xref: /titanic_44/usr/src/uts/sun4u/io/px/px_lib4u.c (revision e4b86885570d77af552e9cf94f142f4d744fb8c8)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/kmem.h>
28 #include <sys/conf.h>
29 #include <sys/ddi.h>
30 #include <sys/sunddi.h>
31 #include <sys/sunndi.h>
32 #include <sys/fm/protocol.h>
33 #include <sys/fm/util.h>
34 #include <sys/modctl.h>
35 #include <sys/disp.h>
36 #include <sys/stat.h>
37 #include <sys/ddi_impldefs.h>
38 #include <sys/vmem.h>
39 #include <sys/iommutsb.h>
40 #include <sys/cpuvar.h>
41 #include <sys/ivintr.h>
42 #include <sys/byteorder.h>
43 #include <sys/hotplug/pci/pciehpc.h>
44 #include <sys/spl.h>
45 #include <px_obj.h>
46 #include <pcie_pwr.h>
47 #include "px_tools_var.h"
48 #include <px_regs.h>
49 #include <px_csr.h>
50 #include <sys/machsystm.h>
51 #include "px_lib4u.h"
52 #include "px_err.h"
53 #include "oberon_regs.h"
54 
55 #pragma weak jbus_stst_order
56 
57 extern void jbus_stst_order();
58 
59 ulong_t px_mmu_dvma_end = 0xfffffffful;
60 uint_t px_ranges_phi_mask = 0xfffffffful;
61 uint64_t *px_oberon_ubc_scratch_regs;
62 uint64_t px_paddr_mask;
63 
64 static int px_goto_l23ready(px_t *px_p);
65 static int px_goto_l0(px_t *px_p);
66 static int px_pre_pwron_check(px_t *px_p);
67 static uint32_t px_identity_init(px_t *px_p);
68 static boolean_t px_cpr_callb(void *arg, int code);
69 static uint_t px_cb_intr(caddr_t arg);
70 
71 /*
72  * px_lib_map_registers
73  *
74  * This function is called from the attach routine to map the registers
75  * accessed by this driver.
76  *
77  * used by: px_attach()
78  *
79  * return value: DDI_FAILURE on failure
80  */
81 int
82 px_lib_map_regs(pxu_t *pxu_p, dev_info_t *dip)
83 {
84 	ddi_device_acc_attr_t	attr;
85 	px_reg_bank_t		reg_bank = PX_REG_CSR;
86 
87 	DBG(DBG_ATTACH, dip, "px_lib_map_regs: pxu_p:0x%p, dip 0x%p\n",
88 	    pxu_p, dip);
89 
90 	attr.devacc_attr_version = DDI_DEVICE_ATTR_V0;
91 	attr.devacc_attr_dataorder = DDI_STRICTORDER_ACC;
92 	attr.devacc_attr_endian_flags = DDI_NEVERSWAP_ACC;
93 
94 	/*
95 	 * PCI CSR Base
96 	 */
97 	if (ddi_regs_map_setup(dip, reg_bank, &pxu_p->px_address[reg_bank],
98 	    0, 0, &attr, &pxu_p->px_ac[reg_bank]) != DDI_SUCCESS) {
99 		goto fail;
100 	}
101 
102 	reg_bank++;
103 
104 	/*
105 	 * XBUS CSR Base
106 	 */
107 	if (ddi_regs_map_setup(dip, reg_bank, &pxu_p->px_address[reg_bank],
108 	    0, 0, &attr, &pxu_p->px_ac[reg_bank]) != DDI_SUCCESS) {
109 		goto fail;
110 	}
111 
112 	pxu_p->px_address[reg_bank] -= FIRE_CONTROL_STATUS;
113 
114 done:
115 	for (; reg_bank >= PX_REG_CSR; reg_bank--) {
116 		DBG(DBG_ATTACH, dip, "reg_bank 0x%x address 0x%p\n",
117 		    reg_bank, pxu_p->px_address[reg_bank]);
118 	}
119 
120 	return (DDI_SUCCESS);
121 
122 fail:
123 	cmn_err(CE_WARN, "%s%d: unable to map reg entry %d\n",
124 	    ddi_driver_name(dip), ddi_get_instance(dip), reg_bank);
125 
126 	for (reg_bank--; reg_bank >= PX_REG_CSR; reg_bank--) {
127 		pxu_p->px_address[reg_bank] = NULL;
128 		ddi_regs_map_free(&pxu_p->px_ac[reg_bank]);
129 	}
130 
131 	return (DDI_FAILURE);
132 }
133 
134 /*
135  * px_lib_unmap_regs:
136  *
137  * This routine unmaps the registers mapped by map_px_registers.
138  *
139  * used by: px_detach(), and error conditions in px_attach()
140  *
141  * return value: none
142  */
143 void
144 px_lib_unmap_regs(pxu_t *pxu_p)
145 {
146 	int i;
147 
148 	for (i = 0; i < PX_REG_MAX; i++) {
149 		if (pxu_p->px_ac[i])
150 			ddi_regs_map_free(&pxu_p->px_ac[i]);
151 	}
152 }
153 
154 int
155 px_lib_dev_init(dev_info_t *dip, devhandle_t *dev_hdl)
156 {
157 
158 	caddr_t			xbc_csr_base, csr_base;
159 	px_dvma_range_prop_t	px_dvma_range;
160 	pxu_t			*pxu_p;
161 	uint8_t			chip_mask;
162 	px_t			*px_p = DIP_TO_STATE(dip);
163 	px_chip_type_t		chip_type = px_identity_init(px_p);
164 
165 	DBG(DBG_ATTACH, dip, "px_lib_dev_init: dip 0x%p", dip);
166 
167 	if (chip_type == PX_CHIP_UNIDENTIFIED) {
168 		cmn_err(CE_WARN, "%s%d: Unrecognized Hardware Version\n",
169 		    NAMEINST(dip));
170 		return (DDI_FAILURE);
171 	}
172 
173 	chip_mask = BITMASK(chip_type);
174 	px_paddr_mask = (chip_type == PX_CHIP_FIRE) ? MMU_FIRE_PADDR_MASK :
175 	    MMU_OBERON_PADDR_MASK;
176 
177 	/*
178 	 * Allocate platform specific structure and link it to
179 	 * the px state structure.
180 	 */
181 	pxu_p = kmem_zalloc(sizeof (pxu_t), KM_SLEEP);
182 	pxu_p->chip_type = chip_type;
183 	pxu_p->portid  = ddi_getprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
184 	    "portid", -1);
185 
186 	/* Map in the registers */
187 	if (px_lib_map_regs(pxu_p, dip) == DDI_FAILURE) {
188 		kmem_free(pxu_p, sizeof (pxu_t));
189 
190 		return (DDI_FAILURE);
191 	}
192 
193 	xbc_csr_base = (caddr_t)pxu_p->px_address[PX_REG_XBC];
194 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
195 
196 	pxu_p->tsb_cookie = iommu_tsb_alloc(pxu_p->portid);
197 	pxu_p->tsb_size = iommu_tsb_cookie_to_size(pxu_p->tsb_cookie);
198 	pxu_p->tsb_vaddr = iommu_tsb_cookie_to_va(pxu_p->tsb_cookie);
199 
200 	pxu_p->tsb_paddr = va_to_pa(pxu_p->tsb_vaddr);
201 
202 	/*
203 	 * Create "virtual-dma" property to support child devices
204 	 * needing to know DVMA range.
205 	 */
206 	px_dvma_range.dvma_base = (uint32_t)px_mmu_dvma_end + 1
207 	    - ((pxu_p->tsb_size >> 3) << MMU_PAGE_SHIFT);
208 	px_dvma_range.dvma_len = (uint32_t)
209 	    px_mmu_dvma_end - px_dvma_range.dvma_base + 1;
210 
211 	(void) ddi_prop_update_int_array(DDI_DEV_T_NONE, dip,
212 	    "virtual-dma", (int *)&px_dvma_range,
213 	    sizeof (px_dvma_range_prop_t) / sizeof (int));
214 	/*
215 	 * Initilize all fire hardware specific blocks.
216 	 */
217 	hvio_cb_init(xbc_csr_base, pxu_p);
218 	hvio_ib_init(csr_base, pxu_p);
219 	hvio_pec_init(csr_base, pxu_p);
220 	hvio_mmu_init(csr_base, pxu_p);
221 
222 	px_p->px_plat_p = (void *)pxu_p;
223 
224 	/*
225 	 * Initialize all the interrupt handlers
226 	 */
227 	switch (PX_CHIP_TYPE(pxu_p)) {
228 	case PX_CHIP_OBERON:
229 		/*
230 		 * Oberon hotplug uses SPARE3 field in ILU Error Log Enable
231 		 * register to indicate the status of leaf reset,
232 		 * we need to preserve the value of this bit, and keep it in
233 		 * px_ilu_log_mask to reflect the state of the bit
234 		 */
235 		if (CSR_BR(csr_base, ILU_ERROR_LOG_ENABLE, SPARE3))
236 			px_ilu_log_mask |= (1ull <<
237 			    ILU_ERROR_LOG_ENABLE_SPARE3);
238 		else
239 			px_ilu_log_mask &= ~(1ull <<
240 			    ILU_ERROR_LOG_ENABLE_SPARE3);
241 
242 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_ENABLE);
243 		break;
244 
245 	case PX_CHIP_FIRE:
246 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_ENABLE);
247 		break;
248 
249 	default:
250 		cmn_err(CE_WARN, "%s%d: PX primary bus Unknown\n",
251 		    ddi_driver_name(dip), ddi_get_instance(dip));
252 		return (DDI_FAILURE);
253 	}
254 
255 	/* Initilize device handle */
256 	*dev_hdl = (devhandle_t)csr_base;
257 
258 	DBG(DBG_ATTACH, dip, "px_lib_dev_init: dev_hdl 0x%llx\n", *dev_hdl);
259 
260 	return (DDI_SUCCESS);
261 }
262 
263 int
264 px_lib_dev_fini(dev_info_t *dip)
265 {
266 	caddr_t			csr_base;
267 	uint8_t			chip_mask;
268 	px_t			*px_p = DIP_TO_STATE(dip);
269 	pxu_t			*pxu_p = (pxu_t *)px_p->px_plat_p;
270 
271 	DBG(DBG_DETACH, dip, "px_lib_dev_fini: dip 0x%p\n", dip);
272 
273 	/*
274 	 * Deinitialize all the interrupt handlers
275 	 */
276 	switch (PX_CHIP_TYPE(pxu_p)) {
277 	case PX_CHIP_OBERON:
278 	case PX_CHIP_FIRE:
279 		chip_mask = BITMASK(PX_CHIP_TYPE(pxu_p));
280 		csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
281 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_DISABLE);
282 		break;
283 
284 	default:
285 		cmn_err(CE_WARN, "%s%d: PX primary bus Unknown\n",
286 		    ddi_driver_name(dip), ddi_get_instance(dip));
287 		return (DDI_FAILURE);
288 	}
289 
290 	iommu_tsb_free(pxu_p->tsb_cookie);
291 
292 	px_lib_unmap_regs((pxu_t *)px_p->px_plat_p);
293 	kmem_free(px_p->px_plat_p, sizeof (pxu_t));
294 	px_p->px_plat_p = NULL;
295 	(void) ddi_prop_remove(DDI_DEV_T_NONE, dip, "virtual-dma");
296 
297 	return (DDI_SUCCESS);
298 }
299 
300 /*ARGSUSED*/
301 int
302 px_lib_intr_devino_to_sysino(dev_info_t *dip, devino_t devino,
303     sysino_t *sysino)
304 {
305 	px_t	*px_p = DIP_TO_STATE(dip);
306 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
307 	uint64_t	ret;
308 
309 	DBG(DBG_LIB_INT, dip, "px_lib_intr_devino_to_sysino: dip 0x%p "
310 	    "devino 0x%x\n", dip, devino);
311 
312 	if ((ret = hvio_intr_devino_to_sysino(DIP_TO_HANDLE(dip),
313 	    pxu_p, devino, sysino)) != H_EOK) {
314 		DBG(DBG_LIB_INT, dip,
315 		    "hvio_intr_devino_to_sysino failed, ret 0x%lx\n", ret);
316 		return (DDI_FAILURE);
317 	}
318 
319 	DBG(DBG_LIB_INT, dip, "px_lib_intr_devino_to_sysino: sysino 0x%llx\n",
320 	    *sysino);
321 
322 	return (DDI_SUCCESS);
323 }
324 
325 /*ARGSUSED*/
326 int
327 px_lib_intr_getvalid(dev_info_t *dip, sysino_t sysino,
328     intr_valid_state_t *intr_valid_state)
329 {
330 	uint64_t	ret;
331 
332 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getvalid: dip 0x%p sysino 0x%llx\n",
333 	    dip, sysino);
334 
335 	if ((ret = hvio_intr_getvalid(DIP_TO_HANDLE(dip),
336 	    sysino, intr_valid_state)) != H_EOK) {
337 		DBG(DBG_LIB_INT, dip, "hvio_intr_getvalid failed, ret 0x%lx\n",
338 		    ret);
339 		return (DDI_FAILURE);
340 	}
341 
342 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getvalid: intr_valid_state 0x%x\n",
343 	    *intr_valid_state);
344 
345 	return (DDI_SUCCESS);
346 }
347 
348 /*ARGSUSED*/
349 int
350 px_lib_intr_setvalid(dev_info_t *dip, sysino_t sysino,
351     intr_valid_state_t intr_valid_state)
352 {
353 	uint64_t	ret;
354 
355 	DBG(DBG_LIB_INT, dip, "px_lib_intr_setvalid: dip 0x%p sysino 0x%llx "
356 	    "intr_valid_state 0x%x\n", dip, sysino, intr_valid_state);
