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
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 /* Mark the pci device driver too */
212 struct pci_dev
*dev
= PCI_DN(dn
)->pcidev
;
214 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
216 if (dev
&& dev
->driver
)
217 dev
->error_state
= pci_channel_io_frozen
;
220 __eeh_mark_slot (dn
->child
, mode_flag
);
226 void eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
229 dn
= find_device_pe (dn
);
231 /* Back up one, since config addrs might be shared */
232 if (PCI_DN(dn
) && PCI_DN(dn
)->eeh_pe_config_addr
)
235 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
237 /* Mark the pci device too */
238 dev
= PCI_DN(dn
)->pcidev
;
240 dev
->error_state
= pci_channel_io_frozen
;
242 __eeh_mark_slot (dn
->child
, mode_flag
);
245 static void __eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
249 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
250 PCI_DN(dn
)->eeh_check_count
= 0;
252 __eeh_clear_slot (dn
->child
, mode_flag
);
258 void eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
261 spin_lock_irqsave(&confirm_error_lock
, flags
);
263 dn
= find_device_pe (dn
);
265 /* Back up one, since config addrs might be shared */
266 if (PCI_DN(dn
) && PCI_DN(dn
)->eeh_pe_config_addr
)
269 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
270 PCI_DN(dn
)->eeh_check_count
= 0;
271 __eeh_clear_slot (dn
->child
, mode_flag
);
272 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
276 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
278 * @dev pci device, if known
280 * Check for an EEH failure for the given device node. Call this
281 * routine if the result of a read was all 0xff's and you want to
282 * find out if this is due to an EEH slot freeze. This routine
283 * will query firmware for the EEH status.
285 * Returns 0 if there has not been an EEH error; otherwise returns
286 * a non-zero value and queues up a slot isolation event notification.
288 * It is safe to call this routine in an interrupt context.
290 int eeh_dn_check_failure(struct device_node
*dn
, struct pci_dev
*dev
)
296 enum pci_channel_state state
;
301 if (!eeh_subsystem_enabled
)
310 /* Access to IO BARs might get this far and still not want checking. */
311 if (!(pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) ||
312 pdn
->eeh_mode
& EEH_MODE_NOCHECK
) {
315 printk ("EEH:ignored check (%x) for %s %s\n",
316 pdn
->eeh_mode
, pci_name (dev
), dn
->full_name
);
321 if (!pdn
->eeh_config_addr
&& !pdn
->eeh_pe_config_addr
) {
326 /* If we already have a pending isolation event for this
327 * slot, we know it's bad already, we don't need to check.
328 * Do this checking under a lock; as multiple PCI devices
329 * in one slot might report errors simultaneously, and we
330 * only want one error recovery routine running.
332 spin_lock_irqsave(&confirm_error_lock
, flags
);
334 if (pdn
->eeh_mode
& EEH_MODE_ISOLATED
) {
335 pdn
->eeh_check_count
++;
336 if (pdn
->eeh_check_count
>= EEH_MAX_FAILS
) {
337 printk (KERN_ERR
"EEH: Device driver ignored %d bad reads, panicing\n",
338 pdn
->eeh_check_count
);
342 /* re-read the slot reset state */
343 if (read_slot_reset_state(pdn
, rets
) != 0)
344 rets
[0] = -1; /* reset state unknown */
346 /* If we are here, then we hit an infinite loop. Stop. */
347 panic("EEH: MMIO halt (%d) on device:%s\n", rets
[0], pci_name(dev
));
353 * Now test for an EEH failure. This is VERY expensive.
354 * Note that the eeh_config_addr may be a parent device
355 * in the case of a device behind a bridge, or it may be
356 * function zero of a multi-function device.
357 * In any case they must share a common PHB.
