xref: /titanic_52/usr/src/uts/i86pc/io/pci/pci_tools.c (revision a60fc142342386d0b786e65fba901234400d7020)
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 2007 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/mkdev.h>
30 #include <sys/stat.h>
31 #include <sys/sunddi.h>
32 #include <vm/seg_kmem.h>
33 #include <sys/machparam.h>
34 #include <sys/sunndi.h>
35 #include <sys/ontrap.h>
36 #include <sys/psm.h>
37 #include <sys/pcie.h>
38 #include <sys/hotplug/pci/pcihp.h>
39 #include <sys/pci_cfgspace.h>
40 #include <sys/pci_tools.h>
41 #include <io/pci/pci_tools_ext.h>
42 #include <sys/apic.h>
43 #include <io/pci/pci_var.h>
44 #include <sys/promif.h>
45 #include <sys/x86_archext.h>
46 #include <sys/cpuvar.h>
47 
48 #define	PCIEX_BDF_OFFSET_DELTA	4
49 #define	PCIEX_REG_FUNC_SHIFT	(PCI_REG_FUNC_SHIFT + PCIEX_BDF_OFFSET_DELTA)
50 #define	PCIEX_REG_DEV_SHIFT	(PCI_REG_DEV_SHIFT + PCIEX_BDF_OFFSET_DELTA)
51 #define	PCIEX_REG_BUS_SHIFT	(PCI_REG_BUS_SHIFT + PCIEX_BDF_OFFSET_DELTA)
52 
53 #define	SUCCESS	0
54 
55 int pcitool_debug = 0;
56 
57 /*
58  * Offsets of BARS in config space.  First entry of 0 means config space.
59  * Entries here correlate to pcitool_bars_t enumerated type.
60  */
61 static uint8_t pci_bars[] = {
62 	0x0,
63 	PCI_CONF_BASE0,
64 	PCI_CONF_BASE1,
65 	PCI_CONF_BASE2,
66 	PCI_CONF_BASE3,
67 	PCI_CONF_BASE4,
68 	PCI_CONF_BASE5,
69 	PCI_CONF_ROM
70 };
71 
72 /* Max offset allowed into config space for a particular device. */
73 static uint64_t max_cfg_size = PCI_CONF_HDR_SIZE;
74 
75 static uint64_t pcitool_swap_endian(uint64_t data, int size);
76 static int pcitool_pciex_cfg_access(dev_info_t *dip, pcitool_reg_t *prg,
77     boolean_t write_flag);
78 static int pcitool_cfg_access(dev_info_t *dip, pcitool_reg_t *prg,
79     boolean_t write_flag);
80 static int pcitool_io_access(dev_info_t *dip, pcitool_reg_t *prg,
81     boolean_t write_flag);
82 static int pcitool_mem_access(dev_info_t *dip, pcitool_reg_t *prg,
83     uint64_t virt_addr, boolean_t write_flag);
84 static uint64_t pcitool_map(uint64_t phys_addr, size_t size, size_t *num_pages);
85 static void pcitool_unmap(uint64_t virt_addr, size_t num_pages);
86 
87 /* Extern declarations */
88 extern int	(*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *,
89 		    psm_intr_op_t, int *);
90 
91 int
92 pcitool_init(dev_info_t *dip, boolean_t is_pciex)
93 {
94 	int instance = ddi_get_instance(dip);
95 
96 	/* Create pcitool nodes for register access and interrupt routing. */
97 
98 	if (ddi_create_minor_node(dip, PCI_MINOR_REG, S_IFCHR,
99 	    PCIHP_AP_MINOR_NUM(instance, PCI_TOOL_REG_MINOR_NUM),
100 	    DDI_NT_REGACC, 0) != DDI_SUCCESS) {
101 		return (DDI_FAILURE);
102 	}
103 
104 	if (ddi_create_minor_node(dip, PCI_MINOR_INTR, S_IFCHR,
105 	    PCIHP_AP_MINOR_NUM(instance, PCI_TOOL_INTR_MINOR_NUM),
106 	    DDI_NT_INTRCTL, 0) != DDI_SUCCESS) {
107 		ddi_remove_minor_node(dip, PCI_MINOR_REG);
108 		return (DDI_FAILURE);
109 	}
110 
111 	if (is_pciex)
112 		max_cfg_size = PCIE_CONF_HDR_SIZE;
113 
114 	return (DDI_SUCCESS);
115 }
116 
117 void
118 pcitool_uninit(dev_info_t *dip)
119 {
120 	ddi_remove_minor_node(dip, PCI_MINOR_INTR);
121 	ddi_remove_minor_node(dip, PCI_MINOR_REG);
122 }
123 
124 /*ARGSUSED*/
125 static int
126 pcitool_set_intr(dev_info_t *dip, void *arg, int mode)
127 {
128 	ddi_intr_handle_impl_t info_hdl;
129 	pcitool_intr_set_t iset;
130 	uint32_t old_cpu;
131 	int ret, result;
132 	size_t copyinout_size;
133 	int rval = SUCCESS;
134 
135 	/* Version 1 of pcitool_intr_set_t doesn't have flags. */
136 	copyinout_size = (size_t)&iset.flags - (size_t)&iset;
137 
138 	if (ddi_copyin(arg, &iset, copyinout_size, mode) != DDI_SUCCESS)
139 		return (EFAULT);
140 
141 	switch (iset.user_version) {
142 	case PCITOOL_V1:
143 		break;
144 
145 	case PCITOOL_V2:
146 		copyinout_size = sizeof (pcitool_intr_set_t);
147 		if (ddi_copyin(arg, &iset, copyinout_size, mode) != DDI_SUCCESS)
148 			return (EFAULT);
149 		break;
150 
151 	default:
152 		iset.status = PCITOOL_OUT_OF_RANGE;
153 		rval = ENOTSUP;
154 		goto done_set_intr;
155 	}
156 
157 	if (iset.ino > APIC_MAX_VECTOR) {
158 		rval = EINVAL;
159 		iset.status = PCITOOL_INVALID_INO;
160 		goto done_set_intr;
161 	}
162 
163 	iset.status = PCITOOL_SUCCESS;
164 
165 	if ((old_cpu = pci_get_cpu_from_vecirq(iset.ino, IS_VEC)) == -1) {
166 		iset.status = PCITOOL_IO_ERROR;
167 		rval = EINVAL;
168 		goto done_set_intr;
169 	}
170 
171 
172 	old_cpu &= ~PSMGI_CPU_USER_BOUND;
173 
174 	/*
175 	 * For this locally-declared and used handle, ih_private will contain a
176 	 * CPU value, not an ihdl_plat_t as used for global interrupt handling.
