?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
8702 PCI addresses with physaddr > 0xffffffff can't be mapped in
[unleashed.git] / usr / src / uts / i86pc / io / pci / pci_tools.c
blob26990cf876c163c89263419c04824f54233ee40a
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 #include <sys/sysmacros.h>
26 #include <sys/types.h>
27 #include <sys/mkdev.h>
28 #include <sys/stat.h>
29 #include <sys/sunddi.h>
30 #include <vm/seg_kmem.h>
31 #include <sys/machparam.h>
32 #include <sys/sunndi.h>
33 #include <sys/ontrap.h>
34 #include <sys/psm.h>
35 #include <sys/pcie.h>
36 #include <sys/pci_cfgspace.h>
37 #include <sys/pci_tools.h>
38 #include <io/pci/pci_tools_ext.h>
39 #include <sys/apic.h>
40 #include <sys/apix.h>
41 #include <io/pci/pci_var.h>
42 #include <sys/pci_impl.h>
43 #include <sys/promif.h>
44 #include <sys/x86_archext.h>
45 #include <sys/cpuvar.h>
46 #include <sys/pci_cfgacc.h>
47
48 #ifdef __xpv
49 #include <sys/hypervisor.h>
50 #endif
51
52 #define PCIEX_BDF_OFFSET_DELTA 4
53 #define PCIEX_REG_FUNC_SHIFT (PCI_REG_FUNC_SHIFT + PCIEX_BDF_OFFSET_DELTA)
54 #define PCIEX_REG_DEV_SHIFT (PCI_REG_DEV_SHIFT + PCIEX_BDF_OFFSET_DELTA)
55 #define PCIEX_REG_BUS_SHIFT (PCI_REG_BUS_SHIFT + PCIEX_BDF_OFFSET_DELTA)
56
57 #define SUCCESS 0
58
59 extern uint64_t mcfg_mem_base;
60 int pcitool_debug = 0;
61
62 /*
63 * Offsets of BARS in config space. First entry of 0 means config space.
64 * Entries here correlate to pcitool_bars_t enumerated type.
65 */
66 static uint8_t pci_bars[] = {
67 0x0,
68 PCI_CONF_BASE0,
69 PCI_CONF_BASE1,
70 PCI_CONF_BASE2,
71 PCI_CONF_BASE3,
72 PCI_CONF_BASE4,
73 PCI_CONF_BASE5,
74 PCI_CONF_ROM
75 };
76
77 /* Max offset allowed into config space for a particular device. */
78 static uint64_t max_cfg_size = PCI_CONF_HDR_SIZE;
79
80 static uint64_t pcitool_swap_endian(uint64_t data, int size);
81 static int pcitool_cfg_access(pcitool_reg_t *prg, boolean_t write_flag,
82 boolean_t io_access);
83 static int pcitool_io_access(pcitool_reg_t *prg, boolean_t write_flag);
84 static int pcitool_mem_access(pcitool_reg_t *prg, uint64_t virt_addr,
85 boolean_t write_flag);
86 static uint64_t pcitool_map(uint64_t phys_addr, size_t size, size_t *num_pages);
87 static void pcitool_unmap(uint64_t virt_addr, size_t num_pages);
88
89 /* Extern declarations */
90 extern int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *,
91 psm_intr_op_t, int *);
92
93 int
94 pcitool_init(dev_info_t *dip, boolean_t is_pciex)
95 {
96 int instance = ddi_get_instance(dip);
97
98 /* Create pcitool nodes for register access and interrupt routing. */
99
100 if (ddi_create_minor_node(dip, PCI_MINOR_REG, S_IFCHR,
101 PCI_MINOR_NUM(instance, PCI_TOOL_REG_MINOR_NUM),
102 DDI_NT_REGACC, 0) != DDI_SUCCESS) {
103 return (DDI_FAILURE);
104 }
105
106 if (ddi_create_minor_node(dip, PCI_MINOR_INTR, S_IFCHR,
107 PCI_MINOR_NUM(instance, PCI_TOOL_INTR_MINOR_NUM),
108 DDI_NT_INTRCTL, 0) != DDI_SUCCESS) {
109 ddi_remove_minor_node(dip, PCI_MINOR_REG);
110 return (DDI_FAILURE);
111 }
112
113 if (is_pciex)
114 max_cfg_size = PCIE_CONF_HDR_SIZE;
115
116 return (DDI_SUCCESS);
117 }
118
119 void
120 pcitool_uninit(dev_info_t *dip)
121 {
122 ddi_remove_minor_node(dip, PCI_MINOR_INTR);
123 ddi_remove_minor_node(dip, PCI_MINOR_REG);
124 }
125
126 /*ARGSUSED*/
127 static int
128 pcitool_set_intr(dev_info_t *dip, void *arg, int mode)
129 {
130 ddi_intr_handle_impl_t info_hdl;
131 pcitool_intr_set_t iset;
132 uint32_t old_cpu;
133 int ret, result;
134 size_t copyinout_size;
135 int rval = SUCCESS;
136 apic_get_type_t type_info;
137
138 /* Version 1 of pcitool_intr_set_t doesn't have flags. */
139 copyinout_size = (size_t)&iset.flags - (size_t)&iset;
140
141 if (ddi_copyin(arg, &iset, copyinout_size, mode) != DDI_SUCCESS)
142 return (EFAULT);
143
144 switch (iset.user_version) {
145 case PCITOOL_V1:
