xref: /illumos-gate/usr/src/uts/sun4u/io/pci/pci_ecc.c (revision 45526e9775395f5d44bad3f5430041f32c84ce1e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * PCI ECC support
31  */
32 
33 #include <sys/types.h>
34 #include <sys/systm.h>		/* for strrchr */
35 #include <sys/kmem.h>
36 #include <sys/sunddi.h>
37 #include <sys/intr.h>
38 #include <sys/async.h>		/* struct async_flt */
39 #include <sys/ddi_impldefs.h>
40 #include <sys/machsystm.h>
41 #include <sys/sysmacros.h>
42 #include <sys/fm/protocol.h>
43 #include <sys/fm/util.h>
44 #include <sys/fm/io/pci.h>
45 #include <sys/fm/io/sun4upci.h>
46 #include <sys/fm/io/ddi.h>
47 #include <sys/pci/pci_obj.h>	/* ld/st physio */
48 #include <sys/cpuvar.h>
49 #include <sys/errclassify.h>
50 
51 /*LINTLIBRARY*/
52 
53 static void ecc_disable(ecc_t *, int);
54 static void ecc_delayed_ce(void *);
55 static uint64_t ecc_read_afsr(ecc_intr_info_t *);
56 static void ecc_ereport_post(dev_info_t *dip, ecc_errstate_t *ecc_err);
57 
58 clock_t pci_ecc_panic_delay = 200;
59 int ecc_ce_delay_secs = 6;	/* number of sec to delay reenabling of CEs */
60 int ecc_ce_delayed = 1;		/* global for enabling/disabling CE delay */
61 
62 void
63 ecc_create(pci_t *pci_p)
64 {
65 #ifdef DEBUG
66 	dev_info_t *dip = pci_p->pci_dip;
67 #endif
68 	uint64_t cb_base_pa = pci_p->pci_cb_p->cb_base_pa;
69 	ecc_t *ecc_p;
70 
71 	ecc_p = (ecc_t *)kmem_zalloc(sizeof (ecc_t), KM_SLEEP);
72 	ecc_p->ecc_pci_cmn_p = pci_p->pci_common_p;
73 	pci_p->pci_ecc_p = ecc_p;
74 
75 	ecc_p->ecc_ue.ecc_p = ecc_p;
76 	ecc_p->ecc_ue.ecc_type = CBNINTR_UE;
77 	ecc_p->ecc_ce.ecc_p = ecc_p;
78 	ecc_p->ecc_ce.ecc_type = CBNINTR_CE;
79 
80 	pci_ecc_setup(ecc_p);
81 
82 	/*
83 	 * Determine the virtual addresses of the streaming cache
84 	 * control/status and flush registers.
85 	 */
86 	ecc_p->ecc_csr_pa = cb_base_pa + COMMON_ECC_CSR_OFFSET;
87 	ecc_p->ecc_ue.ecc_afsr_pa = cb_base_pa + COMMON_UE_AFSR_OFFSET;
88 	ecc_p->ecc_ue.ecc_afar_pa = cb_base_pa + COMMON_UE_AFAR_OFFSET;
89 	ecc_p->ecc_ce.ecc_afsr_pa = cb_base_pa + COMMON_CE_AFSR_OFFSET;
90 	ecc_p->ecc_ce.ecc_afar_pa = cb_base_pa + COMMON_CE_AFAR_OFFSET;
91 
92 	DEBUG1(DBG_ATTACH, dip, "ecc_create: csr=%x\n", ecc_p->ecc_csr_pa);
93 	DEBUG2(DBG_ATTACH, dip, "ecc_create: ue_afsr=%x, ue_afar=%x\n",
94 		ecc_p->ecc_ue.ecc_afsr_pa, ecc_p->ecc_ue.ecc_afar_pa);
95 	DEBUG2(DBG_ATTACH, dip, "ecc_create: ce_afsr=%x, ce_afar=%x\n",
96 		ecc_p->ecc_ce.ecc_afsr_pa, ecc_p->ecc_ce.ecc_afar_pa);
97 
98 	ecc_configure(pci_p);
99 
100 	/*
101 	 * Register routines to be called from system error handling code.
