xref: /freebsd/sys/powerpc/powernv/opal_pci.c (revision 42249ef2)
1 /*-
2  * Copyright (c) 2015-2016 Nathan Whitehorn
3  * Copyright (c) 2017-2018 Semihalf
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/module.h>
33 #include <sys/bus.h>
34 #include <sys/conf.h>
35 #include <sys/kernel.h>
36 #include <sys/pciio.h>
37 #include <sys/endian.h>
38 #include <sys/rman.h>
39 #include <sys/vmem.h>
40 
41 #include <dev/ofw/openfirm.h>
42 #include <dev/ofw/ofw_pci.h>
43 #include <dev/ofw/ofw_bus.h>
44 #include <dev/ofw/ofw_bus_subr.h>
45 #include <dev/ofw/ofwpci.h>
46 
47 #include <dev/pci/pcivar.h>
48 #include <dev/pci/pcireg.h>
49 
50 #include <machine/bus.h>
51 #include <machine/intr_machdep.h>
52 #include <machine/md_var.h>
53 
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 
57 #include "pcib_if.h"
58 #include "pic_if.h"
59 #include "iommu_if.h"
60 #include "opal.h"
61 
62 #define	OPAL_PCI_TCE_MAX_ENTRIES	(1024*1024UL)
63 #define	OPAL_PCI_TCE_DEFAULT_SEG_SIZE	(16*1024*1024UL)
64 #define	OPAL_PCI_TCE_R			(1UL << 0)
65 #define	OPAL_PCI_TCE_W			(1UL << 1)
66 #define	PHB3_TCE_KILL_INVAL_ALL		(1UL << 63)
67 
68 /*
69  * Device interface.
70  */
71 static int		opalpci_probe(device_t);
72 static int		opalpci_attach(device_t);
73 
74 /*
75  * pcib interface.
76  */
77 static uint32_t		opalpci_read_config(device_t, u_int, u_int, u_int,
78 			    u_int, int);
79 static void		opalpci_write_config(device_t, u_int, u_int, u_int,
80 			    u_int, u_int32_t, int);
81 static int		opalpci_alloc_msi(device_t dev, device_t child,
82 			    int count, int maxcount, int *irqs);
83 static int		opalpci_release_msi(device_t dev, device_t child,
84 			    int count, int *irqs);
85 static int		opalpci_alloc_msix(device_t dev, device_t child,
86 			    int *irq);
87 static int		opalpci_release_msix(device_t dev, device_t child,
88 			    int irq);
89 static int		opalpci_map_msi(device_t dev, device_t child,
90 			    int irq, uint64_t *addr, uint32_t *data);
91 static int opalpci_route_interrupt(device_t bus, device_t dev, int pin);
92 
93 /*
94  * MSI PIC interface.
95  */
96 static void opalpic_pic_enable(device_t dev, u_int irq, u_int vector, void **);
97 static void opalpic_pic_eoi(device_t dev, u_int irq, void *);
98 
99 /* Bus interface */
100 static bus_dma_tag_t opalpci_get_dma_tag(device_t dev, device_t child);
101 
102 /*
103  * Commands
104  */
105 #define	OPAL_M32_WINDOW_TYPE		1
106 #define	OPAL_M64_WINDOW_TYPE		2
107 #define	OPAL_IO_WINDOW_TYPE		3
108 
109 #define	OPAL_RESET_PHB_COMPLETE		1
110 #define	OPAL_RESET_PCI_IODA_TABLE	6
111 
112 #define	OPAL_DISABLE_M64		0
113 #define	OPAL_ENABLE_M64_SPLIT		1
114 #define	OPAL_ENABLE_M64_NON_SPLIT	2
115 
116 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO	1
117 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_DMA	2
118 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_ALL	3
119 
120 #define	OPAL_EEH_STOPPED_NOT_FROZEN		0
121 
122 /*
123  * Constants
124  */
125 #define OPAL_PCI_DEFAULT_PE			1
126 
127 #define OPAL_PCI_BUS_SPACE_LOWADDR_32BIT	0x7FFFFFFFUL
128 
129 /*
130  * Driver methods.
