ENXIO;
139 }
140
141 return rets[0];
142 }
143
144 /*
145 * PAPR does describe a process for finding the pe_config_addr that was
146 * used before the ibm,get-config-addr-info calls were added. However,
147 * I haven't found *any* systems that don't have that RTAS call
148 * implemented. If you happen to find one that needs the old DT based
149 * process, patches are welcome!
150 */
151 return -ENOENT;
152}
153
154/**
155 * pseries_eeh_phb_reset - Reset the specified PHB
156 * @phb: PCI controller
157 * @config_addr: the associated config address
158 * @option: reset option
159 *
160 * Reset the specified PHB/PE
161 */
162static int pseries_eeh_phb_reset(struct pci_controller *phb, int config_addr, int option)
163{
164 int ret;
165
166 /* Reset PE through RTAS call */
167 ret = rtas_call(ibm_set_slot_reset, 4, 1, NULL,
168 config_addr, BUID_HI(phb->buid),
169 BUID_LO(phb->buid), option);
170
171 /* If fundamental-reset not supported, try hot-reset */
172 if (option == EEH_RESET_FUNDAMENTAL && ret == -8) {
173 option = EEH_RESET_HOT;
174 ret = rtas_call(ibm_set_slot_reset, 4, 1, NULL,
175 config_addr, BUID_HI(phb->buid),
176 BUID_LO(phb->buid), option);
177 }
178
179 /* We need reset hold or settlement delay */
180 if (option == EEH_RESET_FUNDAMENTAL || option == EEH_RESET_HOT)
181 msleep(msecs: EEH_PE_RST_HOLD_TIME);
182 else
183 msleep(msecs: EEH_PE_RST_SETTLE_TIME);
184
185 return ret;
186}
187
188/**
189 * pseries_eeh_phb_configure_bridge - Configure PCI bridges in the indicated PE
190 * @phb: PCI controller
191 * @config_addr: the associated config address
192 *
193 * The function will be called to reconfigure the bridges included
194 * in the specified PE so that the mulfunctional PE would be recovered
195 * again.
196 */
197static int pseries_eeh_phb_configure_bridge(struct pci_controller *phb, int config_addr)
198{
199 int ret;
200 /* Waiting 0.2s maximum before skipping configuration */
201 int max_wait = 200;
202
203 while (max_wait > 0) {
204 ret = rtas_call(ibm_configure_pe, 3, 1, NULL,
205 config_addr, BUID_HI(phb->buid),
206 BUID_LO(phb->buid));
207
208 if (!ret)
209 return ret;
210 if (ret < 0)
211 break;
212
213 /*
214 * If RTAS returns a delay value that's above 100ms, cut it
215 * down to 100ms in case firmware made a mistake. For more
216 * on how these delay values work see rtas_busy_delay_time
217 */
218 if (ret > RTAS_EXTENDED_DELAY_MIN+2 &&
219 ret <= RTAS_EXTENDED_DELAY_MAX)
220 ret = RTAS_EXTENDED_DELAY_MIN+2;
221
222 max_wait -= rtas_busy_delay_time(ret);
223
224 if (max_wait < 0)
225 break;
226
227 rtas_busy_delay(ret);
228 }
229
230 pr_warn("%s: Unable to configure bridge PHB#%x-PE#%x (%d)\n",
231 __func__, phb->global_number, config_addr, ret);
232 /* PAPR defines -3 as "Parameter Error" for this function: */
233 if (ret == -3)
234 return -EINVAL;
235 else
236 return -EIO;
237}
238
239/*
240 * Buffer for reporting slot-error-detail rtas calls. Its here
241 * in BSS, and not dynamically alloced, so that it ends up in
242 * RMO where RTAS can access it.
