1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * The file intends to implement PE based on the information from
4 * platforms. Basically, there have 3 types of PEs: PHB/Bus/Device.
5 * All the PEs should be organized as hierarchy tree. The first level
6 * of the tree will be associated to existing PHBs since the particular
7 * PE is only meaningful in one PHB domain.
9 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2012.
12 #include <linux/delay.h>
13 #include <linux/export.h>
14 #include <linux/gfp.h>
15 #include <linux/kernel.h>
17 #include <linux/pci.h>
18 #include <linux/string.h>
20 #include <asm/pci-bridge.h>
21 #include <asm/ppc-pci.h>
23 static int eeh_pe_aux_size
= 0;
24 static LIST_HEAD(eeh_phb_pe
);
27 * eeh_set_pe_aux_size - Set PE auxiliary data size
28 * @size: PE auxiliary data size in bytes
30 * Set PE auxiliary data size.
32 void eeh_set_pe_aux_size(int size
)
37 eeh_pe_aux_size
= size
;
41 * eeh_pe_alloc - Allocate PE
42 * @phb: PCI controller
45 * Allocate PE instance dynamically.
47 static struct eeh_pe
*eeh_pe_alloc(struct pci_controller
*phb
, int type
)
52 alloc_size
= sizeof(struct eeh_pe
);
53 if (eeh_pe_aux_size
) {
54 alloc_size
= ALIGN(alloc_size
, cache_line_size());
55 alloc_size
+= eeh_pe_aux_size
;
59 pe
= kzalloc(alloc_size
, GFP_KERNEL
);
62 /* Initialize PHB PE */
65 INIT_LIST_HEAD(&pe
->child_list
);
66 INIT_LIST_HEAD(&pe
->edevs
);
68 pe
->data
= (void *)pe
+ ALIGN(sizeof(struct eeh_pe
),
74 * eeh_phb_pe_create - Create PHB PE
75 * @phb: PCI controller
77 * The function should be called while the PHB is detected during
78 * system boot or PCI hotplug in order to create PHB PE.
80 int eeh_phb_pe_create(struct pci_controller
*phb
)
85 pe
= eeh_pe_alloc(phb
, EEH_PE_PHB
);
87 pr_err("%s: out of memory!\n", __func__
);
91 /* Put it into the list */
92 list_add_tail(&pe
->child
, &eeh_phb_pe
);
94 pr_debug("EEH: Add PE for PHB#%x\n", phb
->global_number
);
100 * eeh_wait_state - Wait for PE state
102 * @max_wait: maximal period in millisecond
104 * Wait for the state of associated PE. It might take some time
105 * to retrieve the PE's state.
107 int eeh_wait_state(struct eeh_pe
*pe
, int max_wait
)
113 * According to PAPR, the state of PE might be temporarily
114 * unavailable. Under the circumstance, we have to wait
115 * for indicated time determined by firmware. The maximal
116 * wait time is 5 minutes, which is acquired from the original
117 * EEH implementation. Also, the original implementation
118 * also defined the minimal wait time as 1 second.
120 #define EEH_STATE_MIN_WAIT_TIME (1000)
121 #define EEH_STATE_MAX_WAIT_TIME (300 * 1000)
124 ret
= eeh_ops
->get_state(pe
, &mwait
);
126 if (ret
!= EEH_STATE_UNAVAILABLE
)
130 pr_warn("%s: Timeout when getting PE's state (%d)\n",
132 return EEH_STATE_NOT_SUPPORT
;
135 if (mwait
< EEH_STATE_MIN_WAIT_TIME
) {
136 pr_warn("%s: Firmware returned bad wait value %d\n",
138 mwait
= EEH_STATE_MIN_WAIT_TIME
;
139 } else if (mwait
> EEH_STATE_MAX_WAIT_TIME
) {
140 pr_warn("%s: Firmware returned too long wait value %d\n",
142 mwait
= EEH_STATE_MAX_WAIT_TIME
;
145 msleep(min(mwait
, max_wait
));
151 * eeh_phb_pe_get - Retrieve PHB PE based on the given PHB
152 * @phb: PCI controller
154 * The overall PEs form hierarchy tree. The first layer of the
155 * hierarchy tree is composed of PHB PEs. The function is used
156 * to retrieve the corresponding PHB PE according to the given PHB.