131  */
132 static device_method_t	opalpci_methods[] = {
133 	/* Device interface */
134 	DEVMETHOD(device_probe,		opalpci_probe),
135 	DEVMETHOD(device_attach,	opalpci_attach),
136 
137 	/* pcib interface */
138 	DEVMETHOD(pcib_read_config,	opalpci_read_config),
139 	DEVMETHOD(pcib_write_config,	opalpci_write_config),
140 
141 	DEVMETHOD(pcib_alloc_msi,	opalpci_alloc_msi),
142 	DEVMETHOD(pcib_release_msi,	opalpci_release_msi),
143 	DEVMETHOD(pcib_alloc_msix,	opalpci_alloc_msix),
144 	DEVMETHOD(pcib_release_msix,	opalpci_release_msix),
145 	DEVMETHOD(pcib_map_msi,		opalpci_map_msi),
146 	DEVMETHOD(pcib_route_interrupt,	opalpci_route_interrupt),
147 
148 	/* PIC interface for MSIs */
149 	DEVMETHOD(pic_enable,		opalpic_pic_enable),
150 	DEVMETHOD(pic_eoi,		opalpic_pic_eoi),
151 
152 	/* Bus interface */
153 	DEVMETHOD(bus_get_dma_tag,	opalpci_get_dma_tag),
154 	DEVMETHOD(bus_get_cpus,		ofw_pcibus_get_cpus),
155 	DEVMETHOD(bus_get_domain,	ofw_pcibus_get_domain),
156 
157 	DEVMETHOD_END
158 };
159 
160 struct opalpci_softc {
161 	struct ofw_pci_softc ofw_sc;
162 	uint64_t phb_id;
163 	vmem_t *msi_vmem;
164 	int msi_base;		/* Base XIVE number */
165 	int base_msi_irq;	/* Base IRQ assigned by FreeBSD to this PIC */
166 	uint64_t *tce;		/* TCE table for 1:1 mapping */
167 	struct resource *r_reg;
168 };
169 
170 static devclass_t	opalpci_devclass;
171 DEFINE_CLASS_1(pcib, opalpci_driver, opalpci_methods,
172     sizeof(struct opalpci_softc), ofw_pci_driver);
173 EARLY_DRIVER_MODULE(opalpci, ofwbus, opalpci_driver, opalpci_devclass, 0, 0,
174     BUS_PASS_BUS);
175 
176 static int
177 opalpci_probe(device_t dev)
178 {
179 	const char	*type;
180 
181 	if (opal_check() != 0)
182 		return (ENXIO);
183 
184 	type = ofw_bus_get_type(dev);
185 
186 	if (type == NULL || (strcmp(type, "pci") != 0 &&
187 	    strcmp(type, "pciex") != 0))
188 		return (ENXIO);
189 
190 	if (!OF_hasprop(ofw_bus_get_node(dev), "ibm,opal-phbid"))
191 		return (ENXIO);
192 
193 	device_set_desc(dev, "OPAL Host-PCI bridge");
194 	return (BUS_PROBE_GENERIC);
195 }
196 
197 static void
198 pci_phb3_tce_invalidate_entire(struct opalpci_softc *sc)
199 {
200 
201 	mb();
202 	bus_write_8(sc->r_reg, 0x210, PHB3_TCE_KILL_INVAL_ALL);
203 	mb();
204 }
205 
206 /* Simple function to round to a power of 2 */
207 static uint64_t
208 round_pow2(uint64_t val)
209 {
210 
211 	return (1 << (flsl(val + (val - 1)) - 1));
212 }
213 
214 /*
215  * Starting with skiboot 5.10 PCIe nodes have a new property,
216  * "ibm,supported-tce-sizes", to denote the TCE sizes available.  This allows us
217  * to avoid hard-coding the maximum TCE size allowed, and instead provide a sane
218  * default (however, the "sane" default, which works for all targets, is 64k,
219  * limiting us to 64GB if we have 1M entries.
