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 DEFINE_PER_CPU(unsigned long, no_device
);
101 static DEFINE_PER_CPU(unsigned long, no_dn
);
102 static DEFINE_PER_CPU(unsigned long, no_cfg_addr
);
103 static DEFINE_PER_CPU(unsigned long, ignored_check
);
104 static DEFINE_PER_CPU(unsigned long, total_mmio_ffs
);
105 static DEFINE_PER_CPU(unsigned long, false_positives
);
106 static DEFINE_PER_CPU(unsigned long, ignored_failures
);
107 static DEFINE_PER_CPU(unsigned long, slot_resets
);
109 /* --------------------------------------------------------------- */
110 /* Below lies the EEH event infrastructure */
112 void eeh_slot_error_detail (struct pci_dn
*pdn
, int severity
)
118 /* Log the error with the rtas logger */
119 spin_lock_irqsave(&slot_errbuf_lock
, flags
);
120 memset(slot_errbuf
, 0, eeh_error_buf_size
);
122 /* Use PE configuration address, if present */
123 config_addr
= pdn
->eeh_config_addr
;
124 if (pdn
->eeh_pe_config_addr
)
125 config_addr
= pdn
->eeh_pe_config_addr
;
127 rc
= rtas_call(ibm_slot_error_detail
,
128 8, 1, NULL
, config_addr
,
129 BUID_HI(pdn
->phb
->buid
),
130 BUID_LO(pdn
->phb
->buid
), NULL
, 0,
131 virt_to_phys(slot_errbuf
),
136 log_error(slot_errbuf
, ERR_TYPE_RTAS_LOG
, 0);
137 spin_unlock_irqrestore(&slot_errbuf_lock
, flags
);
141 * read_slot_reset_state - Read the reset state of a device node's slot
142 * @dn: device node to read
143 * @rets: array to return results in
145 static int read_slot_reset_state(struct pci_dn
*pdn
, int rets
[])
150 if (ibm_read_slot_reset_state2
!= RTAS_UNKNOWN_SERVICE
) {
151 token
= ibm_read_slot_reset_state2
;
154 token
= ibm_read_slot_reset_state
;
155 rets
[2] = 0; /* fake PE Unavailable info */
159 /* Use PE configuration address, if present */
160 config_addr
= pdn
->eeh_config_addr
;
161 if (pdn
->eeh_pe_config_addr
)
162 config_addr
= pdn
->eeh_pe_config_addr
;
164 return rtas_call(token
, 3, outputs
, rets
, config_addr
,
165 BUID_HI(pdn
->phb
->buid
), BUID_LO(pdn
->phb
->buid
));
169 * eeh_token_to_phys - convert EEH address token to phys address
170 * @token i/o token, should be address in the form 0xA....
172 static inline unsigned long eeh_token_to_phys(unsigned long token
)
177 ptep
= find_linux_pte(init_mm
.pgd
, token
);
180 pa
= pte_pfn(*ptep
) << PAGE_SHIFT
;
182 return pa
| (token
& (PAGE_SIZE
-1));
186 * Return the "partitionable endpoint" (pe) under which this device lies
188 struct device_node
* find_device_pe(struct device_node
*dn
)
190 while ((dn
->parent
) && PCI_DN(dn
->parent
) &&
191 (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
197 /** Mark all devices that are peers of this device as failed.
198 * Mark the device driver too, so that it can see the failure
199 * immediately; this is critical, since some drivers poll
200 * status registers in interrupts ... If a driver is polling,
201 * and the slot is frozen, then the driver can deadlock in
202 * an interrupt context, which is bad.
