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 100000
80 static int ibm_set_eeh_option
;
81 static int ibm_set_slot_reset
;
82 static int ibm_read_slot_reset_state
;
83 static int ibm_read_slot_reset_state2
;
84 static int ibm_slot_error_detail
;
85 static int ibm_get_config_addr_info
;
86 static int ibm_configure_bridge
;
88 int eeh_subsystem_enabled
;
89 EXPORT_SYMBOL(eeh_subsystem_enabled
);
91 /* Lock to avoid races due to multiple reports of an error */
92 static DEFINE_SPINLOCK(confirm_error_lock
);
94 /* Buffer for reporting slot-error-detail rtas calls */
95 static unsigned char slot_errbuf
[RTAS_ERROR_LOG_MAX
];
96 static DEFINE_SPINLOCK(slot_errbuf_lock
);
97 static int eeh_error_buf_size
;
99 /* System monitoring statistics */
100 static unsigned long no_device
;
101 static unsigned long no_dn
;
102 static unsigned long no_cfg_addr
;
103 static unsigned long ignored_check
;
104 static unsigned long total_mmio_ffs
;
105 static unsigned long false_positives
;
106 static unsigned long ignored_failures
;
107 static unsigned long slot_resets
;
109 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
111 /* --------------------------------------------------------------- */
112 /* Below lies the EEH event infrastructure */
114 void eeh_slot_error_detail (struct pci_dn
*pdn
, int severity
)
120 /* Log the error with the rtas logger */
121 spin_lock_irqsave(&slot_errbuf_lock
, flags
);
122 memset(slot_errbuf
, 0, eeh_error_buf_size
);
124 /* Use PE configuration address, if present */
125 config_addr
= pdn
->eeh_config_addr
;
126 if (pdn
->eeh_pe_config_addr
)
127 config_addr
= pdn
->eeh_pe_config_addr
;
129 rc
= rtas_call(ibm_slot_error_detail
,
130 8, 1, NULL
, config_addr
,
131 BUID_HI(pdn
->phb
->buid
),
132 BUID_LO(pdn
->phb
->buid
), NULL
, 0,
133 virt_to_phys(slot_errbuf
),
138 log_error(slot_errbuf
, ERR_TYPE_RTAS_LOG
, 0);
139 spin_unlock_irqrestore(&slot_errbuf_lock
, flags
);
143 * read_slot_reset_state - Read the reset state of a device node's slot
144 * @dn: device node to read
145 * @rets: array to return results in
147 static int read_slot_reset_state(struct pci_dn
*pdn
, int rets
[])
152 if (ibm_read_slot_reset_state2
!= RTAS_UNKNOWN_SERVICE
) {
153 token
= ibm_read_slot_reset_state2
;
156 token
= ibm_read_slot_reset_state
;
157 rets
[2] = 0; /* fake PE Unavailable info */
161 /* Use PE configuration address, if present */
162 config_addr
= pdn
->eeh_config_addr
;
163 if (pdn
->eeh_pe_config_addr
)
164 config_addr
= pdn
->eeh_pe_config_addr
;
166 return rtas_call(token
, 3, outputs
, rets
, config_addr
,
167 BUID_HI(pdn
->phb
->buid
), BUID_LO(pdn
->phb
->buid
));
171 * eeh_token_to_phys - convert EEH address token to phys address
172 * @token i/o token, should be address in the form 0xA....
174 static inline unsigned long eeh_token_to_phys(unsigned long token
)
179 ptep
= find_linux_pte(init_mm
.pgd
, token
);
182 pa
= pte_pfn(*ptep
) << PAGE_SHIFT
;
184 return pa
| (token
& (PAGE_SIZE
-1));
188 * Return the "partitionable endpoint" (pe) under which this device lies
190 struct device_node
* find_device_pe(struct device_node
*dn
)
192 while ((dn
->parent
) && PCI_DN(dn
->parent
) &&
193 (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
199 /** Mark all devices that are peers of this device as failed.
