3 * Copyright (C) 2001 Dave Engebretsen & Todd Inglett IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/delay.h>
21 #include <linux/init.h>
22 #include <linux/list.h>
23 #include <linux/pci.h>
24 #include <linux/proc_fs.h>
25 #include <linux/rbtree.h>
26 #include <linux/seq_file.h>
27 #include <linux/spinlock.h>
28 #include <asm/atomic.h>
30 #include <asm/eeh_event.h>
32 #include <asm/machdep.h>
33 #include <asm/ppc-pci.h>
39 * EEH, or "Extended Error Handling" is a PCI bridge technology for
40 * dealing with PCI bus errors that can't be dealt with within the
41 * usual PCI framework, except by check-stopping the CPU. Systems
42 * that are designed for high-availability/reliability cannot afford
43 * to crash due to a "mere" PCI error, thus the need for EEH.
44 * An EEH-capable bridge operates by converting a detected error
45 * into a "slot freeze", taking the PCI adapter off-line, making
46 * the slot behave, from the OS'es point of view, as if the slot
47 * were "empty": all reads return 0xff's and all writes are silently
48 * ignored. EEH slot isolation events can be triggered by parity
49 * errors on the address or data busses (e.g. during posted writes),
50 * which in turn might be caused by low voltage on the bus, dust,
51 * vibration, humidity, radioactivity or plain-old failed hardware.
53 * Note, however, that one of the leading causes of EEH slot
54 * freeze events are buggy device drivers, buggy device microcode,
55 * or buggy device hardware. This is because any attempt by the
56 * device to bus-master data to a memory address that is not
57 * assigned to the device will trigger a slot freeze. (The idea
58 * is to prevent devices-gone-wild from corrupting system memory).
59 * Buggy hardware/drivers will have a miserable time co-existing
62 * Ideally, a PCI device driver, when suspecting that an isolation
63 * event has occured (e.g. by reading 0xff's), will then ask EEH
64 * whether this is the case, and then take appropriate steps to
65 * reset the PCI slot, the PCI device, and then resume operations.
66 * However, until that day, the checking is done here, with the
67 * eeh_check_failure() routine embedded in the MMIO macros. If
68 * the slot is found to be isolated, an "EEH Event" is synthesized
69 * and sent out for processing.
72 /* If a device driver keeps reading an MMIO register in an interrupt
73 * handler after a slot isolation event has occurred, we assume it
74 * is broken and panic. This sets the threshold for how many read
75 * attempts we allow before panicking.
77 #define EEH_MAX_FAILS 2100000
79 /* Time to wait for a PCI slot to retport status, in milliseconds */
80 #define PCI_BUS_RESET_WAIT_MSEC (60*1000)
83 static int ibm_set_eeh_option;
84 static int ibm_set_slot_reset;
85 static int ibm_read_slot_reset_state;
86 static int ibm_read_slot_reset_state2;
87 static int ibm_slot_error_detail;
88 static int ibm_get_config_addr_info;
89 static int ibm_get_config_addr_info2;
90 static int ibm_configure_bridge;
92 int eeh_subsystem_enabled;
93 EXPORT_SYMBOL(eeh_subsystem_enabled);
95 /* Lock to avoid races due to multiple reports of an error */
96 static DEFINE_SPINLOCK(confirm_error_lock);
98 /* Buffer for reporting slot-error-detail rtas calls */
99 static unsigned char slot_errbuf[RTAS_ERROR_LOG_MAX];
100 static DEFINE_SPINLOCK(slot_errbuf_lock);
101 static int eeh_error_buf_size;
103 #define EEH_PCI_REGS_LOG_LEN 4096
104 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
106 /* System monitoring statistics */
107 static unsigned long no_device;
108 static unsigned long no_dn;
109 static unsigned long no_cfg_addr;
110 static unsigned long ignored_check;
111 static unsigned long total_mmio_ffs;
112 static unsigned long false_positives;
113 static unsigned long ignored_failures;
114 static unsigned long slot_resets;
116 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
118 /* --------------------------------------------------------------- */
119 /* Below lies the EEH event infrastructure */
121 static void rtas_slot_error_detail(struct pci_dn *pdn, int severity,
122 char *driver_log, size_t loglen)
128 /* Log the error with the rtas logger */
129 spin_lock_irqsave(&slot_errbuf_lock, flags);
130 memset(slot_errbuf, 0, eeh_error_buf_size);
132 /* Use PE configuration address, if present */
133 config_addr = pdn->eeh_config_addr;
134 if (pdn->eeh_pe_config_addr)
135 config_addr = pdn->eeh_pe_config_addr;
137 rc = rtas_call(ibm_slot_error_detail,
138 8, 1, NULL, config_addr,
139 BUID_HI(pdn->phb->buid),
140 BUID_LO(pdn->phb->buid),
141 virt_to_phys(driver_log), loglen,
142 virt_to_phys(slot_errbuf),
147 log_error(slot_errbuf, ERR_TYPE_RTAS_LOG, 0);
148 spin_unlock_irqrestore(&slot_errbuf_lock, flags);
152 * gather_pci_data - copy assorted PCI config space registers to buff
153 * @pdn: device to report data for
154 * @buf: point to buffer in which to log
155 * @len: amount of room in buffer
157 * This routine captures assorted PCI configuration space data,
158 * and puts them into a buffer for RTAS error logging.