220  */
221 static uint64_t
222 max_tce_size(device_t dev)
223 {
224 	phandle_t node;
225 	cell_t sizes[64]; /* Property is a list of bit-widths, up to 64-bits */
226 	int count;
227 
228 	node = ofw_bus_get_node(dev);
229 
230 	count = OF_getencprop(node, "ibm,supported-tce-sizes",
231 	    sizes, sizeof(sizes));
232 	if (count < (int) sizeof(cell_t))
233 		return OPAL_PCI_TCE_DEFAULT_SEG_SIZE;
234 
235 	count /= sizeof(cell_t);
236 
237 	return (1ULL << sizes[count - 1]);
238 }
239 
240 static int
241 opalpci_attach(device_t dev)
242 {
243 	struct opalpci_softc *sc;
244 	cell_t id[2], m64ranges[2], m64window[6], npe;
245 	phandle_t node;
246 	int i, err;
247 	uint64_t maxmem;
248 	uint64_t entries;
249 	uint64_t tce_size;
250 	uint64_t tce_tbl_size;
251 	int m64bar;
252 	int rid;
253 
254 	sc = device_get_softc(dev);
255 	node = ofw_bus_get_node(dev);
256 
257 	switch (OF_getproplen(node, "ibm,opal-phbid")) {
258 	case 8:
259 		OF_getencprop(node, "ibm,opal-phbid", id, 8);
260 		sc->phb_id = ((uint64_t)id[0] << 32) | id[1];
261 		break;
262 	case 4:
263 		OF_getencprop(node, "ibm,opal-phbid", id, 4);
264 		sc->phb_id = id[0];
265 		break;
266 	default:
267 		device_printf(dev, "PHB ID property had wrong length (%zd)\n",
268 		    OF_getproplen(node, "ibm,opal-phbid"));
269 		return (ENXIO);
270 	}
271 
272 	if (bootverbose)
273 		device_printf(dev, "OPAL ID %#lx\n", sc->phb_id);
274 
275 	rid = 0;
276 	sc->r_reg = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
277 	    &rid, RF_ACTIVE | RF_SHAREABLE);
278 	if (sc->r_reg == NULL) {
279 		device_printf(dev, "Failed to allocate PHB[%jd] registers\n",
280 		    (uintmax_t)sc->phb_id);
281 		return (ENXIO);
282 	}
283 
284 #if 0
285 	/*
286 	 * Reset PCI IODA table
287 	 */
288 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PCI_IODA_TABLE,
289 	    1);
290 	if (err != 0) {
291 		device_printf(dev, "IODA table reset failed: %d\n", err);
292 		return (ENXIO);
293 	}
294 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PHB_COMPLETE,
295 	    1);
296 	if (err < 0) {
297 		device_printf(dev, "PHB reset failed: %d\n", err);
298 		return (ENXIO);
299 	}
300 	if (err > 0) {
301 		while ((err = opal_call(OPAL_PCI_POLL, sc->phb_id)) > 0) {
302 			DELAY(1000*(err + 1)); /* Returns expected delay in ms */
303 		}
304 	}
305 	if (err < 0) {
306 		device_printf(dev, "WARNING: PHB IODA reset poll failed: %d\n", err);
307 	}
308 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PHB_COMPLETE,
309 	    0);
310 	if (err < 0) {
311 		device_printf(dev, "PHB reset failed: %d\n", err);
312 		return (ENXIO);
313 	}
314 	if (err > 0) {
315 		while ((err = opal_call(OPAL_PCI_POLL, sc->phb_id)) > 0) {
316 			DELAY(1000*(err + 1)); /* Returns expected delay in ms */
317 		}
318 	}
319 #endif
320 
321 	/*
322 	 * Map all devices on the bus to partitionable endpoint one until
323 	 * such time as we start wanting to do things like bhyve.