146 break;
147
148 case PCITOOL_V2:
149 copyinout_size = sizeof (pcitool_intr_set_t);
150 if (ddi_copyin(arg, &iset, copyinout_size, mode) != DDI_SUCCESS)
151 return (EFAULT);
152 break;
153
154 default:
155 iset.status = PCITOOL_OUT_OF_RANGE;
156 rval = ENOTSUP;
157 goto done_set_intr;
158 }
159
160 if (iset.flags & PCITOOL_INTR_FLAG_SET_MSI) {
161 rval = ENOTSUP;
162 iset.status = PCITOOL_IO_ERROR;
163 goto done_set_intr;
164 }
165
166 info_hdl.ih_private = &type_info;
167
168 if ((*psm_intr_ops)(NULL, &info_hdl,
169 PSM_INTR_OP_APIC_TYPE, NULL) != PSM_SUCCESS) {
170 rval = ENOTSUP;
171 iset.status = PCITOOL_IO_ERROR;
172 goto done_set_intr;
173 }
174
175 if (strcmp(type_info.avgi_type, APIC_APIX_NAME) == 0) {
176 if (iset.old_cpu > type_info.avgi_num_cpu) {
177 rval = EINVAL;
178 iset.status = PCITOOL_INVALID_CPUID;
179 goto done_set_intr;
180 }
181 old_cpu = iset.old_cpu;
182 } else {
183 if ((old_cpu =
184 pci_get_cpu_from_vecirq(iset.ino, IS_VEC)) == -1) {
185 iset.status = PCITOOL_IO_ERROR;
186 rval = EINVAL;
187 goto done_set_intr;
188 }
189 }
190
191 if (iset.ino > type_info.avgi_num_intr) {
192 rval = EINVAL;
193 iset.status = PCITOOL_INVALID_INO;
194 goto done_set_intr;
195 }
196
197 iset.status = PCITOOL_SUCCESS;
198
199 old_cpu &= ~PSMGI_CPU_USER_BOUND;
200
201 /*
202 * For this locally-declared and used handle, ih_private will contain a
203 * CPU value, not an ihdl_plat_t as used for global interrupt handling.
204 */
205 if (strcmp(type_info.avgi_type, APIC_APIX_NAME) == 0) {
206 info_hdl.ih_vector = APIX_VIRTVECTOR(old_cpu, iset.ino);
207 } else {
208 info_hdl.ih_vector = iset.ino;
209 }
210 info_hdl.ih_private = (void *)(uintptr_t)iset.cpu_id;
211 info_hdl.ih_flags = PSMGI_INTRBY_VEC;
212 if (pcitool_debug)
213 prom_printf("user version:%d, flags:0x%x\n",
214 iset.user_version, iset.flags);
215
216 result = ENOTSUP;
217 if ((iset.user_version >= PCITOOL_V2) &&
218 (iset.flags & PCITOOL_INTR_FLAG_SET_GROUP)) {
219 ret = (*psm_intr_ops)(NULL, &info_hdl, PSM_INTR_OP_GRP_SET_CPU,
220 &result);
221 } else {
222 ret = (*psm_intr_ops)(NULL, &info_hdl, PSM_INTR_OP_SET_CPU,
223 &result);
224 }
225
226 if (ret != PSM_SUCCESS) {
227 switch (result) {
228 case EIO: /* Error making the change */
229 rval = EIO;
230 iset.status = PCITOOL_IO_ERROR;
231 break;
232 case ENXIO: /* Couldn't convert vector to irq */
233 rval = EINVAL;
234 iset.status = PCITOOL_INVALID_INO;
235 break;
236 case EINVAL: /* CPU out of range */
237 rval = EINVAL;
238 iset.status = PCITOOL_INVALID_CPUID;
239 break;
240 case ENOTSUP: /* Requested PSM intr ops missing */
241 rval = ENOTSUP;
242 iset.status = PCITOOL_IO_ERROR;
243 break;
244 }
245 }
246
247 /* Return original CPU. */
248 iset.cpu_id = old_cpu;
249
250 /* Return new vector */
251 if (strcmp(type_info.avgi_type, APIC_APIX_NAME) == 0) {
252 iset.ino = APIX_VIRTVEC_VECTOR(info_hdl.ih_vector);
253 }
254
255 done_set_intr:
256 iset.drvr_version = PCITOOL_VERSION;
257 if (ddi_copyout(&iset, arg, copyinout_size, mode) != DDI_SUCCESS)
258 rval = EFAULT;
259 return (rval);
260 }
261
262
263 /* It is assumed that dip != NULL */
264 static void
265 pcitool_get_intr_dev_info(dev_info_t *dip, pcitool_intr_dev_t *devs)
266 {
267 (void) strncpy(devs->driver_name,
268 ddi_driver_name(dip), MAXMODCONFNAME-2);
269 devs->driver_name[MAXMODCONFNAME-1] = '\0';
270 (void) ddi_pathname(dip, devs->path);
271 devs->dev_inst = ddi_get_instance(dip);
272 }
273
274 static int
275 pcitool_get_intr(dev_info_t *dip, void *arg, int mode)
276 {
277 /* Array part isn't used here, but oh well... */
278 pcitool_intr_get_t partial_iget;
279 pcitool_intr_get_t *iget = &partial_iget;
280 size_t iget_kmem_alloc_size = 0;
281 uint8_t num_devs_ret;
282 int copyout_rval;
283 int rval = SUCCESS;
284 int circ;
285 int i;
286
287 ddi_intr_handle_impl_t info_hdl;
