xref: /illumos-gate/usr/src/uts/sun4u/io/pci/pci_fm.c (revision 00d0963faf2e861a4aef6b1bf28f99a5b2b20755)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/sunddi.h>
30 #include <sys/sunndi.h>
31 #include <sys/ddi_impldefs.h>
32 #include <sys/async.h>
33 #include <sys/membar.h>
34 #include <sys/spl.h>
35 #include <sys/iommu.h>
36 #include <sys/pci/pci_obj.h>
37 #include <sys/fm/util.h>
38 #include <sys/fm/io/pci.h>
39 #include <sys/fm/io/ddi.h>
40 #include <sys/fm/io/sun4upci.h>
41 #include <sys/fm/protocol.h>
42 #include <sys/intr.h>
43 
44 /*LINTLIBRARY*/
45 
46 /*
47  * The routines below are generic sun4u PCI interfaces to support
48  * Fault Management.
49  *
50  * pci_dma_check, pci_acc_check, pci_handle_lookup are functions used
51  * to associate a captured PCI address to a particular dma/acc handle.
52  *
53  * pci_fm_acc_setup, pci_fm_init_child, pci_fm_create,
54  * pci_fm_destroy are constructors/destructors used to setup and teardown
55  * necessary resources.
56  *
57  * pci_bus_enter, pci_bus_exit are registered via busops and are used to
58  * provide exclusive access to the PCI bus.
59  *
60  * pci_err_callback is the registered callback for PCI which is called
61  * by the CPU code when it detects a UE/TO/BERR.
62  *
63  * pbm_ereport_post is used by the PBM code to generically report all
64  * PBM errors.
65  *
66  */
67 
68 /*
69  * Function used to setup access functions depending on level of desired
70  * protection.
71  */
72 void
73 pci_fm_acc_setup(ddi_map_req_t *mp, dev_info_t *rdip)
74 {
75 	uchar_t fflag;
76 	ddi_acc_hdl_t *hp;
77 	ddi_acc_impl_t *ap;
78 
79 	hp = mp->map_handlep;
80 	ap = (ddi_acc_impl_t *)hp->ah_platform_private;
81 	fflag = ap->ahi_common.ah_acc.devacc_attr_access;
82 
83 	if (mp->map_op == DDI_MO_MAP_LOCKED) {
84 		ndi_fmc_insert(rdip, ACC_HANDLE, (void *)hp, NULL);
85 		switch (fflag) {
86 		case DDI_FLAGERR_ACC:
87 			ap->ahi_get8 = i_ddi_prot_get8;
88 			ap->ahi_get16 = i_ddi_prot_get16;
89 			ap->ahi_get32 = i_ddi_prot_get32;
90 			ap->ahi_get64 = i_ddi_prot_get64;
91 			ap->ahi_put8 = i_ddi_prot_put8;
92 			ap->ahi_put16 = i_ddi_prot_put16;
93 			ap->ahi_put32 = i_ddi_prot_put32;
94 			ap->ahi_put64 = i_ddi_prot_put64;
95 			ap->ahi_rep_get8 = i_ddi_prot_rep_get8;
96 			ap->ahi_rep_get16 = i_ddi_prot_rep_get16;
97 			ap->ahi_rep_get32 = i_ddi_prot_rep_get32;
98 			ap->ahi_rep_get64 = i_ddi_prot_rep_get64;
99 			ap->ahi_rep_put8 = i_ddi_prot_rep_put8;
100 			ap->ahi_rep_put16 = i_ddi_prot_rep_put16;
101 			ap->ahi_rep_put32 = i_ddi_prot_rep_put32;
102 			ap->ahi_rep_put64 = i_ddi_prot_rep_put64;
103 			break;
104 		case DDI_CAUTIOUS_ACC :
105 			ap->ahi_get8 = i_ddi_caut_get8;
106 			ap->ahi_get16 = i_ddi_caut_get16;
107 			ap->ahi_get32 = i_ddi_caut_get32;
108 			ap->ahi_get64 = i_ddi_caut_get64;
109 			ap->ahi_put8 = i_ddi_caut_put8;
110 			ap->ahi_put16 = i_ddi_caut_put16;
111 			ap->ahi_put32 = i_ddi_caut_put32;
112 			ap->ahi_put64 = i_ddi_caut_put64;
113 			ap->ahi_rep_get8 = i_ddi_caut_rep_get8;
114 			ap->ahi_rep_get16 = i_ddi_caut_rep_get16;
115 			ap->ahi_rep_get32 = i_ddi_caut_rep_get32;
116 			ap->ahi_rep_get64 = i_ddi_caut_rep_get64;
117 			ap->ahi_rep_put8 = i_ddi_caut_rep_put8;
118 			ap->ahi_rep_put16 = i_ddi_caut_rep_put16;
119 			ap->ahi_rep_put32 = i_ddi_caut_rep_put32;
120 			ap->ahi_rep_put64 = i_ddi_caut_rep_put64;
121 			break;
122 		default:
123 			break;
124 		}
125 	} else if (mp->map_op == DDI_MO_UNMAP) {
126 		ndi_fmc_remove(rdip, ACC_HANDLE, (void *)hp);
127 	}
128 }
129 
130 /*
131  * Function used to initialize FMA for our children nodes. Called
132  * through pci busops when child node calls ddi_fm_init.