200 * Mark the device driver too, so that it can see the failure
201 * immediately; this is critical, since some drivers poll
202 * status registers in interrupts ... If a driver is polling,
203 * and the slot is frozen, then the driver can deadlock in
204 * an interrupt context, which is bad.
207 static void __eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
211 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
213 /* Mark the pci device driver too */
214 struct pci_dev
*dev
= PCI_DN(dn
)->pcidev
;
215 if (dev
&& dev
->driver
)
216 dev
->error_state
= pci_channel_io_frozen
;
219 __eeh_mark_slot (dn
->child
, mode_flag
);
225 void eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
227 dn
= find_device_pe (dn
);
229 /* Back up one, since config addrs might be shared */
230 if (PCI_DN(dn
) && PCI_DN(dn
)->eeh_pe_config_addr
)
233 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
234 __eeh_mark_slot (dn
->child
, mode_flag
);
237 static void __eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
241 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
242 PCI_DN(dn
)->eeh_check_count
= 0;
244 __eeh_clear_slot (dn
->child
, mode_flag
);
250 void eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
253 spin_lock_irqsave(&confirm_error_lock
, flags
);
255 dn
= find_device_pe (dn
);
257 /* Back up one, since config addrs might be shared */
258 if (PCI_DN(dn
) && PCI_DN(dn
)->eeh_pe_config_addr
)
261 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
262 PCI_DN(dn
)->eeh_check_count
= 0;
263 __eeh_clear_slot (dn
->child
, mode_flag
);
264 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
268 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
270 * @dev pci device, if known
272 * Check for an EEH failure for the given device node. Call this
273 * routine if the result of a read was all 0xff's and you want to
274 * find out if this is due to an EEH slot freeze. This routine
275 * will query firmware for the EEH status.
277 * Returns 0 if there has not been an EEH error; otherwise returns
278 * a non-zero value and queues up a slot isolation event notification.
280 * It is safe to call this routine in an interrupt context.
282 int eeh_dn_check_failure(struct device_node
*dn
, struct pci_dev
*dev
)
288 enum pci_channel_state state
;
293 if (!eeh_subsystem_enabled
)
302 /* Access to IO BARs might get this far and still not want checking. */
303 if (!(pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) ||
304 pdn
->eeh_mode
& EEH_MODE_NOCHECK
) {
307 printk ("EEH:ignored check (%x) for %s %s\n",
308 pdn
->eeh_mode
, pci_name (dev
), dn
->full_name
);
313 if (!pdn
->eeh_config_addr
&& !pdn
->eeh_pe_config_addr
) {
318 /* If we already have a pending isolation event for this
319 * slot, we know it's bad already, we don't need to check.
320 * Do this checking under a lock; as multiple PCI devices
321 * in one slot might report errors simultaneously, and we
322 * only want one error recovery routine running.
324 spin_lock_irqsave(&confirm_error_lock
, flags
);
326 if (pdn
->eeh_mode
& EEH_MODE_ISOLATED
) {
327 pdn
->eeh_check_count
++;
328 if (pdn
->eeh_check_count
>= EEH_MAX_FAILS
) {
329 printk (KERN_ERR
"EEH: Device driver ignored %d bad reads, panicing\n",
330 pdn
->eeh_check_count
);
333 /* re-read the slot reset state */
334 if (read_slot_reset_state(pdn
, rets
) != 0)
335 rets
[0] = -1; /* reset state unknown */
337 /* If we are here, then we hit an infinite loop. Stop. */
338 panic("EEH: MMIO halt (%d) on device:%s\n", rets
[0], pci_name(dev
));
344 * Now test for an EEH failure. This is VERY expensive.
345 * Note that the eeh_config_addr may be a parent device
346 * in the case of a device behind a bridge, or it may be
347 * function zero of a multi-function device.
348 * In any case they must share a common PHB.
