xref: /netbsd-src/sys/arch/xen/x86/x86_xpmap.c (revision 9ac63422b666fbe53a067de74d8af2aa4e45a08b)
1 /*	$NetBSD: x86_xpmap.c,v 1.42 2012/03/02 16:37:38 bouyer Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 Mathieu Ropert <mro@adviseo.fr>
5  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 /*
20  * Copyright (c) 2006, 2007 Manuel Bouyer.
21  *
22  * Redistribution and use in source and binary forms, with or without
23  * modification, are permitted provided that the following conditions
24  * are met:
25  * 1. Redistributions of source code must retain the above copyright
26  *    notice, this list of conditions and the following disclaimer.
27  * 2. Redistributions in binary form must reproduce the above copyright
28  *    notice, this list of conditions and the following disclaimer in the
29  *    documentation and/or other materials provided with the distribution.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
32  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
33  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
34  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
35  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
36  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
37  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
38  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
39  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
40  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41  *
42  */
43 
44 /*
45  *
46  * Copyright (c) 2004 Christian Limpach.
47  * All rights reserved.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
59  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
60  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
61  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
62  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
63  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
64  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
65  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
66  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
67  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
68  */
69 
70 
71 #include <sys/cdefs.h>
72 __KERNEL_RCSID(0, "$NetBSD: x86_xpmap.c,v 1.42 2012/03/02 16:37:38 bouyer Exp $");
73 
74 #include "opt_xen.h"
75 #include "opt_ddb.h"
76 #include "ksyms.h"
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/mutex.h>
81 #include <sys/cpu.h>
82 
83 #include <uvm/uvm.h>
84 
85 #include <x86/pmap.h>
86 #include <machine/gdt.h>
87 #include <xen/xenfunc.h>
88 
89 #include <dev/isa/isareg.h>
90 #include <machine/isa_machdep.h>
91 
92 #undef	XENDEBUG
93 /* #define XENDEBUG_SYNC */
94 /* #define	XENDEBUG_LOW */
95 
96 #ifdef XENDEBUG
97 #define	XENPRINTF(x) printf x
98 #define	XENPRINTK(x) printk x
99 #define	XENPRINTK2(x) /* printk x */
100 
101 static char XBUF[256];
102 #else
103 #define	XENPRINTF(x)
104 #define	XENPRINTK(x)
105 #define	XENPRINTK2(x)
106 #endif
107 #define	PRINTF(x) printf x
108 #define	PRINTK(x) printk x
109 
110 volatile shared_info_t *HYPERVISOR_shared_info;
111 /* Xen requires the start_info struct to be page aligned */
112 union start_info_union start_info_union __aligned(PAGE_SIZE);
113 unsigned long *xpmap_phys_to_machine_mapping;
114 kmutex_t pte_lock;
115 
116 void xen_failsafe_handler(void);
117 
118 #define HYPERVISOR_mmu_update_self(req, count, success_count) \
119 	HYPERVISOR_mmu_update((req), (count), (success_count), DOMID_SELF)
120 
121 void
122 xen_failsafe_handler(void)
123 {
124 
125 	panic("xen_failsafe_handler called!\n");
126 }
127 
128 
129 void
130 xen_set_ldt(vaddr_t base, uint32_t entries)
131 {
132 	vaddr_t va;
133 	vaddr_t end;
134 	pt_entry_t *ptp;
135 	int s;
136 
137 #ifdef __x86_64__
138 	end = base + (entries << 3);
139 #else
140 	end = base + entries * sizeof(union descriptor);
141 #endif
142 
143 	for (va = base; va < end; va += PAGE_SIZE) {
144 		KASSERT(va >= VM_MIN_KERNEL_ADDRESS);
145 		ptp = kvtopte(va);
