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
)
166 n
+= scnprintf(buf
+n
, len
-n
, "%s\n", pdn
->node
->full_name
);
167 printk(KERN_WARNING
"EEH: of node=%s\n", pdn
->node
->full_name
);
169 rtas_read_config(pdn
, PCI_VENDOR_ID
, 4, &cfg
);
170 n
+= scnprintf(buf
+n
, len
-n
, "dev/vend:%08x\n", cfg
);
171 printk(KERN_WARNING
"EEH: PCI device/vendor: %08x\n", cfg
);
173 rtas_read_config(pdn
, PCI_COMMAND
, 4, &cfg
);
174 n
+= scnprintf(buf
+n
, len
-n
, "cmd/stat:%x\n", cfg
);
175 printk(KERN_WARNING
"EEH: PCI cmd/status register: %08x\n", cfg
);
177 /* Dump out the PCI-X command and status regs */
178 cap
= pci_find_capability(pdn
->pcidev
, PCI_CAP_ID_PCIX
);
180 rtas_read_config(pdn
, cap
, 4, &cfg
);
181 n
+= scnprintf(buf
+n
, len
-n
, "pcix-cmd:%x\n", cfg
);
182 printk(KERN_WARNING
"EEH: PCI-X cmd: %08x\n", cfg
);
184 rtas_read_config(pdn
, cap
+4, 4, &cfg
);
185 n
+= scnprintf(buf
+n
, len
-n
, "pcix-stat:%x\n", cfg
);
186 printk(KERN_WARNING
"EEH: PCI-X status: %08x\n", cfg
);
189 /* If PCI-E capable, dump PCI-E cap 10, and the AER */
190 cap
= pci_find_capability(pdn
->pcidev
, PCI_CAP_ID_EXP
);
192 n
+= scnprintf(buf
+n
, len
-n
, "pci-e cap10:\n");
194 "EEH: PCI-E capabilities and status follow:\n");
196 for (i
=0; i
<=8; i
++) {
197 rtas_read_config(pdn
, cap
+4*i
, 4, &cfg
);
198 n
+= scnprintf(buf
+n
, len
-n
, "%02x:%x\n", 4*i
, cfg
);
199 printk(KERN_WARNING
"EEH: PCI-E %02x: %08x\n", i
, cfg
);
202 cap
= pci_find_ext_capability(pdn
->pcidev
,PCI_EXT_CAP_ID_ERR
);
204 n
+= scnprintf(buf
+n
, len
-n
, "pci-e AER:\n");
206 "EEH: PCI-E AER capability register set follows:\n");
208 for (i
=0; i
<14; i
++) {
209 rtas_read_config(pdn
, cap
+4*i
, 4, &cfg
);
210 n
+= scnprintf(buf
+n
, len
-n
, "%02x:%x\n", 4*i
, cfg
);
211 printk(KERN_WARNING
"EEH: PCI-E AER %02x: %08x\n", i
, cfg
);
218 void eeh_slot_error_detail(struct pci_dn
*pdn
, int severity
)
221 memset(pci_regs_buf
, 0, EEH_PCI_REGS_LOG_LEN
);
223 rtas_pci_enable(pdn
, EEH_THAW_MMIO
);
224 loglen
= gather_pci_data(pdn
, pci_regs_buf
, EEH_PCI_REGS_LOG_LEN
);
226 rtas_slot_error_detail(pdn
, severity
, pci_regs_buf
, loglen
);
230 * read_slot_reset_state - Read the reset state of a device node's slot
231 * @dn: device node to read
232 * @rets: array to return results in
234 static int read_slot_reset_state(struct pci_dn
*pdn
, int rets
[])
239 if (ibm_read_slot_reset_state2
!= RTAS_UNKNOWN_SERVICE
) {
240 token
= ibm_read_slot_reset_state2
;
243 token
= ibm_read_slot_reset_state
;
244 rets
[2] = 0; /* fake PE Unavailable info */
248 /* Use PE configuration address, if present */
249 config_addr
= pdn
->eeh_config_addr
;
250 if (pdn
->eeh_pe_config_addr
)
251 config_addr
= pdn
->eeh_pe_config_addr
;
253 return rtas_call(token
, 3, outputs
, rets
, config_addr
,
254 BUID_HI(pdn
->phb
->buid
), BUID_LO(pdn
->phb
->buid
));
258 * eeh_wait_for_slot_status - returns error status of slot
259 * @pdn pci device node
260 * @max_wait_msecs maximum number to millisecs to wait
262 * Return negative value if a permanent error, else return
263 * Partition Endpoint (PE) status value.