359 ret
= read_slot_reset_state(pdn
, rets
);
361 /* If the call to firmware failed, punt */
363 printk(KERN_WARNING
"EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
370 /* Note that config-io to empty slots may fail;
371 * they are empty when they don't have children. */
372 if ((rets
[0] == 5) && (dn
->child
== NULL
)) {
378 /* If EEH is not supported on this device, punt. */
380 printk(KERN_WARNING
"EEH: event on unsupported device, rc=%d dn=%s\n",
387 /* If not the kind of error we know about, punt. */
388 if (rets
[0] != 2 && rets
[0] != 4 && rets
[0] != 5) {
396 /* Avoid repeated reports of this failure, including problems
397 * with other functions on this device, and functions under
399 eeh_mark_slot (dn
, EEH_MODE_ISOLATED
);
400 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
402 state
= pci_channel_io_normal
;
403 if ((rets
[0] == 2) || (rets
[0] == 4))
404 state
= pci_channel_io_frozen
;
406 state
= pci_channel_io_perm_failure
;
407 eeh_send_failure_event (dn
, dev
, state
, rets
[2]);
409 /* Most EEH events are due to device driver bugs. Having
410 * a stack trace will help the device-driver authors figure
411 * out what happened. So print that out. */
412 if (rets
[0] != 5) dump_stack();
416 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
420 EXPORT_SYMBOL_GPL(eeh_dn_check_failure
);
423 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
424 * @token i/o token, should be address in the form 0xA....
425 * @val value, should be all 1's (XXX why do we need this arg??)
427 * Check for an EEH failure at the given token address. Call this
428 * routine if the result of a read was all 0xff's and you want to
429 * find out if this is due to an EEH slot freeze event. This routine
430 * will query firmware for the EEH status.
432 * Note this routine is safe to call in an interrupt context.
434 unsigned long eeh_check_failure(const volatile void __iomem
*token
, unsigned long val
)
438 struct device_node
*dn
;
440 /* Finding the phys addr + pci device; this is pretty quick. */
441 addr
= eeh_token_to_phys((unsigned long __force
) token
);
442 dev
= pci_get_device_by_addr(addr
);
448 dn
= pci_device_to_OF_node(dev
);
449 eeh_dn_check_failure (dn
, dev
);
455 EXPORT_SYMBOL(eeh_check_failure
);
457 /* ------------------------------------------------------------- */
458 /* The code below deals with error recovery */
461 * eeh_slot_availability - returns error status of slot
462 * @pdn pci device node
464 * Return negative value if a permanent error, else return
465 * a number of milliseconds to wait until the PCI slot is
469 eeh_slot_availability(struct pci_dn
*pdn
)
474 rc
= read_slot_reset_state(pdn
, rets
);
478 if (rets
[1] == 0) return -1; /* EEH is not supported */
479 if (rets
[0] == 0) return 0; /* Oll Korrect */
481 if (rets
[2] == 0) return -1; /* permanently unavailable */
482 return rets
[2]; /* number of millisecs to wait */
487 printk (KERN_ERR
"EEH: Slot unavailable: rc=%d, rets=%d %d %d\n",
488 rc
, rets
[0], rets
[1], rets
[2]);
493 * rtas_pci_enable - enable MMIO or DMA transfers for this slot
494 * @pdn pci device node
498 rtas_pci_enable(struct pci_dn
*pdn
, int function
)
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_eeh_option
, 4, 1, NULL
,
510 BUID_HI(pdn
->phb
->buid
),
511 BUID_LO(pdn
->phb
->buid
),
515 printk(KERN_WARNING
"EEH: Cannot enable function %d, err=%d dn=%s\n",
516 function
, rc
, pdn
->node
->full_name
);
522 * rtas_pci_slot_reset - raises/lowers the pci #RST line
523 * @pdn pci device node
524 * @state: 1/0 to raise/lower the #RST
526 * Clear the EEH-frozen condition on a slot. This routine
527 * asserts the PCI #RST line if the 'state' argument is '1',
528 * and drops the #RST line if 'state is '0'. This routine is
529 * safe to call in an interrupt context.
534 rtas_pci_slot_reset(struct pci_dn
*pdn
, int state
)
542 printk (KERN_WARNING
"EEH: in slot reset, device node %s has no phb\n",
543 pdn
->node
->full_name
);
547 /* Use PE configuration address, if present */
548 config_addr
= pdn
->eeh_config_addr
;
549 if (pdn
->eeh_pe_config_addr
)
550 config_addr
= pdn
->eeh_pe_config_addr
;
552 rc
= rtas_call(ibm_set_slot_reset
,4,1, NULL
,
554 BUID_HI(pdn
->phb
->buid
),
555 BUID_LO(pdn
->phb
->buid
),
558 printk (KERN_WARNING
"EEH: Unable to reset the failed slot,"
559 " (%d) #RST=%d dn=%s\n",
560 rc
, state
, pdn
->node
->full_name
);
564 * rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
565 * @pdn: pci device node to be reset.
