xref: /freebsd/sys/powerpc/powernv/opal_pci.c (revision d0b2dbfa)
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 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/module.h>
32 #include <sys/bus.h>
33 #include <sys/conf.h>
34 #include <sys/kernel.h>
35 #include <sys/pciio.h>
36 #include <sys/endian.h>
37 #include <sys/rman.h>
38 #include <sys/vmem.h>
39 
40 #include <dev/ofw/openfirm.h>
41 #include <dev/ofw/ofw_pci.h>
42 #include <dev/ofw/ofw_bus.h>
43 #include <dev/ofw/ofw_bus_subr.h>
44 #include <dev/ofw/ofwpci.h>
45 
46 #include <dev/pci/pcivar.h>
47 #include <dev/pci/pcireg.h>
48 
49 #include <machine/bus.h>
50 #include <machine/intr_machdep.h>
51 #include <machine/md_var.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 
56 #include "pcib_if.h"
57 #include "pic_if.h"
58 #include "iommu_if.h"
59 #include "opal.h"
60 
61 #define	OPAL_PCI_TCE_MAX_ENTRIES	(1024*1024UL)
62 #define	OPAL_PCI_TCE_DEFAULT_SEG_SIZE	(16*1024*1024UL)
63 #define	OPAL_PCI_TCE_R			(1UL << 0)
64 #define	OPAL_PCI_TCE_W			(1UL << 1)
65 #define	PHB3_TCE_KILL_INVAL_ALL		(1UL << 63)
66 
67 /*
68  * Device interface.
69  */
70 static int		opalpci_probe(device_t);
71 static int		opalpci_attach(device_t);
72 
73 /*
74  * pcib interface.
75  */
76 static uint32_t		opalpci_read_config(device_t, u_int, u_int, u_int,
77 			    u_int, int);
78 static void		opalpci_write_config(device_t, u_int, u_int, u_int,
79 			    u_int, u_int32_t, int);
80 static int		opalpci_alloc_msi(device_t dev, device_t child,
81 			    int count, int maxcount, int *irqs);
82 static int		opalpci_release_msi(device_t dev, device_t child,
83 			    int count, int *irqs);
84 static int		opalpci_alloc_msix(device_t dev, device_t child,
85 			    int *irq);
86 static int		opalpci_release_msix(device_t dev, device_t child,
87 			    int irq);
88 static int		opalpci_map_msi(device_t dev, device_t child,
89 			    int irq, uint64_t *addr, uint32_t *data);
90 static int opalpci_route_interrupt(device_t bus, device_t dev, int pin);
91 
92 /*
93  * MSI PIC interface.
94  */
95 static void opalpic_pic_enable(device_t dev, u_int irq, u_int vector, void **);
96 static void opalpic_pic_eoi(device_t dev, u_int irq, void *);
97 
98 /* Bus interface */
99 static bus_dma_tag_t opalpci_get_dma_tag(device_t dev, device_t child);
100 
101 /*
102  * Commands
103  */
104 #define	OPAL_M32_WINDOW_TYPE		1
105 #define	OPAL_M64_WINDOW_TYPE		2
106 #define	OPAL_IO_WINDOW_TYPE		3
107 
108 #define	OPAL_RESET_PHB_COMPLETE		1
109 #define	OPAL_RESET_PCI_IODA_TABLE	6
110 
111 #define	OPAL_DISABLE_M64		0
112 #define	OPAL_ENABLE_M64_SPLIT		1
113 #define	OPAL_ENABLE_M64_NON_SPLIT	2
114 
115 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO	1
116 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_DMA	2
117 #define	OPAL_EEH_ACTION_CLEAR_FREEZE_ALL	3
118 
119 #define	OPAL_EEH_STOPPED_NOT_FROZEN		0
120 
121 /*
122  * Constants
123  */
124 #define OPAL_PCI_DEFAULT_PE			1
125 
126 #define OPAL_PCI_BUS_SPACE_LOWADDR_32BIT	0x7FFFFFFFUL
127 
128 /*
129  * Driver methods.