288 apic_get_intr_t intr_info;
289 apic_get_type_t type_info;
290
291 /* Read in just the header part, no array section. */
292 if (ddi_copyin(arg, &partial_iget, PCITOOL_IGET_SIZE(0), mode) !=
293 DDI_SUCCESS)
294 return (EFAULT);
295
296 if (partial_iget.flags & PCITOOL_INTR_FLAG_GET_MSI) {
297 partial_iget.status = PCITOOL_IO_ERROR;
298 partial_iget.num_devs_ret = 0;
299 rval = ENOTSUP;
300 goto done_get_intr;
301 }
302
303 info_hdl.ih_private = &type_info;
304
305 if ((*psm_intr_ops)(NULL, &info_hdl,
306 PSM_INTR_OP_APIC_TYPE, NULL) != PSM_SUCCESS) {
307 iget->status = PCITOOL_IO_ERROR;
308 iget->num_devs_ret = 0;
309 rval = EINVAL;
310 goto done_get_intr;
311 }
312
313 if (strcmp(type_info.avgi_type, APIC_APIX_NAME) == 0) {
314 if (partial_iget.cpu_id > type_info.avgi_num_cpu) {
315 partial_iget.status = PCITOOL_INVALID_CPUID;
316 partial_iget.num_devs_ret = 0;
317 rval = EINVAL;
318 goto done_get_intr;
319 }
320 }
321
322 /* Validate argument. */
323 if ((partial_iget.ino & APIX_VIRTVEC_VECMASK) >
324 type_info.avgi_num_intr) {
325 partial_iget.status = PCITOOL_INVALID_INO;
326 partial_iget.num_devs_ret = 0;
327 rval = EINVAL;
328 goto done_get_intr;
329 }
330
331 num_devs_ret = partial_iget.num_devs_ret;
332 intr_info.avgi_dip_list = NULL;
333 intr_info.avgi_req_flags =
334 PSMGI_REQ_CPUID | PSMGI_REQ_NUM_DEVS | PSMGI_INTRBY_VEC;
335 /*
336 * For this locally-declared and used handle, ih_private will contain a
337 * pointer to apic_get_intr_t, not an ihdl_plat_t as used for
338 * global interrupt handling.
339 */
340 info_hdl.ih_private = &intr_info;
341
342 if (strcmp(type_info.avgi_type, APIC_APIX_NAME) == 0) {
343 info_hdl.ih_vector =
344 APIX_VIRTVECTOR(partial_iget.cpu_id, partial_iget.ino);
345 } else {
346 info_hdl.ih_vector = partial_iget.ino;
347 }
348
349 /* Caller wants device information returned. */
350 if (num_devs_ret > 0) {
351
352 intr_info.avgi_req_flags |= PSMGI_REQ_GET_DEVS;
353
354 /*
355 * Allocate room.
356 * If num_devs_ret == 0 iget remains pointing to partial_iget.
357 */
358 iget_kmem_alloc_size = PCITOOL_IGET_SIZE(num_devs_ret);
359 iget = kmem_alloc(iget_kmem_alloc_size, KM_SLEEP);
360
361 /* Read in whole structure to verify there's room. */
362 if (ddi_copyin(arg, iget, iget_kmem_alloc_size, mode) !=
363 SUCCESS) {
364
365 /* Be consistent and just return EFAULT here. */
366 kmem_free(iget, iget_kmem_alloc_size);
367
368 return (EFAULT);
369 }
370 }
371
372 bzero(iget, PCITOOL_IGET_SIZE(num_devs_ret));
373 iget->ino = info_hdl.ih_vector;
374
375 /*
376 * Lock device tree branch from the pci root nexus on down if info will
377 * be extracted from dips returned from the tree.
378 */
379 if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
380 ndi_devi_enter(dip, &circ);
381 }
382
383 /* Call psm_intr_ops(PSM_INTR_OP_GET_INTR) to get information. */
384 if ((rval = (*psm_intr_ops)(NULL, &info_hdl,
385 PSM_INTR_OP_GET_INTR, NULL)) != PSM_SUCCESS) {
386 iget->status = PCITOOL_IO_ERROR;
387 iget->num_devs_ret = 0;
388 rval = EINVAL;
389 goto done_get_intr;
390 }
391
392 /*
393 * Fill in the pcitool_intr_get_t to be returned,
394 * with the CPU, num_devs_ret and num_devs.
395 */
396 if (intr_info.avgi_cpu_id == IRQ_UNBOUND ||
397 intr_info.avgi_cpu_id == IRQ_UNINIT)
398 iget->cpu_id = 0;
399 else
400 iget->cpu_id = intr_info.avgi_cpu_id & ~PSMGI_CPU_USER_BOUND;
401
402 /* Number of devices returned by apic. */
403 iget->num_devs = intr_info.avgi_num_devs;
404
405 /* Device info was returned. */
406 if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
407
408 /*
409 * num devs returned is num devs ret by apic,
410 * space permitting.
411 */
412 iget->num_devs_ret = min(num_devs_ret, intr_info.avgi_num_devs);
413
414 /*
415 * Loop thru list of dips and extract driver, name and instance.
416 * Fill in the pcitool_intr_dev_t's with this info.