567 * Return 0 if success, else a non-zero value.
570 static void __rtas_set_slot_reset(struct pci_dn
*pdn
)
572 rtas_pci_slot_reset (pdn
, 1);
574 /* The PCI bus requires that the reset be held high for at least
575 * a 100 milliseconds. We wait a bit longer 'just in case'. */
577 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
578 msleep (PCI_BUS_RST_HOLD_TIME_MSEC
);
580 /* We might get hit with another EEH freeze as soon as the
581 * pci slot reset line is dropped. Make sure we don't miss
582 * these, and clear the flag now. */
583 eeh_clear_slot (pdn
->node
, EEH_MODE_ISOLATED
);
585 rtas_pci_slot_reset (pdn
, 0);
587 /* After a PCI slot has been reset, the PCI Express spec requires
588 * a 1.5 second idle time for the bus to stabilize, before starting
590 #define PCI_BUS_SETTLE_TIME_MSEC 1800
591 msleep (PCI_BUS_SETTLE_TIME_MSEC
);
594 int rtas_set_slot_reset(struct pci_dn
*pdn
)
598 __rtas_set_slot_reset(pdn
);
600 /* Now double check with the firmware to make sure the device is
601 * ready to be used; if not, wait for recovery. */
602 for (i
=0; i
<10; i
++) {
603 rc
= eeh_slot_availability (pdn
);
608 printk (KERN_ERR
"EEH: failed (%d) to reset slot %s\n",
609 i
, pdn
->node
->full_name
);
610 __rtas_set_slot_reset(pdn
);
615 printk (KERN_ERR
"EEH: unrecoverable slot failure %s\n",
616 pdn
->node
->full_name
);
623 rc
= eeh_slot_availability (pdn
);
625 printk (KERN_ERR
"EEH: timeout resetting slot %s\n", pdn
->node
->full_name
);
630 /* ------------------------------------------------------- */
631 /** Save and restore of PCI BARs
633 * Although firmware will set up BARs during boot, it doesn't
634 * set up device BAR's after a device reset, although it will,
635 * if requested, set up bridge configuration. Thus, we need to
636 * configure the PCI devices ourselves.
640 * __restore_bars - Restore the Base Address Registers
641 * @pdn: pci device node
643 * Loads the PCI configuration space base address registers,
644 * the expansion ROM base address, the latency timer, and etc.
645 * from the saved values in the device node.
647 static inline void __restore_bars (struct pci_dn
*pdn
)
651 if (NULL
==pdn
->phb
) return;
652 for (i
=4; i
<10; i
++) {
653 rtas_write_config(pdn
, i
*4, 4, pdn
->config_space
[i
]);
656 /* 12 == Expansion ROM Address */
657 rtas_write_config(pdn
, 12*4, 4, pdn
->config_space
[12]);
659 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
660 #define SAVED_BYTE(OFF) (((u8 *)(pdn->config_space))[BYTE_SWAP(OFF)])
662 rtas_write_config (pdn
, PCI_CACHE_LINE_SIZE
, 1,
663 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
665 rtas_write_config (pdn
, PCI_LATENCY_TIMER
, 1,
666 SAVED_BYTE(PCI_LATENCY_TIMER
));
668 /* max latency, min grant, interrupt pin and line */
669 rtas_write_config(pdn
, 15*4, 4, pdn
->config_space
[15]);
673 * eeh_restore_bars - restore the PCI config space info
675 * This routine performs a recursive walk to the children
676 * of this device as well.
678 void eeh_restore_bars(struct pci_dn
*pdn
)
680 struct device_node
*dn
;
684 if ((pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) && !IS_BRIDGE(pdn
->class_code
))
685 __restore_bars (pdn
);
687 dn
= pdn
->node
->child
;
689 eeh_restore_bars (PCI_DN(dn
));
695 * eeh_save_bars - save device bars
697 * Save the values of the device bars. Unlike the restore
698 * routine, this routine is *not* recursive. This is because
699 * PCI devices are added individuallly; but, for the restore,
700 * an entire slot is reset at a time.