350 ret
= read_slot_reset_state(pdn
, rets
);
352 /* If the call to firmware failed, punt */
354 printk(KERN_WARNING
"EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
361 /* If EEH is not supported on this device, punt. */
363 printk(KERN_WARNING
"EEH: event on unsupported device, rc=%d dn=%s\n",
370 /* If not the kind of error we know about, punt. */
371 if (rets
[0] != 2 && rets
[0] != 4 && rets
[0] != 5) {
377 /* Note that config-io to empty slots may fail;
378 * we recognize empty because they don't have children. */
379 if ((rets
[0] == 5) && (dn
->child
== NULL
)) {
387 /* Avoid repeated reports of this failure, including problems
388 * with other functions on this device, and functions under
390 eeh_mark_slot (dn
, EEH_MODE_ISOLATED
);
391 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
393 state
= pci_channel_io_normal
;
394 if ((rets
[0] == 2) || (rets
[0] == 4))
395 state
= pci_channel_io_frozen
;
397 state
= pci_channel_io_perm_failure
;
398 eeh_send_failure_event (dn
, dev
, state
, rets
[2]);
400 /* Most EEH events are due to device driver bugs. Having
401 * a stack trace will help the device-driver authors figure
402 * out what happened. So print that out. */
403 if (rets
[0] != 5) dump_stack();
407 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
411 EXPORT_SYMBOL_GPL(eeh_dn_check_failure
);
414 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
415 * @token i/o token, should be address in the form 0xA....
416 * @val value, should be all 1's (XXX why do we need this arg??)
418 * Check for an EEH failure at the given token address. Call this
419 * routine if the result of a read was all 0xff's and you want to
420 * find out if this is due to an EEH slot freeze event. This routine
421 * will query firmware for the EEH status.
423 * Note this routine is safe to call in an interrupt context.
425 unsigned long eeh_check_failure(const volatile void __iomem
*token
, unsigned long val
)
429 struct device_node
*dn
;
431 /* Finding the phys addr + pci device; this is pretty quick. */
432 addr
= eeh_token_to_phys((unsigned long __force
) token
);
433 dev
= pci_get_device_by_addr(addr
);
439 dn
= pci_device_to_OF_node(dev
);
440 eeh_dn_check_failure (dn
, dev
);
446 EXPORT_SYMBOL(eeh_check_failure
);
448 /* ------------------------------------------------------------- */
449 /* The code below deals with error recovery */
451 /** Return negative value if a permanent error, else return
452 * a number of milliseconds to wait until the PCI slot is
456 eeh_slot_availability(struct pci_dn
*pdn
)
461 rc
= read_slot_reset_state(pdn
, rets
);
465 if (rets
[1] == 0) return -1; /* EEH is not supported */
466 if (rets
[0] == 0) return 0; /* Oll Korrect */
468 if (rets
[2] == 0) return -1; /* permanently unavailable */
469 return rets
[2]; /* number of millisecs to wait */
474 printk (KERN_ERR
"EEH: Slot unavailable: rc=%d, rets=%d %d %d\n",
475 rc
, rets
[0], rets
[1], rets
[2]);
479 /** rtas_pci_slot_reset raises/lowers the pci #RST line
480 * state: 1/0 to raise/lower the #RST
482 * Clear the EEH-frozen condition on a slot. This routine
483 * asserts the PCI #RST line if the 'state' argument is '1',
484 * and drops the #RST line if 'state is '0'. This routine is
485 * safe to call in an interrupt context.
490 rtas_pci_slot_reset(struct pci_dn
*pdn
, int state
)
498 printk (KERN_WARNING
"EEH: in slot reset, device node %s has no phb\n",
499 pdn
->node
->full_name
);
503 /* Use PE configuration address, if present */
504 config_addr
= pdn
->eeh_config_addr
;
505 if (pdn
->eeh_pe_config_addr
)
506 config_addr
= pdn
->eeh_pe_config_addr
;
508 rc
= rtas_call(ibm_set_slot_reset
,4,1, NULL
,
510 BUID_HI(pdn
->phb
->buid
),
511 BUID_LO(pdn
->phb
->buid
),
514 printk (KERN_WARNING
"EEH: Unable to reset the failed slot, (%d) #RST=%d dn=%s\n",
515 rc
, state
, pdn
->node
->full_name
);
520 /** rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
521 * dn -- device node to be reset.
