1/* $NetBSD: pci_resource.c,v 1.4 2023/11/12 13:56:20 jmcneill Exp $ */
2
3/*-
4 * Copyright (c) 2022 Jared McNeill <jmcneill@invisible.ca>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29/*
30 * pci_resource.c --
31 *
32 * Scan current PCI resource allocations and attempt to assign resources
33 * to devices that are not configured WITHOUT changing any configuration
34 * performed by system firmware.
35 */
36
37#include <sys/cdefs.h>
38__KERNEL_RCSID(0, "$NetBSD: pci_resource.c,v 1.4 2023/11/12 13:56:20 jmcneill Exp $");
39
40#include <sys/param.h>
41#include <sys/bus.h>
42#include <sys/systm.h>
43#include <sys/kmem.h>
44#include <sys/vmem.h>
45
46#include <dev/pci/pcireg.h>
47#include <dev/pci/pcivar.h>
48#include <dev/pci/pcidevs.h>
49#include <dev/pci/pci_resource.h>
50
51#define	DPRINT		aprint_debug
52
53#if defined(PCI_RESOURCE_TEST_VENDOR_ID) && \
54    defined(PCI_RESOURCE_TEST_PRODUCT_ID)
55#define IS_TEST_DEVICE(_pd)						      \
56	(PCI_VENDOR(pd->pd_id) == PCI_RESOURCE_TEST_VENDOR_ID &&	      \
57	 PCI_PRODUCT(pd->pd_id) == PCI_RESOURCE_TEST_PRODUCT_ID)
58#else
59#define IS_TEST_DEVICE(_pd)	0
60#endif
61
62#define	PCI_MAX_DEVICE	32
63#define	PCI_MAX_FUNC	8
64
65#define	PCI_MAX_IORES	6
66
67#define	PCI_RANGE_FOREACH(_type)					      \
68	for (u_int _type = PCI_RANGE_BUS; _type < NUM_PCI_RANGES; _type++)
69
70static const char *pci_range_typenames[NUM_PCI_RANGES] = {
71	[PCI_RANGE_BUS]  = "bus",
72	[PCI_RANGE_IO]   = "io",
73	[PCI_RANGE_MEM]  = "mem",
74	[PCI_RANGE_PMEM] = "pmem",
75};
76
77struct pci_bus;
78
79struct pci_iores {
80	uint64_t	pi_base;	/* Base address */
81	uint64_t	pi_size;	/* Resource size */
82	uint8_t		pi_type;	/* PCI_MAPREG_TYPE_* */
83	u_int		pi_bar;		/* PCI bar number */
84	union {
85		struct {
86			uint8_t		memtype;
87			bool		prefetch;
88		} pi_mem;
89	};
90};
91
92struct pci_device {
93	bool		pd_present;	/* Device is present */
94	bool		pd_configured;	/* Device is configured */
95	struct pci_bus *pd_bus;	/* Parent bus */
96	uint8_t		pd_devno;	/* Device number */
97	uint8_t		pd_funcno;	/* Function number */
98	pcitag_t	pd_tag;		/* PCI tag */
99
100	pcireg_t	pd_id;		/* Vendor ID, Device ID */
101	pcireg_t	pd_class;	/* Revision ID, Class Code */
102	pcireg_t	pd_bhlc;	/* BIST, Header Type, Primary Latency
103					 * Timer, Cache Line Size */
104
105	struct pci_iores pd_iores[PCI_MAX_IORES];
106	u_int		pd_niores;
107
108	bool		pd_ppb;		/* PCI-PCI bridge */
109	union {
110		struct {
111			pcireg_t	bridge_bus;
112			struct pci_resource_range ranges[NUM_PCI_RANGES];
113		} pd_bridge;
114	};
115};
116
117struct pci_bus {
118	uint8_t		pb_busno;	/* Bus number */
119	struct pci_device *pb_bridge; /* Parent bridge, or NULL */
120
121	struct pci_device pb_device[PCI_MAX_DEVICE * PCI_MAX_FUNC];
122					/* Devices on bus */
123	u_int		pb_lastdevno;	/* Last device found */
124
125	struct pci_resource_range pb_ranges[NUM_PCI_RANGES];
126	vmem_t		*pb_res[NUM_PCI_RANGES];
127};
128
129struct pci_resources {
130	struct pci_bus **pr_bus;	/* Bus list */
131	pci_chipset_tag_t pr_pc;	/* Chipset tag */
132	uint8_t		pr_startbus;	/* First bus number */
133	uint8_t		pr_endbus;	/* Last bus number */
134
135	struct pci_resource_range pr_ranges[NUM_PCI_RANGES];
136	vmem_t		*pr_res[NUM_PCI_RANGES];
137};
138
139static void	pci_resource_scan_bus(struct pci_resources *,
140		    struct pci_device *, uint8_t);
141
142#define	PCI_SBDF_FMT			"%04x:%02x:%02x.%u"
143#define	PCI_SBDF_FMT_ARGS(_pr, _pd)	\
144	pci_get_segment((_pr)->pr_pc),	\
145	(_pd)->pd_bus->pb_busno,	\
146	(_pd)->pd_devno,		\
147	(_pd)->pd_funcno
148
149#define	PCICONF_RES_BUS(_pr, _busno)				\
150	((_pr)->pr_bus[(_busno) - (_pr)->pr_startbus])
151#define	PCICONF_BUS_DEVICE(_pb, _devno, _funcno)		\
152	(&(_pb)->pb_device[(_devno) * PCI_MAX_FUNC + (_funcno)])
153
154/*
155 * pci_create_vmem --
156 *
157 *   Create a vmem arena covering the specified range, used for tracking
158 *   PCI resources.