130  */
131 static device_method_t	opalpci_methods[] = {
132 	/* Device interface */
133 	DEVMETHOD(device_probe,		opalpci_probe),
134 	DEVMETHOD(device_attach,	opalpci_attach),
135 
136 	/* pcib interface */
137 	DEVMETHOD(pcib_read_config,	opalpci_read_config),
138 	DEVMETHOD(pcib_write_config,	opalpci_write_config),
139 
140 	DEVMETHOD(pcib_alloc_msi,	opalpci_alloc_msi),
141 	DEVMETHOD(pcib_release_msi,	opalpci_release_msi),
142 	DEVMETHOD(pcib_alloc_msix,	opalpci_alloc_msix),
143 	DEVMETHOD(pcib_release_msix,	opalpci_release_msix),
144 	DEVMETHOD(pcib_map_msi,		opalpci_map_msi),
145 	DEVMETHOD(pcib_route_interrupt,	opalpci_route_interrupt),
146 
147 	/* PIC interface for MSIs */
148 	DEVMETHOD(pic_enable,		opalpic_pic_enable),
149 	DEVMETHOD(pic_eoi,		opalpic_pic_eoi),
150 
151 	/* Bus interface */
152 	DEVMETHOD(bus_get_dma_tag,	opalpci_get_dma_tag),
153 	DEVMETHOD(bus_get_cpus,		ofw_pcibus_get_cpus),
154 	DEVMETHOD(bus_get_domain,	ofw_pcibus_get_domain),
155 
156 	DEVMETHOD_END
157 };
158 
159 struct opalpci_softc {
160 	struct ofw_pci_softc ofw_sc;
161 	uint64_t phb_id;
162 	vmem_t *msi_vmem;
163 	int msi_base;		/* Base XIVE number */
164 	int base_msi_irq;	/* Base IRQ assigned by FreeBSD to this PIC */
165 	uint64_t *tce;		/* TCE table for 1:1 mapping */
166 	struct resource *r_reg;
167 };
168 
169 DEFINE_CLASS_1(pcib, opalpci_driver, opalpci_methods,
170     sizeof(struct opalpci_softc), ofw_pcib_driver);
171 EARLY_DRIVER_MODULE(opalpci, ofwbus, opalpci_driver, 0, 0, BUS_PASS_BUS);
172 
173 static int
174 opalpci_probe(device_t dev)
175 {
176 	const char	*type;
177 
178 	if (opal_check() != 0)
179 		return (ENXIO);
180 
181 	type = ofw_bus_get_type(dev);
182 
183 	if (type == NULL || (strcmp(type, "pci") != 0 &&
184 	    strcmp(type, "pciex") != 0))
185 		return (ENXIO);
186 
187 	if (!OF_hasprop(ofw_bus_get_node(dev), "ibm,opal-phbid"))
188 		return (ENXIO);
189 
190 	device_set_desc(dev, "OPAL Host-PCI bridge");
191 	return (BUS_PROBE_GENERIC);
192 }
193 
194 static void
195 pci_phb3_tce_invalidate_entire(struct opalpci_softc *sc)
196 {
197 
198 	mb();
199 	bus_write_8(sc->r_reg, 0x210, PHB3_TCE_KILL_INVAL_ALL);
200 	mb();
201 }
202 
203 /* Simple function to round to a power of 2 */
204 static uint64_t
205 round_pow2(uint64_t val)
206 {
207 
208 	return (1 << (flsl(val + (val - 1)) - 1));
209 }
210 
211 /*
212  * Starting with skiboot 5.10 PCIe nodes have a new property,
213  * "ibm,supported-tce-sizes", to denote the TCE sizes available.  This allows us
214  * to avoid hard-coding the maximum TCE size allowed, and instead provide a sane
215  * default (however, the "sane" default, which works for all targets, is 64k,
216  * limiting us to 64GB if we have 1M entries.