523 * Return 0 if success, else a non-zero value.
527 rtas_set_slot_reset(struct pci_dn
*pdn
)
531 rtas_pci_slot_reset (pdn
, 1);
533 /* The PCI bus requires that the reset be held high for at least
534 * a 100 milliseconds. We wait a bit longer 'just in case'. */
536 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
537 msleep (PCI_BUS_RST_HOLD_TIME_MSEC
);
539 /* We might get hit with another EEH freeze as soon as the
540 * pci slot reset line is dropped. Make sure we don't miss
541 * these, and clear the flag now. */
542 eeh_clear_slot (pdn
->node
, EEH_MODE_ISOLATED
);
544 rtas_pci_slot_reset (pdn
, 0);
546 /* After a PCI slot has been reset, the PCI Express spec requires
547 * a 1.5 second idle time for the bus to stabilize, before starting
549 #define PCI_BUS_SETTLE_TIME_MSEC 1800
550 msleep (PCI_BUS_SETTLE_TIME_MSEC
);
552 /* Now double check with the firmware to make sure the device is
553 * ready to be used; if not, wait for recovery. */
554 for (i
=0; i
<10; i
++) {
555 rc
= eeh_slot_availability (pdn
);
557 printk (KERN_ERR
"EEH: failed (%d) to reset slot %s\n", rc
, pdn
->node
->full_name
);
566 rc
= eeh_slot_availability (pdn
);
568 printk (KERN_ERR
"EEH: timeout resetting slot %s\n", pdn
->node
->full_name
);
573 /* ------------------------------------------------------- */
574 /** Save and restore of PCI BARs
576 * Although firmware will set up BARs during boot, it doesn't
577 * set up device BAR's after a device reset, although it will,
578 * if requested, set up bridge configuration. Thus, we need to
579 * configure the PCI devices ourselves.
583 * __restore_bars - Restore the Base Address Registers
584 * Loads the PCI configuration space base address registers,
585 * the expansion ROM base address, the latency timer, and etc.
586 * from the saved values in the device node.
588 static inline void __restore_bars (struct pci_dn
*pdn
)
592 if (NULL
==pdn
->phb
) return;
593 for (i
=4; i
<10; i
++) {
594 rtas_write_config(pdn
, i
*4, 4, pdn
->config_space
[i
]);
597 /* 12 == Expansion ROM Address */
598 rtas_write_config(pdn
, 12*4, 4, pdn
->config_space
[12]);
600 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
601 #define SAVED_BYTE(OFF) (((u8 *)(pdn->config_space))[BYTE_SWAP(OFF)])
603 rtas_write_config (pdn
, PCI_CACHE_LINE_SIZE
, 1,
604 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
606 rtas_write_config (pdn
, PCI_LATENCY_TIMER
, 1,
607 SAVED_BYTE(PCI_LATENCY_TIMER
));
609 /* max latency, min grant, interrupt pin and line */
610 rtas_write_config(pdn
, 15*4, 4, pdn
->config_space
[15]);
614 * eeh_restore_bars - restore the PCI config space info
616 * This routine performs a recursive walk to the children
617 * of this device as well.
619 void eeh_restore_bars(struct pci_dn
*pdn
)
621 struct device_node
*dn
;
625 if ((pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) && !IS_BRIDGE(pdn
->class_code
))
626 __restore_bars (pdn
);
628 dn
= pdn
->node
->child
;
630 eeh_restore_bars (PCI_DN(dn
));
636 * eeh_save_bars - save device bars
638 * Save the values of the device bars. Unlike the restore
639 * routine, this routine is *not* recursive. This is because
640 * PCI devices are added individuallly; but, for the restore,
641 * an entire slot is reset at a time.