159 */
160static vmem_t *
161pci_create_vmem(const char *name, bus_addr_t start, bus_addr_t end)
162{
163	vmem_t *arena;
164	int error __diagused;
165
166	arena = vmem_create(name, 0, 0, 1, NULL, NULL, NULL, 0, VM_SLEEP,
167	    IPL_NONE);
168	error = vmem_add(arena, start, end - start + 1, VM_SLEEP);
169	KASSERTMSG(error == 0, "error=%d", error);
170
171	return arena;
172}
173
174/*
175 * pci_new_bus --
176 *
177 *   Create a new PCI bus and initialize its resource ranges.
178 */
179static struct pci_bus *
180pci_new_bus(struct pci_resources *pr, uint8_t busno, struct pci_device *bridge)
181{
182	struct pci_bus *pb;
183	struct pci_resource_range *ranges;
184
185	pb = kmem_zalloc(sizeof(*pb), KM_SLEEP);
186	pb->pb_busno = busno;
187	pb->pb_bridge = bridge;
188	if (bridge == NULL) {
189		/*
190		 * No additional constraints on resource allocations for
191		 * the root bus.
192		 */
193		ranges = pr->pr_ranges;
194	} else {
195		/*
196		 * Resource allocations for this bus are constrained by the
197		 * bridge forwarding settings.
198		 */
199		ranges = bridge->pd_bridge.ranges;
200	}
201	memcpy(pb->pb_ranges, ranges, sizeof(pb->pb_ranges));
202
203	return pb;
204}
205
206/*
207 * pci_resource_device_functions --
208 *
209 *   Returns the number of PCI functions for a a given bus and device.
210 */
211static uint8_t
212pci_resource_device_functions(struct pci_resources *pr,
213    uint8_t busno, uint8_t devno)
214{
215	struct pci_bus *pb;
216	struct pci_device *pd;
217
218	pb = PCICONF_RES_BUS(pr, busno);
219	pd = PCICONF_BUS_DEVICE(pb, devno, 0);
220	if (!pd->pd_present) {
221		return 0;
222	}
223
224	return PCI_HDRTYPE_MULTIFN(pd->pd_bhlc) ? 8 : 1;
225}
226
227/*
228 * pci_resource_device_print --
229 *
230 *   Log details about a device.
231 */
232static void
233pci_resource_device_print(struct pci_resources *pr,
234    struct pci_device *pd)
235{
236	struct pci_iores *pi;
237	u_int res;
238
239	DPRINT("PCI: " PCI_SBDF_FMT " %04x:%04x %02x 0x%06x",
240	       PCI_SBDF_FMT_ARGS(pr, pd),
241	       PCI_VENDOR(pd->pd_id), PCI_PRODUCT(pd->pd_id),
242	       PCI_REVISION(pd->pd_class), (pd->pd_class >> 8) & 0xffffff);
243
244	switch (PCI_HDRTYPE_TYPE(pd->pd_bhlc)) {
245	case PCI_HDRTYPE_DEVICE:
246		DPRINT(" (device)\n");
247		break;
248	case PCI_HDRTYPE_PPB:
249		DPRINT(" (bridge %u -> %u-%u)\n",
250		    PCI_BRIDGE_BUS_NUM_PRIMARY(pd->pd_bridge.bridge_bus),
251		    PCI_BRIDGE_BUS_NUM_SECONDARY(pd->pd_bridge.bridge_bus),
252		    PCI_BRIDGE_BUS_NUM_SUBORDINATE(pd->pd_bridge.bridge_bus));
253
254		if (pd->pd_bridge.ranges[PCI_RANGE_IO].end) {
255			DPRINT("PCI: " PCI_SBDF_FMT
256			       " [bridge] window io  %#" PRIx64 "-%#" PRIx64
257			       "\n",
258			       PCI_SBDF_FMT_ARGS(pr, pd),
259			       pd->pd_bridge.ranges[PCI_RANGE_IO].start,
260			       pd->pd_bridge.ranges[PCI_RANGE_IO].end);
261		}
262		if (pd->pd_bridge.ranges[PCI_RANGE_MEM].end) {
263			DPRINT("PCI: " PCI_SBDF_FMT
264			       " [bridge] window mem %#" PRIx64 "-%#" PRIx64
265			       " (non-prefetchable)\n",
266			       PCI_SBDF_FMT_ARGS(pr, pd),
267			       pd->pd_bridge.ranges[PCI_RANGE_MEM].start,
268			       pd->pd_bridge.ranges[PCI_RANGE_MEM].end);
269		}
270		if (pd->pd_bridge.ranges[PCI_RANGE_PMEM].end) {
271			DPRINT("PCI: " PCI_SBDF_FMT
272			       " [bridge] window mem %#" PRIx64 "-%#" PRIx64
273			       " (prefetchable)\n",
274			       PCI_SBDF_FMT_ARGS(pr, pd),
275			       pd->pd_bridge.ranges[PCI_RANGE_PMEM].start,
276			       pd->pd_bridge.ranges[PCI_RANGE_PMEM].end);
277		}
278
279		break;
280	default:
281		DPRINT(" (0x%02x)\n", PCI_HDRTYPE_TYPE(pd->pd_bhlc));
282	}
283
284	for (res = 0; res < pd->pd_niores; res++) {
285		pi = &pd->pd_iores[res];
286
287		DPRINT("PCI: " PCI_SBDF_FMT
288		       " [device] resource BAR%u: %s @ %#" PRIx64 " size %#"
289		       PRIx64,
290		       PCI_SBDF_FMT_ARGS(pr, pd), pi->pi_bar,
291		       pi->pi_type == PCI_MAPREG_TYPE_MEM ? "mem" : "io ",
292		       pi->pi_base, pi->pi_size);
293
294		if (pi->pi_type == PCI_MAPREG_TYPE_MEM) {
295			switch (pi->pi_mem.memtype) {
296			case PCI_MAPREG_MEM_TYPE_32BIT:
297				DPRINT(", 32-bit");
298				break;
299			case PCI_MAPREG_MEM_TYPE_32BIT_1M:
300				DPRINT(", 32-bit (1M)");
301				break;
302			case PCI_MAPREG_MEM_TYPE_64BIT:
303				DPRINT(", 64-bit");
304				break;
305			}
306			DPRINT(" %sprefetchable",
307			    pi->pi_mem.prefetch ? "" : "non-");
308		}
309		DPRINT("\n");
310	}
311}
312
313/*
314 * pci_resource_scan_bar --
315 *
316 *   Determine the current BAR configuration for a given device.
