xref: /illumos-gate/usr/src/uts/sun4u/io/px/px_lib4u.c (revision e3b270024f3e952233b61d3dcb4129ba75bf2720)
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/types.h>
26 #include <sys/kmem.h>
27 #include <sys/conf.h>
28 #include <sys/ddi.h>
29 #include <sys/sunddi.h>
30 #include <sys/sunndi.h>
31 #include <sys/fm/protocol.h>
32 #include <sys/fm/util.h>
33 #include <sys/modctl.h>
34 #include <sys/disp.h>
35 #include <sys/stat.h>
36 #include <sys/ddi_impldefs.h>
37 #include <sys/vmem.h>
38 #include <sys/iommutsb.h>
39 #include <sys/cpuvar.h>
40 #include <sys/ivintr.h>
41 #include <sys/byteorder.h>
42 #include <sys/spl.h>
43 #include <px_obj.h>
44 #include <sys/pcie_pwr.h>
45 #include "px_tools_var.h"
46 #include <px_regs.h>
47 #include <px_csr.h>
48 #include <sys/machsystm.h>
49 #include "px_lib4u.h"
50 #include "px_err.h"
51 #include "oberon_regs.h"
52 #include <sys/hotplug/pci/pcie_hp.h>
53 
54 #pragma weak jbus_stst_order
55 
56 extern void jbus_stst_order();
57 
58 ulong_t px_mmu_dvma_end = 0xfffffffful;
59 uint_t px_ranges_phi_mask = 0xfffffffful;
60 uint64_t *px_oberon_ubc_scratch_regs;
61 uint64_t px_paddr_mask;
62 
63 static int px_goto_l23ready(px_t *px_p);
64 static int px_goto_l0(px_t *px_p);
65 static int px_pre_pwron_check(px_t *px_p);
66 static uint32_t px_identity_init(px_t *px_p);
67 static boolean_t px_cpr_callb(void *arg, int code);
68 static uint_t px_cb_intr(caddr_t arg);
69 
70 /*
71  * ACKNAK Latency Threshold Table.
72  * See Fire PRM 2.0 section 1.2.12.2, table 1-17.
73  */
74 int px_acknak_timer_table[LINK_MAX_PKT_ARR_SIZE][LINK_WIDTH_ARR_SIZE] = {
75 	{0xED,   0x49,  0x43,  0x30},
76 	{0x1A0,  0x76,  0x6B,  0x48},
77 	{0x22F,  0x9A,  0x56,  0x56},
78 	{0x42F,  0x11A, 0x96,  0x96},
79 	{0x82F,  0x21A, 0x116, 0x116},
80 	{0x102F, 0x41A, 0x216, 0x216}
81 };
82 
83 /*
84  * TxLink Replay Timer Latency Table
85  * See Fire PRM 2.0 sections 1.2.12.3, table 1-18.
86  */
87 int px_replay_timer_table[LINK_MAX_PKT_ARR_SIZE][LINK_WIDTH_ARR_SIZE] = {
88 	{0x379,  0x112, 0xFC,  0xB4},
89 	{0x618,  0x1BA, 0x192, 0x10E},
90 	{0x831,  0x242, 0x143, 0x143},
91 	{0xFB1,  0x422, 0x233, 0x233},
92 	{0x1EB0, 0x7E1, 0x412, 0x412},
93 	{0x3CB0, 0xF61, 0x7D2, 0x7D2}
94 };
95 /*
96  * px_lib_map_registers
97  *
98  * This function is called from the attach routine to map the registers
99  * accessed by this driver.
100  *
101  * used by: px_attach()
102  *
103  * return value: DDI_FAILURE on failure
104  */
105 int
106 px_lib_map_regs(pxu_t *pxu_p, dev_info_t *dip)
107 {
108 	ddi_device_acc_attr_t	attr;
109 	px_reg_bank_t		reg_bank = PX_REG_CSR;
110 
111 	DBG(DBG_ATTACH, dip, "px_lib_map_regs: pxu_p:0x%p, dip 0x%p\n",
112 	    pxu_p, dip);
113 
114 	attr.devacc_attr_version = DDI_DEVICE_ATTR_V0;
115 	attr.devacc_attr_dataorder = DDI_STRICTORDER_ACC;
116 	attr.devacc_attr_endian_flags = DDI_NEVERSWAP_ACC;
117 
118 	/*
119 	 * PCI CSR Base
120 	 */
121 	if (ddi_regs_map_setup(dip, reg_bank, &pxu_p->px_address[reg_bank],
122 	    0, 0, &attr, &pxu_p->px_ac[reg_bank]) != DDI_SUCCESS) {
123 		goto fail;
124 	}
125 
126 	reg_bank++;
127 
128 	/*
129 	 * XBUS CSR Base
130 	 */
131 	if (ddi_regs_map_setup(dip, reg_bank, &pxu_p->px_address[reg_bank],
132 	    0, 0, &attr, &pxu_p->px_ac[reg_bank]) != DDI_SUCCESS) {
133 		goto fail;
134 	}
135 
136 	pxu_p->px_address[reg_bank] -= FIRE_CONTROL_STATUS;
137 
138 done:
139 	for (; reg_bank >= PX_REG_CSR; reg_bank--) {
140 		DBG(DBG_ATTACH, dip, "reg_bank 0x%x address 0x%p\n",
141 		    reg_bank, pxu_p->px_address[reg_bank]);
142 	}
143 
144 	return (DDI_SUCCESS);
145 
146 fail:
147 	cmn_err(CE_WARN, "%s%d: unable to map reg entry %d\n",
148 	    ddi_driver_name(dip), ddi_get_instance(dip), reg_bank);
149 
150 	for (reg_bank--; reg_bank >= PX_REG_CSR; reg_bank--) {
151 		pxu_p->px_address[reg_bank] = NULL;
152 		ddi_regs_map_free(&pxu_p->px_ac[reg_bank]);
153 	}
154 
155 	return (DDI_FAILURE);
156 }
157 
158 /*
159  * px_lib_unmap_regs:
160  *
161  * This routine unmaps the registers mapped by map_px_registers.
162  *
163  * used by: px_detach(), and error conditions in px_attach()
164  *
165  * return value: none
166  */
167 void
168 px_lib_unmap_regs(pxu_t *pxu_p)
169 {
170 	int i;
171 
172 	for (i = 0; i < PX_REG_MAX; i++) {
173 		if (pxu_p->px_ac[i])
174 			ddi_regs_map_free(&pxu_p->px_ac[i]);
175 	}
176 }
177 
178 int
179 px_lib_dev_init(dev_info_t *dip, devhandle_t *dev_hdl)
180 {
181 
182 	caddr_t			xbc_csr_base, csr_base;
183 	px_dvma_range_prop_t	px_dvma_range;
184 	pxu_t			*pxu_p;
185 	uint8_t			chip_mask;
186 	px_t			*px_p = DIP_TO_STATE(dip);
187 	px_chip_type_t		chip_type = px_identity_init(px_p);
188 
189 	DBG(DBG_ATTACH, dip, "px_lib_dev_init: dip 0x%p", dip);
190 
191 	if (chip_type == PX_CHIP_UNIDENTIFIED) {
192 		cmn_err(CE_WARN, "%s%d: Unrecognized Hardware Version\n",
193 		    NAMEINST(dip));
194 		return (DDI_FAILURE);
195 	}
196 
197 	chip_mask = BITMASK(chip_type);
198 	px_paddr_mask = (chip_type == PX_CHIP_FIRE) ? MMU_FIRE_PADDR_MASK :
199 	    MMU_OBERON_PADDR_MASK;
200 
201 	/*
202 	 * Allocate platform specific structure and link it to
203 	 * the px state structure.
204 	 */
205 	pxu_p = kmem_zalloc(sizeof (pxu_t), KM_SLEEP);
206 	pxu_p->chip_type = chip_type;
207 	pxu_p->portid  = ddi_getprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
208 	    "portid", -1);
209 
210 	/* Map in the registers */
211 	if (px_lib_map_regs(pxu_p, dip) == DDI_FAILURE) {
212 		kmem_free(pxu_p, sizeof (pxu_t));
213 
214 		return (DDI_FAILURE);
215 	}
216 
217 	xbc_csr_base = (caddr_t)pxu_p->px_address[PX_REG_XBC];
218 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
219 
220 	pxu_p->tsb_cookie = iommu_tsb_alloc(pxu_p->portid);
221 	pxu_p->tsb_size = iommu_tsb_cookie_to_size(pxu_p->tsb_cookie);
222 	pxu_p->tsb_vaddr = iommu_tsb_cookie_to_va(pxu_p->tsb_cookie);
223 
224 	pxu_p->tsb_paddr = va_to_pa(pxu_p->tsb_vaddr);
225 
226 	/*
227 	 * Create "virtual-dma" property to support child devices
228 	 * needing to know DVMA range.
229 	 */
230 	px_dvma_range.dvma_base = (uint32_t)px_mmu_dvma_end + 1
231 	    - ((pxu_p->tsb_size >> 3) << MMU_PAGE_SHIFT);
232 	px_dvma_range.dvma_len = (uint32_t)
233 	    px_mmu_dvma_end - px_dvma_range.dvma_base + 1;
234 
235 	(void) ddi_prop_update_int_array(DDI_DEV_T_NONE, dip,
236 	    "virtual-dma", (int *)&px_dvma_range,
237 	    sizeof (px_dvma_range_prop_t) / sizeof (int));
238 	/*
239 	 * Initilize all fire hardware specific blocks.
240 	 */
241 	hvio_cb_init(xbc_csr_base, pxu_p);
242 	hvio_ib_init(csr_base, pxu_p);
243 	hvio_pec_init(csr_base, pxu_p);
244 	hvio_mmu_init(csr_base, pxu_p);
245 
246 	px_p->px_plat_p = (void *)pxu_p;
247 
248 	/*
249 	 * Initialize all the interrupt handlers
250 	 */
251 	switch (PX_CHIP_TYPE(pxu_p)) {
252 	case PX_CHIP_OBERON:
253 		/*
254 		 * Oberon hotplug uses SPARE3 field in ILU Error Log Enable
255 		 * register to indicate the status of leaf reset,
256 		 * we need to preserve the value of this bit, and keep it in
257 		 * px_ilu_log_mask to reflect the state of the bit
258 		 */
259 		if (CSR_BR(csr_base, ILU_ERROR_LOG_ENABLE, SPARE3))
260 			px_ilu_log_mask |= (1ull <<
261 			    ILU_ERROR_LOG_ENABLE_SPARE3);
262 		else
263 			px_ilu_log_mask &= ~(1ull <<
264 			    ILU_ERROR_LOG_ENABLE_SPARE3);
265 
266 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_ENABLE);
267 		break;
268 
269 	case PX_CHIP_FIRE:
270 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_ENABLE);
271 		break;
272 
273 	default:
274 		cmn_err(CE_WARN, "%s%d: PX primary bus Unknown\n",
275 		    ddi_driver_name(dip), ddi_get_instance(dip));
276 		return (DDI_FAILURE);
277 	}
278 
279 	/* Initilize device handle */
280 	*dev_hdl = (devhandle_t)csr_base;
281 
282 	DBG(DBG_ATTACH, dip, "px_lib_dev_init: dev_hdl 0x%llx\n", *dev_hdl);
283 
284 	/* Sun4u always support fixed interrupt */
285 	px_p->px_supp_intr_types |= DDI_INTR_TYPE_FIXED;
286 
287 	return (DDI_SUCCESS);
288 }
289 
290 int
291 px_lib_dev_fini(dev_info_t *dip)
292 {
293 	caddr_t			csr_base;
294 	uint8_t			chip_mask;
295 	px_t			*px_p = DIP_TO_STATE(dip);
296 	pxu_t			*pxu_p = (pxu_t *)px_p->px_plat_p;
297 
298 	DBG(DBG_DETACH, dip, "px_lib_dev_fini: dip 0x%p\n", dip);
299 
300 	/*
301 	 * Deinitialize all the interrupt handlers
302 	 */
303 	switch (PX_CHIP_TYPE(pxu_p)) {
304 	case PX_CHIP_OBERON:
305 	case PX_CHIP_FIRE:
306 		chip_mask = BITMASK(PX_CHIP_TYPE(pxu_p));
307 		csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
308 		px_err_reg_setup_pcie(chip_mask, csr_base, PX_ERR_DISABLE);
309 		break;
310 
311 	default:
312 		cmn_err(CE_WARN, "%s%d: PX primary bus Unknown\n",
313 		    ddi_driver_name(dip), ddi_get_instance(dip));
314 		return (DDI_FAILURE);
315 	}
316 
317 	iommu_tsb_free(pxu_p->tsb_cookie);
318 
319 	px_lib_unmap_regs((pxu_t *)px_p->px_plat_p);
320 	kmem_free(px_p->px_plat_p, sizeof (pxu_t));
321 	px_p->px_plat_p = NULL;
322 	(void) ddi_prop_remove(DDI_DEV_T_NONE, dip, "virtual-dma");
323 
324 	return (DDI_SUCCESS);
325 }
326 
327 /*ARGSUSED*/
328 int
329 px_lib_intr_devino_to_sysino(dev_info_t *dip, devino_t devino,
330     sysino_t *sysino)
331 {
332 	px_t	*px_p = DIP_TO_STATE(dip);
333 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
334 	uint64_t	ret;
335 
336 	DBG(DBG_LIB_INT, dip, "px_lib_intr_devino_to_sysino: dip 0x%p "
337 	    "devino 0x%x\n", dip, devino);
338 
339 	if ((ret = hvio_intr_devino_to_sysino(DIP_TO_HANDLE(dip),
340 	    pxu_p, devino, sysino)) != H_EOK) {
341 		DBG(DBG_LIB_INT, dip,
342 		    "hvio_intr_devino_to_sysino failed, ret 0x%lx\n", ret);
343 		return (DDI_FAILURE);
344 	}
345 
346 	DBG(DBG_LIB_INT, dip, "px_lib_intr_devino_to_sysino: sysino 0x%llx\n",
347 	    *sysino);
348 
349 	return (DDI_SUCCESS);
350 }
351 
352 /*ARGSUSED*/
353 int
354 px_lib_intr_getvalid(dev_info_t *dip, sysino_t sysino,
355     intr_valid_state_t *intr_valid_state)
356 {
357 	uint64_t	ret;
358 
359 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getvalid: dip 0x%p sysino 0x%llx\n",
360 	    dip, sysino);
361 
362 	if ((ret = hvio_intr_getvalid(DIP_TO_HANDLE(dip),
363 	    sysino, intr_valid_state)) != H_EOK) {
364 		DBG(DBG_LIB_INT, dip, "hvio_intr_getvalid failed, ret 0x%lx\n",
365 		    ret);
366 		return (DDI_FAILURE);
367 	}
368 
369 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getvalid: intr_valid_state 0x%x\n",
370 	    *intr_valid_state);
371 
372 	return (DDI_SUCCESS);