243 */
244static unsigned char slot_errbuf[RTAS_ERROR_LOG_MAX];
245static DEFINE_SPINLOCK(slot_errbuf_lock);
246static int eeh_error_buf_size;
247
248static int pseries_eeh_cap_start(struct pci_dn *pdn)
249{
250 u32 status;
251
252 if (!pdn)
253 return 0;
254
255 rtas_pci_dn_read_config(pdn, PCI_STATUS, 2, &status);
256 if (!(status & PCI_STATUS_CAP_LIST))
257 return 0;
258
259 return PCI_CAPABILITY_LIST;
260}
261
262
263static int pseries_eeh_find_cap(struct pci_dn *pdn, int cap)
264{
265 int pos = pseries_eeh_cap_start(pdn);
266 int cnt = 48; /* Maximal number of capabilities */
267 u32 id;
268
269 if (!pos)
270 return 0;
271
272 while (cnt--) {
273 rtas_pci_dn_read_config(pdn, pos, 1, &pos);
274 if (pos < 0x40)
275 break;
276 pos &= ~3;
277 rtas_pci_dn_read_config(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
278 if (id == 0xff)
279 break;
280 if (id == cap)
281 return pos;
282 pos += PCI_CAP_LIST_NEXT;
283 }
284
285 return 0;
286}
287
288static int pseries_eeh_find_ecap(struct pci_dn *pdn, int cap)
289{
290 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
291 u32 header;
292 int pos = 256;
293 int ttl = (4096 - 256) / 8;
294
295 if (!edev || !edev->pcie_cap)
296 return 0;
297 if (rtas_pci_dn_read_config(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
298 return 0;
299 else if (!header)
300 return 0;
301
302 while (ttl-- > 0) {
303 if (PCI_EXT_CAP_ID(header) == cap && pos)
304 return pos;
305
306 pos = PCI_EXT_CAP_NEXT(header);
307 if (pos < 256)
308 break;
309
310 if (rtas_pci_dn_read_config(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
311 break;
312 }
313
314 return 0;
315}
316
317/**
318 * pseries_eeh_pe_get_parent - Retrieve the parent PE
319 * @edev: EEH device
320 *
321 * The whole PEs existing in the system are organized as hierarchy
322 * tree. The function is used to retrieve the parent PE according
323 * to the parent EEH device.
324 */
325static struct eeh_pe *pseries_eeh_pe_get_parent(struct eeh_dev *edev)
326{
327 struct eeh_dev *parent;
328 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
329
330 /*
331 * It might have the case for the indirect parent
332 * EEH device already having associated PE, but
333 * the direct parent EEH device doesn't have yet.
334 */
335 if (edev->physfn)
336 pdn = pci_get_pdn(edev->physfn);
337 else
338 pdn = pdn ? pdn->parent : NULL;
339 while (pdn) {
340 /* We're poking out of PCI territory */
341 parent = pdn_to_eeh_dev(pdn);
342 if (!parent)
343 return NULL;
344
345 if (parent->pe)
346 return parent->pe;
347
348 pdn = pdn->parent;
349 }
350
351 return NULL;
352}
353
354/**
355 * pseries_eeh_init_edev - initialise the eeh_dev and eeh_pe for a pci_dn
356 *
357 * @pdn: PCI device node
358 *
359 * When we discover a new PCI device via the device-tree we create a
360 * corresponding pci_dn and we allocate, but don't initialise, an eeh_dev.
361 * This function takes care of the initialisation and inserts the eeh_dev
362 * into the correct eeh_pe. If no eeh_pe exists we'll allocate one.
363 */
364static void pseries_eeh_init_edev(struct pci_dn *pdn)
365{
366 struct eeh_pe pe, *parent;
367 struct eeh_dev *edev;
368 u32 pcie_flags;
369 int ret;
370
371 if (WARN_ON_ONCE(!eeh_has_flag(EEH_PROBE_MODE_DEVTREE)))
372 return;
373
374 /*
375 * Find the eeh_dev for this pdn. The storage for the eeh_dev was
376 * allocated at the same time as the pci_dn.
377 *
378 * XXX: We should probably re-visit that.