102 	 */
103 	bus_func_register(BF_TYPE_ERRDIS, (busfunc_t)ecc_disable_nowait, ecc_p);
104 }
105 
106 int
107 ecc_register_intr(pci_t *pci_p)
108 {
109 	ecc_t *ecc_p = pci_p->pci_ecc_p;
110 	int ret;
111 
112 	/*
113 	 * Install the UE and CE error interrupt handlers.
114 	 */
115 	if ((ret = pci_ecc_add_intr(pci_p, CBNINTR_UE, &ecc_p->ecc_ue)) !=
116 	    DDI_SUCCESS)
117 		return (ret);
118 	if ((ret = pci_ecc_add_intr(pci_p, CBNINTR_CE, &ecc_p->ecc_ce)) !=
119 	    DDI_SUCCESS)
120 		return (ret);
121 
122 	return (DDI_SUCCESS);
123 }
124 
125 void
126 ecc_destroy(pci_t *pci_p)
127 {
128 	ecc_t *ecc_p = pci_p->pci_ecc_p;
129 
130 	DEBUG0(DBG_DETACH, pci_p->pci_dip, "ecc_destroy:\n");
131 
132 	/*
133 	 * Disable UE and CE ECC error interrupts.
134 	 */
135 	ecc_disable_wait(ecc_p);
136 
137 	/*
138 	 * Remove the ECC interrupt handlers.
139 	 */
140 	pci_ecc_rem_intr(pci_p, CBNINTR_UE, &ecc_p->ecc_ue);
141 	pci_ecc_rem_intr(pci_p, CBNINTR_CE, &ecc_p->ecc_ce);
142 
143 	/*
144 	 * Unregister our error handling functions.
145 	 */
146 	bus_func_unregister(BF_TYPE_ERRDIS,
147 	    (busfunc_t)ecc_disable_nowait, ecc_p);
148 	/*
149 	 * If a timer has been set, unset it.
150 	 */
151 	(void) untimeout(ecc_p->ecc_to_id);
152 
153 	kmem_free(ecc_p, sizeof (ecc_t));
154 	pci_p->pci_ecc_p = NULL;
155 }
156 
157 void
158 ecc_configure(pci_t *pci_p)
159 {
160 	ecc_t *ecc_p = pci_p->pci_ecc_p;
161 	dev_info_t *dip = pci_p->pci_dip;
162 	uint64_t l;
163 
164 	/*
165 	 * Clear any pending ECC errors.
166 	 */
167 	DEBUG0(DBG_ATTACH, dip, "ecc_configure: clearing UE and CE errors\n");
168 	l = (COMMON_ECC_UE_AFSR_E_MASK << COMMON_ECC_UE_AFSR_PE_SHIFT) |
169 		(COMMON_ECC_UE_AFSR_E_MASK << COMMON_ECC_UE_AFSR_SE_SHIFT);
170 	stdphysio(ecc_p->ecc_ue.ecc_afsr_pa, l);
171 
172 	l = (COMMON_ECC_CE_AFSR_E_MASK << COMMON_ECC_CE_AFSR_PE_SHIFT) |
173 		(COMMON_ECC_CE_AFSR_E_MASK << COMMON_ECC_CE_AFSR_SE_SHIFT);
174 	stdphysio(ecc_p->ecc_ce.ecc_afsr_pa, l);
175 
176 	/*
177 	 * Enable ECC error detections via the control register.