417 */
418 for (i = 0; i < iget->num_devs_ret; i++)
419 pcitool_get_intr_dev_info(intr_info.avgi_dip_list[i],
420 &iget->dev[i]);
421
422 /* Free kmem_alloc'ed memory of the apic_get_intr_t */
423 kmem_free(intr_info.avgi_dip_list,
424 intr_info.avgi_num_devs * sizeof (dev_info_t *));
425 }
426
427 done_get_intr:
428
429 if (intr_info.avgi_req_flags & PSMGI_REQ_GET_DEVS) {
430 ndi_devi_exit(dip, circ);
431 }
432
433 iget->drvr_version = PCITOOL_VERSION;
434 copyout_rval = ddi_copyout(iget, arg,
435 PCITOOL_IGET_SIZE(num_devs_ret), mode);
436
437 if (iget_kmem_alloc_size > 0)
438 kmem_free(iget, iget_kmem_alloc_size);
439
440 if (copyout_rval != DDI_SUCCESS)
441 rval = EFAULT;
442
443 return (rval);
444 }
445
446 /*ARGSUSED*/
447 static int
448 pcitool_intr_info(dev_info_t *dip, void *arg, int mode)
449 {
450 pcitool_intr_info_t intr_info;
451 ddi_intr_handle_impl_t info_hdl;
452 int rval = SUCCESS;
453 apic_get_type_t type_info;
454
455 /* If we need user_version, and to ret same user version as passed in */
456 if (ddi_copyin(arg, &intr_info, sizeof (pcitool_intr_info_t), mode) !=
457 DDI_SUCCESS) {
458 if (pcitool_debug)
459 prom_printf("Error reading arguments\n");
460 return (EFAULT);
461 }
462
463 if (intr_info.flags & PCITOOL_INTR_FLAG_GET_MSI)
464 return (ENOTSUP);
465
466 info_hdl.ih_private = &type_info;
467
468 /* For UPPC systems, psm_intr_ops has no entry for APIC_TYPE. */
469 if ((rval = (*psm_intr_ops)(NULL, &info_hdl,
470 PSM_INTR_OP_APIC_TYPE, NULL)) != PSM_SUCCESS) {
471 intr_info.ctlr_type = PCITOOL_CTLR_TYPE_UPPC;
472 intr_info.ctlr_version = 0;
473 intr_info.num_intr = APIC_MAX_VECTOR;
474 } else {
475 intr_info.ctlr_version = (uint32_t)info_hdl.ih_ver;
476 intr_info.num_cpu = type_info.avgi_num_cpu;
477 if (strcmp(type_info.avgi_type,
478 APIC_PCPLUSMP_NAME) == 0) {
479 intr_info.ctlr_type = PCITOOL_CTLR_TYPE_PCPLUSMP;
480 intr_info.num_intr = type_info.avgi_num_intr;
481 } else if (strcmp(type_info.avgi_type,
482 APIC_APIX_NAME) == 0) {
483 intr_info.ctlr_type = PCITOOL_CTLR_TYPE_APIX;
484 intr_info.num_intr = type_info.avgi_num_intr;
485 } else {
486 intr_info.ctlr_type = PCITOOL_CTLR_TYPE_UNKNOWN;
487 intr_info.num_intr = APIC_MAX_VECTOR;
488 }
489 }
490
491 intr_info.drvr_version = PCITOOL_VERSION;
492 if (ddi_copyout(&intr_info, arg, sizeof (pcitool_intr_info_t), mode) !=
493 DDI_SUCCESS) {
494 if (pcitool_debug)
495 prom_printf("Error returning arguments.\n");
496 rval = EFAULT;
497 }
498
499 return (rval);
500 }
501
502
503
504 /*
505 * Main function for handling interrupt CPU binding requests and queries.
506 * Need to implement later
507 */
508 int
509 pcitool_intr_admn(dev_info_t *dip, void *arg, int cmd, int mode)
510 {
511 int rval;
512
513 switch (cmd) {
514
515 /* Associate a new CPU with a given vector */
516 case PCITOOL_DEVICE_SET_INTR:
517 rval = pcitool_set_intr(dip, arg, mode);
518 break;
519
520 case PCITOOL_DEVICE_GET_INTR:
521 rval = pcitool_get_intr(dip, arg, mode);
522 break;
523
524 case PCITOOL_SYSTEM_INTR_INFO:
525 rval = pcitool_intr_info(dip, arg, mode);
526 break;
527
528 default:
529 rval = ENOTSUP;
530 }
531
532 return (rval);
533 }
534
535 /*
536 * Perform register accesses on the nexus device itself.
537 * No explicit PCI nexus device for X86, so not applicable.
538 */
539
540 /*ARGSUSED*/
541 int
542 pcitool_bus_reg_ops(dev_info_t *dip, void *arg, int cmd, int mode)
543 {
544 return (ENOTSUP);
545 }
546
547 /* Swap endianness. */
548 static uint64_t
549 pcitool_swap_endian(uint64_t data, int size)
550 {
551 typedef union {
552 uint64_t data64;
553 uint8_t data8[8];
554 } data_split_t;
555
556 data_split_t orig_data;
557 data_split_t returned_data;
558 int i;
559
560 orig_data.data64 = data;
561 returned_data.data64 = 0;
562
563 for (i = 0; i < size; i++) {
564 returned_data.data8[i] = orig_data.data8[size - 1 - i];
565 }
566
567 return (returned_data.data64);
568 }
569
570 /*
571 * A note about ontrap handling:
572 *
573 * X86 systems on which this module was tested return FFs instead of bus errors
574 * when accessing devices with invalid addresses. Ontrap handling, which
575 * gracefully handles kernel bus errors, is installed anyway for I/O and mem
576 * space accessing (not for pci config space), in case future X86 platforms
577 * require it.
