xref: /titanic_41/usr/src/uts/sun4/os/ddi_impl.c (revision b0fc0e77220f1fa4c933fd58a4e1dedcd650b0f1)
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 /*
23  * Copyright 2006 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  * sun4 specific DDI implementation
31  */
32 #include <sys/cpuvar.h>
33 #include <sys/ddi_subrdefs.h>
34 #include <sys/machsystm.h>
35 #include <sys/sunndi.h>
36 #include <sys/sysmacros.h>
37 #include <sys/ontrap.h>
38 #include <vm/seg_kmem.h>
39 #include <sys/membar.h>
40 #include <sys/dditypes.h>
41 #include <sys/ndifm.h>
42 #include <sys/fm/io/ddi.h>
43 #include <sys/ivintr.h>
44 #include <sys/bootconf.h>
45 #include <sys/conf.h>
46 #include <sys/ethernet.h>
47 #include <sys/idprom.h>
48 #include <sys/promif.h>
49 #include <sys/prom_plat.h>
50 #include <sys/systeminfo.h>
51 #include <sys/fpu/fpusystm.h>
52 #include <sys/vm.h>
53 #include <sys/fs/dv_node.h>
54 #include <sys/fs/snode.h>
55 #include <sys/ddi_isa.h>
56 #include <sys/modhash.h>
57 
58 dev_info_t *get_intr_parent(dev_info_t *, dev_info_t *,
59     ddi_intr_handle_impl_t *);
60 #pragma weak get_intr_parent
61 
62 int process_intr_ops(dev_info_t *, dev_info_t *, ddi_intr_op_t,
63     ddi_intr_handle_impl_t *, void *);
64 #pragma weak process_intr_ops
65 
66 void cells_1275_copy(prop_1275_cell_t *, prop_1275_cell_t *, int32_t);
67     prop_1275_cell_t *cells_1275_cmp(prop_1275_cell_t *, prop_1275_cell_t *,
68     int32_t len);
69 #pragma weak cells_1275_copy
70 
71 /*
72  * Wrapper for ddi_prop_lookup_int_array().
73  * This is handy because it returns the prop length in
74  * bytes which is what most of the callers require.
75  */
76 
77 static int
78 get_prop_int_array(dev_info_t *di, char *pname, int **pval, uint_t *plen)
79 {
80 	int ret;
81 
82 	if ((ret = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, di,
83 	    DDI_PROP_DONTPASS, pname, pval, plen)) == DDI_PROP_SUCCESS) {
84 		*plen = (*plen) * (uint_t)sizeof (int);
85 	}
86 	return (ret);
87 }
88 
89 /*
90  * SECTION: DDI Node Configuration
91  */
92 
93 /*
94  * init_regspec_64:
95  *
96  * If the parent #size-cells is 2, convert the upa-style or
97  * safari-style reg property from 2-size cells to 1 size cell
98  * format, ignoring the size_hi, which must be zero for devices.
99  * (It won't be zero in the memory list properties in the memory
100  * nodes, but that doesn't matter here.)
101  */
102 struct ddi_parent_private_data *
103 init_regspec_64(dev_info_t *dip)
104 {
105 	struct ddi_parent_private_data *pd;
106 	dev_info_t *parent;
107 	int size_cells;
108 
109 	/*
110 	 * If there are no "reg"s in the child node, return.
111 	 */
112 	pd = ddi_get_parent_data(dip);
113 	if ((pd == NULL) || (pd->par_nreg == 0)) {
114 		return (pd);
115 	}
116 	parent = ddi_get_parent(dip);
117 
118 	size_cells = ddi_prop_get_int(DDI_DEV_T_ANY, parent,
119 	    DDI_PROP_DONTPASS, "#size-cells", 1);
120 
121 	if (size_cells != 1)  {
122 
123 		int n, j;
124 		struct regspec *irp;
125 		struct reg_64 {
126 			uint_t addr_hi, addr_lo, size_hi, size_lo;
127 		};
128 		struct reg_64 *r64_rp;
129 		struct regspec *rp;
130 		uint_t len = 0;
131 		int *reg_prop;
132 
133 		ASSERT(size_cells == 2);
134 
135 		/*
136 		 * We already looked the property up once before if
137 		 * pd is non-NULL.
138 		 */
139 		(void) ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip,
140 		    DDI_PROP_DONTPASS, OBP_REG, &reg_prop, &len);
141 		ASSERT(len != 0);
142 
143 		n = sizeof (struct reg_64) / sizeof (int);
144 		n = len / n;
145 
146 		/*
147 		 * We're allocating a buffer the size of the PROM's property,
148 		 * but we're only using a smaller portion when we assign it
149 		 * to a regspec.  We do this so that in the
150 		 * impl_ddi_sunbus_removechild function, we will
151 		 * always free the right amount of memory.
152 		 */
153 		irp = rp = (struct regspec *)reg_prop;
154 		r64_rp = (struct reg_64 *)pd->par_reg;
155 
156 		for (j = 0; j < n; ++j, ++rp, ++r64_rp) {
157 			ASSERT(r64_rp->size_hi == 0);
158 			rp->regspec_bustype = r64_rp->addr_hi;
159 			rp->regspec_addr = r64_rp->addr_lo;
160 			rp->regspec_size = r64_rp->size_lo;
161 		}
162 
163 		ddi_prop_free((void *)pd->par_reg);
164 		pd->par_nreg = n;
165 		pd->par_reg = irp;
166 	}
167 	return (pd);
168 }
169 
170 /*
171  * Create a ddi_parent_private_data structure from the ddi properties of
172  * the dev_info node.
173  *
174  * The "reg" is required if the driver wishes to create mappings on behalf
175  * of the device. The "reg" property is assumed to be a list of at least
176  * one triplet
177  *
178  *	<bustype, address, size>*1
179  *
180  * The "interrupt" property is no longer part of parent private data on
181  * sun4u. The interrupt parent is may not be the device tree parent.
182  *
183  * The "ranges" property describes the mapping of child addresses to parent
184  * addresses.
185  *
186  * N.B. struct rangespec is defined for the following default values:
187  *			parent  child
188  *	#address-cells	2	2
189  *	#size-cells	1	1
190  * This function doesn't deal with non-default cells and will not create
191  * ranges in such cases.
192  */
193 void
194 make_ddi_ppd(dev_info_t *child, struct ddi_parent_private_data **ppd)
195 {
196 	struct ddi_parent_private_data *pdptr;
197 	int *reg_prop, *rng_prop;
198 	uint_t reg_len = 0, rng_len = 0;
199 	dev_info_t *parent;
200 	int parent_addr_cells, parent_size_cells;
201 	int child_addr_cells, child_size_cells;
202 
203 	*ppd = pdptr = kmem_zalloc(sizeof (*pdptr), KM_SLEEP);
204 
205 	/*
206 	 * root node has no parent private data, so *ppd should
207 	 * be initialized for naming to work properly.
208 	 */
209 	if ((parent = ddi_get_parent(child)) == NULL)
210 		return;
211 
212 	/*
213 	 * Set reg field of parent data from "reg" property
214 	 */
215 	if ((get_prop_int_array(child, OBP_REG, &reg_prop, &reg_len)
216 	    == DDI_PROP_SUCCESS) && (reg_len != 0)) {
217 		pdptr->par_nreg = (int)(reg_len / sizeof (struct regspec));
218 		pdptr->par_reg = (struct regspec *)reg_prop;
219 	}
220 
221 	/*
222 	 * "ranges" property ...
223 	 *
224 	 * This function does not handle cases where #address-cells != 2
225 	 * and * min(parent, child) #size-cells != 1 (see bugid 4211124).
226 	 *
227 	 * Nexus drivers with such exceptions (e.g. pci ranges)
228 	 * should either create a separate function for handling
229 	 * ranges or not use parent private data to store ranges.
230 	 */
231 
232 	/* root node has no ranges */
233 	if ((parent = ddi_get_parent(child)) == NULL)
234 		return;
235 
236 	child_addr_cells = ddi_prop_get_int(DDI_DEV_T_ANY, child,
237 	    DDI_PROP_DONTPASS, "#address-cells", 2);
238 	child_size_cells = ddi_prop_get_int(DDI_DEV_T_ANY, child,
239 	    DDI_PROP_DONTPASS, "#size-cells", 1);
240 	parent_addr_cells = ddi_prop_get_int(DDI_DEV_T_ANY, parent,
241 	    DDI_PROP_DONTPASS, "#address-cells", 2);
242 	parent_size_cells = ddi_prop_get_int(DDI_DEV_T_ANY, parent,
243 	    DDI_PROP_DONTPASS, "#size-cells", 1);
244 	if (child_addr_cells != 2 || parent_addr_cells != 2 ||
245 	    (child_size_cells != 1 && parent_size_cells != 1)) {
246 		NDI_CONFIG_DEBUG((CE_NOTE, "!ranges not made in parent data; "
247 		    "#address-cells or #size-cells have non-default value"));
248 		return;
249 	}
250 
251 	if (get_prop_int_array(child, OBP_RANGES, &rng_prop, &rng_len)
252 	    == DDI_PROP_SUCCESS) {
253 		pdptr->par_nrng = rng_len / (int)(sizeof (struct rangespec));
254 		pdptr->par_rng = (struct rangespec *)rng_prop;
255 	}
256 }
257 
258 /*
259  * Free ddi_parent_private_data structure
260  */
261 void
262 impl_free_ddi_ppd(dev_info_t *dip)
263 {
264 	struct ddi_parent_private_data *pdptr = ddi_get_parent_data(dip);
265 
266 	if (pdptr == NULL)
267 		return;
268 
269 	if (pdptr->par_nrng != 0)
270 		ddi_prop_free((void *)pdptr->par_rng);
271 
272 	if (pdptr->par_nreg != 0)
273 		ddi_prop_free((void *)pdptr->par_reg);
274 
275 	kmem_free(pdptr, sizeof (*pdptr));
276 	ddi_set_parent_data(dip, NULL);
277 }
278 
279 /*
280  * Name a child of sun busses based on the reg spec.
281  * Handles the following properties:
282  *
283  *	Property	value
284  *	Name		type
285  *
286  *	reg		register spec
287  *	interrupts	new (bus-oriented) interrupt spec
288  *	ranges		range spec
289  *
290  * This may be called multiple times, independent of
291  * initchild calls.
292  */
293 static int
294 impl_sunbus_name_child(dev_info_t *child, char *name, int namelen)
295 {
296 	struct ddi_parent_private_data *pdptr;
297 	struct regspec *rp;
298 
299 	/*
300 	 * Fill in parent-private data and this function returns to us
301 	 * an indication if it used "registers" to fill in the data.
302 	 */
303 	if (ddi_get_parent_data(child) == NULL) {
304 		make_ddi_ppd(child, &pdptr);
305 		ddi_set_parent_data(child, pdptr);
306 	}
307 
308 	/*
309 	 * No reg property, return null string as address
310 	 * (e.g. root node)
311 	 */
312 	name[0] = '\0';
313 	if (sparc_pd_getnreg(child) == 0) {
314 		return (DDI_SUCCESS);
315 	}
316 
317 	rp = sparc_pd_getreg(child, 0);
318 	(void) snprintf(name, namelen, "%x,%x",
319 	    rp->regspec_bustype, rp->regspec_addr);
320 	return (DDI_SUCCESS);
321 }
322 
323 
324 /*
325  * Called from the bus_ctl op of some drivers.
326  * to implement the DDI_CTLOPS_INITCHILD operation.
327  *
328  * NEW drivers should NOT use this function, but should declare
329  * there own initchild/uninitchild handlers. (This function assumes
330  * the layout of the parent private data and the format of "reg",
331  * "ranges", "interrupts" properties and that #address-cells and
332  * #size-cells of the parent bus are defined to be default values.)
333  */
334 int
335 impl_ddi_sunbus_initchild(dev_info_t *child)
336 {
337 	char name[MAXNAMELEN];
338 
339 	(void) impl_sunbus_name_child(child, name, MAXNAMELEN);
340 	ddi_set_name_addr(child, name);
341 
342 	/*
343 	 * Try to merge .conf node. If successful, return failure to
344 	 * remove this child.
345 	 */
346 	if ((ndi_dev_is_persistent_node(child) == 0) &&
347 	    (ndi_merge_node(child, impl_sunbus_name_child) == DDI_SUCCESS)) {
348 		impl_ddi_sunbus_removechild(child);
349 		return (DDI_FAILURE);
350 	}
351 	return (DDI_SUCCESS);
352 }
353 
354 /*
355  * A better name for this function would be impl_ddi_sunbus_uninitchild()
356  * It does not remove the child, it uninitializes it, reclaiming the
357  * resources taken by impl_ddi_sunbus_initchild.