265 * If @max_wait_msecs is positive, then this routine will
266 * sleep until a valid status can be obtained, or until
267 * the max allowed wait time is exceeded, in which case
271 eeh_wait_for_slot_status(struct pci_dn
*pdn
, int max_wait_msecs
)
278 rc
= read_slot_reset_state(pdn
, rets
);
280 if (rets
[1] == 0) return -1; /* EEH is not supported */
282 if (rets
[0] != 5) return rets
[0]; /* return actual status */
284 if (rets
[2] == 0) return -1; /* permanently unavailable */
286 if (max_wait_msecs
<= 0) return -1;
291 "EEH: Firmware returned bad wait value=%d\n", mwait
);
293 } else if (mwait
> 300*1000) {
295 "EEH: Firmware is taking too long, time=%d\n", mwait
);
298 max_wait_msecs
-= mwait
;
302 printk(KERN_WARNING
"EEH: Timed out waiting for slot status\n");
307 * eeh_token_to_phys - convert EEH address token to phys address
308 * @token i/o token, should be address in the form 0xA....
310 static inline unsigned long eeh_token_to_phys(unsigned long token
)
315 ptep
= find_linux_pte(init_mm
.pgd
, token
);
318 pa
= pte_pfn(*ptep
) << PAGE_SHIFT
;
320 return pa
| (token
& (PAGE_SIZE
-1));
324 * Return the "partitionable endpoint" (pe) under which this device lies
326 struct device_node
* find_device_pe(struct device_node
*dn
)
328 while ((dn
->parent
) && PCI_DN(dn
->parent
) &&
329 (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
335 /** Mark all devices that are peers of this device as failed.
336 * Mark the device driver too, so that it can see the failure
337 * immediately; this is critical, since some drivers poll
338 * status registers in interrupts ... If a driver is polling,
339 * and the slot is frozen, then the driver can deadlock in
340 * an interrupt context, which is bad.
343 static void __eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
347 /* Mark the pci device driver too */
348 struct pci_dev
*dev
= PCI_DN(dn
)->pcidev
;
350 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
352 if (dev
&& dev
->driver
)
353 dev
->error_state
= pci_channel_io_frozen
;
356 __eeh_mark_slot (dn
->child
, mode_flag
);
362 void eeh_mark_slot (struct device_node
*dn
, int mode_flag
)
365 dn
= find_device_pe (dn
);
367 /* Back up one, since config addrs might be shared */
368 if (!pcibios_find_pci_bus(dn
) && PCI_DN(dn
->parent
))
371 PCI_DN(dn
)->eeh_mode
|= mode_flag
;
373 /* Mark the pci device too */
374 dev
= PCI_DN(dn
)->pcidev
;
376 dev
->error_state
= pci_channel_io_frozen
;
378 __eeh_mark_slot (dn
->child
, mode_flag
);
381 static void __eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
385 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
386 PCI_DN(dn
)->eeh_check_count
= 0;
388 __eeh_clear_slot (dn
->child
, mode_flag
);
394 void eeh_clear_slot (struct device_node
*dn
, int mode_flag
)
397 spin_lock_irqsave(&confirm_error_lock
, flags
);
399 dn
= find_device_pe (dn
);
401 /* Back up one, since config addrs might be shared */
402 if (!pcibios_find_pci_bus(dn
) && PCI_DN(dn
->parent
))
405 PCI_DN(dn
)->eeh_mode
&= ~mode_flag
;
406 PCI_DN(dn
)->eeh_check_count
= 0;
407 __eeh_clear_slot (dn
->child
, mode_flag
);
408 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
412 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
414 * @dev pci device, if known
416 * Check for an EEH failure for the given device node. Call this
417 * routine if the result of a read was all 0xff's and you want to
418 * find out if this is due to an EEH slot freeze. This routine
419 * will query firmware for the EEH status.
421 * Returns 0 if there has not been an EEH error; otherwise returns
422 * a non-zero value and queues up a slot isolation event notification.