160 static size_t gather_pci_data(struct pci_dn *pdn, char * buf, size_t len)
165 n += scnprintf(buf+n, len-n, "%s\n", pdn->node->name);
166 rtas_read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
167 n += scnprintf(buf+n, len-n, "dev/vend:%x\n", cfg);
168 rtas_read_config(pdn, PCI_COMMAND, 4, &cfg);
169 n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
174 void eeh_slot_error_detail(struct pci_dn *pdn, int severity)
177 memset(pci_regs_buf, 0, EEH_PCI_REGS_LOG_LEN);
179 rtas_pci_enable(pdn, EEH_THAW_MMIO);
180 loglen = gather_pci_data(pdn, pci_regs_buf, EEH_PCI_REGS_LOG_LEN);
182 rtas_slot_error_detail(pdn, severity, pci_regs_buf, loglen);
186 * read_slot_reset_state - Read the reset state of a device node's slot
187 * @dn: device node to read
188 * @rets: array to return results in
190 static int read_slot_reset_state(struct pci_dn *pdn, int rets[])
195 if (ibm_read_slot_reset_state2 != RTAS_UNKNOWN_SERVICE) {
196 token = ibm_read_slot_reset_state2;
199 token = ibm_read_slot_reset_state;
200 rets[2] = 0; /* fake PE Unavailable info */
204 /* Use PE configuration address, if present */
205 config_addr = pdn->eeh_config_addr;
206 if (pdn->eeh_pe_config_addr)
207 config_addr = pdn->eeh_pe_config_addr;
209 return rtas_call(token, 3, outputs, rets, config_addr,
210 BUID_HI(pdn->phb->buid), BUID_LO(pdn->phb->buid));
214 * eeh_wait_for_slot_status - returns error status of slot
215 * @pdn pci device node
216 * @max_wait_msecs maximum number to millisecs to wait
218 * Return negative value if a permanent error, else return
219 * Partition Endpoint (PE) status value.
221 * If @max_wait_msecs is positive, then this routine will
222 * sleep until a valid status can be obtained, or until
223 * the max allowed wait time is exceeded, in which case
227 eeh_wait_for_slot_status(struct pci_dn *pdn, int max_wait_msecs)
234 rc = read_slot_reset_state(pdn, rets);
236 if (rets[1] == 0) return -1; /* EEH is not supported */
238 if (rets[0] != 5) return rets[0]; /* return actual status */
240 if (rets[2] == 0) return -1; /* permanently unavailable */
242 if (max_wait_msecs <= 0) return -1;
247 "EEH: Firmware returned bad wait value=%d\n", mwait);
249 } else if (mwait > 300*1000) {
251 "EEH: Firmware is taking too long, time=%d\n", mwait);
254 max_wait_msecs -= mwait;
258 printk(KERN_WARNING "EEH: Timed out waiting for slot status\n");
263 * eeh_token_to_phys - convert EEH address token to phys address
264 * @token i/o token, should be address in the form 0xA....
266 static inline unsigned long eeh_token_to_phys(unsigned long token)
271 ptep = find_linux_pte(init_mm.pgd, token);
274 pa = pte_pfn(*ptep) << PAGE_SHIFT;
276 return pa | (token & (PAGE_SIZE-1));
280 * Return the "partitionable endpoint" (pe) under which this device lies
282 struct device_node * find_device_pe(struct device_node *dn)
284 while ((dn->parent) && PCI_DN(dn->parent) &&
285 (PCI_DN(dn->parent)->eeh_mode & EEH_MODE_SUPPORTED)) {
291 /** Mark all devices that are peers of this device as failed.