358  */
359 void
360 impl_ddi_sunbus_removechild(dev_info_t *dip)
361 {
362 	impl_free_ddi_ppd(dip);
363 	ddi_set_name_addr(dip, NULL);
364 	/*
365 	 * Strip the node to properly convert it back to prototype form
366 	 */
367 	impl_rem_dev_props(dip);
368 }
369 
370 /*
371  * SECTION: DDI Interrupt
372  */
373 
374 void
375 cells_1275_copy(prop_1275_cell_t *from, prop_1275_cell_t *to, int32_t len)
376 {
377 	int i;
378 	for (i = 0; i < len; i++)
379 		*to = *from;
380 }
381 
382 prop_1275_cell_t *
383 cells_1275_cmp(prop_1275_cell_t *cell1, prop_1275_cell_t *cell2, int32_t len)
384 {
385 	prop_1275_cell_t *match_cell = 0;
386 	int32_t i;
387 
388 	for (i = 0; i < len; i++)
389 		if (cell1[i] != cell2[i]) {
390 			match_cell = &cell1[i];
391 			break;
392 		}
393 
394 	return (match_cell);
395 }
396 
397 /*
398  * get_intr_parent() is a generic routine that process a 1275 interrupt
399  * map (imap) property.  This function returns a dev_info_t structure
400  * which claims ownership of the interrupt domain.
401  * It also returns the new interrupt translation within this new domain.
402  * If an interrupt-parent or interrupt-map property are not found,
403  * then we fallback to using the device tree's parent.
404  *
405  * imap entry format:
406  * <reg>,<interrupt>,<phandle>,<translated interrupt>
407  * reg - The register specification in the interrupts domain
408  * interrupt - The interrupt specification
409  * phandle - PROM handle of the device that owns the xlated interrupt domain
410  * translated interrupt - interrupt specifier in the parents domain
411  * note: <reg>,<interrupt> - The reg and interrupt can be combined to create
412  *	a unique entry called a unit interrupt specifier.
413  *
414  * Here's the processing steps:
415  * step1 - If the interrupt-parent property exists, create the ispec and
416  *	return the dip of the interrupt parent.
417  * step2 - Extract the interrupt-map property and the interrupt-map-mask
418  *	If these don't exist, just return the device tree parent.
419  * step3 - build up the unit interrupt specifier to match against the
420  *	interrupt map property
421  * step4 - Scan the interrupt-map property until a match is found
422  * step4a - Extract the interrupt parent
423  * step4b - Compare the unit interrupt specifier
424  */
425 dev_info_t *
426 get_intr_parent(dev_info_t *pdip, dev_info_t *dip, ddi_intr_handle_impl_t *hdlp)
427 {
428 	prop_1275_cell_t *imap, *imap_mask, *scan, *reg_p, *match_req;
429 	int32_t imap_sz, imap_cells, imap_scan_cells, imap_mask_sz,
430 	    addr_cells, intr_cells, reg_len, i, j;
431 	int32_t match_found = 0;
432 	dev_info_t *intr_parent_dip = NULL;
433 	uint32_t *intr = &hdlp->ih_vector;
434 	uint32_t nodeid;
435 #ifdef DEBUG
436 	static int debug = 0;
437 #endif
438 
439 	/*
440 	 * step1
441 	 * If we have an interrupt-parent property, this property represents
442 	 * the nodeid of our interrupt parent.
443 	 */
444 	if ((nodeid = ddi_getprop(DDI_DEV_T_ANY, dip, 0,
445 	    "interrupt-parent", -1)) != -1) {
446 		intr_parent_dip = e_ddi_nodeid_to_dip(nodeid);
447 		ASSERT(intr_parent_dip);
448 
449 		/*
450 		 * Attach the interrupt parent.
451 		 *
452 		 * N.B. e_ddi_nodeid_to_dip() isn't safe under DR.
453 		 *	Also, interrupt parent isn't held. This needs
454 		 *	to be revisited if DR-capable platforms implement
455 		 *	interrupt redirection.
456 		 */
457 		if (i_ddi_attach_node_hierarchy(intr_parent_dip)
458 		    != DDI_SUCCESS) {
459 			ndi_rele_devi(intr_parent_dip);
460 			return (NULL);
461 		}
462 
463 		return (intr_parent_dip);
464 	}
465 
466 	/*
467 	 * step2
468 	 * Get interrupt map structure from PROM property
469 	 */
470 	if (ddi_getlongprop(DDI_DEV_T_ANY, pdip, DDI_PROP_DONTPASS,
471 	    "interrupt-map", (caddr_t)&imap, &imap_sz)
472 	    != DDI_PROP_SUCCESS) {
473 		/*
474 		 * If we don't have an imap property, default to using the
475 		 * device tree.
476 		 */
477 
478 		ndi_hold_devi(pdip);
479 		return (pdip);
480 	}
481 
482 	/* Get the interrupt mask property */
483 	if (ddi_getlongprop(DDI_DEV_T_ANY, pdip, DDI_PROP_DONTPASS,
484 	    "interrupt-map-mask", (caddr_t)&imap_mask, &imap_mask_sz)
485 	    != DDI_PROP_SUCCESS) {
486 		/*
487 		 * If we don't find this property, we have to fail the request
488 		 * because the 1275 imap property wasn't defined correctly.
489 		 */
490 		ASSERT(intr_parent_dip == NULL);
491 		goto exit2;
492 	}
493 
494 	/* Get the address cell size */
495 	addr_cells = ddi_getprop(DDI_DEV_T_ANY, pdip, 0,
496 	    "#address-cells", 2);
497 
498 	/* Get the interrupts cell size */
499 	intr_cells = ddi_getprop(DDI_DEV_T_ANY, pdip, 0,
500 	    "#interrupt-cells", 1);
501 
502 	/*
503 	 * step3
504 	 * Now lets build up the unit interrupt specifier e.g. reg,intr
505 	 * and apply the imap mask.  match_req will hold this when we're
506 	 * through.
507 	 */
508 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, "reg",
509 	    (caddr_t)&reg_p, &reg_len) != DDI_SUCCESS) {
510 		ASSERT(intr_parent_dip == NULL);
511 		goto exit3;
512 	}
513 
514 	match_req = kmem_alloc(CELLS_1275_TO_BYTES(addr_cells) +
515 	    CELLS_1275_TO_BYTES(intr_cells), KM_SLEEP);
516 
517 	for (i = 0; i < addr_cells; i++)
518 		match_req[i] = (reg_p[i] & imap_mask[i]);
519 
520 	for (j = 0; j < intr_cells; i++, j++)
521 		match_req[i] = (intr[j] & imap_mask[i]);
522 
523 	/* Calculate the imap size in cells */
524 	imap_cells = BYTES_TO_1275_CELLS(imap_sz);
525 
526 #ifdef DEBUG
527 	if (debug)
528 		prom_printf("reg cell size 0x%x, intr cell size 0x%x, "
529 		    "match_request 0x%p, imap 0x%p\n", addr_cells, intr_cells,
530 		    match_req, imap);
531 #endif
532 
533 	/*
534 	 * Scan the imap property looking for a match of the interrupt unit
535 	 * specifier.  This loop is rather complex since the data within the
536 	 * imap property may vary in size.
537 	 */
538 	for (scan = imap, imap_scan_cells = i = 0;
539 	    imap_scan_cells < imap_cells; scan += i, imap_scan_cells += i) {
540 		int new_intr_cells;
541 
542 		/* Set the index to the nodeid field */
543 		i = addr_cells + intr_cells;
544 
545 		/*
546 		 * step4a
547 		 * Translate the nodeid field to a dip
548 		 */
549 		ASSERT(intr_parent_dip == NULL);
550 		intr_parent_dip = e_ddi_nodeid_to_dip((uint_t)scan[i++]);
551 
552 		ASSERT(intr_parent_dip != 0);
553 #ifdef DEBUG
554 		if (debug)
555 			prom_printf("scan 0x%p\n", scan);
556 #endif
557 		/*
558 		 * The tmp_dip describes the new domain, get it's interrupt
559 		 * cell size
560 		 */
561 		new_intr_cells = ddi_getprop(DDI_DEV_T_ANY, intr_parent_dip, 0,
562 		    "#interrupts-cells", 1);
563 
564 		/*
565 		 * step4b
566 		 * See if we have a match on the interrupt unit specifier
567 		 */
568 		if (cells_1275_cmp(match_req, scan, addr_cells + intr_cells)
569 		    == 0) {
570 			uint32_t *intr;
571 
572 			match_found = 1;
573 
574 			/*
575 			 * If we have an imap parent whose not in our device
576 			 * tree path, we need to hold and install that driver.
577 			 */
578 			if (i_ddi_attach_node_hierarchy(intr_parent_dip)
579 			    != DDI_SUCCESS) {
580 				ndi_rele_devi(intr_parent_dip);
581 				intr_parent_dip = (dev_info_t *)NULL;
582 				goto exit4;
583 			}
584 
585 			/*
586 			 * We need to handcraft an ispec along with a bus
587 			 * interrupt value, so we can dup it into our
588 			 * standard ispec structure.
589 			 */
590 			/* Extract the translated interrupt information */
591 			intr = kmem_alloc(
592 			    CELLS_1275_TO_BYTES(new_intr_cells), KM_SLEEP);
593 
594 			for (j = 0; j < new_intr_cells; j++, i++)
595 				intr[j] = scan[i];
596 
597 			cells_1275_copy(intr, &hdlp->ih_vector, new_intr_cells);
598 
599 			kmem_free(intr, CELLS_1275_TO_BYTES(new_intr_cells));
600 
601 #ifdef DEBUG
602 			if (debug)
603 				prom_printf("dip 0x%p\n", intr_parent_dip);
604 #endif
605 			break;
606 		} else {
607 #ifdef DEBUG
608 			if (debug)
609 				prom_printf("dip 0x%p\n", intr_parent_dip);
610 #endif
611 			ndi_rele_devi(intr_parent_dip);
612 			intr_parent_dip = NULL;
613 			i += new_intr_cells;
614 		}
615 	}
616 
617 	/*
618 	 * If we haven't found our interrupt parent at this point, fallback
619 	 * to using the device tree.
620 	 */
621 	if (!match_found) {
622 		ndi_hold_devi(pdip);
623 		ASSERT(intr_parent_dip == NULL);
624 		intr_parent_dip = pdip;
625 	}
626 
627 	ASSERT(intr_parent_dip != NULL);
628 
629 exit4:
630 	kmem_free(reg_p, reg_len);
631 	kmem_free(match_req, CELLS_1275_TO_BYTES(addr_cells) +
632 	    CELLS_1275_TO_BYTES(intr_cells));
633 
634 exit3:
635 	kmem_free(imap_mask, imap_mask_sz);
636 
637 exit2:
638 	kmem_free(imap, imap_sz);
639 
640 	return (intr_parent_dip);
641 }
642 
643 /*
644  * process_intr_ops:
645  *
646  * Process the interrupt op via the interrupt parent.
647  */
648 int
649 process_intr_ops(dev_info_t *pdip, dev_info_t *rdip, ddi_intr_op_t op,
650     ddi_intr_handle_impl_t *hdlp, void *result)
651 {
652 	int		ret = DDI_FAILURE;
653 
654 	if (NEXUS_HAS_INTR_OP(pdip)) {
655 		ret = (*(DEVI(pdip)->devi_ops->devo_bus_ops->
656 		    bus_intr_op)) (pdip, rdip, op, hdlp, result);
657 	} else {
658 		cmn_err(CE_WARN, "Failed to process interrupt "
659 		    "for %s%d due to down-rev nexus driver %s%d",
660 		    ddi_get_name(rdip), ddi_get_instance(rdip),
661 		    ddi_get_name(pdip), ddi_get_instance(pdip));
662 	}
663 
664 	return (ret);
665 }
666 
667 /*ARGSUSED*/
668 uint_t
669 softlevel1(caddr_t arg)
670 {
671 	softint();
672 	return (1);
673 }
674 
675 /*
676  * indirection table, to save us some large switch statements
677  * NOTE: This must agree with "INTLEVEL_foo" constants in
678  *	<sys/avintr.h>
679  */
680 struct autovec *const vectorlist[] = { 0 };
681 
682 /*
683  * This value is exported here for the functions in avintr.c
684  */
685 const uint_t maxautovec = (sizeof (vectorlist) / sizeof (vectorlist[0]));
686 
687 /*
688  * Check for machine specific interrupt levels which cannot be reassigned by
689  * settrap(), sun4u version.
690  *
691  * sun4u does not support V8 SPARC "fast trap" handlers.
692  */
693 /*ARGSUSED*/
694 int
695 exclude_settrap(int lvl)
696 {
697 	return (1);
698 }
699 
700 /*
701  * Check for machine specific interrupt levels which cannot have interrupt
702  * handlers added. We allow levels 1 through 15; level 0 is nonsense.
703  */
704 /*ARGSUSED*/
705 int
706 exclude_level(int lvl)
707 {
708 	return ((lvl < 1) || (lvl > 15));
709 }
710 
711 /*
712  * Wrapper functions used by New DDI interrupt framework.
713  */
714 
715 /*
716  * i_ddi_intr_ops:
717  */
718 int
719 i_ddi_intr_ops(dev_info_t *dip, dev_info_t *rdip, ddi_intr_op_t op,
720     ddi_intr_handle_impl_t *hdlp, void *result)
721 {
722 	dev_info_t	*pdip = ddi_get_parent(dip);
723 	int		ret = DDI_FAILURE;
724 
725 	/*
726 	 * The following check is required to address
727 	 * one of the test case of ADDI test suite.