324 	 */
325 	err = opal_call(OPAL_PCI_SET_PE, sc->phb_id, OPAL_PCI_DEFAULT_PE,
326 	    0, OPAL_PCI_BUS_ANY, OPAL_IGNORE_RID_DEVICE_NUMBER,
327 	    OPAL_IGNORE_RID_FUNC_NUMBER, OPAL_MAP_PE);
328 	if (err != 0) {
329 		device_printf(dev, "PE mapping failed: %d\n", err);
330 		return (ENXIO);
331 	}
332 
333 	/*
334 	 * Turn on MMIO, mapped to PE 1
335 	 */
336 	if (OF_getencprop(node, "ibm,opal-num-pes", &npe, 4) != 4)
337 		npe = 1;
338 	for (i = 0; i < npe; i++) {
339 		err = opal_call(OPAL_PCI_MAP_PE_MMIO_WINDOW, sc->phb_id,
340 		    OPAL_PCI_DEFAULT_PE, OPAL_M32_WINDOW_TYPE, 0, i);
341 		if (err != 0)
342 			device_printf(dev, "MMIO %d map failed: %d\n", i, err);
343 	}
344 
345 	if (OF_getencprop(node, "ibm,opal-available-m64-ranges",
346 	    m64ranges, sizeof(m64ranges)) == sizeof(m64ranges))
347 		m64bar = m64ranges[0];
348 	else
349 	    m64bar = 0;
350 
351 	/* XXX: multiple M64 windows? */
352 	if (OF_getencprop(node, "ibm,opal-m64-window",
353 	    m64window, sizeof(m64window)) == sizeof(m64window)) {
354 		opal_call(OPAL_PCI_PHB_MMIO_ENABLE, sc->phb_id,
355 		    OPAL_M64_WINDOW_TYPE, m64bar, 0);
356 		opal_call(OPAL_PCI_SET_PHB_MEM_WINDOW, sc->phb_id,
357 		    OPAL_M64_WINDOW_TYPE, m64bar /* index */,
358 		    ((uint64_t)m64window[2] << 32) | m64window[3], 0,
359 		    ((uint64_t)m64window[4] << 32) | m64window[5]);
360 		opal_call(OPAL_PCI_MAP_PE_MMIO_WINDOW, sc->phb_id,
361 		    OPAL_PCI_DEFAULT_PE, OPAL_M64_WINDOW_TYPE,
362 		    m64bar /* index */, 0);
363 		opal_call(OPAL_PCI_PHB_MMIO_ENABLE, sc->phb_id,
364 		    OPAL_M64_WINDOW_TYPE, m64bar, OPAL_ENABLE_M64_NON_SPLIT);
365 	}
366 
367 	/*
368 	 * Enable IOMMU for PE1 - map everything 1:1 using
369 	 * segments of max_tce_size size
370 	 */
371 	tce_size = max_tce_size(dev);
372 	maxmem = roundup2(powerpc_ptob(Maxmem), tce_size);
373 	entries = round_pow2(maxmem / tce_size);
374 	tce_tbl_size = MAX(entries * sizeof(uint64_t), 4096);
375 	if (entries > OPAL_PCI_TCE_MAX_ENTRIES)
376 		panic("POWERNV supports only %jdGB of memory space\n",
377 		    (uintmax_t)((OPAL_PCI_TCE_MAX_ENTRIES * tce_size) >> 30));
378 	if (bootverbose)
379 		device_printf(dev, "Mapping 0-%#jx for DMA\n", (uintmax_t)maxmem);
380 	sc->tce = contigmalloc(tce_tbl_size,
381 	    M_DEVBUF, M_NOWAIT | M_ZERO, 0,
382 	    BUS_SPACE_MAXADDR, tce_tbl_size, 0);
383 	if (sc->tce == NULL)
384 		panic("Failed to allocate TCE memory for PHB %jd\n",
385 		    (uintmax_t)sc->phb_id);
386 
387 	for (i = 0; i < entries; i++)
388 		sc->tce[i] = (i * tce_size) | OPAL_PCI_TCE_R | OPAL_PCI_TCE_W;
389 
390 	/* Map TCE for every PE. It seems necessary for Power8 */
391 	for (i = 0; i < npe; i++) {
392 		err = opal_call(OPAL_PCI_MAP_PE_DMA_WINDOW, sc->phb_id,
393 		    i, (i << 1),
394 		    1, pmap_kextract((uint64_t)&sc->tce[0]),
395 		    tce_tbl_size, tce_size);
396 		if (err != 0) {
397 			device_printf(dev, "DMA IOMMU mapping failed: %d\n", err);
398 			return (ENXIO);
399 		}
400 
401 		err = opal_call(OPAL_PCI_MAP_PE_DMA_WINDOW_REAL, sc->phb_id,
402 		    i, (i << 1) + 1,
403 		    (1UL << 59), maxmem);
404 		if (err != 0) {
405 			device_printf(dev, "DMA 64b bypass mapping failed: %d\n", err);
406 			return (ENXIO);
407 		}
408 	}
409 
410 	/*
411 	 * Invalidate all previous TCE entries.