424 * It is safe to call this routine in an interrupt context.
426 int eeh_dn_check_failure(struct device_node
*dn
, struct pci_dev
*dev
)
436 if (!eeh_subsystem_enabled
)
445 /* Access to IO BARs might get this far and still not want checking. */
446 if (!(pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) ||
447 pdn
->eeh_mode
& EEH_MODE_NOCHECK
) {
450 printk ("EEH:ignored check (%x) for %s %s\n",
451 pdn
->eeh_mode
, pci_name (dev
), dn
->full_name
);
456 if (!pdn
->eeh_config_addr
&& !pdn
->eeh_pe_config_addr
) {
461 /* If we already have a pending isolation event for this
462 * slot, we know it's bad already, we don't need to check.
463 * Do this checking under a lock; as multiple PCI devices
464 * in one slot might report errors simultaneously, and we
465 * only want one error recovery routine running.
467 spin_lock_irqsave(&confirm_error_lock
, flags
);
469 if (pdn
->eeh_mode
& EEH_MODE_ISOLATED
) {
470 pdn
->eeh_check_count
++;
471 if (pdn
->eeh_check_count
>= EEH_MAX_FAILS
) {
472 printk (KERN_ERR
"EEH: Device driver ignored %d bad reads, panicing\n",
473 pdn
->eeh_check_count
);
477 /* re-read the slot reset state */
478 if (read_slot_reset_state(pdn
, rets
) != 0)
479 rets
[0] = -1; /* reset state unknown */
481 /* If we are here, then we hit an infinite loop. Stop. */
482 panic("EEH: MMIO halt (%d) on device:%s\n", rets
[0], pci_name(dev
));
488 * Now test for an EEH failure. This is VERY expensive.
489 * Note that the eeh_config_addr may be a parent device
490 * in the case of a device behind a bridge, or it may be
491 * function zero of a multi-function device.
492 * In any case they must share a common PHB.
494 ret
= read_slot_reset_state(pdn
, rets
);
496 /* If the call to firmware failed, punt */
498 printk(KERN_WARNING
"EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
505 /* Note that config-io to empty slots may fail;
506 * they are empty when they don't have children. */
507 if ((rets
[0] == 5) && (dn
->child
== NULL
)) {
513 /* If EEH is not supported on this device, punt. */
515 printk(KERN_WARNING
"EEH: event on unsupported device, rc=%d dn=%s\n",
522 /* If not the kind of error we know about, punt. */
523 if (rets
[0] != 1 && rets
[0] != 2 && rets
[0] != 4 && rets
[0] != 5) {
531 /* Avoid repeated reports of this failure, including problems
532 * with other functions on this device, and functions under
534 eeh_mark_slot (dn
, EEH_MODE_ISOLATED
);
535 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
537 eeh_send_failure_event (dn
, dev
);
539 /* Most EEH events are due to device driver bugs. Having
540 * a stack trace will help the device-driver authors figure
541 * out what happened. So print that out. */
546 spin_unlock_irqrestore(&confirm_error_lock
, flags
);
550 EXPORT_SYMBOL_GPL(eeh_dn_check_failure
);
553 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
554 * @token i/o token, should be address in the form 0xA....
555 * @val value, should be all 1's (XXX why do we need this arg??)
557 * Check for an EEH failure at the given token address. Call this
558 * routine if the result of a read was all 0xff's and you want to
559 * find out if this is due to an EEH slot freeze event. This routine
560 * will query firmware for the EEH status.
562 * Note this routine is safe to call in an interrupt context.