133  */
134 /* ARGSUSED */
135 int
136 pci_fm_init_child(dev_info_t *dip, dev_info_t *tdip, int cap,
137     ddi_iblock_cookie_t *ibc)
138 {
139 	pci_t *pci_p = get_pci_soft_state(ddi_get_instance(dip));
140 
141 	ASSERT(ibc != NULL);
142 	*ibc = pci_p->pci_fm_ibc;
143 
144 	return (pci_p->pci_fm_cap);
145 }
146 
147 /*
148  * Lock accesses to the pci bus, to be able to protect against bus errors.
149  */
150 void
151 pci_bus_enter(dev_info_t *dip, ddi_acc_handle_t handle)
152 {
153 	pci_t *pci_p = get_pci_soft_state(ddi_get_instance(dip));
154 	pbm_t *pbm_p = pci_p->pci_pbm_p;
155 
156 	membar_sync();
157 
158 	mutex_enter(&pbm_p->pbm_pokefault_mutex);
159 	pbm_p->pbm_excl_handle = handle;
160 }
161 
162 /*
163  * Unlock access to bus and clear errors before exiting.
164  */
165 /* ARGSUSED */
166 void
167 pci_bus_exit(dev_info_t *dip, ddi_acc_handle_t handle)
168 {
169 	pci_t *pci_p = get_pci_soft_state(ddi_get_instance(dip));
170 	pbm_t *pbm_p = pci_p->pci_pbm_p;
171 	ddi_fm_error_t derr;
172 
173 	ASSERT(MUTEX_HELD(&pbm_p->pbm_pokefault_mutex));
174 
175 	membar_sync();
176 
177 	mutex_enter(&pci_p->pci_common_p->pci_fm_mutex);
178 	ddi_fm_acc_err_get(pbm_p->pbm_excl_handle, &derr, DDI_FME_VERSION);
179 
180 	if (derr.fme_status == DDI_FM_OK) {
181 		if (pci_check_error(pci_p) != 0) {
182 			(void) pci_pbm_err_handler(pci_p->pci_dip, &derr,
183 					(const void *)pci_p, PCI_BUS_EXIT_CALL);
184 		}
185 	}
186 	mutex_exit(&pci_p->pci_common_p->pci_fm_mutex);
187 
188 	pbm_p->pbm_excl_handle = NULL;
189 	mutex_exit(&pbm_p->pbm_pokefault_mutex);
190 }
191 
192 /*
193  * PCI error callback which is registered with our parent to call
194  * for PCI logging when the CPU traps due to BERR/TO/UE.
195  */
196 int
197 pci_err_callback(dev_info_t *dip, ddi_fm_error_t *derr,
198     const void *impl_data)
199 {
200 	pci_t *pci_p = (pci_t *)impl_data;
201 	pci_common_t *cmn_p = pci_p->pci_common_p;
202 	ecc_t *ecc_p = cmn_p->pci_common_ecc_p;
203 	ecc_errstate_t ecc_err;
204 	int fatal = 0;
205 	int nonfatal = 0;
206 	int unknown = 0;
207 	int ret = DDI_FM_OK;
208 
209 	bzero(&ecc_err, sizeof (ecc_err));
210 	mutex_enter(&cmn_p->pci_fm_mutex);
211 	/*
212 	 * Check and log ecc and pbm errors
213 	 */
214 	ecc_err.ecc_ii_p = ecc_p->ecc_ue;
215 	ecc_err.ecc_ena = derr->fme_ena;
216 	ecc_err.ecc_caller = PCI_TRAP_CALL;
217 
218 	if ((ret = ecc_err_handler(&ecc_err)) == DDI_FM_FATAL)
219 		fatal++;
220 	else if (ret == DDI_FM_NONFATAL)
221 		nonfatal++;
222 	else if (ret == DDI_FM_UNKNOWN)
223 		unknown++;
224 
225 	if (pci_check_error(pci_p) != 0) {
226 		int err = pci_pbm_err_handler(pci_p->pci_dip, derr,
227 				(const void *)pci_p, PCI_TRAP_CALL);
228 		if (err == DDI_FM_FATAL)
229 			fatal++;
230 		else if (err == DDI_FM_NONFATAL)
231 			nonfatal++;
232 		else if (err == DDI_FM_UNKNOWN)
233 			unknown++;
234 	}
235 
236 	mutex_exit(&cmn_p->pci_fm_mutex);
237 
238 	if (fatal)
239 		return (DDI_FM_FATAL);
240 	else if (nonfatal)
241 		return (DDI_FM_NONFATAL);
242 	else if (unknown)
243 		return (DDI_FM_UNKNOWN);
244 	else
245 		return (DDI_FM_OK);
246 }
247 
248 void
249 pci_fm_create(pci_t *pci_p)
250 {
251 	pci_common_t *cmn_p = pci_p->pci_common_p;
252 
253 	/*
254 	 * PCI detected ECC errorq, to schedule async handling
255 	 * of ECC errors and logging.