217  */
218 static uint64_t
219 max_tce_size(device_t dev)
220 {
221 	phandle_t node;
222 	cell_t sizes[64]; /* Property is a list of bit-widths, up to 64-bits */
223 	int count;
224 
225 	node = ofw_bus_get_node(dev);
226 
227 	count = OF_getencprop(node, "ibm,supported-tce-sizes",
228 	    sizes, sizeof(sizes));
229 	if (count < (int) sizeof(cell_t))
230 		return OPAL_PCI_TCE_DEFAULT_SEG_SIZE;
231 
232 	count /= sizeof(cell_t);
233 
234 	return (1ULL << sizes[count - 1]);
235 }
236 
237 static int
238 opalpci_attach(device_t dev)
239 {
240 	struct opalpci_softc *sc;
241 	cell_t id[2], m64ranges[2], m64window[6], npe;
242 	phandle_t node;
243 	int i, err;
244 	uint64_t maxmem;
245 	uint64_t entries;
246 	uint64_t tce_size;
247 	uint64_t tce_tbl_size;
248 	int m64bar;
249 	int rid;
250 
251 	sc = device_get_softc(dev);
252 	node = ofw_bus_get_node(dev);
253 
254 	switch (OF_getproplen(node, "ibm,opal-phbid")) {
255 	case 8:
256 		OF_getencprop(node, "ibm,opal-phbid", id, 8);
257 		sc->phb_id = ((uint64_t)id[0] << 32) | id[1];
258 		break;
259 	case 4:
260 		OF_getencprop(node, "ibm,opal-phbid", id, 4);
261 		sc->phb_id = id[0];
262 		break;
263 	default:
264 		device_printf(dev, "PHB ID property had wrong length (%zd)\n",
265 		    OF_getproplen(node, "ibm,opal-phbid"));
266 		return (ENXIO);
267 	}
268 
269 	if (bootverbose)
270 		device_printf(dev, "OPAL ID %#lx\n", sc->phb_id);
271 
272 	rid = 0;
273 	sc->r_reg = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
274 	    &rid, RF_ACTIVE | RF_SHAREABLE);
275 	if (sc->r_reg == NULL) {
276 		device_printf(dev, "Failed to allocate PHB[%jd] registers\n",
277 		    (uintmax_t)sc->phb_id);
278 		return (ENXIO);
279 	}
280 
281 #if 0
282 	/*
283 	 * Reset PCI IODA table
284 	 */
285 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PCI_IODA_TABLE,
286 	    1);
287 	if (err != 0) {
288 		device_printf(dev, "IODA table reset failed: %d\n", err);
289 		return (ENXIO);
290 	}
291 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PHB_COMPLETE,
292 	    1);
293 	if (err < 0) {
294 		device_printf(dev, "PHB reset failed: %d\n", err);
295 		return (ENXIO);
296 	}
297 	if (err > 0) {
298 		while ((err = opal_call(OPAL_PCI_POLL, sc->phb_id)) > 0) {
299 			DELAY(1000*(err + 1)); /* Returns expected delay in ms */
300 		}
301 	}
302 	if (err < 0) {
303 		device_printf(dev, "WARNING: PHB IODA reset poll failed: %d\n", err);
304 	}
305 	err = opal_call(OPAL_PCI_RESET, sc->phb_id, OPAL_RESET_PHB_COMPLETE,
306 	    0);
307 	if (err < 0) {
308 		device_printf(dev, "PHB reset failed: %d\n", err);
309 		return (ENXIO);
310 	}
311 	if (err > 0) {
312 		while ((err = opal_call(OPAL_PCI_POLL, sc->phb_id)) > 0) {
313 			DELAY(1000*(err + 1)); /* Returns expected delay in ms */
314 		}
315 	}
316 #endif
317 
318 	/*
319 	 * Map all devices on the bus to partitionable endpoint one until
320 	 * such time as we start wanting to do things like bhyve.