578 */
579
580 /* Access device. prg is modified. */
581 static int
582 pcitool_cfg_access(pcitool_reg_t *prg, boolean_t write_flag,
583 boolean_t io_access)
584 {
585 int size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
586 boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
587 int rval = SUCCESS;
588 uint64_t local_data;
589 pci_cfgacc_req_t req;
590 uint32_t max_offset;
591
592 if ((size <= 0) || (size > 8) || !ISP2(size)) {
593 prg->status = PCITOOL_INVALID_SIZE;
594 return (ENOTSUP);
595 }
596
597 /*
598 * NOTE: there is no way to verify whether or not the address is
599 * valid other than that it is within the maximum offset. The
600 * put functions return void and the get functions return -1 on error.
601 */
602
603 if (io_access)
604 max_offset = 0xFF;
605 else
606 max_offset = 0xFFF;
607 if (prg->offset + size - 1 > max_offset) {
608 prg->status = PCITOOL_INVALID_ADDRESS;
609 return (ENOTSUP);
610 }
611
612 prg->status = PCITOOL_SUCCESS;
613
614 req.rcdip = NULL;
615 req.bdf = PCI_GETBDF(prg->bus_no, prg->dev_no, prg->func_no);
616 req.offset = prg->offset;
617 req.size = size;
618 req.write = write_flag;
619 req.ioacc = io_access;
620 if (write_flag) {
621 if (big_endian) {
622 local_data = pcitool_swap_endian(prg->data, size);
623 } else {
624 local_data = prg->data;
625 }
626 VAL64(&req) = local_data;
627 pci_cfgacc_acc(&req);
628 } else {
629 pci_cfgacc_acc(&req);
630 switch (size) {
631 case 1:
632 local_data = VAL8(&req);
633 break;
634 case 2:
635 local_data = VAL16(&req);
636 break;
637 case 4:
638 local_data = VAL32(&req);
639 break;
640 case 8:
641 local_data = VAL64(&req);
642 break;
643 }
644 if (big_endian) {
645 prg->data =
646 pcitool_swap_endian(local_data, size);
647 } else {
648 prg->data = local_data;
649 }
650 }
651 /*
652 * Check if legacy IO config access is used, in which case
653 * only first 256 bytes are valid.
654 */
655 if (req.ioacc && (prg->offset + size - 1 > 0xFF)) {
656 prg->status = PCITOOL_INVALID_ADDRESS;
657 return (ENOTSUP);
658 }
659
660 /* Set phys_addr only if MMIO is used */
661 prg->phys_addr = 0;
662 if (!req.ioacc && mcfg_mem_base != 0) {
663 prg->phys_addr = mcfg_mem_base + prg->offset +
664 ((prg->bus_no << PCIEX_REG_BUS_SHIFT) |
665 (prg->dev_no << PCIEX_REG_DEV_SHIFT) |
666 (prg->func_no << PCIEX_REG_FUNC_SHIFT));
667 }
668
669 return (rval);
670 }
671
672 static int
673 pcitool_io_access(pcitool_reg_t *prg, boolean_t write_flag)
674 {
675 int port = (int)prg->phys_addr;
676 size_t size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
677 boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
678 volatile int rval = SUCCESS;
679 on_trap_data_t otd;
680 volatile uint64_t local_data;
681
682
683 /*
684 * on_trap works like setjmp.
685 *
686 * A non-zero return here means on_trap has returned from an error.
687 *
688 * A zero return here means that on_trap has just returned from setup.
689 */
690 if (on_trap(&otd, OT_DATA_ACCESS)) {
691 no_trap();
692 if (pcitool_debug)
693 prom_printf(
694 "pcitool_io_access: on_trap caught an error...\n");
695 prg->status = PCITOOL_INVALID_ADDRESS;
696 return (EFAULT);
697 }
698
699 if (write_flag) {
700
701 if (big_endian) {
702 local_data = pcitool_swap_endian(prg->data, size);
703 } else {
704 local_data = prg->data;
705 }
706
707 if (pcitool_debug)
708 prom_printf("Writing %ld byte(s) to port 0x%x\n",
709 size, port);
710
711 switch (size) {
712 case 1:
713 outb(port, (uint8_t)local_data);
714 break;
715 case 2:
716 outw(port, (uint16_t)local_data);
717 break;
718 case 4:
719 outl(port, (uint32_t)local_data);
720 break;
721 default:
722 rval = ENOTSUP;
723 prg->status = PCITOOL_INVALID_SIZE;
724 break;
725 }
726 } else {
727 if (pcitool_debug)
728 prom_printf("Reading %ld byte(s) from port 0x%x\n",
729 size, port);
730
731 switch (size) {
732 case 1:
733 local_data = inb(port);
734 break;
735 case 2:
736 local_data = inw(port);
737 break;
738 case 4:
739 local_data = inl(port);
740 break;
741 default:
742 rval = ENOTSUP;
743 prg->status = PCITOOL_INVALID_SIZE;
744 break;
745 }
746
747 if (rval == SUCCESS) {
748 if (big_endian) {
749 prg->data =
750 pcitool_swap_endian(local_data, size);
751 } else {
752 prg->data = local_data;
753 }
754 }
755 }
756
757 no_trap();
758 return (rval);
759 }
760
761 static int
762 pcitool_mem_access(pcitool_reg_t *prg, uint64_t virt_addr, boolean_t write_flag)
763 {
764 size_t size = PCITOOL_ACC_ATTR_SIZE(prg->acc_attr);
765 boolean_t big_endian = PCITOOL_ACC_IS_BIG_ENDIAN(prg->acc_attr);
766 volatile int rval = DDI_SUCCESS;
767 on_trap_data_t otd;
768 volatile uint64_t local_data;
769
770 /*
771 * on_trap works like setjmp.