256 	 * The errorq is created here but destroyed when _fini is called
257 	 * for the pci module.
258 	 */
259 	if (pci_ecc_queue == NULL) {
260 		pci_ecc_queue = errorq_create("pci_ecc_queue",
261 				(errorq_func_t)ecc_err_drain,
262 				(void *)pci_p->pci_ecc_p,
263 				ECC_MAX_ERRS, sizeof (ecc_errstate_t),
264 				PIL_2, ERRORQ_VITAL);
265 		if (pci_ecc_queue == NULL)
266 			panic("failed to create required system error queue");
267 	}
268 
269 	/*
270 	 * Initialize pci_target_queue for FMA handling of pci errors.
271 	 */
272 	pci_targetq_init();
273 
274 	/*
275 	 * Initialize FMA support
276 	 * The axq workaround prevents fault management of access errors
277 	 */
278 	if (pci_p->pci_pbm_p->pbm_pio_limit == 0)
279 		pci_p->pci_fm_cap = DDI_FM_EREPORT_CAPABLE |
280 			DDI_FM_ACCCHK_CAPABLE | DDI_FM_DMACHK_CAPABLE |
281 			DDI_FM_ERRCB_CAPABLE;
282 	else
283 		pci_p->pci_fm_cap = DDI_FM_EREPORT_CAPABLE |
284 			DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE;
285 	/*
286 	 * Call parent to get it's capablity
287 	 */
288 	ddi_fm_init(pci_p->pci_dip, &pci_p->pci_fm_cap,
289 			&pci_p->pci_fm_ibc);
290 	/*
291 	 * Need to be ereport and error handler cabable
292 	 */
293 	ASSERT((pci_p->pci_fm_cap & DDI_FM_ERRCB_CAPABLE) &&
294 	    (pci_p->pci_fm_cap & DDI_FM_EREPORT_CAPABLE));
295 	/*
296 	 * Initialize error handling mutex.
297 	 */
298 	if (cmn_p->pci_common_refcnt == 0) {
299 		mutex_init(&cmn_p->pci_fm_mutex, NULL, MUTEX_DRIVER,
300 				(void *)pci_p->pci_fm_ibc);
301 	}
302 
303 	/*
304 	 * Register error callback with our parent.
305 	 */
306 	ddi_fm_handler_register(pci_p->pci_dip, pci_err_callback,
307 			pci_p);
308 
309 }
310 
311 void
312 pci_fm_destroy(pci_t *pci_p)
313 {
314 	pci_common_t *cmn_p = pci_p->pci_common_p;
315 
316 	/* schizo non-shared objects */
317 	ddi_fm_handler_unregister(pci_p->pci_dip);
318 	ddi_fm_fini(pci_p->pci_dip);
319 
320 	if (cmn_p->pci_common_refcnt != 0)
321 		return;
322 
323 	mutex_destroy(&cmn_p->pci_fm_mutex);
324 }
325 
326 /*
327  * Function used to post PCI block module specific ereports.
328  */
329 void
330 pbm_ereport_post(dev_info_t *dip, uint64_t ena, pbm_errstate_t *pbm_err)
331 {
332 	char buf[FM_MAX_CLASS];
333 
334 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s",
335 	    pbm_err->pbm_bridge_type, pbm_err->pbm_err_class);
336 
337 	ena = ena ? ena : fm_ena_generate(0, FM_ENA_FMT1);
338 
339 	ddi_fm_ereport_post(dip, buf, ena, DDI_NOSLEEP,
340 	    FM_VERSION, DATA_TYPE_UINT8, 0,
341 	    PCI_CONFIG_STATUS, DATA_TYPE_UINT16, pbm_err->pbm_pci.pci_cfg_stat,
342 	    PCI_CONFIG_COMMAND, DATA_TYPE_UINT16, pbm_err->pbm_pci.pci_cfg_comm,
343 	    PCI_PBM_CSR, DATA_TYPE_UINT64, pbm_err->pbm_ctl_stat,
344 	    PCI_PBM_AFSR, DATA_TYPE_UINT64, pbm_err->pbm_afsr,
345 	    PCI_PBM_AFAR, DATA_TYPE_UINT64, pbm_err->pbm_afar,
346 	    PCI_PBM_SLOT, DATA_TYPE_UINT64, pbm_err->pbm_err_sl,
347 	    PCI_PBM_VALOG, DATA_TYPE_UINT64, pbm_err->pbm_va_log,
348 	    NULL);
349 }
350