292 * Mark the device driver too, so that it can see the failure
293 * immediately; this is critical, since some drivers poll
294 * status registers in interrupts ... If a driver is polling,
295 * and the slot is frozen, then the driver can deadlock in
296 * an interrupt context, which is bad.
299 static void __eeh_mark_slot (struct device_node *dn, int mode_flag)
303 /* Mark the pci device driver too */
304 struct pci_dev *dev = PCI_DN(dn)->pcidev;
306 PCI_DN(dn)->eeh_mode |= mode_flag;
308 if (dev && dev->driver)
309 dev->error_state = pci_channel_io_frozen;
312 __eeh_mark_slot (dn->child, mode_flag);
318 void eeh_mark_slot (struct device_node *dn, int mode_flag)
321 dn = find_device_pe (dn);
323 /* Back up one, since config addrs might be shared */
324 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
327 PCI_DN(dn)->eeh_mode |= mode_flag;
329 /* Mark the pci device too */
330 dev = PCI_DN(dn)->pcidev;
332 dev->error_state = pci_channel_io_frozen;
334 __eeh_mark_slot (dn->child, mode_flag);
337 static void __eeh_clear_slot (struct device_node *dn, int mode_flag)
341 PCI_DN(dn)->eeh_mode &= ~mode_flag;
342 PCI_DN(dn)->eeh_check_count = 0;
344 __eeh_clear_slot (dn->child, mode_flag);
350 void eeh_clear_slot (struct device_node *dn, int mode_flag)
353 spin_lock_irqsave(&confirm_error_lock, flags);
355 dn = find_device_pe (dn);
357 /* Back up one, since config addrs might be shared */
358 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
361 PCI_DN(dn)->eeh_mode &= ~mode_flag;
362 PCI_DN(dn)->eeh_check_count = 0;
363 __eeh_clear_slot (dn->child, mode_flag);
364 spin_unlock_irqrestore(&confirm_error_lock, flags);
368 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
370 * @dev pci device, if known
372 * Check for an EEH failure for the given device node. Call this
373 * routine if the result of a read was all 0xff's and you want to
374 * find out if this is due to an EEH slot freeze. This routine
375 * will query firmware for the EEH status.
377 * Returns 0 if there has not been an EEH error; otherwise returns
378 * a non-zero value and queues up a slot isolation event notification.
380 * It is safe to call this routine in an interrupt context.
382 int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
392 if (!eeh_subsystem_enabled)
401 /* Access to IO BARs might get this far and still not want checking. */
402 if (!(pdn->eeh_mode & EEH_MODE_SUPPORTED) ||
403 pdn->eeh_mode & EEH_MODE_NOCHECK) {
406 printk ("EEH:ignored check (%x) for %s %s\n",
407 pdn->eeh_mode, pci_name (dev), dn->full_name);
412 if (!pdn->eeh_config_addr && !pdn->eeh_pe_config_addr) {
417 /* If we already have a pending isolation event for this
418 * slot, we know it's bad already, we don't need to check.
419 * Do this checking under a lock; as multiple PCI devices
420 * in one slot might report errors simultaneously, and we
421 * only want one error recovery routine running.
423 spin_lock_irqsave(&confirm_error_lock, flags);
425 if (pdn->eeh_mode & EEH_MODE_ISOLATED) {
426 pdn->eeh_check_count ++;
427 if (pdn->eeh_check_count >= EEH_MAX_FAILS) {
428 printk (KERN_ERR "EEH: Device driver ignored %d bad reads, panicing\n",
429 pdn->eeh_check_count);
433 /* re-read the slot reset state */
434 if (read_slot_reset_state(pdn, rets) != 0)
435 rets[0] = -1; /* reset state unknown */
437 /* If we are here, then we hit an infinite loop. Stop. */
438 panic("EEH: MMIO halt (%d) on device:%s\n", rets[0], pci_name(dev));
444 * Now test for an EEH failure. This is VERY expensive.
445 * Note that the eeh_config_addr may be a parent device
446 * in the case of a device behind a bridge, or it may be
447 * function zero of a multi-function device.
448 * In any case they must share a common PHB.