564 unsigned long eeh_check_failure(const volatile void __iomem
*token
, unsigned long val
)
568 struct device_node
*dn
;
570 /* Finding the phys addr + pci device; this is pretty quick. */
571 addr
= eeh_token_to_phys((unsigned long __force
) token
);
572 dev
= pci_get_device_by_addr(addr
);
578 dn
= pci_device_to_OF_node(dev
);
579 eeh_dn_check_failure (dn
, dev
);
585 EXPORT_SYMBOL(eeh_check_failure
);
587 /* ------------------------------------------------------------- */
588 /* The code below deals with error recovery */
591 * rtas_pci_enable - enable MMIO or DMA transfers for this slot
592 * @pdn pci device node
596 rtas_pci_enable(struct pci_dn
*pdn
, int function
)
601 /* Use PE configuration address, if present */
602 config_addr
= pdn
->eeh_config_addr
;
603 if (pdn
->eeh_pe_config_addr
)
604 config_addr
= pdn
->eeh_pe_config_addr
;
606 rc
= rtas_call(ibm_set_eeh_option
, 4, 1, NULL
,
608 BUID_HI(pdn
->phb
->buid
),
609 BUID_LO(pdn
->phb
->buid
),
613 printk(KERN_WARNING
"EEH: Unexpected state change %d, err=%d dn=%s\n",
614 function
, rc
, pdn
->node
->full_name
);
616 rc
= eeh_wait_for_slot_status (pdn
, PCI_BUS_RESET_WAIT_MSEC
);
617 if ((rc
== 4) && (function
== EEH_THAW_MMIO
))
624 * rtas_pci_slot_reset - raises/lowers the pci #RST line
625 * @pdn pci device node
626 * @state: 1/0 to raise/lower the #RST
628 * Clear the EEH-frozen condition on a slot. This routine
629 * asserts the PCI #RST line if the 'state' argument is '1',
630 * and drops the #RST line if 'state is '0'. This routine is
631 * safe to call in an interrupt context.
636 rtas_pci_slot_reset(struct pci_dn
*pdn
, int state
)
644 printk (KERN_WARNING
"EEH: in slot reset, device node %s has no phb\n",
645 pdn
->node
->full_name
);
649 /* Use PE configuration address, if present */
650 config_addr
= pdn
->eeh_config_addr
;
651 if (pdn
->eeh_pe_config_addr
)
652 config_addr
= pdn
->eeh_pe_config_addr
;
654 rc
= rtas_call(ibm_set_slot_reset
,4,1, NULL
,
656 BUID_HI(pdn
->phb
->buid
),
657 BUID_LO(pdn
->phb
->buid
),
660 printk (KERN_WARNING
"EEH: Unable to reset the failed slot,"
661 " (%d) #RST=%d dn=%s\n",
662 rc
, state
, pdn
->node
->full_name
);
666 * pcibios_set_pcie_slot_reset - Set PCI-E reset state
667 * @dev: pci device struct
668 * @state: reset state to enter
673 int pcibios_set_pcie_reset_state(struct pci_dev
*dev
, enum pcie_reset_state state
)
675 struct device_node
*dn
= pci_device_to_OF_node(dev
);
676 struct pci_dn
*pdn
= PCI_DN(dn
);
679 case pcie_deassert_reset
:
680 rtas_pci_slot_reset(pdn
, 0);
683 rtas_pci_slot_reset(pdn
, 1);
685 case pcie_warm_reset
:
686 rtas_pci_slot_reset(pdn
, 3);
696 * rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
697 * @pdn: pci device node to be reset.
699 * Return 0 if success, else a non-zero value.
702 static void __rtas_set_slot_reset(struct pci_dn
*pdn
)
704 rtas_pci_slot_reset (pdn
, 1);
706 /* The PCI bus requires that the reset be held high for at least
707 * a 100 milliseconds. We wait a bit longer 'just in case'. */
709 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
710 msleep (PCI_BUS_RST_HOLD_TIME_MSEC
);
712 /* We might get hit with another EEH freeze as soon as the
713 * pci slot reset line is dropped. Make sure we don't miss
714 * these, and clear the flag now. */
715 eeh_clear_slot (pdn
->node
, EEH_MODE_ISOLATED
);
717 rtas_pci_slot_reset (pdn
, 0);
719 /* After a PCI slot has been reset, the PCI Express spec requires
720 * a 1.5 second idle time for the bus to stabilize, before starting
722 #define PCI_BUS_SETTLE_TIME_MSEC 1800
723 msleep (PCI_BUS_SETTLE_TIME_MSEC
);
726 int rtas_set_slot_reset(struct pci_dn
*pdn
)
730 /* Take three shots at resetting the bus */
731 for (i
=0; i
<3; i
++) {
732 __rtas_set_slot_reset(pdn
);
734 rc
= eeh_wait_for_slot_status(pdn
, PCI_BUS_RESET_WAIT_MSEC
);
739 printk (KERN_ERR
"EEH: unrecoverable slot failure %s\n",
740 pdn
->node
->full_name
);
743 printk (KERN_ERR
"EEH: bus reset %d failed on slot %s\n",
744 i
+1, pdn
->node
->full_name
);
750 /* ------------------------------------------------------- */
751 /** Save and restore of PCI BARs
753 * Although firmware will set up BARs during boot, it doesn't
754 * set up device BAR's after a device reset, although it will,
755 * if requested, set up bridge configuration. Thus, we need to
756 * configure the PCI devices ourselves.
