xref: /netbsd/sys/arch/shark/ofw/ofw.c (revision bf9ec67e)
1 /*	$NetBSD: ofw.c,v 1.13 2002/04/05 16:58:11 thorpej Exp $	*/
2 
3 /*
4  * Copyright 1997
5  * Digital Equipment Corporation. All rights reserved.
6  *
7  * This software is furnished under license and may be used and
8  * copied only in accordance with the following terms and conditions.
9  * Subject to these conditions, you may download, copy, install,
10  * use, modify and distribute this software in source and/or binary
11  * form. No title or ownership is transferred hereby.
12  *
13  * 1) Any source code used, modified or distributed must reproduce
14  *    and retain this copyright notice and list of conditions as
15  *    they appear in the source file.
16  *
17  * 2) No right is granted to use any trade name, trademark, or logo of
18  *    Digital Equipment Corporation. Neither the "Digital Equipment
19  *    Corporation" name nor any trademark or logo of Digital Equipment
20  *    Corporation may be used to endorse or promote products derived
21  *    from this software without the prior written permission of
22  *    Digital Equipment Corporation.
23  *
24  * 3) This software is provided "AS-IS" and any express or implied
25  *    warranties, including but not limited to, any implied warranties
26  *    of merchantability, fitness for a particular purpose, or
27  *    non-infringement are disclaimed. In no event shall DIGITAL be
28  *    liable for any damages whatsoever, and in particular, DIGITAL
29  *    shall not be liable for special, indirect, consequential, or
30  *    incidental damages or damages for lost profits, loss of
31  *    revenue or loss of use, whether such damages arise in contract,
32  *    negligence, tort, under statute, in equity, at law or otherwise,
33  *    even if advised of the possibility of such damage.
34  */
35 
36 /*
37  *  Routines for interfacing between NetBSD and OFW.
38  *
39  *  Parts of this could be moved to an MI file in time. -JJK
40  *
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/reboot.h>
47 #include <sys/mbuf.h>
48 
49 #include <uvm/uvm_extern.h>
50 
51 #include <dev/cons.h>
52 
53 #include <machine/bus.h>
54 #include <machine/frame.h>
55 #include <machine/bootconfig.h>
56 #include <machine/cpu.h>
57 #include <machine/intr.h>
58 
59 #include <dev/ofw/openfirm.h>
60 #include <machine/ofw.h>
61 
62 #include <netinet/in.h>
63 
64 #if	BOOT_FW_DHCP
65 #include <nfs/bootdata.h>
66 #endif
67 
68 #ifdef SHARK
69 #include "machine/pio.h"
70 #include "machine/isa_machdep.h"
71 #endif
72 
73 #include "pc.h"
74 #include "isadma.h"
75 
76 #define IO_VIRT_BASE (OFW_VIRT_BASE + OFW_VIRT_SIZE)
77 #define IO_VIRT_SIZE 0x01000000
78 
79 #define	KERNEL_IMG_PTS		2
80 #define	KERNEL_VMDATA_PTS	(KERNEL_VM_SIZE >> (L1_S_SHIFT + 2))
81 #define	KERNEL_OFW_PTS		4
82 #define	KERNEL_IO_PTS		4
83 
84 /*
85  *  Imported variables
86  */
87 extern BootConfig bootconfig;	/* temporary, I hope */
88 
89 #ifdef	DIAGNOSTIC
90 /* NOTE: These variables will be removed, well some of them */
91 extern u_int spl_mask;
92 extern u_int current_mask;
93 #endif
94 
95 extern int ofw_handleticks;
96 
97 
98 /*
99  *  Imported routines
100  */
101 extern void dump_spl_masks  __P((void));
102 extern void dumpsys	    __P((void));
103 extern void dotickgrovelling __P((vm_offset_t));
104 #if defined(SHARK) && (NPC > 0)
105 extern void shark_screen_cleanup __P((int));
106 #endif
107 
108 #define WriteWord(a, b) \
109 *((volatile unsigned int *)(a)) = (b)
110 
111 #define ReadWord(a) \
112 (*((volatile unsigned int *)(a)))
113 
114 
115 /*
116  *  Exported variables
117  */
118 /* These should all be in a meminfo structure. */
119 vm_offset_t physical_start;
120 vm_offset_t physical_freestart;
121 vm_offset_t physical_freeend;
122 vm_offset_t physical_end;
123 u_int free_pages;
124 int physmem;
125 pv_addr_t systempage;
126 #ifndef	OFWGENCFG
127 pv_addr_t irqstack;
128 #endif
129 pv_addr_t undstack;
130 pv_addr_t abtstack;
131 pv_addr_t kernelstack;
132 
133 vm_offset_t msgbufphys;
134 
135 /* for storage allocation, used to be local to ofw_construct_proc0_addrspace */
136 static vm_offset_t  virt_freeptr;
137 
138 int ofw_callbacks = 0;		/* debugging counter */
139 
140 /**************************************************************/
141 
142 
143 /*
144  *  Declarations and definitions private to this module
145  *
146  */
147 
148 struct mem_region {
149 	vm_offset_t start;
150 	vm_size_t size;
151 };
152 
153 struct mem_translation {
154 	vm_offset_t virt;
155 	vm_size_t size;
156 	vm_offset_t phys;
157 	unsigned int mode;
158 };
159 
160 struct isa_range {
161 	vm_offset_t isa_phys_hi;
162 	vm_offset_t isa_phys_lo;
163 	vm_offset_t parent_phys_start;
164 	vm_size_t   isa_size;
165 };
166 
167 struct vl_range {
168 	vm_offset_t vl_phys_hi;
169 	vm_offset_t vl_phys_lo;
170 	vm_offset_t parent_phys_start;
171 	vm_size_t   vl_size;
172 };
173 
174 struct vl_isa_range {
175 	vm_offset_t isa_phys_hi;
176 	vm_offset_t isa_phys_lo;
177 	vm_offset_t parent_phys_hi;
178 	vm_offset_t parent_phys_lo;
179 	vm_size_t   isa_size;
180 };
181 
182 struct dma_range {
183 	vm_offset_t start;
184 	vm_size_t   size;
185 };
186 
187 struct ofw_cbargs {
188 	char *name;
189 	int nargs;
190 	int nreturns;
191 	int args_n_results[12];
192 };
193 
194 
195 /* Memory info */
196 static int nOFphysmem;
197 static struct mem_region *OFphysmem;
198 static int nOFphysavail;
199 static struct mem_region *OFphysavail;
200 static int nOFtranslations;
201 static struct mem_translation *OFtranslations;
202 static int nOFdmaranges;
203 static struct dma_range *OFdmaranges;
204 
205 /* The OFW client services handle. */
206 /* Initialized by ofw_init(). */
207 static ofw_handle_t ofw_client_services_handle;
208 
209 
210 static void ofw_callbackhandler __P((struct ofw_cbargs *));
211 static void ofw_construct_proc0_addrspace __P((pv_addr_t *, pv_addr_t *));
212 static void ofw_getphysmeminfo __P((void));
213 static void ofw_getvirttranslations __P((void));
214 static void *ofw_malloc(vm_size_t size);
215 static void ofw_claimpages __P((vm_offset_t *, pv_addr_t *, vm_size_t));
216 static void ofw_discardmappings __P ((vm_offset_t, vm_offset_t, vm_size_t));
217 static int ofw_mem_ihandle  __P((void));
218 static int ofw_mmu_ihandle  __P((void));
219 static vm_offset_t ofw_claimphys __P((vm_offset_t, vm_size_t, vm_offset_t));
220 #if 0
221 static vm_offset_t ofw_releasephys __P((vm_offset_t, vm_size_t));
222 #endif
223 static vm_offset_t ofw_claimvirt __P((vm_offset_t, vm_size_t, vm_offset_t));
224 static void ofw_settranslation __P ((vm_offset_t, vm_offset_t, vm_size_t, int));
225 static void ofw_initallocator __P((void));
226 static void ofw_configisaonly __P((vm_offset_t *, vm_offset_t *));
227 static void ofw_configvl __P((int, vm_offset_t *, vm_offset_t *));
228 static vm_offset_t ofw_valloc __P((vm_offset_t, vm_offset_t));
229 
230 
231 /*
232  * DHCP hooks.  For a first cut, we look to see if there is a DHCP
233  * packet that was saved by the firmware.  If not, we proceed as before,
234  * getting hand-configured data from NVRAM.  If there is one, we get the
235  * packet, and extract the data from it.  For now, we hand that data up
236  * in the boot_args string as before.
237  */
238 
239 
240 /**************************************************************/
241 
242 
243 /*
244  *
245  *  Support routines for xxx_machdep.c
246  *
247  *  The intent is that all OFW-based configurations use the
248  *  exported routines in this file to do their business.  If
249  *  they need to override some function they are free to do so.
250  *
251  *  The exported routines are:
252  *
253  *    openfirmware
254  *    ofw_init
255  *    ofw_boot
256  *    ofw_getbootinfo
257  *    ofw_configmem
258  *    ofw_configisa
259  *    ofw_configisadma
260  *    ofw_gettranslation
261  *    ofw_map
262  *    ofw_getcleaninfo
263  */
264 
265 
266 int
267 openfirmware(args)
268 	void *args;
269 {
270 	int ofw_result;
271 	u_int saved_irq_state;
272 
273 	/* OFW is not re-entrant, so we wrap a mutex around the call. */
274 	saved_irq_state = disable_interrupts(I32_bit);
275 	ofw_result = ofw_client_services_handle(args);
276 	(void)restore_interrupts(saved_irq_state);
277 
278 	return(ofw_result);
279 }
280 
281 
282 void
283 ofw_init(ofw_handle)
284 	ofw_handle_t ofw_handle;
285 {
286 	ofw_client_services_handle = ofw_handle;
287 
288 	/*  Everything we allocate in the remainder of this block is
289 	 *  constrained to be in the "kernel-static" portion of the
290 	 *  virtual address space (i.e., 0xF0000000 - 0xF1000000).