357 
358 	if ((ret = hvio_intr_setvalid(DIP_TO_HANDLE(dip),
359 	    sysino, intr_valid_state)) != H_EOK) {
360 		DBG(DBG_LIB_INT, dip, "hvio_intr_setvalid failed, ret 0x%lx\n",
361 		    ret);
362 		return (DDI_FAILURE);
363 	}
364 
365 	return (DDI_SUCCESS);
366 }
367 
368 /*ARGSUSED*/
369 int
370 px_lib_intr_getstate(dev_info_t *dip, sysino_t sysino,
371     intr_state_t *intr_state)
372 {
373 	uint64_t	ret;
374 
375 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getstate: dip 0x%p sysino 0x%llx\n",
376 	    dip, sysino);
377 
378 	if ((ret = hvio_intr_getstate(DIP_TO_HANDLE(dip),
379 	    sysino, intr_state)) != H_EOK) {
380 		DBG(DBG_LIB_INT, dip, "hvio_intr_getstate failed, ret 0x%lx\n",
381 		    ret);
382 		return (DDI_FAILURE);
383 	}
384 
385 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getstate: intr_state 0x%x\n",
386 	    *intr_state);
387 
388 	return (DDI_SUCCESS);
389 }
390 
391 /*ARGSUSED*/
392 int
393 px_lib_intr_setstate(dev_info_t *dip, sysino_t sysino,
394     intr_state_t intr_state)
395 {
396 	uint64_t	ret;
397 
398 	DBG(DBG_LIB_INT, dip, "px_lib_intr_setstate: dip 0x%p sysino 0x%llx "
399 	    "intr_state 0x%x\n", dip, sysino, intr_state);
400 
401 	if ((ret = hvio_intr_setstate(DIP_TO_HANDLE(dip),
402 	    sysino, intr_state)) != H_EOK) {
403 		DBG(DBG_LIB_INT, dip, "hvio_intr_setstate failed, ret 0x%lx\n",
404 		    ret);
405 		return (DDI_FAILURE);
406 	}
407 
408 	return (DDI_SUCCESS);
409 }
410 
411 /*ARGSUSED*/
412 int
413 px_lib_intr_gettarget(dev_info_t *dip, sysino_t sysino, cpuid_t *cpuid)
414 {
415 	px_t		*px_p = DIP_TO_STATE(dip);
416 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
417 	uint64_t	ret;
418 
419 	DBG(DBG_LIB_INT, dip, "px_lib_intr_gettarget: dip 0x%p sysino 0x%llx\n",
420 	    dip, sysino);
421 
422 	if ((ret = hvio_intr_gettarget(DIP_TO_HANDLE(dip), pxu_p,
423 	    sysino, cpuid)) != H_EOK) {
424 		DBG(DBG_LIB_INT, dip, "hvio_intr_gettarget failed, ret 0x%lx\n",
425 		    ret);
426 		return (DDI_FAILURE);
427 	}
428 
429 	DBG(DBG_LIB_INT, dip, "px_lib_intr_gettarget: cpuid 0x%x\n", cpuid);
430 
431 	return (DDI_SUCCESS);
432 }
433 
434 /*ARGSUSED*/
435 int
436 px_lib_intr_settarget(dev_info_t *dip, sysino_t sysino, cpuid_t cpuid)
437 {
438 	px_t		*px_p = DIP_TO_STATE(dip);
439 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
440 	uint64_t	ret;
441 
442 	DBG(DBG_LIB_INT, dip, "px_lib_intr_settarget: dip 0x%p sysino 0x%llx "
443 	    "cpuid 0x%x\n", dip, sysino, cpuid);
444 
445 	if ((ret = hvio_intr_settarget(DIP_TO_HANDLE(dip), pxu_p,
446 	    sysino, cpuid)) != H_EOK) {
447 		DBG(DBG_LIB_INT, dip, "hvio_intr_settarget failed, ret 0x%lx\n",
448 		    ret);
449 		return (DDI_FAILURE);
450 	}
451 
452 	return (DDI_SUCCESS);
453 }
454 
455 /*ARGSUSED*/
456 int
457 px_lib_intr_reset(dev_info_t *dip)
458 {
459 	devino_t	ino;
460 	sysino_t	sysino;
461 
462 	DBG(DBG_LIB_INT, dip, "px_lib_intr_reset: dip 0x%p\n", dip);
463 
464 	/* Reset all Interrupts */
465 	for (ino = 0; ino < INTERRUPT_MAPPING_ENTRIES; ino++) {
466 		if (px_lib_intr_devino_to_sysino(dip, ino,
467 		    &sysino) != DDI_SUCCESS)
468 			return (BF_FATAL);
469 
470 		if (px_lib_intr_setstate(dip, sysino,
471 		    INTR_IDLE_STATE) != DDI_SUCCESS)
472 			return (BF_FATAL);
473 	}
474 
475 	return (BF_NONE);
476 }
477 
478 /*ARGSUSED*/
479 int
480 px_lib_iommu_map(dev_info_t *dip, tsbid_t tsbid, pages_t pages,
481     io_attributes_t attr, void *addr, size_t pfn_index, int flags)
482 {
483 	px_t		*px_p = DIP_TO_STATE(dip);
484 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
485 	uint64_t	ret;
486 
487 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_map: dip 0x%p tsbid 0x%llx "
488 	    "pages 0x%x attr 0x%x addr 0x%p pfn_index 0x%llx flags 0x%x\n",
489 	    dip, tsbid, pages, attr, addr, pfn_index, flags);
490 
491 	if ((ret = hvio_iommu_map(px_p->px_dev_hdl, pxu_p, tsbid, pages,
492 	    attr, addr, pfn_index, flags)) != H_EOK) {
493 		DBG(DBG_LIB_DMA, dip,
494 		    "px_lib_iommu_map failed, ret 0x%lx\n", ret);
495 		return (DDI_FAILURE);
496 	}
497 
498 	return (DDI_SUCCESS);
499 }
500 
501 /*ARGSUSED*/
502 int
503 px_lib_iommu_demap(dev_info_t *dip, tsbid_t tsbid, pages_t pages)
504 {
505 	px_t		*px_p = DIP_TO_STATE(dip);
506 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
507 	uint64_t	ret;
508 
509 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_demap: dip 0x%p tsbid 0x%llx "
510 	    "pages 0x%x\n", dip, tsbid, pages);
511 
512 	if ((ret = hvio_iommu_demap(px_p->px_dev_hdl, pxu_p, tsbid, pages))
513 	    != H_EOK) {
514 		DBG(DBG_LIB_DMA, dip,
515 		    "px_lib_iommu_demap failed, ret 0x%lx\n", ret);
516 
517 		return (DDI_FAILURE);
518 	}
519 
520 	return (DDI_SUCCESS);
521 }
522 
523 /*ARGSUSED*/
524 int
525 px_lib_iommu_getmap(dev_info_t *dip, tsbid_t tsbid, io_attributes_t *attr_p,
526     r_addr_t *r_addr_p)
527 {
528 	px_t	*px_p = DIP_TO_STATE(dip);
529 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
530 	uint64_t	ret;
531 
532 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getmap: dip 0x%p tsbid 0x%llx\n",
533 	    dip, tsbid);
534 
535 	if ((ret = hvio_iommu_getmap(DIP_TO_HANDLE(dip), pxu_p, tsbid,
536 	    attr_p, r_addr_p)) != H_EOK) {
537 		DBG(DBG_LIB_DMA, dip,
538 		    "hvio_iommu_getmap failed, ret 0x%lx\n", ret);
539 
540 		return ((ret == H_ENOMAP) ? DDI_DMA_NOMAPPING:DDI_FAILURE);
541 	}
542 
543 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getmap: attr 0x%x r_addr 0x%llx\n",
544 	    *attr_p, *r_addr_p);
545 
546 	return (DDI_SUCCESS);
547 }
548 
549 
550 /*
551  * Checks dma attributes against system bypass ranges
552  * The bypass range is determined by the hardware. Return them so the
553  * common code can do generic checking against them.
554  */
555 /*ARGSUSED*/
556 int
557 px_lib_dma_bypass_rngchk(dev_info_t *dip, ddi_dma_attr_t *attr_p,
558     uint64_t *lo_p, uint64_t *hi_p)
559 {
560 	px_t	*px_p = DIP_TO_STATE(dip);
561 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
562 
563 	*lo_p = hvio_get_bypass_base(pxu_p);
564 	*hi_p = hvio_get_bypass_end(pxu_p);
565 
566 	return (DDI_SUCCESS);
567 }
568 
569 
570 /*ARGSUSED*/
571 int
572 px_lib_iommu_getbypass(dev_info_t *dip, r_addr_t ra, io_attributes_t attr,
573     io_addr_t *io_addr_p)
574 {
575 	uint64_t	ret;
576 	px_t	*px_p = DIP_TO_STATE(dip);
577 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
578 
579 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getbypass: dip 0x%p ra 0x%llx "
580 	    "attr 0x%x\n", dip, ra, attr);
581 
582 	if ((ret = hvio_iommu_getbypass(DIP_TO_HANDLE(dip), pxu_p, ra,
583 	    attr, io_addr_p)) != H_EOK) {
584 		DBG(DBG_LIB_DMA, dip,
585 		    "hvio_iommu_getbypass failed, ret 0x%lx\n", ret);
586 		return (DDI_FAILURE);
587 	}
588 
589 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getbypass: io_addr 0x%llx\n",
590 	    *io_addr_p);
591 
592 	return (DDI_SUCCESS);
593 }
594 
595 /*
596  * bus dma sync entry point.
597  */
598 /*ARGSUSED*/
599 int
600 px_lib_dma_sync(dev_info_t *dip, dev_info_t *rdip, ddi_dma_handle_t handle,
601     off_t off, size_t len, uint_t cache_flags)
602 {
603 	ddi_dma_impl_t *mp = (ddi_dma_impl_t *)handle;
604 	px_t	*px_p = DIP_TO_STATE(dip);
605 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
606 
607 	DBG(DBG_LIB_DMA, dip, "px_lib_dma_sync: dip 0x%p rdip 0x%p "
608 	    "handle 0x%llx off 0x%x len 0x%x flags 0x%x\n",
609 	    dip, rdip, handle, off, len, cache_flags);
610 
611 	/*
612 	 * No flush needed for Oberon
613 	 */
614 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON)
615 		return (DDI_SUCCESS);
616 
617 	/*
618 	 * jbus_stst_order is found only in certain cpu modules.
619 	 * Just return success if not present.
620 	 */
621 	if (&jbus_stst_order == NULL)
622 		return (DDI_SUCCESS);
623 
624 	if (!(mp->dmai_flags & PX_DMAI_FLAGS_INUSE)) {
625 		cmn_err(CE_WARN, "%s%d: Unbound dma handle %p.",
626 		    ddi_driver_name(rdip), ddi_get_instance(rdip), (void *)mp);
627 
628 		return (DDI_FAILURE);
629 	}
630 
631 	if (mp->dmai_flags & PX_DMAI_FLAGS_NOSYNC)
632 		return (DDI_SUCCESS);
633 
634 	/*
635 	 * No flush needed when sending data from memory to device.
636 	 * Nothing to do to "sync" memory to what device would already see.
637 	 */
638 	if (!(mp->dmai_rflags & DDI_DMA_READ) ||
639 	    ((cache_flags & PX_DMA_SYNC_DDI_FLAGS) == DDI_DMA_SYNC_FORDEV))
640 		return (DDI_SUCCESS);
641 
642 	/*
643 	 * Perform necessary cpu workaround to ensure jbus ordering.
644 	 * CPU's internal "invalidate FIFOs" are flushed.
645 	 */
646 
647 #if !defined(lint)
648 	kpreempt_disable();
649 #endif
650 	jbus_stst_order();
651 #if !defined(lint)
652 	kpreempt_enable();
653 #endif
654 	return (DDI_SUCCESS);
655 }
656 
657 /*
658  * MSIQ Functions:
659  */
660 /*ARGSUSED*/
661 int
662 px_lib_msiq_init(dev_info_t *dip)
663 {
664 	px_t		*px_p = DIP_TO_STATE(dip);
665 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
666 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
667 	px_dvma_addr_t	pg_index;
668 	size_t		q_sz = msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
669 	size_t		size;
670 	int		i, ret;
671 
672 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_init: dip 0x%p\n", dip);
673 
674 	/* must aligned on q_sz (happens to be !!! page) boundary */
675 	ASSERT(q_sz == 8 * 1024);
676 
677 	/*
678 	 * Map the EQ memory into the Fire MMU (has to be 512KB aligned)
679 	 * and then initialize the base address register.
680 	 *
681 	 * Allocate entries from Fire IOMMU so that the resulting address
682 	 * is properly aligned.  Calculate the index of the first allocated
683 	 * entry.  Note: The size of the mapping is assumed to be a multiple
684 	 * of the page size.