317 */
318static void
319pci_resource_scan_bar(struct pci_resources *pr,
320    struct pci_device *pd, pcireg_t mapreg_start, pcireg_t mapreg_end,
321    bool is_ppb)
322{
323	pci_chipset_tag_t pc = pr->pr_pc;
324	pcitag_t tag = pd->pd_tag;
325	pcireg_t mapreg = mapreg_start;
326	pcireg_t ocmd, cmd, bar[2], mask[2];
327	uint64_t addr, size;
328	struct pci_iores *pi;
329
330	if (!is_ppb) {
331		ocmd = cmd = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
332		cmd &= ~(PCI_COMMAND_MASTER_ENABLE |
333			 PCI_COMMAND_MEM_ENABLE |
334			 PCI_COMMAND_IO_ENABLE);
335		pci_conf_write(pc, tag, PCI_COMMAND_STATUS_REG, cmd);
336	}
337
338	while (mapreg < mapreg_end) {
339		u_int width = 4;
340
341		bar[0] = pci_conf_read(pc, tag, mapreg);
342		pci_conf_write(pc, tag, mapreg, 0xffffffff);
343		mask[0] = pci_conf_read(pc, tag, mapreg);
344		pci_conf_write(pc, tag, mapreg, bar[0]);
345
346		switch (PCI_MAPREG_TYPE(mask[0])) {
347		case PCI_MAPREG_TYPE_MEM:
348			switch (PCI_MAPREG_MEM_TYPE(mask[0])) {
349			case PCI_MAPREG_MEM_TYPE_32BIT:
350			case PCI_MAPREG_MEM_TYPE_32BIT_1M:
351				size = PCI_MAPREG_MEM_SIZE(mask[0]);
352				addr = PCI_MAPREG_MEM_ADDR(bar[0]);
353				break;
354			case PCI_MAPREG_MEM_TYPE_64BIT:
355				bar[1] = pci_conf_read(pc, tag, mapreg + 4);
356				pci_conf_write(pc, tag, mapreg + 4, 0xffffffff);
357				mask[1] = pci_conf_read(pc, tag, mapreg + 4);
358				pci_conf_write(pc, tag, mapreg + 4, bar[1]);
359
360				size = PCI_MAPREG_MEM64_SIZE(
361				    ((uint64_t)mask[1] << 32) | mask[0]);
362				addr = PCI_MAPREG_MEM64_ADDR(
363				    ((uint64_t)bar[1] << 32) | bar[0]);
364				width = 8;
365				break;
366			default:
367				size = 0;
368			}
369			if (size > 0) {
370				pi = &pd->pd_iores[pd->pd_niores++];
371				pi->pi_type = PCI_MAPREG_TYPE_MEM;
372				pi->pi_base = addr;
373				pi->pi_size = size;
374				pi->pi_bar = (mapreg - mapreg_start) / 4;
375				pi->pi_mem.memtype =
376				    PCI_MAPREG_MEM_TYPE(mask[0]);
377				pi->pi_mem.prefetch =
378				    PCI_MAPREG_MEM_PREFETCHABLE(mask[0]);
379			}
380			break;
381		case PCI_MAPREG_TYPE_IO:
382			size = PCI_MAPREG_IO_SIZE(mask[0] | 0xffff0000);
383			addr = PCI_MAPREG_IO_ADDR(bar[0]);
384			if (size > 0) {
385				pi = &pd->pd_iores[pd->pd_niores++];
386				pi->pi_type = PCI_MAPREG_TYPE_IO;
387				pi->pi_base = addr;
388				pi->pi_size = size;
389				pi->pi_bar = (mapreg - mapreg_start) / 4;
390			}
391			break;
392		}
393
394		KASSERT(pd->pd_niores <= PCI_MAX_IORES);
395
396		mapreg += width;
397	}
398
399	if (!is_ppb) {
400		pci_conf_write(pc, tag, PCI_COMMAND_STATUS_REG, ocmd);
401	}
402}
403
404/*
405 * pci_resource_scan_bridge --
406 *
407 *   Determine the current configuration of a PCI-PCI bridge.