205 static void __eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
209 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
211 /* Mark the pci device driver too */
212 struct pci_dev
*dev
= PCI_DN(dn
)->pcidev
;
213 if (dev
&& dev
->driver
)
214 dev
->error_state
= pci_channel_io_frozen
;
217 __eeh_mark_slot (dn
->child
, mode_flag
);
223 void eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
225 dn
= find_device_pe (dn
);
226 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
227 __eeh_mark_slot (dn
->child
, mode_flag
);
230 static void __eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
234 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
235 PCI_DN(dn
)->eeh_check_count
= 0;
237 __eeh_clear_slot (dn
->child
, mode_flag
);
243 void eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
246 spin_lock_irqsave(&confirm_error_lock
, flags
);
247 dn
= find_device_pe (dn
);
248 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
249 PCI_DN(dn
)->eeh_check_count
= 0;
250 __eeh_clear_slot (dn
->child
, mode_flag
);
251 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
255 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
257 * @dev pci device, if known
259 * Check for an EEH failure for the given device node. Call this
260 * routine if the result of a read was all 0xff's and you want to
261 * find out if this is due to an EEH slot freeze. This routine
262 * will query firmware for the EEH status.
264 * Returns 0 if there has not been an EEH error; otherwise returns
265 * a non-zero value and queues up a slot isolation event notification.
267 * It is safe to call this routine in an interrupt context.
269 int eeh_dn_check_failure(struct device_node
*dn
, struct pci_dev
*dev
)
275 enum pci_channel_state state
;
278 __get_cpu_var(total_mmio_ffs
)++;
280 if (!eeh_subsystem_enabled
)
284 __get_cpu_var(no_dn
)++;
289 /* Access to IO BARs might get this far and still not want checking. */
290 if (!(pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) ||
291 pdn
->eeh_mode
& EEH_MODE_NOCHECK
) {
292 __get_cpu_var(ignored_check
)++;
294 printk ("EEH:ignored check (%x) for %s %s\n",
295 pdn
->eeh_mode
, pci_name (dev
), dn
->full_name
);
300 if (!pdn
->eeh_config_addr
&& !pdn
->eeh_pe_config_addr
) {
301 __get_cpu_var(no_cfg_addr
)++;
305 /* If we already have a pending isolation event for this
306 * slot, we know it's bad already, we don't need to check.
307 * Do this checking under a lock; as multiple PCI devices
308 * in one slot might report errors simultaneously, and we
309 * only want one error recovery routine running.
311 spin_lock_irqsave(&confirm_error_lock
, flags
);
313 if (pdn
->eeh_mode
& EEH_MODE_ISOLATED
) {
314 pdn
->eeh_check_count
++;
315 if (pdn
->eeh_check_count
>= EEH_MAX_FAILS
) {
316 printk (KERN_ERR
"EEH: Device driver ignored %d bad reads, panicing\n",
317 pdn
->eeh_check_count
);
320 /* re-read the slot reset state */
321 if (read_slot_reset_state(pdn
, rets
) != 0)
322 rets
[0] = -1; /* reset state unknown */
324 /* If we are here, then we hit an infinite loop. Stop. */
325 panic("EEH: MMIO halt (%d) on device:%s\n", rets
[0], pci_name(dev
));
331 * Now test for an EEH failure. This is VERY expensive.
332 * Note that the eeh_config_addr may be a parent device
333 * in the case of a device behind a bridge, or it may be
334 * function zero of a multi-function device.
335 * In any case they must share a common PHB.
337 ret
= read_slot_reset_state(pdn
, rets
);
339 /* If the call to firmware failed, punt */
341 printk(KERN_WARNING
"EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
343 __get_cpu_var(false_positives
)++;
348 /* If EEH is not supported on this device, punt. */
350 printk(KERN_WARNING
"EEH: event on unsupported device, rc=%d dn=%s\n",
352 __get_cpu_var(false_positives
)++;
357 /* If not the kind of error we know about, punt. */
358 if (rets
[0] != 2 && rets
[0] != 4 && rets
[0] != 5) {
359 __get_cpu_var(false_positives
)++;
364 /* Note that config-io to empty slots may fail;
365 * we recognize empty because they don't have children. */
366 if ((rets
[0] == 5) && (dn
->child
== NULL
)) {
367 __get_cpu_var(false_positives
)++;
372 __get_cpu_var(slot_resets
)++;
374 /* Avoid repeated reports of this failure, including problems
375 * with other functions on this device, and functions under
377 eeh_mark_slot (dn
, EEH_MODE_ISOLATED
);
378 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
380 state
= pci_channel_io_normal
;
381 if ((rets
[0] == 2) || (rets
[0] == 4))
382 state
= pci_channel_io_frozen
;
384 state
= pci_channel_io_perm_failure
;
385 eeh_send_failure_event (dn
, dev
, state
, rets
[2]);
387 /* Most EEH events are due to device driver bugs. Having
388 * a stack trace will help the device-driver authors figure
389 * out what happened. So print that out. */
390 if (rets
[0] != 5) dump_stack();
394 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
398 EXPORT_SYMBOL_GPL(eeh_dn_check_failure
);
401 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
402 * @token i/o token, should be address in the form 0xA....