685 	 */
686 	size = msiq_state_p->msiq_cnt * q_sz;
687 
688 	msiq_state_p->msiq_buf_p = kmem_zalloc(size, KM_SLEEP);
689 
690 	for (i = 0; i < msiq_state_p->msiq_cnt; i++)
691 		msiq_state_p->msiq_p[i].msiq_base_p = (msiqhead_t *)
692 		    ((caddr_t)msiq_state_p->msiq_buf_p + (i * q_sz));
693 
694 	pxu_p->msiq_mapped_p = vmem_xalloc(px_p->px_mmu_p->mmu_dvma_map,
695 	    size, (512 * 1024), 0, 0, NULL, NULL, VM_NOSLEEP | VM_BESTFIT);
696 
697 	if (pxu_p->msiq_mapped_p == NULL)
698 		return (DDI_FAILURE);
699 
700 	pg_index = MMU_PAGE_INDEX(px_p->px_mmu_p,
701 	    MMU_BTOP((ulong_t)pxu_p->msiq_mapped_p));
702 
703 	if ((ret = px_lib_iommu_map(px_p->px_dip, PCI_TSBID(0, pg_index),
704 	    MMU_BTOP(size), PCI_MAP_ATTR_WRITE, msiq_state_p->msiq_buf_p,
705 	    0, MMU_MAP_BUF)) != DDI_SUCCESS) {
706 		DBG(DBG_LIB_MSIQ, dip,
707 		    "px_lib_msiq_init: px_lib_iommu_map failed, "
708 		    "ret 0x%lx\n", ret);
709 
710 		(void) px_lib_msiq_fini(dip);
711 		return (DDI_FAILURE);
712 	}
713 
714 	if ((ret = hvio_msiq_init(DIP_TO_HANDLE(dip),
715 	    pxu_p)) != H_EOK) {
716 		DBG(DBG_LIB_MSIQ, dip,
717 		    "hvio_msiq_init failed, ret 0x%lx\n", ret);
718 
719 		(void) px_lib_msiq_fini(dip);
720 		return (DDI_FAILURE);
721 	}
722 
723 	return (DDI_SUCCESS);
724 }
725 
726 /*ARGSUSED*/
727 int
728 px_lib_msiq_fini(dev_info_t *dip)
729 {
730 	px_t		*px_p = DIP_TO_STATE(dip);
731 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
732 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
733 	px_dvma_addr_t	pg_index;
734 	size_t		size;
735 
736 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_fini: dip 0x%p\n", dip);
737 
738 	/*
739 	 * Unmap and free the EQ memory that had been mapped
740 	 * into the Fire IOMMU.
741 	 */
742 	size = msiq_state_p->msiq_cnt *
743 	    msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
744 
745 	pg_index = MMU_PAGE_INDEX(px_p->px_mmu_p,
746 	    MMU_BTOP((ulong_t)pxu_p->msiq_mapped_p));
747 
748 	(void) px_lib_iommu_demap(px_p->px_dip,
749 	    PCI_TSBID(0, pg_index), MMU_BTOP(size));
750 
751 	/* Free the entries from the Fire MMU */
752 	vmem_xfree(px_p->px_mmu_p->mmu_dvma_map,
753 	    (void *)pxu_p->msiq_mapped_p, size);
754 
755 	kmem_free(msiq_state_p->msiq_buf_p, msiq_state_p->msiq_cnt *
756 	    msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t));
757 
758 	return (DDI_SUCCESS);
759 }
760 
761 /*ARGSUSED*/
762 int
763 px_lib_msiq_info(dev_info_t *dip, msiqid_t msiq_id, r_addr_t *ra_p,
764     uint_t *msiq_rec_cnt_p)
765 {
766 	px_t		*px_p = DIP_TO_STATE(dip);
767 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
768 	size_t		msiq_size;
769 
770 	DBG(DBG_LIB_MSIQ, dip, "px_msiq_info: dip 0x%p msiq_id 0x%x\n",
771 	    dip, msiq_id);
772 
773 	msiq_size = msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
774 	ra_p = (r_addr_t *)((caddr_t)msiq_state_p->msiq_buf_p +
775 	    (msiq_id * msiq_size));
776 
777 	*msiq_rec_cnt_p = msiq_state_p->msiq_rec_cnt;
778 
779 	DBG(DBG_LIB_MSIQ, dip, "px_msiq_info: ra_p 0x%p msiq_rec_cnt 0x%x\n",
780 	    ra_p, *msiq_rec_cnt_p);
781 
782 	return (DDI_SUCCESS);
783 }
784 
785 /*ARGSUSED*/
786 int
787 px_lib_msiq_getvalid(dev_info_t *dip, msiqid_t msiq_id,
788     pci_msiq_valid_state_t *msiq_valid_state)
789 {
790 	uint64_t	ret;
791 
792 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getvalid: dip 0x%p msiq_id 0x%x\n",
793 	    dip, msiq_id);
794 
795 	if ((ret = hvio_msiq_getvalid(DIP_TO_HANDLE(dip),
796 	    msiq_id, msiq_valid_state)) != H_EOK) {
797 		DBG(DBG_LIB_MSIQ, dip,
798 		    "hvio_msiq_getvalid failed, ret 0x%lx\n", ret);
799 		return (DDI_FAILURE);
800 	}
801 
802 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getvalid: msiq_valid_state 0x%x\n",
803 	    *msiq_valid_state);
804 
805 	return (DDI_SUCCESS);
806 }
807 
808 /*ARGSUSED*/
809 int
810 px_lib_msiq_setvalid(dev_info_t *dip, msiqid_t msiq_id,
811     pci_msiq_valid_state_t msiq_valid_state)
812 {
813 	uint64_t	ret;
814 
815 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_setvalid: dip 0x%p msiq_id 0x%x "
816 	    "msiq_valid_state 0x%x\n", dip, msiq_id, msiq_valid_state);
817 
818 	if ((ret = hvio_msiq_setvalid(DIP_TO_HANDLE(dip),
819 	    msiq_id, msiq_valid_state)) != H_EOK) {
820 		DBG(DBG_LIB_MSIQ, dip,
821 		    "hvio_msiq_setvalid failed, ret 0x%lx\n", ret);
822 		return (DDI_FAILURE);
823 	}
824 
825 	return (DDI_SUCCESS);
826 }
827 
828 /*ARGSUSED*/
829 int
830 px_lib_msiq_getstate(dev_info_t *dip, msiqid_t msiq_id,
831     pci_msiq_state_t *msiq_state)
832 {
833 	uint64_t	ret;
834 
835 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getstate: dip 0x%p msiq_id 0x%x\n",
836 	    dip, msiq_id);
837 
838 	if ((ret = hvio_msiq_getstate(DIP_TO_HANDLE(dip),
839 	    msiq_id, msiq_state)) != H_EOK) {
840 		DBG(DBG_LIB_MSIQ, dip,
841 		    "hvio_msiq_getstate failed, ret 0x%lx\n", ret);
842 		return (DDI_FAILURE);
843 	}
844 
845 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getstate: msiq_state 0x%x\n",
846 	    *msiq_state);
847 
848 	return (DDI_SUCCESS);
849 }
850 
851 /*ARGSUSED*/
852 int
853 px_lib_msiq_setstate(dev_info_t *dip, msiqid_t msiq_id,
854     pci_msiq_state_t msiq_state)
855 {
856 	uint64_t	ret;
857 
858 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_setstate: dip 0x%p msiq_id 0x%x "
859 	    "msiq_state 0x%x\n", dip, msiq_id, msiq_state);
860 
861 	if ((ret = hvio_msiq_setstate(DIP_TO_HANDLE(dip),
862 	    msiq_id, msiq_state)) != H_EOK) {
863 		DBG(DBG_LIB_MSIQ, dip,
864 		    "hvio_msiq_setstate failed, ret 0x%lx\n", ret);
865 		return (DDI_FAILURE);
866 	}
867 
868 	return (DDI_SUCCESS);
869 }
870 
871 /*ARGSUSED*/
872 int
873 px_lib_msiq_gethead(dev_info_t *dip, msiqid_t msiq_id,
874     msiqhead_t *msiq_head)
875 {
876 	uint64_t	ret;
877 
878 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gethead: dip 0x%p msiq_id 0x%x\n",
879 	    dip, msiq_id);
880 
881 	if ((ret = hvio_msiq_gethead(DIP_TO_HANDLE(dip),
882 	    msiq_id, msiq_head)) != H_EOK) {
883 		DBG(DBG_LIB_MSIQ, dip,
884 		    "hvio_msiq_gethead failed, ret 0x%lx\n", ret);
885 		return (DDI_FAILURE);
886 	}
887 
888 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gethead: msiq_head 0x%x\n",
889 	    *msiq_head);
890 
891 	return (DDI_SUCCESS);
892 }
893 
894 /*ARGSUSED*/
895 int
896 px_lib_msiq_sethead(dev_info_t *dip, msiqid_t msiq_id,
897     msiqhead_t msiq_head)
898 {
899 	uint64_t	ret;
900 
901 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_sethead: dip 0x%p msiq_id 0x%x "
902 	    "msiq_head 0x%x\n", dip, msiq_id, msiq_head);
903 
904 	if ((ret = hvio_msiq_sethead(DIP_TO_HANDLE(dip),
905 	    msiq_id, msiq_head)) != H_EOK) {
906 		DBG(DBG_LIB_MSIQ, dip,
907 		    "hvio_msiq_sethead failed, ret 0x%lx\n", ret);
908 		return (DDI_FAILURE);
909 	}
910 
911 	return (DDI_SUCCESS);
912 }
913 
914 /*ARGSUSED*/
915 int
916 px_lib_msiq_gettail(dev_info_t *dip, msiqid_t msiq_id,
917     msiqtail_t *msiq_tail)
918 {
919 	uint64_t	ret;
920 
921 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gettail: dip 0x%p msiq_id 0x%x\n",
922 	    dip, msiq_id);
923 
924 	if ((ret = hvio_msiq_gettail(DIP_TO_HANDLE(dip),
925 	    msiq_id, msiq_tail)) != H_EOK) {
926 		DBG(DBG_LIB_MSIQ, dip,
927 		    "hvio_msiq_gettail failed, ret 0x%lx\n", ret);
928 		return (DDI_FAILURE);
929 	}
930 
931 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gettail: msiq_tail 0x%x\n",
932 	    *msiq_tail);
933 
934 	return (DDI_SUCCESS);
935 }
936 
937 /*ARGSUSED*/
938 void
939 px_lib_get_msiq_rec(dev_info_t *dip, msiqhead_t *msiq_head_p,
940     msiq_rec_t *msiq_rec_p)
941 {
942 	eq_rec_t	*eq_rec_p = (eq_rec_t *)msiq_head_p;
943 
944 	DBG(DBG_LIB_MSIQ, dip, "px_lib_get_msiq_rec: dip 0x%p eq_rec_p 0x%p\n",
945 	    dip, eq_rec_p);
946 
947 	if (!eq_rec_p->eq_rec_fmt_type) {
948 		/* Set msiq_rec_type to zero */
949 		msiq_rec_p->msiq_rec_type = 0;
950 
951 		return;
952 	}
953 
954 	DBG(DBG_LIB_MSIQ, dip, "px_lib_get_msiq_rec: EQ RECORD, "
955 	    "eq_rec_rid 0x%llx eq_rec_fmt_type 0x%llx "
956 	    "eq_rec_len 0x%llx eq_rec_addr0 0x%llx "
957 	    "eq_rec_addr1 0x%llx eq_rec_data0 0x%llx "
958 	    "eq_rec_data1 0x%llx\n", eq_rec_p->eq_rec_rid,
959 	    eq_rec_p->eq_rec_fmt_type, eq_rec_p->eq_rec_len,
960 	    eq_rec_p->eq_rec_addr0, eq_rec_p->eq_rec_addr1,
961 	    eq_rec_p->eq_rec_data0, eq_rec_p->eq_rec_data1);
962 
963 	/*
964 	 * Only upper 4 bits of eq_rec_fmt_type is used
965 	 * to identify the EQ record type.