408 */
409static void
410pci_resource_scan_bridge(struct pci_resources *pr,
411    struct pci_device *pd)
412{
413	pci_chipset_tag_t pc = pr->pr_pc;
414	pcitag_t tag = pd->pd_tag;
415	pcireg_t res, reshigh;
416
417	pd->pd_ppb = true;
418
419	res = pci_conf_read(pc, tag, PCI_BRIDGE_BUS_REG);
420	pd->pd_bridge.bridge_bus = res;
421	pd->pd_bridge.ranges[PCI_RANGE_BUS].start =
422	    PCI_BRIDGE_BUS_NUM_SECONDARY(res);
423	pd->pd_bridge.ranges[PCI_RANGE_BUS].end =
424	    PCI_BRIDGE_BUS_NUM_SUBORDINATE(res);
425
426	res = pci_conf_read(pc, tag, PCI_BRIDGE_STATIO_REG);
427	pd->pd_bridge.ranges[PCI_RANGE_IO].start =
428	    PCI_BRIDGE_STATIO_IOBASE_ADDR(res);
429	pd->pd_bridge.ranges[PCI_RANGE_IO].end =
430	    PCI_BRIDGE_STATIO_IOLIMIT_ADDR(res);
431	if (PCI_BRIDGE_IO_32BITS(res)) {
432		reshigh = pci_conf_read(pc, tag, PCI_BRIDGE_IOHIGH_REG);
433		pd->pd_bridge.ranges[PCI_RANGE_IO].start |=
434		    __SHIFTOUT(reshigh, PCI_BRIDGE_IOHIGH_BASE) << 16;
435		pd->pd_bridge.ranges[PCI_RANGE_IO].end |=
436		    __SHIFTOUT(reshigh, PCI_BRIDGE_IOHIGH_LIMIT) << 16;
437	}
438	if (pd->pd_bridge.ranges[PCI_RANGE_IO].start >=
439	    pd->pd_bridge.ranges[PCI_RANGE_IO].end) {
440		pd->pd_bridge.ranges[PCI_RANGE_IO].start = 0;
441		pd->pd_bridge.ranges[PCI_RANGE_IO].end = 0;
442	}
443
444	res = pci_conf_read(pc, tag, PCI_BRIDGE_MEMORY_REG);
445	pd->pd_bridge.ranges[PCI_RANGE_MEM].start =
446	    PCI_BRIDGE_MEMORY_BASE_ADDR(res);
447	pd->pd_bridge.ranges[PCI_RANGE_MEM].end =
448	    PCI_BRIDGE_MEMORY_LIMIT_ADDR(res);
449	if (pd->pd_bridge.ranges[PCI_RANGE_MEM].start >=
450	    pd->pd_bridge.ranges[PCI_RANGE_MEM].end) {
451		pd->pd_bridge.ranges[PCI_RANGE_MEM].start = 0;
452		pd->pd_bridge.ranges[PCI_RANGE_MEM].end = 0;
453	}
454
455	res = pci_conf_read(pc, tag, PCI_BRIDGE_PREFETCHMEM_REG);
456	pd->pd_bridge.ranges[PCI_RANGE_PMEM].start =
457	    PCI_BRIDGE_PREFETCHMEM_BASE_ADDR(res);
458	pd->pd_bridge.ranges[PCI_RANGE_PMEM].end =
459	    PCI_BRIDGE_PREFETCHMEM_LIMIT_ADDR(res);
460	if (PCI_BRIDGE_PREFETCHMEM_64BITS(res)) {
461		reshigh = pci_conf_read(pc, tag,
462		    PCI_BRIDGE_PREFETCHBASEUP32_REG);
463		pd->pd_bridge.ranges[PCI_RANGE_PMEM].start |=
464		    (uint64_t)reshigh << 32;
465		reshigh = pci_conf_read(pc, tag,
466		    PCI_BRIDGE_PREFETCHLIMITUP32_REG);
467		pd->pd_bridge.ranges[PCI_RANGE_PMEM].end |=
468		    (uint64_t)reshigh << 32;
469	}
470	if (pd->pd_bridge.ranges[PCI_RANGE_PMEM].start >=
471	    pd->pd_bridge.ranges[PCI_RANGE_PMEM].end) {
472		pd->pd_bridge.ranges[PCI_RANGE_PMEM].start = 0;
473		pd->pd_bridge.ranges[PCI_RANGE_PMEM].end = 0;
474	}
475}
476
477/*
478 * pci_resource_scan_device --
479 *
480 *   Determine the current configuration of a PCI device.