702 static void eeh_save_bars(struct pci_dn
*pdn
)
709 for (i
= 0; i
< 16; i
++)
710 rtas_read_config(pdn
, i
* 4, 4, &pdn
->config_space
[i
]);
714 rtas_configure_bridge(struct pci_dn
*pdn
)
719 /* Use PE configuration address, if present */
720 config_addr
= pdn
->eeh_config_addr
;
721 if (pdn
->eeh_pe_config_addr
)
722 config_addr
= pdn
->eeh_pe_config_addr
;
724 rc
= rtas_call(ibm_configure_bridge
,3,1, NULL
,
726 BUID_HI(pdn
->phb
->buid
),
727 BUID_LO(pdn
->phb
->buid
));
729 printk (KERN_WARNING
"EEH: Unable to configure device bridge (%d) for %s\n",
730 rc
, pdn
->node
->full_name
);
734 /* ------------------------------------------------------------- */
735 /* The code below deals with enabling EEH for devices during the
736 * early boot sequence. EEH must be enabled before any PCI probing
742 struct eeh_early_enable_info
{
743 unsigned int buid_hi
;
744 unsigned int buid_lo
;
747 /* Enable eeh for the given device node. */
748 static void *early_enable_eeh(struct device_node
*dn
, void *data
)
750 unsigned int rets
[3];
751 struct eeh_early_enable_info
*info
= data
;
753 const char *status
= get_property(dn
, "status", NULL
);
754 const u32
*class_code
= get_property(dn
, "class-code", NULL
);
755 const u32
*vendor_id
= get_property(dn
, "vendor-id", NULL
);
756 const u32
*device_id
= get_property(dn
, "device-id", NULL
);
759 struct pci_dn
*pdn
= PCI_DN(dn
);
763 pdn
->eeh_check_count
= 0;
764 pdn
->eeh_freeze_count
= 0;
766 if (status
&& strcmp(status
, "ok") != 0)
767 return NULL
; /* ignore devices with bad status */
769 /* Ignore bad nodes. */
770 if (!class_code
|| !vendor_id
|| !device_id
)
773 /* There is nothing to check on PCI to ISA bridges */
774 if (dn
->type
&& !strcmp(dn
->type
, "isa")) {
775 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
778 pdn
->class_code
= *class_code
;
781 * Now decide if we are going to "Disable" EEH checking
782 * for this device. We still run with the EEH hardware active,
783 * but we won't be checking for ff's. This means a driver
784 * could return bad data (very bad!), an interrupt handler could
785 * hang waiting on status bits that won't change, etc.
786 * But there are a few cases like display devices that make sense.
788 enable
= 1; /* i.e. we will do checking */
790 if ((*class_code
>> 16) == PCI_BASE_CLASS_DISPLAY
)
795 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
797 /* Ok... see if this device supports EEH. Some do, some don't,
798 * and the only way to find out is to check each and every one. */
799 regs
= get_property(dn
, "reg", NULL
);
801 /* First register entry is addr (00BBSS00) */
802 /* Try to enable eeh */
803 ret
= rtas_call(ibm_set_eeh_option
, 4, 1, NULL
,
804 regs
[0], info
->buid_hi
, info
->buid_lo
,
809 pdn
->eeh_config_addr
= regs
[0];
811 /* If the newer, better, ibm,get-config-addr-info is supported,
812 * then use that instead. */
813 pdn
->eeh_pe_config_addr
= 0;
814 if (ibm_get_config_addr_info
!= RTAS_UNKNOWN_SERVICE
) {
815 ret
= rtas_call (ibm_get_config_addr_info
, 4, 2, rets
,
816 pdn
->eeh_config_addr
,
817 info
->buid_hi
, info
->buid_lo
,
820 pdn
->eeh_pe_config_addr
= rets
[0];
823 /* Some older systems (Power4) allow the
824 * ibm,set-eeh-option call to succeed even on nodes
825 * where EEH is not supported. Verify support
827 ret
= read_slot_reset_state(pdn
, rets
);
828 if ((ret
== 0) && (rets
[1] == 1))
833 eeh_subsystem_enabled
= 1;
834 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
837 printk(KERN_DEBUG
"EEH: %s: eeh enabled, config=%x pe_config=%x\n",
838 dn
->full_name
, pdn
->eeh_config_addr
, pdn
->eeh_pe_config_addr
);
842 /* This device doesn't support EEH, but it may have an
843 * EEH parent, in which case we mark it as supported. */
844 if (dn
->parent
&& PCI_DN(dn
->parent
)
845 && (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
846 /* Parent supports EEH. */
847 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
848 pdn
->eeh_config_addr
= PCI_DN(dn
->parent
)->eeh_config_addr
;
853 printk(KERN_WARNING
"EEH: %s: unable to get reg property.\n",
862 * Initialize EEH by trying to enable it for all of the adapters in the system.