158 struct eeh_pe
*eeh_phb_pe_get(struct pci_controller
*phb
)
162 list_for_each_entry(pe
, &eeh_phb_pe
, child
) {
164 * Actually, we needn't check the type since
165 * the PE for PHB has been determined when that
168 if ((pe
->type
& EEH_PE_PHB
) && pe
->phb
== phb
)
176 * eeh_pe_next - Retrieve the next PE in the tree
180 * The function is used to retrieve the next PE in the
183 struct eeh_pe
*eeh_pe_next(struct eeh_pe
*pe
, struct eeh_pe
*root
)
185 struct list_head
*next
= pe
->child_list
.next
;
187 if (next
== &pe
->child_list
) {
191 next
= pe
->child
.next
;
192 if (next
!= &pe
->parent
->child_list
)
198 return list_entry(next
, struct eeh_pe
, child
);
202 * eeh_pe_traverse - Traverse PEs in the specified PHB
205 * @flag: extra parameter to callback
207 * The function is used to traverse the specified PE and its
208 * child PEs. The traversing is to be terminated once the
209 * callback returns something other than NULL, or no more PEs
212 void *eeh_pe_traverse(struct eeh_pe
*root
,
213 eeh_pe_traverse_func fn
, void *flag
)
218 eeh_for_each_pe(root
, pe
) {
227 * eeh_pe_dev_traverse - Traverse the devices from the PE
229 * @fn: function callback
230 * @flag: extra parameter to callback
232 * The function is used to traverse the devices of the specified
233 * PE and its child PEs.
235 void eeh_pe_dev_traverse(struct eeh_pe
*root
,
236 eeh_edev_traverse_func fn
, void *flag
)
239 struct eeh_dev
*edev
, *tmp
;
242 pr_warn("%s: Invalid PE %p\n",
247 /* Traverse root PE */
248 eeh_for_each_pe(root
, pe
)
249 eeh_pe_for_each_dev(pe
, edev
, tmp
)
254 * __eeh_pe_get - Check the PE address
256 * For one particular PE, it can be identified by PE address
257 * or tranditional BDF address. BDF address is composed of
258 * Bus/Device/Function number. The extra data referred by flag
259 * indicates which type of address should be used.
261 static void *__eeh_pe_get(struct eeh_pe
*pe
, void *flag
)
263 int *target_pe
= flag
;
265 /* PHB PEs are special and should be ignored */
266 if (pe
->type
& EEH_PE_PHB
)
269 if (*target_pe
== pe
->addr
)
276 * eeh_pe_get - Search PE based on the given address
277 * @phb: PCI controller
280 * Search the corresponding PE based on the specified address which
281 * is included in the eeh device. The function is used to check if
282 * the associated PE has been created against the PE address. It's
283 * notable that the PE address has 2 format: traditional PE address
284 * which is composed of PCI bus/device/function number, or unified
287 struct eeh_pe
*eeh_pe_get(struct pci_controller
*phb
, int pe_no
)
289 struct eeh_pe
*root
= eeh_phb_pe_get(phb
);
291 return eeh_pe_traverse(root
, __eeh_pe_get
, &pe_no
);
295 * eeh_pe_tree_insert - Add EEH device to parent PE
297 * @new_pe_parent: PE to create additional PEs under
299 * Add EEH device to the PE in edev->pe_config_addr. If a PE already
300 * exists with that address then @edev is added to that PE. Otherwise
301 * a new PE is created and inserted into the PE tree as a child of
304 * If @new_pe_parent is NULL then the new PE will be inserted under
305 * directly under the PHB.
307 int eeh_pe_tree_insert(struct eeh_dev
*edev
, struct eeh_pe
*new_pe_parent
)
309 struct pci_controller
*hose
= edev
->controller
;
310 struct eeh_pe
*pe
, *parent
;
313 * Search the PE has been existing or not according
314 * to the PE address. If that has been existing, the
315 * PE should be composed of PCI bus and its subordinate
318 pe
= eeh_pe_get(hose
, edev
->pe_config_addr
);
320 if (pe
->type
& EEH_PE_INVALID
) {
321 list_add_tail(&edev
->entry
, &pe
->edevs
);
324 * We're running to here because of PCI hotplug caused by
325 * EEH recovery. We need clear EEH_PE_INVALID until the top.