178 	 */
179 	DEBUG0(DBG_ATTACH, dip, "ecc_configure: enabling UE CE detection\n");
180 	l = COMMON_ECC_CTRL_ECC_EN;
181 	if (ecc_error_intr_enable)
182 		l |= COMMON_ECC_CTRL_UE_INTEN | COMMON_ECC_CTRL_CE_INTEN;
183 	stdphysio(ecc_p->ecc_csr_pa, l);
184 }
185 
186 void
187 ecc_enable_intr(pci_t *pci_p)
188 {
189 	cb_enable_nintr(pci_p, CBNINTR_UE);
190 	cb_enable_nintr(pci_p, CBNINTR_CE);
191 }
192 
193 void
194 ecc_disable_wait(ecc_t *ecc_p)
195 {
196 	ecc_disable(ecc_p, IB_INTR_WAIT);
197 }
198 
199 uint_t
200 ecc_disable_nowait(ecc_t *ecc_p)
201 {
202 	ecc_disable(ecc_p, IB_INTR_NOWAIT);
203 	return (BF_NONE);
204 }
205 
206 static void
207 ecc_disable(ecc_t *ecc_p, int wait)
208 {
209 	cb_t *cb_p = ecc_p->ecc_pci_cmn_p->pci_common_cb_p;
210 	uint64_t csr_pa = ecc_p->ecc_csr_pa;
211 	uint64_t csr = lddphysio(csr_pa);
212 
213 	csr &= ~(COMMON_ECC_CTRL_UE_INTEN | COMMON_ECC_CTRL_CE_INTEN);
214 	stdphysio(csr_pa, csr);
215 
216 	cb_disable_nintr(cb_p, CBNINTR_UE, wait);
217 	cb_disable_nintr(cb_p, CBNINTR_CE, wait);
218 }
219 
220 /*
221  * I/O ECC error handling:
222  *
223  * Below are the generic functions that handle PCI(pcisch, pcipsy) detected
224  * ECC errors.
225  *
226  * The registered interrupt handler for both pcisch and pcipsy is ecc_intr(),
227  * it's function is to receive the error, capture some state, and pass that on
228  * to the ecc_err_handler() for reporting purposes.
229  *
230  * ecc_err_handler() gathers more state(via ecc_errstate_get) and attempts
231  * to handle and report the error. ecc_err_handler() must determine if we need
232  * to panic due to this error (via pci_ecc_classify, which also decodes the
233  * ECC afsr), and if any side effects exist that may have caused or are due
234  * to this error. PBM errors related to the ECC error may exist, to report
235  * them we call pci_pbm_err_handler() and call ndi_fm_handler_dispatch() so
236  * that the child devices can log their pci errors.
237  *
238  * To report the error we must also get the syndrome and unum, which can not
239  * be done in high level interrupted context. Therefore we have an error
240  * queue(pci_ecc_queue) which we dispatch errors to, to report the errors
241  * (ecc_err_drain()).
242  *
243  * ecc_err_drain() will be called when either the softint is triggered
244  * or the system is panicing. Either way it will gather more information
245  * about the error from the CPU(via ecc_cpu_call(), ecc.c), attempt to
246  * retire the faulty page(if error is a UE), and report the detected error.
247  *
248  * ecc_delayed_ce() is called via timeout from ecc_err_handler() following
249  * the receipt of a CE interrupt.  It will be called after 6ms and check to
250  * see if any new CEs are present, if so we will log and another timeout will
251  * be set by(ecc_err_handler()).  If no CEs are present then it will re-enable
252  * CEs by clearing the previous interrupt.  This is to keep the system going
253  * in the event of a CE storm.
254  */
255 
256 /*
257  * Function used to get ECC AFSR register
258  */
259 static uint64_t
260 ecc_read_afsr(ecc_intr_info_t *ecc_ii_p)
261 {
262 	uint_t i;
263 	uint64_t afsr = 0ull;
264 
265 	ASSERT((ecc_ii_p->ecc_type == CBNINTR_UE) ||
266 	    (ecc_ii_p->ecc_type == CBNINTR_CE));
267 	if (!ecc_ii_p->ecc_errpndg_mask)
268 		return (lddphysio(ecc_ii_p->ecc_afsr_pa));
269 
270 	for (i = 0; i < pci_ecc_afsr_retries; i++) {
271 
272 		/*
273 		 * If we timeout, the logging routine will
274 		 * know because it will see the ERRPNDG bits
275 		 * set in the AFSR.
276 		 */
277 		afsr = lddphysio(ecc_ii_p->ecc_afsr_pa);
278 		if ((afsr & ecc_ii_p->ecc_errpndg_mask) == 0)
279 			break;
280 	}
281 	return (afsr);
282 }
283 
284 /*
285  * IO detected ECC error interrupt handler, calls ecc_err_handler to post
286  * error reports and handle the interrupt. Re-entry into ecc_err_handler
287  * is protected by the per-chip mutex pci_fm_mutex.
