2 * The file intends to implement PE based on the information from
3 * platforms. Basically, there have 3 types of PEs: PHB/Bus/Device.
4 * All the PEs should be organized as hierarchy tree. The first level
5 * of the tree will be associated to existing PHBs since the particular
6 * PE is only meaningful in one PHB domain.
8 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2012.
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/gfp.h>
28 #include <linux/kernel.h>
29 #include <linux/pci.h>
30 #include <linux/string.h>
32 #include <asm/pci-bridge.h>
33 #include <asm/ppc-pci.h>
35 static LIST_HEAD(eeh_phb_pe);
38 * eeh_pe_alloc - Allocate PE
39 * @phb: PCI controller
42 * Allocate PE instance dynamically.
44 static struct eeh_pe *eeh_pe_alloc(struct pci_controller *phb, int type)
49 pe = kzalloc(sizeof(struct eeh_pe), GFP_KERNEL);
52 /* Initialize PHB PE */
55 INIT_LIST_HEAD(&pe->child_list);
56 INIT_LIST_HEAD(&pe->child);
57 INIT_LIST_HEAD(&pe->edevs);
63 * eeh_phb_pe_create - Create PHB PE
64 * @phb: PCI controller
66 * The function should be called while the PHB is detected during
67 * system boot or PCI hotplug in order to create PHB PE.
69 int eeh_phb_pe_create(struct pci_controller *phb)
74 pe = eeh_pe_alloc(phb, EEH_PE_PHB);
76 pr_err("%s: out of memory!\n", __func__);
80 /* Put it into the list */
81 list_add_tail(&pe->child, &eeh_phb_pe);
83 pr_debug("EEH: Add PE for PHB#%d\n", phb->global_number);
89 * eeh_phb_pe_get - Retrieve PHB PE based on the given PHB
90 * @phb: PCI controller
92 * The overall PEs form hierarchy tree. The first layer of the
93 * hierarchy tree is composed of PHB PEs. The function is used
94 * to retrieve the corresponding PHB PE according to the given PHB.
96 struct eeh_pe *eeh_phb_pe_get(struct pci_controller *phb)
100 list_for_each_entry(pe, &eeh_phb_pe, child) {
102 * Actually, we needn't check the type since
103 * the PE for PHB has been determined when that
106 if ((pe->type & EEH_PE_PHB) && pe->phb == phb)
114 * eeh_pe_next - Retrieve the next PE in the tree
118 * The function is used to retrieve the next PE in the
121 static struct eeh_pe *eeh_pe_next(struct eeh_pe *pe,
124 struct list_head *next = pe->child_list.next;
126 if (next == &pe->child_list) {
130 next = pe->child.next;
131 if (next != &pe->parent->child_list)
137 return list_entry(next, struct eeh_pe, child);
141 * eeh_pe_traverse - Traverse PEs in the specified PHB
144 * @flag: extra parameter to callback
146 * The function is used to traverse the specified PE and its
147 * child PEs. The traversing is to be terminated once the
148 * callback returns something other than NULL, or no more PEs
151 void *eeh_pe_traverse(struct eeh_pe *root,
152 eeh_traverse_func fn, void *flag)
157 for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
166 * eeh_pe_dev_traverse - Traverse the devices from the PE
168 * @fn: function callback
169 * @flag: extra parameter to callback
171 * The function is used to traverse the devices of the specified
172 * PE and its child PEs.
174 void *eeh_pe_dev_traverse(struct eeh_pe *root,
175 eeh_traverse_func fn, void *flag)
178 struct eeh_dev *edev, *tmp;
182 pr_warning("%s: Invalid PE %p\n", __func__, root);
186 /* Traverse root PE */
187 for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
188 eeh_pe_for_each_dev(pe, edev, tmp) {
189 ret = fn(edev, flag);
199 * __eeh_pe_get - Check the PE address
203 * For one particular PE, it can be identified by PE address
204 * or tranditional BDF address. BDF address is composed of
205 * Bus/Device/Function number. The extra data referred by flag
206 * indicates which type of address should be used.