373 }
374 
375 /*ARGSUSED*/
376 int
377 px_lib_intr_setvalid(dev_info_t *dip, sysino_t sysino,
378     intr_valid_state_t intr_valid_state)
379 {
380 	uint64_t	ret;
381 
382 	DBG(DBG_LIB_INT, dip, "px_lib_intr_setvalid: dip 0x%p sysino 0x%llx "
383 	    "intr_valid_state 0x%x\n", dip, sysino, intr_valid_state);
384 
385 	if ((ret = hvio_intr_setvalid(DIP_TO_HANDLE(dip),
386 	    sysino, intr_valid_state)) != H_EOK) {
387 		DBG(DBG_LIB_INT, dip, "hvio_intr_setvalid failed, ret 0x%lx\n",
388 		    ret);
389 		return (DDI_FAILURE);
390 	}
391 
392 	return (DDI_SUCCESS);
393 }
394 
395 /*ARGSUSED*/
396 int
397 px_lib_intr_getstate(dev_info_t *dip, sysino_t sysino,
398     intr_state_t *intr_state)
399 {
400 	uint64_t	ret;
401 
402 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getstate: dip 0x%p sysino 0x%llx\n",
403 	    dip, sysino);
404 
405 	if ((ret = hvio_intr_getstate(DIP_TO_HANDLE(dip),
406 	    sysino, intr_state)) != H_EOK) {
407 		DBG(DBG_LIB_INT, dip, "hvio_intr_getstate failed, ret 0x%lx\n",
408 		    ret);
409 		return (DDI_FAILURE);
410 	}
411 
412 	DBG(DBG_LIB_INT, dip, "px_lib_intr_getstate: intr_state 0x%x\n",
413 	    *intr_state);
414 
415 	return (DDI_SUCCESS);
416 }
417 
418 /*ARGSUSED*/
419 int
420 px_lib_intr_setstate(dev_info_t *dip, sysino_t sysino,
421     intr_state_t intr_state)
422 {
423 	uint64_t	ret;
424 
425 	DBG(DBG_LIB_INT, dip, "px_lib_intr_setstate: dip 0x%p sysino 0x%llx "
426 	    "intr_state 0x%x\n", dip, sysino, intr_state);
427 
428 	if ((ret = hvio_intr_setstate(DIP_TO_HANDLE(dip),
429 	    sysino, intr_state)) != H_EOK) {
430 		DBG(DBG_LIB_INT, dip, "hvio_intr_setstate failed, ret 0x%lx\n",
431 		    ret);
432 		return (DDI_FAILURE);
433 	}
434 
435 	return (DDI_SUCCESS);
436 }
437 
438 /*ARGSUSED*/
439 int
440 px_lib_intr_gettarget(dev_info_t *dip, sysino_t sysino, cpuid_t *cpuid)
441 {
442 	px_t		*px_p = DIP_TO_STATE(dip);
443 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
444 	uint64_t	ret;
445 
446 	DBG(DBG_LIB_INT, dip, "px_lib_intr_gettarget: dip 0x%p sysino 0x%llx\n",
447 	    dip, sysino);
448 
449 	if ((ret = hvio_intr_gettarget(DIP_TO_HANDLE(dip), pxu_p,
450 	    sysino, cpuid)) != H_EOK) {
451 		DBG(DBG_LIB_INT, dip, "hvio_intr_gettarget failed, ret 0x%lx\n",
452 		    ret);
453 		return (DDI_FAILURE);
454 	}
455 
456 	DBG(DBG_LIB_INT, dip, "px_lib_intr_gettarget: cpuid 0x%x\n", cpuid);
457 
458 	return (DDI_SUCCESS);
459 }
460 
461 /*ARGSUSED*/
462 int
463 px_lib_intr_settarget(dev_info_t *dip, sysino_t sysino, cpuid_t cpuid)
464 {
465 	px_t		*px_p = DIP_TO_STATE(dip);
466 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
467 	uint64_t	ret;
468 
469 	DBG(DBG_LIB_INT, dip, "px_lib_intr_settarget: dip 0x%p sysino 0x%llx "
470 	    "cpuid 0x%x\n", dip, sysino, cpuid);
471 
472 	if ((ret = hvio_intr_settarget(DIP_TO_HANDLE(dip), pxu_p,
473 	    sysino, cpuid)) != H_EOK) {
474 		DBG(DBG_LIB_INT, dip, "hvio_intr_settarget failed, ret 0x%lx\n",
475 		    ret);
476 		return (DDI_FAILURE);
477 	}
478 
479 	return (DDI_SUCCESS);
480 }
481 
482 /*ARGSUSED*/
483 int
484 px_lib_intr_reset(dev_info_t *dip)
485 {
486 	devino_t	ino;
487 	sysino_t	sysino;
488 
489 	DBG(DBG_LIB_INT, dip, "px_lib_intr_reset: dip 0x%p\n", dip);
490 
491 	/* Reset all Interrupts */
492 	for (ino = 0; ino < INTERRUPT_MAPPING_ENTRIES; ino++) {
493 		if (px_lib_intr_devino_to_sysino(dip, ino,
494 		    &sysino) != DDI_SUCCESS)
495 			return (BF_FATAL);
496 
497 		if (px_lib_intr_setstate(dip, sysino,
498 		    INTR_IDLE_STATE) != DDI_SUCCESS)
499 			return (BF_FATAL);
500 	}
501 
502 	return (BF_NONE);
503 }
504 
505 /*ARGSUSED*/
506 int
507 px_lib_iommu_map(dev_info_t *dip, tsbid_t tsbid, pages_t pages,
508     io_attributes_t attr, void *addr, size_t pfn_index, int flags)
509 {
510 	px_t		*px_p = DIP_TO_STATE(dip);
511 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
512 	uint64_t	ret;
513 
514 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_map: dip 0x%p tsbid 0x%llx "
515 	    "pages 0x%x attr 0x%llx addr 0x%p pfn_index 0x%llx flags 0x%x\n",
516 	    dip, tsbid, pages, attr, addr, pfn_index, flags);
517 
518 	if ((ret = hvio_iommu_map(px_p->px_dev_hdl, pxu_p, tsbid, pages,
519 	    attr, addr, pfn_index, flags)) != H_EOK) {
520 		DBG(DBG_LIB_DMA, dip,
521 		    "px_lib_iommu_map failed, ret 0x%lx\n", ret);
522 		return (DDI_FAILURE);
523 	}
524 
525 	return (DDI_SUCCESS);
526 }
527 
528 /*ARGSUSED*/
529 int
530 px_lib_iommu_demap(dev_info_t *dip, tsbid_t tsbid, pages_t pages)
531 {
532 	px_t		*px_p = DIP_TO_STATE(dip);
533 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
534 	uint64_t	ret;
535 
536 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_demap: dip 0x%p tsbid 0x%llx "
537 	    "pages 0x%x\n", dip, tsbid, pages);
538 
539 	if ((ret = hvio_iommu_demap(px_p->px_dev_hdl, pxu_p, tsbid, pages))
540 	    != H_EOK) {
541 		DBG(DBG_LIB_DMA, dip,
542 		    "px_lib_iommu_demap failed, ret 0x%lx\n", ret);
543 
544 		return (DDI_FAILURE);
545 	}
546 
547 	return (DDI_SUCCESS);
548 }
549 
550 /*ARGSUSED*/
551 int
552 px_lib_iommu_getmap(dev_info_t *dip, tsbid_t tsbid, io_attributes_t *attr_p,
553     r_addr_t *r_addr_p)
554 {
555 	px_t	*px_p = DIP_TO_STATE(dip);
556 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
557 	uint64_t	ret;
558 
559 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getmap: dip 0x%p tsbid 0x%llx\n",
560 	    dip, tsbid);
561 
562 	if ((ret = hvio_iommu_getmap(DIP_TO_HANDLE(dip), pxu_p, tsbid,
563 	    attr_p, r_addr_p)) != H_EOK) {
564 		DBG(DBG_LIB_DMA, dip,
565 		    "hvio_iommu_getmap failed, ret 0x%lx\n", ret);
566 
567 		return ((ret == H_ENOMAP) ? DDI_DMA_NOMAPPING:DDI_FAILURE);
568 	}
569 
570 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getmap: attr 0x%llx "
571 	    "r_addr 0x%llx\n", *attr_p, *r_addr_p);
572 
573 	return (DDI_SUCCESS);
574 }
575 
576 int
577 px_lib_iommu_detach(px_t *px_p)
578 {
579 	/*
580 	 * Deallocate DVMA addr space that was reserved for OBP TTE's
581 	 * during Attach.
582 	 */
583 	hvio_obptsb_detach(px_p);
584 
585 	return (DDI_SUCCESS);
586 }
587 
588 /*
589  * Checks dma attributes against system bypass ranges
590  * The bypass range is determined by the hardware. Return them so the
591  * common code can do generic checking against them.
592  */
593 /*ARGSUSED*/
594 int
595 px_lib_dma_bypass_rngchk(dev_info_t *dip, ddi_dma_attr_t *attr_p,
596     uint64_t *lo_p, uint64_t *hi_p)
597 {
598 	px_t	*px_p = DIP_TO_STATE(dip);
599 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
600 
601 	*lo_p = hvio_get_bypass_base(pxu_p);
602 	*hi_p = hvio_get_bypass_end(pxu_p);
603 
604 	return (DDI_SUCCESS);
605 }
606 
607 
608 /*ARGSUSED*/
609 int
610 px_lib_iommu_getbypass(dev_info_t *dip, r_addr_t ra, io_attributes_t attr,
611     io_addr_t *io_addr_p)
612 {
613 	uint64_t	ret;
614 	px_t	*px_p = DIP_TO_STATE(dip);
615 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
616 
617 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getbypass: dip 0x%p ra 0x%llx "
618 	    "attr 0x%llx\n", dip, ra, attr);
619 
620 	if ((ret = hvio_iommu_getbypass(DIP_TO_HANDLE(dip), pxu_p, ra,
621 	    attr, io_addr_p)) != H_EOK) {
622 		DBG(DBG_LIB_DMA, dip,
623 		    "hvio_iommu_getbypass failed, ret 0x%lx\n", ret);
624 		return (DDI_FAILURE);
625 	}
626 
627 	DBG(DBG_LIB_DMA, dip, "px_lib_iommu_getbypass: io_addr 0x%llx\n",
628 	    *io_addr_p);
629 
630 	return (DDI_SUCCESS);
631 }
632 
633 /*
634  * Returns any needed IO address bit(s) for relaxed ordering in IOMMU
635  * bypass mode.
636  */
637 uint64_t
638 px_lib_ro_bypass(dev_info_t *dip, io_attributes_t attr, uint64_t ioaddr)
639 {
640 	px_t	*px_p = DIP_TO_STATE(dip);
641 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
642 
643 	if ((PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON) && (attr & PCI_MAP_ATTR_RO))
644 		return (MMU_OBERON_BYPASS_RO | ioaddr);
645 	else
646 		return (ioaddr);
647 }
648 
649 /*
650  * bus dma sync entry point.
651  */
652 /*ARGSUSED*/
653 int
654 px_lib_dma_sync(dev_info_t *dip, dev_info_t *rdip, ddi_dma_handle_t handle,
655     off_t off, size_t len, uint_t cache_flags)
656 {
657 	ddi_dma_impl_t *mp = (ddi_dma_impl_t *)handle;
658 	px_t	*px_p = DIP_TO_STATE(dip);
659 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
660 
661 	DBG(DBG_LIB_DMA, dip, "px_lib_dma_sync: dip 0x%p rdip 0x%p "
662 	    "handle 0x%llx off 0x%x len 0x%x flags 0x%x\n",
663 	    dip, rdip, handle, off, len, cache_flags);
664 
665 	/*
666 	 * No flush needed for Oberon
667 	 */
668 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON)
669 		return (DDI_SUCCESS);
670 
671 	/*
672 	 * jbus_stst_order is found only in certain cpu modules.
673 	 * Just return success if not present.
674 	 */
675 	if (&jbus_stst_order == NULL)
676 		return (DDI_SUCCESS);
677 
678 	if (!(mp->dmai_flags & PX_DMAI_FLAGS_INUSE)) {
679 		cmn_err(CE_WARN, "%s%d: Unbound dma handle %p.",
680 		    ddi_driver_name(rdip), ddi_get_instance(rdip), (void *)mp);
681 
682 		return (DDI_FAILURE);
683 	}
684 
685 	if (mp->dmai_flags & PX_DMAI_FLAGS_NOSYNC)
686 		return (DDI_SUCCESS);
687 
688 	/*
689 	 * No flush needed when sending data from memory to device.
690 	 * Nothing to do to "sync" memory to what device would already see.
691 	 */
692 	if (!(mp->dmai_rflags & DDI_DMA_READ) ||
693 	    ((cache_flags & PX_DMA_SYNC_DDI_FLAGS) == DDI_DMA_SYNC_FORDEV))
694 		return (DDI_SUCCESS);
695 
696 	/*
697 	 * Perform necessary cpu workaround to ensure jbus ordering.
698 	 * CPU's internal "invalidate FIFOs" are flushed.
699 	 */
700 
701 	kpreempt_disable();
702 	jbus_stst_order();
703 	kpreempt_enable();
704 	return (DDI_SUCCESS);
705 }
706 
707 /*
708  * MSIQ Functions:
709  */
710 /*ARGSUSED*/
711 int
712 px_lib_msiq_init(dev_info_t *dip)
713 {
714 	px_t		*px_p = DIP_TO_STATE(dip);
715 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
716 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
717 	px_dvma_addr_t	pg_index;
718 	size_t		q_sz = msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
719 	size_t		size;
720 	int		i, ret;
721 
722 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_init: dip 0x%p\n", dip);
723 
724 	/* must aligned on q_sz (happens to be !!! page) boundary */
725 	ASSERT(q_sz == 8 * 1024);
726 
727 	/*
728 	 * Map the EQ memory into the Fire MMU (has to be 512KB aligned)
729 	 * and then initialize the base address register.