291 	 *  This is because all such objects are expected to be in
292 	 *  that range by NetBSD, or the objects will be re-mapped
293 	 *  after the page-table-switch to other specific locations.
294 	 *  In the latter case, it's simplest if our pre-switch handles
295 	 *  on those objects are in regions that are already "well-
296 	 *  known."  (Otherwise, the cloning of the OFW-managed address-
297 	 *  space becomes more awkward.)  To minimize the number of L2
298 	 *  page tables that we use, we are further restricting the
299 	 *  remaining allocations in this block to the bottom quarter of
300 	 *  the legal range.  OFW will have loaded the kernel text+data+bss
301 	 *  starting at the bottom of the range, and we will allocate
302 	 *  objects from the top, moving downwards.  The two sub-regions
303 	 *  will collide if their total sizes hit 8MB.  The current total
304 	 *  is <1.5MB, but INSTALL kernels are > 4MB, so hence the 8MB
305 	 *  limit.  The variable virt-freeptr represents the next free va
306 	 *  (moving downwards).
307 	 */
308 	virt_freeptr = KERNEL_BASE + (0x00400000 * KERNEL_IMG_PTS);
309 }
310 
311 
312 void
313 ofw_boot(howto, bootstr)
314 	int howto;
315 	char *bootstr;
316 {
317 
318 #ifdef DIAGNOSTIC
319 	printf("boot: howto=%08x curproc=%p\n", howto, curproc);
320 	printf("current_mask=%08x spl_mask=%08x\n", current_mask, spl_mask);
321 
322 	printf("ipl_bio=%08x ipl_net=%08x ipl_tty=%08x ipl_imp=%08x\n",
323 	    irqmasks[IPL_BIO], irqmasks[IPL_NET], irqmasks[IPL_TTY],
324 	    irqmasks[IPL_IMP]);
325 	printf("ipl_audio=%08x ipl_clock=%08x ipl_none=%08x\n",
326 	    irqmasks[IPL_AUDIO], irqmasks[IPL_CLOCK], irqmasks[IPL_NONE]);
327 
328 	dump_spl_masks();
329 #endif
330 
331 	/*
332 	 * If we are still cold then hit the air brakes
333 	 * and crash to earth fast
334 	 */
335 	if (cold) {
336 		doshutdownhooks();
337 		printf("Halted while still in the ICE age.\n");
338 		printf("The operating system has halted.\n");
339 		goto ofw_exit;
340 		/*NOTREACHED*/
341 	}
342 
343 	/*
344 	 * If RB_NOSYNC was not specified sync the discs.
345 	 * Note: Unless cold is set to 1 here, syslogd will die during the unmount.
346 	 * It looks like syslogd is getting woken up only to find that it cannot
347 	 * page part of the binary in as the filesystem has been unmounted.
348 	 */
349 	if (!(howto & RB_NOSYNC))
350 		bootsync();
351 
352 	/* Say NO to interrupts */
353 	splhigh();
354 
355 	/* Do a dump if requested. */
356 	if ((howto & (RB_DUMP | RB_HALT)) == RB_DUMP)
357 		dumpsys();
358 
359 	/* Run any shutdown hooks */
360 	doshutdownhooks();
361 
362 	/* Make sure IRQ's are disabled */
363 	IRQdisable;
364 
365 	if (howto & RB_HALT) {
366 		printf("The operating system has halted.\n");
367 		goto ofw_exit;
368 	}
369 
370 	/* Tell the user we are booting */
371 	printf("rebooting...\n");
372 
373 	/* Jump into the OFW boot routine. */
374 	{
375 		static char str[256];
376 		char *ap = str, *ap1 = ap;
377 
378 		if (bootstr && *bootstr) {
379 			if (strlen(bootstr) > sizeof str - 5)
380 				printf("boot string too large, ignored\n");
381 			else {
382 				strcpy(str, bootstr);
383 				ap1 = ap = str + strlen(str);
384 				*ap++ = ' ';
385 			}
386 		}
387 		*ap++ = '-';
388 		if (howto & RB_SINGLE)
389 			*ap++ = 's';
390 		if (howto & RB_KDB)
391 			*ap++ = 'd';
392 		*ap++ = 0;
393 		if (ap[-2] == '-')
394 			*ap1 = 0;
395 #if defined(SHARK) && (NPC > 0)
396 		shark_screen_cleanup(0);
397 #endif
398 		OF_boot(str);
399 		/*NOTREACHED*/
400 	}
401 
402 ofw_exit:
403 	printf("Calling OF_exit...\n");
404 #if defined(SHARK) && (NPC > 0)
405 	shark_screen_cleanup(1);
406 #endif
407 	OF_exit();
408 	/*NOTREACHED*/
409 }
410 
411 
412 #if	BOOT_FW_DHCP
413 
414 extern	char	*ip2dotted	__P((struct in_addr));
415 
416 /*
417  * Get DHCP data from OFW
418  */
419 
420 void
421 get_fw_dhcp_data(bdp)
422 	struct bootdata *bdp;
423 {
424 	int chosen;
425 	int dhcplen;
426 
427 	bzero((char *)bdp, sizeof(*bdp));
428 	if ((chosen = OF_finddevice("/chosen")) == -1)
429 		panic("no /chosen from OFW");
430 	if ((dhcplen = OF_getproplen(chosen, "bootp-response")) > 0) {
431 		u_char *cp;
432 		int dhcp_type = 0;
433 		char *ip;
434 
435 		/*
436 		 * OFW saved a DHCP (or BOOTP) packet for us.
437 		 */
438 		if (dhcplen > sizeof(bdp->dhcp_packet))
439 			panic("DHCP packet too large");
440 		OF_getprop(chosen, "bootp-response", &bdp->dhcp_packet,
441 		    sizeof(bdp->dhcp_packet));
442 		SANITY(bdp->dhcp_packet.op == BOOTREPLY, "bogus DHCP packet");
443 		/*
444 		 * Collect the interesting data from DHCP into
445 		 * the bootdata structure.
446 		 */
447 		bdp->ip_address = bdp->dhcp_packet.yiaddr;
448 		ip = ip2dotted(bdp->ip_address);
449 		if (bcmp(bdp->dhcp_packet.options, DHCP_OPTIONS_COOKIE, 4) == 0)
450 			parse_dhcp_options(&bdp->dhcp_packet,
451 			    bdp->dhcp_packet.options + 4,
452 			    &bdp->dhcp_packet.options[dhcplen
453 			    - DHCP_FIXED_NON_UDP], bdp, ip);
454 		if (bdp->root_ip.s_addr == 0)
455 			bdp->root_ip = bdp->dhcp_packet.siaddr;
456 		if (bdp->swap_ip.s_addr == 0)
457 			bdp->swap_ip = bdp->dhcp_packet.siaddr;
458 	}
459 	/*
460 	 * If the DHCP packet did not contain all the necessary data,
461 	 * look in NVRAM for the missing parts.