379 */
380 edev = pdn_to_eeh_dev(pdn);
381 if (!edev)
382 return;
383
384 /*
385 * If ->pe is set then we've already probed this device. We hit
386 * this path when a pci_dev is removed and rescanned while recovering
387 * a PE (i.e. for devices where the driver doesn't support error
388 * recovery).
389 */
390 if (edev->pe)
391 return;
392
393 /* Check class/vendor/device IDs */
394 if (!pdn->vendor_id || !pdn->device_id || !pdn->class_code)
395 return;
396
397 /* Skip for PCI-ISA bridge */
398 if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
399 return;
400
401 eeh_edev_dbg(edev, "Probing device\n");
402
403 /*
404 * Update class code and mode of eeh device. We need
405 * correctly reflects that current device is root port
406 * or PCIe switch downstream port.
407 */
408 edev->pcix_cap = pseries_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
409 edev->pcie_cap = pseries_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
410 edev->aer_cap = pseries_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
411 edev->mode &= 0xFFFFFF00;
412 if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
413 edev->mode |= EEH_DEV_BRIDGE;
414 if (edev->pcie_cap) {
415 rtas_pci_dn_read_config(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
416 2, &pcie_flags);
417 pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
418 if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
419 edev->mode |= EEH_DEV_ROOT_PORT;
420 else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
421 edev->mode |= EEH_DEV_DS_PORT;
422 }
423 }
424
425 /* first up, find the pe_config_addr for the PE containing the device */
426 ret = pseries_eeh_get_pe_config_addr(pdn);
427 if (ret < 0) {
428 eeh_edev_dbg(edev, "Unable to find pe_config_addr\n");
429 goto err;
430 }
431
432 /* Try enable EEH on the fake PE */
433 memset(&pe, 0, sizeof(struct eeh_pe));
434 pe.phb = pdn->phb;
435 pe.addr = ret;
436
437 eeh_edev_dbg(edev, "Enabling EEH on device\n");
438 ret = eeh_ops->set_option(&pe, EEH_OPT_ENABLE);
439 if (ret) {
440 eeh_edev_dbg(edev, "EEH failed to enable on device (code %d)\n", ret);
441 goto err;
442 }
443
444 edev->pe_config_addr = pe.addr;
445
446 eeh_add_flag(EEH_ENABLED);
447
448 parent = pseries_eeh_pe_get_parent(edev);
449 eeh_pe_tree_insert(edev, parent);
450 eeh_save_bars(edev);
451 eeh_edev_dbg(edev, "EEH enabled for device");
452
453 return;
454
455err:
456 eeh_edev_dbg(edev, "EEH is unsupported on device (code = %d)\n", ret);
457}
458
459static struct eeh_dev *pseries_eeh_probe(struct pci_dev *pdev)
460{
461 struct eeh_dev *edev;
462 struct pci_dn *pdn;
463
464 pdn = pci_get_pdn_by_devfn(pdev->bus, pdev->devfn);
465 if (!pdn)
466 return NULL;
467
468 /*
469 * If the system supports EEH on this device then the eeh_dev was
470 * configured and inserted into a PE in pseries_eeh_init_edev()
471 */
472 edev = pdn_to_eeh_dev(pdn);
473 if (!edev || !edev->pe)
474 return NULL;
475
476 return edev;
477}
478
479/**
480 * pseries_eeh_init_edev_recursive - Enable EEH for the indicated device
481 * @pdn: PCI device node
482 *
483 * This routine must be used to perform EEH initialization for the
484 * indicated PCI device that was added after system boot (e.g.
485 * hotplug, dlpar).
486 */
487void pseries_eeh_init_edev_recursive(struct pci_dn *pdn)
488{
489 struct pci_dn *n;
490
491 if (!pdn)
492 return;
493
494 list_for_each_entry(n, &pdn->child_list, list)
495 pseries_eeh_init_edev_recursive(pdn: n);
496
497 pseries_eeh_init_edev(pdn);
498}
499EXPORT_SYMBOL_GPL(pseries_eeh_init_edev_recursive);
500
501/**
502 * pseries_eeh_set_option - Initialize EEH or MMIO/DMA reenable
503 * @pe: EEH PE
504 * @option: operation to be issued
505 *
506 * The function is used to control the EEH functionality globally.