412 	 *
413 	 * TODO: add support for other PHBs than PHB3
414 	 */
415 	pci_phb3_tce_invalidate_entire(sc);
416 
417 	/*
418 	 * Get MSI properties
419 	 */
420 	sc->msi_vmem = NULL;
421 	if (OF_getproplen(node, "ibm,opal-msi-ranges") > 0) {
422 		cell_t msi_ranges[2];
423 		OF_getencprop(node, "ibm,opal-msi-ranges",
424 		    msi_ranges, sizeof(msi_ranges));
425 		sc->msi_base = msi_ranges[0];
426 
427 		sc->msi_vmem = vmem_create("OPAL MSI", msi_ranges[0],
428 		    msi_ranges[1], 1, 16, M_BESTFIT | M_WAITOK);
429 
430 		sc->base_msi_irq = powerpc_register_pic(dev,
431 		    OF_xref_from_node(node),
432 		    msi_ranges[0] + msi_ranges[1], 0, FALSE);
433 
434 		if (bootverbose)
435 			device_printf(dev, "Supports %d MSIs starting at %d\n",
436 			    msi_ranges[1], msi_ranges[0]);
437 	}
438 
439 	/* Create the parent DMA tag */
440 	/*
441 	 * Constrain it to POWER8 PHB (ioda2) for now.  It seems to mess up on
442 	 * POWER9 systems.
443 	 */
444 	if (ofw_bus_is_compatible(dev, "ibm,ioda2-phb")) {
445 		err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
446 		    1, 0,				/* alignment, bounds */
447 		    OPAL_PCI_BUS_SPACE_LOWADDR_32BIT,	/* lowaddr */
448 		    BUS_SPACE_MAXADDR_32BIT,		/* highaddr */
449 		    NULL, NULL,				/* filter, filterarg */
450 		    BUS_SPACE_MAXSIZE,			/* maxsize */
451 		    BUS_SPACE_UNRESTRICTED,		/* nsegments */
452 		    BUS_SPACE_MAXSIZE,			/* maxsegsize */
453 		    0,					/* flags */
454 		    NULL, NULL,				/* lockfunc, lockarg */
455 		    &sc->ofw_sc.sc_dmat);
456 		if (err != 0) {
457 			device_printf(dev, "Failed to create DMA tag\n");
458 			return (err);
459 		}
460 	}
461 
462 	/*
463 	 * General OFW PCI attach
464 	 */
465 	err = ofw_pci_init(dev);
466 	if (err != 0)
467 		return (err);
468 
469 	/*
470 	 * Unfreeze non-config-space PCI operations. Let this fail silently
471 	 * if e.g. there is no current freeze.
472 	 */
473 	opal_call(OPAL_PCI_EEH_FREEZE_CLEAR, sc->phb_id, OPAL_PCI_DEFAULT_PE,
474 	    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
475 
476 	/*
477 	 * OPAL stores 64-bit BARs in a special property rather than "ranges"
478 	 */
479 	if (OF_getencprop(node, "ibm,opal-m64-window",
480 	    m64window, sizeof(m64window)) == sizeof(m64window)) {
481 		struct ofw_pci_range *rp;
482 
483 		sc->ofw_sc.sc_nrange++;
484 		sc->ofw_sc.sc_range = realloc(sc->ofw_sc.sc_range,
485 		    sc->ofw_sc.sc_nrange * sizeof(sc->ofw_sc.sc_range[0]),
486 		    M_DEVBUF, M_WAITOK);
487 		rp = &sc->ofw_sc.sc_range[sc->ofw_sc.sc_nrange-1];
488 		rp->pci_hi = OFW_PCI_PHYS_HI_SPACE_MEM64 |
489 		    OFW_PCI_PHYS_HI_PREFETCHABLE;
490 		rp->pci = ((uint64_t)m64window[0] << 32) | m64window[1];
491 		rp->host = ((uint64_t)m64window[2] << 32) | m64window[3];
492 		rp->size = ((uint64_t)m64window[4] << 32) | m64window[5];