462 	 */
463 	{
464 		int options;
465 		int proplen;
466 #define BOOTJUNKV_SIZE	256
467 		char bootjunkv[BOOTJUNKV_SIZE];	/* minimize stack usage */
468 
469 
470 		if ((options = OF_finddevice("/options")) == -1)
471 			panic("can't find /options");
472 		if (bdp->ip_address.s_addr == 0 &&
473 		    (proplen = OF_getprop(options, "ipaddr",
474 		    bootjunkv, BOOTJUNKV_SIZE - 1)) > 0) {
475 			bootjunkv[proplen] = '\0';
476 			if (dotted2ip(bootjunkv, &bdp->ip_address.s_addr) == 0)
477 				bdp->ip_address.s_addr = 0;
478 		}
479 		if (bdp->ip_mask.s_addr == 0 &&
480 		    (proplen = OF_getprop(options, "netmask",
481 		    bootjunkv, BOOTJUNKV_SIZE - 1)) > 0) {
482 			bootjunkv[proplen] = '\0';
483 			if (dotted2ip(bootjunkv, &bdp->ip_mask.s_addr) == 0)
484 				bdp->ip_mask.s_addr = 0;
485 		}
486 		if (bdp->hostname[0] == '\0' &&
487 		    (proplen = OF_getprop(options, "hostname",
488 		    bdp->hostname, sizeof(bdp->hostname) - 1)) > 0) {
489 			bdp->hostname[proplen] = '\0';
490 		}
491 		if (bdp->root[0] == '\0' &&
492 		    (proplen = OF_getprop(options, "rootfs",
493 		    bootjunkv, BOOTJUNKV_SIZE - 1)) > 0) {
494 			bootjunkv[proplen] = '\0';
495 			parse_server_path(bootjunkv, &bdp->root_ip, bdp->root);
496 		}
497 		if (bdp->swap[0] == '\0' &&
498 		    (proplen = OF_getprop(options, "swapfs",
499 		    bootjunkv, BOOTJUNKV_SIZE - 1)) > 0) {
500 			bootjunkv[proplen] = '\0';
501 			parse_server_path(bootjunkv, &bdp->swap_ip, bdp->swap);
502 		}
503 	}
504 }
505 
506 #endif	/* BOOT_FW_DHCP */
507 
508 void
509 ofw_getbootinfo(bp_pp, ba_pp)
510 	char **bp_pp;
511 	char **ba_pp;
512 {
513 	int chosen;
514 	int bp_len;
515 	int ba_len;
516 	char *bootpathv;
517 	char *bootargsv;
518 
519 	/* Read the bootpath and bootargs out of OFW. */
520 	/* XXX is bootpath still interesting?  --emg */
521 	if ((chosen = OF_finddevice("/chosen")) == -1)
522 		panic("no /chosen from OFW");
523 	bp_len = OF_getproplen(chosen, "bootpath");
524 	ba_len = OF_getproplen(chosen, "bootargs");
525 	if (bp_len < 0 || ba_len < 0)
526 		panic("can't get boot data from OFW");
527 
528 	bootpathv = (char *)ofw_malloc(bp_len);
529 	bootargsv = (char *)ofw_malloc(ba_len);
530 
531 	if (bp_len)
532 		OF_getprop(chosen, "bootpath", bootpathv, bp_len);
533 	else
534 		bootpathv[0] = '\0';
535 
536 	if (ba_len)
537 		OF_getprop(chosen, "bootargs", bootargsv, ba_len);
538 	else
539 		bootargsv[0] = '\0';
540 
541 	*bp_pp = bootpathv;
542 	*ba_pp = bootargsv;
543 #ifdef DIAGNOSTIC
544 	printf("bootpath=<%s>, bootargs=<%s>\n", bootpathv, bootargsv);
545 #endif
546 }
547 
548 vm_offset_t
549 ofw_getcleaninfo(void)
550 {
551 	int cpu;
552 	vm_offset_t vclean, pclean;
553 
554 	if ((cpu = OF_finddevice("/cpu")) == -1)
555 		panic("no /cpu from OFW");
556 
557 	if ((OF_getprop(cpu, "d-cache-flush-address", &vclean,
558 	    sizeof(vclean))) != sizeof(vclean)) {
559 #ifdef DEBUG
560 		printf("no OFW d-cache-flush-address property\n");
561 #endif
562 		return -1;
563 	}
564 
565 	if ((pclean = ofw_gettranslation(
566 	    of_decode_int((unsigned char *)&vclean))) == -1)
567 	panic("OFW failed to translate cache flush address");
568 
569 	return pclean;
570 }
571 
572 void
573 ofw_configisa(pio, pmem)
574 	vm_offset_t *pio;
575 	vm_offset_t *pmem;
576 {
577 	int vl;
578 
579 	if ((vl = OF_finddevice("/vlbus")) == -1) /* old style OFW dev info tree */
580 		ofw_configisaonly(pio, pmem);
581 	else /* old style OFW dev info tree */
582 		ofw_configvl(vl, pio, pmem);
583 }
584 
585 static void
586 ofw_configisaonly(pio, pmem)
587 	vm_offset_t *pio;
588 	vm_offset_t *pmem;
589 {
590 	int isa;
591 	int rangeidx;
592 	int size;
593 	vm_offset_t hi, start;
594 	struct isa_range ranges[2];
595 
596 	if ((isa = OF_finddevice("/isa")) == -1)
597 	panic("OFW has no /isa device node");
598 
599 	/* expect to find two isa ranges: IO/data and memory/data */
600 	if ((size = OF_getprop(isa, "ranges", ranges, sizeof(ranges)))
601 	    != sizeof(ranges))
602 		panic("unexpected size of OFW /isa ranges property: %d", size);
603 
604 	*pio = *pmem = -1;
605 
606 	for (rangeidx = 0; rangeidx < 2; ++rangeidx) {
607 		hi    = of_decode_int((unsigned char *)
608 		    &ranges[rangeidx].isa_phys_hi);
609 		start = of_decode_int((unsigned char *)
610 		    &ranges[rangeidx].parent_phys_start);
611 
612 	if (hi & 1) { /* then I/O space */
613 		*pio = start;
614 	} else {
615 		*pmem = start;
616 	}
617 	} /* END for */
618 
619 	if ((*pio == -1) || (*pmem == -1))
620 		panic("bad OFW /isa ranges property");
621 
622 }
623 
624 static void
625 ofw_configvl(vl, pio, pmem)
626 	int vl;
627 	vm_offset_t *pio;
628 	vm_offset_t *pmem;
629 {
630 	int isa;
631 	int ir, vr;
632 	int size;
633 	vm_offset_t hi, start;
634 	struct vl_isa_range isa_ranges[2];
635 	struct vl_range     vl_ranges[2];
636 
637 	if ((isa = OF_finddevice("/vlbus/isa")) == -1)
638 		panic("OFW has no /vlbus/isa device node");
639 
640 	/* expect to find two isa ranges: IO/data and memory/data */
641 	if ((size = OF_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges)))
642 	    != sizeof(isa_ranges))
643 		panic("unexpected size of OFW /vlbus/isa ranges property: %d",
644 		     size);
645 
646 	/* expect to find two vl ranges: IO/data and memory/data */
647 	if ((size = OF_getprop(vl, "ranges", vl_ranges, sizeof(vl_ranges)))
648 	    != sizeof(vl_ranges))
649 		panic("unexpected size of OFW /vlbus ranges property: %d", size);
650 
651 	*pio = -1;
652 	*pmem = -1;
653 
654 	for (ir = 0; ir < 2; ++ir) {
655 		for (vr = 0; vr < 2; ++vr) {
656 			if ((isa_ranges[ir].parent_phys_hi
657 			    == vl_ranges[vr].vl_phys_hi) &&
658 			    (isa_ranges[ir].parent_phys_lo
659 			    == vl_ranges[vr].vl_phys_lo)) {
660 				hi    = of_decode_int((unsigned char *)
661 				    &isa_ranges[ir].isa_phys_hi);
662 				start = of_decode_int((unsigned char *)
663 				    &vl_ranges[vr].parent_phys_start);
664 
665 				if (hi & 1) { /* then I/O space */
666 					*pio = start;
667 				} else {
668 					*pmem = start;
669 				}
670 			} /* END if */
671 		} /* END for */
672 	} /* END for */
673 
674 	if ((*pio == -1) || (*pmem == -1))
675 		panic("bad OFW /isa ranges property");
676 }
677 
678 void
679 ofw_configisadma(pdma)
680 	vm_offset_t *pdma;
681 {
682 	int root;
683 	int rangeidx;
684 	int size;
685 	struct dma_range *dr;
686 #if NISADMA > 0
687 	extern bus_dma_segment_t *pmap_isa_dma_ranges;
688 	extern int pmap_isa_dma_nranges;
689 #endif
690 
691 	if ((root = OF_finddevice("/")) == -1 ||
692 	    (size = OF_getproplen(root, "dma-ranges")) <= 0 ||
693 	    (OFdmaranges = (struct dma_range *)ofw_malloc(size)) == 0 ||
694  	    OF_getprop(root, "dma-ranges", OFdmaranges, size) != size)
695 		panic("bad / dma-ranges property");
696 
697 	nOFdmaranges = size / sizeof(struct dma_range);
698 
699 #if NISADMA > 0
700 	/* Allocate storage for non-OFW representation of the range. */
701 	pmap_isa_dma_ranges = ofw_malloc(nOFdmaranges *
702 	    sizeof(bus_dma_segment_t));
703 	if (pmap_isa_dma_ranges == NULL)
704 		panic("unable to allocate pmap_isa_dma_ranges");
705 	pmap_isa_dma_nranges = nOFdmaranges;
706 #endif
707 
708 	for (rangeidx = 0, dr = OFdmaranges; rangeidx < nOFdmaranges;
709 	    ++rangeidx, ++dr) {
710 		dr->start = of_decode_int((unsigned char *)&dr->start);
711 		dr->size = of_decode_int((unsigned char *)&dr->size);
712 #if NISADMA > 0
713 		pmap_isa_dma_ranges[rangeidx].ds_addr = dr->start;
714 		pmap_isa_dma_ranges[rangeidx].ds_len  = dr->size;
715 #endif
716 	}
717 
718 #ifdef DEBUG
719 	printf("dma ranges size = %d\n", size);
720 
721 	for (rangeidx = 0; rangeidx < nOFdmaranges; ++rangeidx) {
722 		printf("%08lx %08lx\n",
723 		(u_long)OFdmaranges[rangeidx].start,
724 		(u_long)OFdmaranges[rangeidx].size);
725 	}
726 #endif
727 }
728 
729 /*
730  *  Memory configuration:
731  *
732  *  We start off running in the environment provided by OFW.
733  *  This has the MMU turned on, the kernel code and data
734  *  mapped-in at KERNEL_BASE (0xF0000000), OFW's text and
735  *  data mapped-in at OFW_VIRT_BASE (0xF7000000), and (possibly)
736  *  page0 mapped-in at 0x0.
737  *
738  *  The strategy is to set-up the address space for proc0 --
739  *  including the allocation of space for new page tables -- while
740  *  memory is still managed by OFW.  We then effectively create a
741  *  copy of the address space by dumping all of OFW's translations
742  *  and poking them into the new page tables.  We then notify OFW
743  *  that we are assuming control of memory-management by installing
744  *  our callback-handler, and switch to the NetBSD-managed page
745  *  tables with the setttb() call.
746  *
747  *  This scheme may cause some amount of memory to be wasted within
748  *  OFW as dead page tables, but it shouldn't be more than about
749  *  20-30KB.  (It's also possible that OFW will re-use the space.)
750  */
751 void
752 ofw_configmem(void)
753 {
754 	pv_addr_t proc0_ttbbase;
755 	pv_addr_t proc0_ptpt;
756 
757 	/* Set-up proc0 address space. */
758 	ofw_construct_proc0_addrspace(&proc0_ttbbase, &proc0_ptpt);
759 
760 	/*
761 	 * Get a dump of OFW's picture of physical memory.
762 	 * This is used below to initialize a load of variables used by pmap.
763 	 * We get it now rather than later because we are about to
764 	 * tell OFW to stop managing memory.
765 	 */
766 	ofw_getphysmeminfo();
767 
768 	/* We are about to take control of memory-management from OFW.
769 	 * Establish callbacks for OFW to use for its future memory needs.
770 	 * This is required for us to keep using OFW services.