481 */
482static bool
483pci_resource_scan_device(struct pci_resources *pr,
484    struct pci_bus *parent_bus, uint8_t devno, uint8_t funcno)
485{
486	struct pci_device *pd;
487	pcitag_t tag;
488	pcireg_t id, bridge_bus;
489	uint8_t sec_bus;
490
491	tag = pci_make_tag(pr->pr_pc, parent_bus->pb_busno, devno, funcno);
492	id = pci_conf_read(pr->pr_pc, tag, PCI_ID_REG);
493	if (PCI_VENDOR(id) == PCI_VENDOR_INVALID) {
494		return false;
495	}
496
497	pd = PCICONF_BUS_DEVICE(parent_bus, devno, funcno);
498	pd->pd_present = true;
499	pd->pd_bus = parent_bus;
500	pd->pd_tag = tag;
501	pd->pd_devno = devno;
502	pd->pd_funcno = funcno;
503	pd->pd_id = id;
504	pd->pd_class = pci_conf_read(pr->pr_pc, tag, PCI_CLASS_REG);
505	pd->pd_bhlc = pci_conf_read(pr->pr_pc, tag, PCI_BHLC_REG);
506
507	switch (PCI_HDRTYPE_TYPE(pd->pd_bhlc)) {
508	case PCI_HDRTYPE_DEVICE:
509		pci_resource_scan_bar(pr, pd, PCI_MAPREG_START,
510		    PCI_MAPREG_END, false);
511		break;
512	case PCI_HDRTYPE_PPB:
513		pci_resource_scan_bar(pr, pd, PCI_MAPREG_START,
514		    PCI_MAPREG_PPB_END, true);
515		pci_resource_scan_bridge(pr, pd);
516		break;
517	}
518
519	pci_resource_device_print(pr, pd);
520
521	if (PCI_HDRTYPE_TYPE(pd->pd_bhlc) == PCI_HDRTYPE_PPB &&
522	    PCI_CLASS(pd->pd_class) == PCI_CLASS_BRIDGE &&
523	    PCI_SUBCLASS(pd->pd_class) == PCI_SUBCLASS_BRIDGE_PCI) {
524		bridge_bus = pci_conf_read(pr->pr_pc, tag, PCI_BRIDGE_BUS_REG);
525		sec_bus = PCI_BRIDGE_BUS_NUM_SECONDARY(bridge_bus);
526		if (sec_bus <= pr->pr_endbus) {
527			pci_resource_scan_bus(pr, pd, sec_bus);
528		}
529	}
530
531	return true;
532}
533
534/*
535 * pci_resource_scan_bus --
536 *
537 *   Enumerate devices on a bus, recursively.
538 */
539static void
540pci_resource_scan_bus(struct pci_resources *pr,
541    struct pci_device *bridge_dev, uint8_t busno)
542{
543	struct pci_bus *pb;
544	uint8_t devno, funcno;
545	uint8_t nfunc;
546
547	KASSERT(busno >= pr->pr_startbus);
548	KASSERT(busno <= pr->pr_endbus);
549
550	if (PCICONF_RES_BUS(pr, busno) != NULL) {
551		/*
552		 * Firmware has configured more than one bridge with the
553		 * same secondary bus number.
554		 */
555		panic("Bus %u already scanned (firmware bug!)", busno);
556		return;
557	}
558
559	pb = pci_new_bus(pr, busno, bridge_dev);
560	PCICONF_RES_BUS(pr, busno) = pb;
561
562	for (devno = 0; devno < PCI_MAX_DEVICE; devno++) {
563		if (!pci_resource_scan_device(pr, pb, devno, 0)) {
564			continue;
565		}
566		pb->pb_lastdevno = devno;
567
568		nfunc = pci_resource_device_functions(pr, busno, devno);
569		for (funcno = 1; funcno < nfunc; funcno++) {
570			pci_resource_scan_device(pr, pb, devno, funcno);
571		}
572	}
573}
574
575/*
576 * pci_resource_claim --
577 *
578 *   Claim a resource from a vmem arena. This is called to inform the
579 *   resource manager about resources already configured by system firmware.
580 */
581static int
582pci_resource_claim(vmem_t *arena, vmem_addr_t start, vmem_addr_t end)
583{
584	KASSERT(end >= start);
585
586	return vmem_xalloc(arena, end - start + 1, 0, 0, 0, start, end,
587	    VM_BESTFIT | VM_NOSLEEP, NULL);
588}
589
590/*
591 * pci_resource_alloc --
592 *
593 *   Allocate a resource from a vmem arena. This is called when configuring
594 *   devices that were not already configured by system firmware.
595 */
596static int
597pci_resource_alloc(vmem_t *arena, vmem_size_t size, vmem_size_t align,
598    uint64_t *base)
599{
600	vmem_addr_t addr;
601	int error;
602
603	KASSERT(size != 0);
604
605	error = vmem_xalloc(arena, size, align, 0, 0, VMEM_ADDR_MIN,
606	    VMEM_ADDR_MAX, VM_BESTFIT | VM_NOSLEEP, &addr);
607	if (error == 0) {
608		*base = (uint64_t)addr;
609	}
610
611	return error;
612}
613
614/*
615 * pci_resource_init_device --
616 *
617 *   Discover resources assigned by system firmware, notify the resource
618 *   manager of these ranges, and determine if the device has additional
619 *   resources that need to be allocated.