177 	 */
178 	info_hdl.ih_vector = iset.ino;
179 	info_hdl.ih_private = (void *)(uintptr_t)iset.cpu_id;
180 	if (pcitool_debug)
181 		prom_printf("user version:%d, flags:0x%x\n",
182 		    iset.user_version, iset.flags);
183 
184 	result = ENOTSUP;
185 	if ((iset.user_version >= PCITOOL_V2) &&
186 	    (iset.flags & PCITOOL_INTR_SET_FLAG_GROUP)) {
187 		ret = (*psm_intr_ops)(NULL, &info_hdl, PSM_INTR_OP_GRP_SET_CPU,
188 		    &result);
189 	} else {
190 		ret = (*psm_intr_ops)(NULL, &info_hdl, PSM_INTR_OP_SET_CPU,
191 		    &result);
192 	}
193 
194 	if (ret != PSM_SUCCESS) {
195 		switch (result) {
196 		case EIO:		/* Error making the change */
197 			rval = EIO;
198 			iset.status = PCITOOL_IO_ERROR;
199 			break;
200 		case ENXIO:		/* Couldn't convert vector to irq */
201 			rval = EINVAL;
202 			iset.status = PCITOOL_INVALID_INO;
203 			break;
204 		case EINVAL:		/* CPU out of range */
205 			rval = EINVAL;
206 			iset.status = PCITOOL_INVALID_CPUID;
207 			break;
208 		case ENOTSUP:		/* Requested PSM intr ops missing */
209 			rval = ENOTSUP;
210 			iset.status = PCITOOL_IO_ERROR;
211 			break;
212 		}
213 	}
214 
215 	/* Return original CPU. */
216 	iset.cpu_id = old_cpu;
217 
218 done_set_intr:
219 	iset.drvr_version = PCITOOL_VERSION;
220 	if (ddi_copyout(&iset, arg, copyinout_size, mode) != DDI_SUCCESS)
221 		rval = EFAULT;
222 	return (rval);
223 }
224 
225 
226 /* It is assumed that dip != NULL */
227 static void
228 pcitool_get_intr_dev_info(dev_info_t *dip, pcitool_intr_dev_t *devs)
229 {
230 	(void) strncpy(devs->driver_name,
231 	    ddi_driver_name(dip), MAXMODCONFNAME-1);
232 	devs->driver_name[MAXMODCONFNAME] = '\0';
233 	(void) ddi_pathname(dip, devs->path);
234 	devs->dev_inst = ddi_get_instance(dip);
235 }
236 
237 
238 /*ARGSUSED*/
239 static int
240 pcitool_get_intr(dev_info_t *dip, void *arg, int mode)
241 {
242 	/* Array part isn't used here, but oh well... */
243 	pcitool_intr_get_t partial_iget;
244 	pcitool_intr_get_t *iget = &partial_iget;
245 	size_t	iget_kmem_alloc_size = 0;
246 	uint8_t num_devs_ret;
247 	int copyout_rval;
248 	int rval = SUCCESS;
249 	int circ;
250 	int i;
251 
252 	ddi_intr_handle_impl_t info_hdl;
253 	apic_get_intr_t intr_info;
254 
255 	/* Read in just the header part, no array section. */
256 	if (ddi_copyin(arg, &partial_iget, PCITOOL_IGET_SIZE(0), mode) !=
257 	    DDI_SUCCESS)
258 		return (EFAULT);
259 
260 	/* Validate argument. */
261 	if (partial_iget.ino > APIC_MAX_VECTOR) {
262 		partial_iget.status = PCITOOL_INVALID_INO;
263 		partial_iget.num_devs_ret = 0;
264 		rval = EINVAL;
265 		goto done_get_intr;
266 	}
267 
268 	num_devs_ret = partial_iget.num_devs_ret;
269 	intr_info.avgi_dip_list = NULL;
270 	intr_info.avgi_req_flags =
271 	    PSMGI_REQ_CPUID | PSMGI_REQ_NUM_DEVS | PSMGI_INTRBY_VEC;
272 	/*
273 	 * For this locally-declared and used handle, ih_private will contain a
274 	 * pointer to apic_get_intr_t, not an ihdl_plat_t as used for
275 	 * global interrupt handling.