728 	 */
729 	if (pdip == NULL)
730 		return (DDI_FAILURE);
731 
732 	if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
733 		return (process_intr_ops(pdip, rdip, op, hdlp, result));
734 
735 	if (hdlp->ih_vector == 0)
736 		hdlp->ih_vector = i_ddi_get_inum(rdip, hdlp->ih_inum);
737 
738 	if (hdlp->ih_pri == 0)
739 		hdlp->ih_pri = i_ddi_get_intr_pri(rdip, hdlp->ih_inum);
740 
741 	switch (op) {
742 	case DDI_INTROP_ADDISR:
743 	case DDI_INTROP_REMISR:
744 	case DDI_INTROP_ENABLE:
745 	case DDI_INTROP_DISABLE:
746 	case DDI_INTROP_BLOCKENABLE:
747 	case DDI_INTROP_BLOCKDISABLE:
748 		/*
749 		 * Try and determine our parent and possibly an interrupt
750 		 * translation. intr parent dip returned held
751 		 */
752 		if ((pdip = get_intr_parent(pdip, dip, hdlp)) == NULL)
753 			goto done;
754 	}
755 
756 	ret = process_intr_ops(pdip, rdip, op, hdlp, result);
757 
758 done:
759 	switch (op) {
760 	case DDI_INTROP_ADDISR:
761 	case DDI_INTROP_REMISR:
762 	case DDI_INTROP_ENABLE:
763 	case DDI_INTROP_DISABLE:
764 	case DDI_INTROP_BLOCKENABLE:
765 	case DDI_INTROP_BLOCKDISABLE:
766 		/* Release hold acquired in get_intr_parent() */
767 		if (pdip)
768 			ndi_rele_devi(pdip);
769 	}
770 
771 	hdlp->ih_vector = 0;
772 
773 	return (ret);
774 }
775 
776 /*
777  * i_ddi_add_ivintr:
778  */
779 /*ARGSUSED*/
780 int
781 i_ddi_add_ivintr(ddi_intr_handle_impl_t *hdlp)
782 {
783 	/*
784 	 * If the PIL was set and is valid use it, otherwise
785 	 * default it to 1
786 	 */
787 	if ((hdlp->ih_pri < 1) || (hdlp->ih_pri > PIL_MAX))
788 		hdlp->ih_pri = 1;
789 
790 	VERIFY(add_ivintr(hdlp->ih_vector, hdlp->ih_pri,
791 	    (intrfunc)hdlp->ih_cb_func, hdlp->ih_cb_arg1,
792 	    hdlp->ih_cb_arg2, NULL) == 0);
793 
794 	return (DDI_SUCCESS);
795 }
796 
797 /*
798  * i_ddi_rem_ivintr:
799  */
800 /*ARGSUSED*/
801 void
802 i_ddi_rem_ivintr(ddi_intr_handle_impl_t *hdlp)
803 {
804 	VERIFY(rem_ivintr(hdlp->ih_vector, hdlp->ih_pri) == 0);
805 }
806 
807 /*
808  * i_ddi_get_inum - Get the interrupt number property from the
809  * specified device. Note that this function is called only for
810  * the FIXED interrupt type.
811  */
812 uint32_t
813 i_ddi_get_inum(dev_info_t *dip, uint_t inumber)
814 {
815 	int32_t			intrlen, intr_cells, max_intrs;
816 	prop_1275_cell_t	*ip, intr_sz;
817 	uint32_t		intr = 0;
818 
819 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS |
820 	    DDI_PROP_CANSLEEP,
821 	    "interrupts", (caddr_t)&ip, &intrlen) == DDI_SUCCESS) {
822 
823 		intr_cells = ddi_getprop(DDI_DEV_T_ANY, dip, 0,
824 		    "#interrupt-cells", 1);
825 
826 		/* adjust for number of bytes */
827 		intr_sz = CELLS_1275_TO_BYTES(intr_cells);
828 
829 		/* Calculate the number of interrupts */
830 		max_intrs = intrlen / intr_sz;
831 
832 		if (inumber < max_intrs) {
833 			prop_1275_cell_t *intrp = ip;
834 
835 			/* Index into interrupt property */
836 			intrp += (inumber * intr_cells);
837 
838 			cells_1275_copy(intrp, &intr, intr_cells);
839 		}
840 
841 		kmem_free(ip, intrlen);
842 	}
843 
844 	return (intr);
845 }
846 
847 /*
848  * i_ddi_get_intr_pri - Get the interrupt-priorities property from
849  * the specified device. Note that this function is called only for
850  * the FIXED interrupt type.
851  */
852 uint32_t
853 i_ddi_get_intr_pri(dev_info_t *dip, uint_t inumber)
854 {
855 	uint32_t	*intr_prio_p;
856 	uint32_t	pri = 0;
857 	int32_t		i;
858 
859 	/*
860 	 * Use the "interrupt-priorities" property to determine the
861 	 * the pil/ipl for the interrupt handler.
862 	 */
863 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
864 	    "interrupt-priorities", (caddr_t)&intr_prio_p,
865 	    &i) == DDI_SUCCESS) {
866 		if (inumber < (i / sizeof (int32_t)))
867 			pri = intr_prio_p[inumber];
868 		kmem_free(intr_prio_p, i);
869 	}
870 
871 	return (pri);
872 }
873 
874 int
875 i_ddi_get_intx_nintrs(dev_info_t *dip)
876 {
877 	int32_t intrlen;
878 	prop_1275_cell_t intr_sz;
879 	prop_1275_cell_t *ip;
880 	int32_t ret = 0;
881 
882 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS |
883 	    DDI_PROP_CANSLEEP,
884 	    "interrupts", (caddr_t)&ip, &intrlen) == DDI_SUCCESS) {
885 
886 		intr_sz = ddi_getprop(DDI_DEV_T_ANY, dip, 0,
887 		    "#interrupt-cells", 1);
888 		/* adjust for number of bytes */
889 		intr_sz = CELLS_1275_TO_BYTES(intr_sz);
890 
891 		ret = intrlen / intr_sz;
892 
893 		kmem_free(ip, intrlen);
894 	}
895 
896 	return (ret);
897 }
898 
899 /*
900  * i_ddi_add_softint - allocate and add a software interrupt.
901  *
902  * NOTE: All software interrupts that are registered through DDI
903  *	 should be triggered only on a single target or CPU.
904  */
905 int
906 i_ddi_add_softint(ddi_softint_hdl_impl_t *hdlp)
907 {
908 	if ((hdlp->ih_private = (void *)add_softintr(hdlp->ih_pri,
909 	    hdlp->ih_cb_func, hdlp->ih_cb_arg1, SOFTINT_ST)) == NULL)
910 		return (DDI_FAILURE);
911 
912 	return (DDI_SUCCESS);
913 }
914 
915 /*
916  * i_ddi_remove_softint - remove and free a software interrupt.
917  */
918 void
919 i_ddi_remove_softint(ddi_softint_hdl_impl_t *hdlp)
920 {
921 	ASSERT(hdlp->ih_private != NULL);
922 
923 	if (rem_softintr((uint64_t)hdlp->ih_private) == 0)
924 		hdlp->ih_private = NULL;
925 }
926 
927 /*
928  * i_ddi_trigger_softint - trigger a software interrupt.
929  */
930 int
931 i_ddi_trigger_softint(ddi_softint_hdl_impl_t *hdlp, void *arg2)
932 {
933 	int	ret;
934 
935 	ASSERT(hdlp->ih_private != NULL);
936 
937 	/* Update the second argument for the software interrupt */
938 	if ((ret = update_softint_arg2((uint64_t)hdlp->ih_private, arg2)) == 0)
939 		setsoftint((uint64_t)hdlp->ih_private);
940 
941 	return (ret ? DDI_EPENDING : DDI_SUCCESS);
942 }
943 
944 /*
945  * i_ddi_set_softint_pri - change software interrupt priority.
946  */
947 /* ARGSUSED */
948 int
949 i_ddi_set_softint_pri(ddi_softint_hdl_impl_t *hdlp, uint_t old_pri)
950 {
951 	int	ret;
952 
953 	ASSERT(hdlp->ih_private != NULL);
954 
955 	/* Update the interrupt priority for the software interrupt */
956 	ret = update_softint_pri((uint64_t)hdlp->ih_private, hdlp->ih_pri);
957 
958 	return (ret ? DDI_FAILURE : DDI_SUCCESS);
959 }
960 
961 /*ARGSUSED*/
962 void
963 i_ddi_alloc_intr_phdl(ddi_intr_handle_impl_t *hdlp)
964 {
965 }
966 
967 /*ARGSUSED*/
968 void
969 i_ddi_free_intr_phdl(ddi_intr_handle_impl_t *hdlp)
970 {
971 }
972 
973 /*
974  * SECTION: DDI Memory/DMA
975  */
976 
977 /* set HAT endianess attributes from ddi_device_acc_attr */
978 void
979 i_ddi_devacc_to_hatacc(ddi_device_acc_attr_t *devaccp, uint_t *hataccp)
980 {
981 	if (devaccp != NULL) {
982 		if (devaccp->devacc_attr_endian_flags == DDI_STRUCTURE_LE_ACC) {
983 			*hataccp &= ~HAT_ENDIAN_MASK;
984 			*hataccp |= HAT_STRUCTURE_LE;
985 		}
986 	}
987 }
988 
989 /*
990  * Check if the specified cache attribute is supported on the platform.
991  * This function must be called before i_ddi_cacheattr_to_hatacc().
992  */
993 boolean_t
994 i_ddi_check_cache_attr(uint_t flags)
995 {
996 	/*
997 	 * The cache attributes are mutually exclusive. Any combination of
998 	 * the attributes leads to a failure.
999 	 */
1000 	uint_t cache_attr = IOMEM_CACHE_ATTR(flags);
1001 	if ((cache_attr != 0) && ((cache_attr & (cache_attr - 1)) != 0))
1002 		return (B_FALSE);
1003 
1004 	/*
1005 	 * On the sparc architecture, only IOMEM_DATA_CACHED is meaningful,
1006 	 * but others lead to a failure.
1007 	 */
1008 	if (cache_attr & IOMEM_DATA_CACHED)
1009 		return (B_TRUE);
1010 	else
1011 		return (B_FALSE);
1012 }
1013 
1014 /* set HAT cache attributes from the cache attributes */
1015 void
1016 i_ddi_cacheattr_to_hatacc(uint_t flags, uint_t *hataccp)
1017 {
1018 	uint_t cache_attr = IOMEM_CACHE_ATTR(flags);
1019 	static char *fname = "i_ddi_cacheattr_to_hatacc";
1020 #if defined(lint)
1021 	*hataccp = *hataccp;
1022 #endif
1023 	/*
1024 	 * set HAT attrs according to the cache attrs.
1025 	 */
1026 	switch (cache_attr) {
1027 	/*
1028 	 * The cache coherency is always maintained on SPARC, and
1029 	 * nothing is required.
1030 	 */
1031 	case IOMEM_DATA_CACHED:
1032 		break;
1033 	/*
1034 	 * Both IOMEM_DATA_UC_WRITE_COMBINED and IOMEM_DATA_UNCACHED are
1035 	 * not supported on SPARC -- this case must not occur because the
1036 	 * cache attribute is scrutinized before this function is called.
1037 	 */
1038 	case IOMEM_DATA_UNCACHED:
1039 	case IOMEM_DATA_UC_WR_COMBINE:
1040 	default:
1041 		cmn_err(CE_WARN, "%s: cache_attr=0x%x is ignored.",
1042 		    fname, cache_attr);
1043 	}
1044 }
1045 
1046 static vmem_t *little_endian_arena;
1047 static vmem_t *big_endian_arena;
1048 
1049 static void *
1050 segkmem_alloc_le(vmem_t *vmp, size_t size, int flag)
1051 {
1052 	return (segkmem_xalloc(vmp, NULL, size, flag, HAT_STRUCTURE_LE,
1053 	    segkmem_page_create, NULL));
1054 }
1055 
1056 static void *
1057 segkmem_alloc_be(vmem_t *vmp, size_t size, int flag)
1058 {
1059 	return (segkmem_xalloc(vmp, NULL, size, flag, HAT_STRUCTURE_BE,
1060 	    segkmem_page_create, NULL));
1061 }
1062 
1063 void
1064 ka_init(void)
1065 {
1066 	little_endian_arena = vmem_create("little_endian", NULL, 0, 1,
1067 	    segkmem_alloc_le, segkmem_free, heap_arena, 0, VM_SLEEP);
1068 	big_endian_arena = vmem_create("big_endian", NULL, 0, 1,
1069 	    segkmem_alloc_be, segkmem_free, heap_arena, 0, VM_SLEEP);
1070 }
1071 
1072 /*
1073  * Allocate from the system, aligned on a specific boundary.
1074  * The alignment, if non-zero, must be a power of 2.
1075  */
1076 static void *
1077 kalloca(size_t size, size_t align, int cansleep, uint_t endian_flags)
1078 {
1079 	size_t *addr, *raddr, rsize;
1080 	size_t hdrsize = 4 * sizeof (size_t);	/* must be power of 2 */
1081 
1082 	align = MAX(align, hdrsize);
1083 	ASSERT((align & (align - 1)) == 0);
1084 
1085 	/*
1086 	 * We need to allocate
1087 	 *    rsize = size + hdrsize + align - MIN(hdrsize, buffer_alignment)
1088 	 * bytes to be sure we have enough freedom to satisfy the request.