403 * @val value, should be all 1's (XXX why do we need this arg??)
405 * Check for an EEH failure at the given token address. Call this
406 * routine if the result of a read was all 0xff's and you want to
407 * find out if this is due to an EEH slot freeze event. This routine
408 * will query firmware for the EEH status.
410 * Note this routine is safe to call in an interrupt context.
412 unsigned long eeh_check_failure(const volatile void __iomem
*token
, unsigned long val
)
416 struct device_node
*dn
;
418 /* Finding the phys addr + pci device; this is pretty quick. */
419 addr
= eeh_token_to_phys((unsigned long __force
) token
);
420 dev
= pci_get_device_by_addr(addr
);
422 __get_cpu_var(no_device
)++;
426 dn
= pci_device_to_OF_node(dev
);
427 eeh_dn_check_failure (dn
, dev
);
433 EXPORT_SYMBOL(eeh_check_failure
);
435 /* ------------------------------------------------------------- */
436 /* The code below deals with error recovery */
438 /** Return negative value if a permanent error, else return
439 * a number of milliseconds to wait until the PCI slot is
443 eeh_slot_availability(struct pci_dn
*pdn
)
448 rc
= read_slot_reset_state(pdn
, rets
);
452 if (rets
[1] == 0) return -1; /* EEH is not supported */
453 if (rets
[0] == 0) return 0; /* Oll Korrect */
455 if (rets
[2] == 0) return -1; /* permanently unavailable */
456 return rets
[2]; /* number of millisecs to wait */
461 printk (KERN_ERR
"EEH: Slot unavailable: rc=%d, rets=%d %d %d\n",
462 rc
, rets
[0], rets
[1], rets
[2]);
466 /** rtas_pci_slot_reset raises/lowers the pci #RST line
467 * state: 1/0 to raise/lower the #RST
469 * Clear the EEH-frozen condition on a slot. This routine
470 * asserts the PCI #RST line if the 'state' argument is '1',
471 * and drops the #RST line if 'state is '0'. This routine is
472 * safe to call in an interrupt context.
477 rtas_pci_slot_reset(struct pci_dn
*pdn
, int state
)
485 printk (KERN_WARNING
"EEH: in slot reset, device node %s has no phb\n",
486 pdn
->node
->full_name
);
490 /* Use PE configuration address, if present */
491 config_addr
= pdn
->eeh_config_addr
;
492 if (pdn
->eeh_pe_config_addr
)
493 config_addr
= pdn
->eeh_pe_config_addr
;
495 rc
= rtas_call(ibm_set_slot_reset
,4,1, NULL
,
497 BUID_HI(pdn
->phb
->buid
),
498 BUID_LO(pdn
->phb
->buid
),
501 printk (KERN_WARNING
"EEH: Unable to reset the failed slot, (%d) #RST=%d dn=%s\n",
502 rc
, state
, pdn
->node
->full_name
);
507 /** rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
508 * dn -- device node to be reset.
510 * Return 0 if success, else a non-zero value.