450 ret = read_slot_reset_state(pdn, rets);
452 /* If the call to firmware failed, punt */
454 printk(KERN_WARNING "EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
461 /* Note that config-io to empty slots may fail;
462 * they are empty when they don't have children. */
463 if ((rets[0] == 5) && (dn->child == NULL)) {
469 /* If EEH is not supported on this device, punt. */
471 printk(KERN_WARNING "EEH: event on unsupported device, rc=%d dn=%s\n",
478 /* If not the kind of error we know about, punt. */
479 if (rets[0] != 1 && rets[0] != 2 && rets[0] != 4 && rets[0] != 5) {
487 /* Avoid repeated reports of this failure, including problems
488 * with other functions on this device, and functions under
490 eeh_mark_slot (dn, EEH_MODE_ISOLATED);
491 spin_unlock_irqrestore(&confirm_error_lock, flags);
493 eeh_send_failure_event (dn, dev);
495 /* Most EEH events are due to device driver bugs. Having
496 * a stack trace will help the device-driver authors figure
497 * out what happened. So print that out. */
502 spin_unlock_irqrestore(&confirm_error_lock, flags);
506 EXPORT_SYMBOL_GPL(eeh_dn_check_failure);
509 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
510 * @token i/o token, should be address in the form 0xA....
511 * @val value, should be all 1's (XXX why do we need this arg??)
513 * Check for an EEH failure at the given token address. Call this
514 * routine if the result of a read was all 0xff's and you want to
515 * find out if this is due to an EEH slot freeze event. This routine
516 * will query firmware for the EEH status.
518 * Note this routine is safe to call in an interrupt context.
520 unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
524 struct device_node *dn;
526 /* Finding the phys addr + pci device; this is pretty quick. */
527 addr = eeh_token_to_phys((unsigned long __force) token);
528 dev = pci_get_device_by_addr(addr);
534 dn = pci_device_to_OF_node(dev);
535 eeh_dn_check_failure (dn, dev);
541 EXPORT_SYMBOL(eeh_check_failure);
543 /* ------------------------------------------------------------- */
544 /* The code below deals with error recovery */
547 * rtas_pci_enable - enable MMIO or DMA transfers for this slot
548 * @pdn pci device node
552 rtas_pci_enable(struct pci_dn *pdn, int function)
557 /* Use PE configuration address, if present */
558 config_addr = pdn->eeh_config_addr;
559 if (pdn->eeh_pe_config_addr)
560 config_addr = pdn->eeh_pe_config_addr;
562 rc = rtas_call(ibm_set_eeh_option, 4, 1, NULL,
564 BUID_HI(pdn->phb->buid),
565 BUID_LO(pdn->phb->buid),
569 printk(KERN_WARNING "EEH: Unexpected state change %d, err=%d dn=%s\n",
570 function, rc, pdn->node->full_name);
572 rc = eeh_wait_for_slot_status (pdn, PCI_BUS_RESET_WAIT_MSEC);
573 if ((rc == 4) && (function == EEH_THAW_MMIO))
580 * rtas_pci_slot_reset - raises/lowers the pci #RST line
581 * @pdn pci device node
582 * @state: 1/0 to raise/lower the #RST
584 * Clear the EEH-frozen condition on a slot. This routine
585 * asserts the PCI #RST line if the 'state' argument is '1',
586 * and drops the #RST line if 'state is '0'. This routine is
587 * safe to call in an interrupt context.
592 rtas_pci_slot_reset(struct pci_dn *pdn, int state)
600 printk (KERN_WARNING "EEH: in slot reset, device node %s has no phb\n",
601 pdn->node->full_name);
605 /* Use PE configuration address, if present */
606 config_addr = pdn->eeh_config_addr;
607 if (pdn->eeh_pe_config_addr)
608 config_addr = pdn->eeh_pe_config_addr;
610 rc = rtas_call(ibm_set_slot_reset,4,1, NULL,
612 BUID_HI(pdn->phb->buid),
613 BUID_LO(pdn->phb->buid),
616 printk (KERN_WARNING "EEH: Unable to reset the failed slot,"
617 " (%d) #RST=%d dn=%s\n",
618 rc, state, pdn->node->full_name);
622 * pcibios_set_pcie_slot_reset - Set PCI-E reset state
623 * @dev: pci device struct
624 * @state: reset state to enter
629 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
631 struct device_node *dn = pci_device_to_OF_node(dev);
632 struct pci_dn *pdn = PCI_DN(dn);
635 case pcie_deassert_reset:
636 rtas_pci_slot_reset(pdn, 0);
639 rtas_pci_slot_reset(pdn, 1);
641 case pcie_warm_reset:
642 rtas_pci_slot_reset(pdn, 3);
652 * rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
653 * @pdn: pci device node to be reset.