1089 	 * Since the buffer alignment depends on the request size, this is
1090 	 * not straightforward to use directly.
1091 	 *
1092 	 * kmem guarantees that any allocation of a 64-byte multiple will be
1093 	 * 64-byte aligned.  Since rounding up the request could add more
1094 	 * than we save, we compute the size with and without alignment, and
1095 	 * use the smaller of the two.
1096 	 */
1097 	rsize = size + hdrsize + align;
1098 
1099 	if (endian_flags == DDI_STRUCTURE_LE_ACC) {
1100 		raddr = vmem_alloc(little_endian_arena, rsize,
1101 		    cansleep ? VM_SLEEP : VM_NOSLEEP);
1102 	} else {
1103 		raddr = vmem_alloc(big_endian_arena, rsize,
1104 		    cansleep ? VM_SLEEP : VM_NOSLEEP);
1105 	}
1106 
1107 	if (raddr == NULL)
1108 		return (NULL);
1109 
1110 	addr = (size_t *)P2ROUNDUP((uintptr_t)raddr + hdrsize, align);
1111 	ASSERT((uintptr_t)addr + size - (uintptr_t)raddr <= rsize);
1112 
1113 	addr[-3] = (size_t)endian_flags;
1114 	addr[-2] = (size_t)raddr;
1115 	addr[-1] = rsize;
1116 
1117 	return (addr);
1118 }
1119 
1120 static void
1121 kfreea(void *addr)
1122 {
1123 	size_t *saddr = addr;
1124 
1125 	if (saddr[-3] == DDI_STRUCTURE_LE_ACC)
1126 		vmem_free(little_endian_arena, (void *)saddr[-2], saddr[-1]);
1127 	else
1128 		vmem_free(big_endian_arena, (void *)saddr[-2], saddr[-1]);
1129 }
1130 
1131 int
1132 i_ddi_mem_alloc(dev_info_t *dip, ddi_dma_attr_t *attr,
1133     size_t length, int cansleep, int flags,
1134     ddi_device_acc_attr_t *accattrp,
1135     caddr_t *kaddrp, size_t *real_length, ddi_acc_hdl_t *handlep)
1136 {
1137 	caddr_t a;
1138 	int iomin, align, streaming;
1139 	uint_t endian_flags = DDI_NEVERSWAP_ACC;
1140 
1141 #if defined(lint)
1142 	*handlep = *handlep;
1143 #endif
1144 
1145 	/*
1146 	 * Check legality of arguments
1147 	 */
1148 	if (length == 0 || kaddrp == NULL || attr == NULL) {
1149 		return (DDI_FAILURE);
1150 	}
1151 
1152 	if (attr->dma_attr_minxfer == 0 || attr->dma_attr_align == 0 ||
1153 	    (attr->dma_attr_align & (attr->dma_attr_align - 1)) ||
1154 	    (attr->dma_attr_minxfer & (attr->dma_attr_minxfer - 1))) {
1155 		return (DDI_FAILURE);
1156 	}
1157 
1158 	/*
1159 	 * check if a streaming sequential xfer is requested.
1160 	 */
1161 	streaming = (flags & DDI_DMA_STREAMING) ? 1 : 0;
1162 
1163 	/*
1164 	 * Drivers for 64-bit capable SBus devices will encode
1165 	 * the burtsizes for 64-bit xfers in the upper 16-bits.
1166 	 * For DMA alignment, we use the most restrictive
1167 	 * alignment of 32-bit and 64-bit xfers.
1168 	 */
1169 	iomin = (attr->dma_attr_burstsizes & 0xffff) |
1170 	    ((attr->dma_attr_burstsizes >> 16) & 0xffff);
1171 	/*
1172 	 * If a driver set burtsizes to 0, we give him byte alignment.
1173 	 * Otherwise align at the burtsizes boundary.
1174 	 */
1175 	if (iomin == 0)
1176 		iomin = 1;
1177 	else
1178 		iomin = 1 << (ddi_fls(iomin) - 1);
1179 	iomin = maxbit(iomin, attr->dma_attr_minxfer);
1180 	iomin = maxbit(iomin, attr->dma_attr_align);
1181 	iomin = ddi_iomin(dip, iomin, streaming);
1182 	if (iomin == 0)
1183 		return (DDI_FAILURE);
1184 
1185 	ASSERT((iomin & (iomin - 1)) == 0);
1186 	ASSERT(iomin >= attr->dma_attr_minxfer);
1187 	ASSERT(iomin >= attr->dma_attr_align);
1188 
1189 	length = P2ROUNDUP(length, iomin);
1190 	align = iomin;
1191 
1192 	if (accattrp != NULL)
1193 		endian_flags = accattrp->devacc_attr_endian_flags;
1194 
1195 	a = kalloca(length, align, cansleep, endian_flags);
1196 	if ((*kaddrp = a) == 0) {
1197 		return (DDI_FAILURE);
1198 	} else {
1199 		if (real_length) {
1200 			*real_length = length;
1201 		}
1202 		if (handlep) {
1203 			/*
1204 			 * assign handle information
1205 			 */
1206 			impl_acc_hdl_init(handlep);
1207 		}
1208 		return (DDI_SUCCESS);
1209 	}
1210 }
1211 
1212 /*
1213  * covert old DMA limits structure to DMA attribute structure
1214  * and continue
1215  */
1216 int
1217 i_ddi_mem_alloc_lim(dev_info_t *dip, ddi_dma_lim_t *limits,
1218     size_t length, int cansleep, int streaming,
1219     ddi_device_acc_attr_t *accattrp, caddr_t *kaddrp,
1220     uint_t *real_length, ddi_acc_hdl_t *ap)
1221 {
1222 	ddi_dma_attr_t dma_attr, *attrp;
1223 	size_t rlen;
1224 	int ret;
1225 
1226 	ASSERT(limits);
1227 	attrp = &dma_attr;
1228 	attrp->dma_attr_version = DMA_ATTR_V0;
1229 	attrp->dma_attr_addr_lo = (uint64_t)limits->dlim_addr_lo;
1230 	attrp->dma_attr_addr_hi = (uint64_t)limits->dlim_addr_hi;
1231 	attrp->dma_attr_count_max = (uint64_t)-1;
1232 	attrp->dma_attr_align = 1;
1233 	attrp->dma_attr_burstsizes = (uint_t)limits->dlim_burstsizes;
1234 	attrp->dma_attr_minxfer = (uint32_t)limits->dlim_minxfer;
1235 	attrp->dma_attr_maxxfer = (uint64_t)-1;
1236 	attrp->dma_attr_seg = (uint64_t)limits->dlim_cntr_max;
1237 	attrp->dma_attr_sgllen = 1;
1238 	attrp->dma_attr_granular = 1;
1239 	attrp->dma_attr_flags = 0;
1240 
1241 	ret = i_ddi_mem_alloc(dip, attrp, length, cansleep, streaming,
1242 	    accattrp, kaddrp, &rlen, ap);
1243 	if (ret == DDI_SUCCESS) {
1244 		if (real_length)
1245 			*real_length = (uint_t)rlen;
1246 	}
1247 	return (ret);
1248 }
1249 
1250 /* ARGSUSED */
1251 void
1252 i_ddi_mem_free(caddr_t kaddr, ddi_acc_hdl_t *ap)
1253 {
1254 	kfreea(kaddr);
1255 }
1256 
1257 /*
1258  * SECTION: DDI Data Access
1259  */
1260 
1261 static uintptr_t impl_acc_hdl_id = 0;
1262 
1263 /*
1264  * access handle allocator
1265  */
1266 ddi_acc_hdl_t *
1267 impl_acc_hdl_get(ddi_acc_handle_t hdl)
1268 {
1269 	/*
1270 	 * Extract the access handle address from the DDI implemented
1271 	 * access handle
1272 	 */
1273 	return (&((ddi_acc_impl_t *)hdl)->ahi_common);
1274 }
1275 
1276 ddi_acc_handle_t
1277 impl_acc_hdl_alloc(int (*waitfp)(caddr_t), caddr_t arg)
1278 {
1279 	ddi_acc_impl_t *hp;
1280 	on_trap_data_t *otp;
1281 	int sleepflag;
1282 
1283 	sleepflag = ((waitfp == (int (*)())KM_SLEEP) ? KM_SLEEP : KM_NOSLEEP);
1284 
1285 	/*
1286 	 * Allocate and initialize the data access handle and error status.
1287 	 */
1288 	if ((hp = kmem_zalloc(sizeof (ddi_acc_impl_t), sleepflag)) == NULL)
1289 		goto fail;
1290 	if ((hp->ahi_err = (ndi_err_t *)kmem_zalloc(
1291 	    sizeof (ndi_err_t), sleepflag)) == NULL) {
1292 		kmem_free(hp, sizeof (ddi_acc_impl_t));
1293 		goto fail;
1294 	}
1295 	if ((otp = (on_trap_data_t *)kmem_zalloc(
1296 	    sizeof (on_trap_data_t), sleepflag)) == NULL) {
1297 		kmem_free(hp->ahi_err, sizeof (ndi_err_t));
1298 		kmem_free(hp, sizeof (ddi_acc_impl_t));
1299 		goto fail;
1300 	}
1301 	hp->ahi_err->err_ontrap = otp;
1302 	hp->ahi_common.ah_platform_private = (void *)hp;
1303 
1304 	return ((ddi_acc_handle_t)hp);
1305 fail:
1306 	if ((waitfp != (int (*)())KM_SLEEP) &&
1307 	    (waitfp != (int (*)())KM_NOSLEEP))
1308 		ddi_set_callback(waitfp, arg, &impl_acc_hdl_id);
1309 	return (NULL);
1310 }
1311 
1312 void
1313 impl_acc_hdl_free(ddi_acc_handle_t handle)
1314 {
1315 	ddi_acc_impl_t *hp;
1316 
1317 	/*
1318 	 * The supplied (ddi_acc_handle_t) is actually a (ddi_acc_impl_t *),
1319 	 * because that's what we allocated in impl_acc_hdl_alloc() above.
1320 	 */
1321 	hp = (ddi_acc_impl_t *)handle;
1322 	if (hp) {
1323 		kmem_free(hp->ahi_err->err_ontrap, sizeof (on_trap_data_t));
1324 		kmem_free(hp->ahi_err, sizeof (ndi_err_t));
1325 		kmem_free(hp, sizeof (ddi_acc_impl_t));
1326 		if (impl_acc_hdl_id)
1327 			ddi_run_callback(&impl_acc_hdl_id);
1328 	}
1329 }
1330 
1331 #define	PCI_GET_MP_PFN(mp, page_no)	((mp)->dmai_ndvmapages == 1 ? \
1332 	(pfn_t)(mp)->dmai_iopte:(((pfn_t *)(mp)->dmai_iopte)[page_no]))
1333 
1334 /*
1335  * Function called after a dma fault occurred to find out whether the
1336  * fault address is associated with a driver that is able to handle faults
1337  * and recover from faults.
1338  */
1339 /* ARGSUSED */
1340 int
1341 impl_dma_check(dev_info_t *dip, const void *handle, const void *addr,
1342     const void *not_used)
1343 {
1344 	ddi_dma_impl_t *mp = (ddi_dma_impl_t *)handle;
1345 	pfn_t fault_pfn = mmu_btop(*(uint64_t *)addr);
1346 	pfn_t comp_pfn;
1347 
1348 	/*
1349 	 * The driver has to set DDI_DMA_FLAGERR to recover from dma faults.
1350 	 */
1351 	int page;
1352 
1353 	ASSERT(mp);
1354 	for (page = 0; page < mp->dmai_ndvmapages; page++) {
1355 		comp_pfn = PCI_GET_MP_PFN(mp, page);
1356 		if (fault_pfn == comp_pfn)
1357 			return (DDI_FM_NONFATAL);
1358 	}
1359 	return (DDI_FM_UNKNOWN);
1360 }
1361 
1362 /*
1363  * Function used to check if a given access handle owns the failing address.
1364  * Called by ndi_fmc_error, when we detect a PIO error.