514 rtas_set_slot_reset(struct pci_dn
*pdn
)
518 rtas_pci_slot_reset (pdn
, 1);
520 /* The PCI bus requires that the reset be held high for at least
521 * a 100 milliseconds. We wait a bit longer 'just in case'. */
523 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
524 msleep (PCI_BUS_RST_HOLD_TIME_MSEC
);
526 /* We might get hit with another EEH freeze as soon as the
527 * pci slot reset line is dropped. Make sure we don't miss
528 * these, and clear the flag now. */
529 eeh_clear_slot (pdn
->node
, EEH_MODE_ISOLATED
);
531 rtas_pci_slot_reset (pdn
, 0);
533 /* After a PCI slot has been reset, the PCI Express spec requires
534 * a 1.5 second idle time for the bus to stabilize, before starting
536 #define PCI_BUS_SETTLE_TIME_MSEC 1800
537 msleep (PCI_BUS_SETTLE_TIME_MSEC
);
539 /* Now double check with the firmware to make sure the device is
540 * ready to be used; if not, wait for recovery. */
541 for (i
=0; i
<10; i
++) {
542 rc
= eeh_slot_availability (pdn
);
544 printk (KERN_ERR
"EEH: failed (%d) to reset slot %s\n", rc
, pdn
->node
->full_name
);
553 rc
= eeh_slot_availability (pdn
);
555 printk (KERN_ERR
"EEH: timeout resetting slot %s\n", pdn
->node
->full_name
);
560 /* ------------------------------------------------------- */
561 /** Save and restore of PCI BARs
563 * Although firmware will set up BARs during boot, it doesn't
564 * set up device BAR's after a device reset, although it will,
565 * if requested, set up bridge configuration. Thus, we need to
566 * configure the PCI devices ourselves.
570 * __restore_bars - Restore the Base Address Registers
571 * Loads the PCI configuration space base address registers,
572 * the expansion ROM base address, the latency timer, and etc.
573 * from the saved values in the device node.
575 static inline void __restore_bars (struct pci_dn
*pdn
)
579 if (NULL
==pdn
->phb
) return;
580 for (i
=4; i
<10; i
++) {
581 rtas_write_config(pdn
, i
*4, 4, pdn
->config_space
[i
]);
584 /* 12 == Expansion ROM Address */
585 rtas_write_config(pdn
, 12*4, 4, pdn
->config_space
[12]);
587 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
588 #define SAVED_BYTE(OFF) (((u8 *)(pdn->config_space))[BYTE_SWAP(OFF)])
590 rtas_write_config (pdn
, PCI_CACHE_LINE_SIZE
, 1,
591 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
593 rtas_write_config (pdn
, PCI_LATENCY_TIMER
, 1,
594 SAVED_BYTE(PCI_LATENCY_TIMER
));
596 /* max latency, min grant, interrupt pin and line */
597 rtas_write_config(pdn
, 15*4, 4, pdn
->config_space
[15]);
601 * eeh_restore_bars - restore the PCI config space info
603 * This routine performs a recursive walk to the children
604 * of this device as well.
606 void eeh_restore_bars(struct pci_dn
*pdn
)
608 struct device_node
*dn
;
612 if (! pdn
->eeh_is_bridge
)
613 __restore_bars (pdn
);
615 dn
= pdn
->node
->child
;
617 eeh_restore_bars (PCI_DN(dn
));
623 * eeh_save_bars - save device bars
625 * Save the values of the device bars. Unlike the restore
626 * routine, this routine is *not* recursive. This is because
627 * PCI devices are added individuallly; but, for the restore,
628 * an entire slot is reset at a time.