655 * Return 0 if success, else a non-zero value.
658 static void __rtas_set_slot_reset(struct pci_dn *pdn)
660 rtas_pci_slot_reset (pdn, 1);
662 /* The PCI bus requires that the reset be held high for at least
663 * a 100 milliseconds. We wait a bit longer 'just in case'. */
665 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
666 msleep (PCI_BUS_RST_HOLD_TIME_MSEC);
668 /* We might get hit with another EEH freeze as soon as the
669 * pci slot reset line is dropped. Make sure we don't miss
670 * these, and clear the flag now. */
671 eeh_clear_slot (pdn->node, EEH_MODE_ISOLATED);
673 rtas_pci_slot_reset (pdn, 0);
675 /* After a PCI slot has been reset, the PCI Express spec requires
676 * a 1.5 second idle time for the bus to stabilize, before starting
678 #define PCI_BUS_SETTLE_TIME_MSEC 1800
679 msleep (PCI_BUS_SETTLE_TIME_MSEC);
682 int rtas_set_slot_reset(struct pci_dn *pdn)
686 /* Take three shots at resetting the bus */
687 for (i=0; i<3; i++) {
688 __rtas_set_slot_reset(pdn);
690 rc = eeh_wait_for_slot_status(pdn, PCI_BUS_RESET_WAIT_MSEC);
695 printk (KERN_ERR "EEH: unrecoverable slot failure %s\n",
696 pdn->node->full_name);
699 printk (KERN_ERR "EEH: bus reset %d failed on slot %s\n",
700 i+1, pdn->node->full_name);
706 /* ------------------------------------------------------- */
707 /** Save and restore of PCI BARs
709 * Although firmware will set up BARs during boot, it doesn't
710 * set up device BAR's after a device reset, although it will,
711 * if requested, set up bridge configuration. Thus, we need to
712 * configure the PCI devices ourselves.
716 * __restore_bars - Restore the Base Address Registers
717 * @pdn: pci device node
719 * Loads the PCI configuration space base address registers,
720 * the expansion ROM base address, the latency timer, and etc.
721 * from the saved values in the device node.
723 static inline void __restore_bars (struct pci_dn *pdn)
727 if (NULL==pdn->phb) return;
728 for (i=4; i<10; i++) {
729 rtas_write_config(pdn, i*4, 4, pdn->config_space[i]);
732 /* 12 == Expansion ROM Address */
733 rtas_write_config(pdn, 12*4, 4, pdn->config_space[12]);
735 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
736 #define SAVED_BYTE(OFF) (((u8 *)(pdn->config_space))[BYTE_SWAP(OFF)])
738 rtas_write_config (pdn, PCI_CACHE_LINE_SIZE, 1,
739 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
741 rtas_write_config (pdn, PCI_LATENCY_TIMER, 1,
742 SAVED_BYTE(PCI_LATENCY_TIMER));
744 /* max latency, min grant, interrupt pin and line */
745 rtas_write_config(pdn, 15*4, 4, pdn->config_space[15]);
749 * eeh_restore_bars - restore the PCI config space info
751 * This routine performs a recursive walk to the children
752 * of this device as well.
754 void eeh_restore_bars(struct pci_dn *pdn)
756 struct device_node *dn;
760 if ((pdn->eeh_mode & EEH_MODE_SUPPORTED) && !IS_BRIDGE(pdn->class_code))
761 __restore_bars (pdn);
763 dn = pdn->node->child;
765 eeh_restore_bars (PCI_DN(dn));
771 * eeh_save_bars - save device bars
773 * Save the values of the device bars. Unlike the restore
774 * routine, this routine is *not* recursive. This is because
775 * PCI devices are added individuallly; but, for the restore,
776 * an entire slot is reset at a time.