276 	 */
277 	info_hdl.ih_private = &intr_info;
278 	info_hdl.ih_vector = partial_iget.ino;
279 
280 	/* Caller wants device information returned. */
281 	if (num_devs_ret > 0) {
282 
283 		intr_info.avgi_req_flags |= PSMGI_REQ_GET_DEVS;
284 
285 		/*
286 		 * Allocate room.
287 		 * If num_devs_ret == 0 iget remains pointing to partial_iget.
288 		 */
289 		iget_kmem_alloc_size = PCITOOL_IGET_SIZE(num_devs_ret);
290 		iget = kmem_alloc(iget_kmem_alloc_size, KM_SLEEP);
291 
292 		/* Read in whole structure to verify there's room. */
293 		if (ddi_copyin(arg, iget, iget_kmem_alloc_size, mode) !=
294 		    SUCCESS) {
295 
296 			/* Be consistent and just return EFAULT here. */
297 			kmem_free(iget, iget_kmem_alloc_size);
298 
299 			return (EFAULT);
300 		}
301 	}
302 
303 	bzero(iget, PCITOOL_IGET_SIZE(num_devs_ret));
304 	iget->ino = info_hdl.ih_vector;
305 
306 	/*
307 	 * Lock device tree branch from the pci root nexus on down if info will
308 	 * be extracted from dips returned from the tree.
309 	 */
310 	if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
311 		ndi_devi_enter(dip, &circ);
312 	}
313 
314 	/* Call psm_intr_ops(PSM_INTR_OP_GET_INTR) to get information. */
315 	if ((rval = (*psm_intr_ops)(NULL, &info_hdl,
316 	    PSM_INTR_OP_GET_INTR, NULL)) != PSM_SUCCESS) {
317 		iget->status = PCITOOL_IO_ERROR;
318 		iget->num_devs_ret = 0;
319 		rval = EINVAL;
320 		goto done_get_intr;
321 	}
322 
323 	/*
324 	 * Fill in the pcitool_intr_get_t to be returned,
325 	 * with the CPU, num_devs_ret and num_devs.
326 	 */
327 	iget->cpu_id = intr_info.avgi_cpu_id & ~PSMGI_CPU_USER_BOUND;
328 
329 	/* Number of devices returned by apic. */
330 	iget->num_devs = intr_info.avgi_num_devs;
331 
332 	/* Device info was returned. */
333 	if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
334 
335 		/*
336 		 * num devs returned is num devs ret by apic,
337 		 * space permitting.
338 		 */
339 		iget->num_devs_ret = min(num_devs_ret, intr_info.avgi_num_devs);
340 
341 		/*
342 		 * Loop thru list of dips and extract driver, name and instance.
343 		 * Fill in the pcitool_intr_dev_t's with this info.
344 		 */
345 		for (i = 0; i < iget->num_devs_ret; i++)
346 			pcitool_get_intr_dev_info(intr_info.avgi_dip_list[i],
347 			    &iget->dev[i]);
348 
349 		/* Free kmem_alloc'ed memory of the apic_get_intr_t */
350 		kmem_free(intr_info.avgi_dip_list,
351 		    intr_info.avgi_num_devs * sizeof (dev_info_t *));
352 	}
353 
354 done_get_intr:
355 
356 	if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
357 		ndi_devi_exit(dip, circ);
358 	}
359 
360 	iget->drvr_version = PCITOOL_VERSION;
361 	copyout_rval = ddi_copyout(iget, arg,
362 	    PCITOOL_IGET_SIZE(num_devs_ret), mode);
363 
364 	if (iget_kmem_alloc_size > 0)
365 		kmem_free(iget, iget_kmem_alloc_size);
366 
367 	if (copyout_rval != DDI_SUCCESS)
368 		rval = EFAULT;
369 
370 	return (rval);
371 }
372 
373 /*ARGSUSED*/
374 static int
375 pcitool_intr_info(dev_info_t *dip, void *arg, int mode)
376 {
377 	pcitool_intr_info_t intr_info;
378 	ddi_intr_handle_impl_t info_hdl;
379 	int rval = SUCCESS;
380 
381 	/* If we need user_version, and to ret same user version as passed in */
382 	if (ddi_copyin(arg, &intr_info, sizeof (pcitool_intr_info_t), mode) !=
383 	    DDI_SUCCESS) {
384 		if (pcitool_debug)
385 			prom_printf("Error reading arguments\n");
386 		return (EFAULT);
387 	}
388 
389 	/* For UPPC systems, psm_intr_ops has no entry for APIC_TYPE. */
390 	if ((rval = (*psm_intr_ops)(NULL, &info_hdl,
391 	    PSM_INTR_OP_APIC_TYPE, NULL)) != PSM_SUCCESS) {
392 		intr_info.ctlr_type = PCITOOL_CTLR_TYPE_UPPC;
393 		intr_info.ctlr_version = 0;
394 
395 	} else {
396 		intr_info.ctlr_version = (uint32_t)info_hdl.ih_ver;
397 		if (strcmp((char *)info_hdl.ih_private,
398 		    APIC_PCPLUSMP_NAME) == 0)
399 			intr_info.ctlr_type = PCITOOL_CTLR_TYPE_PCPLUSMP;
400 		else
401 			intr_info.ctlr_type = PCITOOL_CTLR_TYPE_UNKNOWN;
402 	}
403 
404 	intr_info.num_intr = APIC_MAX_VECTOR;
405 	intr_info.drvr_version = PCITOOL_VERSION;
406 	if (ddi_copyout(&intr_info, arg, sizeof (pcitool_intr_info_t), mode) !=
407 	    DDI_SUCCESS) {
408 		if (pcitool_debug)
409 			prom_printf("Error returning arguments.\n");
410 		rval = EFAULT;
411 	}
412 
413 	return (rval);
414 }
415 
416 
417 
418 /*
419  * Main function for handling interrupt CPU binding requests and queries.
