2 * The file intends to implement the functions needed by EEH, which is
3 * built on IODA compliant chip. Actually, lots of functions related
4 * to EEH would be built based on the OPAL APIs.
6 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
14 #include <linux/bootmem.h>
15 #include <linux/debugfs.h>
16 #include <linux/delay.h>
18 #include <linux/irq.h>
19 #include <linux/kernel.h>
20 #include <linux/msi.h>
21 #include <linux/notifier.h>
22 #include <linux/pci.h>
23 #include <linux/string.h>
26 #include <asm/eeh_event.h>
28 #include <asm/iommu.h>
29 #include <asm/msi_bitmap.h>
31 #include <asm/pci-bridge.h>
32 #include <asm/ppc-pci.h>
38 static int ioda_eeh_nb_init = 0;
40 static int ioda_eeh_event(struct notifier_block *nb,
41 unsigned long events, void *change)
43 uint64_t changed_evts = (uint64_t)change;
46 * We simply send special EEH event if EEH has
47 * been enabled, or clear pending events in
48 * case that we enable EEH soon
50 if (!(changed_evts & OPAL_EVENT_PCI_ERROR) ||
51 !(events & OPAL_EVENT_PCI_ERROR))
55 eeh_send_failure_event(NULL);
57 opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
62 static struct notifier_block ioda_eeh_nb = {
63 .notifier_call = ioda_eeh_event,
68 #ifdef CONFIG_DEBUG_FS
69 static int ioda_eeh_dbgfs_set(void *data, int offset, u64 val)
71 struct pci_controller *hose = data;
72 struct pnv_phb *phb = hose->private_data;
74 out_be64(phb->regs + offset, val);
78 static int ioda_eeh_dbgfs_get(void *data, int offset, u64 *val)
80 struct pci_controller *hose = data;
81 struct pnv_phb *phb = hose->private_data;
83 *val = in_be64(phb->regs + offset);
87 static int ioda_eeh_outb_dbgfs_set(void *data, u64 val)
89 return ioda_eeh_dbgfs_set(data, 0xD10, val);
92 static int ioda_eeh_outb_dbgfs_get(void *data, u64 *val)
94 return ioda_eeh_dbgfs_get(data, 0xD10, val);
97 static int ioda_eeh_inbA_dbgfs_set(void *data, u64 val)
99 return ioda_eeh_dbgfs_set(data, 0xD90, val);
102 static int ioda_eeh_inbA_dbgfs_get(void *data, u64 *val)
104 return ioda_eeh_dbgfs_get(data, 0xD90, val);
107 static int ioda_eeh_inbB_dbgfs_set(void *data, u64 val)
109 return ioda_eeh_dbgfs_set(data, 0xE10, val);
112 static int ioda_eeh_inbB_dbgfs_get(void *data, u64 *val)
114 return ioda_eeh_dbgfs_get(data, 0xE10, val);
117 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_outb_dbgfs_ops, ioda_eeh_outb_dbgfs_get,
118 ioda_eeh_outb_dbgfs_set, "0x%llx\n");
119 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbA_dbgfs_ops, ioda_eeh_inbA_dbgfs_get,
120 ioda_eeh_inbA_dbgfs_set, "0x%llx\n");
121 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbB_dbgfs_ops, ioda_eeh_inbB_dbgfs_get,
122 ioda_eeh_inbB_dbgfs_set, "0x%llx\n");
123 #endif /* CONFIG_DEBUG_FS */
127 * ioda_eeh_post_init - Chip dependent post initialization
128 * @hose: PCI controller
130 * The function will be called after eeh PEs and devices
131 * have been built. That means the EEH is ready to supply
132 * service with I/O cache.
134 static int ioda_eeh_post_init(struct pci_controller *hose)
136 struct pnv_phb *phb = hose->private_data;
139 /* Register OPAL event notifier */
140 if (!ioda_eeh_nb_init) {
141 ret = opal_notifier_register(&ioda_eeh_nb);
143 pr_err("%s: Can't register OPAL event notifier (%d)\n",
148 ioda_eeh_nb_init = 1;
151 #ifdef CONFIG_DEBUG_FS
152 if (!phb->has_dbgfs && phb->dbgfs) {
155 debugfs_create_file("err_injct_outbound", 0600,
157 &ioda_eeh_outb_dbgfs_ops);
158 debugfs_create_file("err_injct_inboundA", 0600,
160 &ioda_eeh_inbA_dbgfs_ops);
161 debugfs_create_file("err_injct_inboundB", 0600,
163 &ioda_eeh_inbB_dbgfs_ops);
167 /* If EEH is enabled, we're going to rely on that.