643 static void eeh_save_bars(struct pci_dn
*pdn
)
650 for (i
= 0; i
< 16; i
++)
651 rtas_read_config(pdn
, i
* 4, 4, &pdn
->config_space
[i
]);
655 rtas_configure_bridge(struct pci_dn
*pdn
)
660 /* Use PE configuration address, if present */
661 config_addr
= pdn
->eeh_config_addr
;
662 if (pdn
->eeh_pe_config_addr
)
663 config_addr
= pdn
->eeh_pe_config_addr
;
665 rc
= rtas_call(ibm_configure_bridge
,3,1, NULL
,
667 BUID_HI(pdn
->phb
->buid
),
668 BUID_LO(pdn
->phb
->buid
));
670 printk (KERN_WARNING
"EEH: Unable to configure device bridge (%d) for %s\n",
671 rc
, pdn
->node
->full_name
);
675 /* ------------------------------------------------------------- */
676 /* The code below deals with enabling EEH for devices during the
677 * early boot sequence. EEH must be enabled before any PCI probing
683 struct eeh_early_enable_info
{
684 unsigned int buid_hi
;
685 unsigned int buid_lo
;
688 /* Enable eeh for the given device node. */
689 static void *early_enable_eeh(struct device_node
*dn
, void *data
)
691 struct eeh_early_enable_info
*info
= data
;
693 char *status
= get_property(dn
, "status", NULL
);
694 u32
*class_code
= (u32
*)get_property(dn
, "class-code", NULL
);
695 u32
*vendor_id
= (u32
*)get_property(dn
, "vendor-id", NULL
);
696 u32
*device_id
= (u32
*)get_property(dn
, "device-id", NULL
);
699 struct pci_dn
*pdn
= PCI_DN(dn
);
701 pdn
->class_code
= *class_code
;
703 pdn
->eeh_check_count
= 0;
704 pdn
->eeh_freeze_count
= 0;
706 if (status
&& strcmp(status
, "ok") != 0)
707 return NULL
; /* ignore devices with bad status */
709 /* Ignore bad nodes. */
710 if (!class_code
|| !vendor_id
|| !device_id
)
713 /* There is nothing to check on PCI to ISA bridges */
714 if (dn
->type
&& !strcmp(dn
->type
, "isa")) {
715 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
720 * Now decide if we are going to "Disable" EEH checking
721 * for this device. We still run with the EEH hardware active,
722 * but we won't be checking for ff's. This means a driver
723 * could return bad data (very bad!), an interrupt handler could
724 * hang waiting on status bits that won't change, etc.
725 * But there are a few cases like display devices that make sense.
727 enable
= 1; /* i.e. we will do checking */
729 if ((*class_code
>> 16) == PCI_BASE_CLASS_DISPLAY
)
734 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
736 /* Ok... see if this device supports EEH. Some do, some don't,
737 * and the only way to find out is to check each and every one. */
738 regs
= (u32
*)get_property(dn
, "reg", NULL
);
740 /* First register entry is addr (00BBSS00) */
741 /* Try to enable eeh */
742 ret
= rtas_call(ibm_set_eeh_option
, 4, 1, NULL
,
743 regs
[0], info
->buid_hi
, info
->buid_lo
,
747 eeh_subsystem_enabled
= 1;
748 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
749 pdn
->eeh_config_addr
= regs
[0];
751 /* If the newer, better, ibm,get-config-addr-info is supported,
752 * then use that instead. */
753 pdn
->eeh_pe_config_addr
= 0;
754 if (ibm_get_config_addr_info
!= RTAS_UNKNOWN_SERVICE
) {
755 unsigned int rets
[2];
756 ret
= rtas_call (ibm_get_config_addr_info
, 4, 2, rets
,
757 pdn
->eeh_config_addr
,
758 info
->buid_hi
, info
->buid_lo
,
761 pdn
->eeh_pe_config_addr
= rets
[0];
764 printk(KERN_DEBUG
"EEH: %s: eeh enabled, config=%x pe_config=%x\n",
765 dn
->full_name
, pdn
->eeh_config_addr
, pdn
->eeh_pe_config_addr
);
769 /* This device doesn't support EEH, but it may have an
770 * EEH parent, in which case we mark it as supported. */
771 if (dn
->parent
&& PCI_DN(dn
->parent
)
772 && (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
773 /* Parent supports EEH. */
774 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
775 pdn
->eeh_config_addr
= PCI_DN(dn
->parent
)->eeh_config_addr
;
780 printk(KERN_WARNING
"EEH: %s: unable to get reg property.\n",
789 * Initialize EEH by trying to enable it for all of the adapters in the system.