420  * Need to implement later
421  */
422 /*ARGSUSED*/
423 int
424 pcitool_intr_admn(dev_info_t *dip, void *arg, int cmd, int mode)
425 {
426 	int rval;
427 
428 	switch (cmd) {
429 
430 	/* Associate a new CPU with a given vector */
431 	case PCITOOL_DEVICE_SET_INTR:
432 		rval = pcitool_set_intr(dip, arg, mode);
433 		break;
434 
435 	case PCITOOL_DEVICE_GET_INTR:
436 		rval = pcitool_get_intr(dip, arg, mode);
437 		break;
438 
439 	case PCITOOL_SYSTEM_INTR_INFO:
440 		rval = pcitool_intr_info(dip, arg, mode);
441 		break;
442 
443 	default:
444 		rval = ENOTSUP;
445 	}
446 
447 	return (rval);
448 }
449 
450 
451 /*
452  * A note about ontrap handling:
453  *
454  * X86 systems on which this module was tested return FFs instead of bus errors
455  * when accessing devices with invalid addresses.  Ontrap handling, which
456  * gracefully handles kernel bus errors, is installed anyway, in case future
457  * X86 platforms require it.
458  */
459 
460 /*
461  * Perform register accesses on the nexus device itself.
462  * No explicit PCI nexus device for X86, so not applicable.
463  */
464 
465 /*ARGSUSED*/
466 int
467 pcitool_bus_reg_ops(dev_info_t *dip, void *arg, int cmd, int mode)
468 {
469 	return (ENOTSUP);
470 }
471 
472 /* Swap endianness. */
473 static uint64_t
474 pcitool_swap_endian(uint64_t data, int size)
475 {
476 	typedef union {
477 		uint64_t data64;
478 		uint8_t data8[8];
479 	} data_split_t;
480 
481 	data_split_t orig_data;
482 	data_split_t returned_data;
483 	int i;
484 
485 	orig_data.data64 = data;
486 	returned_data.data64 = 0;
487 
488 	for (i = 0; i < size; i++) {
489 		returned_data.data8[i] = orig_data.data8[size - 1 - i];
490 	}
491 
492 	return (returned_data.data64);
493 }
494 
495 
496 /*
497  * Access device.  prg is modified.
498  *
499  * Extended config space is available only through memory-mapped access.
500  * Standard config space on pci express devices is available either way,
501  * so do it memory-mapped here too, for simplicity.
502  */
503 /*ARGSUSED*/
504 static int
505 pcitool_pciex_cfg_access(dev_info_t *dip, pcitool_reg_t *prg,
506     boolean_t write_flag)
507 {
508 	int rval = SUCCESS;
509 	uint64_t virt_addr;
510 	size_t	num_virt_pages;
511 
512 	prg->status = PCITOOL_SUCCESS;
513 
514 	prg->phys_addr = ddi_prop_get_int64(DDI_DEV_T_ANY, dip, 0,
515 	    "ecfga-base-address", 0);
516 	if (prg->phys_addr == 0) {
517 		prg->status = PCITOOL_IO_ERROR;
518 		return (EIO);
519 	}
520 
521 	prg->phys_addr += prg->offset +
522 	    ((prg->bus_no << PCIEX_REG_BUS_SHIFT) |
523 	    (prg->dev_no << PCIEX_REG_DEV_SHIFT) |
524 	    (prg->func_no << PCIEX_REG_FUNC_SHIFT));
525 
526 	virt_addr = pcitool_map(prg->phys_addr,
527 	    PCITOOL_ACC_ATTR_SIZE(prg->acc_attr), &num_virt_pages);
528 	if (virt_addr == NULL) {
529 		prg->status = PCITOOL_IO_ERROR;
530 		return (EIO);
531 	}
532 
533 	rval = pcitool_mem_access(dip, prg, virt_addr, write_flag);
534 	pcitool_unmap(virt_addr, num_virt_pages);
535 	return (rval);
536 }
537 
538 /* Access device.  prg is modified. */
539 /*ARGSUSED*/
540 static int
541 pcitool_cfg_access(dev_info_t *dip, pcitool_reg_t *prg, boolean_t write_flag)
542 {
543 	int size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
544 	boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
545 	int rval = SUCCESS;
546 	uint64_t local_data;
547 
548 	/*
549 	 * NOTE: there is no way to verify whether or not the address is valid.
550 	 * The put functions return void and the get functions return ff on
551 	 * error.