321 	 */
322 	err = opal_call(OPAL_PCI_SET_PE, sc->phb_id, OPAL_PCI_DEFAULT_PE,
323 	    0, OPAL_PCI_BUS_ANY, OPAL_IGNORE_RID_DEVICE_NUMBER,
324 	    OPAL_IGNORE_RID_FUNC_NUMBER, OPAL_MAP_PE);
325 	if (err != 0) {
326 		device_printf(dev, "PE mapping failed: %d\n", err);
327 		return (ENXIO);
328 	}
329 
330 	/*
331 	 * Turn on MMIO, mapped to PE 1
332 	 */
333 	if (OF_getencprop(node, "ibm,opal-num-pes", &npe, 4) != 4)
334 		npe = 1;
335 	for (i = 0; i < npe; i++) {
336 		err = opal_call(OPAL_PCI_MAP_PE_MMIO_WINDOW, sc->phb_id,
337 		    OPAL_PCI_DEFAULT_PE, OPAL_M32_WINDOW_TYPE, 0, i);
338 		if (err != 0)
339 			device_printf(dev, "MMIO %d map failed: %d\n", i, err);
340 	}
341 
342 	if (OF_getencprop(node, "ibm,opal-available-m64-ranges",
343 	    m64ranges, sizeof(m64ranges)) == sizeof(m64ranges))
344 		m64bar = m64ranges[0];
345 	else
346 	    m64bar = 0;
347 
348 	/* XXX: multiple M64 windows? */
349 	if (OF_getencprop(node, "ibm,opal-m64-window",
350 	    m64window, sizeof(m64window)) == sizeof(m64window)) {
351 		opal_call(OPAL_PCI_PHB_MMIO_ENABLE, sc->phb_id,
352 		    OPAL_M64_WINDOW_TYPE, m64bar, 0);
353 		opal_call(OPAL_PCI_SET_PHB_MEM_WINDOW, sc->phb_id,
354 		    OPAL_M64_WINDOW_TYPE, m64bar /* index */,
355 		    ((uint64_t)m64window[2] << 32) | m64window[3], 0,
356 		    ((uint64_t)m64window[4] << 32) | m64window[5]);
357 		opal_call(OPAL_PCI_MAP_PE_MMIO_WINDOW, sc->phb_id,
358 		    OPAL_PCI_DEFAULT_PE, OPAL_M64_WINDOW_TYPE,
359 		    m64bar /* index */, 0);
360 		opal_call(OPAL_PCI_PHB_MMIO_ENABLE, sc->phb_id,
361 		    OPAL_M64_WINDOW_TYPE, m64bar, OPAL_ENABLE_M64_NON_SPLIT);
362 	}
363 
364 	/*
365 	 * Enable IOMMU for PE1 - map everything 1:1 using
366 	 * segments of max_tce_size size
367 	 */
368 	tce_size = max_tce_size(dev);
369 	maxmem = roundup2(powerpc_ptob(Maxmem), tce_size);
370 	entries = round_pow2(maxmem / tce_size);
371 	tce_tbl_size = MAX(entries * sizeof(uint64_t), 4096);
372 	if (entries > OPAL_PCI_TCE_MAX_ENTRIES)
373 		panic("POWERNV supports only %jdGB of memory space\n",
374 		    (uintmax_t)((OPAL_PCI_TCE_MAX_ENTRIES * tce_size) >> 30));
375 	if (bootverbose)
376 		device_printf(dev, "Mapping 0-%#jx for DMA\n", (uintmax_t)maxmem);
377 	sc->tce = contigmalloc(tce_tbl_size,
378 	    M_DEVBUF, M_NOWAIT | M_ZERO, 0,
379 	    BUS_SPACE_MAXADDR, tce_tbl_size, 0);
380 	if (sc->tce == NULL)
381 		panic("Failed to allocate TCE memory for PHB %jd\n",
382 		    (uintmax_t)sc->phb_id);
383 
384 	for (i = 0; i < entries; i++)
385 		sc->tce[i] = htobe64((i * tce_size) | OPAL_PCI_TCE_R | OPAL_PCI_TCE_W);
386 
387 	/* Map TCE for every PE. It seems necessary for Power8 */
388 	for (i = 0; i < npe; i++) {
389 		err = opal_call(OPAL_PCI_MAP_PE_DMA_WINDOW, sc->phb_id,
390 		    i, (i << 1),
391 		    1, pmap_kextract((uint64_t)&sc->tce[0]),
392 		    tce_tbl_size, tce_size);
393 		if (err != 0) {
394 			device_printf(dev, "DMA IOMMU mapping failed: %d\n", err);
395 			return (ENXIO);
396 		}
397 
398 		err = opal_call(OPAL_PCI_MAP_PE_DMA_WINDOW_REAL, sc->phb_id,
399 		    i, (i << 1) + 1,
400 		    (1UL << 59), maxmem);
401 		if (err != 0) {
402 			device_printf(dev, "DMA 64b bypass mapping failed: %d\n", err);
403 			return (ENXIO);
404 		}
405 	}
406 
407 	/*
408 	 * Invalidate all previous TCE entries.