630 void eeh_save_bars(struct pci_dev
* pdev
, struct pci_dn
*pdn
)
637 for (i
= 0; i
< 16; i
++)
638 pci_read_config_dword(pdev
, i
* 4, &pdn
->config_space
[i
]);
640 if (pdev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
)
641 pdn
->eeh_is_bridge
= 1;
645 rtas_configure_bridge(struct pci_dn
*pdn
)
650 /* Use PE configuration address, if present */
651 config_addr
= pdn
->eeh_config_addr
;
652 if (pdn
->eeh_pe_config_addr
)
653 config_addr
= pdn
->eeh_pe_config_addr
;
655 rc
= rtas_call(ibm_configure_bridge
,3,1, NULL
,
657 BUID_HI(pdn
->phb
->buid
),
658 BUID_LO(pdn
->phb
->buid
));
660 printk (KERN_WARNING
"EEH: Unable to configure device bridge (%d) for %s\n",
661 rc
, pdn
->node
->full_name
);
665 /* ------------------------------------------------------------- */
666 /* The code below deals with enabling EEH for devices during the
667 * early boot sequence. EEH must be enabled before any PCI probing
673 struct eeh_early_enable_info
{
674 unsigned int buid_hi
;
675 unsigned int buid_lo
;
678 /* Enable eeh for the given device node. */
679 static void *early_enable_eeh(struct device_node
*dn
, void *data
)
681 struct eeh_early_enable_info
*info
= data
;
683 char *status
= get_property(dn
, "status", NULL
);
684 u32
*class_code
= (u32
*)get_property(dn
, "class-code", NULL
);
685 u32
*vendor_id
= (u32
*)get_property(dn
, "vendor-id", NULL
);
686 u32
*device_id
= (u32
*)get_property(dn
, "device-id", NULL
);
689 struct pci_dn
*pdn
= PCI_DN(dn
);
692 pdn
->eeh_check_count
= 0;
693 pdn
->eeh_freeze_count
= 0;
695 if (status
&& strcmp(status
, "ok") != 0)
696 return NULL
; /* ignore devices with bad status */
698 /* Ignore bad nodes. */
699 if (!class_code
|| !vendor_id
|| !device_id
)
702 /* There is nothing to check on PCI to ISA bridges */
703 if (dn
->type
&& !strcmp(dn
->type
, "isa")) {
704 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
709 * Now decide if we are going to "Disable" EEH checking
710 * for this device. We still run with the EEH hardware active,
711 * but we won't be checking for ff's. This means a driver
712 * could return bad data (very bad!), an interrupt handler could
713 * hang waiting on status bits that won't change, etc.
714 * But there are a few cases like display devices that make sense.
716 enable
= 1; /* i.e. we will do checking */
718 if ((*class_code
>> 16) == PCI_BASE_CLASS_DISPLAY
)
723 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
725 /* Ok... see if this device supports EEH. Some do, some don't,
726 * and the only way to find out is to check each and every one. */
727 regs
= (u32
*)get_property(dn
, "reg", NULL
);
729 /* First register entry is addr (00BBSS00) */
730 /* Try to enable eeh */
731 ret
= rtas_call(ibm_set_eeh_option
, 4, 1, NULL
,
732 regs
[0], info
->buid_hi
, info
->buid_lo
,
736 eeh_subsystem_enabled
= 1;
737 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
738 pdn
->eeh_config_addr
= regs
[0];
740 /* If the newer, better, ibm,get-config-addr-info is supported,
741 * then use that instead. */
742 pdn
->eeh_pe_config_addr
= 0;
743 if (ibm_get_config_addr_info
!= RTAS_UNKNOWN_SERVICE
) {
744 unsigned int rets
[2];
745 ret
= rtas_call (ibm_get_config_addr_info
, 4, 2, rets
,
746 pdn
->eeh_config_addr
,
747 info
->buid_hi
, info
->buid_lo
,
750 pdn
->eeh_pe_config_addr
= rets
[0];
753 printk(KERN_DEBUG
"EEH: %s: eeh enabled, config=%x pe_config=%x\n",
754 dn
->full_name
, pdn
->eeh_config_addr
, pdn
->eeh_pe_config_addr
);
758 /* This device doesn't support EEH, but it may have an
759 * EEH parent, in which case we mark it as supported. */
760 if (dn
->parent
&& PCI_DN(dn
->parent
)
761 && (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
762 /* Parent supports EEH. */
763 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
764 pdn
->eeh_config_addr
= PCI_DN(dn
->parent
)->eeh_config_addr
;
769 printk(KERN_WARNING
"EEH: %s: unable to get reg property.\n",
777 * Initialize EEH by trying to enable it for all of the adapters in the system.
778 * As a side effect we can determine here if eeh is supported at all.
779 * Note that we leave EEH on so failed config cycles won't cause a machine
780 * check. If a user turns off EEH for a particular adapter they are really
781 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
782 * grant access to a slot if EEH isn't enabled, and so we always enable
783 * EEH for all slots/all devices.
785 * The eeh-force-off option disables EEH checking globally, for all slots.
786 * Even if force-off is set, the EEH hardware is still enabled, so that
787 * newer systems can boot.