552 	 */
553 	prg->status = PCITOOL_SUCCESS;
554 
555 	if (write_flag) {
556 
557 		if (big_endian) {
558 			local_data = pcitool_swap_endian(prg->data, size);
559 		} else {
560 			local_data = prg->data;
561 		}
562 
563 		switch (size) {
564 		case 1:
565 			(*pci_putb_func)(prg->bus_no, prg->dev_no,
566 			    prg->func_no, prg->offset, local_data);
567 			break;
568 		case 2:
569 			(*pci_putw_func)(prg->bus_no, prg->dev_no,
570 			    prg->func_no, prg->offset, local_data);
571 			break;
572 		case 4:
573 			(*pci_putl_func)(prg->bus_no, prg->dev_no,
574 			    prg->func_no, prg->offset, local_data);
575 			break;
576 		default:
577 			rval = ENOTSUP;
578 			prg->status = PCITOOL_INVALID_SIZE;
579 			break;
580 		}
581 	} else {
582 		switch (size) {
583 		case 1:
584 			local_data = (*pci_getb_func)(prg->bus_no, prg->dev_no,
585 			    prg->func_no, prg->offset);
586 			break;
587 		case 2:
588 			local_data = (*pci_getw_func)(prg->bus_no, prg->dev_no,
589 			    prg->func_no, prg->offset);
590 			break;
591 		case 4:
592 			local_data = (*pci_getl_func)(prg->bus_no, prg->dev_no,
593 			    prg->func_no, prg->offset);
594 			break;
595 		default:
596 			rval = ENOTSUP;
597 			prg->status = PCITOOL_INVALID_SIZE;
598 			break;
599 		}
600 
601 		if (rval == SUCCESS) {
602 			if (big_endian) {
603 				prg->data =
604 				    pcitool_swap_endian(local_data, size);
605 			} else {
606 				prg->data = local_data;
607 			}
608 		}
609 	}
610 	prg->phys_addr = 0;	/* Config space is not memory mapped on X86. */
611 	return (rval);
612 }
613 
614 
615 /*ARGSUSED*/
616 static int
617 pcitool_io_access(dev_info_t *dip, pcitool_reg_t *prg, boolean_t write_flag)
618 {
619 	int port = (int)prg->phys_addr;
620 	size_t size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
621 	boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
622 	int rval = SUCCESS;
623 	on_trap_data_t otd;
624 	uint64_t local_data;
625 
626 
627 	/*
628 	 * on_trap works like setjmp.
629 	 *
630 	 * A non-zero return here means on_trap has returned from an error.
631 	 *
632 	 * A zero return here means that on_trap has just returned from setup.
633 	 */
634 	if (on_trap(&otd, OT_DATA_ACCESS)) {
635 		no_trap();
636 		if (pcitool_debug)
637 			prom_printf(
638 			    "pcitool_io_access: on_trap caught an error...\n");
639 		prg->status = PCITOOL_INVALID_ADDRESS;
640 		return (EFAULT);
641 	}
642 
643 	if (write_flag) {
644 
645 		if (big_endian) {
646 			local_data = pcitool_swap_endian(prg->data, size);
647 		} else {
648 			local_data = prg->data;
649 		}
650 
651 		if (pcitool_debug)
652 			prom_printf("Writing %ld byte(s) to port 0x%x\n",
653 			    size, port);
654 
655 		switch (size) {
656 		case 1:
657 			outb(port, (uint8_t)local_data);
658 			break;
659 		case 2:
660 			outw(port, (uint16_t)local_data);
661 			break;
662 		case 4:
663 			outl(port, (uint32_t)local_data);
664 			break;
665 		default:
666 			rval = ENOTSUP;
667 			prg->status = PCITOOL_INVALID_SIZE;
668 			break;
669 		}
670 	} else {
671 		if (pcitool_debug)
672 			prom_printf("Reading %ld byte(s) from port 0x%x\n",
673 			    size, port);
674 
675 		switch (size) {
676 		case 1:
677 			local_data = inb(port);
678 			break;
679 		case 2:
680 			local_data = inw(port);
681 			break;
682 		case 4:
683 			local_data = inl(port);
684 			break;
685 		default:
686 			rval = ENOTSUP;
687 			prg->status = PCITOOL_INVALID_SIZE;
688 			break;
689 		}
690 
691 		if (rval == SUCCESS) {
692 			if (big_endian) {
693 				prg->data =
694 				    pcitool_swap_endian(local_data, size);
695 			} else {
696 				prg->data = local_data;
697 			}
698 		}
699 	}
700 
701 	no_trap();
702 	return (rval);
703 }
704 
705 /*ARGSUSED*/
706 static int
707 pcitool_mem_access(dev_info_t *dip, pcitool_reg_t *prg, uint64_t virt_addr,
708 	boolean_t write_flag)
709 {
710 	size_t size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
711 	boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
712 	int rval = DDI_SUCCESS;
713 	on_trap_data_t otd;
714 	uint64_t local_data;
715 
716 	/*
717 	 * on_trap works like setjmp.
718 	 *
719 	 * A non-zero return here means on_trap has returned from an error.
720 	 *
721 	 * A zero return here means that on_trap has just returned from setup.