772 *
773 * A non-zero return here means on_trap has returned from an error.
774 *
775 * A zero return here means that on_trap has just returned from setup.
776 */
777 if (on_trap(&otd, OT_DATA_ACCESS)) {
778 no_trap();
779 if (pcitool_debug)
780 prom_printf(
781 "pcitool_mem_access: on_trap caught an error...\n");
782 prg->status = PCITOOL_INVALID_ADDRESS;
783 return (EFAULT);
784 }
785
786 if (write_flag) {
787
788 if (big_endian) {
789 local_data = pcitool_swap_endian(prg->data, size);
790 } else {
791 local_data = prg->data;
792 }
793
794 switch (size) {
795 case 1:
796 *((uint8_t *)(uintptr_t)virt_addr) = local_data;
797 break;
798 case 2:
799 *((uint16_t *)(uintptr_t)virt_addr) = local_data;
800 break;
801 case 4:
802 *((uint32_t *)(uintptr_t)virt_addr) = local_data;
803 break;
804 case 8:
805 *((uint64_t *)(uintptr_t)virt_addr) = local_data;
806 break;
807 default:
808 rval = ENOTSUP;
809 prg->status = PCITOOL_INVALID_SIZE;
810 break;
811 }
812 } else {
813 switch (size) {
814 case 1:
815 local_data = *((uint8_t *)(uintptr_t)virt_addr);
816 break;
817 case 2:
818 local_data = *((uint16_t *)(uintptr_t)virt_addr);
819 break;
820 case 4:
821 local_data = *((uint32_t *)(uintptr_t)virt_addr);
822 break;
823 case 8:
824 local_data = *((uint64_t *)(uintptr_t)virt_addr);
825 break;
826 default:
827 rval = ENOTSUP;
828 prg->status = PCITOOL_INVALID_SIZE;
829 break;
830 }
831
832 if (rval == SUCCESS) {
833 if (big_endian) {
834 prg->data =
835 pcitool_swap_endian(local_data, size);
836 } else {
837 prg->data = local_data;
838 }
839 }
840 }
841
842 no_trap();
843 return (rval);
844 }
845
846 /*
847 * Map up to 2 pages which contain the address we want to access.
848 *
849 * Mapping should span no more than 8 bytes. With X86 it is possible for an
850 * 8 byte value to start on a 4 byte boundary, so it can cross a page boundary.
851 * We'll never have to map more than two pages.
852 */
853
854 static uint64_t
855 pcitool_map(uint64_t phys_addr, size_t size, size_t *num_pages)
856 {
857
858 uint64_t page_base = phys_addr & ~MMU_PAGEOFFSET;
859 uint64_t offset = phys_addr & MMU_PAGEOFFSET;
860 void *virt_base;
861 uint64_t returned_addr;
862 pfn_t pfn;
863
864 if (pcitool_debug)
865 prom_printf("pcitool_map: Called with PA:0x%p\n",
866 (void *)(uintptr_t)phys_addr);
867
868 *num_pages = 1;
869
870 /* Desired mapping would span more than two pages. */
871 if ((offset + size) > (MMU_PAGESIZE * 2)) {
872 if (pcitool_debug)
873 prom_printf("boundary violation: "
874 "offset:0x%" PRIx64 ", size:%ld, pagesize:0x%lx\n",
875 offset, (uintptr_t)size, (uintptr_t)MMU_PAGESIZE);
876 return (NULL);
877
878 } else if ((offset + size) > MMU_PAGESIZE) {
879 (*num_pages)++;
880 }
881
882 /* Get page(s) of virtual space. */
883 virt_base = vmem_alloc(heap_arena, ptob(*num_pages), VM_NOSLEEP);
884 if (virt_base == NULL) {
885 if (pcitool_debug)
886 prom_printf("Couldn't get virtual base address.\n");
887 return (NULL);
888 }
889
890 if (pcitool_debug)
891 prom_printf("Got base virtual address:0x%p\n", virt_base);
892
893 #ifdef __xpv
894 /*
895 * We should only get here if we are dom0.
896 * We're using a real device so we need to translate the MA to a PFN.