760 * __restore_bars - Restore the Base Address Registers
761 * @pdn: pci device node
763 * Loads the PCI configuration space base address registers,
764 * the expansion ROM base address, the latency timer, and etc.
765 * from the saved values in the device node.
767 static inline void __restore_bars (struct pci_dn
*pdn
)
771 if (NULL
==pdn
->phb
) return;
772 for (i
=4; i
<10; i
++) {
773 rtas_write_config(pdn
, i
*4, 4, pdn
->config_space
[i
]);
776 /* 12 == Expansion ROM Address */
777 rtas_write_config(pdn
, 12*4, 4, pdn
->config_space
[12]);
779 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
780 #define SAVED_BYTE(OFF) (((u8 *)(pdn->config_space))[BYTE_SWAP(OFF)])
782 rtas_write_config (pdn
, PCI_CACHE_LINE_SIZE
, 1,
783 SAVED_BYTE(PCI_CACHE_LINE_SIZE
));
785 rtas_write_config (pdn
, PCI_LATENCY_TIMER
, 1,
786 SAVED_BYTE(PCI_LATENCY_TIMER
));
788 /* max latency, min grant, interrupt pin and line */
789 rtas_write_config(pdn
, 15*4, 4, pdn
->config_space
[15]);
793 * eeh_restore_bars - restore the PCI config space info
795 * This routine performs a recursive walk to the children
796 * of this device as well.
798 void eeh_restore_bars(struct pci_dn
*pdn
)
800 struct device_node
*dn
;
804 if ((pdn
->eeh_mode
& EEH_MODE_SUPPORTED
) && !IS_BRIDGE(pdn
->class_code
))
805 __restore_bars (pdn
);
807 dn
= pdn
->node
->child
;
809 eeh_restore_bars (PCI_DN(dn
));
815 * eeh_save_bars - save device bars
817 * Save the values of the device bars. Unlike the restore
818 * routine, this routine is *not* recursive. This is because
819 * PCI devices are added individuallly; but, for the restore,
820 * an entire slot is reset at a time.
822 static void eeh_save_bars(struct pci_dn
*pdn
)
829 for (i
= 0; i
< 16; i
++)
830 rtas_read_config(pdn
, i
* 4, 4, &pdn
->config_space
[i
]);
834 rtas_configure_bridge(struct pci_dn
*pdn
)
839 /* Use PE configuration address, if present */
840 config_addr
= pdn
->eeh_config_addr
;
841 if (pdn
->eeh_pe_config_addr
)
842 config_addr
= pdn
->eeh_pe_config_addr
;
844 rc
= rtas_call(ibm_configure_bridge
,3,1, NULL
,
846 BUID_HI(pdn
->phb
->buid
),
847 BUID_LO(pdn
->phb
->buid
));
849 printk (KERN_WARNING
"EEH: Unable to configure device bridge (%d) for %s\n",
850 rc
, pdn
->node
->full_name
);
854 /* ------------------------------------------------------------- */
855 /* The code below deals with enabling EEH for devices during the
856 * early boot sequence. EEH must be enabled before any PCI probing
862 struct eeh_early_enable_info
{
863 unsigned int buid_hi
;
864 unsigned int buid_lo
;
867 static int get_pe_addr (int config_addr
,
868 struct eeh_early_enable_info
*info
)
870 unsigned int rets
[3];
873 /* Use latest config-addr token on power6 */
874 if (ibm_get_config_addr_info2
!= RTAS_UNKNOWN_SERVICE
) {
875 /* Make sure we have a PE in hand */
876 ret
= rtas_call (ibm_get_config_addr_info2
, 4, 2, rets
,
877 config_addr
, info
->buid_hi
, info
->buid_lo
, 1);
878 if (ret
|| (rets
[0]==0))
881 ret
= rtas_call (ibm_get_config_addr_info2
, 4, 2, rets
,
882 config_addr
, info
->buid_hi
, info
->buid_lo
, 0);
888 /* Use older config-addr token on power5 */
889 if (ibm_get_config_addr_info
!= RTAS_UNKNOWN_SERVICE
) {
890 ret
= rtas_call (ibm_get_config_addr_info
, 4, 2, rets
,
891 config_addr
, info
->buid_hi
, info
->buid_lo
, 0);
899 /* Enable eeh for the given device node. */