863 * As a side effect we can determine here if eeh is supported at all.
864 * Note that we leave EEH on so failed config cycles won't cause a machine
865 * check. If a user turns off EEH for a particular adapter they are really
866 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
867 * grant access to a slot if EEH isn't enabled, and so we always enable
868 * EEH for all slots/all devices.
870 * The eeh-force-off option disables EEH checking globally, for all slots.
871 * Even if force-off is set, the EEH hardware is still enabled, so that
872 * newer systems can boot.
874 void __init
eeh_init(void)
876 struct device_node
*phb
, *np
;
877 struct eeh_early_enable_info info
;
879 spin_lock_init(&confirm_error_lock
);
880 spin_lock_init(&slot_errbuf_lock
);
882 np
= of_find_node_by_path("/rtas");
886 ibm_set_eeh_option
= rtas_token("ibm,set-eeh-option");
887 ibm_set_slot_reset
= rtas_token("ibm,set-slot-reset");
888 ibm_read_slot_reset_state2
= rtas_token("ibm,read-slot-reset-state2");
889 ibm_read_slot_reset_state
= rtas_token("ibm,read-slot-reset-state");
890 ibm_slot_error_detail
= rtas_token("ibm,slot-error-detail");
891 ibm_get_config_addr_info
= rtas_token("ibm,get-config-addr-info");
892 ibm_configure_bridge
= rtas_token ("ibm,configure-bridge");
894 if (ibm_set_eeh_option
== RTAS_UNKNOWN_SERVICE
)
897 eeh_error_buf_size
= rtas_token("rtas-error-log-max");
898 if (eeh_error_buf_size
== RTAS_UNKNOWN_SERVICE
) {
899 eeh_error_buf_size
= 1024;
901 if (eeh_error_buf_size
> RTAS_ERROR_LOG_MAX
) {
902 printk(KERN_WARNING
"EEH: rtas-error-log-max is bigger than allocated "
903 "buffer ! (%d vs %d)", eeh_error_buf_size
, RTAS_ERROR_LOG_MAX
);
904 eeh_error_buf_size
= RTAS_ERROR_LOG_MAX
;
907 /* Enable EEH for all adapters. Note that eeh requires buid's */
908 for (phb
= of_find_node_by_name(NULL
, "pci"); phb
;
909 phb
= of_find_node_by_name(phb
, "pci")) {
912 buid
= get_phb_buid(phb
);
913 if (buid
== 0 || PCI_DN(phb
) == NULL
)
916 info
.buid_lo
= BUID_LO(buid
);
917 info
.buid_hi
= BUID_HI(buid
);
918 traverse_pci_devices(phb
, early_enable_eeh
, &info
);
921 if (eeh_subsystem_enabled
)
922 printk(KERN_INFO
"EEH: PCI Enhanced I/O Error Handling Enabled\n");
924 printk(KERN_WARNING
"EEH: No capable adapters found\n");
928 * eeh_add_device_early - enable EEH for the indicated device_node
929 * @dn: device node for which to set up EEH
931 * This routine must be used to perform EEH initialization for PCI
932 * devices that were added after system boot (e.g. hotplug, dlpar).
933 * This routine must be called before any i/o is performed to the
934 * adapter (inluding any config-space i/o).
935 * Whether this actually enables EEH or not for this device depends
936 * on the CEC architecture, type of the device, on earlier boot
937 * command-line arguments & etc.
939 static void eeh_add_device_early(struct device_node
*dn
)
941 struct pci_controller
*phb
;
942 struct eeh_early_enable_info info
;
944 if (!dn
|| !PCI_DN(dn
))
946 phb
= PCI_DN(dn
)->phb
;
948 /* USB Bus children of PCI devices will not have BUID's */
949 if (NULL
== phb
|| 0 == phb
->buid
)
952 info
.buid_hi
= BUID_HI(phb
->buid
);
953 info
.buid_lo
= BUID_LO(phb
->buid
);
954 early_enable_eeh(dn
, &info
);
957 void eeh_add_device_tree_early(struct device_node
*dn
)
959 struct device_node
*sib
;
960 for (sib
= dn
->child
; sib
; sib
= sib
->sibling
)
961 eeh_add_device_tree_early(sib
);
962 eeh_add_device_early(dn
);
964 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early
);
967 * eeh_add_device_late - perform EEH initialization for the indicated pci device
968 * @dev: pci device for which to set up EEH
970 * This routine must be used to complete EEH initialization for PCI
971 * devices that were added after system boot (e.g. hotplug, dlpar).