168 * Otherwise, we restore to conventional mechanism
169 * to clear frozen PE during PCI config access.
172 phb->flags |= PNV_PHB_FLAG_EEH;
174 phb->flags &= ~PNV_PHB_FLAG_EEH;
180 * ioda_eeh_set_option - Set EEH operation or I/O setting
184 * Enable or disable EEH option for the indicated PE. The
185 * function also can be used to enable I/O or DMA for the
188 static int ioda_eeh_set_option(struct eeh_pe *pe, int option)
192 struct pci_controller *hose = pe->phb;
193 struct pnv_phb *phb = hose->private_data;
195 /* Check on PE number */
196 if (pe->addr < 0 || pe->addr >= phb->ioda.total_pe) {
197 pr_err("%s: PE address %x out of range [0, %x] "
199 __func__, pe->addr, phb->ioda.total_pe,
200 hose->global_number);
206 case EEH_OPT_DISABLE:
212 case EEH_OPT_THAW_MMIO:
213 ret = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
214 OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO);
216 pr_warning("%s: Failed to enable MMIO for "
217 "PHB#%x-PE#%x, err=%lld\n",
218 __func__, hose->global_number, pe_no, ret);
223 case EEH_OPT_THAW_DMA:
224 ret = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
225 OPAL_EEH_ACTION_CLEAR_FREEZE_DMA);
227 pr_warning("%s: Failed to enable DMA for "
228 "PHB#%x-PE#%x, err=%lld\n",
229 __func__, hose->global_number, pe_no, ret);
235 pr_warning("%s: Invalid option %d\n", __func__, option);
242 static void ioda_eeh_phb_diag(struct pci_controller *hose)
244 struct pnv_phb *phb = hose->private_data;
247 rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
248 PNV_PCI_DIAG_BUF_SIZE);
249 if (rc != OPAL_SUCCESS) {
250 pr_warning("%s: Failed to get diag-data for PHB#%x (%ld)\n",
251 __func__, hose->global_number, rc);
255 pnv_pci_dump_phb_diag_data(hose, phb->diag.blob);
259 * ioda_eeh_get_state - Retrieve the state of PE
262 * The PE's state should be retrieved from the PEEV, PEST
263 * IODA tables. Since the OPAL has exported the function
264 * to do it, it'd better to use that.
266 static int ioda_eeh_get_state(struct eeh_pe *pe)
273 struct pci_controller *hose = pe->phb;
274 struct pnv_phb *phb = hose->private_data;
277 * Sanity check on PE address. The PHB PE address should
280 if (pe->addr < 0 || pe->addr >= phb->ioda.total_pe) {
281 pr_err("%s: PE address %x out of range [0, %x] "
283 __func__, pe->addr, phb->ioda.total_pe,
284 hose->global_number);
285 return EEH_STATE_NOT_SUPPORT;
289 * If we're in middle of PE reset, return normal
290 * state to keep EEH core going. For PHB reset, we
291 * still expect to have fenced PHB cleared with
294 if (!(pe->type & EEH_PE_PHB) &&
295 (pe->state & EEH_PE_RESET)) {
296 result = (EEH_STATE_MMIO_ACTIVE |
297 EEH_STATE_DMA_ACTIVE |
298 EEH_STATE_MMIO_ENABLED |
299 EEH_STATE_DMA_ENABLED);
303 /* Retrieve PE status through OPAL */
305 ret = opal_pci_eeh_freeze_status(phb->opal_id, pe_no,
306 &fstate, &pcierr, NULL);
308 pr_err("%s: Failed to get EEH status on "
309 "PHB#%x-PE#%x\n, err=%lld\n",
310 __func__, hose->global_number, pe_no, ret);
311 return EEH_STATE_NOT_SUPPORT;
314 /* Check PHB status */
315 if (pe->type & EEH_PE_PHB) {
317 result &= ~EEH_STATE_RESET_ACTIVE;
319 if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