966 	 */
967 	switch (eq_rec_p->eq_rec_fmt_type >> 3) {
968 	case EQ_REC_MSI32:
969 		msiq_rec_p->msiq_rec_type = MSI32_REC;
970 
971 		msiq_rec_p->msiq_rec_data.msi.msi_data =
972 		    eq_rec_p->eq_rec_data0;
973 		break;
974 	case EQ_REC_MSI64:
975 		msiq_rec_p->msiq_rec_type = MSI64_REC;
976 
977 		msiq_rec_p->msiq_rec_data.msi.msi_data =
978 		    eq_rec_p->eq_rec_data0;
979 		break;
980 	case EQ_REC_MSG:
981 		msiq_rec_p->msiq_rec_type = MSG_REC;
982 
983 		msiq_rec_p->msiq_rec_data.msg.msg_route =
984 		    eq_rec_p->eq_rec_fmt_type & 7;
985 		msiq_rec_p->msiq_rec_data.msg.msg_targ = eq_rec_p->eq_rec_rid;
986 		msiq_rec_p->msiq_rec_data.msg.msg_code = eq_rec_p->eq_rec_data0;
987 		break;
988 	default:
989 		cmn_err(CE_WARN, "%s%d: px_lib_get_msiq_rec: "
990 		    "0x%x is an unknown EQ record type",
991 		    ddi_driver_name(dip), ddi_get_instance(dip),
992 		    (int)eq_rec_p->eq_rec_fmt_type);
993 		break;
994 	}
995 
996 	msiq_rec_p->msiq_rec_rid = eq_rec_p->eq_rec_rid;
997 	msiq_rec_p->msiq_rec_msi_addr = ((eq_rec_p->eq_rec_addr1 << 16) |
998 	    (eq_rec_p->eq_rec_addr0 << 2));
999 }
1000 
1001 /*ARGSUSED*/
1002 void
1003 px_lib_clr_msiq_rec(dev_info_t *dip, msiqhead_t *msiq_head_p)
1004 {
1005 	eq_rec_t	*eq_rec_p = (eq_rec_t *)msiq_head_p;
1006 
1007 	DBG(DBG_LIB_MSIQ, dip, "px_lib_clr_msiq_rec: dip 0x%p eq_rec_p 0x%p\n",
1008 	    dip, eq_rec_p);
1009 
1010 	if (eq_rec_p->eq_rec_fmt_type) {
1011 		/* Zero out eq_rec_fmt_type field */
1012 		eq_rec_p->eq_rec_fmt_type = 0;
1013 	}
1014 }
1015 
1016 /*
1017  * MSI Functions:
1018  */
1019 /*ARGSUSED*/
1020 int
1021 px_lib_msi_init(dev_info_t *dip)
1022 {
1023 	px_t		*px_p = DIP_TO_STATE(dip);
1024 	px_msi_state_t	*msi_state_p = &px_p->px_ib_p->ib_msi_state;
1025 	uint64_t	ret;
1026 
1027 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_init: dip 0x%p\n", dip);
1028 
1029 	if ((ret = hvio_msi_init(DIP_TO_HANDLE(dip),
1030 	    msi_state_p->msi_addr32, msi_state_p->msi_addr64)) != H_EOK) {
1031 		DBG(DBG_LIB_MSIQ, dip, "px_lib_msi_init failed, ret 0x%lx\n",
1032 		    ret);
1033 		return (DDI_FAILURE);
1034 	}
1035 
1036 	return (DDI_SUCCESS);
1037 }
1038 
1039 /*ARGSUSED*/
1040 int
1041 px_lib_msi_getmsiq(dev_info_t *dip, msinum_t msi_num,
1042     msiqid_t *msiq_id)
1043 {
1044 	uint64_t	ret;
1045 
1046 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getmsiq: dip 0x%p msi_num 0x%x\n",
1047 	    dip, msi_num);
1048 
1049 	if ((ret = hvio_msi_getmsiq(DIP_TO_HANDLE(dip),
1050 	    msi_num, msiq_id)) != H_EOK) {
1051 		DBG(DBG_LIB_MSI, dip,
1052 		    "hvio_msi_getmsiq failed, ret 0x%lx\n", ret);
1053 		return (DDI_FAILURE);
1054 	}
1055 
1056 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getmsiq: msiq_id 0x%x\n",
1057 	    *msiq_id);
1058 
1059 	return (DDI_SUCCESS);
1060 }
1061 
1062 /*ARGSUSED*/
1063 int
1064 px_lib_msi_setmsiq(dev_info_t *dip, msinum_t msi_num,
1065     msiqid_t msiq_id, msi_type_t msitype)
1066 {
1067 	uint64_t	ret;
1068 
1069 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setmsiq: dip 0x%p msi_num 0x%x "
1070 	    "msq_id 0x%x\n", dip, msi_num, msiq_id);
1071 
1072 	if ((ret = hvio_msi_setmsiq(DIP_TO_HANDLE(dip),
1073 	    msi_num, msiq_id)) != H_EOK) {
1074 		DBG(DBG_LIB_MSI, dip,
1075 		    "hvio_msi_setmsiq failed, ret 0x%lx\n", ret);
1076 		return (DDI_FAILURE);
1077 	}
1078 
1079 	return (DDI_SUCCESS);
1080 }
1081 
1082 /*ARGSUSED*/
1083 int
1084 px_lib_msi_getvalid(dev_info_t *dip, msinum_t msi_num,
1085     pci_msi_valid_state_t *msi_valid_state)
1086 {
1087 	uint64_t	ret;
1088 
1089 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getvalid: dip 0x%p msi_num 0x%x\n",
1090 	    dip, msi_num);
1091 
1092 	if ((ret = hvio_msi_getvalid(DIP_TO_HANDLE(dip),
1093 	    msi_num, msi_valid_state)) != H_EOK) {
1094 		DBG(DBG_LIB_MSI, dip,
1095 		    "hvio_msi_getvalid failed, ret 0x%lx\n", ret);
1096 		return (DDI_FAILURE);
1097 	}
1098 
1099 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getvalid: msiq_id 0x%x\n",
1100 	    *msi_valid_state);
1101 
1102 	return (DDI_SUCCESS);
1103 }
1104 
1105 /*ARGSUSED*/
1106 int
1107 px_lib_msi_setvalid(dev_info_t *dip, msinum_t msi_num,
1108     pci_msi_valid_state_t msi_valid_state)
1109 {
1110 	uint64_t	ret;
1111 
1112 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setvalid: dip 0x%p msi_num 0x%x "
1113 	    "msi_valid_state 0x%x\n", dip, msi_num, msi_valid_state);
1114 
1115 	if ((ret = hvio_msi_setvalid(DIP_TO_HANDLE(dip),
1116 	    msi_num, msi_valid_state)) != H_EOK) {
1117 		DBG(DBG_LIB_MSI, dip,
1118 		    "hvio_msi_setvalid failed, ret 0x%lx\n", ret);
1119 		return (DDI_FAILURE);
1120 	}
1121 
1122 	return (DDI_SUCCESS);
1123 }
1124 
1125 /*ARGSUSED*/
1126 int
1127 px_lib_msi_getstate(dev_info_t *dip, msinum_t msi_num,
1128     pci_msi_state_t *msi_state)
1129 {
1130 	uint64_t	ret;
1131 
1132 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getstate: dip 0x%p msi_num 0x%x\n",
1133 	    dip, msi_num);
1134 
1135 	if ((ret = hvio_msi_getstate(DIP_TO_HANDLE(dip),
1136 	    msi_num, msi_state)) != H_EOK) {
1137 		DBG(DBG_LIB_MSI, dip,
1138 		    "hvio_msi_getstate failed, ret 0x%lx\n", ret);
1139 		return (DDI_FAILURE);
1140 	}
1141 
1142 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getstate: msi_state 0x%x\n",
1143 	    *msi_state);
1144 
1145 	return (DDI_SUCCESS);
1146 }
1147 
1148 /*ARGSUSED*/
1149 int
1150 px_lib_msi_setstate(dev_info_t *dip, msinum_t msi_num,
1151     pci_msi_state_t msi_state)
1152 {
1153 	uint64_t	ret;
1154 
1155 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setstate: dip 0x%p msi_num 0x%x "
1156 	    "msi_state 0x%x\n", dip, msi_num, msi_state);
1157 
1158 	if ((ret = hvio_msi_setstate(DIP_TO_HANDLE(dip),
1159 	    msi_num, msi_state)) != H_EOK) {
1160 		DBG(DBG_LIB_MSI, dip,
1161 		    "hvio_msi_setstate failed, ret 0x%lx\n", ret);
1162 		return (DDI_FAILURE);
1163 	}
1164 
1165 	return (DDI_SUCCESS);
1166 }
1167 
1168 /*
1169  * MSG Functions:
1170  */
1171 /*ARGSUSED*/
1172 int
1173 px_lib_msg_getmsiq(dev_info_t *dip, pcie_msg_type_t msg_type,
1174     msiqid_t *msiq_id)
1175 {
1176 	uint64_t	ret;
1177 
1178 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_getmsiq: dip 0x%p msg_type 0x%x\n",
1179 	    dip, msg_type);
1180 
1181 	if ((ret = hvio_msg_getmsiq(DIP_TO_HANDLE(dip),
1182 	    msg_type, msiq_id)) != H_EOK) {
1183 		DBG(DBG_LIB_MSG, dip,
1184 		    "hvio_msg_getmsiq failed, ret 0x%lx\n", ret);
1185 		return (DDI_FAILURE);
1186 	}
1187 
1188 	DBG(DBG_LIB_MSI, dip, "px_lib_msg_getmsiq: msiq_id 0x%x\n",
1189 	    *msiq_id);
1190 
1191 	return (DDI_SUCCESS);
1192 }
1193 
1194 /*ARGSUSED*/
1195 int
1196 px_lib_msg_setmsiq(dev_info_t *dip, pcie_msg_type_t msg_type,
1197     msiqid_t msiq_id)
1198 {
1199 	uint64_t	ret;
1200 
1201 	DBG(DBG_LIB_MSG, dip, "px_lib_msi_setstate: dip 0x%p msg_type 0x%x "
1202 	    "msiq_id 0x%x\n", dip, msg_type, msiq_id);
1203 
1204 	if ((ret = hvio_msg_setmsiq(DIP_TO_HANDLE(dip),
1205 	    msg_type, msiq_id)) != H_EOK) {
1206 		DBG(DBG_LIB_MSG, dip,
1207 		    "hvio_msg_setmsiq failed, ret 0x%lx\n", ret);
1208 		return (DDI_FAILURE);
1209 	}
1210 
1211 	return (DDI_SUCCESS);
1212 }
1213 
1214 /*ARGSUSED*/
1215 int
1216 px_lib_msg_getvalid(dev_info_t *dip, pcie_msg_type_t msg_type,
1217     pcie_msg_valid_state_t *msg_valid_state)
1218 {
1219 	uint64_t	ret;
1220 
1221 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_getvalid: dip 0x%p msg_type 0x%x\n",
1222 	    dip, msg_type);
1223 
1224 	if ((ret = hvio_msg_getvalid(DIP_TO_HANDLE(dip), msg_type,
1225 	    msg_valid_state)) != H_EOK) {
1226 		DBG(DBG_LIB_MSG, dip,
1227 		    "hvio_msg_getvalid failed, ret 0x%lx\n", ret);
1228 		return (DDI_FAILURE);
1229 	}
1230 
1231 	DBG(DBG_LIB_MSI, dip, "px_lib_msg_getvalid: msg_valid_state 0x%x\n",
1232 	    *msg_valid_state);
1233 
1234 	return (DDI_SUCCESS);
1235 }
1236 
1237 /*ARGSUSED*/
1238 int
1239 px_lib_msg_setvalid(dev_info_t *dip, pcie_msg_type_t msg_type,
1240     pcie_msg_valid_state_t msg_valid_state)
1241 {
1242 	uint64_t	ret;
1243 
1244 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_setvalid: dip 0x%p msg_type 0x%x "
1245 	    "msg_valid_state 0x%x\n", dip, msg_type, msg_valid_state);
1246 
1247 	if ((ret = hvio_msg_setvalid(DIP_TO_HANDLE(dip), msg_type,
1248 	    msg_valid_state)) != H_EOK) {
1249 		DBG(DBG_LIB_MSG, dip,
1250 		    "hvio_msg_setvalid failed, ret 0x%lx\n", ret);
1251 		return (DDI_FAILURE);
1252 	}
1253 
1254 	return (DDI_SUCCESS);
1255 }
1256 
1257 /*
1258  * Suspend/Resume Functions:
1259  * Currently unsupported by hypervisor
1260  */
1261 int
1262 px_lib_suspend(dev_info_t *dip)
1263 {
1264 	px_t		*px_p = DIP_TO_STATE(dip);
1265 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1266 	px_cb_t		*cb_p = PX2CB(px_p);
1267 	devhandle_t	dev_hdl, xbus_dev_hdl;
1268 	uint64_t	ret = H_EOK;
1269 
1270 	DBG(DBG_DETACH, dip, "px_lib_suspend: dip 0x%p\n", dip);
1271 
1272 	dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_CSR];
1273 	xbus_dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_XBC];
1274 
1275 	if ((ret = hvio_suspend(dev_hdl, pxu_p)) != H_EOK)
1276 		goto fail;
1277 
1278 	if (--cb_p->attachcnt == 0) {
1279 		ret = hvio_cb_suspend(xbus_dev_hdl, pxu_p);
1280 		if (ret != H_EOK)
1281 			cb_p->attachcnt++;
1282 	}
1283 	pxu_p->cpr_flag = PX_ENTERED_CPR;
1284 
1285 fail:
1286 	return ((ret != H_EOK) ? DDI_FAILURE: DDI_SUCCESS);
1287 }
1288 
1289 void
1290 px_lib_resume(dev_info_t *dip)
1291 {
1292 	px_t		*px_p = DIP_TO_STATE(dip);
1293 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1294 	px_cb_t		*cb_p = PX2CB(px_p);
1295 	devhandle_t	dev_hdl, xbus_dev_hdl;
1296 	devino_t	pec_ino = px_p->px_inos[PX_INTR_PEC];
1297 	devino_t	xbc_ino = px_p->px_inos[PX_INTR_XBC];
1298 
1299 	DBG(DBG_ATTACH, dip, "px_lib_resume: dip 0x%p\n", dip);
1300 
1301 	dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_CSR];
1302 	xbus_dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_XBC];
1303 
1304 	if (++cb_p->attachcnt == 1)
1305 		hvio_cb_resume(dev_hdl, xbus_dev_hdl, xbc_ino, pxu_p);
1306 
1307 	hvio_resume(dev_hdl, pec_ino, pxu_p);
1308 }
1309 
1310 /*
1311  * Generate a unique Oberon UBC ID based on the Logicial System Board and
1312  * the IO Channel from the portid property field.
1313  */
1314 static uint64_t
1315 oberon_get_ubc_id(dev_info_t *dip)
1316 {
1317 	px_t	*px_p = DIP_TO_STATE(dip);
1318 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1319 	uint64_t	ubc_id;
1320 
1321 	/*
1322 	 * Generate a unique 6 bit UBC ID using the 2 IO_Channel#[1:0] bits and
1323 	 * the 4 LSB_ID[3:0] bits from the Oberon's portid property.
1324 	 */
1325 	ubc_id = (((pxu_p->portid >> OBERON_PORT_ID_IOC) &
1326 	    OBERON_PORT_ID_IOC_MASK) | (((pxu_p->portid >>
1327 	    OBERON_PORT_ID_LSB) & OBERON_PORT_ID_LSB_MASK)
1328 	    << OBERON_UBC_ID_LSB));
1329 
1330 	return (ubc_id);
1331 }
1332 
1333 /*
1334  * Oberon does not have a UBC scratch register, so alloc an array of scratch
1335  * registers when needed and use a unique UBC ID as an index. This code
1336  * can be simplified if we use a pre-allocated array. They are currently
1337  * being dynamically allocated because it's only needed by the Oberon.