620 */
621static void
622pci_resource_init_device(struct pci_resources *pr,
623    struct pci_device *pd)
624{
625	struct pci_iores *pi;
626	struct pci_bus *pb = pd->pd_bus;
627	vmem_t *res_io = pb->pb_res[PCI_RANGE_IO];
628	vmem_t *res_mem = pb->pb_res[PCI_RANGE_MEM];
629	vmem_t *res_pmem = pb->pb_res[PCI_RANGE_PMEM];
630	pcireg_t cmd;
631	u_int enabled, required;
632	u_int iores;
633	int error;
634
635	KASSERT(pd->pd_present);
636
637	if (IS_TEST_DEVICE(pd)) {
638		cmd = pci_conf_read(pr->pr_pc, pd->pd_tag,
639		    PCI_COMMAND_STATUS_REG);
640		cmd &= ~(PCI_COMMAND_MEM_ENABLE|PCI_COMMAND_IO_ENABLE|
641			 PCI_COMMAND_MASTER_ENABLE);
642		pci_conf_write(pr->pr_pc, pd->pd_tag, PCI_COMMAND_STATUS_REG,
643		    cmd);
644	}
645
646	enabled = required = 0;
647	cmd = pci_conf_read(pr->pr_pc, pd->pd_tag, PCI_COMMAND_STATUS_REG);
648	if ((cmd & PCI_COMMAND_MEM_ENABLE) != 0) {
649		enabled |= __BIT(PCI_MAPREG_TYPE_MEM);
650	}
651	if ((cmd & PCI_COMMAND_IO_ENABLE) != 0) {
652		enabled |= __BIT(PCI_MAPREG_TYPE_IO);
653	}
654
655	for (iores = 0; iores < pd->pd_niores; iores++) {
656		pi = &pd->pd_iores[iores];
657
658		required |= __BIT(pi->pi_type);
659
660		if (IS_TEST_DEVICE(pd)) {
661			pci_conf_write(pr->pr_pc, pd->pd_tag,
662			    PCI_BAR(pi->pi_bar), 0);
663			continue;
664		}
665		if ((enabled & __BIT(pi->pi_type)) == 0) {
666			continue;
667		}
668
669		if (pi->pi_type == PCI_MAPREG_TYPE_IO) {
670			error = res_io == NULL ? ERANGE :
671			    pci_resource_claim(res_io, pi->pi_base,
672				pi->pi_base + pi->pi_size - 1);
673			if (error) {
674				DPRINT("PCI: " PCI_SBDF_FMT " [device] io "
675				       " %#" PRIx64 "-%#" PRIx64
676				       " invalid (%d)\n",
677				       PCI_SBDF_FMT_ARGS(pr, pd),
678				       pi->pi_base,
679				       pi->pi_base + pi->pi_size - 1,
680				       error);
681			}
682			continue;
683		}
684
685		KASSERT(pi->pi_type == PCI_MAPREG_TYPE_MEM);
686		error = ERANGE;
687		if (pi->pi_mem.prefetch) {
688			/*
689			 * Prefetchable memory must be allocated from the
690			 * bridge's prefetchable region.
691			 */
692			if (res_pmem != NULL) {
693				error = pci_resource_claim(res_pmem, pi->pi_base,
694				    pi->pi_base + pi->pi_size - 1);
695			}
696		} else if (pi->pi_mem.memtype == PCI_MAPREG_MEM_TYPE_64BIT) {
697			/*
698			 * Non-prefetchable 64-bit memory can be allocated from
699			 * any range. Prefer allocations from the prefetchable
700			 * region to save 32-bit only resources for 32-bit BARs.
701			 */
702			if (res_pmem != NULL) {
703				error = pci_resource_claim(res_pmem, pi->pi_base,
704				    pi->pi_base + pi->pi_size - 1);
705			}
706			if (error && res_mem != NULL) {
707				error = pci_resource_claim(res_mem, pi->pi_base,
708				    pi->pi_base + pi->pi_size - 1);
709			}
710		} else {
711			/*
712			 * Non-prefetchable 32-bit memory can be allocated from
713			 * any range, provided that the range is below 4GB. Try
714			 * the non-prefetchable range first, and if that fails,
715			 * make one last attempt at allocating from the
716			 * prefetchable range in case the platform provides
717			 * memory below 4GB.
718			 */
719			if (res_mem != NULL) {
720				error = pci_resource_claim(res_mem, pi->pi_base,
721				    pi->pi_base + pi->pi_size - 1);
722			}
723			if (error && res_pmem != NULL) {
724				error = pci_resource_claim(res_pmem, pi->pi_base,
725				    pi->pi_base + pi->pi_size - 1);
726			}
727		}
728		if (error) {
729			DPRINT("PCI: " PCI_SBDF_FMT " [device] mem"
730			       " (%sprefetchable)"
731			       " %#" PRIx64 "-%#" PRIx64
732			       " invalid (%d)\n",
733			       PCI_SBDF_FMT_ARGS(pr, pd),
734			       pi->pi_mem.prefetch ? "" : "non-",
735			       pi->pi_base,
736			       pi->pi_base + pi->pi_size - 1,
737			       error);
738		}
739	}
740
741	pd->pd_configured = (enabled & required) == required;
742
743	if (!pd->pd_configured) {
744		DPRINT("PCI: " PCI_SBDF_FMT " [device] "
745		       "not configured by firmware\n",
746		       PCI_SBDF_FMT_ARGS(pr, pd));
747	}
748}
749
750/*
751 * pci_resource_init_bus --
752 *
753 *   Discover resources in use on a given bus, recursively.