320 result |= EEH_STATE_MMIO_ACTIVE;
321 result |= EEH_STATE_DMA_ACTIVE;
322 result |= EEH_STATE_MMIO_ENABLED;
323 result |= EEH_STATE_DMA_ENABLED;
324 } else if (!(pe->state & EEH_PE_ISOLATED)) {
325 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
326 ioda_eeh_phb_diag(hose);
332 /* Parse result out */
335 case OPAL_EEH_STOPPED_NOT_FROZEN:
336 result &= ~EEH_STATE_RESET_ACTIVE;
337 result |= EEH_STATE_MMIO_ACTIVE;
338 result |= EEH_STATE_DMA_ACTIVE;
339 result |= EEH_STATE_MMIO_ENABLED;
340 result |= EEH_STATE_DMA_ENABLED;
342 case OPAL_EEH_STOPPED_MMIO_FREEZE:
343 result &= ~EEH_STATE_RESET_ACTIVE;
344 result |= EEH_STATE_DMA_ACTIVE;
345 result |= EEH_STATE_DMA_ENABLED;
347 case OPAL_EEH_STOPPED_DMA_FREEZE:
348 result &= ~EEH_STATE_RESET_ACTIVE;
349 result |= EEH_STATE_MMIO_ACTIVE;
350 result |= EEH_STATE_MMIO_ENABLED;
352 case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
353 result &= ~EEH_STATE_RESET_ACTIVE;
355 case OPAL_EEH_STOPPED_RESET:
356 result |= EEH_STATE_RESET_ACTIVE;
358 case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
359 result |= EEH_STATE_UNAVAILABLE;
361 case OPAL_EEH_STOPPED_PERM_UNAVAIL:
362 result |= EEH_STATE_NOT_SUPPORT;
365 pr_warning("%s: Unexpected EEH status 0x%x "
367 __func__, fstate, hose->global_number, pe_no);
370 /* Dump PHB diag-data for frozen PE */
371 if (result != EEH_STATE_NOT_SUPPORT &&
372 (result & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) !=
373 (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE) &&
374 !(pe->state & EEH_PE_ISOLATED)) {
375 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
376 ioda_eeh_phb_diag(hose);
382 static s64 ioda_eeh_phb_poll(struct pnv_phb *phb)
384 s64 rc = OPAL_HARDWARE;
387 rc = opal_pci_poll(phb->opal_id);
391 if (system_state < SYSTEM_RUNNING)
400 int ioda_eeh_phb_reset(struct pci_controller *hose, int option)
402 struct pnv_phb *phb = hose->private_data;
403 s64 rc = OPAL_HARDWARE;
405 pr_debug("%s: Reset PHB#%x, option=%d\n",
406 __func__, hose->global_number, option);
408 /* Issue PHB complete reset request */
409 if (option == EEH_RESET_FUNDAMENTAL ||
410 option == EEH_RESET_HOT)
411 rc = opal_pci_reset(phb->opal_id,
414 else if (option == EEH_RESET_DEACTIVATE)
415 rc = opal_pci_reset(phb->opal_id,
417 OPAL_DEASSERT_RESET);
422 * Poll state of the PHB until the request is done
423 * successfully. The PHB reset is usually PHB complete
424 * reset followed by hot reset on root bus. So we also
425 * need the PCI bus settlement delay.
427 rc = ioda_eeh_phb_poll(phb);
428 if (option == EEH_RESET_DEACTIVATE) {
429 if (system_state < SYSTEM_RUNNING)
430 udelay(1000 * EEH_PE_RST_SETTLE_TIME);
432 msleep(EEH_PE_RST_SETTLE_TIME);
435 if (rc != OPAL_SUCCESS)
441 static int ioda_eeh_root_reset(struct pci_controller *hose, int option)
443 struct pnv_phb *phb = hose->private_data;
444 s64 rc = OPAL_SUCCESS;
446 pr_debug("%s: Reset PHB#%x, option=%d\n",
447 __func__, hose->global_number, option);
450 * During the reset deassert time, we needn't care
451 * the reset scope because the firmware does nothing
452 * for fundamental or hot reset during deassert phase.