1338  */
1339 static void
1340 oberon_set_cb(dev_info_t *dip, uint64_t val)
1341 {
1342 	uint64_t	ubc_id;
1343 
1344 	if (px_oberon_ubc_scratch_regs == NULL)
1345 		px_oberon_ubc_scratch_regs =
1346 		    (uint64_t *)kmem_zalloc(sizeof (uint64_t)*
1347 		    OBERON_UBC_ID_MAX, KM_SLEEP);
1348 
1349 	ubc_id = oberon_get_ubc_id(dip);
1350 
1351 	px_oberon_ubc_scratch_regs[ubc_id] = val;
1352 
1353 	/*
1354 	 * Check if any scratch registers are still in use. If all scratch
1355 	 * registers are currently set to zero, then deallocate the scratch
1356 	 * register array.
1357 	 */
1358 	for (ubc_id = 0; ubc_id < OBERON_UBC_ID_MAX; ubc_id++) {
1359 		if (px_oberon_ubc_scratch_regs[ubc_id] != NULL)
1360 			return;
1361 	}
1362 
1363 	/*
1364 	 * All scratch registers are set to zero so deallocate the scratch
1365 	 * register array and set the pointer to NULL.
1366 	 */
1367 	kmem_free(px_oberon_ubc_scratch_regs,
1368 	    (sizeof (uint64_t)*OBERON_UBC_ID_MAX));
1369 
1370 	px_oberon_ubc_scratch_regs = NULL;
1371 }
1372 
1373 /*
1374  * Oberon does not have a UBC scratch register, so use an allocated array of
1375  * scratch registers and use the unique UBC ID as an index into that array.
1376  */
1377 static uint64_t
1378 oberon_get_cb(dev_info_t *dip)
1379 {
1380 	uint64_t	ubc_id;
1381 
1382 	if (px_oberon_ubc_scratch_regs == NULL)
1383 		return (0);
1384 
1385 	ubc_id = oberon_get_ubc_id(dip);
1386 
1387 	return (px_oberon_ubc_scratch_regs[ubc_id]);
1388 }
1389 
1390 /*
1391  * Misc Functions:
1392  * Currently unsupported by hypervisor
1393  */
1394 static uint64_t
1395 px_get_cb(dev_info_t *dip)
1396 {
1397 	px_t	*px_p = DIP_TO_STATE(dip);
1398 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1399 
1400 	/*
1401 	 * Oberon does not currently have Scratchpad registers.
1402 	 */
1403 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON)
1404 		return (oberon_get_cb(dip));
1405 
1406 	return (CSR_XR((caddr_t)pxu_p->px_address[PX_REG_XBC], JBUS_SCRATCH_1));
1407 }
1408 
1409 static void
1410 px_set_cb(dev_info_t *dip, uint64_t val)
1411 {
1412 	px_t	*px_p = DIP_TO_STATE(dip);
1413 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1414 
1415 	/*
1416 	 * Oberon does not currently have Scratchpad registers.
1417 	 */
1418 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON) {
1419 		oberon_set_cb(dip, val);
1420 		return;
1421 	}
1422 
1423 	CSR_XS((caddr_t)pxu_p->px_address[PX_REG_XBC], JBUS_SCRATCH_1, val);
1424 }
1425 
1426 /*ARGSUSED*/
1427 int
1428 px_lib_map_vconfig(dev_info_t *dip,
1429 	ddi_map_req_t *mp, pci_config_offset_t off,
1430 		pci_regspec_t *rp, caddr_t *addrp)
1431 {
1432 	/*
1433 	 * No special config space access services in this layer.
1434 	 */
1435 	return (DDI_FAILURE);
1436 }
1437 
1438 void
1439 px_lib_map_attr_check(ddi_map_req_t *mp)
1440 {
1441 	ddi_acc_hdl_t *hp = mp->map_handlep;
1442 
1443 	/* fire does not accept byte masks from PIO store merge */
1444 	if (hp->ah_acc.devacc_attr_dataorder == DDI_STORECACHING_OK_ACC)
1445 		hp->ah_acc.devacc_attr_dataorder = DDI_STRICTORDER_ACC;
1446 }
1447 
1448 /* This function is called only by poke, caut put and pxtool poke. */
1449 void
1450 px_lib_clr_errs(px_t *px_p, dev_info_t *rdip, uint64_t addr)
1451 {
1452 	px_pec_t	*pec_p = px_p->px_pec_p;
1453 	dev_info_t	*rpdip = px_p->px_dip;
1454 	int		rc_err, fab_err, i;
1455 	int		acctype = pec_p->pec_safeacc_type;
1456 	ddi_fm_error_t	derr;
1457 	px_ranges_t	*ranges_p;
1458 	int		range_len;
1459 	uint32_t	addr_high, addr_low;
1460 	pcie_req_id_t	bdf = 0;
1461 
1462 	/* Create the derr */
1463 	bzero(&derr, sizeof (ddi_fm_error_t));
1464 	derr.fme_version = DDI_FME_VERSION;
1465 	derr.fme_ena = fm_ena_generate(0, FM_ENA_FMT1);
1466 	derr.fme_flag = acctype;
1467 
1468 	if (acctype == DDI_FM_ERR_EXPECTED) {
1469 		derr.fme_status = DDI_FM_NONFATAL;
1470 		ndi_fm_acc_err_set(pec_p->pec_acc_hdl, &derr);
1471 	}
1472 
1473 	if (px_fm_enter(px_p) != DDI_SUCCESS)
1474 		return;
1475 
1476 	/* send ereport/handle/clear fire registers */
1477 	rc_err = px_err_cmn_intr(px_p, &derr, PX_LIB_CALL, PX_FM_BLOCK_ALL);
1478 
1479 	/* Figure out if this is a cfg or mem32 access */
1480 	addr_high = (uint32_t)(addr >> 32);
1481 	addr_low = (uint32_t)addr;
1482 	range_len = px_p->px_ranges_length / sizeof (px_ranges_t);
1483 	i = 0;
1484 	for (ranges_p = px_p->px_ranges_p; i < range_len; i++, ranges_p++) {
1485 		if (ranges_p->parent_high == addr_high) {
1486 			switch (ranges_p->child_high & PCI_ADDR_MASK) {
1487 			case PCI_ADDR_CONFIG:
1488 				bdf = (pcie_req_id_t)(addr_low >> 12);
1489 				addr_low = 0;
1490 				break;
1491 			case PCI_ADDR_MEM32:
1492 				if (rdip)
1493 					bdf = PCI_GET_BDF(rdip);
1494 				else
1495 					bdf = NULL;
1496 				break;
1497 			}
1498 			break;
1499 		}
1500 	}
1501 
1502 	px_rp_en_q(px_p, bdf, addr_low, NULL);
1503 
1504 	/*
1505 	 * XXX - Current code scans the fabric for all px_tool accesses.
1506 	 * In future, do not scan fabric for px_tool access to IO Root Nexus
1507 	 */
1508 	fab_err = px_scan_fabric(px_p, rpdip, &derr);
1509 
1510 	px_err_panic(rc_err, PX_RC, fab_err, B_TRUE);
1511 	px_fm_exit(px_p);
1512 	px_err_panic(rc_err, PX_RC, fab_err, B_FALSE);
1513 }
1514 
1515 #ifdef  DEBUG
1516 int	px_peekfault_cnt = 0;
1517 int	px_pokefault_cnt = 0;
1518 #endif  /* DEBUG */
1519 
1520 /*ARGSUSED*/
1521 static int
1522 px_lib_do_poke(dev_info_t *dip, dev_info_t *rdip,
1523     peekpoke_ctlops_t *in_args)
1524 {
1525 	px_t *px_p = DIP_TO_STATE(dip);
1526 	px_pec_t *pec_p = px_p->px_pec_p;
1527 	int err = DDI_SUCCESS;
1528 	on_trap_data_t otd;
1529 
1530 	mutex_enter(&pec_p->pec_pokefault_mutex);
1531 	pec_p->pec_ontrap_data = &otd;
1532 	pec_p->pec_safeacc_type = DDI_FM_ERR_POKE;
1533 
1534 	/* Set up protected environment. */
1535 	if (!on_trap(&otd, OT_DATA_ACCESS)) {
1536 		uintptr_t tramp = otd.ot_trampoline;
1537 
1538 		otd.ot_trampoline = (uintptr_t)&poke_fault;
1539 		err = do_poke(in_args->size, (void *)in_args->dev_addr,
1540 		    (void *)in_args->host_addr);
1541 		otd.ot_trampoline = tramp;
1542 	} else
1543 		err = DDI_FAILURE;
1544 
1545 	px_lib_clr_errs(px_p, rdip, in_args->dev_addr);
1546 
1547 	if (otd.ot_trap & OT_DATA_ACCESS)
1548 		err = DDI_FAILURE;
1549 
1550 	/* Take down protected environment. */
1551 	no_trap();
1552 
1553 	pec_p->pec_ontrap_data = NULL;
1554 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1555 	mutex_exit(&pec_p->pec_pokefault_mutex);
1556 
1557 #ifdef  DEBUG
1558 	if (err == DDI_FAILURE)
1559 		px_pokefault_cnt++;
1560 #endif
1561 	return (err);
1562 }
1563 
1564 /*ARGSUSED*/
1565 static int
1566 px_lib_do_caut_put(dev_info_t *dip, dev_info_t *rdip,
1567     peekpoke_ctlops_t *cautacc_ctlops_arg)
1568 {
1569 	size_t size = cautacc_ctlops_arg->size;
1570 	uintptr_t dev_addr = cautacc_ctlops_arg->dev_addr;
1571 	uintptr_t host_addr = cautacc_ctlops_arg->host_addr;
1572 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)cautacc_ctlops_arg->handle;
1573 	size_t repcount = cautacc_ctlops_arg->repcount;
1574 	uint_t flags = cautacc_ctlops_arg->flags;
1575 
1576 	px_t *px_p = DIP_TO_STATE(dip);
1577 	px_pec_t *pec_p = px_p->px_pec_p;
1578 	int err = DDI_SUCCESS;
1579 
1580 	/*
1581 	 * Note that i_ndi_busop_access_enter ends up grabbing the pokefault
1582 	 * mutex.
1583 	 */
1584 	i_ndi_busop_access_enter(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1585 
1586 	pec_p->pec_ontrap_data = (on_trap_data_t *)hp->ahi_err->err_ontrap;
1587 	pec_p->pec_safeacc_type = DDI_FM_ERR_EXPECTED;
1588 	hp->ahi_err->err_expected = DDI_FM_ERR_EXPECTED;
1589 
1590 	if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1591 		for (; repcount; repcount--) {
1592 			switch (size) {
1593 
1594 			case sizeof (uint8_t):
1595 				i_ddi_put8(hp, (uint8_t *)dev_addr,
1596 				    *(uint8_t *)host_addr);
1597 				break;
1598 
1599 			case sizeof (uint16_t):
1600 				i_ddi_put16(hp, (uint16_t *)dev_addr,
1601 				    *(uint16_t *)host_addr);
1602 				break;
1603 
1604 			case sizeof (uint32_t):
1605 				i_ddi_put32(hp, (uint32_t *)dev_addr,
1606 				    *(uint32_t *)host_addr);
1607 				break;
1608 
1609 			case sizeof (uint64_t):
1610 				i_ddi_put64(hp, (uint64_t *)dev_addr,
1611 				    *(uint64_t *)host_addr);
1612 				break;
1613 			}
1614 
1615 			host_addr += size;
1616 
1617 			if (flags == DDI_DEV_AUTOINCR)
1618 				dev_addr += size;
1619 
1620 			px_lib_clr_errs(px_p, rdip, dev_addr);
1621 
1622 			if (pec_p->pec_ontrap_data->ot_trap & OT_DATA_ACCESS) {
1623 				err = DDI_FAILURE;
1624 #ifdef  DEBUG
1625 				px_pokefault_cnt++;
1626 #endif
1627 				break;
1628 			}
1629 		}
1630 	}
1631 
1632 	i_ddi_notrap((ddi_acc_handle_t)hp);
1633 	pec_p->pec_ontrap_data = NULL;
1634 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1635 	i_ndi_busop_access_exit(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1636 	hp->ahi_err->err_expected = DDI_FM_ERR_UNEXPECTED;
1637 
1638 	return (err);
1639 }
1640 
1641 
1642 int
1643 px_lib_ctlops_poke(dev_info_t *dip, dev_info_t *rdip,
1644     peekpoke_ctlops_t *in_args)
1645 {
1646 	return (in_args->handle ? px_lib_do_caut_put(dip, rdip, in_args) :
1647 	    px_lib_do_poke(dip, rdip, in_args));
1648 }
1649 
1650 
1651 /*ARGSUSED*/
1652 static int
1653 px_lib_do_peek(dev_info_t *dip, peekpoke_ctlops_t *in_args)
1654 {
1655 	px_t *px_p = DIP_TO_STATE(dip);
1656 	px_pec_t *pec_p = px_p->px_pec_p;
1657 	int err = DDI_SUCCESS;
1658 	on_trap_data_t otd;
1659 
1660 	mutex_enter(&pec_p->pec_pokefault_mutex);
1661 	if (px_fm_enter(px_p) != DDI_SUCCESS)
1662 		return (DDI_FAILURE);
1663 	pec_p->pec_safeacc_type = DDI_FM_ERR_PEEK;
1664 	px_fm_exit(px_p);
1665 
1666 	if (!on_trap(&otd, OT_DATA_ACCESS)) {
1667 		uintptr_t tramp = otd.ot_trampoline;
1668 
1669 		otd.ot_trampoline = (uintptr_t)&peek_fault;
1670 		err = do_peek(in_args->size, (void *)in_args->dev_addr,
1671 		    (void *)in_args->host_addr);
1672 		otd.ot_trampoline = tramp;
1673 	} else
1674 		err = DDI_FAILURE;
1675 
1676 	no_trap();
1677 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1678 	mutex_exit(&pec_p->pec_pokefault_mutex);
1679 
1680 #ifdef  DEBUG
1681 	if (err == DDI_FAILURE)
1682 		px_peekfault_cnt++;
1683 #endif
1684 	return (err);
1685 }
1686 
1687 
1688 static int
1689 px_lib_do_caut_get(dev_info_t *dip, peekpoke_ctlops_t *cautacc_ctlops_arg)
1690 {
1691 	size_t size = cautacc_ctlops_arg->size;
1692 	uintptr_t dev_addr = cautacc_ctlops_arg->dev_addr;
1693 	uintptr_t host_addr = cautacc_ctlops_arg->host_addr;
1694 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)cautacc_ctlops_arg->handle;
1695 	size_t repcount = cautacc_ctlops_arg->repcount;
1696 	uint_t flags = cautacc_ctlops_arg->flags;
1697 
1698 	px_t *px_p = DIP_TO_STATE(dip);
1699 	px_pec_t *pec_p = px_p->px_pec_p;
1700 	int err = DDI_SUCCESS;
1701 
1702 	/*
1703 	 * Note that i_ndi_busop_access_enter ends up grabbing the pokefault
1704 	 * mutex.