329 if (!(parent
->type
& EEH_PE_INVALID
))
331 parent
->type
&= ~EEH_PE_INVALID
;
332 parent
= parent
->parent
;
335 eeh_edev_dbg(edev
, "Added to existing PE (parent: PE#%x)\n",
338 /* Mark the PE as type of PCI bus */
339 pe
->type
= EEH_PE_BUS
;
342 /* Put the edev to PE */
343 list_add_tail(&edev
->entry
, &pe
->edevs
);
344 eeh_edev_dbg(edev
, "Added to bus PE\n");
349 /* Create a new EEH PE */
351 pe
= eeh_pe_alloc(hose
, EEH_PE_VF
);
353 pe
= eeh_pe_alloc(hose
, EEH_PE_DEVICE
);
355 pr_err("%s: out of memory!\n", __func__
);
359 pe
->addr
= edev
->pe_config_addr
;
362 * Put the new EEH PE into hierarchy tree. If the parent
363 * can't be found, the newly created PE will be attached
364 * to PHB directly. Otherwise, we have to associate the
365 * PE with its parent.
367 if (!new_pe_parent
) {
368 new_pe_parent
= eeh_phb_pe_get(hose
);
369 if (!new_pe_parent
) {
370 pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
371 __func__
, hose
->global_number
);
378 /* link new PE into the tree */
379 pe
->parent
= new_pe_parent
;
380 list_add_tail(&pe
->child
, &new_pe_parent
->child_list
);
383 * Put the newly created PE into the child list and
384 * link the EEH device accordingly.
386 list_add_tail(&edev
->entry
, &pe
->edevs
);
388 eeh_edev_dbg(edev
, "Added to new (parent: PE#%x)\n",
389 new_pe_parent
->addr
);
395 * eeh_pe_tree_remove - Remove one EEH device from the associated PE
398 * The PE hierarchy tree might be changed when doing PCI hotplug.
399 * Also, the PCI devices or buses could be removed from the system
400 * during EEH recovery. So we have to call the function remove the
401 * corresponding PE accordingly if necessary.
403 int eeh_pe_tree_remove(struct eeh_dev
*edev
)
405 struct eeh_pe
*pe
, *parent
, *child
;
409 pe
= eeh_dev_to_pe(edev
);
411 eeh_edev_dbg(edev
, "No PE found for device.\n");
415 /* Remove the EEH device */
417 list_del(&edev
->entry
);
420 * Check if the parent PE includes any EEH devices.
421 * If not, we should delete that. Also, we should
422 * delete the parent PE if it doesn't have associated
423 * child PEs and EEH devices.
428 /* PHB PEs should never be removed */
429 if (pe
->type
& EEH_PE_PHB
)
433 * XXX: KEEP is set while resetting a PE. I don't think it's
434 * ever set without RECOVERING also being set. I could
435 * be wrong though so catch that with a WARN.
437 keep
= !!(pe
->state
& EEH_PE_KEEP
);
438 recover
= !!(pe
->state
& EEH_PE_RECOVERING
);
439 WARN_ON(keep
&& !recover
);
441 if (!keep
&& !recover
) {
442 if (list_empty(&pe
->edevs
) &&
443 list_empty(&pe
->child_list
)) {
444 list_del(&pe
->child
);
451 * Mark the PE as invalid. At the end of the recovery
452 * process any invalid PEs will be garbage collected.
454 * We need to delay the free()ing of them since we can
455 * remove edev's while traversing the PE tree which
456 * might trigger the removal of a PE and we can't
457 * deal with that (yet).
459 if (list_empty(&pe
->edevs
)) {
461 list_for_each_entry(child
, &pe
->child_list
, child
) {
462 if (!(child
->type
& EEH_PE_INVALID
)) {
469 pe
->type
|= EEH_PE_INVALID
;
482 * eeh_pe_update_time_stamp - Update PE's frozen time stamp
485 * We have time stamp for each PE to trace its time of getting
486 * frozen in last hour. The function should be called to update
487 * the time stamp on first error of the specific PE. On the other
488 * handle, we needn't account for errors happened in last hour.