507 * Currently, following options are support according to PAPR:
508 * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
509 */
510static int pseries_eeh_set_option(struct eeh_pe *pe, int option)
511{
512 int ret = 0;
513
514 /*
515 * When we're enabling or disabling EEH functionality on
516 * the particular PE, the PE config address is possibly
517 * unavailable. Therefore, we have to figure it out from
518 * the FDT node.
519 */
520 switch (option) {
521 case EEH_OPT_DISABLE:
522 case EEH_OPT_ENABLE:
523 case EEH_OPT_THAW_MMIO:
524 case EEH_OPT_THAW_DMA:
525 break;
526 case EEH_OPT_FREEZE_PE:
527 /* Not support */
528 return 0;
529 default:
530 pr_err("%s: Invalid option %d\n", __func__, option);
531 return -EINVAL;
532 }
533
534 ret = rtas_call(ibm_set_eeh_option, 4, 1, NULL,
535 pe->addr, BUID_HI(pe->phb->buid),
536 BUID_LO(pe->phb->buid), option);
537
538 return ret;
539}
540
541/**
542 * pseries_eeh_get_state - Retrieve PE state
543 * @pe: EEH PE
544 * @delay: suggested time to wait if state is unavailable
545 *
546 * Retrieve the state of the specified PE. On RTAS compliant
547 * pseries platform, there already has one dedicated RTAS function
548 * for the purpose. It's notable that the associated PE config address
549 * might be ready when calling the function. Therefore, endeavour to
550 * use the PE config address if possible. Further more, there're 2
551 * RTAS calls for the purpose, we need to try the new one and back
552 * to the old one if the new one couldn't work properly.
553 */
554static int pseries_eeh_get_state(struct eeh_pe *pe, int *delay)
555{
556 int ret;
557 int rets[4];
558 int result;
559
560 if (ibm_read_slot_reset_state2 != RTAS_UNKNOWN_SERVICE) {
561 ret = rtas_call(ibm_read_slot_reset_state2, 3, 4, rets,
562 pe->addr, BUID_HI(pe->phb->buid),
563 BUID_LO(pe->phb->buid));
564 } else if (ibm_read_slot_reset_state != RTAS_UNKNOWN_SERVICE) {
565 /* Fake PE unavailable info */
566 rets[2] = 0;
567 ret = rtas_call(ibm_read_slot_reset_state, 3, 3, rets,
568 pe->addr, BUID_HI(pe->phb->buid),
569 BUID_LO(pe->phb->buid));
570 } else {
571 return EEH_STATE_NOT_SUPPORT;
572 }
573
574 if (ret)
575 return ret;
576
577 /* Parse the result out */
578 if (!rets[1])
579 return EEH_STATE_NOT_SUPPORT;
580
581 switch(rets[0]) {
582 case 0:
583 result = EEH_STATE_MMIO_ACTIVE |
584 EEH_STATE_DMA_ACTIVE;
585 break;
586 case 1:
587 result = EEH_STATE_RESET_ACTIVE |
588 EEH_STATE_MMIO_ACTIVE |
589 EEH_STATE_DMA_ACTIVE;
590 break;
591 case 2:
592 result = 0;
593 break;
594 case 4:
595 result = EEH_STATE_MMIO_ENABLED;
596 break;
597 case 5:
598 if (rets[2]) {
599 if (delay)
600 *delay = rets[2];
601 result = EEH_STATE_UNAVAILABLE;
602 } else {
603 result = EEH_STATE_NOT_SUPPORT;
604 }
605 break;
606 default:
607 result = EEH_STATE_NOT_SUPPORT;
608 }
609
610 return result;
611}
612
613/**
614 * pseries_eeh_reset - Reset the specified PE
615 * @pe: EEH PE
616 * @option: reset option
617 *
618 * Reset the specified PE
619 */
620static int pseries_eeh_reset(struct eeh_pe *pe, int option)
621{
622 return pseries_eeh_phb_reset(phb: pe->phb, config_addr: pe->addr, option);
623}
624
625/**
626 * pseries_eeh_get_log - Retrieve error log
627 * @pe: EEH PE
628 * @severity: temporary or permanent error log
629 * @drv_log: driver log to be combined with retrieved error log
630 * @len: length of driver log
631 *
632 * Retrieve the temporary or permanent error from the PE.