1705 	 */
1706 	i_ndi_busop_access_enter(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1707 
1708 	pec_p->pec_ontrap_data = (on_trap_data_t *)hp->ahi_err->err_ontrap;
1709 	pec_p->pec_safeacc_type = DDI_FM_ERR_EXPECTED;
1710 	hp->ahi_err->err_expected = DDI_FM_ERR_EXPECTED;
1711 
1712 	if (repcount == 1) {
1713 		if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1714 			i_ddi_caut_get(size, (void *)dev_addr,
1715 			    (void *)host_addr);
1716 		} else {
1717 			int i;
1718 			uint8_t *ff_addr = (uint8_t *)host_addr;
1719 			for (i = 0; i < size; i++)
1720 				*ff_addr++ = 0xff;
1721 
1722 			err = DDI_FAILURE;
1723 #ifdef  DEBUG
1724 			px_peekfault_cnt++;
1725 #endif
1726 		}
1727 	} else {
1728 		if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1729 			for (; repcount; repcount--) {
1730 				i_ddi_caut_get(size, (void *)dev_addr,
1731 				    (void *)host_addr);
1732 
1733 				host_addr += size;
1734 
1735 				if (flags == DDI_DEV_AUTOINCR)
1736 					dev_addr += size;
1737 			}
1738 		} else {
1739 			err = DDI_FAILURE;
1740 #ifdef  DEBUG
1741 			px_peekfault_cnt++;
1742 #endif
1743 		}
1744 	}
1745 
1746 	i_ddi_notrap((ddi_acc_handle_t)hp);
1747 	pec_p->pec_ontrap_data = NULL;
1748 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1749 	i_ndi_busop_access_exit(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1750 	hp->ahi_err->err_expected = DDI_FM_ERR_UNEXPECTED;
1751 
1752 	return (err);
1753 }
1754 
1755 /*ARGSUSED*/
1756 int
1757 px_lib_ctlops_peek(dev_info_t *dip, dev_info_t *rdip,
1758     peekpoke_ctlops_t *in_args, void *result)
1759 {
1760 	result = (void *)in_args->host_addr;
1761 	return (in_args->handle ? px_lib_do_caut_get(dip, in_args) :
1762 	    px_lib_do_peek(dip, in_args));
1763 }
1764 
1765 /*
1766  * implements PPM interface
1767  */
1768 int
1769 px_lib_pmctl(int cmd, px_t *px_p)
1770 {
1771 	ASSERT((cmd & ~PPMREQ_MASK) == PPMREQ);
1772 	switch (cmd) {
1773 	case PPMREQ_PRE_PWR_OFF:
1774 		/*
1775 		 * Currently there is no device power management for
1776 		 * the root complex (fire). When there is we need to make
1777 		 * sure that it is at full power before trying to send the
1778 		 * PME_Turn_Off message.
1779 		 */
1780 		DBG(DBG_PWR, px_p->px_dip,
1781 		    "ioctl: request to send PME_Turn_Off\n");
1782 		return (px_goto_l23ready(px_p));
1783 
1784 	case PPMREQ_PRE_PWR_ON:
1785 		DBG(DBG_PWR, px_p->px_dip, "ioctl: PRE_PWR_ON request\n");
1786 		return (px_pre_pwron_check(px_p));
1787 
1788 	case PPMREQ_POST_PWR_ON:
1789 		DBG(DBG_PWR, px_p->px_dip, "ioctl: POST_PWR_ON request\n");
1790 		return (px_goto_l0(px_p));
1791 
1792 	default:
1793 		return (DDI_FAILURE);
1794 	}
1795 }
1796 
1797 /*
1798  * sends PME_Turn_Off message to put the link in L2/L3 ready state.
1799  * called by px_ioctl.
1800  * returns DDI_SUCCESS or DDI_FAILURE
1801  * 1. Wait for link to be in L1 state (link status reg)
1802  * 2. write to PME_Turn_off reg to boradcast
1803  * 3. set timeout
1804  * 4. If timeout, return failure.
1805  * 5. If PM_TO_Ack, wait till link is in L2/L3 ready
1806  */
1807 static int
1808 px_goto_l23ready(px_t *px_p)
1809 {
1810 	pcie_pwr_t	*pwr_p;
1811 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1812 	caddr_t	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
1813 	int		ret = DDI_SUCCESS;
1814 	clock_t		end, timeleft;
1815 	int		mutex_held = 1;
1816 
1817 	/* If no PM info, return failure */
1818 	if (!PCIE_PMINFO(px_p->px_dip) ||
1819 	    !(pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)))
1820 		return (DDI_FAILURE);
1821 
1822 	mutex_enter(&pwr_p->pwr_lock);
1823 	mutex_enter(&px_p->px_l23ready_lock);
1824 	/* Clear the PME_To_ACK receieved flag */
1825 	px_p->px_pm_flags &= ~PX_PMETOACK_RECVD;
1826 	/*
1827 	 * When P25 is the downstream device, after receiving
1828 	 * PME_To_ACK, fire will go to Detect state, which causes
1829 	 * the link down event. Inform FMA that this is expected.
1830 	 * In case of all other cards complaint with the pci express
1831 	 * spec, this will happen when the power is re-applied. FMA
1832 	 * code will clear this flag after one instance of LDN. Since
1833 	 * there will not be a LDN event for the spec compliant cards,
1834 	 * we need to clear the flag after receiving PME_To_ACK.
1835 	 */
1836 	px_p->px_pm_flags |= PX_LDN_EXPECTED;
1837 	if (px_send_pme_turnoff(csr_base) != DDI_SUCCESS) {
1838 		ret = DDI_FAILURE;
1839 		goto l23ready_done;
1840 	}
1841 	px_p->px_pm_flags |= PX_PME_TURNOFF_PENDING;
1842 
1843 	end = ddi_get_lbolt() + drv_usectohz(px_pme_to_ack_timeout);
1844 	while (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1845 		timeleft = cv_timedwait(&px_p->px_l23ready_cv,
1846 		    &px_p->px_l23ready_lock, end);
1847 		/*
1848 		 * if cv_timedwait returns -1, it is either
1849 		 * 1) timed out or
1850 		 * 2) there was a pre-mature wakeup but by the time
1851 		 * cv_timedwait is called again end < lbolt i.e.
1852 		 * end is in the past.
1853 		 * 3) By the time we make first cv_timedwait call,
1854 		 * end < lbolt is true.
1855 		 */
1856 		if (timeleft == -1)
1857 			break;
1858 	}
1859 	if (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1860 		/*
1861 		 * Either timedout or interrupt didn't get a
1862 		 * chance to grab the mutex and set the flag.
1863 		 * release the mutex and delay for sometime.
1864 		 * This will 1) give a chance for interrupt to
1865 		 * set the flag 2) creates a delay between two
1866 		 * consequetive requests.
1867 		 */
1868 		mutex_exit(&px_p->px_l23ready_lock);
1869 		delay(drv_usectohz(50 * PX_MSEC_TO_USEC));
1870 		mutex_held = 0;
1871 		if (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1872 			ret = DDI_FAILURE;
1873 			DBG(DBG_PWR, px_p->px_dip, " Timed out while waiting"
1874 			    " for PME_TO_ACK\n");
1875 		}
1876 	}
1877 	px_p->px_pm_flags &=
1878 	    ~(PX_PME_TURNOFF_PENDING | PX_PMETOACK_RECVD | PX_LDN_EXPECTED);
1879 
1880 l23ready_done:
1881 	if (mutex_held)
1882 		mutex_exit(&px_p->px_l23ready_lock);
1883 	/*
1884 	 * Wait till link is in L1 idle, if sending PME_Turn_Off
1885 	 * was succesful.
1886 	 */
1887 	if (ret == DDI_SUCCESS) {
1888 		if (px_link_wait4l1idle(csr_base) != DDI_SUCCESS) {
1889 			DBG(DBG_PWR, px_p->px_dip, " Link is not at L1"
1890 			    " even though we received PME_To_ACK.\n");
1891 			/*
1892 			 * Workaround for hardware bug with P25.
1893 			 * Due to a hardware bug with P25, link state
1894 			 * will be Detect state rather than L1 after
1895 			 * link is transitioned to L23Ready state. Since
1896 			 * we don't know whether link is L23ready state
1897 			 * without Fire's state being L1_idle, we delay
1898 			 * here just to make sure that we wait till link
1899 			 * is transitioned to L23Ready state.
1900 			 */
1901 			delay(drv_usectohz(100 * PX_MSEC_TO_USEC));
1902 		}
1903 		pwr_p->pwr_link_lvl = PM_LEVEL_L3;
1904 
1905 	}
1906 	mutex_exit(&pwr_p->pwr_lock);
1907 	return (ret);
1908 }
1909 
1910 /*
1911  * Message interrupt handler intended to be shared for both
1912  * PME and PME_TO_ACK msg handling, currently only handles
1913  * PME_To_ACK message.
1914  */
1915 uint_t
1916 px_pmeq_intr(caddr_t arg)
1917 {
1918 	px_t	*px_p = (px_t *)arg;
1919 
1920 	DBG(DBG_PWR, px_p->px_dip, " PME_To_ACK received \n");
1921 	mutex_enter(&px_p->px_l23ready_lock);
1922 	cv_broadcast(&px_p->px_l23ready_cv);
1923 	if (px_p->px_pm_flags & PX_PME_TURNOFF_PENDING) {
1924 		px_p->px_pm_flags |= PX_PMETOACK_RECVD;
1925 	} else {
1926 		/*
1927 		 * This maybe the second ack received. If so then,
1928 		 * we should be receiving it during wait4L1 stage.
1929 		 */
1930 		px_p->px_pmetoack_ignored++;
1931 	}
1932 	mutex_exit(&px_p->px_l23ready_lock);
1933 	return (DDI_INTR_CLAIMED);
1934 }
1935 
1936 static int
1937 px_pre_pwron_check(px_t *px_p)
1938 {
1939 	pcie_pwr_t	*pwr_p;
1940 
1941 	/* If no PM info, return failure */
1942 	if (!PCIE_PMINFO(px_p->px_dip) ||
1943 	    !(pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)))
1944 		return (DDI_FAILURE);
1945 
1946 	/*
1947 	 * For the spec compliant downstream cards link down
1948 	 * is expected when the device is powered on.
1949 	 */
1950 	px_p->px_pm_flags |= PX_LDN_EXPECTED;
1951 	return (pwr_p->pwr_link_lvl == PM_LEVEL_L3 ? DDI_SUCCESS : DDI_FAILURE);
1952 }
1953 
1954 static int
1955 px_goto_l0(px_t *px_p)
1956 {
1957 	pcie_pwr_t	*pwr_p;
1958 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1959 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
1960 	int		ret = DDI_SUCCESS;
1961 	uint64_t	time_spent = 0;
1962 
1963 	/* If no PM info, return failure */
1964 	if (!PCIE_PMINFO(px_p->px_dip) ||
1965 	    !(pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)))
1966 		return (DDI_FAILURE);
1967 
1968 	mutex_enter(&pwr_p->pwr_lock);
1969 	/*
1970 	 * The following link retrain activity will cause LDN and LUP event.
1971 	 * Receiving LDN prior to receiving LUP is expected, not an error in
1972 	 * this case.  Receiving LUP indicates link is fully up to support
1973 	 * powering up down stream device, and of course any further LDN and
1974 	 * LUP outside this context will be error.
1975 	 */
1976 	px_p->px_lup_pending = 1;
1977 	if (px_link_retrain(csr_base) != DDI_SUCCESS) {
1978 		ret = DDI_FAILURE;
1979 		goto l0_done;
1980 	}
1981 
1982 	/* LUP event takes the order of 15ms amount of time to occur */
1983 	for (; px_p->px_lup_pending && (time_spent < px_lup_poll_to);
1984 	    time_spent += px_lup_poll_interval)
1985 		drv_usecwait(px_lup_poll_interval);
1986 	if (px_p->px_lup_pending)
1987 		ret = DDI_FAILURE;
1988 l0_done:
1989 	px_enable_detect_quiet(csr_base);
1990 	if (ret == DDI_SUCCESS)
1991 		pwr_p->pwr_link_lvl = PM_LEVEL_L0;
1992 	mutex_exit(&pwr_p->pwr_lock);
1993 	return (ret);
1994 }
1995 
1996 /*
1997  * Extract the drivers binding name to identify which chip we're binding to.