490 void eeh_pe_update_time_stamp(struct eeh_pe
*pe
)
496 if (pe
->freeze_count
<= 0) {
497 pe
->freeze_count
= 0;
498 pe
->tstamp
= ktime_get_seconds();
500 tstamp
= ktime_get_seconds();
501 if (tstamp
- pe
->tstamp
> 3600) {
503 pe
->freeze_count
= 0;
509 * eeh_pe_state_mark - Mark specified state for PE and its associated device
512 * EEH error affects the current PE and its child PEs. The function
513 * is used to mark appropriate state for the affected PEs and the
514 * associated devices.
516 void eeh_pe_state_mark(struct eeh_pe
*root
, int state
)
520 eeh_for_each_pe(root
, pe
)
521 if (!(pe
->state
& EEH_PE_REMOVED
))
524 EXPORT_SYMBOL_GPL(eeh_pe_state_mark
);
527 * eeh_pe_mark_isolated
530 * Record that a PE has been isolated by marking the PE and its children as
531 * EEH_PE_ISOLATED (and EEH_PE_CFG_BLOCKED, if required) and their PCI devices
532 * as pci_channel_io_frozen.
534 void eeh_pe_mark_isolated(struct eeh_pe
*root
)
537 struct eeh_dev
*edev
;
538 struct pci_dev
*pdev
;
540 eeh_pe_state_mark(root
, EEH_PE_ISOLATED
);
541 eeh_for_each_pe(root
, pe
) {
542 list_for_each_entry(edev
, &pe
->edevs
, entry
) {
543 pdev
= eeh_dev_to_pci_dev(edev
);
545 pdev
->error_state
= pci_channel_io_frozen
;
547 /* Block PCI config access if required */
548 if (pe
->state
& EEH_PE_CFG_RESTRICTED
)
549 pe
->state
|= EEH_PE_CFG_BLOCKED
;
552 EXPORT_SYMBOL_GPL(eeh_pe_mark_isolated
);
554 static void __eeh_pe_dev_mode_mark(struct eeh_dev
*edev
, void *flag
)
556 int mode
= *((int *)flag
);
562 * eeh_pe_dev_state_mark - Mark state for all device under the PE
565 * Mark specific state for all child devices of the PE.
567 void eeh_pe_dev_mode_mark(struct eeh_pe
*pe
, int mode
)
569 eeh_pe_dev_traverse(pe
, __eeh_pe_dev_mode_mark
, &mode
);
573 * eeh_pe_state_clear - Clear state for the PE
576 * @include_passed: include passed-through devices?
578 * The function is used to clear the indicated state from the
579 * given PE. Besides, we also clear the check count of the PE
582 void eeh_pe_state_clear(struct eeh_pe
*root
, int state
, bool include_passed
)
585 struct eeh_dev
*edev
, *tmp
;
586 struct pci_dev
*pdev
;
588 eeh_for_each_pe(root
, pe
) {
589 /* Keep the state of permanently removed PE intact */
590 if (pe
->state
& EEH_PE_REMOVED
)
593 if (!include_passed
&& eeh_pe_passed(pe
))
599 * Special treatment on clearing isolated state. Clear
600 * check count since last isolation and put all affected
601 * devices to normal state.
603 if (!(state
& EEH_PE_ISOLATED
))
607 eeh_pe_for_each_dev(pe
, edev
, tmp
) {
608 pdev
= eeh_dev_to_pci_dev(edev
);
612 pdev
->error_state
= pci_channel_io_normal
;
615 /* Unblock PCI config access if required */
616 if (pe
->state
& EEH_PE_CFG_RESTRICTED
)
617 pe
->state
&= ~EEH_PE_CFG_BLOCKED
;
622 * Some PCI bridges (e.g. PLX bridges) have primary/secondary
623 * buses assigned explicitly by firmware, and we probably have
624 * lost that after reset. So we have to delay the check until
625 * the PCI-CFG registers have been restored for the parent
628 * Don't use normal PCI-CFG accessors, which probably has been
629 * blocked on normal path during the stage. So we need utilize
630 * eeh operations, which is always permitted.