754 */
755static void
756pci_resource_init_bus(struct pci_resources *pr, uint8_t busno)
757{
758	struct pci_bus *pb, *parent_bus;
759	struct pci_device *pd, *bridge;
760	uint8_t devno, funcno;
761	uint8_t nfunc;
762	int error;
763
764	KASSERT(busno >= pr->pr_startbus);
765	KASSERT(busno <= pr->pr_endbus);
766
767	pb = PCICONF_RES_BUS(pr, busno);
768	bridge = pb->pb_bridge;
769
770	KASSERT(pb != NULL);
771	KASSERT((busno == pr->pr_startbus) == (bridge == NULL));
772
773	if (bridge == NULL) {
774		/* Use resources provided by firmware. */
775		PCI_RANGE_FOREACH(prtype) {
776			pb->pb_res[prtype] = pr->pr_res[prtype];
777			pr->pr_res[prtype] = NULL;
778		}
779	} else {
780		/*
781		 * Using the resources configured in to the bridge by
782		 * firmware, claim the resources on the parent bus and
783		 * create a new vmem arena for the secondary bus.
784		 */
785		KASSERT(bridge->pd_bus != NULL);
786		parent_bus = bridge->pd_bus;
787		PCI_RANGE_FOREACH(prtype) {
788			if (parent_bus->pb_res[prtype] == NULL ||
789			    !bridge->pd_bridge.ranges[prtype].end) {
790				continue;
791			}
792			error = pci_resource_claim(
793			    parent_bus->pb_res[prtype],
794			    bridge->pd_bridge.ranges[prtype].start,
795			    bridge->pd_bridge.ranges[prtype].end);
796			if (error == 0) {
797				pb->pb_res[prtype] = pci_create_vmem(
798				    pci_resource_typename(prtype),
799				    bridge->pd_bridge.ranges[prtype].start,
800				    bridge->pd_bridge.ranges[prtype].end);
801				KASSERT(pb->pb_res[prtype] != NULL);
802			} else {
803				DPRINT("PCI: " PCI_SBDF_FMT " bridge (bus %u)"
804				       " %-4s %#" PRIx64 "-%#" PRIx64
805				       " invalid\n",
806				       PCI_SBDF_FMT_ARGS(pr, bridge), busno,
807				       pci_resource_typename(prtype),
808				       bridge->pd_bridge.ranges[prtype].start,
809				       bridge->pd_bridge.ranges[prtype].end);
810			}
811		}
812	}
813
814	for (devno = 0; devno <= pb->pb_lastdevno; devno++) {
815		KASSERT(devno < PCI_MAX_DEVICE);
816		nfunc = pci_resource_device_functions(pr, busno, devno);
817		for (funcno = 0; funcno < nfunc; funcno++) {
818			pd = PCICONF_BUS_DEVICE(pb, devno, funcno);
819			if (!pd->pd_present) {
820				continue;
821			}
822			if (pd->pd_ppb) {
823				uint8_t sec_bus = PCI_BRIDGE_BUS_NUM_SECONDARY(
824				    pd->pd_bridge.bridge_bus);
825				pci_resource_init_bus(pr, sec_bus);
826			}
827			pci_resource_init_device(pr, pd);
828		}
829	}
830}
831
832/*
833 * pci_resource_probe --
834 *
835 *   Scan for PCI devices and initialize the resource manager.
836 */
837static void
838pci_resource_probe(struct pci_resources *pr,
839    const struct pci_resource_info *info)
840{
841	uint8_t startbus = (uint8_t)info->ranges[PCI_RANGE_BUS].start;
842	uint8_t endbus = (uint8_t)info->ranges[PCI_RANGE_BUS].end;
843	u_int nbus;
844
845	KASSERT(startbus <= endbus);
846	KASSERT(pr->pr_bus == NULL);
847
848	nbus = endbus - startbus + 1;
849
850	pr->pr_pc = info->pc;
851	pr->pr_startbus = startbus;
852	pr->pr_endbus = endbus;
853	pr->pr_bus = kmem_zalloc(nbus * sizeof(struct pci_bus *), KM_SLEEP);
854	memcpy(pr->pr_ranges, info->ranges, sizeof(pr->pr_ranges));
855	PCI_RANGE_FOREACH(prtype) {
856		if (prtype == PCI_RANGE_BUS || info->ranges[prtype].end) {
857			pr->pr_res[prtype] = pci_create_vmem(
858			    pci_resource_typename(prtype),
859			    info->ranges[prtype].start,
860			    info->ranges[prtype].end);
861			KASSERT(pr->pr_res[prtype] != NULL);
862		}
863	}
864
865	/* Scan devices */
866	pci_resource_scan_bus(pr, NULL, pr->pr_startbus);
867
868	/*
869	 * Create per-bus resource pools and remove ranges that are already
870	 * in use by devices and downstream bridges.
871	 */
872	pci_resource_init_bus(pr, pr->pr_startbus);
873}
874
875/*
876 * pci_resource_alloc_device --
877 *
878 *   Attempt to allocate resources for a given device.