1998  * Whenever a new bus bridge is created, the driver alias entry should be
1999  * added here to identify the device if needed.  If a device isn't added,
2000  * the identity defaults to PX_CHIP_UNIDENTIFIED.
2001  */
2002 static uint32_t
2003 px_identity_init(px_t *px_p)
2004 {
2005 	dev_info_t	*dip = px_p->px_dip;
2006 	char		*name = ddi_binding_name(dip);
2007 	uint32_t	revision = 0;
2008 
2009 	revision = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2010 	    "module-revision#", 0);
2011 
2012 	/* Check for Fire driver binding name */
2013 	if (strcmp(name, "pciex108e,80f0") == 0) {
2014 		DBG(DBG_ATTACH, dip, "px_identity_init: %s%d: "
2015 		    "(FIRE), module-revision %d\n", NAMEINST(dip),
2016 		    revision);
2017 
2018 		return ((revision >= FIRE_MOD_REV_20) ?
2019 		    PX_CHIP_FIRE : PX_CHIP_UNIDENTIFIED);
2020 	}
2021 
2022 	/* Check for Oberon driver binding name */
2023 	if (strcmp(name, "pciex108e,80f8") == 0) {
2024 		DBG(DBG_ATTACH, dip, "px_identity_init: %s%d: "
2025 		    "(OBERON), module-revision %d\n", NAMEINST(dip),
2026 		    revision);
2027 
2028 		return (PX_CHIP_OBERON);
2029 	}
2030 
2031 	DBG(DBG_ATTACH, dip, "%s%d: Unknown PCI Express Host bridge %s %x\n",
2032 	    ddi_driver_name(dip), ddi_get_instance(dip), name, revision);
2033 
2034 	return (PX_CHIP_UNIDENTIFIED);
2035 }
2036 
2037 int
2038 px_err_add_intr(px_fault_t *px_fault_p)
2039 {
2040 	dev_info_t	*dip = px_fault_p->px_fh_dip;
2041 	px_t		*px_p = DIP_TO_STATE(dip);
2042 
2043 	VERIFY(add_ivintr(px_fault_p->px_fh_sysino, PX_ERR_PIL,
2044 	    (intrfunc)px_fault_p->px_err_func, (caddr_t)px_fault_p,
2045 	    NULL, NULL) == 0);
2046 
2047 	px_ib_intr_enable(px_p, intr_dist_cpuid(), px_fault_p->px_intr_ino);
2048 
2049 	return (DDI_SUCCESS);
2050 }
2051 
2052 void
2053 px_err_rem_intr(px_fault_t *px_fault_p)
2054 {
2055 	dev_info_t	*dip = px_fault_p->px_fh_dip;
2056 	px_t		*px_p = DIP_TO_STATE(dip);
2057 
2058 	px_ib_intr_disable(px_p->px_ib_p, px_fault_p->px_intr_ino,
2059 	    IB_INTR_WAIT);
2060 
2061 	VERIFY(rem_ivintr(px_fault_p->px_fh_sysino, PX_ERR_PIL) == 0);
2062 }
2063 
2064 /*
2065  * px_cb_intr_redist() - sun4u only, CB interrupt redistribution
2066  */
2067 void
2068 px_cb_intr_redist(void *arg)
2069 {
2070 	px_cb_t		*cb_p = (px_cb_t *)arg;
2071 	px_cb_list_t	*pxl;
2072 	px_t		*pxp = NULL;
2073 	px_fault_t	*f_p = NULL;
2074 	uint32_t	new_cpuid;
2075 	intr_valid_state_t	enabled = 0;
2076 
2077 	mutex_enter(&cb_p->cb_mutex);
2078 
2079 	pxl = cb_p->pxl;
2080 	if (!pxl)
2081 		goto cb_done;
2082 
2083 	pxp = pxl->pxp;
2084 	f_p = &pxp->px_cb_fault;
2085 	for (; pxl && (f_p->px_fh_sysino != cb_p->sysino); ) {
2086 		pxl = pxl->next;
2087 		pxp = pxl->pxp;
2088 		f_p = &pxp->px_cb_fault;
2089 	}
2090 	if (pxl == NULL)
2091 		goto cb_done;
2092 
2093 	new_cpuid =  intr_dist_cpuid();
2094 	if (new_cpuid == cb_p->cpuid)
2095 		goto cb_done;
2096 
2097 	if ((px_lib_intr_getvalid(pxp->px_dip, f_p->px_fh_sysino, &enabled)
2098 	    != DDI_SUCCESS) || !enabled) {
2099 		DBG(DBG_IB, pxp->px_dip, "px_cb_intr_redist: CB not enabled, "
2100 		    "sysino(0x%x)\n", f_p->px_fh_sysino);
2101 		goto cb_done;
2102 	}
2103 
2104 	PX_INTR_DISABLE(pxp->px_dip, f_p->px_fh_sysino);
2105 
2106 	cb_p->cpuid = new_cpuid;
2107 	cb_p->sysino = f_p->px_fh_sysino;
2108 	PX_INTR_ENABLE(pxp->px_dip, cb_p->sysino, cb_p->cpuid);
2109 
2110 cb_done:
2111 	mutex_exit(&cb_p->cb_mutex);
2112 }
2113 
2114 /*
2115  * px_cb_add_intr() - Called from attach(9E) to create CB if not yet
2116  * created, to add CB interrupt vector always, but enable only once.
2117  */
2118 int
2119 px_cb_add_intr(px_fault_t *fault_p)
2120 {
2121 	px_t		*px_p = DIP_TO_STATE(fault_p->px_fh_dip);
2122 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2123 	px_cb_t		*cb_p = (px_cb_t *)px_get_cb(fault_p->px_fh_dip);
2124 	px_cb_list_t	*pxl, *pxl_new;
2125 	boolean_t	is_proxy = B_FALSE;
2126 
2127 	/* create cb */
2128 	if (cb_p == NULL) {
2129 		cb_p = kmem_zalloc(sizeof (px_cb_t), KM_SLEEP);
2130 
2131 		mutex_init(&cb_p->cb_mutex, NULL, MUTEX_DRIVER,
2132 		    (void *) ipltospl(FM_ERR_PIL));
2133 
2134 		cb_p->px_cb_func = px_cb_intr;
2135 		pxu_p->px_cb_p = cb_p;
2136 		px_set_cb(fault_p->px_fh_dip, (uint64_t)cb_p);
2137 
2138 		/* px_lib_dev_init allows only FIRE and OBERON */
2139 		px_err_reg_enable(
2140 		    (pxu_p->chip_type == PX_CHIP_FIRE) ?
2141 		    PX_ERR_JBC : PX_ERR_UBC,
2142 		    pxu_p->px_address[PX_REG_XBC]);
2143 	} else
2144 		pxu_p->px_cb_p = cb_p;
2145 
2146 	/* register cb interrupt */
2147 	VERIFY(add_ivintr(fault_p->px_fh_sysino, PX_ERR_PIL,
2148 	    (intrfunc)cb_p->px_cb_func, (caddr_t)cb_p, NULL, NULL) == 0);
2149 
2150 
2151 	/* update cb list */
2152 	mutex_enter(&cb_p->cb_mutex);
2153 	if (cb_p->pxl == NULL) {
2154 		is_proxy = B_TRUE;
2155 		pxl = kmem_zalloc(sizeof (px_cb_list_t), KM_SLEEP);
2156 		pxl->pxp = px_p;
2157 		cb_p->pxl = pxl;
2158 		cb_p->sysino = fault_p->px_fh_sysino;
2159 		cb_p->cpuid = intr_dist_cpuid();
2160 	} else {
2161 		/*
2162 		 * Find the last pxl or
2163 		 * stop short at encountering a redundent entry, or
2164 		 * both.
2165 		 */
2166 		pxl = cb_p->pxl;
2167 		for (; !(pxl->pxp == px_p) && pxl->next; pxl = pxl->next) {};
2168 		ASSERT(pxl->pxp != px_p);
2169 
2170 		/* add to linked list */
2171 		pxl_new = kmem_zalloc(sizeof (px_cb_list_t), KM_SLEEP);
2172 		pxl_new->pxp = px_p;
2173 		pxl->next = pxl_new;
2174 	}
2175 	cb_p->attachcnt++;
2176 	mutex_exit(&cb_p->cb_mutex);
2177 
2178 	if (is_proxy) {
2179 		/* add to interrupt redistribution list */
2180 		intr_dist_add(px_cb_intr_redist, cb_p);
2181 
2182 		/* enable cb hw interrupt */
2183 		px_ib_intr_enable(px_p, cb_p->cpuid, fault_p->px_intr_ino);
2184 	}
2185 
2186 	return (DDI_SUCCESS);
2187 }
2188 
2189 /*
2190  * px_cb_rem_intr() - Called from detach(9E) to remove its CB
2191  * interrupt vector, to shift proxy to the next available px,
2192  * or disable CB interrupt when itself is the last.
2193  */
2194 void
2195 px_cb_rem_intr(px_fault_t *fault_p)
2196 {
2197 	px_t		*px_p = DIP_TO_STATE(fault_p->px_fh_dip), *pxp;
2198 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2199 	px_cb_t		*cb_p = PX2CB(px_p);
2200 	px_cb_list_t	*pxl, *prev;
2201 	px_fault_t	*f_p;
2202 
2203 	ASSERT(cb_p->pxl);
2204 
2205 	/* find and remove this px, and update cb list */
2206 	mutex_enter(&cb_p->cb_mutex);
2207 
2208 	pxl = cb_p->pxl;
2209 	if (pxl->pxp == px_p) {
2210 		cb_p->pxl = pxl->next;
2211 	} else {
2212 		prev = pxl;
2213 		pxl = pxl->next;
2214 		for (; pxl && (pxl->pxp != px_p); prev = pxl, pxl = pxl->next) {
2215 		};
2216 		if (!pxl) {
2217 			cmn_err(CE_WARN, "px_cb_rem_intr: can't find px_p 0x%p "
2218 			    "in registered CB list.", (void *)px_p);
2219 			mutex_exit(&cb_p->cb_mutex);
2220 			return;
2221 		}
2222 		prev->next = pxl->next;
2223 	}
2224 	pxu_p->px_cb_p = NULL;
2225 	cb_p->attachcnt--;
2226 	kmem_free(pxl, sizeof (px_cb_list_t));
2227 	mutex_exit(&cb_p->cb_mutex);
2228 
2229 	/* disable cb hw interrupt */
2230 	if (fault_p->px_fh_sysino == cb_p->sysino)
2231 		px_ib_intr_disable(px_p->px_ib_p, fault_p->px_intr_ino,
2232 		    IB_INTR_WAIT);
2233 
2234 	/* if last px, remove from interrupt redistribution list */
2235 	if (cb_p->pxl == NULL)
2236 		intr_dist_rem(px_cb_intr_redist, cb_p);
2237 
2238 	/* de-register interrupt */
2239 	VERIFY(rem_ivintr(fault_p->px_fh_sysino, PX_ERR_PIL) == 0);
2240 
2241 	/* if not last px, assign next px to manage cb */
2242 	mutex_enter(&cb_p->cb_mutex);
2243 	if (cb_p->pxl) {
2244 		if (fault_p->px_fh_sysino == cb_p->sysino) {
2245 			pxp = cb_p->pxl->pxp;
2246 			f_p = &pxp->px_cb_fault;
2247 			cb_p->sysino = f_p->px_fh_sysino;
2248 
2249 			PX_INTR_ENABLE(pxp->px_dip, cb_p->sysino, cb_p->cpuid);
2250 			(void) px_lib_intr_setstate(pxp->px_dip, cb_p->sysino,
2251 			    INTR_IDLE_STATE);
2252 		}
2253 		mutex_exit(&cb_p->cb_mutex);
2254 		return;
2255 	}
2256 
2257 	/* clean up after the last px */
2258 	mutex_exit(&cb_p->cb_mutex);
2259 
2260 	/* px_lib_dev_init allows only FIRE and OBERON */
2261 	px_err_reg_disable(
2262 	    (pxu_p->chip_type == PX_CHIP_FIRE) ? PX_ERR_JBC : PX_ERR_UBC,
2263 	    pxu_p->px_address[PX_REG_XBC]);
2264 
2265 	mutex_destroy(&cb_p->cb_mutex);
2266 	px_set_cb(fault_p->px_fh_dip, 0ull);
2267 	kmem_free(cb_p, sizeof (px_cb_t));
2268 }
2269 
2270 /*
2271  * px_cb_intr() - sun4u only,  CB interrupt dispatcher
2272  */
2273 uint_t
2274 px_cb_intr(caddr_t arg)
2275 {
2276 	px_cb_t		*cb_p = (px_cb_t *)arg;
2277 	px_t		*pxp;
2278 	px_fault_t	*f_p;
2279 	int		ret;
2280 
2281 	mutex_enter(&cb_p->cb_mutex);
2282 
2283 	if (!cb_p->pxl) {
2284 		mutex_exit(&cb_p->cb_mutex);
2285 		return (DDI_INTR_UNCLAIMED);
2286 	}
2287 
2288 	pxp = cb_p->pxl->pxp;
2289 	f_p = &pxp->px_cb_fault;
2290 
2291 	ret = f_p->px_err_func((caddr_t)f_p);
2292 
2293 	mutex_exit(&cb_p->cb_mutex);
2294 	return (ret);
2295 }
2296 
2297 #ifdef	FMA
2298 void
2299 px_fill_rc_status(px_fault_t *px_fault_p, pciex_rc_error_regs_t *rc_status)
2300 {
2301 	/* populate the rc_status by reading the registers - TBD */
2302 }
2303 #endif /* FMA */
2304 
2305 /*
2306  * Unprotected raw reads/writes of fabric device's config space.