771 	 */
772 
773 	/* First initialize our callback memory allocator. */
774 	ofw_initallocator();
775 
776 	OF_set_callback((void(*)())ofw_callbackhandler);
777 
778 	/* Switch to the proc0 pagetables. */
779 	setttb(proc0_ttbbase.pv_pa);
780 
781 	/* Aaaaaaaah, running in the proc0 address space! */
782 	/* I feel good... */
783 
784 	/* Set-up the various globals which describe physical memory for pmap. */
785 	{
786 		struct mem_region *mp;
787 		int totalcnt;
788 		int availcnt;
789 		int i;
790 
791 		/* physmem, physical_start, physical_end */
792 		physmem = 0;
793 		for (totalcnt = 0, mp = OFphysmem; totalcnt < nOFphysmem;
794 		    totalcnt++, mp++) {
795 #ifdef	OLDPRINTFS
796 			printf("physmem: %x, %x\n", mp->start, mp->size);
797 #endif
798 			physmem += btoc(mp->size);
799 		}
800 		physical_start = OFphysmem[0].start;
801 		mp--;
802 		physical_end = mp->start + mp->size;
803 
804 		/* free_pages, physical_freestart, physical_freeend */
805 		free_pages = 0;
806 		for (availcnt = 0, mp = OFphysavail; availcnt < nOFphysavail;
807 		    availcnt++, mp++) {
808 #ifdef	OLDPRINTFS
809 			printf("physavail: %x, %x\n", mp->start, mp->size);
810 #endif
811 			free_pages += btoc(mp->size);
812 		}
813 		physical_freestart = OFphysavail[0].start;
814 		mp--;
815 		physical_freeend = mp->start + mp->size;
816 #ifdef	OLDPRINTFS
817 		printf("pmap_bootstrap:  physmem = %x, free_pages = %x\n",
818 		    physmem, free_pages);
819 #endif
820 
821 		/*
822 		 *  This is a hack to work with the existing pmap code.
823 		 *  That code depends on a RiscPC BootConfig structure
824 		 *  containing, among other things, an array describing
825 		 *  the regions of physical memory.  So, for now, we need
826 		 *  to stuff our OFW-derived physical memory info into a
827 		 *  "fake" BootConfig structure.
828 		 *
829 		 *  An added twist is that we initialize the BootConfig
830 		 *  structure with our "available" physical memory regions
831 		 *  rather than the "total" physical memory regions.  Why?
832 		 *  Because:
833 		 *
834 		 *   (a) the VM code requires that the "free" pages it is
835 		 *       initialized with have consecutive indices.  This
836 		 *       allows it to use more efficient data structures
837 		 *       (presumably).
838 		 *   (b) the current pmap routines which report the initial
839 		 *       set of free page indices (pmap_next_page) and
840 		 *       which map addresses to indices (pmap_page_index)
841 		 *       assume that the free pages are consecutive across
842 		 *       memory region boundaries.
843 		 *
844 		 *  This means that memory which is "stolen" at startup time
845 		 *  (say, for page descriptors) MUST come from either the
846 		 *  bottom of the first region or the top of the last.
847 		 *
848 		 *  This requirement doesn't mesh well with OFW (or at least
849 		 *  our use of it).  We can get around it for the time being
850 		 *  by pretending that our "available" region array describes
851 		 *  all of our physical memory.  This may cause some important
852 		 *  information to be excluded from a dump file, but so far
853 		 *  I haven't come across any other negative effects.
854 		 *
855 		 *  In the long-run we should fix the index
856 		 *  generation/translation code in the pmap module.
857 		 */
858 
859 		if (DRAM_BLOCKS < (availcnt + 1))
860 			panic("more ofw memory regions than bootconfig blocks");
861 
862 		for (i = 0, mp = OFphysavail; i < nOFphysavail; i++, mp++) {
863 			bootconfig.dram[i].address = mp->start;
864 			bootconfig.dram[i].pages = btoc(mp->size);
865 		}
866 		bootconfig.dramblocks = availcnt;
867 	}
868 
869 	/* Initialize pmap module. */
870 	pmap_bootstrap((pd_entry_t *)proc0_ttbbase.pv_va, proc0_ptpt);
871 }
872 
873 
874 /*
875  ************************************************************
876 
877   Routines private to this module
878 
879  ************************************************************
880  */
881 
882 /* N.B.  Not supposed to call printf in callback-handler!  Could deadlock! */
883 static void
884 ofw_callbackhandler(args)
885 	struct ofw_cbargs *args;
886 {
887 	char *name = args->name;
888 	int nargs = args->nargs;
889 	int nreturns = args->nreturns;
890 	int *args_n_results = args->args_n_results;
891 
892 	ofw_callbacks++;
893 
894 #if defined(OFWGENCFG)
895 	/* Check this first, so that we don't waste IRQ time parsing. */
896 	if (strcmp(name, "tick") == 0) {
897 		vm_offset_t frame;
898 
899 		/* Check format. */
900 		if (nargs != 1 || nreturns < 1) {
901 			args_n_results[nargs] = -1;
902 			args->nreturns = 1;
903 			return;
904 		}
905 		args_n_results[nargs] =	0;	/* properly formatted request */
906 
907 		/*
908 		 *  Note that we are running in the IRQ frame, with interrupts
909 		 *  disabled.
910 		 *
911 		 *  We need to do two things here:
912 		 *    - copy a few words out of the input frame into a global
913 		 *      area, for later use by our real tick-handling code
914 		 *    - patch a few words in the frame so that when OFW returns
915 		 *      from the interrupt it will resume with our handler
916 		 *      rather than the code that was actually interrupted.
917 		 *      Our handler will resume when it finishes with the code
918 		 *      that was actually interrupted.
919 		 *
920 		 *  It's simplest to do this in assembler, since it requires
921 		 *  switching frames and grovelling about with registers.
922 		 */
923 		frame = (vm_offset_t)args_n_results[0];
924 		if (ofw_handleticks)
925 			dotickgrovelling(frame);
926 		args_n_results[nargs + 1] = frame;
927 		args->nreturns = 1;
928 	} else
929 #endif
930 
931 	if (strcmp(name, "map") == 0) {
932 		vm_offset_t va;
933 		vm_offset_t pa;
934 		vm_size_t size;
935 		int mode;
936 		int ap_bits;
937 		int dom_bits;
938 		int cb_bits;
939 
940 		/* Check format. */
941 		if (nargs != 4 || nreturns < 2) {
942 			args_n_results[nargs] = -1;
943 			args->nreturns = 1;
944 			return;
945 		}
946 		args_n_results[nargs] =	0;	/* properly formatted request */
947 
948 		pa = (vm_offset_t)args_n_results[0];
949 		va = (vm_offset_t)args_n_results[1];
950 		size = (vm_size_t)args_n_results[2];
951 		mode = args_n_results[3];
952 		ap_bits =  (mode & 0x00000C00);
953 		dom_bits = (mode & 0x000001E0);
954 		cb_bits =  (mode & 0x000000C0);
955 
956 		/* Sanity checks. */
957 		if ((va & PGOFSET) != 0 || va < OFW_VIRT_BASE ||
958 		    (va + size) > (OFW_VIRT_BASE + OFW_VIRT_SIZE) ||
959 		    (pa & PGOFSET) != 0 || (size & PGOFSET) != 0 ||
960 		    size == 0 || (dom_bits >> 5) != 0) {
961 			args_n_results[nargs + 1] = -1;
962 			args->nreturns = 1;
963 			return;
964 		}
965 
966 		/* Write-back anything stuck in the cache. */
967 		cpu_idcache_wbinv_all();
968 
969 		/* Install new mappings. */
970 		{
971 			pt_entry_t *pte = vtopte(va);
972 			int npages = size >> PGSHIFT;
973 
974 			ap_bits >>= 10;
975 			for (; npages > 0; pte++, pa += NBPG, npages--)
976 				*pte = (pa | L2_AP(ap_bits) | L2_TYPE_S | cb_bits);
977 		}
978 
979 		/* Clean out tlb. */
980 		tlb_flush();
981 
982 		args_n_results[nargs + 1] = 0;
983 		args->nreturns = 2;
984 	} else if (strcmp(name, "unmap") == 0) {
985 		vm_offset_t va;
986 		vm_size_t size;
987 
988 		/* Check format. */
989 		if (nargs != 2 || nreturns < 1) {
990 			args_n_results[nargs] = -1;
991 			args->nreturns = 1;
992 			return;
993 		}
994 		args_n_results[nargs] =	0;	/* properly formatted request */
995 
996 		va = (vm_offset_t)args_n_results[0];
997 		size = (vm_size_t)args_n_results[1];
998 
999 		/* Sanity checks. */
1000 		if ((va & PGOFSET) != 0 || va < OFW_VIRT_BASE ||
1001 		    (va + size) > (OFW_VIRT_BASE + OFW_VIRT_SIZE) ||
1002 		    (size & PGOFSET) != 0 || size == 0) {
1003 			args_n_results[nargs + 1] = -1;
1004 			args->nreturns = 1;
1005 			return;
1006 		}
1007 
1008 		/* Write-back anything stuck in the cache. */
1009 		cpu_idcache_wbinv_all();
1010 
1011 		/* Zero the mappings. */
1012 		{
1013 			pt_entry_t *pte = vtopte(va);
1014 			int npages = size >> PGSHIFT;
1015 
1016 			for (; npages > 0; pte++, npages--)
1017 				*pte = 0;
1018 		}
1019 
1020 		/* Clean out tlb. */
1021 		tlb_flush();
1022 
1023 		args->nreturns = 1;
1024 	} else if (strcmp(name, "translate") == 0) {
1025 		vm_offset_t va;
1026 		vm_offset_t pa;
1027 		int mode;
1028 		pt_entry_t pte;
1029 
1030 		/* Check format. */
1031 		if (nargs != 1 || nreturns < 4) {
1032 			args_n_results[nargs] = -1;
1033 			args->nreturns = 1;
1034 			return;
1035 		}
1036 		args_n_results[nargs] =	0;	/* properly formatted request */
1037 
1038 		va = (vm_offset_t)args_n_results[0];
1039 
1040 		/* Sanity checks.