789 void __init
eeh_init(void)
791 struct device_node
*phb
, *np
;
792 struct eeh_early_enable_info info
;
794 spin_lock_init(&confirm_error_lock
);
795 spin_lock_init(&slot_errbuf_lock
);
797 np
= of_find_node_by_path("/rtas");
801 ibm_set_eeh_option
= rtas_token("ibm,set-eeh-option");
802 ibm_set_slot_reset
= rtas_token("ibm,set-slot-reset");
803 ibm_read_slot_reset_state2
= rtas_token("ibm,read-slot-reset-state2");
804 ibm_read_slot_reset_state
= rtas_token("ibm,read-slot-reset-state");
805 ibm_slot_error_detail
= rtas_token("ibm,slot-error-detail");
806 ibm_get_config_addr_info
= rtas_token("ibm,get-config-addr-info");
807 ibm_configure_bridge
= rtas_token ("ibm,configure-bridge");
809 if (ibm_set_eeh_option
== RTAS_UNKNOWN_SERVICE
)
812 eeh_error_buf_size
= rtas_token("rtas-error-log-max");
813 if (eeh_error_buf_size
== RTAS_UNKNOWN_SERVICE
) {
814 eeh_error_buf_size
= 1024;
816 if (eeh_error_buf_size
> RTAS_ERROR_LOG_MAX
) {
817 printk(KERN_WARNING
"EEH: rtas-error-log-max is bigger than allocated "
818 "buffer ! (%d vs %d)", eeh_error_buf_size
, RTAS_ERROR_LOG_MAX
);
819 eeh_error_buf_size
= RTAS_ERROR_LOG_MAX
;
822 /* Enable EEH for all adapters. Note that eeh requires buid's */
823 for (phb
= of_find_node_by_name(NULL
, "pci"); phb
;
824 phb
= of_find_node_by_name(phb
, "pci")) {
827 buid
= get_phb_buid(phb
);
828 if (buid
== 0 || PCI_DN(phb
) == NULL
)
831 info
.buid_lo
= BUID_LO(buid
);
832 info
.buid_hi
= BUID_HI(buid
);
833 traverse_pci_devices(phb
, early_enable_eeh
, &info
);
836 if (eeh_subsystem_enabled
)
837 printk(KERN_INFO
"EEH: PCI Enhanced I/O Error Handling Enabled\n");
839 printk(KERN_WARNING
"EEH: No capable adapters found\n");
843 * eeh_add_device_early - enable EEH for the indicated device_node
844 * @dn: device node for which to set up EEH
846 * This routine must be used to perform EEH initialization for PCI
847 * devices that were added after system boot (e.g. hotplug, dlpar).
848 * This routine must be called before any i/o is performed to the
849 * adapter (inluding any config-space i/o).
850 * Whether this actually enables EEH or not for this device depends
851 * on the CEC architecture, type of the device, on earlier boot
852 * command-line arguments & etc.
854 void eeh_add_device_early(struct device_node
*dn
)
856 struct pci_controller
*phb
;
857 struct eeh_early_enable_info info
;
859 if (!dn
|| !PCI_DN(dn
))
861 phb
= PCI_DN(dn
)->phb
;
863 /* USB Bus children of PCI devices will not have BUID's */
864 if (NULL
== phb
|| 0 == phb
->buid
)
867 info
.buid_hi
= BUID_HI(phb
->buid
);
868 info
.buid_lo
= BUID_LO(phb
->buid
);
869 early_enable_eeh(dn
, &info
);
871 EXPORT_SYMBOL_GPL(eeh_add_device_early
);
873 void eeh_add_device_tree_early(struct device_node
*dn
)
875 struct device_node
*sib
;
876 for (sib
= dn
->child
; sib
; sib
= sib
->sibling
)
877 eeh_add_device_tree_early(sib
);
878 eeh_add_device_early(dn
);
880 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early
);
883 * eeh_add_device_late - perform EEH initialization for the indicated pci device
884 * @dev: pci device for which to set up EEH
886 * This routine must be used to complete EEH initialization for PCI
887 * devices that were added after system boot (e.g. hotplug, dlpar).