208 static void *__eeh_pe_get(void *data, void *flag)
210 struct eeh_pe *pe = (struct eeh_pe *)data;
211 struct eeh_dev *edev = (struct eeh_dev *)flag;
213 /* Unexpected PHB PE */
214 if (pe->type & EEH_PE_PHB)
217 /* We prefer PE address */
218 if (edev->pe_config_addr &&
219 (edev->pe_config_addr == pe->addr))
222 /* Try BDF address */
223 if (edev->config_addr &&
224 (edev->config_addr == pe->config_addr))
231 * eeh_pe_get - Search PE based on the given address
234 * Search the corresponding PE based on the specified address which
235 * is included in the eeh device. The function is used to check if
236 * the associated PE has been created against the PE address. It's
237 * notable that the PE address has 2 format: traditional PE address
238 * which is composed of PCI bus/device/function number, or unified
241 struct eeh_pe *eeh_pe_get(struct eeh_dev *edev)
243 struct eeh_pe *root = eeh_phb_pe_get(edev->phb);
246 pe = eeh_pe_traverse(root, __eeh_pe_get, edev);
252 * eeh_pe_get_parent - Retrieve the parent PE
255 * The whole PEs existing in the system are organized as hierarchy
256 * tree. The function is used to retrieve the parent PE according
257 * to the parent EEH device.
259 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
261 struct device_node *dn;
262 struct eeh_dev *parent;
265 * It might have the case for the indirect parent
266 * EEH device already having associated PE, but
267 * the direct parent EEH device doesn't have yet.
269 dn = edev->dn->parent;
271 /* We're poking out of PCI territory */
272 if (!PCI_DN(dn)) return NULL;
274 parent = of_node_to_eeh_dev(dn);
275 /* We're poking out of PCI territory */
276 if (!parent) return NULL;
288 * eeh_add_to_parent_pe - Add EEH device to parent PE
291 * Add EEH device to the parent PE. If the parent PE already
292 * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
293 * we have to create new PE to hold the EEH device and the new
294 * PE will be linked to its parent PE as well.
296 int eeh_add_to_parent_pe(struct eeh_dev *edev)
298 struct eeh_pe *pe, *parent;
301 * Search the PE has been existing or not according
302 * to the PE address. If that has been existing, the
303 * PE should be composed of PCI bus and its subordinate
306 pe = eeh_pe_get(edev);
307 if (pe && !(pe->type & EEH_PE_INVALID)) {
308 if (!edev->pe_config_addr) {
309 pr_err("%s: PE with addr 0x%x already exists\n",
310 __func__, edev->config_addr);
314 /* Mark the PE as type of PCI bus */
315 pe->type = EEH_PE_BUS;
318 /* Put the edev to PE */
319 list_add_tail(&edev->list, &pe->edevs);
320 pr_debug("EEH: Add %s to Bus PE#%x\n",
321 edev->dn->full_name, pe->addr);
324 } else if (pe && (pe->type & EEH_PE_INVALID)) {
325 list_add_tail(&edev->list, &pe->edevs);
328 * We're running to here because of PCI hotplug caused by
329 * EEH recovery. We need clear EEH_PE_INVALID until the top.
333 if (!(parent->type & EEH_PE_INVALID))
335 parent->type &= ~(EEH_PE_INVALID | EEH_PE_KEEP);
336 parent = parent->parent;
338 pr_debug("EEH: Add %s to Device PE#%x, Parent PE#%x\n",
339 edev->dn->full_name, pe->addr, pe->parent->addr);
344 /* Create a new EEH PE */
345 pe = eeh_pe_alloc(edev->phb, EEH_PE_DEVICE);
347 pr_err("%s: out of memory!\n", __func__);
350 pe->addr = edev->pe_config_addr;
351 pe->config_addr = edev->config_addr;
354 * While doing PE reset, we probably hot-reset the
355 * upstream bridge. However, the PCI devices including
356 * the associated EEH devices might be removed when EEH
357 * core is doing recovery. So that won't safe to retrieve
358 * the bridge through downstream EEH device. We have to
359 * trace the parent PCI bus, then the upstream bridge.