879 */
880static void
881pci_resource_alloc_device(struct pci_resources *pr, struct pci_device *pd)
882{
883	struct pci_iores *pi;
884	vmem_t *arena;
885	pcireg_t cmd, ocmd, base;
886	uint64_t addr;
887	u_int enabled;
888	u_int res;
889	u_int align;
890	int error;
891
892	enabled = 0;
893	ocmd = cmd = pci_conf_read(pr->pr_pc, pd->pd_tag,
894	    PCI_COMMAND_STATUS_REG);
895	if ((cmd & PCI_COMMAND_MEM_ENABLE) != 0) {
896		enabled |= __BIT(PCI_MAPREG_TYPE_MEM);
897	}
898	if ((cmd & PCI_COMMAND_IO_ENABLE) != 0) {
899		enabled |= __BIT(PCI_MAPREG_TYPE_IO);
900	}
901
902	for (res = 0; res < pd->pd_niores; res++) {
903		pi = &pd->pd_iores[res];
904
905		if ((enabled & __BIT(pi->pi_type)) != 0) {
906			continue;
907		}
908
909		if (pi->pi_type == PCI_MAPREG_TYPE_IO) {
910			arena = pd->pd_bus->pb_res[PCI_RANGE_IO];
911			align = uimax(pi->pi_size, 4);
912		} else {
913			KASSERT(pi->pi_type == PCI_MAPREG_TYPE_MEM);
914			arena = NULL;
915			align = uimax(pi->pi_size, 16);
916			if (pi->pi_mem.prefetch) {
917				arena = pd->pd_bus->pb_res[PCI_RANGE_PMEM];
918			}
919			if (arena == NULL) {
920				arena = pd->pd_bus->pb_res[PCI_RANGE_MEM];
921			}
922		}
923		if (arena == NULL) {
924			DPRINT("PCI: " PCI_SBDF_FMT " BAR%u failed to"
925			       " allocate %#" PRIx64 " bytes (no arena)\n",
926			       PCI_SBDF_FMT_ARGS(pr, pd),
927			       pi->pi_bar, pi->pi_size);
928			return;
929		}
930		error = pci_resource_alloc(arena, pi->pi_size, align, &addr);
931		if (error != 0) {
932			DPRINT("PCI: " PCI_SBDF_FMT " BAR%u failed to"
933			       " allocate %#" PRIx64 " bytes (no space)\n",
934			       PCI_SBDF_FMT_ARGS(pr, pd),
935			       pi->pi_bar, pi->pi_size);
936			return;
937		}
938		DPRINT("PCI: " PCI_SBDF_FMT " BAR%u assigned range"
939		       " 0x%#" PRIx64 "-0x%#" PRIx64 "\n",
940		       PCI_SBDF_FMT_ARGS(pr, pd),
941		       pi->pi_bar, addr, addr + pi->pi_size - 1);
942
943		if (pi->pi_type == PCI_MAPREG_TYPE_IO) {
944			cmd |= PCI_COMMAND_IO_ENABLE;
945			pci_conf_write(pr->pr_pc, pd->pd_tag,
946			    PCI_BAR(pi->pi_bar),
947			    PCI_MAPREG_IO_ADDR(addr) | PCI_MAPREG_TYPE_IO);
948		} else {
949			cmd |= PCI_COMMAND_MEM_ENABLE;
950			base = pci_conf_read(pr->pr_pc, pd->pd_tag,
951			    PCI_BAR(pi->pi_bar));
952			base = PCI_MAPREG_MEM_ADDR(addr) |
953			    PCI_MAPREG_MEM_TYPE(base);
954			pci_conf_write(pr->pr_pc, pd->pd_tag,
955			    PCI_BAR(pi->pi_bar), base);
956			if (pi->pi_mem.memtype == PCI_MAPREG_MEM_TYPE_64BIT) {
957				base = (pcireg_t)
958				    (PCI_MAPREG_MEM64_ADDR(addr) >> 32);
959				pci_conf_write(pr->pr_pc, pd->pd_tag,
960				    PCI_BAR(pi->pi_bar + 1), base);
961			}
962		}
963	}
964
965	if (ocmd != cmd) {
966		pci_conf_write(pr->pr_pc, pd->pd_tag,
967		    PCI_COMMAND_STATUS_REG, cmd);
968	}
969}
970
971/*
972 * pci_resource_alloc_bus --
973 *
974 *   Attempt to assign resources to all devices on a given bus, recursively.
975 */
976static void
977pci_resource_alloc_bus(struct pci_resources *pr, uint8_t busno)
978{
979	struct pci_bus *pb = PCICONF_RES_BUS(pr, busno);
980	struct pci_device *pd;
981	uint8_t devno, funcno;
982
983	for (devno = 0; devno <= pb->pb_lastdevno; devno++) {
984		for (funcno = 0; funcno < 8; funcno++) {
985			pd = PCICONF_BUS_DEVICE(pb, devno, funcno);
986			if (!pd->pd_present) {
987				if (funcno == 0) {
988					break;
989				}
990				continue;
991			}
992			if (!pd->pd_configured) {
993				pci_resource_alloc_device(pr, pd);
994			}
995			if (pd->pd_ppb) {
996				uint8_t sec_bus = PCI_BRIDGE_BUS_NUM_SECONDARY(
997				    pd->pd_bridge.bridge_bus);
998				pci_resource_alloc_bus(pr, sec_bus);
999			}
1000		}
1001	}
1002}
1003
1004/*
1005 * pci_resource_init --
1006 *
1007 *   Public interface to PCI resource manager. Scans for available devices
1008 *   and assigns resources.
1009 */
1010void
1011pci_resource_init(const struct pci_resource_info *info)
1012{
1013	struct pci_resources pr = {};
1014
1015	pci_resource_probe(&pr, info);
1016	pci_resource_alloc_bus(&pr, pr.pr_startbus);
1017}
1018
1019/*
1020 * pci_resource_typename --
1021 *
1022 *   Return a string description of a PCI range type.
1023 */
1024const char *
1025pci_resource_typename(enum pci_range_type prtype)
1026{
1027	KASSERT(prtype < NUM_PCI_RANGES);
1028	return pci_range_typenames[prtype];
1029}
1030