288  */
289 uint_t
290 ecc_intr(caddr_t a)
291 {
292 	ecc_intr_info_t *ecc_ii_p = (ecc_intr_info_t *)a;
293 	ecc_t *ecc_p = ecc_ii_p->ecc_p;
294 	pci_common_t *cmn_p = ecc_p->ecc_pci_cmn_p;
295 	ecc_errstate_t ecc_err;
296 	int ret = DDI_FM_OK;
297 
298 	bzero(&ecc_err, sizeof (ecc_errstate_t));
299 	ecc_err.ecc_ena = fm_ena_generate(0, FM_ENA_FMT1);
300 	ecc_err.ecc_ii_p = *ecc_ii_p;
301 	ecc_err.ecc_p = ecc_p;
302 	ecc_err.ecc_caller = PCI_ECC_CALL;
303 
304 	mutex_enter(&cmn_p->pci_fm_mutex);
305 	ret = ecc_err_handler(&ecc_err);
306 	mutex_exit(&cmn_p->pci_fm_mutex);
307 	if (ret == DDI_FM_FATAL) {
308 		/*
309 		 * Need delay here to allow CPUs to handle related traps,
310 		 * such as FRUs for USIIIi systems.
311 		 */
312 		DELAY(pci_ecc_panic_delay);
313 		fm_panic("Fatal PCI UE Error");
314 	}
315 
316 	return (DDI_INTR_CLAIMED);
317 }
318 
319 /*
320  * Function used to gather IO ECC error state.
321  */
322 static void
323 ecc_errstate_get(ecc_errstate_t *ecc_err_p)
324 {
325 	ecc_t *ecc_p;
326 	uint_t bus_id;
327 
328 	ASSERT(ecc_err_p);
329 
330 	ecc_p = ecc_err_p->ecc_ii_p.ecc_p;
331 	bus_id = ecc_p->ecc_pci_cmn_p->pci_common_id;
332 
333 	ASSERT(MUTEX_HELD(&ecc_p->ecc_pci_cmn_p->pci_fm_mutex));
334 	/*
335 	 * Read the fault registers.
336 	 */
337 	ecc_err_p->ecc_afsr = ecc_read_afsr(&ecc_err_p->ecc_ii_p);
338 	ecc_err_p->ecc_afar = lddphysio(ecc_err_p->ecc_ii_p.ecc_afar_pa);
339 
340 	ecc_err_p->ecc_offset = ((ecc_err_p->ecc_afsr &
341 				ecc_err_p->ecc_ii_p.ecc_offset_mask) >>
342 			ecc_err_p->ecc_ii_p.ecc_offset_shift) <<
343 		ecc_err_p->ecc_ii_p.ecc_size_log2;
344 
345 	ecc_err_p->ecc_aflt.flt_id = gethrtime();
346 	ecc_err_p->ecc_aflt.flt_stat = ecc_err_p->ecc_afsr;
347 	ecc_err_p->ecc_aflt.flt_addr = P2ALIGN(ecc_err_p->ecc_afar, 64) +
348 		ecc_err_p->ecc_offset;
349 	ecc_err_p->ecc_aflt.flt_bus_id = bus_id;
350 	ecc_err_p->ecc_aflt.flt_inst = CPU->cpu_id;
351 	ecc_err_p->ecc_aflt.flt_status = ECC_IOBUS;
352 	ecc_err_p->ecc_aflt.flt_in_memory = (pf_is_memory
353 			(ecc_err_p->ecc_afar >> MMU_PAGESHIFT))? 1: 0;
354 	ecc_err_p->ecc_aflt.flt_class = BUS_FAULT;
355 }
356 
357 /*
358  * ecc_pci_check: Called by ecc_err_handler() this function is responsible
359  * for calling pci_pbm_err_handler() for both sides of the schizo/psycho
360  * and calling their children error handlers(via ndi_fm_handler_dispatch()).