361 if (eeh_probe_mode_dev())
362 pe->bus = eeh_dev_to_pci_dev(edev)->bus;
365 * Put the new EEH PE into hierarchy tree. If the parent
366 * can't be found, the newly created PE will be attached
367 * to PHB directly. Otherwise, we have to associate the
368 * PE with its parent.
370 parent = eeh_pe_get_parent(edev);
372 parent = eeh_phb_pe_get(edev->phb);
374 pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
375 __func__, edev->phb->global_number);
384 * Put the newly created PE into the child list and
385 * link the EEH device accordingly.
387 list_add_tail(&pe->child, &parent->child_list);
388 list_add_tail(&edev->list, &pe->edevs);
390 pr_debug("EEH: Add %s to Device PE#%x, Parent PE#%x\n",
391 edev->dn->full_name, pe->addr, pe->parent->addr);
397 * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
400 * The PE hierarchy tree might be changed when doing PCI hotplug.
401 * Also, the PCI devices or buses could be removed from the system
402 * during EEH recovery. So we have to call the function remove the
403 * corresponding PE accordingly if necessary.
405 int eeh_rmv_from_parent_pe(struct eeh_dev *edev)
407 struct eeh_pe *pe, *parent, *child;
411 pr_debug("%s: No PE found for EEH device %s\n",
412 __func__, edev->dn->full_name);
416 /* Remove the EEH device */
419 list_del(&edev->list);
422 * Check if the parent PE includes any EEH devices.
423 * If not, we should delete that. Also, we should
424 * delete the parent PE if it doesn't have associated
425 * child PEs and EEH devices.
429 if (pe->type & EEH_PE_PHB)
432 if (!(pe->state & EEH_PE_KEEP)) {
433 if (list_empty(&pe->edevs) &&
434 list_empty(&pe->child_list)) {
435 list_del(&pe->child);
441 if (list_empty(&pe->edevs)) {
443 list_for_each_entry(child, &pe->child_list, child) {
444 if (!(child->type & EEH_PE_INVALID)) {
451 pe->type |= EEH_PE_INVALID;
464 * eeh_pe_update_time_stamp - Update PE's frozen time stamp
467 * We have time stamp for each PE to trace its time of getting
468 * frozen in last hour. The function should be called to update
469 * the time stamp on first error of the specific PE. On the other
470 * handle, we needn't account for errors happened in last hour.
472 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
474 struct timeval tstamp;
478 if (pe->freeze_count <= 0) {
479 pe->freeze_count = 0;
480 do_gettimeofday(&pe->tstamp);
482 do_gettimeofday(&tstamp);
483 if (tstamp.tv_sec - pe->tstamp.tv_sec > 3600) {
485 pe->freeze_count = 0;
491 * __eeh_pe_state_mark - Mark the state for the PE
495 * The function is used to mark the indicated state for the given
496 * PE. Also, the associated PCI devices will be put into IO frozen
499 static void *__eeh_pe_state_mark(void *data, void *flag)
501 struct eeh_pe *pe = (struct eeh_pe *)data;
502 int state = *((int *)flag);
503 struct eeh_dev *edev, *tmp;
504 struct pci_dev *pdev;
507 * Mark the PE with the indicated state. Also,
508 * the associated PCI device will be put into
509 * I/O frozen state to avoid I/O accesses from
510 * the PCI device driver.
513 eeh_pe_for_each_dev(pe, edev, tmp) {
514 pdev = eeh_dev_to_pci_dev(edev);
516 pdev->error_state = pci_channel_io_frozen;
523 * eeh_pe_state_mark - Mark specified state for PE and its associated device
526 * EEH error affects the current PE and its child PEs. The function
527 * is used to mark appropriate state for the affected PEs and the
528 * associated devices.