633 * Actually, the error will be retrieved through the dedicated
634 * RTAS call.
635 */
636static int pseries_eeh_get_log(struct eeh_pe *pe, int severity, char *drv_log, unsigned long len)
637{
638 unsigned long flags;
639 int ret;
640
641 spin_lock_irqsave(&slot_errbuf_lock, flags);
642 memset(slot_errbuf, 0, eeh_error_buf_size);
643
644 ret = rtas_call(ibm_slot_error_detail, 8, 1, NULL, pe->addr,
645 BUID_HI(pe->phb->buid), BUID_LO(pe->phb->buid),
646 virt_to_phys(address: drv_log), len,
647 virt_to_phys(address: slot_errbuf), eeh_error_buf_size,
648 severity);
649 if (!ret)
650 log_error(slot_errbuf, ERR_TYPE_RTAS_LOG, 0);
651 spin_unlock_irqrestore(lock: &slot_errbuf_lock, flags);
652
653 return ret;
654}
655
656/**
657 * pseries_eeh_configure_bridge - Configure PCI bridges in the indicated PE
658 * @pe: EEH PE
659 *
660 */
661static int pseries_eeh_configure_bridge(struct eeh_pe *pe)
662{
663 return pseries_eeh_phb_configure_bridge(phb: pe->phb, config_addr: pe->addr);
664}
665
666/**
667 * pseries_eeh_read_config - Read PCI config space
668 * @edev: EEH device handle
669 * @where: PCI config space offset
670 * @size: size to read
671 * @val: return value
672 *
673 * Read config space from the speicifed device
674 */
675static int pseries_eeh_read_config(struct eeh_dev *edev, int where, int size, u32 *val)
676{
677 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
678
679 return rtas_pci_dn_read_config(pdn, where, size, val);
680}
681
682/**
683 * pseries_eeh_write_config - Write PCI config space
684 * @edev: EEH device handle
685 * @where: PCI config space offset
686 * @size: size to write
687 * @val: value to be written
688 *
689 * Write config space to the specified device
690 */
691static int pseries_eeh_write_config(struct eeh_dev *edev, int where, int size, u32 val)
692{
693 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
694
695 return rtas_pci_dn_write_config(pdn, where, size, val);
696}
697
698#ifdef CONFIG_PCI_IOV
699static int pseries_send_allow_unfreeze(struct pci_dn *pdn, u16 *vf_pe_array, int cur_vfs)
700{
701 int rc;
702 int ibm_allow_unfreeze = rtas_function_token(RTAS_FN_IBM_OPEN_SRIOV_ALLOW_UNFREEZE);
703 unsigned long buid, addr;
704
705 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
706 buid = pdn->phb->buid;
707 spin_lock(&rtas_data_buf_lock);
708 memcpy(rtas_data_buf, vf_pe_array, RTAS_DATA_BUF_SIZE);
709 rc = rtas_call(ibm_allow_unfreeze, 5, 1, NULL,
710 addr,
711 BUID_HI(buid),
712 BUID_LO(buid),
713 rtas_data_buf, cur_vfs * sizeof(u16));
714 spin_unlock(&rtas_data_buf_lock);
715 if (rc)
716 pr_warn("%s: Failed to allow unfreeze for PHB#%x-PE#%lx, rc=%x\n",
717 __func__,
718 pdn->phb->global_number, addr, rc);