897 */
898 ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
899 pfn = xen_assign_pfn(mmu_btop(page_base));
900 #else
901 pfn = btop(page_base);
902 #endif
903
904 /* Now map the allocated virtual space to the physical address. */
905 hat_devload(kas.a_hat, virt_base, mmu_ptob(*num_pages), pfn,
906 PROT_READ | PROT_WRITE | HAT_STRICTORDER,
907 HAT_LOAD_LOCK);
908
909 returned_addr = ((uintptr_t)(virt_base)) + offset;
910
911 if (pcitool_debug)
912 prom_printf("pcitool_map: returning VA:0x%p\n",
913 (void *)(uintptr_t)returned_addr);
914
915 return (returned_addr);
916 }
917
918 /* Unmap the mapped page(s). */
919 static void
920 pcitool_unmap(uint64_t virt_addr, size_t num_pages)
921 {
922 void *base_virt_addr = (void *)(uintptr_t)(virt_addr & ~MMU_PAGEOFFSET);
923
924 hat_unload(kas.a_hat, base_virt_addr, ptob(num_pages),
925 HAT_UNLOAD_UNLOCK);
926 vmem_free(heap_arena, base_virt_addr, ptob(num_pages));
927 }
928
929
930 /* Perform register accesses on PCI leaf devices. */
931 /*ARGSUSED*/
932 int
933 pcitool_dev_reg_ops(dev_info_t *dip, void *arg, int cmd, int mode)
934 {
935 boolean_t write_flag = B_FALSE;
936 boolean_t io_access = B_TRUE;
937 int rval = 0;
938 pcitool_reg_t prg;
939 uint8_t size;
940
941 uint64_t base_addr;
942 uint64_t virt_addr;
943 size_t num_virt_pages;
944
945 switch (cmd) {
946 case (PCITOOL_DEVICE_SET_REG):
947 write_flag = B_TRUE;
948
949 /*FALLTHRU*/
950 case (PCITOOL_DEVICE_GET_REG):
951 if (pcitool_debug)
952 prom_printf("pci_dev_reg_ops set/get reg\n");
953 if (ddi_copyin(arg, &prg, sizeof (pcitool_reg_t), mode) !=
954 DDI_SUCCESS) {
955 if (pcitool_debug)
956 prom_printf("Error reading arguments\n");
957 return (EFAULT);
958 }
959
960 if (prg.barnum >= (sizeof (pci_bars) / sizeof (pci_bars[0]))) {
961 prg.status = PCITOOL_OUT_OF_RANGE;
962 rval = EINVAL;
963 goto done_reg;
964 }
965
966 if (pcitool_debug)
967 prom_printf("raw bus:0x%x, dev:0x%x, func:0x%x\n",
968 prg.bus_no, prg.dev_no, prg.func_no);
969 /* Validate address arguments of bus / dev / func */
970 if (((prg.bus_no &
971 (PCI_REG_BUS_M >> PCI_REG_BUS_SHIFT)) !=
972 prg.bus_no) ||
973 ((prg.dev_no &
974 (PCI_REG_DEV_M >> PCI_REG_DEV_SHIFT)) !=
975 prg.dev_no) ||
976 ((prg.func_no &
977 (PCI_REG_FUNC_M >> PCI_REG_FUNC_SHIFT)) !=
978 prg.func_no)) {
979 prg.status = PCITOOL_INVALID_ADDRESS;
980 rval = EINVAL;
981 goto done_reg;
982 }
983
984 size = PCITOOL_ACC_ATTR_SIZE(prg.acc_attr);
985
986 /* Proper config space desired. */
987 if (prg.barnum == 0) {
988
989 if (pcitool_debug)
990 prom_printf(
991 "config access: offset:0x%" PRIx64 ", "
992 "phys_addr:0x%" PRIx64 "\n",
993 prg.offset, prg.phys_addr);
994
995 if (prg.offset >= max_cfg_size) {
996 prg.status = PCITOOL_OUT_OF_RANGE;
997 rval = EINVAL;
998 goto done_reg;
999 }
1000 if (max_cfg_size == PCIE_CONF_HDR_SIZE)
1001 io_access = B_FALSE;
1002
1003 rval = pcitool_cfg_access(&prg, write_flag, io_access);
1004 if (pcitool_debug)
1005 prom_printf(
1006 "config access: data:0x%" PRIx64 "\n",
1007 prg.data);
1008
1009 /* IO/ MEM/ MEM64 space. */
1010 } else {
1011
1012 pcitool_reg_t prg2;
1013 bcopy(&prg, &prg2, sizeof (pcitool_reg_t));
1014
1015 /*
1016 * Translate BAR number into offset of the BAR in
1017 * the device's config space.
1018 */
1019 prg2.offset = pci_bars[prg2.barnum];
1020 prg2.acc_attr =
1021 PCITOOL_ACC_ATTR_SIZE_4 | PCITOOL_ACC_ATTR_ENDN_LTL;
1022
1023 if (pcitool_debug)
1024 prom_printf(
1025 "barnum:%d, bar_offset:0x%" PRIx64 "\n",
1026 prg2.barnum, prg2.offset);
1027 /*
1028 * Get Bus Address Register (BAR) from config space.
1029 * prg2.offset is the offset into config space of the
1030 * BAR desired. prg.status is modified on error.
1031 */
1032 rval = pcitool_cfg_access(&prg2, B_FALSE, B_TRUE);
1033 if (rval != SUCCESS) {
1034 if (pcitool_debug)
1035 prom_printf("BAR access failed\n");
1036 prg.status = prg2.status;
1037 goto done_reg;
1038 }
1039 /*
1040 * Reference proper PCI space based on the BAR.
1041 * If 64 bit MEM space, need to load other half of the
1042 * BAR first.