454 if (option == EEH_RESET_FUNDAMENTAL)
455 rc = opal_pci_reset(phb->opal_id,
456 OPAL_PCI_FUNDAMENTAL_RESET,
458 else if (option == EEH_RESET_HOT)
459 rc = opal_pci_reset(phb->opal_id,
462 else if (option == EEH_RESET_DEACTIVATE)
463 rc = opal_pci_reset(phb->opal_id,
465 OPAL_DEASSERT_RESET);
469 /* Poll state of the PHB until the request is done */
470 rc = ioda_eeh_phb_poll(phb);
471 if (option == EEH_RESET_DEACTIVATE)
472 msleep(EEH_PE_RST_SETTLE_TIME);
474 if (rc != OPAL_SUCCESS)
480 static int ioda_eeh_bridge_reset(struct pci_dev *dev, int option)
483 struct device_node *dn = pci_device_to_OF_node(dev);
484 struct eeh_dev *edev = of_node_to_eeh_dev(dn);
485 int aer = edev ? edev->aer_cap : 0;
488 pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
489 __func__, pci_domain_nr(dev->bus),
490 dev->bus->number, option);
493 case EEH_RESET_FUNDAMENTAL:
495 /* Don't report linkDown event */
497 eeh_ops->read_config(dn, aer + PCI_ERR_UNCOR_MASK,
499 ctrl |= PCI_ERR_UNC_SURPDN;
500 eeh_ops->write_config(dn, aer + PCI_ERR_UNCOR_MASK,
504 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
505 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
506 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
507 msleep(EEH_PE_RST_HOLD_TIME);
510 case EEH_RESET_DEACTIVATE:
511 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
512 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
513 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
514 msleep(EEH_PE_RST_SETTLE_TIME);
516 /* Continue reporting linkDown event */
518 eeh_ops->read_config(dn, aer + PCI_ERR_UNCOR_MASK,
520 ctrl &= ~PCI_ERR_UNC_SURPDN;
521 eeh_ops->write_config(dn, aer + PCI_ERR_UNCOR_MASK,
531 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
533 struct pci_controller *hose;
535 if (pci_is_root_bus(dev->bus)) {
536 hose = pci_bus_to_host(dev->bus);
537 ioda_eeh_root_reset(hose, EEH_RESET_HOT);
538 ioda_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
540 ioda_eeh_bridge_reset(dev, EEH_RESET_HOT);
541 ioda_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
546 * ioda_eeh_reset - Reset the indicated PE
548 * @option: reset option
550 * Do reset on the indicated PE. For PCI bus sensitive PE,
551 * we need to reset the parent p2p bridge. The PHB has to
552 * be reinitialized if the p2p bridge is root bridge. For
553 * PCI device sensitive PE, we will try to reset the device
554 * through FLR. For now, we don't have OPAL APIs to do HARD
555 * reset yet, so all reset would be SOFT (HOT) reset.
557 static int ioda_eeh_reset(struct eeh_pe *pe, int option)
559 struct pci_controller *hose = pe->phb;
564 * For PHB reset, we always have complete reset. For those PEs whose
565 * primary bus derived from root complex (root bus) or root port
566 * (usually bus#1), we apply hot or fundamental reset on the root port.
567 * For other PEs, we always have hot reset on the PE primary bus.
569 * Here, we have different design to pHyp, which always clear the
570 * frozen state during PE reset. However, the good idea here from
571 * benh is to keep frozen state before we get PE reset done completely
572 * (until BAR restore). With the frozen state, HW drops illegal IO
573 * or MMIO access, which can incur recrusive frozen PE during PE
574 * reset. The side effect is that EEH core has to clear the frozen
575 * state explicitly after BAR restore.
577 if (pe->type & EEH_PE_PHB) {
578 ret = ioda_eeh_phb_reset(hose, option);
580 bus = eeh_pe_bus_get(pe);
581 if (pci_is_root_bus(bus) ||
582 pci_is_root_bus(bus->parent))
583 ret = ioda_eeh_root_reset(hose, option);
585 ret = ioda_eeh_bridge_reset(bus->self, option);
592 * ioda_eeh_configure_bridge - Configure the PCI bridges for the indicated PE
595 * For particular PE, it might have included PCI bridges. In order
596 * to make the PE work properly, those PCI bridges should be configured
597 * correctly. However, we need do nothing on P7IOC since the reset
598 * function will do everything that should be covered by the function.