719 return rc;
720}
721
722static int pseries_call_allow_unfreeze(struct eeh_dev *edev)
723{
724 int cur_vfs = 0, rc = 0, vf_index, bus, devfn, vf_pe_num;
725 struct pci_dn *pdn, *tmp, *parent, *physfn_pdn;
726 u16 *vf_pe_array;
727
728 vf_pe_array = kzalloc(RTAS_DATA_BUF_SIZE, GFP_KERNEL);
729 if (!vf_pe_array)
730 return -ENOMEM;
731 if (pci_num_vf(dev: edev->physfn ? edev->physfn : edev->pdev)) {
732 if (edev->pdev->is_physfn) {
733 cur_vfs = pci_num_vf(dev: edev->pdev);
734 pdn = eeh_dev_to_pdn(edev);
735 parent = pdn->parent;
736 for (vf_index = 0; vf_index < cur_vfs; vf_index++)
737 vf_pe_array[vf_index] =
738 cpu_to_be16(pdn->pe_num_map[vf_index]);
739 rc = pseries_send_allow_unfreeze(pdn, vf_pe_array,
740 cur_vfs);
741 pdn->last_allow_rc = rc;
742 for (vf_index = 0; vf_index < cur_vfs; vf_index++) {
743 list_for_each_entry_safe(pdn, tmp,
744 &parent->child_list,
745 list) {
746 bus = pci_iov_virtfn_bus(dev: edev->pdev,
747 id: vf_index);
748 devfn = pci_iov_virtfn_devfn(dev: edev->pdev,
749 id: vf_index);
750 if (pdn->busno != bus ||
751 pdn->devfn != devfn)
752 continue;
753 pdn->last_allow_rc = rc;
754 }
755 }
756 } else {
757 pdn = pci_get_pdn(edev->pdev);
758 physfn_pdn = pci_get_pdn(edev->physfn);
759
760 vf_pe_num = physfn_pdn->pe_num_map[edev->vf_index];
761 vf_pe_array[0] = cpu_to_be16(vf_pe_num);
762 rc = pseries_send_allow_unfreeze(pdn: physfn_pdn,
763 vf_pe_array, cur_vfs: 1);
764 pdn->last_allow_rc = rc;
765 }
766 }
767
768 kfree(objp: vf_pe_array);
769 return rc;
770}
771
772static int pseries_notify_resume(struct eeh_dev *edev)
773{
774 if (!edev)
775 return -EEXIST;
776
777 if (rtas_function_token(RTAS_FN_IBM_OPEN_SRIOV_ALLOW_UNFREEZE) == RTAS_UNKNOWN_SERVICE)
778 return -EINVAL;
779
780 if (edev->pdev->is_physfn || edev->pdev->is_virtfn)
781 return pseries_call_allow_unfreeze(edev);
782
783 return 0;
784}
785#endif
786
787static struct eeh_ops pseries_eeh_ops = {
788 .name = "pseries",
789 .probe = pseries_eeh_probe,
790 .set_option = pseries_eeh_set_option,
791 .get_state = pseries_eeh_get_state,
792 .reset = pseries_eeh_reset,
793 .get_log = pseries_eeh_get_log,
794 .configure_bridge = pseries_eeh_configure_bridge,
795 .err_inject = NULL,
796 .read_config = pseries_eeh_read_config,
797 .write_config = pseries_eeh_write_config,
798 .next_error = NULL,
799 .restore_config = NULL, /* NB: configure_bridge() does this */
800#ifdef CONFIG_PCI_IOV
801 .notify_resume = pseries_notify_resume
802#endif
803};
804
805/**
806 * eeh_pseries_init - Register platform dependent EEH operations
807 *
808 * EEH initialization on pseries platform. This function should be
809 * called before any EEH related functions.