493 		rman_manage_region(&sc->ofw_sc.sc_mem_rman, rp->pci,
494 		   rp->pci + rp->size - 1);
495 	}
496 
497 	return (ofw_pci_attach(dev));
498 }
499 
500 static uint32_t
501 opalpci_read_config(device_t dev, u_int bus, u_int slot, u_int func, u_int reg,
502     int width)
503 {
504 	struct opalpci_softc *sc;
505 	uint64_t config_addr;
506 	uint8_t byte, eeh_state;
507 	uint16_t half;
508 	uint32_t word;
509 	int error;
510 	uint16_t err_type;
511 
512 	sc = device_get_softc(dev);
513 
514 	config_addr = (bus << 8) | ((slot & 0x1f) << 3) | (func & 0x7);
515 
516 	switch (width) {
517 	case 1:
518 		error = opal_call(OPAL_PCI_CONFIG_READ_BYTE, sc->phb_id,
519 		    config_addr, reg, vtophys(&byte));
520 		word = byte;
521 		break;
522 	case 2:
523 		error = opal_call(OPAL_PCI_CONFIG_READ_HALF_WORD, sc->phb_id,
524 		    config_addr, reg, vtophys(&half));
525 		word = half;
526 		break;
527 	case 4:
528 		error = opal_call(OPAL_PCI_CONFIG_READ_WORD, sc->phb_id,
529 		    config_addr, reg, vtophys(&word));
530 		break;
531 	default:
532 		error = OPAL_SUCCESS;
533 		word = 0xffffffff;
534 		width = 4;
535 	}
536 
537 	/*
538 	 * Poking config state for non-existant devices can make
539 	 * the host bridge hang up. Clear any errors.
540 	 */
541 
542 	if (error != OPAL_SUCCESS ||
543 	    (word == ((1UL << (8 * width)) - 1))) {
544 		if (error != OPAL_HARDWARE) {
545 			opal_call(OPAL_PCI_EEH_FREEZE_STATUS, sc->phb_id,
546 			    OPAL_PCI_DEFAULT_PE, vtophys(&eeh_state),
547 			    vtophys(&err_type), NULL);
548 			if (eeh_state != OPAL_EEH_STOPPED_NOT_FROZEN)
549 				opal_call(OPAL_PCI_EEH_FREEZE_CLEAR,
550 				    sc->phb_id, OPAL_PCI_DEFAULT_PE,
551 				    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
552 		}
553 		if (error != OPAL_SUCCESS)
554 			word = 0xffffffff;
555 	}
556 
557 	return (word);
558 }
559 
560 static void
561 opalpci_write_config(device_t dev, u_int bus, u_int slot, u_int func,
562     u_int reg, uint32_t val, int width)
563 {
564 	struct opalpci_softc *sc;
565 	uint64_t config_addr;
566 	int error = OPAL_SUCCESS;
567 
568 	sc = device_get_softc(dev);
569 
570 	config_addr = (bus << 8) | ((slot & 0x1f) << 3) | (func & 0x7);
571 
572 	switch (width) {
573 	case 1:
574 		error = opal_call(OPAL_PCI_CONFIG_WRITE_BYTE, sc->phb_id,
575 		    config_addr, reg, val);
576 		break;
577 	case 2:
578 		error = opal_call(OPAL_PCI_CONFIG_WRITE_HALF_WORD, sc->phb_id,
579 		    config_addr, reg, val);
580 		break;
581 	case 4:
582 		error = opal_call(OPAL_PCI_CONFIG_WRITE_WORD, sc->phb_id,
583 		    config_addr, reg, val);
584 		break;
585 	}
586 
587 	if (error != OPAL_SUCCESS) {
588 		/*
589 		 * Poking config state for non-existant devices can make
590 		 * the host bridge hang up. Clear any errors.