530 void eeh_pe_state_mark(struct eeh_pe *pe, int state)
532 eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
536 * __eeh_pe_state_clear - Clear state for the PE
540 * The function is used to clear the indicated state from the
541 * given PE. Besides, we also clear the check count of the PE
544 static void *__eeh_pe_state_clear(void *data, void *flag)
546 struct eeh_pe *pe = (struct eeh_pe *)data;
547 int state = *((int *)flag);
556 * eeh_pe_state_clear - Clear state for the PE and its children
558 * @state: state to be cleared
560 * When the PE and its children has been recovered from error,
561 * we need clear the error state for that. The function is used
564 void eeh_pe_state_clear(struct eeh_pe *pe, int state)
566 eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
570 * Some PCI bridges (e.g. PLX bridges) have primary/secondary
571 * buses assigned explicitly by firmware, and we probably have
572 * lost that after reset. So we have to delay the check until
573 * the PCI-CFG registers have been restored for the parent
576 * Don't use normal PCI-CFG accessors, which probably has been
577 * blocked on normal path during the stage. So we need utilize
578 * eeh operations, which is always permitted.
580 static void eeh_bridge_check_link(struct eeh_dev *edev,
581 struct device_node *dn)
588 * We only check root port and downstream ports of
591 if (!(edev->mode & (EEH_DEV_ROOT_PORT | EEH_DEV_DS_PORT)))
594 pr_debug("%s: Check PCIe link for %04x:%02x:%02x.%01x ...\n",
595 __func__, edev->phb->global_number,
596 edev->config_addr >> 8,
597 PCI_SLOT(edev->config_addr & 0xFF),
598 PCI_FUNC(edev->config_addr & 0xFF));
600 /* Check slot status */
601 cap = edev->pcie_cap;
602 eeh_ops->read_config(dn, cap + PCI_EXP_SLTSTA, 2, &val);
603 if (!(val & PCI_EXP_SLTSTA_PDS)) {
604 pr_debug(" No card in the slot (0x%04x) !\n", val);
608 /* Check power status if we have the capability */
609 eeh_ops->read_config(dn, cap + PCI_EXP_SLTCAP, 2, &val);
610 if (val & PCI_EXP_SLTCAP_PCP) {
611 eeh_ops->read_config(dn, cap + PCI_EXP_SLTCTL, 2, &val);
612 if (val & PCI_EXP_SLTCTL_PCC) {
613 pr_debug(" In power-off state, power it on ...\n");
614 val &= ~(PCI_EXP_SLTCTL_PCC | PCI_EXP_SLTCTL_PIC);
615 val |= (0x0100 & PCI_EXP_SLTCTL_PIC);
616 eeh_ops->write_config(dn, cap + PCI_EXP_SLTCTL, 2, val);
622 eeh_ops->read_config(dn, cap + PCI_EXP_LNKCTL, 2, &val);
623 val &= ~PCI_EXP_LNKCTL_LD;
624 eeh_ops->write_config(dn, cap + PCI_EXP_LNKCTL, 2, val);
627 eeh_ops->read_config(dn, cap + PCI_EXP_LNKCAP, 4, &val);
628 if (!(val & PCI_EXP_LNKCAP_DLLLARC)) {
629 pr_debug(" No link reporting capability (0x%08x) \n", val);
634 /* Wait the link is up until timeout (5s) */
636 while (timeout < 5000) {
640 eeh_ops->read_config(dn, cap + PCI_EXP_LNKSTA, 2, &val);
641 if (val & PCI_EXP_LNKSTA_DLLLA)
645 if (val & PCI_EXP_LNKSTA_DLLLA)
646 pr_debug(" Link up (%s)\n",
647 (val & PCI_EXP_LNKSTA_CLS_2_5GB) ? "2.5GB" : "5GB");
649 pr_debug(" Link not ready (0x%04x)\n", val);
652 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
653 #define SAVED_BYTE(OFF) (((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
655 static void eeh_restore_bridge_bars(struct eeh_dev *edev,
656 struct device_node *dn)
661 * Device BARs: 0x10 - 0x18
662 * Bus numbers and windows: 0x18 - 0x30
664 for (i = 4; i < 13; i++)