2307  * Only used for temporary PCI-E Fabric Error Handling.
2308  */
2309 uint32_t
2310 px_fab_get(px_t *px_p, pcie_req_id_t bdf, uint16_t offset)
2311 {
2312 	px_ranges_t	*rp = px_p->px_ranges_p;
2313 	uint64_t	range_prop, base_addr;
2314 	int		bank = PCI_REG_ADDR_G(PCI_ADDR_CONFIG);
2315 	uint32_t	val;
2316 
2317 	/* Get Fire's Physical Base Address */
2318 	range_prop = px_get_range_prop(px_p, rp, bank);
2319 
2320 	/* Get config space first. */
2321 	base_addr = range_prop + PX_BDF_TO_CFGADDR(bdf, offset);
2322 
2323 	val = ldphysio(base_addr);
2324 
2325 	return (LE_32(val));
2326 }
2327 
2328 void
2329 px_fab_set(px_t *px_p, pcie_req_id_t bdf, uint16_t offset,
2330     uint32_t val) {
2331 	px_ranges_t	*rp = px_p->px_ranges_p;
2332 	uint64_t	range_prop, base_addr;
2333 	int		bank = PCI_REG_ADDR_G(PCI_ADDR_CONFIG);
2334 
2335 	/* Get Fire's Physical Base Address */
2336 	range_prop = px_get_range_prop(px_p, rp, bank);
2337 
2338 	/* Get config space first. */
2339 	base_addr = range_prop + PX_BDF_TO_CFGADDR(bdf, offset);
2340 
2341 	stphysio(base_addr, LE_32(val));
2342 }
2343 
2344 /*
2345  * cpr callback
2346  *
2347  * disable fabric error msg interrupt prior to suspending
2348  * all device drivers; re-enable fabric error msg interrupt
2349  * after all devices are resumed.
2350  */
2351 static boolean_t
2352 px_cpr_callb(void *arg, int code)
2353 {
2354 	px_t		*px_p = (px_t *)arg;
2355 	px_ib_t		*ib_p = px_p->px_ib_p;
2356 	px_pec_t	*pec_p = px_p->px_pec_p;
2357 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2358 	caddr_t		csr_base;
2359 	devino_t	ce_ino, nf_ino, f_ino;
2360 	px_ino_t	*ce_ino_p, *nf_ino_p, *f_ino_p;
2361 	uint64_t	imu_log_enable, imu_intr_enable;
2362 	uint64_t	imu_log_mask, imu_intr_mask;
2363 
2364 	ce_ino = px_msiqid_to_devino(px_p, pec_p->pec_corr_msg_msiq_id);
2365 	nf_ino = px_msiqid_to_devino(px_p, pec_p->pec_non_fatal_msg_msiq_id);
2366 	f_ino = px_msiqid_to_devino(px_p, pec_p->pec_fatal_msg_msiq_id);
2367 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2368 
2369 	imu_log_enable = CSR_XR(csr_base, IMU_ERROR_LOG_ENABLE);
2370 	imu_intr_enable = CSR_XR(csr_base, IMU_INTERRUPT_ENABLE);
2371 
2372 	imu_log_mask = BITMASK(IMU_ERROR_LOG_ENABLE_FATAL_MES_NOT_EN_LOG_EN) |
2373 	    BITMASK(IMU_ERROR_LOG_ENABLE_NONFATAL_MES_NOT_EN_LOG_EN) |
2374 	    BITMASK(IMU_ERROR_LOG_ENABLE_COR_MES_NOT_EN_LOG_EN);
2375 
2376 	imu_intr_mask =
2377 	    BITMASK(IMU_INTERRUPT_ENABLE_FATAL_MES_NOT_EN_S_INT_EN) |
2378 	    BITMASK(IMU_INTERRUPT_ENABLE_NONFATAL_MES_NOT_EN_S_INT_EN) |
2379 	    BITMASK(IMU_INTERRUPT_ENABLE_COR_MES_NOT_EN_S_INT_EN) |
2380 	    BITMASK(IMU_INTERRUPT_ENABLE_FATAL_MES_NOT_EN_P_INT_EN) |
2381 	    BITMASK(IMU_INTERRUPT_ENABLE_NONFATAL_MES_NOT_EN_P_INT_EN) |
2382 	    BITMASK(IMU_INTERRUPT_ENABLE_COR_MES_NOT_EN_P_INT_EN);
2383 
2384 	switch (code) {
2385 	case CB_CODE_CPR_CHKPT:
2386 		/* disable imu rbne on corr/nonfatal/fatal errors */
2387 		CSR_XS(csr_base, IMU_ERROR_LOG_ENABLE,
2388 		    imu_log_enable & (~imu_log_mask));
2389 
2390 		CSR_XS(csr_base, IMU_INTERRUPT_ENABLE,
2391 		    imu_intr_enable & (~imu_intr_mask));
2392 
2393 		/* disable CORR intr mapping */
2394 		px_ib_intr_disable(ib_p, ce_ino, IB_INTR_NOWAIT);
2395 
2396 		/* disable NON FATAL intr mapping */
2397 		px_ib_intr_disable(ib_p, nf_ino, IB_INTR_NOWAIT);
2398 
2399 		/* disable FATAL intr mapping */
2400 		px_ib_intr_disable(ib_p, f_ino, IB_INTR_NOWAIT);
2401 
2402 		break;
2403 
2404 	case CB_CODE_CPR_RESUME:
2405 		pxu_p->cpr_flag = PX_NOT_CPR;
2406 		mutex_enter(&ib_p->ib_ino_lst_mutex);
2407 
2408 		ce_ino_p = px_ib_locate_ino(ib_p, ce_ino);
2409 		nf_ino_p = px_ib_locate_ino(ib_p, nf_ino);
2410 		f_ino_p = px_ib_locate_ino(ib_p, f_ino);
2411 
2412 		/* enable CORR intr mapping */
2413 		if (ce_ino_p)
2414 			px_ib_intr_enable(px_p, ce_ino_p->ino_cpuid, ce_ino);
2415 		else
2416 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2417 			    "reenable PCIe Correctable msg intr.\n");
2418 
2419 		/* enable NON FATAL intr mapping */
2420 		if (nf_ino_p)
2421 			px_ib_intr_enable(px_p, nf_ino_p->ino_cpuid, nf_ino);
2422 		else
2423 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2424 			    "reenable PCIe Non Fatal msg intr.\n");
2425 
2426 		/* enable FATAL intr mapping */
2427 		if (f_ino_p)
2428 			px_ib_intr_enable(px_p, f_ino_p->ino_cpuid, f_ino);
2429 		else
2430 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2431 			    "reenable PCIe Fatal msg intr.\n");
2432 
2433 		mutex_exit(&ib_p->ib_ino_lst_mutex);
2434 
2435 		/* enable corr/nonfatal/fatal not enable error */
2436 		CSR_XS(csr_base, IMU_ERROR_LOG_ENABLE, (imu_log_enable |
2437 		    (imu_log_mask & px_imu_log_mask)));
2438 		CSR_XS(csr_base, IMU_INTERRUPT_ENABLE, (imu_intr_enable |
2439 		    (imu_intr_mask & px_imu_intr_mask)));
2440 
2441 		break;
2442 	}
2443 
2444 	return (B_TRUE);
2445 }
2446 
2447 uint64_t
2448 px_get_rng_parent_hi_mask(px_t *px_p)
2449 {
2450 	pxu_t *pxu_p = (pxu_t *)px_p->px_plat_p;
2451 	uint64_t mask;
2452 
2453 	switch (PX_CHIP_TYPE(pxu_p)) {
2454 	case PX_CHIP_OBERON:
2455 		mask = OBERON_RANGE_PROP_MASK;
2456 		break;
2457 	case PX_CHIP_FIRE:
2458 		mask = PX_RANGE_PROP_MASK;
2459 		break;
2460 	default:
2461 		mask = PX_RANGE_PROP_MASK;
2462 	}
2463 
2464 	return (mask);
2465 }
2466 
2467 /*
2468  * fetch chip's range propery's value
2469  */
2470 uint64_t
2471 px_get_range_prop(px_t *px_p, px_ranges_t *rp, int bank)
2472 {
2473 	uint64_t mask, range_prop;
2474 
2475 	mask = px_get_rng_parent_hi_mask(px_p);
2476 	range_prop = (((uint64_t)(rp[bank].parent_high & mask)) << 32) |
2477 	    rp[bank].parent_low;
2478 
2479 	return (range_prop);
2480 }
2481 
2482 /*
2483  * add cpr callback
2484  */
2485 void
2486 px_cpr_add_callb(px_t *px_p)
2487 {
2488 	px_p->px_cprcb_id = callb_add(px_cpr_callb, (void *)px_p,
2489 	    CB_CL_CPR_POST_USER, "px_cpr");
2490 }
2491 
2492 /*
2493  * remove cpr callback
2494  */
2495 void
2496 px_cpr_rem_callb(px_t *px_p)
2497 {
2498 	(void) callb_delete(px_p->px_cprcb_id);
2499 }
2500 
2501 /*ARGSUSED*/
2502 static uint_t
2503 px_hp_intr(caddr_t arg1, caddr_t arg2)
2504 {
2505 	px_t	*px_p = (px_t *)arg1;
2506 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2507 	int	rval;
2508 
2509 	rval = pciehpc_intr(px_p->px_dip);
2510 
2511 #ifdef  DEBUG
2512 	if (rval == DDI_INTR_UNCLAIMED)
2513 		cmn_err(CE_WARN, "%s%d: UNCLAIMED intr\n",
2514 		    ddi_driver_name(px_p->px_dip),
2515 		    ddi_get_instance(px_p->px_dip));
2516 #endif
2517 
2518 	/* Set the interrupt state to idle */
2519 	if (px_lib_intr_setstate(px_p->px_dip,
2520 	    pxu_p->hp_sysino, INTR_IDLE_STATE) != DDI_SUCCESS)
2521 		return (DDI_INTR_UNCLAIMED);
2522 
2523 	return (rval);
2524 }
2525 
2526 int
2527 px_lib_hotplug_init(dev_info_t *dip, void *arg)
2528 {
2529 	px_t	*px_p = DIP_TO_STATE(dip);
2530 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2531 	uint64_t ret;
2532 
2533 	if ((ret = hvio_hotplug_init(dip, arg)) == DDI_SUCCESS) {
2534 		if (px_lib_intr_devino_to_sysino(px_p->px_dip,
2535 		    px_p->px_inos[PX_INTR_HOTPLUG], &pxu_p->hp_sysino) !=
2536 		    DDI_SUCCESS) {
2537 #ifdef	DEBUG
2538 			cmn_err(CE_WARN, "%s%d: devino_to_sysino fails\n",
2539 			    ddi_driver_name(px_p->px_dip),
2540 			    ddi_get_instance(px_p->px_dip));
2541 #endif
2542 			return (DDI_FAILURE);
2543 		}
2544 
2545 		VERIFY(add_ivintr(pxu_p->hp_sysino, PX_PCIEHP_PIL,
2546 		    (intrfunc)px_hp_intr, (caddr_t)px_p, NULL, NULL) == 0);
2547 
2548 		px_ib_intr_enable(px_p, intr_dist_cpuid(),
2549 		    px_p->px_inos[PX_INTR_HOTPLUG]);
2550 	}
2551 
2552 	return (ret);
2553 }
2554 
2555 void
2556 px_lib_hotplug_uninit(dev_info_t *dip)
2557 {
2558 	if (hvio_hotplug_uninit(dip) == DDI_SUCCESS) {
2559 		px_t	*px_p = DIP_TO_STATE(dip);
2560 		pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2561 
2562 		px_ib_intr_disable(px_p->px_ib_p,
2563 		    px_p->px_inos[PX_INTR_HOTPLUG], IB_INTR_WAIT);
2564 
2565 		VERIFY(rem_ivintr(pxu_p->hp_sysino, PX_PCIEHP_PIL) == 0);
2566 	}
2567 }
2568 
2569 /*
2570  * px_hp_intr_redist() - sun4u only, HP interrupt redistribution
2571  */
2572 void
2573 px_hp_intr_redist(px_t *px_p)
2574 {
2575 	if (px_p && (px_p->px_dev_caps & PX_HOTPLUG_CAPABLE)) {
2576 		px_ib_intr_dist_en(px_p->px_dip, intr_dist_cpuid(),
2577 		    px_p->px_inos[PX_INTR_HOTPLUG], B_FALSE);
2578 	}
2579 }
2580 
2581 boolean_t
2582 px_lib_is_in_drain_state(px_t *px_p)
2583 {
2584 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2585 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2586 	uint64_t drain_status;
2587 
2588 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON) {
2589 		drain_status = CSR_BR(csr_base, DRAIN_CONTROL_STATUS, DRAIN);
2590 	} else {
2591 		drain_status = CSR_BR(csr_base, TLU_STATUS, DRAIN);
2592 	}
2593 
2594 	return (drain_status);
2595 }
2596 
2597 pcie_req_id_t
2598 px_lib_get_bdf(px_t *px_p)
2599 {
2600 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2601 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2602 	pcie_req_id_t bdf;
2603 
2604 	bdf = CSR_BR(csr_base, DMC_PCI_EXPRESS_CONFIGURATION, REQ_ID);
2605 
2606 	return (bdf);
2607 }
2608