900 static void *early_enable_eeh(struct device_node
*dn
, void *data
)
902 unsigned int rets
[3];
903 struct eeh_early_enable_info
*info
= data
;
905 const char *status
= of_get_property(dn
, "status", NULL
);
906 const u32
*class_code
= of_get_property(dn
, "class-code", NULL
);
907 const u32
*vendor_id
= of_get_property(dn
, "vendor-id", NULL
);
908 const u32
*device_id
= of_get_property(dn
, "device-id", NULL
);
911 struct pci_dn
*pdn
= PCI_DN(dn
);
915 pdn
->eeh_check_count
= 0;
916 pdn
->eeh_freeze_count
= 0;
918 if (status
&& strcmp(status
, "ok") != 0)
919 return NULL
; /* ignore devices with bad status */
921 /* Ignore bad nodes. */
922 if (!class_code
|| !vendor_id
|| !device_id
)
925 /* There is nothing to check on PCI to ISA bridges */
926 if (dn
->type
&& !strcmp(dn
->type
, "isa")) {
927 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
930 pdn
->class_code
= *class_code
;
933 * Now decide if we are going to "Disable" EEH checking
934 * for this device. We still run with the EEH hardware active,
935 * but we won't be checking for ff's. This means a driver
936 * could return bad data (very bad!), an interrupt handler could
937 * hang waiting on status bits that won't change, etc.
938 * But there are a few cases like display devices that make sense.
940 enable
= 1; /* i.e. we will do checking */
942 if ((*class_code
>> 16) == PCI_BASE_CLASS_DISPLAY
)
947 pdn
->eeh_mode
|= EEH_MODE_NOCHECK
;
949 /* Ok... see if this device supports EEH. Some do, some don't,
950 * and the only way to find out is to check each and every one. */
951 regs
= of_get_property(dn
, "reg", NULL
);
953 /* First register entry is addr (00BBSS00) */
954 /* Try to enable eeh */
955 ret
= rtas_call(ibm_set_eeh_option
, 4, 1, NULL
,
956 regs
[0], info
->buid_hi
, info
->buid_lo
,
961 pdn
->eeh_config_addr
= regs
[0];
963 /* If the newer, better, ibm,get-config-addr-info is supported,
964 * then use that instead. */
965 pdn
->eeh_pe_config_addr
= get_pe_addr(pdn
->eeh_config_addr
, info
);
967 /* Some older systems (Power4) allow the
968 * ibm,set-eeh-option call to succeed even on nodes
969 * where EEH is not supported. Verify support
971 ret
= read_slot_reset_state(pdn
, rets
);
972 if ((ret
== 0) && (rets
[1] == 1))
977 eeh_subsystem_enabled
= 1;
978 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
981 printk(KERN_DEBUG
"EEH: %s: eeh enabled, config=%x pe_config=%x\n",
982 dn
->full_name
, pdn
->eeh_config_addr
, pdn
->eeh_pe_config_addr
);
986 /* This device doesn't support EEH, but it may have an
987 * EEH parent, in which case we mark it as supported. */
988 if (dn
->parent
&& PCI_DN(dn
->parent
)
989 && (PCI_DN(dn
->parent
)->eeh_mode
& EEH_MODE_SUPPORTED
)) {
990 /* Parent supports EEH. */
991 pdn
->eeh_mode
|= EEH_MODE_SUPPORTED
;
992 pdn
->eeh_config_addr
= PCI_DN(dn
->parent
)->eeh_config_addr
;
997 printk(KERN_WARNING
"EEH: %s: unable to get reg property.\n",
1006 * Initialize EEH by trying to enable it for all of the adapters in the system.
1007 * As a side effect we can determine here if eeh is supported at all.
1008 * Note that we leave EEH on so failed config cycles won't cause a machine
1009 * check. If a user turns off EEH for a particular adapter they are really
1010 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
1011 * grant access to a slot if EEH isn't enabled, and so we always enable
1012 * EEH for all slots/all devices.