632 static void eeh_bridge_check_link(struct eeh_dev
*edev
)
639 * We only check root port and downstream ports of
642 if (!(edev
->mode
& (EEH_DEV_ROOT_PORT
| EEH_DEV_DS_PORT
)))
645 eeh_edev_dbg(edev
, "Checking PCIe link...\n");
647 /* Check slot status */
648 cap
= edev
->pcie_cap
;
649 eeh_ops
->read_config(edev
, cap
+ PCI_EXP_SLTSTA
, 2, &val
);
650 if (!(val
& PCI_EXP_SLTSTA_PDS
)) {
651 eeh_edev_dbg(edev
, "No card in the slot (0x%04x) !\n", val
);
655 /* Check power status if we have the capability */
656 eeh_ops
->read_config(edev
, cap
+ PCI_EXP_SLTCAP
, 2, &val
);
657 if (val
& PCI_EXP_SLTCAP_PCP
) {
658 eeh_ops
->read_config(edev
, cap
+ PCI_EXP_SLTCTL
, 2, &val
);
659 if (val
& PCI_EXP_SLTCTL_PCC
) {
660 eeh_edev_dbg(edev
, "In power-off state, power it on ...\n");
661 val
&= ~(PCI_EXP_SLTCTL_PCC
| PCI_EXP_SLTCTL_PIC
);
662 val
|= (0x0100 & PCI_EXP_SLTCTL_PIC
);
663 eeh_ops
->write_config(edev
, cap
+ PCI_EXP_SLTCTL
, 2, val
);
669 eeh_ops
->read_config(edev
, cap
+ PCI_EXP_LNKCTL
, 2, &val
);
670 val
&= ~PCI_EXP_LNKCTL_LD
;
671 eeh_ops
->write_config(edev
, cap
+ PCI_EXP_LNKCTL
, 2, val
);
674 if (!edev
->pdev
->link_active_reporting
) {
675 eeh_edev_dbg(edev
, "No link reporting capability\n");
680 /* Wait the link is up until timeout (5s) */
682 while (timeout
< 5000) {
686 eeh_ops
->read_config(edev
, cap
+ PCI_EXP_LNKSTA
, 2, &val
);
687 if (val
& PCI_EXP_LNKSTA_DLLLA
)
691 if (val
& PCI_EXP_LNKSTA_DLLLA
)
692 eeh_edev_dbg(edev
, "Link up (%s)\n",
693 (val
& PCI_EXP_LNKSTA_CLS_2_5GB
) ? "2.5GB" : "5GB");
695 eeh_edev_dbg(edev
, "Link not ready (0x%04x)\n", val
);
698 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
699 #define SAVED_BYTE(OFF) (((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
701 static void eeh_restore_bridge_bars(struct eeh_dev
*edev
)
706 * Device BARs: 0x10 - 0x18
707 * Bus numbers and windows: 0x18 - 0x30
709 for (i
= 4; i
< 13; i
++)
710 eeh_ops
->write_config(edev
, i
*4, 4, edev
->config_space
[i
]);
712 eeh_ops
->write_config(edev
, 14*4, 4, edev
->config_space
[14]);
714 /* Cache line & Latency timer: 0xC 0xD */
715 eeh_ops
->write_config(edev
, PCI_CACHE_LINE_SIZE
, 1,
716 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
717 eeh_ops
->write_config(edev
, PCI_LATENCY_TIMER
, 1,
718 SAVED_BYTE(PCI_LATENCY_TIMER
));
719 /* Max latency, min grant, interrupt ping and line: 0x3C */
720 eeh_ops
->write_config(edev
, 15*4, 4, edev
->config_space
[15]);
722 /* PCI Command: 0x4 */
723 eeh_ops
->write_config(edev
, PCI_COMMAND
, 4, edev
->config_space
[1] |
724 PCI_COMMAND_MEMORY
| PCI_COMMAND_MASTER
);
726 /* Check the PCIe link is ready */
727 eeh_bridge_check_link(edev
);
730 static void eeh_restore_device_bars(struct eeh_dev
*edev
)
735 for (i
= 4; i
< 10; i
++)
736 eeh_ops
->write_config(edev
, i
*4, 4, edev
->config_space
[i
]);
737 /* 12 == Expansion ROM Address */
738 eeh_ops
->write_config(edev
, 12*4, 4, edev
->config_space
[12]);
740 eeh_ops
->write_config(edev
, PCI_CACHE_LINE_SIZE
, 1,
741 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