361  */
362 static int
363 ecc_pci_check(ecc_t *ecc_p, uint64_t fme_ena)
364 {
365 	ddi_fm_error_t derr;
366 	int i;
367 	int ret;
368 
369 	ASSERT(MUTEX_HELD(&ecc_p->ecc_pci_cmn_p->pci_fm_mutex));
370 
371 	bzero(&derr, sizeof (ddi_fm_error_t));
372 	derr.fme_version = DDI_FME_VERSION;
373 	derr.fme_ena = fme_ena;
374 	ret = DDI_FM_NONFATAL;
375 
376 	/*
377 	 * Need to report any PBM errors which may have caused or
378 	 * resulted from this error.
379 	 *
380 	 * Each psycho or schizo is represented by a pair of pci nodes
381 	 * in the device tree.
382 	 */
383 	for (i = 0; i < 2; i++) {
384 		dev_info_t *dip;
385 		pci_t *pci_p;
386 
387 		/* Make sure PBM PCI node exists */
388 		pci_p = ecc_p->ecc_pci_cmn_p->pci_p[i];
389 		if (pci_p == NULL)
390 			continue;
391 
392 		dip = pci_p->pci_dip;
393 		if (pci_pbm_err_handler(dip, &derr, (void *)pci_p,
394 		    PCI_ECC_CALL) == DDI_FM_FATAL)
395 			ret = DDI_FM_FATAL;
396 	}
397 	if (ret == DDI_FM_FATAL)
398 		return (DDI_FM_FATAL);
399 	else
400 		return (DDI_FM_NONFATAL);
401 }
402 
403 /*
404  * Function used to handle and log IO detected ECC errors, can be called by
405  * ecc_intr and pci_err_callback(trap callback). Protected by pci_fm_mutex.
406  */
407 int
408 ecc_err_handler(ecc_errstate_t *ecc_err_p)
409 {
410 	uint64_t pri_err, sec_err;
411 	ecc_intr_info_t *ecc_ii_p = &ecc_err_p->ecc_ii_p;
412 	ecc_t *ecc_p = ecc_ii_p->ecc_p;
413 	pci_t *pci_p;
414 	cb_t *cb_p;
415 	int fatal = 0;
416 	int nonfatal = 0;
417 
418 	ASSERT(MUTEX_HELD(&ecc_p->ecc_pci_cmn_p->pci_fm_mutex));
419 
420 	pci_p = ecc_p->ecc_pci_cmn_p->pci_p[0];
421 	if (pci_p == NULL)
422 		pci_p = ecc_p->ecc_pci_cmn_p->pci_p[1];
423 
424 	cb_p = ecc_p->ecc_pci_cmn_p->pci_common_cb_p;
425 
426 	ecc_errstate_get(ecc_err_p);
427 	pri_err = (ecc_err_p->ecc_afsr >> COMMON_ECC_UE_AFSR_PE_SHIFT) &
428 		COMMON_ECC_UE_AFSR_E_MASK;
429 
430 	sec_err = (ecc_err_p->ecc_afsr >> COMMON_ECC_UE_AFSR_SE_SHIFT) &
431 		COMMON_ECC_UE_AFSR_E_MASK;
432 
433 	switch (ecc_ii_p->ecc_type) {
434 	case CBNINTR_UE:
435 		if (pri_err) {
436 			ecc_err_p->ecc_aflt.flt_synd =
437 				pci_ecc_get_synd(ecc_err_p->ecc_afsr);
438 			ecc_err_p->ecc_pri = 1;
439 			pci_ecc_classify(pri_err, ecc_err_p);
440 			errorq_dispatch(pci_ecc_queue, (void *)ecc_err_p,
441 				sizeof (ecc_errstate_t),
442 				ecc_err_p->ecc_aflt.flt_panic);
443 		}
444 		if (sec_err) {
445 			ecc_errstate_t ecc_sec_err;
446 
447 			ecc_sec_err = *ecc_err_p;
448 			ecc_sec_err.ecc_pri = 0;
449 			pci_ecc_classify(sec_err, &ecc_sec_err);
450 			ecc_ereport_post(pci_p->pci_dip,
451 					&ecc_sec_err);
452 		}
453 		/*
454 		 * Check for PCI bus errors that may have resulted from or
455 		 * caused this UE.