889 void eeh_add_device_late(struct pci_dev
*dev
)
891 struct device_node
*dn
;
894 if (!dev
|| !eeh_subsystem_enabled
)
898 printk(KERN_DEBUG
"EEH: adding device %s\n", pci_name(dev
));
902 dn
= pci_device_to_OF_node(dev
);
906 pci_addr_cache_insert_device (dev
);
907 eeh_save_bars(dev
, pdn
);
909 EXPORT_SYMBOL_GPL(eeh_add_device_late
);
912 * eeh_remove_device - undo EEH setup for the indicated pci device
913 * @dev: pci device to be removed
915 * This routine should be when a device is removed from a running
916 * system (e.g. by hotplug or dlpar).
918 void eeh_remove_device(struct pci_dev
*dev
)
920 struct device_node
*dn
;
921 if (!dev
|| !eeh_subsystem_enabled
)
924 /* Unregister the device with the EEH/PCI address search system */
926 printk(KERN_DEBUG
"EEH: remove device %s\n", pci_name(dev
));
928 pci_addr_cache_remove_device(dev
);
930 dn
= pci_device_to_OF_node(dev
);
931 PCI_DN(dn
)->pcidev
= NULL
;
934 EXPORT_SYMBOL_GPL(eeh_remove_device
);
936 void eeh_remove_bus_device(struct pci_dev
*dev
)
938 eeh_remove_device(dev
);
939 if (dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
940 struct pci_bus
*bus
= dev
->subordinate
;
941 struct list_head
*ln
;
944 for (ln
= bus
->devices
.next
; ln
!= &bus
->devices
; ln
= ln
->next
) {
945 struct pci_dev
*pdev
= pci_dev_b(ln
);
947 eeh_remove_bus_device(pdev
);
951 EXPORT_SYMBOL_GPL(eeh_remove_bus_device
);
953 static int proc_eeh_show(struct seq_file
*m
, void *v
)
956 unsigned long ffs
= 0, positives
= 0, failures
= 0;
957 unsigned long resets
= 0;
958 unsigned long no_dev
= 0, no_dn
= 0, no_cfg
= 0, no_check
= 0;
961 ffs
+= per_cpu(total_mmio_ffs
, cpu
);
962 positives
+= per_cpu(false_positives
, cpu
);
963 failures
+= per_cpu(ignored_failures
, cpu
);
964 resets
+= per_cpu(slot_resets
, cpu
);
965 no_dev
+= per_cpu(no_device
, cpu
);
966 no_dn
+= per_cpu(no_dn
, cpu
);
967 no_cfg
+= per_cpu(no_cfg_addr
, cpu
);
968 no_check
+= per_cpu(ignored_check
, cpu
);
971 if (0 == eeh_subsystem_enabled
) {
972 seq_printf(m
, "EEH Subsystem is globally disabled\n");
973 seq_printf(m
, "eeh_total_mmio_ffs=%ld\n", ffs
);
975 seq_printf(m
, "EEH Subsystem is enabled\n");
978 "no device node=%ld\n"
979 "no config address=%ld\n"
980 "check not wanted=%ld\n"
981 "eeh_total_mmio_ffs=%ld\n"
982 "eeh_false_positives=%ld\n"
983 "eeh_ignored_failures=%ld\n"
984 "eeh_slot_resets=%ld\n",
985 no_dev
, no_dn
, no_cfg
, no_check
,
986 ffs
, positives
, failures
, resets
);
992 static int proc_eeh_open(struct inode
*inode
, struct file
*file
)
994 return single_open(file
, proc_eeh_show
, NULL
);
997 static struct file_operations proc_eeh_operations
= {
998 .open
= proc_eeh_open
,
1000 .llseek
= seq_lseek
,
1001 .release
= single_release
,
1004 static int __init
eeh_init_proc(void)
1006 struct proc_dir_entry
*e
;
1008 if (platform_is_pseries()) {
1009 e
= create_proc_entry("ppc64/eeh", 0, NULL
);
1011 e
->proc_fops
= &proc_eeh_operations
;
1016 __initcall(eeh_init_proc
);