1043 */
1044
1045 if (pcitool_debug)
1046 prom_printf("bar returned is 0x%" PRIx64 "\n",
1047 prg2.data);
1048 if (!prg2.data) {
1049 if (pcitool_debug)
1050 prom_printf("BAR data == 0\n");
1051 rval = EINVAL;
1052 prg.status = PCITOOL_INVALID_ADDRESS;
1053 goto done_reg;
1054 }
1055 if (prg2.data == 0xffffffff) {
1056 if (pcitool_debug)
1057 prom_printf("BAR data == -1\n");
1058 rval = EINVAL;
1059 prg.status = PCITOOL_INVALID_ADDRESS;
1060 goto done_reg;
1061 }
1062
1063 /*
1064 * BAR has bits saying this space is IO space, unless
1065 * this is the ROM address register.
1066 */
1067 if (((PCI_BASE_SPACE_M & prg2.data) ==
1068 PCI_BASE_SPACE_IO) &&
1069 (prg2.offset != PCI_CONF_ROM)) {
1070 if (pcitool_debug)
1071 prom_printf("IO space\n");
1072
1073 prg2.data &= PCI_BASE_IO_ADDR_M;
1074 prg.phys_addr = prg2.data + prg.offset;
1075
1076 rval = pcitool_io_access(&prg, write_flag);
1077 if ((rval != SUCCESS) && (pcitool_debug))
1078 prom_printf("IO access failed\n");
1079
1080 goto done_reg;
1081
1082
1083 /*
1084 * BAR has bits saying this space is 64 bit memory
1085 * space, unless this is the ROM address register.
1086 *
1087 * The 64 bit address stored in two BAR cells is not
1088 * necessarily aligned on an 8-byte boundary.
1089 * Need to keep the first 4 bytes read,
1090 * and do a separate read of the high 4 bytes.
1091 */
1092
1093 } else if ((PCI_BASE_TYPE_ALL & prg2.data) &&
1094 (prg2.offset != PCI_CONF_ROM)) {
1095
1096 uint32_t low_bytes =
1097 (uint32_t)(prg2.data & ~PCI_BASE_TYPE_ALL);
1098
1099 /*
1100 * Don't try to read the next 4 bytes
1101 * past the end of BARs.
1102 */
1103 if (prg2.offset >= PCI_CONF_BASE5) {
1104 prg.status = PCITOOL_OUT_OF_RANGE;
1105 rval = EIO;
1106 goto done_reg;
1107 }
1108
1109 /*
1110 * Access device.
1111 * prg2.status is modified on error.
1112 */
1113 prg2.offset += 4;
1114 rval = pcitool_cfg_access(&prg2,
1115 B_FALSE, B_TRUE);
1116 if (rval != SUCCESS) {
1117 prg.status = prg2.status;
1118 goto done_reg;
1119 }
1120
1121 if (prg2.data == 0xffffffff) {
1122 prg.status = PCITOOL_INVALID_ADDRESS;
1123 prg.status = EFAULT;
1124 goto done_reg;
1125 }
1126
1127 prg2.data = (prg2.data << 32) + low_bytes;
1128 if (pcitool_debug)
1129 prom_printf(
1130 "64 bit mem space. "
1131 "64-bit bar is 0x%" PRIx64 "\n",
1132 prg2.data);
1133
1134 /* Mem32 space, including ROM */
1135 } else {
1136
1137 if (prg2.offset == PCI_CONF_ROM) {
1138 if (pcitool_debug)
1139 prom_printf(
1140 "Additional ROM "
1141 "checking\n");
1142 /* Can't write to ROM */
1143 if (write_flag) {
1144 prg.status = PCITOOL_ROM_WRITE;
1145 rval = EIO;
1146 goto done_reg;
1147
1148 /* ROM disabled for reading */
1149 } else if (!(prg2.data & 0x00000001)) {
1150 prg.status =
1151 PCITOOL_ROM_DISABLED;
1152 rval = EIO;
1153 goto done_reg;
1154 }
1155 }
1156
1157 if (pcitool_debug)
1158 prom_printf("32 bit mem space\n");
1159 }
1160
1161 /* Common code for all IO/MEM range spaces. */
1162
1163 base_addr = prg2.data;
1164 if (pcitool_debug)
1165 prom_printf(
1166 "addr portion of bar is 0x%" PRIx64 ", "
1167 "base=0x%" PRIx64 ", "
1168 "offset:0x%" PRIx64 "\n",
1169 prg2.data, base_addr, prg.offset);
1170 /*
1171 * Use offset provided by caller to index into
1172 * desired space, then access.
1173 * Note that prg.status is modified on error.
1174 */
1175 prg.phys_addr = base_addr + prg.offset;
1176
1177 virt_addr = pcitool_map(prg.phys_addr, size,
1178 &num_virt_pages);
1179 if (virt_addr == NULL) {
1180 prg.status = PCITOOL_IO_ERROR;
1181 rval = EIO;
1182 goto done_reg;
1183 }
1184
1185 rval = pcitool_mem_access(&prg, virt_addr, write_flag);
1186 pcitool_unmap(virt_addr, num_virt_pages);
1187 }
1188 done_reg:
1189 prg.drvr_version = PCITOOL_VERSION;
1190 if (ddi_copyout(&prg, arg, sizeof (pcitool_reg_t), mode) !=
1191 DDI_SUCCESS) {
1192 if (pcitool_debug)
1193 prom_printf("Error returning arguments.\n");
1194 rval = EFAULT;
1195 }
1196 break;
1197 default:
1198 rval = ENOTTY;
1199 break;
1200 }
1201 return (rval);
1202 }
Unleashed OS