730 	 *
731 	 * Allocate entries from Fire IOMMU so that the resulting address
732 	 * is properly aligned.  Calculate the index of the first allocated
733 	 * entry.  Note: The size of the mapping is assumed to be a multiple
734 	 * of the page size.
735 	 */
736 	size = msiq_state_p->msiq_cnt * q_sz;
737 
738 	msiq_state_p->msiq_buf_p = kmem_zalloc(size, KM_SLEEP);
739 
740 	for (i = 0; i < msiq_state_p->msiq_cnt; i++)
741 		msiq_state_p->msiq_p[i].msiq_base_p = (msiqhead_t *)
742 		    ((caddr_t)msiq_state_p->msiq_buf_p + (i * q_sz));
743 
744 	pxu_p->msiq_mapped_p = vmem_xalloc(px_p->px_mmu_p->mmu_dvma_map,
745 	    size, (512 * 1024), 0, 0, NULL, NULL, VM_NOSLEEP | VM_BESTFIT);
746 
747 	if (pxu_p->msiq_mapped_p == NULL)
748 		return (DDI_FAILURE);
749 
750 	pg_index = MMU_PAGE_INDEX(px_p->px_mmu_p,
751 	    MMU_BTOP((ulong_t)pxu_p->msiq_mapped_p));
752 
753 	if ((ret = px_lib_iommu_map(px_p->px_dip, PCI_TSBID(0, pg_index),
754 	    MMU_BTOP(size), PCI_MAP_ATTR_WRITE, msiq_state_p->msiq_buf_p,
755 	    0, MMU_MAP_BUF)) != DDI_SUCCESS) {
756 		DBG(DBG_LIB_MSIQ, dip,
757 		    "px_lib_msiq_init: px_lib_iommu_map failed, "
758 		    "ret 0x%lx\n", ret);
759 
760 		(void) px_lib_msiq_fini(dip);
761 		return (DDI_FAILURE);
762 	}
763 
764 	if ((ret = hvio_msiq_init(DIP_TO_HANDLE(dip),
765 	    pxu_p)) != H_EOK) {
766 		DBG(DBG_LIB_MSIQ, dip,
767 		    "hvio_msiq_init failed, ret 0x%lx\n", ret);
768 
769 		(void) px_lib_msiq_fini(dip);
770 		return (DDI_FAILURE);
771 	}
772 
773 	return (DDI_SUCCESS);
774 }
775 
776 /*ARGSUSED*/
777 int
778 px_lib_msiq_fini(dev_info_t *dip)
779 {
780 	px_t		*px_p = DIP_TO_STATE(dip);
781 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
782 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
783 	px_dvma_addr_t	pg_index;
784 	size_t		size;
785 
786 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_fini: dip 0x%p\n", dip);
787 
788 	/*
789 	 * Unmap and free the EQ memory that had been mapped
790 	 * into the Fire IOMMU.
791 	 */
792 	size = msiq_state_p->msiq_cnt *
793 	    msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
794 
795 	pg_index = MMU_PAGE_INDEX(px_p->px_mmu_p,
796 	    MMU_BTOP((ulong_t)pxu_p->msiq_mapped_p));
797 
798 	(void) px_lib_iommu_demap(px_p->px_dip,
799 	    PCI_TSBID(0, pg_index), MMU_BTOP(size));
800 
801 	/* Free the entries from the Fire MMU */
802 	vmem_xfree(px_p->px_mmu_p->mmu_dvma_map,
803 	    (void *)pxu_p->msiq_mapped_p, size);
804 
805 	kmem_free(msiq_state_p->msiq_buf_p, msiq_state_p->msiq_cnt *
806 	    msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t));
807 
808 	return (DDI_SUCCESS);
809 }
810 
811 /*ARGSUSED*/
812 int
813 px_lib_msiq_info(dev_info_t *dip, msiqid_t msiq_id, r_addr_t *ra_p,
814     uint_t *msiq_rec_cnt_p)
815 {
816 	px_t		*px_p = DIP_TO_STATE(dip);
817 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
818 	size_t		msiq_size;
819 
820 	DBG(DBG_LIB_MSIQ, dip, "px_msiq_info: dip 0x%p msiq_id 0x%x\n",
821 	    dip, msiq_id);
822 
823 	msiq_size = msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t);
824 	ra_p = (r_addr_t *)((caddr_t)msiq_state_p->msiq_buf_p +
825 	    (msiq_id * msiq_size));
826 
827 	*msiq_rec_cnt_p = msiq_state_p->msiq_rec_cnt;
828 
829 	DBG(DBG_LIB_MSIQ, dip, "px_msiq_info: ra_p 0x%p msiq_rec_cnt 0x%x\n",
830 	    ra_p, *msiq_rec_cnt_p);
831 
832 	return (DDI_SUCCESS);
833 }
834 
835 /*ARGSUSED*/
836 int
837 px_lib_msiq_getvalid(dev_info_t *dip, msiqid_t msiq_id,
838     pci_msiq_valid_state_t *msiq_valid_state)
839 {
840 	uint64_t	ret;
841 
842 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getvalid: dip 0x%p msiq_id 0x%x\n",
843 	    dip, msiq_id);
844 
845 	if ((ret = hvio_msiq_getvalid(DIP_TO_HANDLE(dip),
846 	    msiq_id, msiq_valid_state)) != H_EOK) {
847 		DBG(DBG_LIB_MSIQ, dip,
848 		    "hvio_msiq_getvalid failed, ret 0x%lx\n", ret);
849 		return (DDI_FAILURE);
850 	}
851 
852 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getvalid: msiq_valid_state 0x%x\n",
853 	    *msiq_valid_state);
854 
855 	return (DDI_SUCCESS);
856 }
857 
858 /*ARGSUSED*/
859 int
860 px_lib_msiq_setvalid(dev_info_t *dip, msiqid_t msiq_id,
861     pci_msiq_valid_state_t msiq_valid_state)
862 {
863 	uint64_t	ret;
864 
865 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_setvalid: dip 0x%p msiq_id 0x%x "
866 	    "msiq_valid_state 0x%x\n", dip, msiq_id, msiq_valid_state);
867 
868 	if ((ret = hvio_msiq_setvalid(DIP_TO_HANDLE(dip),
869 	    msiq_id, msiq_valid_state)) != H_EOK) {
870 		DBG(DBG_LIB_MSIQ, dip,
871 		    "hvio_msiq_setvalid failed, ret 0x%lx\n", ret);
872 		return (DDI_FAILURE);
873 	}
874 
875 	return (DDI_SUCCESS);
876 }
877 
878 /*ARGSUSED*/
879 int
880 px_lib_msiq_getstate(dev_info_t *dip, msiqid_t msiq_id,
881     pci_msiq_state_t *msiq_state)
882 {
883 	uint64_t	ret;
884 
885 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getstate: dip 0x%p msiq_id 0x%x\n",
886 	    dip, msiq_id);
887 
888 	if ((ret = hvio_msiq_getstate(DIP_TO_HANDLE(dip),
889 	    msiq_id, msiq_state)) != H_EOK) {
890 		DBG(DBG_LIB_MSIQ, dip,
891 		    "hvio_msiq_getstate failed, ret 0x%lx\n", ret);
892 		return (DDI_FAILURE);
893 	}
894 
895 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_getstate: msiq_state 0x%x\n",
896 	    *msiq_state);
897 
898 	return (DDI_SUCCESS);
899 }
900 
901 /*ARGSUSED*/
902 int
903 px_lib_msiq_setstate(dev_info_t *dip, msiqid_t msiq_id,
904     pci_msiq_state_t msiq_state)
905 {
906 	uint64_t	ret;
907 
908 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_setstate: dip 0x%p msiq_id 0x%x "
909 	    "msiq_state 0x%x\n", dip, msiq_id, msiq_state);
910 
911 	if ((ret = hvio_msiq_setstate(DIP_TO_HANDLE(dip),
912 	    msiq_id, msiq_state)) != H_EOK) {
913 		DBG(DBG_LIB_MSIQ, dip,
914 		    "hvio_msiq_setstate failed, ret 0x%lx\n", ret);
915 		return (DDI_FAILURE);
916 	}
917 
918 	return (DDI_SUCCESS);
919 }
920 
921 /*ARGSUSED*/
922 int
923 px_lib_msiq_gethead(dev_info_t *dip, msiqid_t msiq_id,
924     msiqhead_t *msiq_head)
925 {
926 	uint64_t	ret;
927 
928 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gethead: dip 0x%p msiq_id 0x%x\n",
929 	    dip, msiq_id);
930 
931 	if ((ret = hvio_msiq_gethead(DIP_TO_HANDLE(dip),
932 	    msiq_id, msiq_head)) != H_EOK) {
933 		DBG(DBG_LIB_MSIQ, dip,
934 		    "hvio_msiq_gethead failed, ret 0x%lx\n", ret);
935 		return (DDI_FAILURE);
936 	}
937 
938 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gethead: msiq_head 0x%x\n",
939 	    *msiq_head);
940 
941 	return (DDI_SUCCESS);
942 }
943 
944 /*ARGSUSED*/
945 int
946 px_lib_msiq_sethead(dev_info_t *dip, msiqid_t msiq_id,
947     msiqhead_t msiq_head)
948 {
949 	uint64_t	ret;
950 
951 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_sethead: dip 0x%p msiq_id 0x%x "
952 	    "msiq_head 0x%x\n", dip, msiq_id, msiq_head);
953 
954 	if ((ret = hvio_msiq_sethead(DIP_TO_HANDLE(dip),
955 	    msiq_id, msiq_head)) != H_EOK) {
956 		DBG(DBG_LIB_MSIQ, dip,
957 		    "hvio_msiq_sethead failed, ret 0x%lx\n", ret);
958 		return (DDI_FAILURE);
959 	}
960 
961 	return (DDI_SUCCESS);
962 }
963 
964 /*ARGSUSED*/
965 int
966 px_lib_msiq_gettail(dev_info_t *dip, msiqid_t msiq_id,
967     msiqtail_t *msiq_tail)
968 {
969 	uint64_t	ret;
970 
971 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gettail: dip 0x%p msiq_id 0x%x\n",
972 	    dip, msiq_id);
973 
974 	if ((ret = hvio_msiq_gettail(DIP_TO_HANDLE(dip),
975 	    msiq_id, msiq_tail)) != H_EOK) {
976 		DBG(DBG_LIB_MSIQ, dip,
977 		    "hvio_msiq_gettail failed, ret 0x%lx\n", ret);
978 		return (DDI_FAILURE);
979 	}
980 
981 	DBG(DBG_LIB_MSIQ, dip, "px_lib_msiq_gettail: msiq_tail 0x%x\n",
982 	    *msiq_tail);
983 
984 	return (DDI_SUCCESS);
985 }
986 
987 /*ARGSUSED*/
988 void
989 px_lib_get_msiq_rec(dev_info_t *dip, msiqhead_t *msiq_head_p,
990     msiq_rec_t *msiq_rec_p)
991 {
992 	eq_rec_t	*eq_rec_p = (eq_rec_t *)msiq_head_p;
993 
994 	DBG(DBG_LIB_MSIQ, dip, "px_lib_get_msiq_rec: dip 0x%p eq_rec_p 0x%p\n",
995 	    dip, eq_rec_p);
996 
997 	if (!eq_rec_p->eq_rec_fmt_type) {
998 		/* Set msiq_rec_type to zero */
999 		msiq_rec_p->msiq_rec_type = 0;
1000 
1001 		return;
1002 	}
1003 
1004 	DBG(DBG_LIB_MSIQ, dip, "px_lib_get_msiq_rec: EQ RECORD, "
1005 	    "eq_rec_rid 0x%llx eq_rec_fmt_type 0x%llx "
1006 	    "eq_rec_len 0x%llx eq_rec_addr0 0x%llx "
1007 	    "eq_rec_addr1 0x%llx eq_rec_data0 0x%llx "
1008 	    "eq_rec_data1 0x%llx\n", eq_rec_p->eq_rec_rid,
1009 	    eq_rec_p->eq_rec_fmt_type, eq_rec_p->eq_rec_len,
1010 	    eq_rec_p->eq_rec_addr0, eq_rec_p->eq_rec_addr1,
1011 	    eq_rec_p->eq_rec_data0, eq_rec_p->eq_rec_data1);
1012 
1013 	/*
1014 	 * Only upper 4 bits of eq_rec_fmt_type is used
1015 	 * to identify the EQ record type.