722 	 */
723 	if (on_trap(&otd, OT_DATA_ACCESS)) {
724 		no_trap();
725 		if (pcitool_debug)
726 			prom_printf(
727 			    "pcitool_mem_access: on_trap caught an error...\n");
728 		prg->status = PCITOOL_INVALID_ADDRESS;
729 		return (EFAULT);
730 	}
731 
732 	if (write_flag) {
733 
734 		if (big_endian) {
735 			local_data = pcitool_swap_endian(prg->data, size);
736 		} else {
737 			local_data = prg->data;
738 		}
739 
740 		switch (size) {
741 		case 1:
742 			*((uint8_t *)(uintptr_t)virt_addr) = local_data;
743 			break;
744 		case 2:
745 			*((uint16_t *)(uintptr_t)virt_addr) = local_data;
746 			break;
747 		case 4:
748 			*((uint32_t *)(uintptr_t)virt_addr) = local_data;
749 			break;
750 		case 8:
751 			*((uint64_t *)(uintptr_t)virt_addr) = local_data;
752 			break;
753 		default:
754 			rval = ENOTSUP;
755 			prg->status = PCITOOL_INVALID_SIZE;
756 			break;
757 		}
758 	} else {
759 		switch (size) {
760 		case 1:
761 			local_data = *((uint8_t *)(uintptr_t)virt_addr);
762 			break;
763 		case 2:
764 			local_data = *((uint16_t *)(uintptr_t)virt_addr);
765 			break;
766 		case 4:
767 			local_data = *((uint32_t *)(uintptr_t)virt_addr);
768 			break;
769 		case 8:
770 			local_data = *((uint64_t *)(uintptr_t)virt_addr);
771 			break;
772 		default:
773 			rval = ENOTSUP;
774 			prg->status = PCITOOL_INVALID_SIZE;
775 			break;
776 		}
777 
778 		if (rval == SUCCESS) {
779 			if (big_endian) {
780 				prg->data =
781 				    pcitool_swap_endian(local_data, size);
782 			} else {
783 				prg->data = local_data;
784 			}
785 		}
786 	}
787 
788 	no_trap();
789 	return (rval);
790 }
791 
792 /*
793  * Map up to 2 pages which contain the address we want to access.
794  *
795  * Mapping should span no more than 8 bytes.  With X86 it is possible for an
796  * 8 byte value to start on a 4 byte boundary, so it can cross a page boundary.
797  * We'll never have to map more than two pages.
798  */
799 
800 static uint64_t
801 pcitool_map(uint64_t phys_addr, size_t size, size_t *num_pages)
802 {
803 
804 	uint64_t page_base = phys_addr & ~MMU_PAGEOFFSET;
805 	uint64_t offset = phys_addr & MMU_PAGEOFFSET;
806 	void *virt_base;
807 	uint64_t returned_addr;
808 	pfn_t pfn;
809 
810 	if (pcitool_debug)
811 		prom_printf("pcitool_map: Called with PA:0x%p\n",
812 		    (uint8_t *)(uintptr_t)phys_addr);
813 
814 	*num_pages = 1;
815 
816 	/* Desired mapping would span more than two pages. */
817 	if ((offset + size) > (MMU_PAGESIZE * 2)) {
818 		if (pcitool_debug)
819 			prom_printf("boundary violation: "
820 			    "offset:0x%" PRIx64 ", size:%ld, pagesize:0x%lx\n",
821 			    offset, (uintptr_t)size, (uintptr_t)MMU_PAGESIZE);
822 		return (NULL);
823 
824 	} else if ((offset + size) > MMU_PAGESIZE) {
825 		(*num_pages)++;
826 	}
827 
828 	/* Get page(s) of virtual space. */
829 	virt_base = vmem_alloc(heap_arena, ptob(*num_pages), VM_NOSLEEP);
830 	if (virt_base == NULL) {
831 		if (pcitool_debug)
832 			prom_printf("Couldn't get virtual base address.\n");
833 		return (NULL);
834 	}
835 
836 	if (pcitool_debug)
837 		prom_printf("Got base virtual address:0x%p\n", virt_base);
838 
839 	pfn = btop(page_base);
840 
841 	/* Now map the allocated virtual space to the physical address. */
842 	hat_devload(kas.a_hat, virt_base, mmu_ptob(*num_pages), pfn,
843 	    PROT_READ | PROT_WRITE | HAT_STRICTORDER,
844 	    HAT_LOAD_LOCK);
845 
846 	returned_addr = ((uintptr_t)(virt_base)) + offset;
847 
848 	if (pcitool_debug)
849 		prom_printf("pcitool_map: returning VA:0x%p\n",
850 		    (void *)(uintptr_t)returned_addr);
851 
852 	return (returned_addr);
853 }
854 
855 /* Unmap the mapped page(s). */
856 static void
857 pcitool_unmap(uint64_t virt_addr, size_t num_pages)
858 {
859 	void *base_virt_addr = (void *)(uintptr_t)(virt_addr & ~MMU_PAGEOFFSET);
860 
861 	hat_unload(kas.a_hat, base_virt_addr, ptob(num_pages),
862 	    HAT_UNLOAD_UNLOCK);
863 	vmem_free(heap_arena, base_virt_addr, ptob(num_pages));
864 }
865 
866 
867 /* Perform register accesses on PCI leaf devices. */
868 int
869 pcitool_dev_reg_ops(dev_info_t *dip, void *arg, int cmd, int mode)
870 {
871 	boolean_t	write_flag = B_FALSE;
872 	int		rval = 0;
873 	pcitool_reg_t	prg;
874 	uint8_t		size;
875 
876 	uint64_t	base_addr;
877 	uint64_t	virt_addr;
878 	size_t		num_virt_pages;
879 