790 * As a side effect we can determine here if eeh is supported at all.
791 * Note that we leave EEH on so failed config cycles won't cause a machine
792 * check. If a user turns off EEH for a particular adapter they are really
793 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
794 * grant access to a slot if EEH isn't enabled, and so we always enable
795 * EEH for all slots/all devices.
797 * The eeh-force-off option disables EEH checking globally, for all slots.
798 * Even if force-off is set, the EEH hardware is still enabled, so that
799 * newer systems can boot.
801 void __init
eeh_init(void)
803 struct device_node
*phb
, *np
;
804 struct eeh_early_enable_info info
;
806 spin_lock_init(&confirm_error_lock
);
807 spin_lock_init(&slot_errbuf_lock
);
809 np
= of_find_node_by_path("/rtas");
813 ibm_set_eeh_option
= rtas_token("ibm,set-eeh-option");
814 ibm_set_slot_reset
= rtas_token("ibm,set-slot-reset");
815 ibm_read_slot_reset_state2
= rtas_token("ibm,read-slot-reset-state2");
816 ibm_read_slot_reset_state
= rtas_token("ibm,read-slot-reset-state");
817 ibm_slot_error_detail
= rtas_token("ibm,slot-error-detail");
818 ibm_get_config_addr_info
= rtas_token("ibm,get-config-addr-info");
819 ibm_configure_bridge
= rtas_token ("ibm,configure-bridge");
821 if (ibm_set_eeh_option
== RTAS_UNKNOWN_SERVICE
)
824 eeh_error_buf_size
= rtas_token("rtas-error-log-max");
825 if (eeh_error_buf_size
== RTAS_UNKNOWN_SERVICE
) {
826 eeh_error_buf_size
= 1024;
828 if (eeh_error_buf_size
> RTAS_ERROR_LOG_MAX
) {
829 printk(KERN_WARNING
"EEH: rtas-error-log-max is bigger than allocated "
830 "buffer ! (%d vs %d)", eeh_error_buf_size
, RTAS_ERROR_LOG_MAX
);
831 eeh_error_buf_size
= RTAS_ERROR_LOG_MAX
;
834 /* Enable EEH for all adapters. Note that eeh requires buid's */
835 for (phb
= of_find_node_by_name(NULL
, "pci"); phb
;
836 phb
= of_find_node_by_name(phb
, "pci")) {
839 buid
= get_phb_buid(phb
);
840 if (buid
== 0 || PCI_DN(phb
) == NULL
)
843 info
.buid_lo
= BUID_LO(buid
);
844 info
.buid_hi
= BUID_HI(buid
);
845 traverse_pci_devices(phb
, early_enable_eeh
, &info
);
848 if (eeh_subsystem_enabled
)
849 printk(KERN_INFO
"EEH: PCI Enhanced I/O Error Handling Enabled\n");
851 printk(KERN_WARNING
"EEH: No capable adapters found\n");
855 * eeh_add_device_early - enable EEH for the indicated device_node
856 * @dn: device node for which to set up EEH
858 * This routine must be used to perform EEH initialization for PCI
859 * devices that were added after system boot (e.g. hotplug, dlpar).
860 * This routine must be called before any i/o is performed to the
861 * adapter (inluding any config-space i/o).
862 * Whether this actually enables EEH or not for this device depends
863 * on the CEC architecture, type of the device, on earlier boot
864 * command-line arguments & etc.