665 eeh_ops->write_config(dn, i*4, 4, edev->config_space[i]);
667 eeh_ops->write_config(dn, 14*4, 4, edev->config_space[14]);
669 /* Cache line & Latency timer: 0xC 0xD */
670 eeh_ops->write_config(dn, PCI_CACHE_LINE_SIZE, 1,
671 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
672 eeh_ops->write_config(dn, PCI_LATENCY_TIMER, 1,
673 SAVED_BYTE(PCI_LATENCY_TIMER));
674 /* Max latency, min grant, interrupt ping and line: 0x3C */
675 eeh_ops->write_config(dn, 15*4, 4, edev->config_space[15]);
677 /* PCI Command: 0x4 */
678 eeh_ops->write_config(dn, PCI_COMMAND, 4, edev->config_space[1]);
680 /* Check the PCIe link is ready */
681 eeh_bridge_check_link(edev, dn);
684 static void eeh_restore_device_bars(struct eeh_dev *edev,
685 struct device_node *dn)
690 for (i = 4; i < 10; i++)
691 eeh_ops->write_config(dn, i*4, 4, edev->config_space[i]);
692 /* 12 == Expansion ROM Address */
693 eeh_ops->write_config(dn, 12*4, 4, edev->config_space[12]);
695 eeh_ops->write_config(dn, PCI_CACHE_LINE_SIZE, 1,
696 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
697 eeh_ops->write_config(dn, PCI_LATENCY_TIMER, 1,
698 SAVED_BYTE(PCI_LATENCY_TIMER));
700 /* max latency, min grant, interrupt pin and line */
701 eeh_ops->write_config(dn, 15*4, 4, edev->config_space[15]);
704 * Restore PERR & SERR bits, some devices require it,
705 * don't touch the other command bits
707 eeh_ops->read_config(dn, PCI_COMMAND, 4, &cmd);
708 if (edev->config_space[1] & PCI_COMMAND_PARITY)
709 cmd |= PCI_COMMAND_PARITY;
711 cmd &= ~PCI_COMMAND_PARITY;
712 if (edev->config_space[1] & PCI_COMMAND_SERR)
713 cmd |= PCI_COMMAND_SERR;
715 cmd &= ~PCI_COMMAND_SERR;
716 eeh_ops->write_config(dn, PCI_COMMAND, 4, cmd);
720 * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
724 * Loads the PCI configuration space base address registers,
725 * the expansion ROM base address, the latency timer, and etc.
726 * from the saved values in the device node.
728 static void *eeh_restore_one_device_bars(void *data, void *flag)
730 struct eeh_dev *edev = (struct eeh_dev *)data;
731 struct device_node *dn = eeh_dev_to_of_node(edev);
733 /* Do special restore for bridges */
734 if (edev->mode & EEH_DEV_BRIDGE)
735 eeh_restore_bridge_bars(edev, dn);
737 eeh_restore_device_bars(edev, dn);
739 if (eeh_ops->restore_config)
740 eeh_ops->restore_config(dn);
746 * eeh_pe_restore_bars - Restore the PCI config space info
749 * This routine performs a recursive walk to the children
750 * of this device as well.
752 void eeh_pe_restore_bars(struct eeh_pe *pe)
755 * We needn't take the EEH lock since eeh_pe_dev_traverse()
758 eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
762 * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
765 * Retrieve the PCI bus according to the given PE. Basically,
766 * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
767 * primary PCI bus will be retrieved. The parent bus will be
768 * returned for BUS PE. However, we don't have associated PCI
771 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
773 struct pci_bus *bus = NULL;
774 struct eeh_dev *edev;
775 struct pci_dev *pdev;
777 if (pe->type & EEH_PE_PHB) {
779 } else if (pe->type & EEH_PE_BUS ||
780 pe->type & EEH_PE_DEVICE) {
786 edev = list_first_entry(&pe->edevs, struct eeh_dev, list);
787 pdev = eeh_dev_to_pci_dev(edev);