880 	switch (cmd) {
881 	case (PCITOOL_DEVICE_SET_REG):
882 		write_flag = B_TRUE;
883 
884 	/*FALLTHRU*/
885 	case (PCITOOL_DEVICE_GET_REG):
886 		if (pcitool_debug)
887 			prom_printf("pci_dev_reg_ops set/get reg\n");
888 		if (ddi_copyin(arg, &prg, sizeof (pcitool_reg_t), mode) !=
889 		    DDI_SUCCESS) {
890 			if (pcitool_debug)
891 				prom_printf("Error reading arguments\n");
892 			return (EFAULT);
893 		}
894 
895 		if (prg.barnum >= (sizeof (pci_bars) / sizeof (pci_bars[0]))) {
896 			prg.status = PCITOOL_OUT_OF_RANGE;
897 			rval = EINVAL;
898 			goto done_reg;
899 		}
900 
901 		if (pcitool_debug)
902 			prom_printf("raw bus:0x%x, dev:0x%x, func:0x%x\n",
903 			    prg.bus_no, prg.dev_no, prg.func_no);
904 		/* Validate address arguments of bus / dev / func */
905 		if (((prg.bus_no &
906 		    (PCI_REG_BUS_M >> PCI_REG_BUS_SHIFT)) !=
907 		    prg.bus_no) ||
908 		    ((prg.dev_no &
909 		    (PCI_REG_DEV_M >> PCI_REG_DEV_SHIFT)) !=
910 		    prg.dev_no) ||
911 		    ((prg.func_no &
912 		    (PCI_REG_FUNC_M >> PCI_REG_FUNC_SHIFT)) !=
913 		    prg.func_no)) {
914 			prg.status = PCITOOL_INVALID_ADDRESS;
915 			rval = EINVAL;
916 			goto done_reg;
917 		}
918 
919 		size = PCITOOL_ACC_ATTR_SIZE(prg.acc_attr);
920 
921 		/* Proper config space desired. */
922 		if (prg.barnum == 0) {
923 
924 			if (pcitool_debug)
925 				prom_printf(
926 				    "config access: offset:0x%" PRIx64 ", "
927 				    "phys_addr:0x%" PRIx64 "\n",
928 				    prg.offset, prg.phys_addr);
929 
930 			if (prg.offset >= max_cfg_size) {
931 				prg.status = PCITOOL_OUT_OF_RANGE;
932 				rval = EINVAL;
933 				goto done_reg;
934 			}
935 
936 			/*
937 			 * Access device.  prg is modified.
938 			 * First, check for AMD northbridges for I/O access
939 			 * (This fix will move in future to pcitool user-land)
940 			 * Next, check for PCIe devices and do
941 			 * memory-mapped access
942 			 * Lastly, check for PCI devices and do I/O access
943 			 */
944 			if ((prg.bus_no == 0) &&
945 			    (prg.dev_no >= 0x18) &&
946 			    (prg.dev_no < (0x18 + ncpus))) {
947 				if (cpuid_getvendor(CPU) == X86_VENDOR_AMD)
948 					rval = pcitool_cfg_access(dip, &prg,
949 					    write_flag);
950 			} else if (max_cfg_size == PCIE_CONF_HDR_SIZE)
951 				rval = pcitool_pciex_cfg_access(dip, &prg,
952 				    write_flag);
953 			else
954 				rval = pcitool_cfg_access(dip, &prg,
955 				    write_flag);
956 
957 			if (pcitool_debug)
958 				prom_printf(
959 				    "config access: data:0x%" PRIx64 "\n",
960 				    prg.data);
961 
962 		/* IO/ MEM/ MEM64 space. */
963 		} else {
964 
965 			pcitool_reg_t	prg2;
966 			bcopy(&prg, &prg2, sizeof (pcitool_reg_t));
967 
968 			/*
969 			 * Translate BAR number into offset of the BAR in
970 			 * the device's config space.
971 			 */
972 			prg2.offset = pci_bars[prg2.barnum];
973 			prg2.acc_attr =
974 			    PCITOOL_ACC_ATTR_SIZE_4 | PCITOOL_ACC_ATTR_ENDN_LTL;
975 
976 			if (pcitool_debug)
977 				prom_printf(
978 				    "barnum:%d, bar_offset:0x%" PRIx64 "\n",
979 				    prg2.barnum, prg2.offset);
980 			/*
981 			 * Get Bus Address Register (BAR) from config space.
982 			 * prg2.offset is the offset into config space of the
983 			 * BAR desired.  prg.status is modified on error.
984 			 */
985 			rval = pcitool_cfg_access(dip, &prg2, B_FALSE);
986 			if (rval != SUCCESS) {
987 				if (pcitool_debug)
988 					prom_printf("BAR access failed\n");
989 				prg.status = prg2.status;
990 				goto done_reg;
991 			}
992 			/*
993 			 * Reference proper PCI space based on the BAR.
994 			 * If 64 bit MEM space, need to load other half of the
995 			 * BAR first.
996 			 */
997 
998 			if (pcitool_debug)
999 				prom_printf("bar returned is 0x%" PRIx64 "\n",
1000 				    prg2.data);
1001 			if (!prg2.data) {
1002 				if (pcitool_debug)
1003 					prom_printf("BAR data == 0\n");
1004 				rval = EINVAL;
1005 				prg.status = PCITOOL_INVALID_ADDRESS;
1006 				goto done_reg;
1007 			}
1008 			if (prg2.data == 0xffffffff) {
1009 				if (pcitool_debug)
1010 					prom_printf("BAR data == -1\n");
1011 				rval = EINVAL;
1012 				prg.status = PCITOOL_INVALID_ADDRESS;
1013 				goto done_reg;
1014 			}
1015 
1016 			/*
1017 			 * BAR has bits saying this space is IO space, unless
1018 			 * this is the ROM address register.