456 		 */
457 		if (ecc_err_p->ecc_caller == PCI_ECC_CALL &&
458 		    ecc_pci_check(ecc_p, ecc_err_p->ecc_ena) == DDI_FM_FATAL)
459 			ecc_err_p->ecc_aflt.flt_panic = 1;
460 
461 		if (ecc_err_p->ecc_aflt.flt_panic &&
462 				ecc_err_p->ecc_aflt.flt_in_memory)
463 			panic_aflt = ecc_err_p->ecc_aflt;
464 
465 		if (ecc_err_p->ecc_aflt.flt_panic) {
466 			/*
467 			 * Disable all further errors since this will be
468 			 * treated as a fatal error.
469 			 */
470 			(void) ecc_disable_nowait(ecc_p);
471 			fatal++;
472 		}
473 		break;
474 
475 	case CBNINTR_CE:
476 		if (pri_err) {
477 			ecc_err_p->ecc_pri = 1;
478 			pci_ecc_classify(pri_err, ecc_err_p);
479 			ecc_err_p->ecc_aflt.flt_synd =
480 				pci_ecc_get_synd(ecc_err_p->ecc_afsr);
481 			ce_scrub(&ecc_err_p->ecc_aflt);
482 			errorq_dispatch(pci_ecc_queue, (void *)ecc_err_p,
483 					sizeof (ecc_errstate_t), ERRORQ_ASYNC);
484 			nonfatal++;
485 		}
486 		if (sec_err) {
487 			ecc_errstate_t ecc_sec_err;
488 
489 			ecc_sec_err = *ecc_err_p;
490 			ecc_sec_err.ecc_pri = 0;
491 			pci_ecc_classify(sec_err, &ecc_sec_err);
492 			ecc_ereport_post(pci_p->pci_dip, &ecc_sec_err);
493 			nonfatal++;
494 		}
495 		break;
496 
497 	default:
498 		return (DDI_FM_OK);
499 	}
500 	/* Clear the errors */
501 	stdphysio(ecc_ii_p->ecc_afsr_pa, ecc_err_p->ecc_afsr);
502 	/*
503 	 * Clear the interrupt if called by ecc_intr and UE error or if called
504 	 * by ecc_intr and CE error and delayed CE interrupt handling is
505 	 * turned off.
506 	 */
507 	if ((ecc_err_p->ecc_caller == PCI_ECC_CALL &&
508 	    ecc_ii_p->ecc_type == CBNINTR_UE && !fatal) ||
509 	    (ecc_err_p->ecc_caller == PCI_ECC_CALL &&
510 	    ecc_ii_p->ecc_type == CBNINTR_CE && !ecc_ce_delayed))
511 		cb_clear_nintr(cb_p, ecc_ii_p->ecc_type);
512 	if (!fatal && !nonfatal)
513 		return (DDI_FM_OK);
514 	else if (fatal)
515 		return (DDI_FM_FATAL);
516 	return (DDI_FM_NONFATAL);
517 }
518 
519 /*
520  * Called from ecc_err_drain below for CBINTR_CE case.
521  */
522 static int
523 ecc_err_cexdiag(page_t *pp, ecc_errstate_t *ecc_err,
524 		errorq_elem_t *eqep)
525 {
526 	struct async_flt *ecc = &ecc_err->ecc_aflt;
527 
528 	if (!pp) {
529 		CE_XDIAG_SETSKIPCODE(ecc->flt_disp, CE_XDIAG_SKIP_NOPP);
530 		return (0);
531 	} else if (page_isretired(pp) || page_deteriorating(pp)) {
532 		CE_XDIAG_SETSKIPCODE(ecc->flt_disp, CE_XDIAG_SKIP_PAGEDET);
533 		return (0);
534 	}
535 
536 	return (ce_scrub_xdiag_recirc(ecc, pci_ecc_queue, eqep,
537 	    offsetof(ecc_errstate_t, ecc_aflt)));
538 }
539 
540 /*
541  * Function used to drain pci_ecc_queue, either during panic or after softint
542  * is generated, to log IO detected ECC errors.