1014 * The eeh-force-off option disables EEH checking globally, for all slots.
1015 * Even if force-off is set, the EEH hardware is still enabled, so that
1016 * newer systems can boot.
1018 void __init
eeh_init(void)
1020 struct device_node
*phb
, *np
;
1021 struct eeh_early_enable_info info
;
1023 spin_lock_init(&confirm_error_lock
);
1024 spin_lock_init(&slot_errbuf_lock
);
1026 np
= of_find_node_by_path("/rtas");
1030 ibm_set_eeh_option
= rtas_token("ibm,set-eeh-option");
1031 ibm_set_slot_reset
= rtas_token("ibm,set-slot-reset");
1032 ibm_read_slot_reset_state2
= rtas_token("ibm,read-slot-reset-state2");
1033 ibm_read_slot_reset_state
= rtas_token("ibm,read-slot-reset-state");
1034 ibm_slot_error_detail
= rtas_token("ibm,slot-error-detail");
1035 ibm_get_config_addr_info
= rtas_token("ibm,get-config-addr-info");
1036 ibm_get_config_addr_info2
= rtas_token("ibm,get-config-addr-info2");
1037 ibm_configure_bridge
= rtas_token ("ibm,configure-bridge");
1039 if (ibm_set_eeh_option
== RTAS_UNKNOWN_SERVICE
)
1042 eeh_error_buf_size
= rtas_token("rtas-error-log-max");
1043 if (eeh_error_buf_size
== RTAS_UNKNOWN_SERVICE
) {
1044 eeh_error_buf_size
= 1024;
1046 if (eeh_error_buf_size
> RTAS_ERROR_LOG_MAX
) {
1047 printk(KERN_WARNING
"EEH: rtas-error-log-max is bigger than allocated "
1048 "buffer ! (%d vs %d)", eeh_error_buf_size
, RTAS_ERROR_LOG_MAX
);
1049 eeh_error_buf_size
= RTAS_ERROR_LOG_MAX
;
1052 /* Enable EEH for all adapters. Note that eeh requires buid's */
1053 for (phb
= of_find_node_by_name(NULL
, "pci"); phb
;
1054 phb
= of_find_node_by_name(phb
, "pci")) {
1057 buid
= get_phb_buid(phb
);
1058 if (buid
== 0 || PCI_DN(phb
) == NULL
)
1061 info
.buid_lo
= BUID_LO(buid
);
1062 info
.buid_hi
= BUID_HI(buid
);
1063 traverse_pci_devices(phb
, early_enable_eeh
, &info
);
1066 if (eeh_subsystem_enabled
)
1067 printk(KERN_INFO
"EEH: PCI Enhanced I/O Error Handling Enabled\n");
1069 printk(KERN_WARNING
"EEH: No capable adapters found\n");
1073 * eeh_add_device_early - enable EEH for the indicated device_node
1074 * @dn: device node for which to set up EEH
1076 * This routine must be used to perform EEH initialization for PCI
1077 * devices that were added after system boot (e.g. hotplug, dlpar).
1078 * This routine must be called before any i/o is performed to the
1079 * adapter (inluding any config-space i/o).
1080 * Whether this actually enables EEH or not for this device depends
1081 * on the CEC architecture, type of the device, on earlier boot
1082 * command-line arguments & etc.
1084 static void eeh_add_device_early(struct device_node
*dn
)
1086 struct pci_controller
*phb
;
1087 struct eeh_early_enable_info info
;
1089 if (!dn
|| !PCI_DN(dn
))
1091 phb
= PCI_DN(dn
)->phb
;
1093 /* USB Bus children of PCI devices will not have BUID's */
1094 if (NULL
== phb
|| 0 == phb
->buid
)
1097 info
.buid_hi
= BUID_HI(phb
->buid
);
1098 info
.buid_lo
= BUID_LO(phb
->buid
);
1099 early_enable_eeh(dn
, &info
);
1102 void eeh_add_device_tree_early(struct device_node
*dn
)
1104 struct device_node
*sib
;
1105 for (sib
= dn
->child
; sib
; sib
= sib
->sibling
)
1106 eeh_add_device_tree_early(sib
);
1107 eeh_add_device_early(dn
);
1109 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early
);
1112 * eeh_add_device_late - perform EEH initialization for the indicated pci device
1113 * @dev: pci device for which to set up EEH
1115 * This routine must be used to complete EEH initialization for PCI
1116 * devices that were added after system boot (e.g. hotplug, dlpar).