778 static void eeh_save_bars(struct pci_dn *pdn)
785 for (i = 0; i < 16; i++)
786 rtas_read_config(pdn, i * 4, 4, &pdn->config_space[i]);
790 rtas_configure_bridge(struct pci_dn *pdn)
795 /* Use PE configuration address, if present */
796 config_addr = pdn->eeh_config_addr;
797 if (pdn->eeh_pe_config_addr)
798 config_addr = pdn->eeh_pe_config_addr;
800 rc = rtas_call(ibm_configure_bridge,3,1, NULL,
802 BUID_HI(pdn->phb->buid),
803 BUID_LO(pdn->phb->buid));
805 printk (KERN_WARNING "EEH: Unable to configure device bridge (%d) for %s\n",
806 rc, pdn->node->full_name);
810 /* ------------------------------------------------------------- */
811 /* The code below deals with enabling EEH for devices during the
812 * early boot sequence. EEH must be enabled before any PCI probing
818 struct eeh_early_enable_info {
819 unsigned int buid_hi;
820 unsigned int buid_lo;
823 static int get_pe_addr (int config_addr,
824 struct eeh_early_enable_info *info)
826 unsigned int rets[3];
829 /* Use latest config-addr token on power6 */
830 if (ibm_get_config_addr_info2 != RTAS_UNKNOWN_SERVICE) {
831 /* Make sure we have a PE in hand */
832 ret = rtas_call (ibm_get_config_addr_info2, 4, 2, rets,
833 config_addr, info->buid_hi, info->buid_lo, 1);
834 if (ret || (rets[0]==0))
837 ret = rtas_call (ibm_get_config_addr_info2, 4, 2, rets,
838 config_addr, info->buid_hi, info->buid_lo, 0);
844 /* Use older config-addr token on power5 */
845 if (ibm_get_config_addr_info != RTAS_UNKNOWN_SERVICE) {
846 ret = rtas_call (ibm_get_config_addr_info, 4, 2, rets,
847 config_addr, info->buid_hi, info->buid_lo, 0);
855 /* Enable eeh for the given device node. */
856 static void *early_enable_eeh(struct device_node *dn, void *data)
858 unsigned int rets[3];
859 struct eeh_early_enable_info *info = data;
861 const char *status = of_get_property(dn, "status", NULL);
862 const u32 *class_code = of_get_property(dn, "class-code", NULL);
863 const u32 *vendor_id = of_get_property(dn, "vendor-id", NULL);
864 const u32 *device_id = of_get_property(dn, "device-id", NULL);
867 struct pci_dn *pdn = PCI_DN(dn);
871 pdn->eeh_check_count = 0;
872 pdn->eeh_freeze_count = 0;
874 if (status && strcmp(status, "ok") != 0)
875 return NULL; /* ignore devices with bad status */
877 /* Ignore bad nodes. */
878 if (!class_code || !vendor_id || !device_id)
881 /* There is nothing to check on PCI to ISA bridges */
882 if (dn->type && !strcmp(dn->type, "isa")) {
883 pdn->eeh_mode |= EEH_MODE_NOCHECK;
886 pdn->class_code = *class_code;
889 * Now decide if we are going to "Disable" EEH checking
890 * for this device. We still run with the EEH hardware active,
891 * but we won't be checking for ff's. This means a driver
892 * could return bad data (very bad!), an interrupt handler could
893 * hang waiting on status bits that won't change, etc.
894 * But there are a few cases like display devices that make sense.
896 enable = 1; /* i.e. we will do checking */
898 if ((*class_code >> 16) == PCI_BASE_CLASS_DISPLAY)
903 pdn->eeh_mode |= EEH_MODE_NOCHECK;
905 /* Ok... see if this device supports EEH. Some do, some don't,
906 * and the only way to find out is to check each and every one. */
907 regs = of_get_property(dn, "reg", NULL);
909 /* First register entry is addr (00BBSS00) */
910 /* Try to enable eeh */
911 ret = rtas_call(ibm_set_eeh_option, 4, 1, NULL,
912 regs[0], info->buid_hi, info->buid_lo,
917 pdn->eeh_config_addr = regs[0];
919 /* If the newer, better, ibm,get-config-addr-info is supported,
920 * then use that instead. */
921 pdn->eeh_pe_config_addr = get_pe_addr(pdn->eeh_config_addr, info);
923 /* Some older systems (Power4) allow the
924 * ibm,set-eeh-option call to succeed even on nodes
925 * where EEH is not supported. Verify support
927 ret = read_slot_reset_state(pdn, rets);
928 if ((ret == 0) && (rets[1] == 1))
933 eeh_subsystem_enabled = 1;
934 pdn->eeh_mode |= EEH_MODE_SUPPORTED;
937 printk(KERN_DEBUG "EEH: %s: eeh enabled, config=%x pe_config=%x\n",
938 dn->full_name, pdn->eeh_config_addr, pdn->eeh_pe_config_addr);
942 /* This device doesn't support EEH, but it may have an
943 * EEH parent, in which case we mark it as supported. */
944 if (dn->parent && PCI_DN(dn->parent)
945 && (PCI_DN(dn->parent)->eeh_mode & EEH_MODE_SUPPORTED)) {
946 /* Parent supports EEH. */
947 pdn->eeh_mode |= EEH_MODE_SUPPORTED;
948 pdn->eeh_config_addr = PCI_DN(dn->parent)->eeh_config_addr;
953 printk(KERN_WARNING "EEH: %s: unable to get reg property.\n",
962 * Initialize EEH by trying to enable it for all of the adapters in the system.