409 	 */
410 	if (ofw_bus_is_compatible(dev, "power8-pciex"))
411 		pci_phb3_tce_invalidate_entire(sc);
412 	else
413 		opal_call(OPAL_PCI_TCE_KILL, sc->phb_id, OPAL_PCI_TCE_KILL_ALL,
414 		    OPAL_PCI_DEFAULT_PE, 0, 0, 0);
415 
416 	/*
417 	 * Get MSI properties
418 	 */
419 	sc->msi_vmem = NULL;
420 	if (OF_getproplen(node, "ibm,opal-msi-ranges") > 0) {
421 		cell_t msi_ranges[2];
422 		OF_getencprop(node, "ibm,opal-msi-ranges",
423 		    msi_ranges, sizeof(msi_ranges));
424 		sc->msi_base = msi_ranges[0];
425 
426 		sc->msi_vmem = vmem_create("OPAL MSI", msi_ranges[0],
427 		    msi_ranges[1], 1, 0, M_BESTFIT | M_WAITOK);
428 
429 		sc->base_msi_irq = powerpc_register_pic(dev,
430 		    OF_xref_from_node(node),
431 		    msi_ranges[0] + msi_ranges[1], 0, FALSE);
432 
433 		if (bootverbose)
434 			device_printf(dev, "Supports %d MSIs starting at %d\n",
435 			    msi_ranges[1], msi_ranges[0]);
436 	}
437 
438 	/* Create the parent DMA tag */
439 	/*
440 	 * Constrain it to POWER8 PHB (ioda2) for now.  It seems to mess up on
441 	 * POWER9 systems.
442 	 */
443 	if (ofw_bus_is_compatible(dev, "ibm,ioda2-phb")) {
444 		err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
445 		    1, 0,				/* alignment, bounds */
446 		    OPAL_PCI_BUS_SPACE_LOWADDR_32BIT,	/* lowaddr */
447 		    BUS_SPACE_MAXADDR_32BIT,		/* highaddr */
448 		    NULL, NULL,				/* filter, filterarg */
449 		    BUS_SPACE_MAXSIZE,			/* maxsize */
450 		    BUS_SPACE_UNRESTRICTED,		/* nsegments */
451 		    BUS_SPACE_MAXSIZE,			/* maxsegsize */
452 		    0,					/* flags */
453 		    NULL, NULL,				/* lockfunc, lockarg */
454 		    &sc->ofw_sc.sc_dmat);
455 		if (err != 0) {
456 			device_printf(dev, "Failed to create DMA tag\n");
457 			return (err);
458 		}
459 	}
460 
461 	/*
462 	 * General OFW PCI attach
463 	 */
464 	err = ofw_pcib_init(dev);
465 	if (err != 0)
466 		return (err);
467 
468 	/*
469 	 * Unfreeze non-config-space PCI operations. Let this fail silently
470 	 * if e.g. there is no current freeze.