973 static void eeh_add_device_late(struct pci_dev
*dev
)
975 struct device_node
*dn
;
978 if (!dev
|| !eeh_subsystem_enabled
)
982 printk(KERN_DEBUG
"EEH: adding device %s\n", pci_name(dev
));
986 dn
= pci_device_to_OF_node(dev
);
990 pci_addr_cache_insert_device (dev
);
993 void eeh_add_device_tree_late(struct pci_bus
*bus
)
997 list_for_each_entry(dev
, &bus
->devices
, bus_list
) {
998 eeh_add_device_late(dev
);
999 if (dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
1000 struct pci_bus
*subbus
= dev
->subordinate
;
1002 eeh_add_device_tree_late(subbus
);
1006 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late
);
1009 * eeh_remove_device - undo EEH setup for the indicated pci device
1010 * @dev: pci device to be removed
1012 * This routine should be called when a device is removed from
1013 * a running system (e.g. by hotplug or dlpar). It unregisters
1014 * the PCI device from the EEH subsystem. I/O errors affecting
1015 * this device will no longer be detected after this call; thus,
1016 * i/o errors affecting this slot may leave this device unusable.
1018 static void eeh_remove_device(struct pci_dev
*dev
)
1020 struct device_node
*dn
;
1021 if (!dev
|| !eeh_subsystem_enabled
)
1024 /* Unregister the device with the EEH/PCI address search system */
1026 printk(KERN_DEBUG
"EEH: remove device %s\n", pci_name(dev
));
1028 pci_addr_cache_remove_device(dev
);
1030 dn
= pci_device_to_OF_node(dev
);
1031 if (PCI_DN(dn
)->pcidev
) {
1032 PCI_DN(dn
)->pcidev
= NULL
;
1037 void eeh_remove_bus_device(struct pci_dev
*dev
)
1039 struct pci_bus
*bus
= dev
->subordinate
;
1040 struct pci_dev
*child
, *tmp
;
1042 eeh_remove_device(dev
);
1044 if (bus
&& dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
1045 list_for_each_entry_safe(child
, tmp
, &bus
->devices
, bus_list
)
1046 eeh_remove_bus_device(child
);
1049 EXPORT_SYMBOL_GPL(eeh_remove_bus_device
);
1051 static int proc_eeh_show(struct seq_file
*m
, void *v
)
1053 if (0 == eeh_subsystem_enabled
) {
1054 seq_printf(m
, "EEH Subsystem is globally disabled\n");
1055 seq_printf(m
, "eeh_total_mmio_ffs=%ld\n", total_mmio_ffs
);
1057 seq_printf(m
, "EEH Subsystem is enabled\n");
1060 "no device node=%ld\n"
1061 "no config address=%ld\n"
1062 "check not wanted=%ld\n"
1063 "eeh_total_mmio_ffs=%ld\n"
1064 "eeh_false_positives=%ld\n"
1065 "eeh_ignored_failures=%ld\n"
1066 "eeh_slot_resets=%ld\n",
1067 no_device
, no_dn
, no_cfg_addr
,
1068 ignored_check
, total_mmio_ffs
,
1069 false_positives
, ignored_failures
,
1076 static int proc_eeh_open(struct inode
*inode
, struct file
*file
)
1078 return single_open(file
, proc_eeh_show
, NULL
);
1081 static const struct file_operations proc_eeh_operations
= {
1082 .open
= proc_eeh_open
,
1084 .llseek
= seq_lseek
,
1085 .release
= single_release
,
1088 static int __init
eeh_init_proc(void)
1090 struct proc_dir_entry
*e
;
1092 if (machine_is(pseries
)) {
1093 e
= create_proc_entry("ppc64/eeh", 0, NULL
);
1095 e
->proc_fops
= &proc_eeh_operations
;
1100 __initcall(eeh_init_proc
);