1365  */
1366 /* ARGSUSED */
1367 static int
1368 impl_acc_check(dev_info_t *dip, const void *handle, const void *addr,
1369     const void *not_used)
1370 {
1371 	pfn_t pfn, fault_pfn;
1372 	ddi_acc_hdl_t *hp;
1373 
1374 	hp = impl_acc_hdl_get((ddi_acc_handle_t)handle);
1375 
1376 	ASSERT(hp);
1377 
1378 	if (addr != NULL) {
1379 		pfn = hp->ah_pfn;
1380 		fault_pfn = mmu_btop(*(uint64_t *)addr);
1381 		if (fault_pfn >= pfn && fault_pfn < (pfn + hp->ah_pnum))
1382 			return (DDI_FM_NONFATAL);
1383 	}
1384 	return (DDI_FM_UNKNOWN);
1385 }
1386 
1387 void
1388 impl_acc_err_init(ddi_acc_hdl_t *handlep)
1389 {
1390 	int fmcap;
1391 	ndi_err_t *errp;
1392 	on_trap_data_t *otp;
1393 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)handlep;
1394 
1395 	fmcap = ddi_fm_capable(handlep->ah_dip);
1396 
1397 	if (handlep->ah_acc.devacc_attr_version < DDI_DEVICE_ATTR_V1 ||
1398 	    !DDI_FM_ACC_ERR_CAP(fmcap)) {
1399 		handlep->ah_acc.devacc_attr_access = DDI_DEFAULT_ACC;
1400 	} else if (DDI_FM_ACC_ERR_CAP(fmcap)) {
1401 		if (handlep->ah_acc.devacc_attr_access == DDI_DEFAULT_ACC) {
1402 			i_ddi_drv_ereport_post(handlep->ah_dip, DVR_EFMCAP,
1403 			    NULL, DDI_NOSLEEP);
1404 		} else {
1405 			errp = hp->ahi_err;
1406 			otp = (on_trap_data_t *)errp->err_ontrap;
1407 			otp->ot_handle = (void *)(hp);
1408 			otp->ot_prot = OT_DATA_ACCESS;
1409 			if (handlep->ah_acc.devacc_attr_access ==
1410 			    DDI_CAUTIOUS_ACC)
1411 				otp->ot_trampoline =
1412 				    (uintptr_t)&i_ddi_caut_trampoline;
1413 			else
1414 				otp->ot_trampoline =
1415 				    (uintptr_t)&i_ddi_prot_trampoline;
1416 			errp->err_status = DDI_FM_OK;
1417 			errp->err_expected = DDI_FM_ERR_UNEXPECTED;
1418 			errp->err_cf = impl_acc_check;
1419 		}
1420 	}
1421 }
1422 
1423 void
1424 impl_acc_hdl_init(ddi_acc_hdl_t *handlep)
1425 {
1426 	ddi_acc_impl_t *hp;
1427 
1428 	ASSERT(handlep);
1429 
1430 	hp = (ddi_acc_impl_t *)handlep;
1431 
1432 	/*
1433 	 * check for SW byte-swapping
1434 	 */
1435 	hp->ahi_get8 = i_ddi_get8;
1436 	hp->ahi_put8 = i_ddi_put8;
1437 	hp->ahi_rep_get8 = i_ddi_rep_get8;
1438 	hp->ahi_rep_put8 = i_ddi_rep_put8;
1439 	if (handlep->ah_acc.devacc_attr_endian_flags & DDI_STRUCTURE_LE_ACC) {
1440 		hp->ahi_get16 = i_ddi_swap_get16;
1441 		hp->ahi_get32 = i_ddi_swap_get32;
1442 		hp->ahi_get64 = i_ddi_swap_get64;
1443 		hp->ahi_put16 = i_ddi_swap_put16;
1444 		hp->ahi_put32 = i_ddi_swap_put32;
1445 		hp->ahi_put64 = i_ddi_swap_put64;
1446 		hp->ahi_rep_get16 = i_ddi_swap_rep_get16;
1447 		hp->ahi_rep_get32 = i_ddi_swap_rep_get32;
1448 		hp->ahi_rep_get64 = i_ddi_swap_rep_get64;
1449 		hp->ahi_rep_put16 = i_ddi_swap_rep_put16;
1450 		hp->ahi_rep_put32 = i_ddi_swap_rep_put32;
1451 		hp->ahi_rep_put64 = i_ddi_swap_rep_put64;
1452 	} else {
1453 		hp->ahi_get16 = i_ddi_get16;
1454 		hp->ahi_get32 = i_ddi_get32;
1455 		hp->ahi_get64 = i_ddi_get64;
1456 		hp->ahi_put16 = i_ddi_put16;
1457 		hp->ahi_put32 = i_ddi_put32;
1458 		hp->ahi_put64 = i_ddi_put64;
1459 		hp->ahi_rep_get16 = i_ddi_rep_get16;
1460 		hp->ahi_rep_get32 = i_ddi_rep_get32;
1461 		hp->ahi_rep_get64 = i_ddi_rep_get64;
1462 		hp->ahi_rep_put16 = i_ddi_rep_put16;
1463 		hp->ahi_rep_put32 = i_ddi_rep_put32;
1464 		hp->ahi_rep_put64 = i_ddi_rep_put64;
1465 	}
1466 
1467 	/* Legacy fault flags and support */
1468 	hp->ahi_fault_check = i_ddi_acc_fault_check;
1469 	hp->ahi_fault_notify = i_ddi_acc_fault_notify;
1470 	hp->ahi_fault = 0;
1471 	impl_acc_err_init(handlep);
1472 }
1473 
1474 void
1475 i_ddi_acc_set_fault(ddi_acc_handle_t handle)
1476 {
1477 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)handle;
1478 
1479 	if (!hp->ahi_fault) {
1480 		hp->ahi_fault = 1;
1481 			(*hp->ahi_fault_notify)(hp);
1482 	}
1483 }
1484 
1485 void
1486 i_ddi_acc_clr_fault(ddi_acc_handle_t handle)
1487 {
1488 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)handle;
1489 
1490 	if (hp->ahi_fault) {
1491 		hp->ahi_fault = 0;
1492 			(*hp->ahi_fault_notify)(hp);
1493 	}
1494 }
1495 
1496 /* ARGSUSED */
1497 void
1498 i_ddi_acc_fault_notify(ddi_acc_impl_t *hp)
1499 {
1500 	/* Default version, does nothing */
1501 }
1502 
1503 /*
1504  * SECTION: Misc functions
1505  */
1506 
1507 /*
1508  * instance wrappers
1509  */
1510 /*ARGSUSED*/
1511 uint_t
1512 impl_assign_instance(dev_info_t *dip)
1513 {
1514 	return ((uint_t)-1);
1515 }
1516 
1517 /*ARGSUSED*/
1518 int
1519 impl_keep_instance(dev_info_t *dip)
1520 {
1521 	return (DDI_FAILURE);
1522 }
1523 
1524 /*ARGSUSED*/
1525 int
1526 impl_free_instance(dev_info_t *dip)
1527 {
1528 	return (DDI_FAILURE);
1529 }
1530 
1531 /*ARGSUSED*/
1532 int
1533 impl_check_cpu(dev_info_t *devi)
1534 {
1535 	return (DDI_SUCCESS);
1536 }
1537 
1538 
1539 static const char *nocopydevs[] = {
1540 	"SUNW,ffb",
1541 	"SUNW,afb",
1542 	NULL
1543 };
1544 
1545 /*
1546  * Perform a copy from a memory mapped device (whose devinfo pointer is devi)
1547  * separately mapped at devaddr in the kernel to a kernel buffer at kaddr.
1548  */
1549 /*ARGSUSED*/
1550 int
1551 e_ddi_copyfromdev(dev_info_t *devi,
1552     off_t off, const void *devaddr, void *kaddr, size_t len)
1553 {
1554 	const char **argv;
1555 
1556 	for (argv = nocopydevs; *argv; argv++)
1557 		if (strcmp(ddi_binding_name(devi), *argv) == 0) {
1558 			bzero(kaddr, len);
1559 			return (0);
1560 		}
1561 
1562 	bcopy(devaddr, kaddr, len);
1563 	return (0);
1564 }
1565 
1566 /*
1567  * Perform a copy to a memory mapped device (whose devinfo pointer is devi)
1568  * separately mapped at devaddr in the kernel from a kernel buffer at kaddr.
1569  */
1570 /*ARGSUSED*/
1571 int
1572 e_ddi_copytodev(dev_info_t *devi,
1573     off_t off, const void *kaddr, void *devaddr, size_t len)
1574 {
1575 	const char **argv;
1576 
1577 	for (argv = nocopydevs; *argv; argv++)
1578 		if (strcmp(ddi_binding_name(devi), *argv) == 0)
1579 			return (1);
1580 
1581 	bcopy(kaddr, devaddr, len);
1582 	return (0);
1583 }
1584 
1585 /*
1586  * Boot Configuration
1587  */
1588 idprom_t idprom;
1589 
1590 /*
1591  * Configure the hardware on the system.
1592  * Called before the rootfs is mounted
1593  */
1594 void
1595 configure(void)
1596 {
1597 	extern void i_ddi_init_root();
1598 
1599 	/* We better have released boot by this time! */
1600 	ASSERT(!bootops);
1601 
1602 	/*
1603 	 * Determine whether or not to use the fpu, V9 SPARC cpus
1604 	 * always have one. Could check for existence of a fp queue,
1605 	 * Ultra I, II and IIa do not have a fp queue.
1606 	 */
1607 	if (fpu_exists)
1608 		fpu_probe();
1609 	else
1610 		cmn_err(CE_CONT, "FPU not in use\n");
1611 
1612 #if 0 /* XXXQ - not necessary for sun4u */
1613 	/*
1614 	 * This following line fixes bugid 1041296; we need to do a
1615 	 * prom_nextnode(0) because this call ALSO patches the DMA+
1616 	 * bug in Campus-B and Phoenix. The prom uncaches the traptable
1617 	 * page as a side-effect of devr_next(0) (which prom_nextnode calls),
1618 	 * so this *must* be executed early on. (XXX This is untrue for sun4u)
1619 	 */
1620 	(void) prom_nextnode((pnode_t)0);
1621 #endif
1622 
1623 	/*
1624 	 * Initialize devices on the machine.
1625 	 * Uses configuration tree built by the PROMs to determine what
1626 	 * is present, and builds a tree of prototype dev_info nodes
1627 	 * corresponding to the hardware which identified itself.
1628 	 */
1629 	i_ddi_init_root();
1630 
1631 #ifdef	DDI_PROP_DEBUG
1632 	(void) ddi_prop_debug(1);	/* Enable property debugging */
1633 #endif	/* DDI_PROP_DEBUG */
1634 }
1635 
1636 /*
1637  * The "status" property indicates the operational status of a device.
1638  * If this property is present, the value is a string indicating the
1639  * status of the device as follows:
1640  *
1641  *	"okay"		operational.
1642  *	"disabled"	not operational, but might become operational.
1643  *	"fail"		not operational because a fault has been detected,
1644  *			and it is unlikely that the device will become
1645  *			operational without repair. no additional details
1646  *			are available.
1647  *	"fail-xxx"	not operational because a fault has been detected,
1648  *			and it is unlikely that the device will become
1649  *			operational without repair. "xxx" is additional
1650  *			human-readable information about the particular
1651  *			fault condition that was detected.
1652  *
1653  * The absence of this property means that the operational status is
1654  * unknown or okay.
1655  *
1656  * This routine checks the status property of the specified device node
1657  * and returns 0 if the operational status indicates failure, and 1 otherwise.
1658  *
1659  * The property may exist on plug-in cards the existed before IEEE 1275-1994.
1660  * And, in that case, the property may not even be a string. So we carefully
1661  * check for the value "fail", in the beginning of the string, noting
1662  * the property length.
1663  */
1664 int
1665 status_okay(int id, char *buf, int buflen)
1666 {
1667 	char status_buf[OBP_MAXPROPNAME];
1668 	char *bufp = buf;
1669 	int len = buflen;
1670 	int proplen;
1671 	static const char *status = "status";
1672 	static const char *fail = "fail";
1673 	size_t fail_len = strlen(fail);
1674 
1675 	/*
1676 	 * Get the proplen ... if it's smaller than "fail",
1677 	 * or doesn't exist ... then we don't care, since
1678 	 * the value can't begin with the char string "fail".
1679 	 *
1680 	 * NB: proplen, if it's a string, includes the NULL in the
1681 	 * the size of the property, and fail_len does not.
1682 	 */
1683 	proplen = prom_getproplen((pnode_t)id, (caddr_t)status);
1684 	if (proplen <= fail_len)	/* nonexistent or uninteresting len */
1685 		return (1);
1686 
1687 	/*
1688 	 * if a buffer was provided, use it
1689 	 */
1690 	if ((buf == (char *)NULL) || (buflen <= 0)) {
1691 		bufp = status_buf;
1692 		len = sizeof (status_buf);
1693 	}
1694 	*bufp = (char)0;
1695 
1696 	/*
1697 	 * Get the property into the buffer, to the extent of the buffer,
1698 	 * and in case the buffer is smaller than the property size,
1699 	 * NULL terminate the buffer. (This handles the case where
1700 	 * a buffer was passed in and the caller wants to print the
1701 	 * value, but the buffer was too small).
1702 	 */
1703 	(void) prom_bounded_getprop((pnode_t)id, (caddr_t)status,
1704 	    (caddr_t)bufp, len);
1705 	*(bufp + len - 1) = (char)0;
1706 
1707 	/*
1708 	 * If the value begins with the char string "fail",
1709 	 * then it means the node is failed. We don't care
1710 	 * about any other values. We assume the node is ok
1711 	 * although it might be 'disabled'.
1712 	 */
1713 	if (strncmp(bufp, fail, fail_len) == 0)
1714 		return (0);
1715 
1716 	return (1);
1717 }
1718 
1719 
1720 /*
1721  * We set the cpu type from the idprom, if we can.
1722  * Note that we just read out the contents of it, for the most part.
1723  */
1724 void
1725 setcputype(void)
1726 {
1727 	/*
1728 	 * We cache the idprom info early on so that we don't
1729 	 * rummage through the NVRAM unnecessarily later.