471 	 */
472 	opal_call(OPAL_PCI_EEH_FREEZE_CLEAR, sc->phb_id, OPAL_PCI_DEFAULT_PE,
473 	    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
474 
475 	/*
476 	 * OPAL stores 64-bit BARs in a special property rather than "ranges"
477 	 */
478 	if (OF_getencprop(node, "ibm,opal-m64-window",
479 	    m64window, sizeof(m64window)) == sizeof(m64window)) {
480 		struct ofw_pci_range *rp;
481 
482 		sc->ofw_sc.sc_nrange++;
483 		sc->ofw_sc.sc_range = realloc(sc->ofw_sc.sc_range,
484 		    sc->ofw_sc.sc_nrange * sizeof(sc->ofw_sc.sc_range[0]),
485 		    M_DEVBUF, M_WAITOK);
486 		rp = &sc->ofw_sc.sc_range[sc->ofw_sc.sc_nrange-1];
487 		rp->pci_hi = OFW_PCI_PHYS_HI_SPACE_MEM64 |
488 		    OFW_PCI_PHYS_HI_PREFETCHABLE;
489 		rp->pci = ((uint64_t)m64window[0] << 32) | m64window[1];
490 		rp->host = ((uint64_t)m64window[2] << 32) | m64window[3];
491 		rp->size = ((uint64_t)m64window[4] << 32) | m64window[5];
492 		rman_manage_region(&sc->ofw_sc.sc_mem_rman, rp->pci,
493 		   rp->pci + rp->size - 1);
494 	}
495 
496 	return (ofw_pcib_attach(dev));
497 }
498 
499 static uint32_t
500 opalpci_read_config(device_t dev, u_int bus, u_int slot, u_int func, u_int reg,
501     int width)
502 {
503 	struct opalpci_softc *sc;
504 	uint64_t config_addr;
505 	uint8_t byte, eeh_state;
506 	uint16_t half;
507 	uint32_t word;
508 	int error;
509 	uint16_t err_type;
510 
511 	sc = device_get_softc(dev);
512 
513 	config_addr = (bus << 8) | ((slot & 0x1f) << 3) | (func & 0x7);
514 
515 	switch (width) {
516 	case 1:
517 		error = opal_call(OPAL_PCI_CONFIG_READ_BYTE, sc->phb_id,
518 		    config_addr, reg, vtophys(&byte));
519 		word = byte;
520 		break;
521 	case 2:
522 		error = opal_call(OPAL_PCI_CONFIG_READ_HALF_WORD, sc->phb_id,
523 		    config_addr, reg, vtophys(&half));
524 		word = be16toh(half);
525 		break;
526 	case 4:
527 		error = opal_call(OPAL_PCI_CONFIG_READ_WORD, sc->phb_id,
528 		    config_addr, reg, vtophys(&word));
529 		word = be32toh(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 			err_type = be16toh(err_type); /* XXX unused */
549 			if (eeh_state != OPAL_EEH_STOPPED_NOT_FROZEN)
550 				opal_call(OPAL_PCI_EEH_FREEZE_CLEAR,
551 				    sc->phb_id, OPAL_PCI_DEFAULT_PE,
552 				    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
553 		}
554 		if (error != OPAL_SUCCESS)
555 			word = 0xffffffff;
556 	}
557 
558 	return (word);
559 }
560 
561 static void
562 opalpci_write_config(device_t dev, u_int bus, u_int slot, u_int func,
563     u_int reg, uint32_t val, int width)
564 {
565 	struct opalpci_softc *sc;
566 	uint64_t config_addr;
567 	int error = OPAL_SUCCESS;
568 
569 	sc = device_get_softc(dev);
570 
571 	config_addr = (bus << 8) | ((slot & 0x1f) << 3) | (func & 0x7);
572 
573 	switch (width) {
574 	case 1:
575 		error = opal_call(OPAL_PCI_CONFIG_WRITE_BYTE, sc->phb_id,
576 		    config_addr, reg, val);
577 		break;
578 	case 2:
579 		error = opal_call(OPAL_PCI_CONFIG_WRITE_HALF_WORD, sc->phb_id,
580 		    config_addr, reg, val);
581 		break;
582 	case 4:
583 		error = opal_call(OPAL_PCI_CONFIG_WRITE_WORD, sc->phb_id,
584 		    config_addr, reg, val);
585 		break;
586 	}
587 
588 	if (error != OPAL_SUCCESS) {
589 		/*
590 		 * Poking config state for non-existant devices can make
591 		 * the host bridge hang up. Clear any errors.