963 * As a side effect we can determine here if eeh is supported at all.
964 * Note that we leave EEH on so failed config cycles won't cause a machine
965 * check. If a user turns off EEH for a particular adapter they are really
966 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
967 * grant access to a slot if EEH isn't enabled, and so we always enable
968 * EEH for all slots/all devices.
970 * The eeh-force-off option disables EEH checking globally, for all slots.
971 * Even if force-off is set, the EEH hardware is still enabled, so that
972 * newer systems can boot.
974 void __init eeh_init(void)
976 struct device_node *phb, *np;
977 struct eeh_early_enable_info info;
979 spin_lock_init(&confirm_error_lock);
980 spin_lock_init(&slot_errbuf_lock);
982 np = of_find_node_by_path("/rtas");
986 ibm_set_eeh_option = rtas_token("ibm,set-eeh-option");
987 ibm_set_slot_reset = rtas_token("ibm,set-slot-reset");
988 ibm_read_slot_reset_state2 = rtas_token("ibm,read-slot-reset-state2");
989 ibm_read_slot_reset_state = rtas_token("ibm,read-slot-reset-state");
990 ibm_slot_error_detail = rtas_token("ibm,slot-error-detail");
991 ibm_get_config_addr_info = rtas_token("ibm,get-config-addr-info");
992 ibm_get_config_addr_info2 = rtas_token("ibm,get-config-addr-info2");
993 ibm_configure_bridge = rtas_token ("ibm,configure-bridge");
995 if (ibm_set_eeh_option == RTAS_UNKNOWN_SERVICE)
998 eeh_error_buf_size = rtas_token("rtas-error-log-max");
999 if (eeh_error_buf_size == RTAS_UNKNOWN_SERVICE) {
1000 eeh_error_buf_size = 1024;
1002 if (eeh_error_buf_size > RTAS_ERROR_LOG_MAX) {
1003 printk(KERN_WARNING "EEH: rtas-error-log-max is bigger than allocated "
1004 "buffer ! (%d vs %d)", eeh_error_buf_size, RTAS_ERROR_LOG_MAX);
1005 eeh_error_buf_size = RTAS_ERROR_LOG_MAX;
1008 /* Enable EEH for all adapters. Note that eeh requires buid's */
1009 for (phb = of_find_node_by_name(NULL, "pci"); phb;
1010 phb = of_find_node_by_name(phb, "pci")) {
1013 buid = get_phb_buid(phb);
1014 if (buid == 0 || PCI_DN(phb) == NULL)
1017 info.buid_lo = BUID_LO(buid);
1018 info.buid_hi = BUID_HI(buid);
1019 traverse_pci_devices(phb, early_enable_eeh, &info);
1022 if (eeh_subsystem_enabled)
1023 printk(KERN_INFO "EEH: PCI Enhanced I/O Error Handling Enabled\n");
1025 printk(KERN_WARNING "EEH: No capable adapters found\n");
1029 * eeh_add_device_early - enable EEH for the indicated device_node
1030 * @dn: device node for which to set up EEH
1032 * This routine must be used to perform EEH initialization for PCI
1033 * devices that were added after system boot (e.g. hotplug, dlpar).
1034 * This routine must be called before any i/o is performed to the
1035 * adapter (inluding any config-space i/o).
1036 * Whether this actually enables EEH or not for this device depends
1037 * on the CEC architecture, type of the device, on earlier boot
1038 * command-line arguments & etc.