1730 	 */
1731 	(void) prom_getidprom((caddr_t)&idprom, sizeof (idprom));
1732 }
1733 
1734 /*
1735  *  Here is where we actually infer meanings to the members of idprom_t
1736  */
1737 void
1738 parse_idprom(void)
1739 {
1740 	if (idprom.id_format == IDFORM_1) {
1741 		uint_t i;
1742 
1743 		(void) localetheraddr((struct ether_addr *)idprom.id_ether,
1744 		    (struct ether_addr *)NULL);
1745 
1746 		i = idprom.id_machine << 24;
1747 		i = i + idprom.id_serial;
1748 		numtos((ulong_t)i, hw_serial);
1749 	} else
1750 		prom_printf("Invalid format code in IDprom.\n");
1751 }
1752 
1753 /*
1754  * Allow for implementation specific correction of PROM property values.
1755  */
1756 /*ARGSUSED*/
1757 void
1758 impl_fix_props(dev_info_t *dip, dev_info_t *ch_dip, char *name, int len,
1759     caddr_t buffer)
1760 {
1761 	/*
1762 	 * There are no adjustments needed in this implementation.
1763 	 */
1764 }
1765 
1766 /*
1767  * The following functions ready a cautious request to go up to the nexus
1768  * driver.  It is up to the nexus driver to decide how to process the request.
1769  * It may choose to call i_ddi_do_caut_get/put in this file, or do it
1770  * differently.
1771  */
1772 
1773 static void
1774 i_ddi_caut_getput_ctlops(
1775     ddi_acc_impl_t *hp, uint64_t host_addr, uint64_t dev_addr, size_t size,
1776     size_t repcount, uint_t flags, ddi_ctl_enum_t cmd)
1777 {
1778 	peekpoke_ctlops_t	cautacc_ctlops_arg;
1779 
1780 	cautacc_ctlops_arg.size = size;
1781 	cautacc_ctlops_arg.dev_addr = dev_addr;
1782 	cautacc_ctlops_arg.host_addr = host_addr;
1783 	cautacc_ctlops_arg.handle = (ddi_acc_handle_t)hp;
1784 	cautacc_ctlops_arg.repcount = repcount;
1785 	cautacc_ctlops_arg.flags = flags;
1786 
1787 	(void) ddi_ctlops(hp->ahi_common.ah_dip, hp->ahi_common.ah_dip, cmd,
1788 	    &cautacc_ctlops_arg, NULL);
1789 }
1790 
1791 uint8_t
1792 i_ddi_caut_get8(ddi_acc_impl_t *hp, uint8_t *addr)
1793 {
1794 	uint8_t value;
1795 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1796 	    sizeof (uint8_t), 1, 0, DDI_CTLOPS_PEEK);
1797 
1798 	return (value);
1799 }
1800 
1801 uint16_t
1802 i_ddi_caut_get16(ddi_acc_impl_t *hp, uint16_t *addr)
1803 {
1804 	uint16_t value;
1805 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1806 	    sizeof (uint16_t), 1, 0, DDI_CTLOPS_PEEK);
1807 
1808 	return (value);
1809 }
1810 
1811 uint32_t
1812 i_ddi_caut_get32(ddi_acc_impl_t *hp, uint32_t *addr)
1813 {
1814 	uint32_t value;
1815 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1816 	    sizeof (uint32_t), 1, 0, DDI_CTLOPS_PEEK);
1817 
1818 	return (value);
1819 }
1820 
1821 uint64_t
1822 i_ddi_caut_get64(ddi_acc_impl_t *hp, uint64_t *addr)
1823 {
1824 	uint64_t value;
1825 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1826 	    sizeof (uint64_t), 1, 0, DDI_CTLOPS_PEEK);
1827 
1828 	return (value);
1829 }
1830 
1831 void
1832 i_ddi_caut_put8(ddi_acc_impl_t *hp, uint8_t *addr, uint8_t value)
1833 {
1834 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1835 	    sizeof (uint8_t), 1, 0, DDI_CTLOPS_POKE);
1836 }
1837 
1838 void
1839 i_ddi_caut_put16(ddi_acc_impl_t *hp, uint16_t *addr, uint16_t value)
1840 {
1841 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1842 	    sizeof (uint16_t), 1, 0, DDI_CTLOPS_POKE);
1843 }
1844 
1845 void
1846 i_ddi_caut_put32(ddi_acc_impl_t *hp, uint32_t *addr, uint32_t value)
1847 {
1848 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1849 	    sizeof (uint32_t), 1, 0, DDI_CTLOPS_POKE);
1850 }
1851 
1852 void
1853 i_ddi_caut_put64(ddi_acc_impl_t *hp, uint64_t *addr, uint64_t value)
1854 {
1855 	i_ddi_caut_getput_ctlops(hp, (uint64_t)&value, (uint64_t)addr,
1856 	    sizeof (uint64_t), 1, 0, DDI_CTLOPS_POKE);
1857 }
1858 
1859 void
1860 i_ddi_caut_rep_get8(ddi_acc_impl_t *hp, uint8_t *host_addr, uint8_t *dev_addr,
1861 	size_t repcount, uint_t flags)
1862 {
1863 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1864 	    sizeof (uint8_t), repcount, flags, DDI_CTLOPS_PEEK);
1865 }
1866 
1867 void
1868 i_ddi_caut_rep_get16(ddi_acc_impl_t *hp, uint16_t *host_addr,
1869     uint16_t *dev_addr, size_t repcount, uint_t flags)
1870 {
1871 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1872 	    sizeof (uint16_t), repcount, flags, DDI_CTLOPS_PEEK);
1873 }
1874 
1875 void
1876 i_ddi_caut_rep_get32(ddi_acc_impl_t *hp, uint32_t *host_addr,
1877     uint32_t *dev_addr, size_t repcount, uint_t flags)
1878 {
1879 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1880 	    sizeof (uint32_t), repcount, flags, DDI_CTLOPS_PEEK);
1881 }
1882 
1883 void
1884 i_ddi_caut_rep_get64(ddi_acc_impl_t *hp, uint64_t *host_addr,
1885     uint64_t *dev_addr, size_t repcount, uint_t flags)
1886 {
1887 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1888 	    sizeof (uint64_t), repcount, flags, DDI_CTLOPS_PEEK);
1889 }
1890 
1891 void
1892 i_ddi_caut_rep_put8(ddi_acc_impl_t *hp, uint8_t *host_addr, uint8_t *dev_addr,
1893 	size_t repcount, uint_t flags)
1894 {
1895 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1896 	    sizeof (uint8_t), repcount, flags, DDI_CTLOPS_POKE);
1897 }
1898 
1899 void
1900 i_ddi_caut_rep_put16(ddi_acc_impl_t *hp, uint16_t *host_addr,
1901     uint16_t *dev_addr, size_t repcount, uint_t flags)
1902 {
1903 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1904 	    sizeof (uint16_t), repcount, flags, DDI_CTLOPS_POKE);
1905 }
1906 
1907 void
1908 i_ddi_caut_rep_put32(ddi_acc_impl_t *hp, uint32_t *host_addr,
1909     uint32_t *dev_addr, size_t repcount, uint_t flags)
1910 {
1911 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1912 	    sizeof (uint32_t), repcount, flags, DDI_CTLOPS_POKE);
1913 }
1914 
1915 void
1916 i_ddi_caut_rep_put64(ddi_acc_impl_t *hp, uint64_t *host_addr,
1917     uint64_t *dev_addr, size_t repcount, uint_t flags)
1918 {
1919 	i_ddi_caut_getput_ctlops(hp, (uint64_t)host_addr, (uint64_t)dev_addr,
1920 	    sizeof (uint64_t), repcount, flags, DDI_CTLOPS_POKE);
1921 }
1922 
1923 /*
1924  * This is called only to process peek/poke when the DIP is NULL.
1925  * Assume that this is for memory, as nexi take care of device safe accesses.
1926  */
1927 int
1928 peekpoke_mem(ddi_ctl_enum_t cmd, peekpoke_ctlops_t *in_args)
1929 {
1930 	int err = DDI_SUCCESS;
1931 	on_trap_data_t otd;
1932 
1933 	/* Set up protected environment. */
1934 	if (!on_trap(&otd, OT_DATA_ACCESS)) {
1935 		uintptr_t tramp = otd.ot_trampoline;
1936 
1937 		if (cmd == DDI_CTLOPS_POKE) {
1938 			otd.ot_trampoline = (uintptr_t)&poke_fault;
1939 			err = do_poke(in_args->size, (void *)in_args->dev_addr,
1940 			    (void *)in_args->host_addr);
1941 		} else {
1942 			otd.ot_trampoline = (uintptr_t)&peek_fault;
1943 			err = do_peek(in_args->size, (void *)in_args->dev_addr,
1944 			    (void *)in_args->host_addr);
1945 		}
1946 		otd.ot_trampoline = tramp;
1947 	} else
1948 		err = DDI_FAILURE;
1949 
1950 	/* Take down protected environment. */
1951 	no_trap();
1952 
1953 	return (err);
1954 }
1955 
1956 /*
1957  * Platform independent DR routines
1958  */
1959 
1960 static int
1961 ndi2errno(int n)
1962 {
1963 	int err = 0;
1964 
1965 	switch (n) {
1966 		case NDI_NOMEM:
1967 			err = ENOMEM;
1968 			break;
1969 		case NDI_BUSY:
1970 			err = EBUSY;
1971 			break;
1972 		case NDI_FAULT:
1973 			err = EFAULT;
1974 			break;
1975 		case NDI_FAILURE:
1976 			err = EIO;
1977 			break;
1978 		case NDI_SUCCESS:
1979 			break;
1980 		case NDI_BADHANDLE:
1981 		default:
1982 			err = EINVAL;
1983 			break;
1984 	}
1985 	return (err);
1986 }
1987 
1988 /*
1989  * Prom tree node list
1990  */
1991 struct ptnode {
1992 	pnode_t		nodeid;
1993 	struct ptnode	*next;
1994 };
1995 
1996 /*
1997  * Prom tree walk arg
1998  */
1999 struct pta {
2000 	dev_info_t	*pdip;
2001 	devi_branch_t	*bp;
2002 	uint_t		flags;
2003 	dev_info_t	*fdip;
2004 	struct ptnode	*head;
2005 };
2006 
2007 static void
2008 visit_node(pnode_t nodeid, struct pta *ap)
2009 {
2010 	struct ptnode	**nextp;
2011 	int		(*select)(pnode_t, void *, uint_t);
2012 
2013 	ASSERT(nodeid != OBP_NONODE && nodeid != OBP_BADNODE);
2014 
2015 	select = ap->bp->create.prom_branch_select;
2016 
2017 	ASSERT(select);
2018 
2019 	if (select(nodeid, ap->bp->arg, 0) == DDI_SUCCESS) {
2020 
2021 		for (nextp = &ap->head; *nextp; nextp = &(*nextp)->next)
2022 			;
2023 
2024 		*nextp = kmem_zalloc(sizeof (struct ptnode), KM_SLEEP);
2025 
2026 		(*nextp)->nodeid = nodeid;
2027 	}
2028 
2029 	if ((ap->flags & DEVI_BRANCH_CHILD) == DEVI_BRANCH_CHILD)
2030 		return;
2031 
2032 	nodeid = prom_childnode(nodeid);
2033 	while (nodeid != OBP_NONODE && nodeid != OBP_BADNODE) {
2034 		visit_node(nodeid, ap);
2035 		nodeid = prom_nextnode(nodeid);
2036 	}
2037 }
2038 
2039 /*ARGSUSED*/
2040 static int
2041 set_dip_offline(dev_info_t *dip, void *arg)
2042 {
2043 	ASSERT(dip);
2044 
2045 	mutex_enter(&(DEVI(dip)->devi_lock));
2046 	if (!DEVI_IS_DEVICE_OFFLINE(dip))
2047 		DEVI_SET_DEVICE_OFFLINE(dip);
2048 	mutex_exit(&(DEVI(dip)->devi_lock));
2049 
2050 	return (DDI_WALK_CONTINUE);
2051 }
2052 
2053 /*ARGSUSED*/
2054 static int
2055 create_prom_branch(void *arg, int has_changed)
2056 {
2057 	int		circ, c;
2058 	int		exists, rv;
2059 	pnode_t		nodeid;
2060 	struct ptnode	*tnp;
2061 	dev_info_t	*dip;
2062 	struct pta	*ap = arg;
2063 	devi_branch_t	*bp;
2064 
2065 	ASSERT(ap);
2066 	ASSERT(ap->fdip == NULL);
2067 	ASSERT(ap->pdip && ndi_dev_is_prom_node(ap->pdip));
2068 
2069 	bp = ap->bp;
2070 
2071 	nodeid = ddi_get_nodeid(ap->pdip);
2072 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE) {
2073 		cmn_err(CE_WARN, "create_prom_branch: invalid "
2074 		    "nodeid: 0x%x", nodeid);
2075 		return (EINVAL);
2076 	}
2077 
2078 	ap->head = NULL;
2079 
2080 	nodeid = prom_childnode(nodeid);
2081 	while (nodeid != OBP_NONODE && nodeid != OBP_BADNODE) {
2082 		visit_node(nodeid, ap);
2083 		nodeid = prom_nextnode(nodeid);
2084 	}
2085 
2086 	if (ap->head == NULL)
2087 		return (ENODEV);
2088 
2089 	rv = 0;
2090 	while ((tnp = ap->head) != NULL) {
2091 		ap->head = tnp->next;
2092 
2093 		ndi_devi_enter(ap->pdip, &circ);
2094 
2095 		/*
2096 		 * Check if the branch already exists.