1118 static void eeh_add_device_late(struct pci_dev
*dev
)
1120 struct device_node
*dn
;
1123 if (!dev
|| !eeh_subsystem_enabled
)
1127 printk(KERN_DEBUG
"EEH: adding device %s\n", pci_name(dev
));
1131 dn
= pci_device_to_OF_node(dev
);
1135 pci_addr_cache_insert_device (dev
);
1138 void eeh_add_device_tree_late(struct pci_bus
*bus
)
1140 struct pci_dev
*dev
;
1142 list_for_each_entry(dev
, &bus
->devices
, bus_list
) {
1143 eeh_add_device_late(dev
);
1144 if (dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
1145 struct pci_bus
*subbus
= dev
->subordinate
;
1147 eeh_add_device_tree_late(subbus
);
1151 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late
);
1154 * eeh_remove_device - undo EEH setup for the indicated pci device
1155 * @dev: pci device to be removed
1157 * This routine should be called when a device is removed from
1158 * a running system (e.g. by hotplug or dlpar). It unregisters
1159 * the PCI device from the EEH subsystem. I/O errors affecting
1160 * this device will no longer be detected after this call; thus,
1161 * i/o errors affecting this slot may leave this device unusable.
1163 static void eeh_remove_device(struct pci_dev
*dev
)
1165 struct device_node
*dn
;
1166 if (!dev
|| !eeh_subsystem_enabled
)
1169 /* Unregister the device with the EEH/PCI address search system */
1171 printk(KERN_DEBUG
"EEH: remove device %s\n", pci_name(dev
));
1173 pci_addr_cache_remove_device(dev
);
1175 dn
= pci_device_to_OF_node(dev
);
1176 if (PCI_DN(dn
)->pcidev
) {
1177 PCI_DN(dn
)->pcidev
= NULL
;
1182 void eeh_remove_bus_device(struct pci_dev
*dev
)
1184 struct pci_bus
*bus
= dev
->subordinate
;
1185 struct pci_dev
*child
, *tmp
;
1187 eeh_remove_device(dev
);
1189 if (bus
&& dev
->hdr_type
== PCI_HEADER_TYPE_BRIDGE
) {
1190 list_for_each_entry_safe(child
, tmp
, &bus
->devices
, bus_list
)
1191 eeh_remove_bus_device(child
);
1194 EXPORT_SYMBOL_GPL(eeh_remove_bus_device
);
1196 static int proc_eeh_show(struct seq_file
*m
, void *v
)
1198 if (0 == eeh_subsystem_enabled
) {
1199 seq_printf(m
, "EEH Subsystem is globally disabled\n");
1200 seq_printf(m
, "eeh_total_mmio_ffs=%ld\n", total_mmio_ffs
);
1202 seq_printf(m
, "EEH Subsystem is enabled\n");
1205 "no device node=%ld\n"
1206 "no config address=%ld\n"
1207 "check not wanted=%ld\n"
1208 "eeh_total_mmio_ffs=%ld\n"
1209 "eeh_false_positives=%ld\n"
1210 "eeh_ignored_failures=%ld\n"
1211 "eeh_slot_resets=%ld\n",
1212 no_device
, no_dn
, no_cfg_addr
,
1213 ignored_check
, total_mmio_ffs
,
1214 false_positives
, ignored_failures
,
1221 static int proc_eeh_open(struct inode
*inode
, struct file
*file
)
1223 return single_open(file
, proc_eeh_show
, NULL
);
1226 static const struct file_operations proc_eeh_operations
= {
1227 .open
= proc_eeh_open
,
1229 .llseek
= seq_lseek
,
1230 .release
= single_release
,
1233 static int __init
eeh_init_proc(void)
1235 struct proc_dir_entry
*e
;
1237 if (machine_is(pseries
)) {
1238 e
= create_proc_entry("ppc64/eeh", 0, NULL
);
1240 e
->proc_fops
= &proc_eeh_operations
;
1245 __initcall(eeh_init_proc
);