592 		 */
593 		if (error != OPAL_HARDWARE) {
594 			opal_call(OPAL_PCI_EEH_FREEZE_CLEAR,
595 			    sc->phb_id, OPAL_PCI_DEFAULT_PE,
596 			    OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
597 		}
598 	}
599 }
600 
601 static int
602 opalpci_route_interrupt(device_t bus, device_t dev, int pin)
603 {
604 
605 	return (pin);
606 }
607 
608 static int
609 opalpci_alloc_msi(device_t dev, device_t child, int count, int maxcount,
610     int *irqs)
611 {
612 	struct opalpci_softc *sc;
613 	vmem_addr_t start;
614 	phandle_t xref;
615 	int err, i;
616 
617 	sc = device_get_softc(dev);
618 	if (sc->msi_vmem == NULL)
619 		return (ENODEV);
620 
621 	err = vmem_xalloc(sc->msi_vmem, count, powerof2(count), 0, 0,
622 	    VMEM_ADDR_MIN, VMEM_ADDR_MAX, M_BESTFIT | M_WAITOK, &start);
623 
624 	if (err)
625 		return (err);
626 
627 	xref = OF_xref_from_node(ofw_bus_get_node(dev));
628 	for (i = 0; i < count; i++)
629 		irqs[i] = MAP_IRQ(xref, start + i);
630 
631 	return (0);
632 }
633 
634 static int
635 opalpci_release_msi(device_t dev, device_t child, int count, int *irqs)
636 {
637 	struct opalpci_softc *sc;
638 
639 	sc = device_get_softc(dev);
640 	if (sc->msi_vmem == NULL)
641 		return (ENODEV);
642 
643 	vmem_xfree(sc->msi_vmem, irqs[0] - sc->base_msi_irq, count);
644 	return (0);
645 }
646 
647 static int
648 opalpci_alloc_msix(device_t dev, device_t child, int *irq)
649 {
650 	return (opalpci_alloc_msi(dev, child, 1, 1, irq));
651 }
652 
653 static int
654 opalpci_release_msix(device_t dev, device_t child, int irq)
655 {
656 	return (opalpci_release_msi(dev, child, 1, &irq));
657 }
658 
659 static int
660 opalpci_map_msi(device_t dev, device_t child, int irq, uint64_t *addr,
661     uint32_t *data)
662 {
663 	struct opalpci_softc *sc;
664 	struct pci_devinfo *dinfo;
665 	int err, xive;
666 
667 	sc = device_get_softc(dev);
668 	if (sc->msi_vmem == NULL)
669 		return (ENODEV);
670 
671 	xive = irq - sc->base_msi_irq - sc->msi_base;
672 	opal_call(OPAL_PCI_SET_XIVE_PE, sc->phb_id, OPAL_PCI_DEFAULT_PE, xive);
673 
674 	dinfo = device_get_ivars(child);
675 	if (dinfo->cfg.msi.msi_alloc > 0 &&
676 	    (dinfo->cfg.msi.msi_ctrl & PCIM_MSICTRL_64BIT) == 0) {
677 		uint32_t msi32;
678 		err = opal_call(OPAL_GET_MSI_32, sc->phb_id,
679 		    OPAL_PCI_DEFAULT_PE, xive, 1, vtophys(&msi32),
680 		    vtophys(data));
681 		*addr = be32toh(msi32);
682 	} else {
683 		err = opal_call(OPAL_GET_MSI_64, sc->phb_id,
684 		    OPAL_PCI_DEFAULT_PE, xive, 1, vtophys(addr), vtophys(data));
685 		*addr = be64toh(*addr);
686 	}
687 	*data = be32toh(*data);
688 
689 	if (bootverbose && err != 0)
690 		device_printf(child, "OPAL MSI mapping error: %d\n", err);
691 
692 	return ((err == 0) ? 0 : ENXIO);
693 }
694 
695 static void
696 opalpic_pic_enable(device_t dev, u_int irq, u_int vector, void **priv)
697 {
698 	struct opalpci_softc *sc = device_get_softc(dev);
699 
700 	PIC_ENABLE(root_pic, irq, vector, priv);
701 	opal_call(OPAL_PCI_MSI_EOI, sc->phb_id, irq, priv);
702 }
703 
704 static void opalpic_pic_eoi(device_t dev, u_int irq, void *priv)
705 {
706 	struct opalpci_softc *sc;
707 
708 	sc = device_get_softc(dev);
709 	opal_call(OPAL_PCI_MSI_EOI, sc->phb_id, irq);
710 
711 	PIC_EOI(root_pic, irq, priv);
712 }
713 
714 static bus_dma_tag_t
715 opalpci_get_dma_tag(device_t dev, device_t child)
716 {
717 	struct opalpci_softc *sc;
718 
719 	sc = device_get_softc(dev);
720 	return (sc->ofw_sc.sc_dmat);
721 }
722