1041 		 * For now, I am only willing to translate va's in the
1042 		 * "ofw range." Eventually, I may be more generous. -JJK
1043 		 */
1044 		if ((va & PGOFSET) != 0 ||  va < OFW_VIRT_BASE ||
1045 		    va >= (OFW_VIRT_BASE + OFW_VIRT_SIZE)) {
1046 			args_n_results[nargs + 1] = -1;
1047 			args->nreturns = 1;
1048 			return;
1049 		}
1050 
1051 		/* Lookup mapping. */
1052 		pte = *vtopte(va);
1053 		if (pte == 0) {
1054 			/* No mapping. */
1055 			args_n_results[nargs + 1] = -1;
1056 			args->nreturns = 2;
1057 		} else {
1058 			/* Existing mapping. */
1059 			pa = (pte & L2_S_FRAME) | (va & L2_S_OFFSET);
1060 			mode = (pte & 0x0C00) | (0 << 5) | (pte & 0x000C);	/* AP | DOM | CB */
1061 
1062 			args_n_results[nargs + 1] = 0;
1063 			args_n_results[nargs + 2] = pa;
1064 			args_n_results[nargs + 3] =	mode;
1065 			args->nreturns = 4;
1066 		}
1067 	} else if (strcmp(name, "claim-phys") == 0) {
1068 		struct pglist alloclist = TAILQ_HEAD_INITIALIZER(alloclist);
1069 		vm_offset_t low, high;
1070 		vm_size_t align, size;
1071 
1072 		/*
1073 		 * XXX
1074 		 * XXX THIS IS A GROSS HACK AND NEEDS TO BE REWRITTEN. -- cgd
1075 		 * XXX
1076 		 */
1077 
1078 		/* Check format. */
1079 		if (nargs != 4 || nreturns < 3) {
1080 			args_n_results[nargs] = -1;
1081 			args->nreturns = 1;
1082 			return;
1083 		}
1084 		args_n_results[nargs] =	0;	/* properly formatted request */
1085 
1086 		low = args_n_results[0];
1087 		size = args_n_results[2];
1088 		align = args_n_results[3];
1089 		high = args_n_results[1] + size;
1090 
1091 #if 0
1092 		printf("claim-phys: low = 0x%x, size = 0x%x, align = 0x%x, high = 0x%x\n",
1093 		    low, size, align, high);
1094 		align = size;
1095 		printf("forcing align to be 0x%x\n", align);
1096 #endif
1097 
1098 		args_n_results[nargs + 1] =
1099 		uvm_pglistalloc(size, low, high, align, 0, &alloclist, 1, 0);
1100 #if 0
1101 		printf(" -> 0x%lx", args_n_results[nargs + 1]);
1102 #endif
1103 		if (args_n_results[nargs + 1] != 0) {
1104 #if 0
1105 			printf("(failed)\n");
1106 #endif
1107 			args_n_results[nargs + 1] = -1;
1108 			args->nreturns = 2;
1109 			return;
1110 		}
1111 		args_n_results[nargs + 2] = alloclist.tqh_first->phys_addr;
1112 #if 0
1113 		printf("(succeeded: pa = 0x%lx)\n", args_n_results[nargs + 2]);
1114 #endif
1115 		args->nreturns = 3;
1116 
1117 	} else if (strcmp(name, "release-phys") == 0) {
1118 		printf("unimplemented ofw callback - %s\n", name);
1119 		args_n_results[nargs] = -1;
1120 		args->nreturns = 1;
1121 	} else if (strcmp(name, "claim-virt") == 0) {
1122 		vm_offset_t va;
1123 		vm_size_t size;
1124 		vm_offset_t align;
1125 
1126 		/* XXX - notyet */
1127 /*		printf("unimplemented ofw callback - %s\n", name);*/
1128 		args_n_results[nargs] = -1;
1129 		args->nreturns = 1;
1130 		return;
1131 
1132 		/* Check format. */
1133 		if (nargs != 2 || nreturns < 3) {
1134 		    args_n_results[nargs] = -1;
1135 		    args->nreturns = 1;
1136 		    return;
1137 		}
1138 		args_n_results[nargs] =	0;	/* properly formatted request */
1139 
1140 		/* Allocate size bytes with specified alignment. */
1141 		size = (vm_size_t)args_n_results[0];
1142 		align = (vm_offset_t)args_n_results[1];
1143 		if (align % NBPG != 0) {
1144 			args_n_results[nargs + 1] = -1;
1145 			args->nreturns = 2;
1146 			return;
1147 		}
1148 
1149 		if (va == 0) {
1150 			/* Couldn't allocate. */
1151 			args_n_results[nargs + 1] = -1;
1152 			args->nreturns = 2;
1153 		} else {
1154 			/* Successful allocation. */
1155 			args_n_results[nargs + 1] = 0;
1156 			args_n_results[nargs + 2] = va;
1157 			args->nreturns = 3;
1158 		}
1159 	} else if (strcmp(name, "release-virt") == 0) {
1160 		vm_offset_t va;
1161 		vm_size_t size;
1162 
1163 		/* XXX - notyet */
1164 		printf("unimplemented ofw callback - %s\n", name);
1165 		args_n_results[nargs] = -1;
1166 		args->nreturns = 1;
1167 		return;
1168 
1169 		/* Check format. */
1170 		if (nargs != 2 || nreturns < 1) {
1171 			args_n_results[nargs] = -1;
1172 			args->nreturns = 1;
1173 			return;
1174 		}
1175 		args_n_results[nargs] =	0;	/* properly formatted request */
1176 
1177 		/* Release bytes. */
1178 		va = (vm_offset_t)args_n_results[0];
1179 		size = (vm_size_t)args_n_results[1];
1180 
1181 		args->nreturns = 1;
1182 	} else {
1183 		args_n_results[nargs] = -1;
1184 		args->nreturns = 1;
1185 	}
1186 }
1187 
1188 static void
1189 ofw_construct_proc0_addrspace(proc0_ttbbase, proc0_ptpt)
1190 	pv_addr_t *proc0_ttbbase;
1191 	pv_addr_t *proc0_ptpt;
1192 {
1193 	int i, oft;
1194 	pv_addr_t proc0_pagedir;
1195 	pv_addr_t proc0_pt_pte;
1196 	pv_addr_t proc0_pt_sys;
1197 	pv_addr_t proc0_pt_kernel[KERNEL_IMG_PTS];
1198 	pv_addr_t proc0_pt_vmdata[KERNEL_VMDATA_PTS];
1199 	pv_addr_t proc0_pt_ofw[KERNEL_OFW_PTS];
1200 	pv_addr_t proc0_pt_io[KERNEL_IO_PTS];
1201 	pv_addr_t msgbuf;
1202 	vm_offset_t L1pagetable;
1203 	struct mem_translation *tp;
1204 
1205 	/* Set-up the system page. */
1206 	KASSERT(vector_page == 0);	/* XXX for now */
1207 	systempage.pv_va = ofw_claimvirt(vector_page, NBPG, 0);
1208 	if (systempage.pv_va == -1) {
1209 		/* Something was already mapped to vector_page's VA. */
1210 		systempage.pv_va = vector_page;
1211 		systempage.pv_pa = ofw_gettranslation(vector_page);
1212 		if (systempage.pv_pa == -1)
1213 			panic("bogus result from gettranslation(vector_page)");
1214 	} else {
1215 		/* We were just allocated the page-length range at VA 0. */
1216 		if (systempage.pv_va != vector_page)
1217 			panic("bogus result from claimvirt(vector_page, NBPG, 0)");
1218 
1219 		/* Now allocate a physical page, and establish the mapping. */
1220 		systempage.pv_pa = ofw_claimphys(0, NBPG, NBPG);
1221 		if (systempage.pv_pa == -1)
1222 			panic("bogus result from claimphys(0, NBPG, NBPG)");
1223 		ofw_settranslation(systempage.pv_va, systempage.pv_pa,
1224 		    NBPG, -1);	/* XXX - mode? -JJK */
1225 
1226 		/* Zero the memory. */
1227 		bzero((char *)systempage.pv_va, NBPG);
1228 	}
1229 
1230 	/* Allocate/initialize space for the proc0, NetBSD-managed */
1231 	/* page tables that we will be switching to soon. */
1232 	ofw_claimpages(&virt_freeptr, &proc0_pagedir, L1_TABLE_SIZE);
1233 	ofw_claimpages(&virt_freeptr, &proc0_pt_pte, L2_TABLE_SIZE);
1234 	ofw_claimpages(&virt_freeptr, &proc0_pt_sys, L2_TABLE_SIZE);
1235 	for (i = 0; i < KERNEL_IMG_PTS; i++)
1236 		ofw_claimpages(&virt_freeptr, &proc0_pt_kernel[i], L2_TABLE_SIZE);
1237 	for (i = 0; i < KERNEL_VMDATA_PTS; i++)
1238 		ofw_claimpages(&virt_freeptr, &proc0_pt_vmdata[i], L2_TABLE_SIZE);
1239 	for (i = 0; i < KERNEL_OFW_PTS; i++)
1240 		ofw_claimpages(&virt_freeptr, &proc0_pt_ofw[i], L2_TABLE_SIZE);
1241 	for (i = 0; i < KERNEL_IO_PTS; i++)
1242 		ofw_claimpages(&virt_freeptr, &proc0_pt_io[i], L2_TABLE_SIZE);
1243 
1244 	/* Allocate/initialize space for stacks. */
1245 #ifndef	OFWGENCFG
1246 	ofw_claimpages(&virt_freeptr, &irqstack, NBPG);
1247 #endif
1248 	ofw_claimpages(&virt_freeptr, &undstack, NBPG);
1249 	ofw_claimpages(&virt_freeptr, &abtstack, NBPG);
1250 	ofw_claimpages(&virt_freeptr, &kernelstack, UPAGES * NBPG);
1251 
1252 	/* Allocate/initialize space for msgbuf area. */
1253 	ofw_claimpages(&virt_freeptr, &msgbuf, MSGBUFSIZE);
1254 	msgbufphys = msgbuf.pv_pa;
1255 
1256 	/* Construct the proc0 L1 pagetable. */
1257 	L1pagetable = proc0_pagedir.pv_va;
1258 
1259 	pmap_link_l2pt(L1pagetable, 0x0, &proc0_pt_sys);
1260 	for (i = 0; i < KERNEL_IMG_PTS; i++)
1261 		pmap_link_l2pt(L1pagetable, KERNEL_BASE + i * 0x00400000,
1262 		    &proc0_pt_kernel[i]);
1263 	pmap_link_l2pt(L1pagetable, PTE_BASE,
1264 	    &proc0_pt_pte);
1265 	for (i = 0; i < KERNEL_VMDATA_PTS; i++)
1266 		pmap_link_l2pt(L1pagetable, KERNEL_VM_BASE + i * 0x00400000,
1267 		    &proc0_pt_vmdata[i]);
1268 	for (i = 0; i < KERNEL_OFW_PTS; i++)
1269 		pmap_link_l2pt(L1pagetable, OFW_VIRT_BASE + i * 0x00400000,
1270 		    &proc0_pt_ofw[i]);
1271 	for (i = 0; i < KERNEL_IO_PTS; i++)
1272 		pmap_link_l2pt(L1pagetable, IO_VIRT_BASE + i * 0x00400000,
1273 		    &proc0_pt_io[i]);
1274 
1275 	/*
1276 	 * OK, we're done allocating.