866 void eeh_add_device_early(struct device_node
*dn
)
868 struct pci_controller
*phb
;
869 struct eeh_early_enable_info info
;
871 if (!dn
|| !PCI_DN(dn
))
873 phb
= PCI_DN(dn
)->phb
;
875 /* USB Bus children of PCI devices will not have BUID's */
876 if (NULL
== phb
|| 0 == phb
->buid
)
879 info
.buid_hi
= BUID_HI(phb
->buid
);
880 info
.buid_lo
= BUID_LO(phb
->buid
);
881 early_enable_eeh(dn
, &info
);
883 EXPORT_SYMBOL_GPL(eeh_add_device_early
);
885 void eeh_add_device_tree_early(struct device_node
*dn
)
887 struct device_node
*sib
;
888 for (sib
= dn
->child
; sib
; sib
= sib
->sibling
)
889 eeh_add_device_tree_early(sib
);
890 eeh_add_device_early(dn
);
892 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early
);
895 * eeh_add_device_late - perform EEH initialization for the indicated pci device
896 * @dev: pci device for which to set up EEH
898 * This routine must be used to complete EEH initialization for PCI
899 * devices that were added after system boot (e.g. hotplug, dlpar).
901 void eeh_add_device_late(struct pci_dev
*dev
)
903 struct device_node
*dn
;
906 if (!dev
|| !eeh_subsystem_enabled
)
910 printk(KERN_DEBUG
"EEH: adding device %s\n", pci_name(dev
));
914 dn
= pci_device_to_OF_node(dev
);
918 pci_addr_cache_insert_device (dev
);
920 EXPORT_SYMBOL_GPL(eeh_add_device_late
);
923 * eeh_remove_device - undo EEH setup for the indicated pci device
924 * @dev: pci device to be removed
926 * This routine should be when a device is removed from a running
927 * system (e.g. by hotplug or dlpar).
929 void eeh_remove_device(struct pci_dev
*dev
)
931 struct device_node
*dn
;
932 if (!dev
|| !eeh_subsystem_enabled
)
935 /* Unregister the device with the EEH/PCI address search system */
937 printk(KERN_DEBUG
"EEH: remove device %s\n", pci_name(dev
));
939 pci_addr_cache_remove_device(dev
);
941 dn
= pci_device_to_OF_node(dev
);
942 PCI_DN(dn
)->pcidev
= NULL
;
945 EXPORT_SYMBOL_GPL(eeh_remove_device
);
947 void eeh_remove_bus_device(struct pci_dev
*dev
)
949 eeh_remove_device(dev
);
950 if (dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
951 struct pci_bus
*bus
= dev
->subordinate
;
952 struct list_head
*ln
;
955 for (ln
= bus
->devices
.next
; ln
!= &bus
->devices
; ln
= ln
->next
) {
956 struct pci_dev
*pdev
= pci_dev_b(ln
);
958 eeh_remove_bus_device(pdev
);
962 EXPORT_SYMBOL_GPL(eeh_remove_bus_device
);
964 static int proc_eeh_show(struct seq_file
*m
, void *v
)
966 if (0 == eeh_subsystem_enabled
) {
967 seq_printf(m
, "EEH Subsystem is globally disabled\n");
968 seq_printf(m
, "eeh_total_mmio_ffs=%ld\n", total_mmio_ffs
);
970 seq_printf(m
, "EEH Subsystem is enabled\n");
973 "no device node=%ld\n"
974 "no config address=%ld\n"
975 "check not wanted=%ld\n"
976 "eeh_total_mmio_ffs=%ld\n"
977 "eeh_false_positives=%ld\n"
978 "eeh_ignored_failures=%ld\n"
979 "eeh_slot_resets=%ld\n",
980 no_device
, no_dn
, no_cfg_addr
,
981 ignored_check
, total_mmio_ffs
,
982 false_positives
, ignored_failures
,
989 static int proc_eeh_open(struct inode
*inode
, struct file
*file
)
991 return single_open(file
, proc_eeh_show
, NULL
);
994 static struct file_operations proc_eeh_operations
= {
995 .open
= proc_eeh_open
,
998 .release
= single_release
,
1001 static int __init
eeh_init_proc(void)
1003 struct proc_dir_entry
*e
;
1005 if (platform_is_pseries()) {
1006 e
= create_proc_entry("ppc64/eeh", 0, NULL
);
1008 e
->proc_fops
= &proc_eeh_operations
;
1013 __initcall(eeh_init_proc
);