1016 	 */
1017 	switch (eq_rec_p->eq_rec_fmt_type >> 3) {
1018 	case EQ_REC_MSI32:
1019 		msiq_rec_p->msiq_rec_type = MSI32_REC;
1020 
1021 		msiq_rec_p->msiq_rec_data.msi.msi_data =
1022 		    eq_rec_p->eq_rec_data0;
1023 		break;
1024 	case EQ_REC_MSI64:
1025 		msiq_rec_p->msiq_rec_type = MSI64_REC;
1026 
1027 		msiq_rec_p->msiq_rec_data.msi.msi_data =
1028 		    eq_rec_p->eq_rec_data0;
1029 		break;
1030 	case EQ_REC_MSG:
1031 		msiq_rec_p->msiq_rec_type = MSG_REC;
1032 
1033 		msiq_rec_p->msiq_rec_data.msg.msg_route =
1034 		    eq_rec_p->eq_rec_fmt_type & 7;
1035 		msiq_rec_p->msiq_rec_data.msg.msg_targ = eq_rec_p->eq_rec_rid;
1036 		msiq_rec_p->msiq_rec_data.msg.msg_code = eq_rec_p->eq_rec_data0;
1037 		break;
1038 	default:
1039 		cmn_err(CE_WARN, "%s%d: px_lib_get_msiq_rec: "
1040 		    "0x%x is an unknown EQ record type",
1041 		    ddi_driver_name(dip), ddi_get_instance(dip),
1042 		    (int)eq_rec_p->eq_rec_fmt_type);
1043 		break;
1044 	}
1045 
1046 	msiq_rec_p->msiq_rec_rid = eq_rec_p->eq_rec_rid;
1047 	msiq_rec_p->msiq_rec_msi_addr = ((eq_rec_p->eq_rec_addr1 << 16) |
1048 	    (eq_rec_p->eq_rec_addr0 << 2));
1049 }
1050 
1051 /*ARGSUSED*/
1052 void
1053 px_lib_clr_msiq_rec(dev_info_t *dip, msiqhead_t *msiq_head_p)
1054 {
1055 	eq_rec_t	*eq_rec_p = (eq_rec_t *)msiq_head_p;
1056 
1057 	DBG(DBG_LIB_MSIQ, dip, "px_lib_clr_msiq_rec: dip 0x%p eq_rec_p 0x%p\n",
1058 	    dip, eq_rec_p);
1059 
1060 	if (eq_rec_p->eq_rec_fmt_type) {
1061 		/* Zero out eq_rec_fmt_type field */
1062 		eq_rec_p->eq_rec_fmt_type = 0;
1063 	}
1064 }
1065 
1066 /*
1067  * MSI Functions:
1068  */
1069 /*ARGSUSED*/
1070 int
1071 px_lib_msi_init(dev_info_t *dip)
1072 {
1073 	px_t		*px_p = DIP_TO_STATE(dip);
1074 	px_msi_state_t	*msi_state_p = &px_p->px_ib_p->ib_msi_state;
1075 	uint64_t	ret;
1076 
1077 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_init: dip 0x%p\n", dip);
1078 
1079 	if ((ret = hvio_msi_init(DIP_TO_HANDLE(dip),
1080 	    msi_state_p->msi_addr32, msi_state_p->msi_addr64)) != H_EOK) {
1081 		DBG(DBG_LIB_MSIQ, dip, "px_lib_msi_init failed, ret 0x%lx\n",
1082 		    ret);
1083 		return (DDI_FAILURE);
1084 	}
1085 
1086 	return (DDI_SUCCESS);
1087 }
1088 
1089 /*ARGSUSED*/
1090 int
1091 px_lib_msi_getmsiq(dev_info_t *dip, msinum_t msi_num,
1092     msiqid_t *msiq_id)
1093 {
1094 	uint64_t	ret;
1095 
1096 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getmsiq: dip 0x%p msi_num 0x%x\n",
1097 	    dip, msi_num);
1098 
1099 	if ((ret = hvio_msi_getmsiq(DIP_TO_HANDLE(dip),
1100 	    msi_num, msiq_id)) != H_EOK) {
1101 		DBG(DBG_LIB_MSI, dip,
1102 		    "hvio_msi_getmsiq failed, ret 0x%lx\n", ret);
1103 		return (DDI_FAILURE);
1104 	}
1105 
1106 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getmsiq: msiq_id 0x%x\n",
1107 	    *msiq_id);
1108 
1109 	return (DDI_SUCCESS);
1110 }
1111 
1112 /*ARGSUSED*/
1113 int
1114 px_lib_msi_setmsiq(dev_info_t *dip, msinum_t msi_num,
1115     msiqid_t msiq_id, msi_type_t msitype)
1116 {
1117 	uint64_t	ret;
1118 
1119 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setmsiq: dip 0x%p msi_num 0x%x "
1120 	    "msq_id 0x%x\n", dip, msi_num, msiq_id);
1121 
1122 	if ((ret = hvio_msi_setmsiq(DIP_TO_HANDLE(dip),
1123 	    msi_num, msiq_id)) != H_EOK) {
1124 		DBG(DBG_LIB_MSI, dip,
1125 		    "hvio_msi_setmsiq failed, ret 0x%lx\n", ret);
1126 		return (DDI_FAILURE);
1127 	}
1128 
1129 	return (DDI_SUCCESS);
1130 }
1131 
1132 /*ARGSUSED*/
1133 int
1134 px_lib_msi_getvalid(dev_info_t *dip, msinum_t msi_num,
1135     pci_msi_valid_state_t *msi_valid_state)
1136 {
1137 	uint64_t	ret;
1138 
1139 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getvalid: dip 0x%p msi_num 0x%x\n",
1140 	    dip, msi_num);
1141 
1142 	if ((ret = hvio_msi_getvalid(DIP_TO_HANDLE(dip),
1143 	    msi_num, msi_valid_state)) != H_EOK) {
1144 		DBG(DBG_LIB_MSI, dip,
1145 		    "hvio_msi_getvalid failed, ret 0x%lx\n", ret);
1146 		return (DDI_FAILURE);
1147 	}
1148 
1149 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getvalid: msiq_id 0x%x\n",
1150 	    *msi_valid_state);
1151 
1152 	return (DDI_SUCCESS);
1153 }
1154 
1155 /*ARGSUSED*/
1156 int
1157 px_lib_msi_setvalid(dev_info_t *dip, msinum_t msi_num,
1158     pci_msi_valid_state_t msi_valid_state)
1159 {
1160 	uint64_t	ret;
1161 
1162 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setvalid: dip 0x%p msi_num 0x%x "
1163 	    "msi_valid_state 0x%x\n", dip, msi_num, msi_valid_state);
1164 
1165 	if ((ret = hvio_msi_setvalid(DIP_TO_HANDLE(dip),
1166 	    msi_num, msi_valid_state)) != H_EOK) {
1167 		DBG(DBG_LIB_MSI, dip,
1168 		    "hvio_msi_setvalid failed, ret 0x%lx\n", ret);
1169 		return (DDI_FAILURE);
1170 	}
1171 
1172 	return (DDI_SUCCESS);
1173 }
1174 
1175 /*ARGSUSED*/
1176 int
1177 px_lib_msi_getstate(dev_info_t *dip, msinum_t msi_num,
1178     pci_msi_state_t *msi_state)
1179 {
1180 	uint64_t	ret;
1181 
1182 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getstate: dip 0x%p msi_num 0x%x\n",
1183 	    dip, msi_num);
1184 
1185 	if ((ret = hvio_msi_getstate(DIP_TO_HANDLE(dip),
1186 	    msi_num, msi_state)) != H_EOK) {
1187 		DBG(DBG_LIB_MSI, dip,
1188 		    "hvio_msi_getstate failed, ret 0x%lx\n", ret);
1189 		return (DDI_FAILURE);
1190 	}
1191 
1192 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_getstate: msi_state 0x%x\n",
1193 	    *msi_state);
1194 
1195 	return (DDI_SUCCESS);
1196 }
1197 
1198 /*ARGSUSED*/
1199 int
1200 px_lib_msi_setstate(dev_info_t *dip, msinum_t msi_num,
1201     pci_msi_state_t msi_state)
1202 {
1203 	uint64_t	ret;
1204 
1205 	DBG(DBG_LIB_MSI, dip, "px_lib_msi_setstate: dip 0x%p msi_num 0x%x "
1206 	    "msi_state 0x%x\n", dip, msi_num, msi_state);
1207 
1208 	if ((ret = hvio_msi_setstate(DIP_TO_HANDLE(dip),
1209 	    msi_num, msi_state)) != H_EOK) {
1210 		DBG(DBG_LIB_MSI, dip,
1211 		    "hvio_msi_setstate failed, ret 0x%lx\n", ret);
1212 		return (DDI_FAILURE);
1213 	}
1214 
1215 	return (DDI_SUCCESS);
1216 }
1217 
1218 /*
1219  * MSG Functions:
1220  */
1221 /*ARGSUSED*/
1222 int
1223 px_lib_msg_getmsiq(dev_info_t *dip, pcie_msg_type_t msg_type,
1224     msiqid_t *msiq_id)
1225 {
1226 	uint64_t	ret;
1227 
1228 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_getmsiq: dip 0x%p msg_type 0x%x\n",
1229 	    dip, msg_type);
1230 
1231 	if ((ret = hvio_msg_getmsiq(DIP_TO_HANDLE(dip),
1232 	    msg_type, msiq_id)) != H_EOK) {
1233 		DBG(DBG_LIB_MSG, dip,
1234 		    "hvio_msg_getmsiq failed, ret 0x%lx\n", ret);
1235 		return (DDI_FAILURE);
1236 	}
1237 
1238 	DBG(DBG_LIB_MSI, dip, "px_lib_msg_getmsiq: msiq_id 0x%x\n",
1239 	    *msiq_id);
1240 
1241 	return (DDI_SUCCESS);
1242 }
1243 
1244 /*ARGSUSED*/
1245 int
1246 px_lib_msg_setmsiq(dev_info_t *dip, pcie_msg_type_t msg_type,
1247     msiqid_t msiq_id)
1248 {
1249 	uint64_t	ret;
1250 
1251 	DBG(DBG_LIB_MSG, dip, "px_lib_msi_setstate: dip 0x%p msg_type 0x%x "
1252 	    "msiq_id 0x%x\n", dip, msg_type, msiq_id);
1253 
1254 	if ((ret = hvio_msg_setmsiq(DIP_TO_HANDLE(dip),
1255 	    msg_type, msiq_id)) != H_EOK) {
1256 		DBG(DBG_LIB_MSG, dip,
1257 		    "hvio_msg_setmsiq failed, ret 0x%lx\n", ret);
1258 		return (DDI_FAILURE);
1259 	}
1260 
1261 	return (DDI_SUCCESS);
1262 }
1263 
1264 /*ARGSUSED*/
1265 int
1266 px_lib_msg_getvalid(dev_info_t *dip, pcie_msg_type_t msg_type,
1267     pcie_msg_valid_state_t *msg_valid_state)
1268 {
1269 	uint64_t	ret;
1270 
1271 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_getvalid: dip 0x%p msg_type 0x%x\n",
1272 	    dip, msg_type);
1273 
1274 	if ((ret = hvio_msg_getvalid(DIP_TO_HANDLE(dip), msg_type,
1275 	    msg_valid_state)) != H_EOK) {
1276 		DBG(DBG_LIB_MSG, dip,
1277 		    "hvio_msg_getvalid failed, ret 0x%lx\n", ret);
1278 		return (DDI_FAILURE);
1279 	}
1280 
1281 	DBG(DBG_LIB_MSI, dip, "px_lib_msg_getvalid: msg_valid_state 0x%x\n",
1282 	    *msg_valid_state);
1283 
1284 	return (DDI_SUCCESS);
1285 }
1286 
1287 /*ARGSUSED*/
1288 int
1289 px_lib_msg_setvalid(dev_info_t *dip, pcie_msg_type_t msg_type,
1290     pcie_msg_valid_state_t msg_valid_state)
1291 {
1292 	uint64_t	ret;
1293 
1294 	DBG(DBG_LIB_MSG, dip, "px_lib_msg_setvalid: dip 0x%p msg_type 0x%x "
1295 	    "msg_valid_state 0x%x\n", dip, msg_type, msg_valid_state);
1296 
1297 	if ((ret = hvio_msg_setvalid(DIP_TO_HANDLE(dip), msg_type,
1298 	    msg_valid_state)) != H_EOK) {
1299 		DBG(DBG_LIB_MSG, dip,
1300 		    "hvio_msg_setvalid failed, ret 0x%lx\n", ret);
1301 		return (DDI_FAILURE);
1302 	}
1303 
1304 	return (DDI_SUCCESS);
1305 }
1306 
1307 /*ARGSUSED*/
1308 void
1309 px_panic_domain(px_t *px_p, pcie_req_id_t bdf)
1310 {
1311 }
1312 
1313 /*
1314  * Suspend/Resume Functions:
1315  * Currently unsupported by hypervisor
1316  */
1317 int
1318 px_lib_suspend(dev_info_t *dip)
1319 {
1320 	px_t		*px_p = DIP_TO_STATE(dip);
1321 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1322 	px_cb_t		*cb_p = PX2CB(px_p);
1323 	devhandle_t	dev_hdl, xbus_dev_hdl;
1324 	uint64_t	ret = H_EOK;
1325 
1326 	DBG(DBG_DETACH, dip, "px_lib_suspend: dip 0x%p\n", dip);
1327 
1328 	dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_CSR];
1329 	xbus_dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_XBC];
1330 
1331 	if ((ret = hvio_suspend(dev_hdl, pxu_p)) != H_EOK)
1332 		goto fail;
1333 
1334 	if (--cb_p->attachcnt == 0) {
1335 		ret = hvio_cb_suspend(xbus_dev_hdl, pxu_p);
1336 		if (ret != H_EOK)
1337 			cb_p->attachcnt++;
1338 	}
1339 	pxu_p->cpr_flag = PX_ENTERED_CPR;
1340 
1341 fail:
1342 	return ((ret != H_EOK) ? DDI_FAILURE: DDI_SUCCESS);
1343 }
1344 
1345 void
1346 px_lib_resume(dev_info_t *dip)
1347 {
1348 	px_t		*px_p = DIP_TO_STATE(dip);
1349 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1350 	px_cb_t		*cb_p = PX2CB(px_p);
1351 	devhandle_t	dev_hdl, xbus_dev_hdl;
1352 	devino_t	pec_ino = px_p->px_inos[PX_INTR_PEC];
1353 	devino_t	xbc_ino = px_p->px_inos[PX_INTR_XBC];
1354 
1355 	DBG(DBG_ATTACH, dip, "px_lib_resume: dip 0x%p\n", dip);
1356 
1357 	dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_CSR];
1358 	xbus_dev_hdl = (devhandle_t)pxu_p->px_address[PX_REG_XBC];
1359 
1360 	if (++cb_p->attachcnt == 1)
1361 		hvio_cb_resume(dev_hdl, xbus_dev_hdl, xbc_ino, pxu_p);
1362 
1363 	hvio_resume(dev_hdl, pec_ino, pxu_p);
1364 }
1365 
1366 /*
1367  * Generate a unique Oberon UBC ID based on the Logicial System Board and
1368  * the IO Channel from the portid property field.
1369  */
1370 static uint64_t
1371 oberon_get_ubc_id(dev_info_t *dip)
1372 {
1373 	px_t	*px_p = DIP_TO_STATE(dip);
1374 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1375 	uint64_t	ubc_id;
1376 
1377 	/*
1378 	 * Generate a unique 6 bit UBC ID using the 2 IO_Channel#[1:0] bits and
1379 	 * the 4 LSB_ID[3:0] bits from the Oberon's portid property.
1380 	 */
1381 	ubc_id = (((pxu_p->portid >> OBERON_PORT_ID_IOC) &
1382 	    OBERON_PORT_ID_IOC_MASK) | (((pxu_p->portid >>
1383 	    OBERON_PORT_ID_LSB) & OBERON_PORT_ID_LSB_MASK)
1384 	    << OBERON_UBC_ID_LSB));
1385 
1386 	return (ubc_id);
1387 }
1388 
1389 /*
1390  * Oberon does not have a UBC scratch register, so alloc an array of scratch
1391  * registers when needed and use a unique UBC ID as an index. This code
1392  * can be simplified if we use a pre-allocated array. They are currently
1393  * being dynamically allocated because it's only needed by the Oberon.