1019 			 */
1020 			if (((PCI_BASE_SPACE_M & prg2.data) ==
1021 			    PCI_BASE_SPACE_IO) &&
1022 			    (prg2.offset != PCI_CONF_ROM)) {
1023 				if (pcitool_debug)
1024 					prom_printf("IO space\n");
1025 
1026 				prg2.data &= PCI_BASE_IO_ADDR_M;
1027 				prg.phys_addr = prg2.data + prg.offset;
1028 
1029 				rval = pcitool_io_access(dip, &prg, write_flag);
1030 				if ((rval != SUCCESS) && (pcitool_debug))
1031 					prom_printf("IO access failed\n");
1032 
1033 				goto done_reg;
1034 
1035 
1036 			/*
1037 			 * BAR has bits saying this space is 64 bit memory
1038 			 * space, unless this is the ROM address register.
1039 			 *
1040 			 * The 64 bit address stored in two BAR cells is not
1041 			 * necessarily aligned on an 8-byte boundary.
1042 			 * Need to keep the first 4 bytes read,
1043 			 * and do a separate read of the high 4 bytes.
1044 			 */
1045 
1046 			} else if ((PCI_BASE_TYPE_ALL & prg2.data) &&
1047 			    (prg2.offset != PCI_CONF_ROM)) {
1048 
1049 				uint32_t low_bytes =
1050 				    (uint32_t)(prg2.data & ~PCI_BASE_TYPE_ALL);
1051 
1052 				/*
1053 				 * Don't try to read the next 4 bytes
1054 				 * past the end of BARs.
1055 				 */
1056 				if (prg2.offset >= PCI_CONF_BASE5) {
1057 					prg.status = PCITOOL_OUT_OF_RANGE;
1058 					rval = EIO;
1059 					goto done_reg;
1060 				}
1061 
1062 				/*
1063 				 * Access device.
1064 				 * prg2.status is modified on error.
1065 				 */
1066 				prg2.offset += 4;
1067 				rval = pcitool_cfg_access(dip, &prg2, B_FALSE);
1068 				if (rval != SUCCESS) {
1069 					prg.status = prg2.status;
1070 					goto done_reg;
1071 				}
1072 
1073 				if (prg2.data == 0xffffffff) {
1074 					prg.status = PCITOOL_INVALID_ADDRESS;
1075 					prg.status = EFAULT;
1076 					goto done_reg;
1077 				}
1078 
1079 				prg2.data = (prg2.data << 32) + low_bytes;
1080 				if (pcitool_debug)
1081 					prom_printf(
1082 					    "64 bit mem space.  "
1083 					    "64-bit bar is 0x%" PRIx64 "\n",
1084 					    prg2.data);
1085 
1086 			/* Mem32 space, including ROM */
1087 			} else {
1088 
1089 				if (prg2.offset == PCI_CONF_ROM) {
1090 					if (pcitool_debug)
1091 						prom_printf(
1092 						    "Additional ROM "
1093 						    "checking\n");
1094 					/* Can't write to ROM */
1095 					if (write_flag) {
1096 						prg.status = PCITOOL_ROM_WRITE;
1097 						rval = EIO;
1098 						goto done_reg;
1099 
1100 					/* ROM disabled for reading */
1101 					} else if (!(prg2.data & 0x00000001)) {
1102 						prg.status =
1103 						    PCITOOL_ROM_DISABLED;
1104 						rval = EIO;
1105 						goto done_reg;
1106 					}
1107 				}
1108 
1109 				if (pcitool_debug)
1110 					prom_printf("32 bit mem space\n");
1111 			}
1112 
1113 			/* Common code for all IO/MEM range spaces. */
1114 
1115 			base_addr = prg2.data;
1116 			if (pcitool_debug)
1117 				prom_printf(
1118 				    "addr portion of bar is 0x%" PRIx64 ", "
1119 				    "base=0x%" PRIx64 ", "
1120 				    "offset:0x%" PRIx64 "\n",
1121 				    prg2.data, base_addr, prg.offset);
1122 			/*
1123 			 * Use offset provided by caller to index into
1124 			 * desired space, then access.
1125 			 * Note that prg.status is modified on error.
1126 			 */
1127 			prg.phys_addr = base_addr + prg.offset;
1128 
1129 			virt_addr = pcitool_map(prg.phys_addr, size,
1130 			    &num_virt_pages);
1131 			if (virt_addr == NULL) {
1132 				prg.status = PCITOOL_IO_ERROR;
1133 				rval = EIO;
1134 				goto done_reg;
1135 			}
1136 
1137 			rval = pcitool_mem_access(dip, &prg, virt_addr,
1138 			    write_flag);
1139 			pcitool_unmap(virt_addr, num_virt_pages);
1140 		}
1141 done_reg:
1142 		prg.drvr_version = PCITOOL_VERSION;
1143 		if (ddi_copyout(&prg, arg, sizeof (pcitool_reg_t), mode) !=
1144 		    DDI_SUCCESS) {
1145 			if (pcitool_debug)
1146 				prom_printf("Error returning arguments.\n");
1147 			rval = EFAULT;
1148 		}
1149 		break;
1150 	default:
1151 		rval = ENOTTY;
1152 		break;
1153 	}
1154 	return (rval);
1155 }
1156