1277 	 * Get a dump of OFW's translations, and make the appropriate
1278 	 * entries in the L2 pagetables that we just allocated.
1279 	 */
1280 
1281 	ofw_getvirttranslations();
1282 
1283 	for (oft = 0,  tp = OFtranslations; oft < nOFtranslations;
1284 	    oft++, tp++) {
1285 
1286 		vm_offset_t va, pa;
1287 		int npages = tp->size / NBPG;
1288 
1289 		/* Size must be an integral number of pages. */
1290 		if (npages == 0 || tp->size % NBPG != 0)
1291 			panic("illegal ofw translation (size)");
1292 
1293 		/* Make an entry for each page in the appropriate table. */
1294 		for (va = tp->virt, pa = tp->phys; npages > 0;
1295 		    va += NBPG, pa += NBPG, npages--) {
1296 			/*
1297 			 * Map the top bits to the appropriate L2 pagetable.
1298 			 * The only allowable regions are page0, the
1299 			 * kernel-static area, and the ofw area.
1300 			 */
1301 			switch (va >> (L1_S_SHIFT + 2)) {
1302 			case 0:
1303 				/* page0 */
1304 				break;
1305 
1306 #if KERNEL_IMG_PTS != 2
1307 #error "Update ofw translation range list"
1308 #endif
1309 			case ( KERNEL_BASE                 >> (L1_S_SHIFT + 2)):
1310 			case ((KERNEL_BASE   + 0x00400000) >> (L1_S_SHIFT + 2)):
1311 				/* kernel static area */
1312 				break;
1313 
1314 			case ( OFW_VIRT_BASE               >> (L1_S_SHIFT + 2)):
1315 			case ((OFW_VIRT_BASE + 0x00400000) >> (L1_S_SHIFT + 2)):
1316 			case ((OFW_VIRT_BASE + 0x00800000) >> (L1_S_SHIFT + 2)):
1317 			case ((OFW_VIRT_BASE + 0x00C00000) >> (L1_S_SHIFT + 2)):
1318 				/* ofw area */
1319 				break;
1320 
1321 			case ( IO_VIRT_BASE               >> (L1_S_SHIFT + 2)):
1322 			case ((IO_VIRT_BASE + 0x00400000) >> (L1_S_SHIFT + 2)):
1323 			case ((IO_VIRT_BASE + 0x00800000) >> (L1_S_SHIFT + 2)):
1324 			case ((IO_VIRT_BASE + 0x00C00000) >> (L1_S_SHIFT + 2)):
1325 				/* io area */
1326 				break;
1327 
1328 			default:
1329 				/* illegal */
1330 				panic("illegal ofw translation (addr) %#lx",
1331 				    va);
1332 			}
1333 
1334 			/* Make the entry. */
1335 			pmap_map_entry(L1pagetable, va, pa,
1336 			    VM_PROT_READ|VM_PROT_WRITE,
1337 			    (tp->mode & 0xC) == 0xC ? PTE_CACHE
1338 						    : PTE_NOCACHE);
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * We don't actually want some of the mappings that we just
1344 	 * set up to appear in proc0's address space.  In particular,
1345 	 * we don't want aliases to physical addresses that the kernel
1346 	 * has-mapped/will-map elsewhere.
1347 	 */
1348 	ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1349 	    proc0_pt_sys.pv_va, L2_TABLE_SIZE);
1350 	for (i = 0; i < KERNEL_IMG_PTS; i++)
1351 		ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1352 		    proc0_pt_kernel[i].pv_va, L2_TABLE_SIZE);
1353 	for (i = 0; i < KERNEL_VMDATA_PTS; i++)
1354 		ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1355 		    proc0_pt_vmdata[i].pv_va, L2_TABLE_SIZE);
1356 	for (i = 0; i < KERNEL_OFW_PTS; i++)
1357 		ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1358 		    proc0_pt_ofw[i].pv_va, L2_TABLE_SIZE);
1359 	for (i = 0; i < KERNEL_IO_PTS; i++)
1360 		ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1361 		    proc0_pt_io[i].pv_va, L2_TABLE_SIZE);
1362 	ofw_discardmappings(proc0_pt_kernel[KERNEL_IMG_PTS - 1].pv_va,
1363 	    msgbuf.pv_va, MSGBUFSIZE);
1364 
1365 	/*
1366 	 * We did not throw away the proc0_pt_pte and proc0_pagedir
1367 	 * mappings as well still want them. However we don't want them
1368 	 * cached ...
1369 	 * Really these should be uncached when allocated.
1370 	 */
1371 	pmap_map_entry(L1pagetable, proc0_pt_pte.pv_va,
1372 	    proc0_pt_pte.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1373 	for (i = 0; i < (L1_TABLE_SIZE / NBPG); ++i)
1374 		pmap_map_entry(L1pagetable,
1375 		    proc0_pagedir.pv_va + NBPG * i,
1376 		    proc0_pagedir.pv_pa + NBPG * i,
1377 		    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1378 
1379 	/*
1380 	 * Construct the proc0 L2 pagetables that map page tables.
1381 	 */
1382 
1383 	/* Map entries in the L2pagetable used to map L2PTs. */
1384 	pmap_map_entry(L1pagetable,
1385 	    PTE_BASE + (0x00000000 >> (PGSHIFT-2)),
1386 	    proc0_pt_sys.pv_pa,
1387 	    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1388 	for (i = 0; i < KERNEL_IMG_PTS; i++)
1389 		pmap_map_entry(L1pagetable,
1390 		    PTE_BASE + ((KERNEL_BASE + i * 0x00400000) >> (PGSHIFT-2)),
1391 		    proc0_pt_kernel[i].pv_pa,
1392 		    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1393 	pmap_map_entry(L1pagetable,
1394 	    PTE_BASE + (PTE_BASE >> (PGSHIFT-2)),
1395 	    proc0_pt_pte.pv_pa,
1396 	    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1397 	for (i = 0; i < KERNEL_VMDATA_PTS; i++)
1398 		pmap_map_entry(L1pagetable,
1399 		    PTE_BASE + ((KERNEL_VM_BASE + i * 0x00400000)
1400 		    >> (PGSHIFT-2)), proc0_pt_vmdata[i].pv_pa,
1401 		    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1402 	for (i = 0; i < KERNEL_OFW_PTS; i++)
1403 		pmap_map_entry(L1pagetable,
1404 		    PTE_BASE + ((OFW_VIRT_BASE + i * 0x00400000)
1405 		    >> (PGSHIFT-2)), proc0_pt_ofw[i].pv_pa,
1406 		    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1407 	for (i = 0; i < KERNEL_IO_PTS; i++)
1408 		pmap_map_entry(L1pagetable,
1409 		    PTE_BASE + ((IO_VIRT_BASE + i * 0x00400000)
1410 		    >> (PGSHIFT-2)), proc0_pt_io[i].pv_pa,
1411 		    VM_PROT_READ|VM_PROT_WRITE, PTE_NOCACHE);
1412 
1413 	/* update the top of the kernel VM */
1414 	pmap_curmaxkvaddr =
1415 	    KERNEL_VM_BASE + (KERNEL_VMDATA_PTS * 0x00400000);
1416 
1417 	/*
1418          * gross hack for the sake of not thrashing the TLB and making
1419 	 * cache flush more efficient: blast l1 ptes for sections.
1420          */
1421 	for (oft = 0, tp = OFtranslations; oft < nOFtranslations; oft++, tp++) {
1422 		vm_offset_t va = tp->virt;
1423 		vm_offset_t pa = tp->phys;
1424 
1425 		if (((va | pa) & L1_S_OFFSET) == 0) {
1426 			int nsections = tp->size / L1_S_SIZE;
1427 
1428 			while (nsections--) {
1429 				/* XXXJRT prot?? */
1430 				pmap_map_section(L1pagetable, va, pa,
1431 				    VM_PROT_READ|VM_PROT_WRITE,
1432 				    (tp->mode & 0xC) == 0xC ? PTE_CACHE
1433 							    : PTE_NOCACHE);
1434 				va += L1_S_SIZE;
1435 				pa += L1_S_SIZE;
1436 			}
1437 		}
1438 	}
1439 
1440 	/* OUT parameters are the new ttbbase and the pt which maps pts. */
1441 	*proc0_ttbbase = proc0_pagedir;
1442 	*proc0_ptpt = proc0_pt_pte;
1443 }
1444 
1445 
1446 static void
1447 ofw_getphysmeminfo()
1448 {
1449 	int phandle;
1450 	int mem_len;
1451 	int avail_len;
1452 	int i;
1453 
1454 	if ((phandle = OF_finddevice("/memory")) == -1 ||
1455 	    (mem_len = OF_getproplen(phandle, "reg")) <= 0 ||
1456 	    (OFphysmem = (struct mem_region *)ofw_malloc(mem_len)) == 0 ||
1457 	    OF_getprop(phandle, "reg", OFphysmem, mem_len) != mem_len ||
1458 	    (avail_len = OF_getproplen(phandle, "available")) <= 0 ||
1459  	    (OFphysavail = (struct mem_region *)ofw_malloc(avail_len)) == 0 ||
1460 	    OF_getprop(phandle, "available", OFphysavail, avail_len)
1461 	    != avail_len)
1462 		panic("can't get physmeminfo from OFW");
1463 
1464 	nOFphysmem = mem_len / sizeof(struct mem_region);
1465 	nOFphysavail = avail_len / sizeof(struct mem_region);
1466 
1467 	/*
1468 	 * Sort the blocks in each array into ascending address order.