600 static int ioda_eeh_configure_bridge(struct eeh_pe *pe)
605 static void ioda_eeh_hub_diag_common(struct OpalIoP7IOCErrorData *data)
608 pr_info(" GEM XFIR: %016llx\n", data->gemXfir);
609 pr_info(" GEM RFIR: %016llx\n", data->gemRfir);
610 pr_info(" GEM RIRQFIR: %016llx\n", data->gemRirqfir);
611 pr_info(" GEM Mask: %016llx\n", data->gemMask);
612 pr_info(" GEM RWOF: %016llx\n", data->gemRwof);
615 pr_info(" LEM FIR: %016llx\n", data->lemFir);
616 pr_info(" LEM Error Mask: %016llx\n", data->lemErrMask);
617 pr_info(" LEM Action 0: %016llx\n", data->lemAction0);
618 pr_info(" LEM Action 1: %016llx\n", data->lemAction1);
619 pr_info(" LEM WOF: %016llx\n", data->lemWof);
622 static void ioda_eeh_hub_diag(struct pci_controller *hose)
624 struct pnv_phb *phb = hose->private_data;
625 struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag;
628 rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
629 if (rc != OPAL_SUCCESS) {
630 pr_warning("%s: Failed to get HUB#%llx diag-data (%ld)\n",
631 __func__, phb->hub_id, rc);
635 switch (data->type) {
636 case OPAL_P7IOC_DIAG_TYPE_RGC:
637 pr_info("P7IOC diag-data for RGC\n\n");
638 ioda_eeh_hub_diag_common(data);
639 pr_info(" RGC Status: %016llx\n", data->rgc.rgcStatus);
640 pr_info(" RGC LDCP: %016llx\n", data->rgc.rgcLdcp);
642 case OPAL_P7IOC_DIAG_TYPE_BI:
643 pr_info("P7IOC diag-data for BI %s\n\n",
644 data->bi.biDownbound ? "Downbound" : "Upbound");
645 ioda_eeh_hub_diag_common(data);
646 pr_info(" BI LDCP 0: %016llx\n", data->bi.biLdcp0);
647 pr_info(" BI LDCP 1: %016llx\n", data->bi.biLdcp1);
648 pr_info(" BI LDCP 2: %016llx\n", data->bi.biLdcp2);
649 pr_info(" BI Fence Status: %016llx\n", data->bi.biFenceStatus);
651 case OPAL_P7IOC_DIAG_TYPE_CI:
652 pr_info("P7IOC diag-data for CI Port %d\\nn",
654 ioda_eeh_hub_diag_common(data);
655 pr_info(" CI Port Status: %016llx\n", data->ci.ciPortStatus);
656 pr_info(" CI Port LDCP: %016llx\n", data->ci.ciPortLdcp);
658 case OPAL_P7IOC_DIAG_TYPE_MISC:
659 pr_info("P7IOC diag-data for MISC\n\n");
660 ioda_eeh_hub_diag_common(data);
662 case OPAL_P7IOC_DIAG_TYPE_I2C:
663 pr_info("P7IOC diag-data for I2C\n\n");
664 ioda_eeh_hub_diag_common(data);
667 pr_warning("%s: Invalid type of HUB#%llx diag-data (%d)\n",
668 __func__, phb->hub_id, data->type);
672 static int ioda_eeh_get_pe(struct pci_controller *hose,
673 u16 pe_no, struct eeh_pe **pe)
675 struct eeh_pe *phb_pe, *dev_pe;
678 /* Find the PHB PE */
679 phb_pe = eeh_phb_pe_get(hose);
683 /* Find the PE according to PE# */
684 memset(&dev, 0, sizeof(struct eeh_dev));
686 dev.pe_config_addr = pe_no;
687 dev_pe = eeh_pe_get(&dev);
688 if (!dev_pe) return -EEXIST;
695 * ioda_eeh_next_error - Retrieve next error for EEH core to handle
696 * @pe: The affected PE
698 * The function is expected to be called by EEH core while it gets
699 * special EEH event (without binding PE). The function calls to
700 * OPAL APIs for next error to handle. The informational error is
701 * handled internally by platform. However, the dead IOC, dead PHB,
702 * fenced PHB and frozen PE should be handled by EEH core eventually.
704 static int ioda_eeh_next_error(struct eeh_pe **pe)
706 struct pci_controller *hose;
708 struct eeh_pe *phb_pe, *parent_pe;
710 __be16 err_type, severity;
711 int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
713 int state, ret = EEH_NEXT_ERR_NONE;
716 * While running here, it's safe to purge the event queue.
717 * And we should keep the cached OPAL notifier event sychronized
718 * between the kernel and firmware.
720 eeh_remove_event(NULL);
721 opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
723 list_for_each_entry(hose, &hose_list, list_node) {
725 * If the subordinate PCI buses of the PHB has been
726 * removed or is exactly under error recovery, we
727 * needn't take care of it any more.