2097 		 */
2098 		exists = 0;
2099 		dip = e_ddi_nodeid_to_dip(tnp->nodeid);
2100 		if (dip != NULL) {
2101 			exists = 1;
2102 
2103 			/* Parent is held busy, so release hold */
2104 			ndi_rele_devi(dip);
2105 #ifdef	DEBUG
2106 			cmn_err(CE_WARN, "create_prom_branch: dip(%p) exists"
2107 			    " for nodeid 0x%x", (void *)dip, tnp->nodeid);
2108 #endif
2109 		} else {
2110 			dip = i_ddi_create_branch(ap->pdip, tnp->nodeid);
2111 		}
2112 
2113 		kmem_free(tnp, sizeof (struct ptnode));
2114 
2115 		if (dip == NULL) {
2116 			ndi_devi_exit(ap->pdip, circ);
2117 			rv = EIO;
2118 			continue;
2119 		}
2120 
2121 		ASSERT(ddi_get_parent(dip) == ap->pdip);
2122 
2123 		/*
2124 		 * Hold the branch if it is not already held
2125 		 */
2126 		if (!exists)
2127 			e_ddi_branch_hold(dip);
2128 
2129 		ASSERT(e_ddi_branch_held(dip));
2130 
2131 		/*
2132 		 * Set all dips in the branch offline so that
2133 		 * only a "configure" operation can attach
2134 		 * the branch
2135 		 */
2136 		(void) set_dip_offline(dip, NULL);
2137 
2138 		ndi_devi_enter(dip, &c);
2139 		ddi_walk_devs(ddi_get_child(dip), set_dip_offline, NULL);
2140 		ndi_devi_exit(dip, c);
2141 
2142 		ndi_devi_exit(ap->pdip, circ);
2143 
2144 		if (ap->flags & DEVI_BRANCH_CONFIGURE) {
2145 			int error = e_ddi_branch_configure(dip, &ap->fdip, 0);
2146 			if (error && rv == 0)
2147 				rv = error;
2148 		}
2149 
2150 		/*
2151 		 * Invoke devi_branch_callback() (if it exists) only for
2152 		 * newly created branches
2153 		 */
2154 		if (bp->devi_branch_callback && !exists)
2155 			bp->devi_branch_callback(dip, bp->arg, 0);
2156 	}
2157 
2158 	return (rv);
2159 }
2160 
2161 static int
2162 sid_node_create(dev_info_t *pdip, devi_branch_t *bp, dev_info_t **rdipp)
2163 {
2164 	int			rv, circ, len;
2165 	int			i, flags;
2166 	dev_info_t		*dip;
2167 	char			*nbuf;
2168 	static const char	*noname = "<none>";
2169 
2170 	ASSERT(pdip);
2171 	ASSERT(DEVI_BUSY_OWNED(pdip));
2172 
2173 	flags = 0;
2174 
2175 	/*
2176 	 * Creating the root of a branch ?
2177 	 */
2178 	if (rdipp) {
2179 		*rdipp = NULL;
2180 		flags = DEVI_BRANCH_ROOT;
2181 	}
2182 
2183 	ndi_devi_alloc_sleep(pdip, (char *)noname, DEVI_SID_NODEID, &dip);
2184 	rv = bp->create.sid_branch_create(dip, bp->arg, flags);
2185 
2186 	nbuf = kmem_alloc(OBP_MAXDRVNAME, KM_SLEEP);
2187 
2188 	if (rv == DDI_WALK_ERROR) {
2189 		cmn_err(CE_WARN, "e_ddi_branch_create: Error setting"
2190 		    " properties on devinfo node %p",  (void *)dip);
2191 		goto fail;
2192 	}
2193 
2194 	len = OBP_MAXDRVNAME;
2195 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2196 	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM, "name", nbuf, &len)
2197 	    != DDI_PROP_SUCCESS) {
2198 		cmn_err(CE_WARN, "e_ddi_branch_create: devinfo node %p has"
2199 		    "no name property", (void *)dip);
2200 		goto fail;
2201 	}
2202 
2203 	ASSERT(i_ddi_node_state(dip) == DS_PROTO);
2204 	if (ndi_devi_set_nodename(dip, nbuf, 0) != NDI_SUCCESS) {
2205 		cmn_err(CE_WARN, "e_ddi_branch_create: cannot set name (%s)"
2206 		    " for devinfo node %p", nbuf, (void *)dip);
2207 		goto fail;
2208 	}
2209 
2210 	kmem_free(nbuf, OBP_MAXDRVNAME);
2211 
2212 	/*
2213 	 * Ignore bind failures just like boot does
2214 	 */
2215 	(void) ndi_devi_bind_driver(dip, 0);
2216 
2217 	switch (rv) {
2218 	case DDI_WALK_CONTINUE:
2219 	case DDI_WALK_PRUNESIB:
2220 		ndi_devi_enter(dip, &circ);
2221 
2222 		i = DDI_WALK_CONTINUE;
2223 		for (; i == DDI_WALK_CONTINUE; ) {
2224 			i = sid_node_create(dip, bp, NULL);
2225 		}
2226 
2227 		ASSERT(i == DDI_WALK_ERROR || i == DDI_WALK_PRUNESIB);
2228 		if (i == DDI_WALK_ERROR)
2229 			rv = i;
2230 		/*
2231 		 * If PRUNESIB stop creating siblings
2232 		 * of dip's child. Subsequent walk behavior
2233 		 * is determined by rv returned by dip.
2234 		 */
2235 
2236 		ndi_devi_exit(dip, circ);
2237 		break;
2238 	case DDI_WALK_TERMINATE:
2239 		/*
2240 		 * Don't create children and ask our parent
2241 		 * to not create siblings either.
2242 		 */
2243 		rv = DDI_WALK_PRUNESIB;
2244 		break;
2245 	case DDI_WALK_PRUNECHILD:
2246 		/*
2247 		 * Don't create children, but ask parent to continue
2248 		 * with siblings.
2249 		 */
2250 		rv = DDI_WALK_CONTINUE;
2251 		break;
2252 	default:
2253 		ASSERT(0);
2254 		break;
2255 	}
2256 
2257 	if (rdipp)
2258 		*rdipp = dip;
2259 
2260 	/*
2261 	 * Set device offline - only the "configure" op should cause an attach
2262 	 */
2263 	(void) set_dip_offline(dip, NULL);
2264 
2265 	return (rv);
2266 fail:
2267 	(void) ndi_devi_free(dip);
2268 	kmem_free(nbuf, OBP_MAXDRVNAME);
2269 	return (DDI_WALK_ERROR);
2270 }
2271 
2272 static int
2273 create_sid_branch(
2274 	dev_info_t	*pdip,
2275 	devi_branch_t	*bp,
2276 	dev_info_t	**dipp,
2277 	uint_t		flags)
2278 {
2279 	int		rv = 0, state = DDI_WALK_CONTINUE;
2280 	dev_info_t	*rdip;
2281 
2282 	while (state == DDI_WALK_CONTINUE) {
2283 		int	circ;
2284 
2285 		ndi_devi_enter(pdip, &circ);
2286 
2287 		state = sid_node_create(pdip, bp, &rdip);
2288 		if (rdip == NULL) {
2289 			ndi_devi_exit(pdip, circ);
2290 			ASSERT(state == DDI_WALK_ERROR);
2291 			break;
2292 		}
2293 
2294 		e_ddi_branch_hold(rdip);
2295 
2296 		ndi_devi_exit(pdip, circ);
2297 
2298 		if (flags & DEVI_BRANCH_CONFIGURE) {
2299 			int error = e_ddi_branch_configure(rdip, dipp, 0);
2300 			if (error && rv == 0)
2301 				rv = error;
2302 		}
2303 
2304 		/*
2305 		 * devi_branch_callback() is optional
2306 		 */
2307 		if (bp->devi_branch_callback)
2308 			bp->devi_branch_callback(rdip, bp->arg, 0);
2309 	}
2310 
2311 	ASSERT(state == DDI_WALK_ERROR || state == DDI_WALK_PRUNESIB);
2312 
2313 	return (state == DDI_WALK_ERROR ? EIO : rv);
2314 }
2315 
2316 int
2317 e_ddi_branch_create(
2318 	dev_info_t	*pdip,
2319 	devi_branch_t	*bp,
2320 	dev_info_t	**dipp,
2321 	uint_t		flags)
2322 {
2323 	int prom_devi, sid_devi, error;
2324 
2325 	if (pdip == NULL || bp == NULL || bp->type == 0)
2326 		return (EINVAL);
2327 
2328 	prom_devi = (bp->type == DEVI_BRANCH_PROM) ? 1 : 0;
2329 	sid_devi = (bp->type == DEVI_BRANCH_SID) ? 1 : 0;
2330 
2331 	if (prom_devi && bp->create.prom_branch_select == NULL)
2332 		return (EINVAL);
2333 	else if (sid_devi && bp->create.sid_branch_create == NULL)
2334 		return (EINVAL);
2335 	else if (!prom_devi && !sid_devi)
2336 		return (EINVAL);
2337 
2338 	if (flags & DEVI_BRANCH_EVENT)
2339 		return (EINVAL);
2340 
2341 	if (prom_devi) {
2342 		struct pta pta = {0};
2343 
2344 		pta.pdip = pdip;
2345 		pta.bp = bp;
2346 		pta.flags = flags;
2347 
2348 		error = prom_tree_access(create_prom_branch, &pta, NULL);
2349 
2350 		if (dipp)
2351 			*dipp = pta.fdip;
2352 		else if (pta.fdip)
2353 			ndi_rele_devi(pta.fdip);
2354 	} else {
2355 		error = create_sid_branch(pdip, bp, dipp, flags);
2356 	}
2357 
2358 	return (error);
2359 }
2360 
2361 int
2362 e_ddi_branch_configure(dev_info_t *rdip, dev_info_t **dipp, uint_t flags)
2363 {
2364 	int		circ, rv;
2365 	char		*devnm;
2366 	dev_info_t	*pdip;
2367 
2368 	if (dipp)
2369 		*dipp = NULL;
2370 
2371 	if (rdip == NULL || flags != 0 || (flags & DEVI_BRANCH_EVENT))
2372 		return (EINVAL);
2373 
2374 	pdip = ddi_get_parent(rdip);
2375 
2376 	ndi_devi_enter(pdip, &circ);
2377 
2378 	if (!e_ddi_branch_held(rdip)) {
2379 		ndi_devi_exit(pdip, circ);
2380 		cmn_err(CE_WARN, "e_ddi_branch_configure: "
2381 		    "dip(%p) not held", (void *)rdip);
2382 		return (EINVAL);
2383 	}
2384 
2385 	if (i_ddi_node_state(rdip) < DS_INITIALIZED) {
2386 		/*
2387 		 * First attempt to bind a driver. If we fail, return
2388 		 * success (On some platforms, dips for some device
2389 		 * types (CPUs) may not have a driver)
2390 		 */
2391 		if (ndi_devi_bind_driver(rdip, 0) != NDI_SUCCESS) {
2392 			ndi_devi_exit(pdip, circ);
2393 			return (0);
2394 		}
2395 
2396 		if (ddi_initchild(pdip, rdip) != DDI_SUCCESS) {
2397 			rv = NDI_FAILURE;
2398 			goto out;
2399 		}
2400 	}
2401 
2402 	ASSERT(i_ddi_node_state(rdip) >= DS_INITIALIZED);
2403 
2404 	devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2405 
2406 	(void) ddi_deviname(rdip, devnm);
2407 
2408 	if ((rv = ndi_devi_config_one(pdip, devnm+1, &rdip,
2409 	    NDI_DEVI_ONLINE | NDI_CONFIG)) == NDI_SUCCESS) {
2410 		/* release hold from ndi_devi_config_one() */
2411 		ndi_rele_devi(rdip);
2412 	}
2413 
2414 	kmem_free(devnm, MAXNAMELEN + 1);
2415 out:
2416 	if (rv != NDI_SUCCESS && dipp) {
2417 		ndi_hold_devi(rdip);
2418 		*dipp = rdip;
2419 	}
2420 	ndi_devi_exit(pdip, circ);
2421 	return (ndi2errno(rv));
2422 }
2423 
2424 void
2425 e_ddi_branch_hold(dev_info_t *rdip)
2426 {
2427 	if (e_ddi_branch_held(rdip)) {
2428 		cmn_err(CE_WARN, "e_ddi_branch_hold: branch already held");
2429 		return;
2430 	}
2431 
2432 	mutex_enter(&DEVI(rdip)->devi_lock);
2433 	if ((DEVI(rdip)->devi_flags & DEVI_BRANCH_HELD) == 0) {
2434 		DEVI(rdip)->devi_flags |= DEVI_BRANCH_HELD;
2435 		DEVI(rdip)->devi_ref++;
2436 	}
2437 	ASSERT(DEVI(rdip)->devi_ref > 0);
2438 	mutex_exit(&DEVI(rdip)->devi_lock);
2439 }
2440 
2441 int
2442 e_ddi_branch_held(dev_info_t *rdip)
2443 {
2444 	int rv = 0;
2445 
2446 	mutex_enter(&DEVI(rdip)->devi_lock);
2447 	if ((DEVI(rdip)->devi_flags & DEVI_BRANCH_HELD) &&
2448 	    DEVI(rdip)->devi_ref > 0) {
2449 		rv = 1;
2450 	}
2451 	mutex_exit(&DEVI(rdip)->devi_lock);
2452 
2453 	return (rv);
2454 }
2455 void
2456 e_ddi_branch_rele(dev_info_t *rdip)
2457 {
2458 	mutex_enter(&DEVI(rdip)->devi_lock);
2459 	DEVI(rdip)->devi_flags &= ~DEVI_BRANCH_HELD;
2460 	DEVI(rdip)->devi_ref--;
2461 	mutex_exit(&DEVI(rdip)->devi_lock);
2462 }
2463 
2464 int
2465 e_ddi_branch_unconfigure(
2466 	dev_info_t *rdip,
2467 	dev_info_t **dipp,
2468 	uint_t flags)
2469 {
2470 	int	circ, rv;
2471 	int	destroy;
2472 	char	*devnm;
2473 	uint_t	nflags;
2474 	dev_info_t *pdip;
2475 
2476 	if (dipp)
2477 		*dipp = NULL;
2478 
2479 	if (rdip == NULL)
2480 		return (EINVAL);
2481 
2482 	pdip = ddi_get_parent(rdip);
2483 
2484 	ASSERT(pdip);
2485 
2486 	/*
2487 	 * Check if caller holds pdip busy - can cause deadlocks during
2488 	 * devfs_clean()
2489 	 */
2490 	if (DEVI_BUSY_OWNED(pdip)) {
2491 		cmn_err(CE_WARN, "e_ddi_branch_unconfigure: failed: parent"
2492 		    " devinfo node(%p) is busy held", (void *)pdip);
2493 		return (EINVAL);
2494 	}
2495 
2496 	destroy = (flags & DEVI_BRANCH_DESTROY) ? 1 : 0;
2497 
2498 	devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2499 
2500 	ndi_devi_enter(pdip, &circ);
2501 	(void) ddi_deviname(rdip, devnm);
2502 	ndi_devi_exit(pdip, circ);
2503 
2504 	/*
2505 	 * ddi_deviname() returns a component name with / prepended.