1394  */
1395 static void
1396 oberon_set_cb(dev_info_t *dip, uint64_t val)
1397 {
1398 	uint64_t	ubc_id;
1399 
1400 	if (px_oberon_ubc_scratch_regs == NULL)
1401 		px_oberon_ubc_scratch_regs =
1402 		    (uint64_t *)kmem_zalloc(sizeof (uint64_t)*
1403 		    OBERON_UBC_ID_MAX, KM_SLEEP);
1404 
1405 	ubc_id = oberon_get_ubc_id(dip);
1406 
1407 	px_oberon_ubc_scratch_regs[ubc_id] = val;
1408 
1409 	/*
1410 	 * Check if any scratch registers are still in use. If all scratch
1411 	 * registers are currently set to zero, then deallocate the scratch
1412 	 * register array.
1413 	 */
1414 	for (ubc_id = 0; ubc_id < OBERON_UBC_ID_MAX; ubc_id++) {
1415 		if (px_oberon_ubc_scratch_regs[ubc_id] != 0)
1416 			return;
1417 	}
1418 
1419 	/*
1420 	 * All scratch registers are set to zero so deallocate the scratch
1421 	 * register array and set the pointer to NULL.
1422 	 */
1423 	kmem_free(px_oberon_ubc_scratch_regs,
1424 	    (sizeof (uint64_t)*OBERON_UBC_ID_MAX));
1425 
1426 	px_oberon_ubc_scratch_regs = NULL;
1427 }
1428 
1429 /*
1430  * Oberon does not have a UBC scratch register, so use an allocated array of
1431  * scratch registers and use the unique UBC ID as an index into that array.
1432  */
1433 static uint64_t
1434 oberon_get_cb(dev_info_t *dip)
1435 {
1436 	uint64_t	ubc_id;
1437 
1438 	if (px_oberon_ubc_scratch_regs == NULL)
1439 		return (0);
1440 
1441 	ubc_id = oberon_get_ubc_id(dip);
1442 
1443 	return (px_oberon_ubc_scratch_regs[ubc_id]);
1444 }
1445 
1446 /*
1447  * Misc Functions:
1448  * Currently unsupported by hypervisor
1449  */
1450 static uint64_t
1451 px_get_cb(dev_info_t *dip)
1452 {
1453 	px_t	*px_p = DIP_TO_STATE(dip);
1454 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1455 
1456 	/*
1457 	 * Oberon does not currently have Scratchpad registers.
1458 	 */
1459 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON)
1460 		return (oberon_get_cb(dip));
1461 
1462 	return (CSR_XR((caddr_t)pxu_p->px_address[PX_REG_XBC], JBUS_SCRATCH_1));
1463 }
1464 
1465 static void
1466 px_set_cb(dev_info_t *dip, uint64_t val)
1467 {
1468 	px_t	*px_p = DIP_TO_STATE(dip);
1469 	pxu_t	*pxu_p = (pxu_t *)px_p->px_plat_p;
1470 
1471 	/*
1472 	 * Oberon does not currently have Scratchpad registers.
1473 	 */
1474 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON) {
1475 		oberon_set_cb(dip, val);
1476 		return;
1477 	}
1478 
1479 	CSR_XS((caddr_t)pxu_p->px_address[PX_REG_XBC], JBUS_SCRATCH_1, val);
1480 }
1481 
1482 /*ARGSUSED*/
1483 int
1484 px_lib_map_vconfig(dev_info_t *dip,
1485 	ddi_map_req_t *mp, pci_config_offset_t off,
1486 		pci_regspec_t *rp, caddr_t *addrp)
1487 {
1488 	/*
1489 	 * No special config space access services in this layer.
1490 	 */
1491 	return (DDI_FAILURE);
1492 }
1493 
1494 void
1495 px_lib_map_attr_check(ddi_map_req_t *mp)
1496 {
1497 	ddi_acc_hdl_t *hp = mp->map_handlep;
1498 
1499 	/* fire does not accept byte masks from PIO store merge */
1500 	if (hp->ah_acc.devacc_attr_dataorder == DDI_STORECACHING_OK_ACC)
1501 		hp->ah_acc.devacc_attr_dataorder = DDI_STRICTORDER_ACC;
1502 }
1503 
1504 /* This function is called only by poke, caut put and pxtool poke. */
1505 void
1506 px_lib_clr_errs(px_t *px_p, dev_info_t *rdip, uint64_t addr)
1507 {
1508 	px_pec_t	*pec_p = px_p->px_pec_p;
1509 	dev_info_t	*rpdip = px_p->px_dip;
1510 	int		rc_err, fab_err, i;
1511 	int		acctype = pec_p->pec_safeacc_type;
1512 	ddi_fm_error_t	derr;
1513 	pci_ranges_t	*ranges_p;
1514 	int		range_len;
1515 	uint32_t	addr_high, addr_low;
1516 	pcie_req_id_t	bdf = PCIE_INVALID_BDF;
1517 
1518 	/* Create the derr */
1519 	bzero(&derr, sizeof (ddi_fm_error_t));
1520 	derr.fme_version = DDI_FME_VERSION;
1521 	derr.fme_ena = fm_ena_generate(0, FM_ENA_FMT1);
1522 	derr.fme_flag = acctype;
1523 
1524 	if (acctype == DDI_FM_ERR_EXPECTED) {
1525 		derr.fme_status = DDI_FM_NONFATAL;
1526 		ndi_fm_acc_err_set(pec_p->pec_acc_hdl, &derr);
1527 	}
1528 
1529 	if (px_fm_enter(px_p) != DDI_SUCCESS)
1530 		return;
1531 
1532 	/* send ereport/handle/clear fire registers */
1533 	rc_err = px_err_cmn_intr(px_p, &derr, PX_LIB_CALL, PX_FM_BLOCK_ALL);
1534 
1535 	/* Figure out if this is a cfg or mem32 access */
1536 	addr_high = (uint32_t)(addr >> 32);
1537 	addr_low = (uint32_t)addr;
1538 	range_len = px_p->px_ranges_length / sizeof (pci_ranges_t);
1539 	i = 0;
1540 	for (ranges_p = px_p->px_ranges_p; i < range_len; i++, ranges_p++) {
1541 		if (ranges_p->parent_high == addr_high) {
1542 			switch (ranges_p->child_high & PCI_ADDR_MASK) {
1543 			case PCI_ADDR_CONFIG:
1544 				bdf = (pcie_req_id_t)(addr_low >> 12);
1545 				addr_low = 0;
1546 				break;
1547 			case PCI_ADDR_MEM32:
1548 				if (rdip)
1549 					bdf = PCI_GET_BDF(rdip);
1550 				else
1551 					bdf = PCIE_INVALID_BDF;
1552 				break;
1553 			}
1554 			break;
1555 		}
1556 	}
1557 
1558 	(void) px_rp_en_q(px_p, bdf, addr_low, 0);
1559 
1560 	/*
1561 	 * XXX - Current code scans the fabric for all px_tool accesses.
1562 	 * In future, do not scan fabric for px_tool access to IO Root Nexus
1563 	 */
1564 	fab_err = px_scan_fabric(px_p, rpdip, &derr);
1565 
1566 	px_err_panic(rc_err, PX_RC, fab_err, B_TRUE);
1567 	px_fm_exit(px_p);
1568 	px_err_panic(rc_err, PX_RC, fab_err, B_FALSE);
1569 }
1570 
1571 #ifdef  DEBUG
1572 int	px_peekfault_cnt = 0;
1573 int	px_pokefault_cnt = 0;
1574 #endif  /* DEBUG */
1575 
1576 /*ARGSUSED*/
1577 static int
1578 px_lib_do_poke(dev_info_t *dip, dev_info_t *rdip,
1579     peekpoke_ctlops_t *in_args)
1580 {
1581 	px_t *px_p = DIP_TO_STATE(dip);
1582 	px_pec_t *pec_p = px_p->px_pec_p;
1583 	int err = DDI_SUCCESS;
1584 	on_trap_data_t otd;
1585 
1586 	mutex_enter(&pec_p->pec_pokefault_mutex);
1587 	pec_p->pec_ontrap_data = &otd;
1588 	pec_p->pec_safeacc_type = DDI_FM_ERR_POKE;
1589 
1590 	/* Set up protected environment. */
1591 	if (!on_trap(&otd, OT_DATA_ACCESS)) {
1592 		uintptr_t tramp = otd.ot_trampoline;
1593 
1594 		otd.ot_trampoline = (uintptr_t)&poke_fault;
1595 		err = do_poke(in_args->size, (void *)in_args->dev_addr,
1596 		    (void *)in_args->host_addr);
1597 		otd.ot_trampoline = tramp;
1598 	} else
1599 		err = DDI_FAILURE;
1600 
1601 	px_lib_clr_errs(px_p, rdip, in_args->dev_addr);
1602 
1603 	if (otd.ot_trap & OT_DATA_ACCESS)
1604 		err = DDI_FAILURE;
1605 
1606 	/* Take down protected environment. */
1607 	no_trap();
1608 
1609 	pec_p->pec_ontrap_data = NULL;
1610 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1611 	mutex_exit(&pec_p->pec_pokefault_mutex);
1612 
1613 #ifdef  DEBUG
1614 	if (err == DDI_FAILURE)
1615 		px_pokefault_cnt++;
1616 #endif
1617 	return (err);
1618 }
1619 
1620 /*ARGSUSED*/
1621 static int
1622 px_lib_do_caut_put(dev_info_t *dip, dev_info_t *rdip,
1623     peekpoke_ctlops_t *cautacc_ctlops_arg)
1624 {
1625 	size_t size = cautacc_ctlops_arg->size;
1626 	uintptr_t dev_addr = cautacc_ctlops_arg->dev_addr;
1627 	uintptr_t host_addr = cautacc_ctlops_arg->host_addr;
1628 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)cautacc_ctlops_arg->handle;
1629 	size_t repcount = cautacc_ctlops_arg->repcount;
1630 	uint_t flags = cautacc_ctlops_arg->flags;
1631 
1632 	px_t *px_p = DIP_TO_STATE(dip);
1633 	px_pec_t *pec_p = px_p->px_pec_p;
1634 	int err = DDI_SUCCESS;
1635 
1636 	/*
1637 	 * Note that i_ndi_busop_access_enter ends up grabbing the pokefault
1638 	 * mutex.
1639 	 */
1640 	i_ndi_busop_access_enter(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1641 
1642 	pec_p->pec_ontrap_data = (on_trap_data_t *)hp->ahi_err->err_ontrap;
1643 	pec_p->pec_safeacc_type = DDI_FM_ERR_EXPECTED;
1644 	hp->ahi_err->err_expected = DDI_FM_ERR_EXPECTED;
1645 
1646 	if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1647 		for (; repcount; repcount--) {
1648 			switch (size) {
1649 
1650 			case sizeof (uint8_t):
1651 				i_ddi_put8(hp, (uint8_t *)dev_addr,
1652 				    *(uint8_t *)host_addr);
1653 				break;
1654 
1655 			case sizeof (uint16_t):
1656 				i_ddi_put16(hp, (uint16_t *)dev_addr,
1657 				    *(uint16_t *)host_addr);
1658 				break;
1659 
1660 			case sizeof (uint32_t):
1661 				i_ddi_put32(hp, (uint32_t *)dev_addr,
1662 				    *(uint32_t *)host_addr);
1663 				break;
1664 
1665 			case sizeof (uint64_t):
1666 				i_ddi_put64(hp, (uint64_t *)dev_addr,
1667 				    *(uint64_t *)host_addr);
1668 				break;
1669 			}
1670 
1671 			host_addr += size;
1672 
1673 			if (flags == DDI_DEV_AUTOINCR)
1674 				dev_addr += size;
1675 
1676 			px_lib_clr_errs(px_p, rdip, dev_addr);
1677 
1678 			if (pec_p->pec_ontrap_data->ot_trap & OT_DATA_ACCESS) {
1679 				err = DDI_FAILURE;
1680 #ifdef  DEBUG
1681 				px_pokefault_cnt++;
1682 #endif
1683 				break;
1684 			}
1685 		}
1686 	}
1687 
1688 	i_ddi_notrap((ddi_acc_handle_t)hp);
1689 	pec_p->pec_ontrap_data = NULL;
1690 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1691 	i_ndi_busop_access_exit(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1692 	hp->ahi_err->err_expected = DDI_FM_ERR_UNEXPECTED;
1693 
1694 	return (err);
1695 }
1696 
1697 
1698 int
1699 px_lib_ctlops_poke(dev_info_t *dip, dev_info_t *rdip,
1700     peekpoke_ctlops_t *in_args)
1701 {
1702 	return (in_args->handle ? px_lib_do_caut_put(dip, rdip, in_args) :
1703 	    px_lib_do_poke(dip, rdip, in_args));
1704 }
1705 
1706 
1707 /*ARGSUSED*/
1708 static int
1709 px_lib_do_peek(dev_info_t *dip, peekpoke_ctlops_t *in_args)
1710 {
1711 	px_t *px_p = DIP_TO_STATE(dip);
1712 	px_pec_t *pec_p = px_p->px_pec_p;
1713 	int err = DDI_SUCCESS;
1714 	on_trap_data_t otd;
1715 
1716 	mutex_enter(&pec_p->pec_pokefault_mutex);
1717 	if (px_fm_enter(px_p) != DDI_SUCCESS) {
1718 		mutex_exit(&pec_p->pec_pokefault_mutex);
1719 		return (DDI_FAILURE);
1720 	}
1721 	pec_p->pec_safeacc_type = DDI_FM_ERR_PEEK;
1722 	px_fm_exit(px_p);
1723 
1724 	if (!on_trap(&otd, OT_DATA_ACCESS)) {
1725 		uintptr_t tramp = otd.ot_trampoline;
1726 
1727 		otd.ot_trampoline = (uintptr_t)&peek_fault;
1728 		err = do_peek(in_args->size, (void *)in_args->dev_addr,
1729 		    (void *)in_args->host_addr);
1730 		otd.ot_trampoline = tramp;
1731 	} else
1732 		err = DDI_FAILURE;
1733 
1734 	no_trap();
1735 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1736 	mutex_exit(&pec_p->pec_pokefault_mutex);
1737 
1738 #ifdef  DEBUG
1739 	if (err == DDI_FAILURE)
1740 		px_peekfault_cnt++;
1741 #endif
1742 	return (err);
1743 }
1744 
1745 
1746 static int
1747 px_lib_do_caut_get(dev_info_t *dip, peekpoke_ctlops_t *cautacc_ctlops_arg)
1748 {
1749 	size_t size = cautacc_ctlops_arg->size;
1750 	uintptr_t dev_addr = cautacc_ctlops_arg->dev_addr;
1751 	uintptr_t host_addr = cautacc_ctlops_arg->host_addr;
1752 	ddi_acc_impl_t *hp = (ddi_acc_impl_t *)cautacc_ctlops_arg->handle;
1753 	size_t repcount = cautacc_ctlops_arg->repcount;
1754 	uint_t flags = cautacc_ctlops_arg->flags;
1755 
1756 	px_t *px_p = DIP_TO_STATE(dip);
1757 	px_pec_t *pec_p = px_p->px_pec_p;
1758 	int err = DDI_SUCCESS;
1759 
1760 	/*
1761 	 * Note that i_ndi_busop_access_enter ends up grabbing the pokefault
1762 	 * mutex.