1469 	 * Also, page-align all blocks.
1470 	 */
1471 	for (i = 0; i < 2; i++) {
1472 		struct mem_region *tmp = (i == 0) ? OFphysmem : OFphysavail;
1473 		struct mem_region *mp;
1474 		int cnt =  (i == 0) ? nOFphysmem : nOFphysavail;
1475 		int j;
1476 
1477 #ifdef	OLDPRINTFS
1478 		printf("ofw_getphysmeminfo:  %d blocks\n", cnt);
1479 #endif
1480 
1481 		/* XXX - Convert all the values to host order. -JJK */
1482 		for (j = 0, mp = tmp; j < cnt; j++, mp++) {
1483 			mp->start = of_decode_int((unsigned char *)&mp->start);
1484 			mp->size = of_decode_int((unsigned char *)&mp->size);
1485 		}
1486 
1487 		for (j = 0, mp = tmp; j < cnt; j++, mp++) {
1488 			u_int s, sz;
1489 			struct mem_region *mp1;
1490 
1491 			/* Page-align start of the block. */
1492 			s = mp->start % NBPG;
1493 			if (s != 0) {
1494 				s = (NBPG - s);
1495 
1496 				if (mp->size >= s) {
1497 					mp->start += s;
1498 					mp->size -= s;
1499 				}
1500 			}
1501 
1502 			/* Page-align the size. */
1503 			mp->size -= mp->size % NBPG;
1504 
1505 			/* Handle empty block. */
1506 			if (mp->size == 0) {
1507 				bcopy(mp + 1, mp, (cnt - (mp - tmp))
1508 				    * sizeof(struct mem_region));
1509 				cnt--;
1510 				mp--;
1511 				continue;
1512 			}
1513 
1514 			/* Bubble sort. */
1515 			s = mp->start;
1516 			sz = mp->size;
1517 			for (mp1 = tmp; mp1 < mp; mp1++)
1518 				if (s < mp1->start)
1519 					break;
1520 			if (mp1 < mp) {
1521 				bcopy(mp1, mp1 + 1, (char *)mp - (char *)mp1);
1522 				mp1->start = s;
1523 				mp1->size = sz;
1524 			}
1525 		}
1526 
1527 #ifdef	OLDPRINTFS
1528 		for (mp = tmp; mp->size; mp++) {
1529 			printf("%x, %x\n", mp->start, mp->size);
1530 		}
1531 #endif
1532 	}
1533 }
1534 
1535 
1536 static void
1537 ofw_getvirttranslations(void)
1538 {
1539 	int mmu_phandle;
1540 	int mmu_ihandle;
1541 	int trans_len;
1542 	int over, len;
1543 	int i;
1544 	struct mem_translation *tp;
1545 
1546 	mmu_ihandle = ofw_mmu_ihandle();
1547 
1548 	/* overallocate to avoid increases during allocation */
1549 	over = 4 * sizeof(struct mem_translation);
1550 	if ((mmu_phandle = OF_instance_to_package(mmu_ihandle)) == -1 ||
1551 	    (len = OF_getproplen(mmu_phandle, "translations")) <= 0 ||
1552 	    (OFtranslations = ofw_malloc(len + over)) == 0 ||
1553 	    (trans_len = OF_getprop(mmu_phandle, "translations",
1554 	    OFtranslations, len + over)) > (len + over))
1555 		panic("can't get virttranslations from OFW");
1556 
1557 	/* XXX - Convert all the values to host order. -JJK */
1558 	nOFtranslations = trans_len / sizeof(struct mem_translation);
1559 #ifdef	OLDPRINTFS
1560 	printf("ofw_getvirtmeminfo:  %d blocks\n", nOFtranslations);
1561 #endif
1562 	for (i = 0, tp = OFtranslations; i < nOFtranslations; i++, tp++) {
1563 		tp->virt = of_decode_int((unsigned char *)&tp->virt);
1564 		tp->size = of_decode_int((unsigned char *)&tp->size);
1565 		tp->phys = of_decode_int((unsigned char *)&tp->phys);
1566 		tp->mode = of_decode_int((unsigned char *)&tp->mode);
1567 	}
1568 }
1569 
1570 /*
1571  * ofw_valloc: allocate blocks of VM for IO and other special purposes
1572  */
1573 typedef struct _vfree {
1574 	struct _vfree *pNext;
1575 	vm_offset_t start;
1576 	vm_size_t   size;
1577 } VFREE, *PVFREE;
1578 
1579 static VFREE vfinitial = { NULL, IO_VIRT_BASE, IO_VIRT_SIZE };
1580 
1581 static PVFREE vflist = &vfinitial;
1582 
1583 static vm_offset_t
1584 ofw_valloc(size, align)
1585 	vm_offset_t size;
1586 	vm_offset_t align;
1587 {
1588 	PVFREE        *ppvf;
1589 	PVFREE        pNew;
1590 	vm_offset_t   new;
1591 	vm_offset_t   lead;
1592 
1593 	for (ppvf = &vflist; *ppvf; ppvf = &((*ppvf)->pNext)) {
1594 		if (align == 0) {
1595 			new = (*ppvf)->start;
1596 			lead = 0;
1597 		} else {
1598 			new  = ((*ppvf)->start + (align - 1)) & ~(align - 1);
1599 			lead = new - (*ppvf)->start;
1600 		}
1601 
1602 		if (((*ppvf)->size - lead) >= size) {
1603  			if (lead == 0) {
1604 				/* using whole block */
1605 				if (size == (*ppvf)->size) {
1606 					/* splice out of list */
1607 					(*ppvf) = (*ppvf)->pNext;
1608 				} else { /* tail of block is free */
1609 					(*ppvf)->start = new + size;
1610 					(*ppvf)->size -= size;
1611 				}
1612 			} else {
1613 				vm_size_t tail = ((*ppvf)->start
1614 				    + (*ppvf)->size) - (new + size);
1615 				/* free space at beginning */
1616 				(*ppvf)->size = lead;
1617 
1618 				if (tail != 0) {
1619 					/* free space at tail */
1620 					pNew = ofw_malloc(sizeof(VFREE));
1621 					pNew->pNext  = (*ppvf)->pNext;
1622 					(*ppvf)->pNext = pNew;
1623 					pNew->start  = new + size;
1624 					pNew->size   = tail;
1625 				}
1626 			}
1627 			return new;
1628 		} /* END if */
1629 	} /* END for */
1630 
1631 	return -1;
1632 }
1633 
1634 vm_offset_t
1635 ofw_map(pa, size, cb_bits)
1636 	vm_offset_t pa;
1637 	vm_size_t size;
1638 	int cb_bits;
1639 {
1640 	vm_offset_t va;
1641 
1642 	if ((va = ofw_valloc(size, size)) == -1)
1643 		panic("cannot alloc virtual memory for %#lx", pa);
1644 
1645 	ofw_claimvirt(va, size, 0); /* make sure OFW knows about the memory */
1646 
1647 	ofw_settranslation(va, pa, size, L2_AP(AP_KRW) | cb_bits);
1648 
1649 	return va;
1650 }
1651 
1652 static int
1653 ofw_mem_ihandle(void)
1654 {
1655 	static int mem_ihandle = 0;
1656 	int chosen;
1657 
1658 	if (mem_ihandle != 0)
1659 		return(mem_ihandle);
1660 
1661 	if ((chosen = OF_finddevice("/chosen")) == -1 ||
1662 	    OF_getprop(chosen, "memory", &mem_ihandle, sizeof(int)) < 0)
1663 		panic("ofw_mem_ihandle");
1664 
1665 	mem_ihandle = of_decode_int((unsigned char *)&mem_ihandle);
1666 
1667 	return(mem_ihandle);
1668 }
1669 
1670 
1671 static int
1672 ofw_mmu_ihandle(void)
1673 {
1674 	static int mmu_ihandle = 0;
1675 	int chosen;
1676 
1677 	if (mmu_ihandle != 0)
1678 		return(mmu_ihandle);
1679 
1680 	if ((chosen = OF_finddevice("/chosen")) == -1 ||
1681 	    OF_getprop(chosen, "mmu", &mmu_ihandle, sizeof(int)) < 0)
1682 		panic("ofw_mmu_ihandle");
1683 
1684 	mmu_ihandle = of_decode_int((unsigned char *)&mmu_ihandle);
1685 
1686 	return(mmu_ihandle);
1687 }
1688 
1689 
1690 /* Return -1 on failure. */
1691 static vm_offset_t
1692 ofw_claimphys(pa, size, align)
1693 	vm_offset_t pa;
1694 	vm_size_t size;
1695 	vm_offset_t align;
1696 {
1697 	int mem_ihandle = ofw_mem_ihandle();
1698 
1699 /*	printf("ofw_claimphys (%x, %x, %x) --> ", pa, size, align);*/
1700 	if (align == 0) {
1701 		/* Allocate at specified base; alignment is ignored. */
1702 		pa = OF_call_method_1("claim", mem_ihandle, 3, pa, size, align);
1703 	} else {
1704 		/* Allocate anywhere, with specified alignment. */
1705 		pa = OF_call_method_1("claim", mem_ihandle, 2, size, align);
1706 	}
1707 
1708 /*	printf("%x\n", pa);*/
1709 	return(pa);
1710 }
1711 
1712 
1713 #if 0
1714 /* Return -1 on failure. */
1715 static vm_offset_t
1716 ofw_releasephys(pa, size)
1717 	vm_offset_t pa;
1718 	vm_size_t size;
1719 {
1720 	int mem_ihandle = ofw_mem_ihandle();
1721 
1722 /*	printf("ofw_releasephys (%x, %x)\n", pa, size);*/