146 		XENPRINTF(("xen_set_ldt %#" PRIxVADDR " %d %p\n",
147 		    base, entries, ptp));
148 		pmap_pte_clearbits(ptp, PG_RW);
149 	}
150 	s = splvm();
151 	xpq_queue_set_ldt(base, entries);
152 	splx(s);
153 }
154 
155 #ifdef XENDEBUG
156 void xpq_debug_dump(void);
157 #endif
158 
159 #define XPQUEUE_SIZE 2048
160 static mmu_update_t xpq_queue_array[MAXCPUS][XPQUEUE_SIZE];
161 static int xpq_idx_array[MAXCPUS];
162 
163 #ifdef i386
164 extern union descriptor tmpgdt[];
165 #endif /* i386 */
166 void
167 xpq_flush_queue(void)
168 {
169 	int i, ok = 0, ret;
170 
171 	mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid];
172 	int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid];
173 
174 	XENPRINTK2(("flush queue %p entries %d\n", xpq_queue, xpq_idx));
175 	for (i = 0; i < xpq_idx; i++)
176 		XENPRINTK2(("%d: 0x%08" PRIx64 " 0x%08" PRIx64 "\n", i,
177 		    xpq_queue[i].ptr, xpq_queue[i].val));
178 
179 retry:
180 	ret = HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok);
181 
182 	if (xpq_idx != 0 && ret < 0) {
183 		struct cpu_info *ci;
184 		CPU_INFO_ITERATOR cii;
185 
186 		printf("xpq_flush_queue: %d entries (%d successful) on "
187 		    "cpu%d (%ld)\n",
188 		    xpq_idx, ok, curcpu()->ci_index, curcpu()->ci_cpuid);
189 
190 		if (ok != 0) {
191 			xpq_queue += ok;
192 			xpq_idx -= ok;
193 			ok = 0;
194 			goto retry;
195 		}
196 
197 		for (CPU_INFO_FOREACH(cii, ci)) {
198 			xpq_queue = xpq_queue_array[ci->ci_cpuid];
199 			xpq_idx = xpq_idx_array[ci->ci_cpuid];
200 			printf("cpu%d (%ld):\n", ci->ci_index, ci->ci_cpuid);
201 			for (i = 0; i < xpq_idx; i++) {
202 				printf("  0x%016" PRIx64 ": 0x%016" PRIx64 "\n",
203 				   xpq_queue[i].ptr, xpq_queue[i].val);
204 			}
205 #ifdef __x86_64__
206 			for (i = 0; i < PDIR_SLOT_PTE; i++) {
207 				if (ci->ci_kpm_pdir[i] == 0)
208 					continue;
209 				printf(" kpm_pdir[%d]: 0x%" PRIx64 "\n",
210 				    i, ci->ci_kpm_pdir[i]);
211 			}
212 #endif
213 		}
214 		panic("HYPERVISOR_mmu_update failed, ret: %d\n", ret);
215 	}
216 	xpq_idx_array[curcpu()->ci_cpuid] = 0;
217 }
218 
219 static inline void
220 xpq_increment_idx(void)
221 {
222 
223 	if (__predict_false(++xpq_idx_array[curcpu()->ci_cpuid] == XPQUEUE_SIZE))
224 		xpq_flush_queue();
225 }
226 
227 void
228 xpq_queue_machphys_update(paddr_t ma, paddr_t pa)
229 {
230 
231 	mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid];
232 	int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid];
233 
234 	XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64
235 	    "\n", (int64_t)ma, (int64_t)pa));
236 
237 	xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE;
238 	xpq_queue[xpq_idx].val = (pa - XPMAP_OFFSET) >> PAGE_SHIFT;
239 	xpq_increment_idx();
240 #ifdef XENDEBUG_SYNC
241 	xpq_flush_queue();
242 #endif
243 }
244 
245 void
246 xpq_queue_pte_update(paddr_t ptr, pt_entry_t val)
247 {
248 
249 	mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid];
250 	int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid];
251 
252 	KASSERT((ptr & 3) == 0);
253 	xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE;
254 	xpq_queue[xpq_idx].val = val;
255 	xpq_increment_idx();
256 #ifdef XENDEBUG_SYNC
257 	xpq_flush_queue();
258 #endif
259 }
260 
261 void
262 xpq_queue_pt_switch(paddr_t pa)
263 {
264 	struct mmuext_op op;
265 	xpq_flush_queue();
266 
267 	XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n",
268 	    (int64_t)pa, (int64_t)pa));
269 	op.cmd = MMUEXT_NEW_BASEPTR;
270 	op.arg1.mfn = pa >> PAGE_SHIFT;
271 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
272 		panic("xpq_queue_pt_switch");
273 }
274 
275 void
276 xpq_queue_pin_table(paddr_t pa, int lvl)
277 {
278 	struct mmuext_op op;
279 
280 	xpq_flush_queue();
281 
282 	XENPRINTK2(("xpq_queue_pin_l%d_table: %#" PRIxPADDR "\n",
283 	    lvl + 1, pa));
284 
285 	op.arg1.mfn = pa >> PAGE_SHIFT;
286 	op.cmd = lvl;
287 
288 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
289 		panic("xpq_queue_pin_table");
290 }
291 
292 void