2506 	 */
2507 	rv = devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2508 	if (rv) {
2509 		kmem_free(devnm, MAXNAMELEN + 1);
2510 		return (rv);
2511 	}
2512 
2513 	ndi_devi_enter(pdip, &circ);
2514 
2515 	/*
2516 	 * Recreate device name as it may have changed state (init/uninit)
2517 	 * when parent busy lock was dropped for devfs_clean()
2518 	 */
2519 	(void) ddi_deviname(rdip, devnm);
2520 
2521 	if (!e_ddi_branch_held(rdip)) {
2522 		kmem_free(devnm, MAXNAMELEN + 1);
2523 		ndi_devi_exit(pdip, circ);
2524 		cmn_err(CE_WARN, "e_ddi_%s_branch: dip(%p) not held",
2525 		    destroy ? "destroy" : "unconfigure", (void *)rdip);
2526 		return (EINVAL);
2527 	}
2528 
2529 	/*
2530 	 * Release hold on the branch. This is ok since we are holding the
2531 	 * parent busy. If rdip is not removed, we must do a hold on the
2532 	 * branch before returning.
2533 	 */
2534 	e_ddi_branch_rele(rdip);
2535 
2536 	nflags = NDI_DEVI_OFFLINE;
2537 	if (destroy || (flags & DEVI_BRANCH_DESTROY)) {
2538 		nflags |= NDI_DEVI_REMOVE;
2539 		destroy = 1;
2540 	} else {
2541 		nflags |= NDI_UNCONFIG;		/* uninit but don't remove */
2542 	}
2543 
2544 	if (flags & DEVI_BRANCH_EVENT)
2545 		nflags |= NDI_POST_EVENT;
2546 
2547 	if (i_ddi_devi_attached(pdip) &&
2548 	    (i_ddi_node_state(rdip) >= DS_INITIALIZED)) {
2549 		rv = ndi_devi_unconfig_one(pdip, devnm+1, dipp, nflags);
2550 	} else {
2551 		rv = e_ddi_devi_unconfig(rdip, dipp, nflags);
2552 		if (rv == NDI_SUCCESS) {
2553 			ASSERT(!destroy || ddi_get_child(rdip) == NULL);
2554 			rv = ndi_devi_offline(rdip, nflags);
2555 		}
2556 	}
2557 
2558 	if (!destroy || rv != NDI_SUCCESS) {
2559 		/* The dip still exists, so do a hold */
2560 		e_ddi_branch_hold(rdip);
2561 	}
2562 out:
2563 	kmem_free(devnm, MAXNAMELEN + 1);
2564 	ndi_devi_exit(pdip, circ);
2565 	return (ndi2errno(rv));
2566 }
2567 
2568 int
2569 e_ddi_branch_destroy(dev_info_t *rdip, dev_info_t **dipp, uint_t flag)
2570 {
2571 	return (e_ddi_branch_unconfigure(rdip, dipp,
2572 	    flag|DEVI_BRANCH_DESTROY));
2573 }
2574 
2575 /*
2576  * Number of chains for hash table
2577  */
2578 #define	NUMCHAINS	17
2579 
2580 /*
2581  * Devinfo busy arg
2582  */
2583 struct devi_busy {
2584 	int dv_total;
2585 	int s_total;
2586 	mod_hash_t *dv_hash;
2587 	mod_hash_t *s_hash;
2588 	int (*callback)(dev_info_t *, void *, uint_t);
2589 	void *arg;
2590 };
2591 
2592 static int
2593 visit_dip(dev_info_t *dip, void *arg)
2594 {
2595 	uintptr_t sbusy, dvbusy, ref;
2596 	struct devi_busy *bsp = arg;
2597 
2598 	ASSERT(bsp->callback);
2599 
2600 	/*
2601 	 * A dip cannot be busy if its reference count is 0
2602 	 */
2603 	if ((ref = e_ddi_devi_holdcnt(dip)) == 0) {
2604 		return (bsp->callback(dip, bsp->arg, 0));
2605 	}
2606 
2607 	if (mod_hash_find(bsp->dv_hash, dip, (mod_hash_val_t *)&dvbusy))
2608 		dvbusy = 0;
2609 
2610 	/*
2611 	 * To catch device opens currently maintained on specfs common snodes.
2612 	 */
2613 	if (mod_hash_find(bsp->s_hash, dip, (mod_hash_val_t *)&sbusy))
2614 		sbusy = 0;
2615 
2616 #ifdef	DEBUG
2617 	if (ref < sbusy || ref < dvbusy) {
2618 		cmn_err(CE_WARN, "dip(%p): sopen = %lu, dvopen = %lu "
2619 		    "dip ref = %lu\n", (void *)dip, sbusy, dvbusy, ref);
2620 	}
2621 #endif
2622 
2623 	dvbusy = (sbusy > dvbusy) ? sbusy : dvbusy;
2624 
2625 	return (bsp->callback(dip, bsp->arg, dvbusy));
2626 }
2627 
2628 static int
2629 visit_snode(struct snode *sp, void *arg)
2630 {
2631 	uintptr_t sbusy;
2632 	dev_info_t *dip;
2633 	int count;
2634 	struct devi_busy *bsp = arg;
2635 
2636 	ASSERT(sp);
2637 
2638 	/*
2639 	 * The stable lock is held. This prevents
2640 	 * the snode and its associated dip from
2641 	 * going away.
2642 	 */
2643 	dip = NULL;
2644 	count = spec_devi_open_count(sp, &dip);
2645 
2646 	if (count <= 0)
2647 		return (DDI_WALK_CONTINUE);
2648 
2649 	ASSERT(dip);
2650 
2651 	if (mod_hash_remove(bsp->s_hash, dip, (mod_hash_val_t *)&sbusy))
2652 		sbusy = count;
2653 	else
2654 		sbusy += count;
2655 
2656 	if (mod_hash_insert(bsp->s_hash, dip, (mod_hash_val_t)sbusy)) {
2657 		cmn_err(CE_WARN, "%s: s_hash insert failed: dip=0x%p, "
2658 		    "sbusy = %lu", "e_ddi_branch_referenced",
2659 		    (void *)dip, sbusy);
2660 	}
2661 
2662 	bsp->s_total += count;
2663 
2664 	return (DDI_WALK_CONTINUE);
2665 }
2666 
2667 static void
2668 visit_dvnode(struct dv_node *dv, void *arg)
2669 {
2670 	uintptr_t dvbusy;
2671 	uint_t count;
2672 	struct vnode *vp;
2673 	struct devi_busy *bsp = arg;
2674 
2675 	ASSERT(dv && dv->dv_devi);
2676 
2677 	vp = DVTOV(dv);
2678 
2679 	mutex_enter(&vp->v_lock);
2680 	count = vp->v_count;
2681 	mutex_exit(&vp->v_lock);
2682 
2683 	if (!count)
2684 		return;
2685 
2686 	if (mod_hash_remove(bsp->dv_hash, dv->dv_devi,
2687 	    (mod_hash_val_t *)&dvbusy))
2688 		dvbusy = count;
2689 	else
2690 		dvbusy += count;
2691 
2692 	if (mod_hash_insert(bsp->dv_hash, dv->dv_devi,
2693 	    (mod_hash_val_t)dvbusy)) {
2694 		cmn_err(CE_WARN, "%s: dv_hash insert failed: dip=0x%p, "
2695 		    "dvbusy=%lu", "e_ddi_branch_referenced",
2696 		    (void *)dv->dv_devi, dvbusy);
2697 	}
2698 
2699 	bsp->dv_total += count;
2700 }
2701 
2702 /*
2703  * Returns reference count on success or -1 on failure.
2704  */
2705 int
2706 e_ddi_branch_referenced(
2707 	dev_info_t *rdip,
2708 	int (*callback)(dev_info_t *dip, void *arg, uint_t ref),
2709 	void *arg)
2710 {
2711 	int circ;
2712 	char *path;
2713 	dev_info_t *pdip;
2714 	struct devi_busy bsa = {0};
2715 
2716 	ASSERT(rdip);
2717 
2718 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2719 
2720 	ndi_hold_devi(rdip);
2721 
2722 	pdip = ddi_get_parent(rdip);
2723 
2724 	ASSERT(pdip);
2725 
2726 	/*
2727 	 * Check if caller holds pdip busy - can cause deadlocks during
2728 	 * devfs_walk()
2729 	 */
2730 	if (!e_ddi_branch_held(rdip) || DEVI_BUSY_OWNED(pdip)) {
2731 		cmn_err(CE_WARN, "e_ddi_branch_referenced: failed: "
2732 		    "devinfo branch(%p) not held or parent busy held",
2733 		    (void *)rdip);
2734 		ndi_rele_devi(rdip);
2735 		kmem_free(path, MAXPATHLEN);
2736 		return (-1);
2737 	}
2738 
2739 	ndi_devi_enter(pdip, &circ);
2740 	(void) ddi_pathname(rdip, path);
2741 	ndi_devi_exit(pdip, circ);
2742 
2743 	bsa.dv_hash = mod_hash_create_ptrhash("dv_node busy hash", NUMCHAINS,
2744 	    mod_hash_null_valdtor, sizeof (struct dev_info));
2745 
2746 	bsa.s_hash = mod_hash_create_ptrhash("snode busy hash", NUMCHAINS,
2747 	    mod_hash_null_valdtor, sizeof (struct snode));
2748 
2749 	if (devfs_walk(path, visit_dvnode, &bsa)) {
2750 		cmn_err(CE_WARN, "e_ddi_branch_referenced: "
2751 		    "devfs walk failed for: %s", path);
2752 		kmem_free(path, MAXPATHLEN);
2753 		bsa.s_total = bsa.dv_total = -1;
2754 		goto out;
2755 	}
2756 
2757 	kmem_free(path, MAXPATHLEN);
2758 
2759 	/*
2760 	 * Walk the snode table to detect device opens, which are currently
2761 	 * maintained on specfs common snodes.
2762 	 */
2763 	spec_snode_walk(visit_snode, &bsa);
2764 
2765 	if (callback == NULL)
2766 		goto out;
2767 
2768 	bsa.callback = callback;
2769 	bsa.arg = arg;
2770 
2771 	if (visit_dip(rdip, &bsa) == DDI_WALK_CONTINUE) {
2772 		ndi_devi_enter(rdip, &circ);
2773 		ddi_walk_devs(ddi_get_child(rdip), visit_dip, &bsa);
2774 		ndi_devi_exit(rdip, circ);
2775 	}
2776 
2777 out:
2778 	ndi_rele_devi(rdip);
2779 	mod_hash_destroy_ptrhash(bsa.s_hash);
2780 	mod_hash_destroy_ptrhash(bsa.dv_hash);
2781 	return (bsa.s_total > bsa.dv_total ? bsa.s_total : bsa.dv_total);
2782 }
2783