1763 	 */
1764 	i_ndi_busop_access_enter(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1765 
1766 	pec_p->pec_ontrap_data = (on_trap_data_t *)hp->ahi_err->err_ontrap;
1767 	pec_p->pec_safeacc_type = DDI_FM_ERR_EXPECTED;
1768 	hp->ahi_err->err_expected = DDI_FM_ERR_EXPECTED;
1769 
1770 	if (repcount == 1) {
1771 		if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1772 			i_ddi_caut_get(size, (void *)dev_addr,
1773 			    (void *)host_addr);
1774 		} else {
1775 			int i;
1776 			uint8_t *ff_addr = (uint8_t *)host_addr;
1777 			for (i = 0; i < size; i++)
1778 				*ff_addr++ = 0xff;
1779 
1780 			err = DDI_FAILURE;
1781 #ifdef  DEBUG
1782 			px_peekfault_cnt++;
1783 #endif
1784 		}
1785 	} else {
1786 		if (!i_ddi_ontrap((ddi_acc_handle_t)hp)) {
1787 			for (; repcount; repcount--) {
1788 				i_ddi_caut_get(size, (void *)dev_addr,
1789 				    (void *)host_addr);
1790 
1791 				host_addr += size;
1792 
1793 				if (flags == DDI_DEV_AUTOINCR)
1794 					dev_addr += size;
1795 			}
1796 		} else {
1797 			err = DDI_FAILURE;
1798 #ifdef  DEBUG
1799 			px_peekfault_cnt++;
1800 #endif
1801 		}
1802 	}
1803 
1804 	i_ddi_notrap((ddi_acc_handle_t)hp);
1805 	pec_p->pec_ontrap_data = NULL;
1806 	pec_p->pec_safeacc_type = DDI_FM_ERR_UNEXPECTED;
1807 	i_ndi_busop_access_exit(hp->ahi_common.ah_dip, (ddi_acc_handle_t)hp);
1808 	hp->ahi_err->err_expected = DDI_FM_ERR_UNEXPECTED;
1809 
1810 	return (err);
1811 }
1812 
1813 /*ARGSUSED*/
1814 int
1815 px_lib_ctlops_peek(dev_info_t *dip, dev_info_t *rdip,
1816     peekpoke_ctlops_t *in_args, void *result)
1817 {
1818 	result = (void *)in_args->host_addr;
1819 	return (in_args->handle ? px_lib_do_caut_get(dip, in_args) :
1820 	    px_lib_do_peek(dip, in_args));
1821 }
1822 
1823 /*
1824  * implements PPM interface
1825  */
1826 int
1827 px_lib_pmctl(int cmd, px_t *px_p)
1828 {
1829 	ASSERT((cmd & ~PPMREQ_MASK) == PPMREQ);
1830 	switch (cmd) {
1831 	case PPMREQ_PRE_PWR_OFF:
1832 		/*
1833 		 * Currently there is no device power management for
1834 		 * the root complex (fire). When there is we need to make
1835 		 * sure that it is at full power before trying to send the
1836 		 * PME_Turn_Off message.
1837 		 */
1838 		DBG(DBG_PWR, px_p->px_dip,
1839 		    "ioctl: request to send PME_Turn_Off\n");
1840 		return (px_goto_l23ready(px_p));
1841 
1842 	case PPMREQ_PRE_PWR_ON:
1843 		DBG(DBG_PWR, px_p->px_dip, "ioctl: PRE_PWR_ON request\n");
1844 		return (px_pre_pwron_check(px_p));
1845 
1846 	case PPMREQ_POST_PWR_ON:
1847 		DBG(DBG_PWR, px_p->px_dip, "ioctl: POST_PWR_ON request\n");
1848 		return (px_goto_l0(px_p));
1849 
1850 	default:
1851 		return (DDI_FAILURE);
1852 	}
1853 }
1854 
1855 /*
1856  * sends PME_Turn_Off message to put the link in L2/L3 ready state.
1857  * called by px_ioctl.
1858  * returns DDI_SUCCESS or DDI_FAILURE
1859  * 1. Wait for link to be in L1 state (link status reg)
1860  * 2. write to PME_Turn_off reg to boradcast
1861  * 3. set timeout
1862  * 4. If timeout, return failure.
1863  * 5. If PM_TO_Ack, wait till link is in L2/L3 ready
1864  */
1865 static int
1866 px_goto_l23ready(px_t *px_p)
1867 {
1868 	pcie_pwr_t	*pwr_p;
1869 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
1870 	caddr_t	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
1871 	int		ret = DDI_SUCCESS;
1872 	clock_t		end, timeleft;
1873 	int		mutex_held = 1;
1874 
1875 	/* If no PM info, return failure */
1876 	if (PCIE_PMINFO(px_p->px_dip) == NULL ||
1877 	    (pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)) == NULL)
1878 		return (DDI_FAILURE);
1879 
1880 	mutex_enter(&pwr_p->pwr_lock);
1881 	mutex_enter(&px_p->px_l23ready_lock);
1882 	/* Clear the PME_To_ACK receieved flag */
1883 	px_p->px_pm_flags &= ~PX_PMETOACK_RECVD;
1884 	/*
1885 	 * When P25 is the downstream device, after receiving
1886 	 * PME_To_ACK, fire will go to Detect state, which causes
1887 	 * the link down event. Inform FMA that this is expected.
1888 	 * In case of all other cards complaint with the pci express
1889 	 * spec, this will happen when the power is re-applied. FMA
1890 	 * code will clear this flag after one instance of LDN. Since
1891 	 * there will not be a LDN event for the spec compliant cards,
1892 	 * we need to clear the flag after receiving PME_To_ACK.
1893 	 */
1894 	px_p->px_pm_flags |= PX_LDN_EXPECTED;
1895 	if (px_send_pme_turnoff(csr_base) != DDI_SUCCESS) {
1896 		ret = DDI_FAILURE;
1897 		goto l23ready_done;
1898 	}
1899 	px_p->px_pm_flags |= PX_PME_TURNOFF_PENDING;
1900 
1901 	end = ddi_get_lbolt() + drv_usectohz(px_pme_to_ack_timeout);
1902 	while (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1903 		timeleft = cv_timedwait(&px_p->px_l23ready_cv,
1904 		    &px_p->px_l23ready_lock, end);
1905 		/*
1906 		 * if cv_timedwait returns -1, it is either
1907 		 * 1) timed out or
1908 		 * 2) there was a pre-mature wakeup but by the time
1909 		 * cv_timedwait is called again end < lbolt i.e.
1910 		 * end is in the past.
1911 		 * 3) By the time we make first cv_timedwait call,
1912 		 * end < lbolt is true.
1913 		 */
1914 		if (timeleft == -1)
1915 			break;
1916 	}
1917 	if (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1918 		/*
1919 		 * Either timedout or interrupt didn't get a
1920 		 * chance to grab the mutex and set the flag.
1921 		 * release the mutex and delay for sometime.
1922 		 * This will 1) give a chance for interrupt to
1923 		 * set the flag 2) creates a delay between two
1924 		 * consequetive requests.
1925 		 */
1926 		mutex_exit(&px_p->px_l23ready_lock);
1927 		delay(drv_usectohz(50 * PX_MSEC_TO_USEC));
1928 		mutex_held = 0;
1929 		if (!(px_p->px_pm_flags & PX_PMETOACK_RECVD)) {
1930 			ret = DDI_FAILURE;
1931 			DBG(DBG_PWR, px_p->px_dip, " Timed out while waiting"
1932 			    " for PME_TO_ACK\n");
1933 		}
1934 	}
1935 	px_p->px_pm_flags &=
1936 	    ~(PX_PME_TURNOFF_PENDING | PX_PMETOACK_RECVD | PX_LDN_EXPECTED);
1937 
1938 l23ready_done:
1939 	if (mutex_held)
1940 		mutex_exit(&px_p->px_l23ready_lock);
1941 	/*
1942 	 * Wait till link is in L1 idle, if sending PME_Turn_Off
1943 	 * was succesful.
1944 	 */
1945 	if (ret == DDI_SUCCESS) {
1946 		if (px_link_wait4l1idle(csr_base) != DDI_SUCCESS) {
1947 			DBG(DBG_PWR, px_p->px_dip, " Link is not at L1"
1948 			    " even though we received PME_To_ACK.\n");
1949 			/*
1950 			 * Workaround for hardware bug with P25.
1951 			 * Due to a hardware bug with P25, link state
1952 			 * will be Detect state rather than L1 after
1953 			 * link is transitioned to L23Ready state. Since
1954 			 * we don't know whether link is L23ready state
1955 			 * without Fire's state being L1_idle, we delay
1956 			 * here just to make sure that we wait till link
1957 			 * is transitioned to L23Ready state.
1958 			 */
1959 			delay(drv_usectohz(100 * PX_MSEC_TO_USEC));
1960 		}
1961 		pwr_p->pwr_link_lvl = PM_LEVEL_L3;
1962 
1963 	}
1964 	mutex_exit(&pwr_p->pwr_lock);
1965 	return (ret);
1966 }
1967 
1968 /*
1969  * Message interrupt handler intended to be shared for both
1970  * PME and PME_TO_ACK msg handling, currently only handles
1971  * PME_To_ACK message.
1972  */
1973 uint_t
1974 px_pmeq_intr(caddr_t arg)
1975 {
1976 	px_t	*px_p = (px_t *)arg;
1977 
1978 	DBG(DBG_PWR, px_p->px_dip, " PME_To_ACK received \n");
1979 	mutex_enter(&px_p->px_l23ready_lock);
1980 	cv_broadcast(&px_p->px_l23ready_cv);
1981 	if (px_p->px_pm_flags & PX_PME_TURNOFF_PENDING) {
1982 		px_p->px_pm_flags |= PX_PMETOACK_RECVD;
1983 	} else {
1984 		/*
1985 		 * This maybe the second ack received. If so then,
1986 		 * we should be receiving it during wait4L1 stage.
1987 		 */
1988 		px_p->px_pmetoack_ignored++;
1989 	}
1990 	mutex_exit(&px_p->px_l23ready_lock);
1991 	return (DDI_INTR_CLAIMED);
1992 }
1993 
1994 static int
1995 px_pre_pwron_check(px_t *px_p)
1996 {
1997 	pcie_pwr_t	*pwr_p;
1998 
1999 	/* If no PM info, return failure */
2000 	if (PCIE_PMINFO(px_p->px_dip) == NULL ||
2001 	    (pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)) == NULL)
2002 		return (DDI_FAILURE);
2003 
2004 	/*
2005 	 * For the spec compliant downstream cards link down
2006 	 * is expected when the device is powered on.
2007 	 */
2008 	px_p->px_pm_flags |= PX_LDN_EXPECTED;
2009 	return (pwr_p->pwr_link_lvl == PM_LEVEL_L3 ? DDI_SUCCESS : DDI_FAILURE);
2010 }
2011 
2012 static int
2013 px_goto_l0(px_t *px_p)
2014 {
2015 	pcie_pwr_t	*pwr_p;
2016 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2017 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2018 	int		ret = DDI_SUCCESS;
2019 	uint64_t	time_spent = 0;
2020 
2021 	/* If no PM info, return failure */
2022 	if (PCIE_PMINFO(px_p->px_dip) == NULL ||
2023 	    (pwr_p = PCIE_NEXUS_PMINFO(px_p->px_dip)) == NULL)
2024 		return (DDI_FAILURE);
2025 
2026 	mutex_enter(&pwr_p->pwr_lock);
2027 	/*
2028 	 * The following link retrain activity will cause LDN and LUP event.
2029 	 * Receiving LDN prior to receiving LUP is expected, not an error in
2030 	 * this case.  Receiving LUP indicates link is fully up to support
2031 	 * powering up down stream device, and of course any further LDN and
2032 	 * LUP outside this context will be error.
2033 	 */
2034 	px_p->px_lup_pending = 1;
2035 	if (px_link_retrain(csr_base) != DDI_SUCCESS) {
2036 		ret = DDI_FAILURE;
2037 		goto l0_done;
2038 	}
2039 
2040 	/* LUP event takes the order of 15ms amount of time to occur */
2041 	for (; px_p->px_lup_pending && (time_spent < px_lup_poll_to);
2042 	    time_spent += px_lup_poll_interval)
2043 		drv_usecwait(px_lup_poll_interval);
2044 	if (px_p->px_lup_pending)
2045 		ret = DDI_FAILURE;
2046 l0_done:
2047 	px_enable_detect_quiet(csr_base);
2048 	if (ret == DDI_SUCCESS)
2049 		pwr_p->pwr_link_lvl = PM_LEVEL_L0;
2050 	mutex_exit(&pwr_p->pwr_lock);
2051 	return (ret);
2052 }
2053 
2054 /*
2055  * Extract the drivers binding name to identify which chip we're binding to.
2056  * Whenever a new bus bridge is created, the driver alias entry should be
2057  * added here to identify the device if needed.  If a device isn't added,
2058  * the identity defaults to PX_CHIP_UNIDENTIFIED.
2059  */
2060 static uint32_t
2061 px_identity_init(px_t *px_p)
2062 {
2063 	dev_info_t	*dip = px_p->px_dip;
2064 	char		*name = ddi_binding_name(dip);
2065 	uint32_t	revision = 0;
2066 
2067 	revision = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2068 	    "module-revision#", 0);
2069 
2070 	/* Check for Fire driver binding name */
2071 	if (strcmp(name, "pciex108e,80f0") == 0) {
2072 		DBG(DBG_ATTACH, dip, "px_identity_init: %s%d: "
2073 		    "(FIRE), module-revision %d\n", NAMEINST(dip),
2074 		    revision);
2075 
2076 		return ((revision >= FIRE_MOD_REV_20) ?