543  */
544 /*ARGSUSED*/
545 void
546 ecc_err_drain(void *not_used, ecc_errstate_t *ecc_err, errorq_elem_t *eqep)
547 {
548 	struct async_flt *ecc = &ecc_err->ecc_aflt;
549 	pci_t *pci_p = ecc_err->ecc_p->ecc_pci_cmn_p->pci_p[0];
550 	page_t *pp;
551 	int ecc_type = ecc_err->ecc_ii_p.ecc_type;
552 
553 	if (pci_p == NULL)
554 		pci_p = ecc_err->ecc_p->ecc_pci_cmn_p->pci_p[1];
555 
556 	if (ecc->flt_class == RECIRC_BUS_FAULT) {
557 		/*
558 		 * Perform any additional actions that occur after the
559 		 * ecc_err_cexdiag below and post the ereport.
560 		 */
561 		ecc->flt_class = BUS_FAULT;
562 		ecc_err->ecc_err_type = flt_to_error_type(ecc);
563 		ecc_ereport_post(pci_p->pci_dip, ecc_err);
564 		return;
565 	}
566 
567 	ecc_cpu_call(ecc, ecc_err->ecc_unum, (ecc_type == CBNINTR_UE) ?
568 			ECC_IO_UE : ECC_IO_CE);
569 
570 	pp = page_numtopp_nolock(ecc->flt_addr >> MMU_PAGESHIFT);
571 
572 	switch (ecc_type) {
573 	case CBNINTR_UE:
574 		if (pp && ecc_err->ecc_pg_ret == 1) {
575 			page_settoxic(pp, PAGE_IS_FAULTY);
576 			(void) page_retire(pp, PAGE_IS_TOXIC);
577 		}
578 		ecc_err->ecc_err_type = flt_to_error_type(ecc);
579 		break;
580 
581 	case CBNINTR_CE:
582 		/*
583 		 * Setup timeout (if CE detected via interrupt) to
584 		 * re-enable CE interrupts if no more CEs are detected.
585 		 * This is to protect against CE storms.
586 		 */
587 		if (ecc_ce_delayed &&
588 		    ecc_err->ecc_caller == PCI_ECC_CALL &&
589 		    ecc_err->ecc_p->ecc_to_id == 0) {
590 			ecc_err->ecc_p->ecc_to_id = timeout(ecc_delayed_ce,
591 			    (void *)ecc_err->ecc_p,
592 			    drv_usectohz((clock_t)ecc_ce_delay_secs *
593 			    MICROSEC));
594 		}
595 
596 		/* ecc_err_cexdiag returns nonzero to recirculate */
597 		if (CE_XDIAG_EXT_ALG_APPLIED(ecc->flt_disp) &&
598 		    ecc_err_cexdiag(pp, ecc_err, eqep))
599 			return;
600 		ecc_err->ecc_err_type = flt_to_error_type(ecc);
601 		break;
602 	}
603 
604 	ecc_ereport_post(pci_p->pci_dip, ecc_err);
605 }
606 
607 static void
608 ecc_delayed_ce(void *arg)
609 {
610 	ecc_t *ecc_p = (ecc_t *)arg;
611 	pci_common_t *cmn_p;
612 	cb_t *cb_p;
613 
614 	ASSERT(ecc_p);
615 
616 	cmn_p = ecc_p->ecc_pci_cmn_p;
617 	cb_p = cmn_p->pci_common_cb_p;
618 	/*
619 	 * If no more CE errors are found then enable interrupts(by
620 	 * clearing the previous interrupt), else send in for logging
621 	 * and the timeout should be set again.
622 	 */
623 	ecc_p->ecc_to_id = 0;
624 	if (!((ecc_read_afsr(&ecc_p->ecc_ce) >>
625 	    COMMON_ECC_UE_AFSR_PE_SHIFT) & COMMON_ECC_UE_AFSR_E_MASK)) {
626 		cb_clear_nintr(cb_p, ecc_p->ecc_ce.ecc_type);
627 	} else {
628 		ecc_errstate_t ecc_err;
629 
630 		bzero(&ecc_err, sizeof (ecc_errstate_t));
631 		ecc_err.ecc_ena = fm_ena_generate(0, FM_ENA_FMT1);
632 		ecc_err.ecc_ii_p = ecc_p->ecc_ce;
633 		ecc_err.ecc_p = ecc_p;
634 		ecc_err.ecc_caller = PCI_ECC_CALL;
635 
636 		mutex_enter(&cmn_p->pci_fm_mutex);
637 		(void) ecc_err_handler(&ecc_err);
638 		mutex_exit(&cmn_p->pci_fm_mutex);
639 	}
640 }
641 
642 /*
643  * Function used to post IO detected ECC ereports.