591 		 */
592 		if (error != OPAL_HARDWARE) {
593 			opal_call(OPAL_PCI_EEH_FREEZE_CLEAR,
594 			    sc->phb_id, OPAL_PCI_DEFAULT_PE,
595 			    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
596 		}
597 	}
598 }
599 
600 static int
601 opalpci_route_interrupt(device_t bus, device_t dev, int pin)
602 {
603 
604 	return (pin);
605 }
606 
607 static int
608 opalpci_alloc_msi(device_t dev, device_t child, int count, int maxcount,
609     int *irqs)
610 {
611 	struct opalpci_softc *sc;
612 	vmem_addr_t start;
613 	phandle_t xref;
614 	int err, i;
615 
616 	sc = device_get_softc(dev);
617 	if (sc->msi_vmem == NULL)
618 		return (ENODEV);
619 
620 	err = vmem_xalloc(sc->msi_vmem, count, powerof2(count), 0, 0,
621 	    VMEM_ADDR_MIN, VMEM_ADDR_MAX, M_BESTFIT | M_WAITOK, &start);
622 
623 	if (err)
624 		return (err);
625 
626 	xref = OF_xref_from_node(ofw_bus_get_node(dev));
627 	for (i = 0; i < count; i++)
628 		irqs[i] = MAP_IRQ(xref, start + i);
629 
630 	return (0);
631 }
632 
633 static int
634 opalpci_release_msi(device_t dev, device_t child, int count, int *irqs)
635 {
636 	struct opalpci_softc *sc;
637 
638 	sc = device_get_softc(dev);
639 	if (sc->msi_vmem == NULL)
640 		return (ENODEV);
641 
642 	vmem_xfree(sc->msi_vmem, irqs[0] - sc->base_msi_irq, count);
643 	return (0);
644 }
645 
646 static int
647 opalpci_alloc_msix(device_t dev, device_t child, int *irq)
648 {
649 	return (opalpci_alloc_msi(dev, child, 1, 1, irq));
650 }
651 
652 static int
653 opalpci_release_msix(device_t dev, device_t child, int irq)
654 {
655 	return (opalpci_release_msi(dev, child, 1, &irq));
656 }
657 
658 static int
659 opalpci_map_msi(device_t dev, device_t child, int irq, uint64_t *addr,
660     uint32_t *data)
661 {
662 	struct opalpci_softc *sc;
663 	struct pci_devinfo *dinfo;
664 	int err, xive;
665 
666 	sc = device_get_softc(dev);
667 	if (sc->msi_vmem == NULL)
668 		return (ENODEV);
669 
670 	xive = irq - sc->base_msi_irq - sc->msi_base;
671 	opal_call(OPAL_PCI_SET_XIVE_PE, sc->phb_id, OPAL_PCI_DEFAULT_PE, xive);
672 
673 	dinfo = device_get_ivars(child);
674 	if (dinfo->cfg.msi.msi_alloc > 0 &&
675 	    (dinfo->cfg.msi.msi_ctrl & PCIM_MSICTRL_64BIT) == 0) {
676 		uint32_t msi32;
677 		err = opal_call(OPAL_GET_MSI_32, sc->phb_id,
678 		    OPAL_PCI_DEFAULT_PE, xive, 1, vtophys(&msi32),
679 		    vtophys(data));
680 		*addr = be32toh(msi32);
681 	} else {
682 		err = opal_call(OPAL_GET_MSI_64, sc->phb_id,
683 		    OPAL_PCI_DEFAULT_PE, xive, 1, vtophys(addr), vtophys(data));
684 		*addr = be64toh(*addr);
685 	}
686 	*data = be32toh(*data);
687 
688 	if (bootverbose && err != 0)
689 		device_printf(child, "OPAL MSI mapping error: %d\n", err);
690 
691 	return ((err == 0) ? 0 : ENXIO);
692 }
693 
694 static void
695 opalpic_pic_enable(device_t dev, u_int irq, u_int vector, void **priv)
696 {
697 	struct opalpci_softc *sc = device_get_softc(dev);
698 
699 	PIC_ENABLE(root_pic, irq, vector, priv);
700 	opal_call(OPAL_PCI_MSI_EOI, sc->phb_id, irq, priv);
701 }
702 
703 static void opalpic_pic_eoi(device_t dev, u_int irq, void *priv)
704 {
705 	struct opalpci_softc *sc;
706 
707 	sc = device_get_softc(dev);
708 	opal_call(OPAL_PCI_MSI_EOI, sc->phb_id, irq);
709 
710 	PIC_EOI(root_pic, irq, priv);
711 }
712 
713 static bus_dma_tag_t
714 opalpci_get_dma_tag(device_t dev, device_t child)
715 {
716 	struct opalpci_softc *sc;
717 
718 	sc = device_get_softc(dev);
719 	return (sc->ofw_sc.sc_dmat);
720 }
721