2077 		    PX_CHIP_FIRE : PX_CHIP_UNIDENTIFIED);
2078 	}
2079 
2080 	/* Check for Oberon driver binding name */
2081 	if (strcmp(name, "pciex108e,80f8") == 0) {
2082 		DBG(DBG_ATTACH, dip, "px_identity_init: %s%d: "
2083 		    "(OBERON), module-revision %d\n", NAMEINST(dip),
2084 		    revision);
2085 
2086 		return (PX_CHIP_OBERON);
2087 	}
2088 
2089 	DBG(DBG_ATTACH, dip, "%s%d: Unknown PCI Express Host bridge %s %x\n",
2090 	    ddi_driver_name(dip), ddi_get_instance(dip), name, revision);
2091 
2092 	return (PX_CHIP_UNIDENTIFIED);
2093 }
2094 
2095 int
2096 px_err_add_intr(px_fault_t *px_fault_p)
2097 {
2098 	dev_info_t	*dip = px_fault_p->px_fh_dip;
2099 	px_t		*px_p = DIP_TO_STATE(dip);
2100 
2101 	VERIFY(add_ivintr(px_fault_p->px_fh_sysino, PX_ERR_PIL,
2102 	    (intrfunc)px_fault_p->px_err_func, (caddr_t)px_fault_p,
2103 	    NULL, NULL) == 0);
2104 
2105 	px_ib_intr_enable(px_p, intr_dist_cpuid(), px_fault_p->px_intr_ino);
2106 
2107 	return (DDI_SUCCESS);
2108 }
2109 
2110 void
2111 px_err_rem_intr(px_fault_t *px_fault_p)
2112 {
2113 	dev_info_t	*dip = px_fault_p->px_fh_dip;
2114 	px_t		*px_p = DIP_TO_STATE(dip);
2115 
2116 	px_ib_intr_disable(px_p->px_ib_p, px_fault_p->px_intr_ino,
2117 	    IB_INTR_WAIT);
2118 
2119 	VERIFY(rem_ivintr(px_fault_p->px_fh_sysino, PX_ERR_PIL) == 0);
2120 }
2121 
2122 /*
2123  * px_cb_intr_redist() - sun4u only, CB interrupt redistribution
2124  */
2125 void
2126 px_cb_intr_redist(void *arg)
2127 {
2128 	px_cb_t		*cb_p = (px_cb_t *)arg;
2129 	px_cb_list_t	*pxl;
2130 	px_t		*pxp = NULL;
2131 	px_fault_t	*f_p = NULL;
2132 	uint32_t	new_cpuid;
2133 	intr_valid_state_t	enabled = 0;
2134 
2135 	mutex_enter(&cb_p->cb_mutex);
2136 
2137 	pxl = cb_p->pxl;
2138 	if (pxl == NULL)
2139 		goto cb_done;
2140 
2141 	pxp = pxl->pxp;
2142 	f_p = &pxp->px_cb_fault;
2143 	for (; pxl && (f_p->px_fh_sysino != cb_p->sysino); ) {
2144 		pxl = pxl->next;
2145 		pxp = pxl->pxp;
2146 		f_p = &pxp->px_cb_fault;
2147 	}
2148 	if (pxl == NULL)
2149 		goto cb_done;
2150 
2151 	new_cpuid =  intr_dist_cpuid();
2152 	if (new_cpuid == cb_p->cpuid)
2153 		goto cb_done;
2154 
2155 	if ((px_lib_intr_getvalid(pxp->px_dip, f_p->px_fh_sysino, &enabled)
2156 	    != DDI_SUCCESS) || !enabled) {
2157 		DBG(DBG_IB, pxp->px_dip, "px_cb_intr_redist: CB not enabled, "
2158 		    "sysino(0x%x)\n", f_p->px_fh_sysino);
2159 		goto cb_done;
2160 	}
2161 
2162 	PX_INTR_DISABLE(pxp->px_dip, f_p->px_fh_sysino);
2163 
2164 	cb_p->cpuid = new_cpuid;
2165 	cb_p->sysino = f_p->px_fh_sysino;
2166 	PX_INTR_ENABLE(pxp->px_dip, cb_p->sysino, cb_p->cpuid);
2167 
2168 cb_done:
2169 	mutex_exit(&cb_p->cb_mutex);
2170 }
2171 
2172 /*
2173  * px_cb_add_intr() - Called from attach(9E) to create CB if not yet
2174  * created, to add CB interrupt vector always, but enable only once.
2175  */
2176 int
2177 px_cb_add_intr(px_fault_t *fault_p)
2178 {
2179 	px_t		*px_p = DIP_TO_STATE(fault_p->px_fh_dip);
2180 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2181 	px_cb_t		*cb_p = (px_cb_t *)px_get_cb(fault_p->px_fh_dip);
2182 	px_cb_list_t	*pxl, *pxl_new;
2183 	boolean_t	is_proxy = B_FALSE;
2184 
2185 	/* create cb */
2186 	if (cb_p == NULL) {
2187 		cb_p = kmem_zalloc(sizeof (px_cb_t), KM_SLEEP);
2188 
2189 		mutex_init(&cb_p->cb_mutex, NULL, MUTEX_DRIVER,
2190 		    (void *) ipltospl(FM_ERR_PIL));
2191 
2192 		cb_p->px_cb_func = px_cb_intr;
2193 		pxu_p->px_cb_p = cb_p;
2194 		px_set_cb(fault_p->px_fh_dip, (uint64_t)cb_p);
2195 
2196 		/* px_lib_dev_init allows only FIRE and OBERON */
2197 		px_err_reg_enable(
2198 		    (pxu_p->chip_type == PX_CHIP_FIRE) ?
2199 		    PX_ERR_JBC : PX_ERR_UBC,
2200 		    pxu_p->px_address[PX_REG_XBC]);
2201 	} else
2202 		pxu_p->px_cb_p = cb_p;
2203 
2204 	/* register cb interrupt */
2205 	VERIFY(add_ivintr(fault_p->px_fh_sysino, PX_ERR_PIL,
2206 	    (intrfunc)cb_p->px_cb_func, (caddr_t)cb_p, NULL, NULL) == 0);
2207 
2208 
2209 	/* update cb list */
2210 	mutex_enter(&cb_p->cb_mutex);
2211 	if (cb_p->pxl == NULL) {
2212 		is_proxy = B_TRUE;
2213 		pxl = kmem_zalloc(sizeof (px_cb_list_t), KM_SLEEP);
2214 		pxl->pxp = px_p;
2215 		cb_p->pxl = pxl;
2216 		cb_p->sysino = fault_p->px_fh_sysino;
2217 		cb_p->cpuid = intr_dist_cpuid();
2218 	} else {
2219 		/*
2220 		 * Find the last pxl or
2221 		 * stop short at encountering a redundent entry, or
2222 		 * both.
2223 		 */
2224 		pxl = cb_p->pxl;
2225 		for (; !(pxl->pxp == px_p) && pxl->next; pxl = pxl->next) {};
2226 		ASSERT(pxl->pxp != px_p);
2227 
2228 		/* add to linked list */
2229 		pxl_new = kmem_zalloc(sizeof (px_cb_list_t), KM_SLEEP);
2230 		pxl_new->pxp = px_p;
2231 		pxl->next = pxl_new;
2232 	}
2233 	cb_p->attachcnt++;
2234 	mutex_exit(&cb_p->cb_mutex);
2235 
2236 	if (is_proxy) {
2237 		/* add to interrupt redistribution list */
2238 		intr_dist_add(px_cb_intr_redist, cb_p);
2239 
2240 		/* enable cb hw interrupt */
2241 		px_ib_intr_enable(px_p, cb_p->cpuid, fault_p->px_intr_ino);
2242 	}
2243 
2244 	return (DDI_SUCCESS);
2245 }
2246 
2247 /*
2248  * px_cb_rem_intr() - Called from detach(9E) to remove its CB
2249  * interrupt vector, to shift proxy to the next available px,
2250  * or disable CB interrupt when itself is the last.
2251  */
2252 void
2253 px_cb_rem_intr(px_fault_t *fault_p)
2254 {
2255 	px_t		*px_p = DIP_TO_STATE(fault_p->px_fh_dip), *pxp;
2256 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2257 	px_cb_t		*cb_p = PX2CB(px_p);
2258 	px_cb_list_t	*pxl, *prev;
2259 	px_fault_t	*f_p;
2260 
2261 	ASSERT(cb_p->pxl);
2262 
2263 	/* find and remove this px, and update cb list */
2264 	mutex_enter(&cb_p->cb_mutex);
2265 
2266 	pxl = cb_p->pxl;
2267 	if (pxl->pxp == px_p) {
2268 		cb_p->pxl = pxl->next;
2269 	} else {
2270 		prev = pxl;
2271 		pxl = pxl->next;
2272 		for (; pxl && (pxl->pxp != px_p); prev = pxl, pxl = pxl->next) {
2273 		};
2274 		if (pxl == NULL) {
2275 			cmn_err(CE_WARN, "px_cb_rem_intr: can't find px_p 0x%p "
2276 			    "in registered CB list.", (void *)px_p);
2277 			mutex_exit(&cb_p->cb_mutex);
2278 			return;
2279 		}
2280 		prev->next = pxl->next;
2281 	}
2282 	pxu_p->px_cb_p = NULL;
2283 	cb_p->attachcnt--;
2284 	kmem_free(pxl, sizeof (px_cb_list_t));
2285 	mutex_exit(&cb_p->cb_mutex);
2286 
2287 	/* disable cb hw interrupt */
2288 	if (fault_p->px_fh_sysino == cb_p->sysino)
2289 		px_ib_intr_disable(px_p->px_ib_p, fault_p->px_intr_ino,
2290 		    IB_INTR_WAIT);
2291 
2292 	/* if last px, remove from interrupt redistribution list */
2293 	if (cb_p->pxl == NULL)
2294 		intr_dist_rem(px_cb_intr_redist, cb_p);
2295 
2296 	/* de-register interrupt */
2297 	VERIFY(rem_ivintr(fault_p->px_fh_sysino, PX_ERR_PIL) == 0);
2298 
2299 	/* if not last px, assign next px to manage cb */
2300 	mutex_enter(&cb_p->cb_mutex);
2301 	if (cb_p->pxl) {
2302 		if (fault_p->px_fh_sysino == cb_p->sysino) {
2303 			pxp = cb_p->pxl->pxp;
2304 			f_p = &pxp->px_cb_fault;
2305 			cb_p->sysino = f_p->px_fh_sysino;
2306 
2307 			PX_INTR_ENABLE(pxp->px_dip, cb_p->sysino, cb_p->cpuid);
2308 			(void) px_lib_intr_setstate(pxp->px_dip, cb_p->sysino,
2309 			    INTR_IDLE_STATE);
2310 		}
2311 		mutex_exit(&cb_p->cb_mutex);
2312 		return;
2313 	}
2314 
2315 	/* clean up after the last px */
2316 	mutex_exit(&cb_p->cb_mutex);
2317 
2318 	/* px_lib_dev_init allows only FIRE and OBERON */
2319 	px_err_reg_disable(
2320 	    (pxu_p->chip_type == PX_CHIP_FIRE) ? PX_ERR_JBC : PX_ERR_UBC,
2321 	    pxu_p->px_address[PX_REG_XBC]);
2322 
2323 	mutex_destroy(&cb_p->cb_mutex);
2324 	px_set_cb(fault_p->px_fh_dip, 0ull);
2325 	kmem_free(cb_p, sizeof (px_cb_t));
2326 }
2327 
2328 /*
2329  * px_cb_intr() - sun4u only,  CB interrupt dispatcher
2330  */
2331 uint_t
2332 px_cb_intr(caddr_t arg)
2333 {
2334 	px_cb_t		*cb_p = (px_cb_t *)arg;
2335 	px_t		*pxp;
2336 	px_fault_t	*f_p;
2337 	int		ret;
2338 
2339 	mutex_enter(&cb_p->cb_mutex);
2340 
2341 	if (cb_p->pxl == NULL) {
2342 		mutex_exit(&cb_p->cb_mutex);
2343 		return (DDI_INTR_UNCLAIMED);
2344 	}
2345 
2346 	pxp = cb_p->pxl->pxp;
2347 	f_p = &pxp->px_cb_fault;
2348 
2349 	ret = f_p->px_err_func((caddr_t)f_p);
2350 
2351 	mutex_exit(&cb_p->cb_mutex);
2352 	return (ret);
2353 }
2354 
2355 #ifdef	FMA
2356 void
2357 px_fill_rc_status(px_fault_t *px_fault_p, pciex_rc_error_regs_t *rc_status)
2358 {
2359 	/* populate the rc_status by reading the registers - TBD */
2360 }
2361 #endif /* FMA */
2362 
2363 /*
2364  * cpr callback
2365  *
2366  * disable fabric error msg interrupt prior to suspending
2367  * all device drivers; re-enable fabric error msg interrupt
2368  * after all devices are resumed.