644  */
645 static void
646 ecc_ereport_post(dev_info_t *dip, ecc_errstate_t *ecc_err)
647 {
648 	char buf[FM_MAX_CLASS], dev_path[MAXPATHLEN], *ptr;
649 	struct i_ddi_fmhdl *fmhdl = DEVI(dip)->devi_fmhdl;
650 	nvlist_t *ereport, *detector;
651 	nv_alloc_t *nva;
652 	errorq_elem_t *eqep;
653 
654 	/*
655 	 * We do not use ddi_fm_ereport_post because we need to set a
656 	 * special detector here. Since we do not have a device path for
657 	 * the bridge chip we use what we think it should be to aid in
658 	 * diagnosis. This path fmri is created by pci_fmri_create()
659 	 * during initialization.
660 	 */
661 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s.%s", DDI_IO_CLASS,
662 	    ecc_err->ecc_bridge_type, ecc_err->ecc_aflt.flt_erpt_class);
663 
664 	ecc_err->ecc_ena = ecc_err->ecc_ena ? ecc_err->ecc_ena :
665 		fm_ena_generate(0, FM_ENA_FMT1);
666 
667 	eqep = errorq_reserve(fmhdl->fh_errorq);
668 	if (eqep == NULL)
669 		return;
670 
671 	ereport = errorq_elem_nvl(fmhdl->fh_errorq, eqep);
672 	nva = errorq_elem_nva(fmhdl->fh_errorq, eqep);
673 	detector = fm_nvlist_create(nva);
674 
675 	ASSERT(ereport);
676 	ASSERT(nva);
677 	ASSERT(detector);
678 
679 	ddi_pathname(dip, dev_path);
680 	ptr = strrchr(dev_path, (int)',');
681 
682 	if (ptr)
683 		*ptr = '\0';
684 
685 	fm_fmri_dev_set(detector, FM_DEV_SCHEME_VERSION, NULL, dev_path, NULL);
686 
687 	if (ecc_err->ecc_pri) {
688 		if ((ecc_err->ecc_fmri = fm_nvlist_create(nva)) != NULL)
689 			fm_fmri_mem_set(ecc_err->ecc_fmri,
690 			    FM_MEM_SCHEME_VERSION, NULL, ecc_err->ecc_unum,
691 			    NULL);
692 
693 		fm_ereport_set(ereport, FM_EREPORT_VERSION, buf,
694 		    ecc_err->ecc_ena, detector,
695 		    PCI_ECC_AFSR, DATA_TYPE_UINT64, ecc_err->ecc_afsr,
696 		    PCI_ECC_AFAR, DATA_TYPE_UINT64, ecc_err->ecc_aflt.flt_addr,
697 		    PCI_ECC_CTRL, DATA_TYPE_UINT64, ecc_err->ecc_ctrl,
698 		    PCI_ECC_SYND, DATA_TYPE_UINT16, ecc_err->ecc_aflt.flt_synd,
699 		    PCI_ECC_TYPE, DATA_TYPE_STRING, ecc_err->ecc_err_type,
700 		    PCI_ECC_DISP, DATA_TYPE_UINT64, ecc_err->ecc_aflt.flt_disp,
701 		    PCI_ECC_RESOURCE, DATA_TYPE_NVLIST, ecc_err->ecc_fmri,
702 		    NULL);
703 	} else {
704 		fm_ereport_set(ereport, FM_EREPORT_VERSION, buf,
705 		    ecc_err->ecc_ena, detector,
706 		    PCI_ECC_AFSR, DATA_TYPE_UINT64, ecc_err->ecc_afsr,
707 		    PCI_ECC_CTRL, DATA_TYPE_UINT64, ecc_err->ecc_ctrl,
708 		    NULL);
709 	}
710 	errorq_commit(fmhdl->fh_errorq, eqep, ERRORQ_ASYNC);
711 }
712