810 */
811static int __init eeh_pseries_init(void)
812{
813 struct pci_controller *phb;
814 struct pci_dn *pdn;
815 int ret, config_addr;
816
817 /* figure out EEH RTAS function call tokens */
818 ibm_set_eeh_option = rtas_function_token(RTAS_FN_IBM_SET_EEH_OPTION);
819 ibm_set_slot_reset = rtas_function_token(RTAS_FN_IBM_SET_SLOT_RESET);
820 ibm_read_slot_reset_state2 = rtas_function_token(RTAS_FN_IBM_READ_SLOT_RESET_STATE2);
821 ibm_read_slot_reset_state = rtas_function_token(RTAS_FN_IBM_READ_SLOT_RESET_STATE);
822 ibm_slot_error_detail = rtas_function_token(RTAS_FN_IBM_SLOT_ERROR_DETAIL);
823 ibm_get_config_addr_info2 = rtas_function_token(RTAS_FN_IBM_GET_CONFIG_ADDR_INFO2);
824 ibm_get_config_addr_info = rtas_function_token(RTAS_FN_IBM_GET_CONFIG_ADDR_INFO);
825 ibm_configure_pe = rtas_function_token(RTAS_FN_IBM_CONFIGURE_PE);
826
827 /*
828 * ibm,configure-pe and ibm,configure-bridge have the same semantics,
829 * however ibm,configure-pe can be faster. If we can't find
830 * ibm,configure-pe then fall back to using ibm,configure-bridge.
831 */
832 if (ibm_configure_pe == RTAS_UNKNOWN_SERVICE)
833 ibm_configure_pe = rtas_function_token(RTAS_FN_IBM_CONFIGURE_BRIDGE);
834
835 /*
836 * Necessary sanity check. We needn't check "get-config-addr-info"
837 * and its variant since the old firmware probably support address
838 * of domain/bus/slot/function for EEH RTAS operations.
839 */
840 if (ibm_set_eeh_option == RTAS_UNKNOWN_SERVICE ||
841 ibm_set_slot_reset == RTAS_UNKNOWN_SERVICE ||
842 (ibm_read_slot_reset_state2 == RTAS_UNKNOWN_SERVICE &&
843 ibm_read_slot_reset_state == RTAS_UNKNOWN_SERVICE) ||
844 ibm_slot_error_detail == RTAS_UNKNOWN_SERVICE ||
845 ibm_configure_pe == RTAS_UNKNOWN_SERVICE) {
846 pr_info("EEH functionality not supported\n");
847 return -EINVAL;
848 }
849
850 /* Initialize error log size */
851 eeh_error_buf_size = rtas_get_error_log_max();
852
853 /* Set EEH probe mode */
854 eeh_add_flag(EEH_PROBE_MODE_DEVTREE | EEH_ENABLE_IO_FOR_LOG);
855
856 /* Set EEH machine dependent code */
857 ppc_md.pcibios_bus_add_device = pseries_pcibios_bus_add_device;
858
859 if (is_kdump_kernel() || reset_devices) {
860 pr_info("Issue PHB reset ...\n");
861 list_for_each_entry(phb, &hose_list, list_node) {
862 // Skip if the slot is empty
863 if (list_empty(&PCI_DN(phb->dn)->child_list))
864 continue;
865
866 pdn = list_first_entry(&PCI_DN(phb->dn)->child_list, struct pci_dn, list);
867 config_addr = pseries_eeh_get_pe_config_addr(pdn);
868
869 /* invalid PE config addr */
870 if (config_addr < 0)
871 continue;
872
873 pseries_eeh_phb_reset(phb, config_addr, EEH_RESET_FUNDAMENTAL);
874 pseries_eeh_phb_reset(phb, config_addr, EEH_RESET_DEACTIVATE);
875 pseries_eeh_phb_configure_bridge(phb, config_addr);
876 }
877 }
878
879 ret = eeh_init(&pseries_eeh_ops);
880 if (!ret)
881 pr_info("EEH: pSeries platform initialized\n");
882 else
883 pr_info("EEH: pSeries platform initialization failure (%d)\n",
884 ret);
885 return ret;
886}
887machine_arch_initcall(pseries, eeh_pseries_init);
888

source code of linux/arch/powerpc/platforms/pseries/eeh_pseries.c