1723 
1724 	return (OF_call_method_1("release", mem_ihandle, 2, pa, size));
1725 }
1726 #endif
1727 
1728 /* Return -1 on failure. */
1729 static vm_offset_t
1730 ofw_claimvirt(va, size, align)
1731 	vm_offset_t va;
1732 	vm_size_t size;
1733 	vm_offset_t align;
1734 {
1735 	int mmu_ihandle = ofw_mmu_ihandle();
1736 
1737 	/*printf("ofw_claimvirt (%x, %x, %x) --> ", va, size, align);*/
1738 	if (align == 0) {
1739 		/* Allocate at specified base; alignment is ignored. */
1740 		va = OF_call_method_1("claim", mmu_ihandle, 3, va, size, align);
1741 	} else {
1742 		/* Allocate anywhere, with specified alignment. */
1743 		va = OF_call_method_1("claim", mmu_ihandle, 2, size, align);
1744 	}
1745 
1746 	/*printf("%x\n", va);*/
1747 	return(va);
1748 }
1749 
1750 
1751 /* Return -1 if no mapping. */
1752 vm_offset_t
1753 ofw_gettranslation(va)
1754 	vm_offset_t va;
1755 {
1756 	int mmu_ihandle = ofw_mmu_ihandle();
1757 	vm_offset_t pa;
1758 	int mode;
1759 	int exists;
1760 
1761 	/*printf("ofw_gettranslation (%x) --> ", va);*/
1762 	exists = 0;	    /* gets set to true if translation exists */
1763 	if (OF_call_method("translate", mmu_ihandle, 1, 3, va, &pa, &mode,
1764 	    &exists) != 0)
1765 		return(-1);
1766 
1767 	/*printf("%x\n", exists ? pa : -1);*/
1768 	return(exists ? pa : -1);
1769 }
1770 
1771 
1772 static void
1773 ofw_settranslation(va, pa, size, mode)
1774 	vm_offset_t va;
1775 	vm_offset_t pa;
1776 	vm_size_t size;
1777 	int mode;
1778 {
1779 	int mmu_ihandle = ofw_mmu_ihandle();
1780 
1781 /*printf("ofw_settranslation (%x, %x, %x, %x) --> void", va, pa, size, mode);*/
1782 	if (OF_call_method("map", mmu_ihandle, 4, 0, pa, va, size, mode) != 0)
1783 		panic("ofw_settranslation failed");
1784 }
1785 
1786 /*
1787  *  Allocation routine used before the kernel takes over memory.
1788  *  Use this for efficient storage for things that aren't rounded to
1789  *  page size.
1790  *
1791  *  The point here is not necessarily to be very efficient (even though
1792  *  that's sort of nice), but to do proper dynamic allocation to avoid
1793  *  size-limitation errors.
1794  *
1795  */
1796 
1797 typedef struct _leftover {
1798 	struct _leftover *pNext;
1799 	vm_size_t size;
1800 } LEFTOVER, *PLEFTOVER;
1801 
1802 /* leftover bits of pages.  first word is pointer to next.
1803    second word is size of leftover */
1804 static PLEFTOVER leftovers = NULL;
1805 
1806 static void *
1807 ofw_malloc(size)
1808 	vm_size_t size;
1809 {
1810 	PLEFTOVER   *ppLeftover;
1811 	PLEFTOVER   pLeft;
1812 	pv_addr_t   new;
1813 	vm_size_t   newSize, claim_size;
1814 
1815 	/* round and set minimum size */
1816 	size = max(sizeof(LEFTOVER),
1817 	    ((size + (sizeof(LEFTOVER) - 1)) & ~(sizeof(LEFTOVER) - 1)));
1818 
1819 	for (ppLeftover = &leftovers; *ppLeftover;
1820 	    ppLeftover = &((*ppLeftover)->pNext))
1821 		if ((*ppLeftover)->size >= size)
1822 			break;
1823 
1824 	if (*ppLeftover) { /* have a leftover of the right size */
1825 		/* remember the leftover */
1826 		new.pv_va = (vm_offset_t)*ppLeftover;
1827 		if ((*ppLeftover)->size < (size + sizeof(LEFTOVER))) {
1828 			/* splice out of chain */
1829 			*ppLeftover = (*ppLeftover)->pNext;
1830 		} else {
1831 			/* remember the next pointer */
1832 			pLeft = (*ppLeftover)->pNext;
1833 			newSize = (*ppLeftover)->size - size; /* reduce size */
1834 			/* move pointer */
1835 			*ppLeftover = (PLEFTOVER)(((vm_offset_t)*ppLeftover)
1836 			    + size);
1837 			(*ppLeftover)->pNext = pLeft;
1838 			(*ppLeftover)->size  = newSize;
1839 		}
1840 	} else {
1841 		claim_size = (size + NBPG - 1) & ~(NBPG - 1);
1842 		ofw_claimpages(&virt_freeptr, &new, claim_size);
1843 		if ((size + sizeof(LEFTOVER)) <= claim_size) {
1844 			pLeft = (PLEFTOVER)(new.pv_va + size);
1845 			pLeft->pNext = leftovers;
1846 			pLeft->size = claim_size - size;
1847 			leftovers = pLeft;
1848 		}
1849 	}
1850 
1851 	return (void *)(new.pv_va);
1852 }
1853 
1854 /*
1855  *  Here is a really, really sleazy free.  It's not used right now,
1856  *  because it's not worth the extra complexity for just a few bytes.
1857  *
1858  */
1859 #if 0
1860 static void
1861 ofw_free(addr, size)
1862 	vm_offset_t addr;
1863 	vm_size_t size;
1864 {
1865 	PLEFTOVER pLeftover = (PLEFTOVER)addr;
1866 
1867 	/* splice right into list without checks or compaction */
1868 	pLeftover->pNext = leftovers;
1869 	pLeftover->size  = size;
1870 	leftovers        = pLeftover;
1871 }
1872 #endif
1873 
1874 /*
1875  *  Allocate and zero round(size)/NBPG pages of memory.
1876  *  We guarantee that the allocated memory will be
1877  *  aligned to a boundary equal to the smallest power of
1878  *  2 greater than or equal to size.
1879  *  free_pp is an IN/OUT parameter which points to the
1880  *  last allocated virtual address in an allocate-downwards
1881  *  stack.  pv_p is an OUT parameter which contains the
1882  *  virtual and physical base addresses of the allocated
1883  *  memory.
1884  */
1885 static void
1886 ofw_claimpages(free_pp, pv_p, size)
1887 	vm_offset_t *free_pp;
1888 	pv_addr_t *pv_p;
1889 	vm_size_t size;
1890 {
1891 	/* round-up to page boundary */
1892 	vm_size_t alloc_size = (size + NBPG - 1) & ~(NBPG - 1);
1893 	vm_size_t aligned_size;
1894 	vm_offset_t va, pa;
1895 
1896 	if (alloc_size == 0)
1897 		panic("ofw_claimpages zero");
1898 
1899 	for (aligned_size = 1; aligned_size < alloc_size; aligned_size <<= 1)
1900 		;
1901 
1902 	/*  The only way to provide the alignment guarantees is to
1903 	 *  allocate the virtual and physical ranges separately,
1904 	 *  then do an explicit map call.
1905 	 */
1906 	va = (*free_pp & ~(aligned_size - 1)) - aligned_size;
1907 	if (ofw_claimvirt(va, alloc_size, 0) != va)
1908 		panic("ofw_claimpages va alloc");
1909 	pa = ofw_claimphys(0, alloc_size, aligned_size);
1910 	if (pa == -1)
1911 		panic("ofw_claimpages pa alloc");
1912 	/* XXX - what mode? -JJK */
1913 	ofw_settranslation(va, pa, alloc_size, -1);
1914 
1915 	/* The memory's mapped-in now, so we can zero it. */
1916 	bzero((char *)va, alloc_size);
1917 
1918 	/* Set OUT parameters. */
1919 	*free_pp = va;
1920 	pv_p->pv_va = va;
1921 	pv_p->pv_pa = pa;
1922 }
1923 
1924 
1925 static void
1926 ofw_discardmappings(L2pagetable, va, size)
1927 	vm_offset_t L2pagetable;
1928 	vm_offset_t va;
1929 	vm_size_t size;
1930 {
1931 	/* round-up to page boundary */
1932 	vm_size_t alloc_size = (size + NBPG - 1) & ~(NBPG - 1);
1933 	int npages = alloc_size / NBPG;
1934 
1935 	if (npages == 0)
1936 		panic("ofw_discardmappings zero");
1937 
1938 	/* Discard each mapping. */
1939 	for (; npages > 0; va += NBPG, npages--) {
1940 		/* Sanity. The current entry should be non-null. */
1941 		if (ReadWord(L2pagetable + ((va >> 10) & 0x00000FFC)) == 0)
1942 			panic("ofw_discardmappings zero entry");
1943 
1944 		/* Clear the entry. */
1945 		WriteWord(L2pagetable + ((va >> 10) & 0x00000FFC), 0);
1946 	}
1947 }
1948 
1949 
1950 static void
1951 ofw_initallocator(void)
1952 {
1953 
1954 }
1955