729 phb = hose->private_data;
730 phb_pe = eeh_phb_pe_get(hose);
731 if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
734 rc = opal_pci_next_error(phb->opal_id,
735 &frozen_pe_no, &err_type, &severity);
737 /* If OPAL API returns error, we needn't proceed */
738 if (rc != OPAL_SUCCESS) {
739 pr_devel("%s: Invalid return value on "
740 "PHB#%x (0x%lx) from opal_pci_next_error",
741 __func__, hose->global_number, rc);
745 /* If the PHB doesn't have error, stop processing */
746 if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
747 be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
748 pr_devel("%s: No error found on PHB#%x\n",
749 __func__, hose->global_number);
754 * Processing the error. We're expecting the error with
755 * highest priority reported upon multiple errors on the
758 pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
759 __func__, be16_to_cpu(err_type), be16_to_cpu(severity),
760 be64_to_cpu(frozen_pe_no), hose->global_number);
761 switch (be16_to_cpu(err_type)) {
762 case OPAL_EEH_IOC_ERROR:
763 if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
764 pr_err("EEH: dead IOC detected\n");
765 ret = EEH_NEXT_ERR_DEAD_IOC;
766 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
767 pr_info("EEH: IOC informative error "
769 ioda_eeh_hub_diag(hose);
770 ret = EEH_NEXT_ERR_NONE;
774 case OPAL_EEH_PHB_ERROR:
775 if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
777 pr_err("EEH: dead PHB#%x detected\n",
778 hose->global_number);
779 ret = EEH_NEXT_ERR_DEAD_PHB;
780 } else if (be16_to_cpu(severity) ==
781 OPAL_EEH_SEV_PHB_FENCED) {
783 pr_err("EEH: fenced PHB#%x detected\n",
784 hose->global_number);
785 ret = EEH_NEXT_ERR_FENCED_PHB;
786 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
787 pr_info("EEH: PHB#%x informative error "
789 hose->global_number);
790 ioda_eeh_phb_diag(hose);
791 ret = EEH_NEXT_ERR_NONE;
795 case OPAL_EEH_PE_ERROR:
797 * If we can't find the corresponding PE, we
798 * just try to unfreeze.
800 if (ioda_eeh_get_pe(hose,
801 be64_to_cpu(frozen_pe_no), pe)) {
802 /* Try best to clear it */
803 pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
804 hose->global_number, frozen_pe_no);
805 opal_pci_eeh_freeze_clear(phb->opal_id, frozen_pe_no,
806 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
807 ret = EEH_NEXT_ERR_NONE;
808 } else if ((*pe)->state & EEH_PE_ISOLATED) {
809 ret = EEH_NEXT_ERR_NONE;
811 pr_err("EEH: Frozen PE#%x on PHB#%x detected\n",
812 (*pe)->addr, (*pe)->phb->global_number);
813 ret = EEH_NEXT_ERR_FROZEN_PE;
818 pr_warn("%s: Unexpected error type %d\n",
819 __func__, be16_to_cpu(err_type));
823 * EEH core will try recover from fenced PHB or
824 * frozen PE. In the time for frozen PE, EEH core
825 * enable IO path for that before collecting logs,
826 * but it ruins the site. So we have to dump the
827 * log in advance here.
829 if ((ret == EEH_NEXT_ERR_FROZEN_PE ||
830 ret == EEH_NEXT_ERR_FENCED_PHB) &&
831 !((*pe)->state & EEH_PE_ISOLATED)) {
832 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
833 ioda_eeh_phb_diag(hose);
837 * We probably have the frozen parent PE out there and
838 * we need have to handle frozen parent PE firstly.
840 if (ret == EEH_NEXT_ERR_FROZEN_PE) {
841 parent_pe = (*pe)->parent;
843 /* Hit the ceiling ? */
844 if (parent_pe->type & EEH_PE_PHB)
847 /* Frozen parent PE ? */
848 state = ioda_eeh_get_state(parent_pe);
850 (state & active_flags) != active_flags)
853 /* Next parent level */
854 parent_pe = parent_pe->parent;
857 /* We possibly migrate to another PE */
858 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
862 * If we have no errors on the specific PHB or only
863 * informative error there, we continue poking it.
864 * Otherwise, we need actions to be taken by upper
867 if (ret > EEH_NEXT_ERR_INF)
874 struct pnv_eeh_ops ioda_eeh_ops = {
875 .post_init = ioda_eeh_post_init,
876 .set_option = ioda_eeh_set_option,
877 .get_state = ioda_eeh_get_state,
878 .reset = ioda_eeh_reset,
879 .configure_bridge = ioda_eeh_configure_bridge,
880 .next_error = ioda_eeh_next_error