742 eeh_ops
->write_config(edev
, PCI_LATENCY_TIMER
, 1,
743 SAVED_BYTE(PCI_LATENCY_TIMER
));
745 /* max latency, min grant, interrupt pin and line */
746 eeh_ops
->write_config(edev
, 15*4, 4, edev
->config_space
[15]);
749 * Restore PERR & SERR bits, some devices require it,
750 * don't touch the other command bits
752 eeh_ops
->read_config(edev
, PCI_COMMAND
, 4, &cmd
);
753 if (edev
->config_space
[1] & PCI_COMMAND_PARITY
)
754 cmd
|= PCI_COMMAND_PARITY
;
756 cmd
&= ~PCI_COMMAND_PARITY
;
757 if (edev
->config_space
[1] & PCI_COMMAND_SERR
)
758 cmd
|= PCI_COMMAND_SERR
;
760 cmd
&= ~PCI_COMMAND_SERR
;
761 eeh_ops
->write_config(edev
, PCI_COMMAND
, 4, cmd
);
765 * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
769 * Loads the PCI configuration space base address registers,
770 * the expansion ROM base address, the latency timer, and etc.
771 * from the saved values in the device node.
773 static void eeh_restore_one_device_bars(struct eeh_dev
*edev
, void *flag
)
775 /* Do special restore for bridges */
776 if (edev
->mode
& EEH_DEV_BRIDGE
)
777 eeh_restore_bridge_bars(edev
);
779 eeh_restore_device_bars(edev
);
781 if (eeh_ops
->restore_config
)
782 eeh_ops
->restore_config(edev
);
786 * eeh_pe_restore_bars - Restore the PCI config space info
789 * This routine performs a recursive walk to the children
790 * of this device as well.
792 void eeh_pe_restore_bars(struct eeh_pe
*pe
)
795 * We needn't take the EEH lock since eeh_pe_dev_traverse()
798 eeh_pe_dev_traverse(pe
, eeh_restore_one_device_bars
, NULL
);
802 * eeh_pe_loc_get - Retrieve location code binding to the given PE
805 * Retrieve the location code of the given PE. If the primary PE bus
806 * is root bus, we will grab location code from PHB device tree node
807 * or root port. Otherwise, the upstream bridge's device tree node
808 * of the primary PE bus will be checked for the location code.
810 const char *eeh_pe_loc_get(struct eeh_pe
*pe
)
812 struct pci_bus
*bus
= eeh_pe_bus_get(pe
);
813 struct device_node
*dn
;
814 const char *loc
= NULL
;
817 dn
= pci_bus_to_OF_node(bus
);
823 if (pci_is_root_bus(bus
))
824 loc
= of_get_property(dn
, "ibm,io-base-loc-code", NULL
);
826 loc
= of_get_property(dn
, "ibm,slot-location-code",
839 * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
842 * Retrieve the PCI bus according to the given PE. Basically,
843 * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
844 * primary PCI bus will be retrieved. The parent bus will be
845 * returned for BUS PE. However, we don't have associated PCI
848 struct pci_bus
*eeh_pe_bus_get(struct eeh_pe
*pe
)
850 struct eeh_dev
*edev
;
851 struct pci_dev
*pdev
;
852 struct pci_bus
*bus
= NULL
;
854 if (pe
->type
& EEH_PE_PHB
)
857 /* The primary bus might be cached during probe time */
858 if (pe
->state
& EEH_PE_PRI_BUS
)
861 /* Retrieve the parent PCI bus of first (top) PCI device */
862 edev
= list_first_entry_or_null(&pe
->edevs
, struct eeh_dev
, entry
);
863 pci_lock_rescan_remove();
864 pdev
= eeh_dev_to_pci_dev(edev
);
867 pci_unlock_rescan_remove();