2369  */
2370 static boolean_t
2371 px_cpr_callb(void *arg, int code)
2372 {
2373 	px_t		*px_p = (px_t *)arg;
2374 	px_ib_t		*ib_p = px_p->px_ib_p;
2375 	px_pec_t	*pec_p = px_p->px_pec_p;
2376 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2377 	caddr_t		csr_base;
2378 	devino_t	ce_ino, nf_ino, f_ino;
2379 	px_ino_t	*ce_ino_p, *nf_ino_p, *f_ino_p;
2380 	uint64_t	imu_log_enable, imu_intr_enable;
2381 	uint64_t	imu_log_mask, imu_intr_mask;
2382 
2383 	ce_ino = px_msiqid_to_devino(px_p, pec_p->pec_corr_msg_msiq_id);
2384 	nf_ino = px_msiqid_to_devino(px_p, pec_p->pec_non_fatal_msg_msiq_id);
2385 	f_ino = px_msiqid_to_devino(px_p, pec_p->pec_fatal_msg_msiq_id);
2386 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2387 
2388 	imu_log_enable = CSR_XR(csr_base, IMU_ERROR_LOG_ENABLE);
2389 	imu_intr_enable = CSR_XR(csr_base, IMU_INTERRUPT_ENABLE);
2390 
2391 	imu_log_mask = BITMASK(IMU_ERROR_LOG_ENABLE_FATAL_MES_NOT_EN_LOG_EN) |
2392 	    BITMASK(IMU_ERROR_LOG_ENABLE_NONFATAL_MES_NOT_EN_LOG_EN) |
2393 	    BITMASK(IMU_ERROR_LOG_ENABLE_COR_MES_NOT_EN_LOG_EN);
2394 
2395 	imu_intr_mask =
2396 	    BITMASK(IMU_INTERRUPT_ENABLE_FATAL_MES_NOT_EN_S_INT_EN) |
2397 	    BITMASK(IMU_INTERRUPT_ENABLE_NONFATAL_MES_NOT_EN_S_INT_EN) |
2398 	    BITMASK(IMU_INTERRUPT_ENABLE_COR_MES_NOT_EN_S_INT_EN) |
2399 	    BITMASK(IMU_INTERRUPT_ENABLE_FATAL_MES_NOT_EN_P_INT_EN) |
2400 	    BITMASK(IMU_INTERRUPT_ENABLE_NONFATAL_MES_NOT_EN_P_INT_EN) |
2401 	    BITMASK(IMU_INTERRUPT_ENABLE_COR_MES_NOT_EN_P_INT_EN);
2402 
2403 	switch (code) {
2404 	case CB_CODE_CPR_CHKPT:
2405 		/* disable imu rbne on corr/nonfatal/fatal errors */
2406 		CSR_XS(csr_base, IMU_ERROR_LOG_ENABLE,
2407 		    imu_log_enable & (~imu_log_mask));
2408 
2409 		CSR_XS(csr_base, IMU_INTERRUPT_ENABLE,
2410 		    imu_intr_enable & (~imu_intr_mask));
2411 
2412 		/* disable CORR intr mapping */
2413 		px_ib_intr_disable(ib_p, ce_ino, IB_INTR_NOWAIT);
2414 
2415 		/* disable NON FATAL intr mapping */
2416 		px_ib_intr_disable(ib_p, nf_ino, IB_INTR_NOWAIT);
2417 
2418 		/* disable FATAL intr mapping */
2419 		px_ib_intr_disable(ib_p, f_ino, IB_INTR_NOWAIT);
2420 
2421 		break;
2422 
2423 	case CB_CODE_CPR_RESUME:
2424 		pxu_p->cpr_flag = PX_NOT_CPR;
2425 		mutex_enter(&ib_p->ib_ino_lst_mutex);
2426 
2427 		ce_ino_p = px_ib_locate_ino(ib_p, ce_ino);
2428 		nf_ino_p = px_ib_locate_ino(ib_p, nf_ino);
2429 		f_ino_p = px_ib_locate_ino(ib_p, f_ino);
2430 
2431 		/* enable CORR intr mapping */
2432 		if (ce_ino_p)
2433 			px_ib_intr_enable(px_p, ce_ino_p->ino_cpuid, ce_ino);
2434 		else
2435 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2436 			    "reenable PCIe Correctable msg intr.\n");
2437 
2438 		/* enable NON FATAL intr mapping */
2439 		if (nf_ino_p)
2440 			px_ib_intr_enable(px_p, nf_ino_p->ino_cpuid, nf_ino);
2441 		else
2442 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2443 			    "reenable PCIe Non Fatal msg intr.\n");
2444 
2445 		/* enable FATAL intr mapping */
2446 		if (f_ino_p)
2447 			px_ib_intr_enable(px_p, f_ino_p->ino_cpuid, f_ino);
2448 		else
2449 			cmn_err(CE_WARN, "px_cpr_callb: RESUME unable to "
2450 			    "reenable PCIe Fatal msg intr.\n");
2451 
2452 		mutex_exit(&ib_p->ib_ino_lst_mutex);
2453 
2454 		/* enable corr/nonfatal/fatal not enable error */
2455 		CSR_XS(csr_base, IMU_ERROR_LOG_ENABLE, (imu_log_enable |
2456 		    (imu_log_mask & px_imu_log_mask)));
2457 		CSR_XS(csr_base, IMU_INTERRUPT_ENABLE, (imu_intr_enable |
2458 		    (imu_intr_mask & px_imu_intr_mask)));
2459 
2460 		break;
2461 	}
2462 
2463 	return (B_TRUE);
2464 }
2465 
2466 uint64_t
2467 px_get_rng_parent_hi_mask(px_t *px_p)
2468 {
2469 	pxu_t *pxu_p = (pxu_t *)px_p->px_plat_p;
2470 	uint64_t mask;
2471 
2472 	switch (PX_CHIP_TYPE(pxu_p)) {
2473 	case PX_CHIP_OBERON:
2474 		mask = OBERON_RANGE_PROP_MASK;
2475 		break;
2476 	case PX_CHIP_FIRE:
2477 		mask = PX_RANGE_PROP_MASK;
2478 		break;
2479 	default:
2480 		mask = PX_RANGE_PROP_MASK;
2481 	}
2482 
2483 	return (mask);
2484 }
2485 
2486 /*
2487  * fetch chip's range propery's value
2488  */
2489 uint64_t
2490 px_get_range_prop(px_t *px_p, pci_ranges_t *rp, int bank)
2491 {
2492 	uint64_t mask, range_prop;
2493 
2494 	mask = px_get_rng_parent_hi_mask(px_p);
2495 	range_prop = (((uint64_t)(rp[bank].parent_high & mask)) << 32) |
2496 	    rp[bank].parent_low;
2497 
2498 	return (range_prop);
2499 }
2500 
2501 /*
2502  * fetch the config space base addr of the root complex
2503  * note this depends on px structure being initialized
2504  */
2505 uint64_t
2506 px_lib_get_cfgacc_base(dev_info_t *dip)
2507 {
2508 	int		instance = DIP_TO_INST(dip);
2509 	px_t		*px_p = INST_TO_STATE(instance);
2510 	pci_ranges_t	*rp = px_p->px_ranges_p;
2511 	int		bank = PCI_REG_ADDR_G(PCI_ADDR_CONFIG);
2512 
2513 	/* Get Fire's Physical Base Address */
2514 	return (px_get_range_prop(px_p, rp, bank));
2515 }
2516 
2517 /*
2518  * add cpr callback
2519  */
2520 void
2521 px_cpr_add_callb(px_t *px_p)
2522 {
2523 	px_p->px_cprcb_id = callb_add(px_cpr_callb, (void *)px_p,
2524 	    CB_CL_CPR_POST_USER, "px_cpr");
2525 }
2526 
2527 /*
2528  * remove cpr callback
2529  */
2530 void
2531 px_cpr_rem_callb(px_t *px_p)
2532 {
2533 	(void) callb_delete(px_p->px_cprcb_id);
2534 }
2535 
2536 /*ARGSUSED*/
2537 static uint_t
2538 px_hp_intr(caddr_t arg1, caddr_t arg2)
2539 {
2540 	px_t		*px_p = (px_t *)arg1;
2541 	pxu_t		*pxu_p = (pxu_t *)px_p->px_plat_p;
2542 	int		rval;
2543 
2544 	rval = pcie_intr(px_p->px_dip);
2545 
2546 #ifdef  DEBUG
2547 	if (rval == DDI_INTR_UNCLAIMED)
2548 		cmn_err(CE_WARN, "%s%d: UNCLAIMED intr\n",
2549 		    ddi_driver_name(px_p->px_dip),
2550 		    ddi_get_instance(px_p->px_dip));
2551 #endif
2552 
2553 	/* Set the interrupt state to idle */
2554 	if (px_lib_intr_setstate(px_p->px_dip,
2555 	    pxu_p->hp_sysino, INTR_IDLE_STATE) != DDI_SUCCESS)
2556 		return (DDI_INTR_UNCLAIMED);
2557 
2558 	return (rval);
2559 }
2560 
2561 int
2562 px_lib_hotplug_init(dev_info_t *dip, void *arg)
2563 {
2564 	px_t	*px_p = DIP_TO_STATE(dip);
2565 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2566 	uint64_t ret;
2567 
2568 	if (ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2569 	    "hotplug-capable") == 0)
2570 		return (DDI_FAILURE);
2571 
2572 	if ((ret = hvio_hotplug_init(dip, arg)) == DDI_SUCCESS) {
2573 		if (px_lib_intr_devino_to_sysino(px_p->px_dip,
2574 		    px_p->px_inos[PX_INTR_HOTPLUG], &pxu_p->hp_sysino) !=
2575 		    DDI_SUCCESS) {
2576 #ifdef	DEBUG
2577 			cmn_err(CE_WARN, "%s%d: devino_to_sysino fails\n",
2578 			    ddi_driver_name(px_p->px_dip),
2579 			    ddi_get_instance(px_p->px_dip));
2580 #endif
2581 			return (DDI_FAILURE);
2582 		}
2583 
2584 		VERIFY(add_ivintr(pxu_p->hp_sysino, PCIE_INTR_PRI,
2585 		    (intrfunc)px_hp_intr, (caddr_t)px_p, NULL, NULL) == 0);
2586 
2587 		px_ib_intr_enable(px_p, intr_dist_cpuid(),
2588 		    px_p->px_inos[PX_INTR_HOTPLUG]);
2589 	}
2590 
2591 	return (ret);
2592 }
2593 
2594 void
2595 px_lib_hotplug_uninit(dev_info_t *dip)
2596 {
2597 	if (hvio_hotplug_uninit(dip) == DDI_SUCCESS) {
2598 		px_t	*px_p = DIP_TO_STATE(dip);
2599 		pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2600 
2601 		px_ib_intr_disable(px_p->px_ib_p,
2602 		    px_p->px_inos[PX_INTR_HOTPLUG], IB_INTR_WAIT);
2603 
2604 		VERIFY(rem_ivintr(pxu_p->hp_sysino, PCIE_INTR_PRI) == 0);
2605 	}
2606 }
2607 
2608 /*
2609  * px_hp_intr_redist() - sun4u only, HP interrupt redistribution
2610  */
2611 void
2612 px_hp_intr_redist(px_t *px_p)
2613 {
2614 	pcie_bus_t	*bus_p = PCIE_DIP2BUS(px_p->px_dip);
2615 
2616 	if (px_p && PCIE_IS_PCIE_HOTPLUG_ENABLED(bus_p)) {
2617 		px_ib_intr_dist_en(px_p->px_dip, intr_dist_cpuid(),
2618 		    px_p->px_inos[PX_INTR_HOTPLUG], B_FALSE);
2619 	}
2620 }
2621 
2622 boolean_t
2623 px_lib_is_in_drain_state(px_t *px_p)
2624 {
2625 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2626 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2627 	uint64_t drain_status;
2628 
2629 	if (PX_CHIP_TYPE(pxu_p) == PX_CHIP_OBERON) {
2630 		drain_status = CSR_BR(csr_base, DRAIN_CONTROL_STATUS, DRAIN);
2631 	} else {
2632 		drain_status = CSR_BR(csr_base, TLU_STATUS, DRAIN);
2633 	}
2634 
2635 	return (drain_status);
2636 }
2637 
2638 pcie_req_id_t
2639 px_lib_get_bdf(px_t *px_p)
2640 {
2641 	pxu_t 	*pxu_p = (pxu_t *)px_p->px_plat_p;
2642 	caddr_t csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2643 	pcie_req_id_t bdf;
2644 
2645 	bdf = CSR_BR(csr_base, DMC_PCI_EXPRESS_CONFIGURATION, REQ_ID);
2646 
2647 	return (bdf);
2648 }
2649 
2650 /*ARGSUSED*/
2651 int
2652 px_lib_get_root_complex_mps(px_t *px_p, dev_info_t *dip, int *mps)
2653 {
2654 	pxu_t	*pxu_p;
2655 	caddr_t csr_base;
2656 
2657 	pxu_p = (pxu_t *)px_p->px_plat_p;
2658 
2659 	if (pxu_p == NULL)
2660 		return (DDI_FAILURE);
2661 
2662 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2663 
2664 
2665 	*mps = CSR_XR(csr_base, TLU_DEVICE_CAPABILITIES) &
2666 	    TLU_DEVICE_CAPABILITIES_MPS_MASK;
2667 
2668 	return (DDI_SUCCESS);
2669 }
2670 
2671 /*ARGSUSED*/
2672 int
2673 px_lib_set_root_complex_mps(px_t *px_p,  dev_info_t *dip, int mps)
2674 {
2675 	pxu_t	*pxu_p;
2676 	caddr_t csr_base;
2677 	uint64_t dev_ctrl;
2678 	int link_width, val;
2679 	px_chip_type_t chip_type = px_identity_init(px_p);
2680 
2681 	pxu_p = (pxu_t *)px_p->px_plat_p;
2682 
2683 	if (pxu_p == NULL)
2684 		return (DDI_FAILURE);
2685 
2686 	csr_base = (caddr_t)pxu_p->px_address[PX_REG_CSR];
2687 
2688 	dev_ctrl = CSR_XR(csr_base, TLU_DEVICE_CONTROL);
2689 	dev_ctrl |= (mps << TLU_DEVICE_CONTROL_MPS);
2690 
2691 	CSR_XS(csr_base, TLU_DEVICE_CONTROL, dev_ctrl);
2692 
2693 	link_width = CSR_FR(csr_base, TLU_LINK_STATUS, WIDTH);
2694 
2695 	/*
2696 	 * Convert link_width to match timer array configuration.
2697 	 */
2698 	switch (link_width) {
2699 	case 1:
2700 		link_width = 0;
2701 		break;
2702 	case 4:
2703 		link_width = 1;
2704 		break;
2705 	case 8:
2706 		link_width = 2;
2707 		break;
2708 	case 16:
2709 		link_width = 3;
2710 		break;
2711 	default:
2712 		link_width = 0;
2713 	}
2714 
2715 	val = px_replay_timer_table[mps][link_width];
2716 	CSR_XS(csr_base, LPU_TXLINK_REPLAY_TIMER_THRESHOLD, val);
2717 
2718 	if (chip_type == PX_CHIP_OBERON)
2719 		return (DDI_SUCCESS);
2720 
2721 	val = px_acknak_timer_table[mps][link_width];
2722 	CSR_XS(csr_base, LPU_TXLINK_FREQUENT_NAK_LATENCY_TIMER_THRESHOLD, val);
2723 
2724 	return (DDI_SUCCESS);
2725 }
2726 
2727 /*ARGSUSED*/
2728 int
2729 px_lib_fabric_sync(dev_info_t *dip)
2730 {
2731 	/* an no-op on sun4u platform */
2732 	return (DDI_SUCCESS);
2733 }
2734