1040 static void eeh_add_device_early(struct device_node *dn)
1042 struct pci_controller *phb;
1043 struct eeh_early_enable_info info;
1045 if (!dn || !PCI_DN(dn))
1047 phb = PCI_DN(dn)->phb;
1049 /* USB Bus children of PCI devices will not have BUID's */
1050 if (NULL == phb || 0 == phb->buid)
1053 info.buid_hi = BUID_HI(phb->buid);
1054 info.buid_lo = BUID_LO(phb->buid);
1055 early_enable_eeh(dn, &info);
1058 void eeh_add_device_tree_early(struct device_node *dn)
1060 struct device_node *sib;
1061 for (sib = dn->child; sib; sib = sib->sibling)
1062 eeh_add_device_tree_early(sib);
1063 eeh_add_device_early(dn);
1065 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1068 * eeh_add_device_late - perform EEH initialization for the indicated pci device
1069 * @dev: pci device for which to set up EEH
1071 * This routine must be used to complete EEH initialization for PCI
1072 * devices that were added after system boot (e.g. hotplug, dlpar).
1074 static void eeh_add_device_late(struct pci_dev *dev)
1076 struct device_node *dn;
1079 if (!dev || !eeh_subsystem_enabled)
1083 printk(KERN_DEBUG "EEH: adding device %s\n", pci_name(dev));
1087 dn = pci_device_to_OF_node(dev);
1091 pci_addr_cache_insert_device (dev);
1094 void eeh_add_device_tree_late(struct pci_bus *bus)
1096 struct pci_dev *dev;
1098 list_for_each_entry(dev, &bus->devices, bus_list) {
1099 eeh_add_device_late(dev);
1100 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1101 struct pci_bus *subbus = dev->subordinate;
1103 eeh_add_device_tree_late(subbus);
1107 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1110 * eeh_remove_device - undo EEH setup for the indicated pci device
1111 * @dev: pci device to be removed
1113 * This routine should be called when a device is removed from
1114 * a running system (e.g. by hotplug or dlpar). It unregisters
1115 * the PCI device from the EEH subsystem. I/O errors affecting
1116 * this device will no longer be detected after this call; thus,
1117 * i/o errors affecting this slot may leave this device unusable.
1119 static void eeh_remove_device(struct pci_dev *dev)
1121 struct device_node *dn;
1122 if (!dev || !eeh_subsystem_enabled)
1125 /* Unregister the device with the EEH/PCI address search system */
1127 printk(KERN_DEBUG "EEH: remove device %s\n", pci_name(dev));
1129 pci_addr_cache_remove_device(dev);
1131 dn = pci_device_to_OF_node(dev);
1132 if (PCI_DN(dn)->pcidev) {
1133 PCI_DN(dn)->pcidev = NULL;
1138 void eeh_remove_bus_device(struct pci_dev *dev)
1140 struct pci_bus *bus = dev->subordinate;
1141 struct pci_dev *child, *tmp;
1143 eeh_remove_device(dev);
1145 if (bus && dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1146 list_for_each_entry_safe(child, tmp, &bus->devices, bus_list)
1147 eeh_remove_bus_device(child);
1150 EXPORT_SYMBOL_GPL(eeh_remove_bus_device);
1152 static int proc_eeh_show(struct seq_file *m, void *v)
1154 if (0 == eeh_subsystem_enabled) {
1155 seq_printf(m, "EEH Subsystem is globally disabled\n");
1156 seq_printf(m, "eeh_total_mmio_ffs=%ld\n", total_mmio_ffs);
1158 seq_printf(m, "EEH Subsystem is enabled\n");
1161 "no device node=%ld\n"
1162 "no config address=%ld\n"
1163 "check not wanted=%ld\n"
1164 "eeh_total_mmio_ffs=%ld\n"
1165 "eeh_false_positives=%ld\n"
1166 "eeh_ignored_failures=%ld\n"
1167 "eeh_slot_resets=%ld\n",
1168 no_device, no_dn, no_cfg_addr,
1169 ignored_check, total_mmio_ffs,
1170 false_positives, ignored_failures,
1177 static int proc_eeh_open(struct inode *inode, struct file *file)
1179 return single_open(file, proc_eeh_show, NULL);
1182 static const struct file_operations proc_eeh_operations = {
1183 .open = proc_eeh_open,
1185 .llseek = seq_lseek,
1186 .release = single_release,
1189 static int __init eeh_init_proc(void)
1191 struct proc_dir_entry *e;
1193 if (machine_is(pseries)) {
1194 e = create_proc_entry("ppc64/eeh", 0, NULL);
1196 e->proc_fops = &proc_eeh_operations;
1201 __initcall(eeh_init_proc);