293 xpq_queue_unpin_table(paddr_t pa)
294 {
295 	struct mmuext_op op;
296 
297 	xpq_flush_queue();
298 
299 	XENPRINTK2(("xpq_queue_unpin_table: %#" PRIxPADDR "\n", pa));
300 	op.arg1.mfn = pa >> PAGE_SHIFT;
301 	op.cmd = MMUEXT_UNPIN_TABLE;
302 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
303 		panic("xpq_queue_unpin_table");
304 }
305 
306 void
307 xpq_queue_set_ldt(vaddr_t va, uint32_t entries)
308 {
309 	struct mmuext_op op;
310 
311 	xpq_flush_queue();
312 
313 	XENPRINTK2(("xpq_queue_set_ldt\n"));
314 	KASSERT(va == (va & ~PAGE_MASK));
315 	op.cmd = MMUEXT_SET_LDT;
316 	op.arg1.linear_addr = va;
317 	op.arg2.nr_ents = entries;
318 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
319 		panic("xpq_queue_set_ldt");
320 }
321 
322 void
323 xpq_queue_tlb_flush(void)
324 {
325 	struct mmuext_op op;
326 
327 	xpq_flush_queue();
328 
329 	XENPRINTK2(("xpq_queue_tlb_flush\n"));
330 	op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
331 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
332 		panic("xpq_queue_tlb_flush");
333 }
334 
335 void
336 xpq_flush_cache(void)
337 {
338 	struct mmuext_op op;
339 	int s = splvm(), err;
340 
341 	xpq_flush_queue();
342 
343 	XENPRINTK2(("xpq_queue_flush_cache\n"));
344 	op.cmd = MMUEXT_FLUSH_CACHE;
345 	if ((err = HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) < 0) {
346 		panic("xpq_flush_cache, err %d", err);
347 	}
348 	splx(s); /* XXX: removeme */
349 }
350 
351 void
352 xpq_queue_invlpg(vaddr_t va)
353 {
354 	struct mmuext_op op;
355 	xpq_flush_queue();
356 
357 	XENPRINTK2(("xpq_queue_invlpg %#" PRIxVADDR "\n", va));
358 	op.cmd = MMUEXT_INVLPG_LOCAL;
359 	op.arg1.linear_addr = (va & ~PAGE_MASK);
360 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
361 		panic("xpq_queue_invlpg");
362 }
363 
364 void
365 xen_mcast_invlpg(vaddr_t va, uint32_t cpumask)
366 {
367 	mmuext_op_t op;
368 	u_long xcpumask = cpumask;
369 
370 	/* Flush pending page updates */
371 	xpq_flush_queue();
372 
373 	op.cmd = MMUEXT_INVLPG_MULTI;
374 	op.arg1.linear_addr = va;
375 	op.arg2.vcpumask = &xcpumask;
376 
377 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) {
378 		panic("xpq_queue_invlpg_all");
379 	}
380 
381 	return;
382 }
383 
384 void
385 xen_bcast_invlpg(vaddr_t va)
386 {
387 	mmuext_op_t op;
388 
389 	/* Flush pending page updates */
390 	xpq_flush_queue();
391 
392 	op.cmd = MMUEXT_INVLPG_ALL;
393 	op.arg1.linear_addr = va;
394 
395 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) {
396 		panic("xpq_queue_invlpg_all");
397 	}
398 
399 	return;
400 }
401 
402 /* This is a synchronous call. */
403 void
404 xen_mcast_tlbflush(uint32_t cpumask)
405 {
406 	mmuext_op_t op;
407 	u_long xcpumask = cpumask;
408 
409 	/* Flush pending page updates */
410 	xpq_flush_queue();
411 
412 	op.cmd = MMUEXT_TLB_FLUSH_MULTI;
413 	op.arg2.vcpumask = &xcpumask;
414 
415 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) {
416 		panic("xpq_queue_invlpg_all");
417 	}
418 
419 	return;
420 }
421 
422 /* This is a synchronous call. */
423 void
424 xen_bcast_tlbflush(void)
425 {
426 	mmuext_op_t op;
427 
428 	/* Flush pending page updates */
429 	xpq_flush_queue();
430 
431 	op.cmd = MMUEXT_TLB_FLUSH_ALL;
432 
433 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) {
434 		panic("xpq_queue_invlpg_all");
435 	}
436 
437 	return;
438 }
439 
440 /* This is a synchronous call. */
441 void
442 xen_vcpu_mcast_invlpg(vaddr_t sva, vaddr_t eva, uint32_t cpumask)
443 {
444 	KASSERT(eva > sva);
445 
446 	/* Flush pending page updates */
447 	xpq_flush_queue();
448 
449 	/* Align to nearest page boundary */
450 	sva &= ~PAGE_MASK;
451 	eva &= ~PAGE_MASK;
452 
453 	for ( ; sva <= eva; sva += PAGE_SIZE) {
454 		xen_mcast_invlpg(sva, cpumask);
455 	}
456 
457 	return;
458 }
459 
460 /* This is a synchronous call. */
461 void
462 xen_vcpu_bcast_invlpg(vaddr_t sva, vaddr_t eva)
463 {
464 	KASSERT(eva > sva);
465 
466 	/* Flush pending page updates */
467 	xpq_flush_queue();
468 
469 	/* Align to nearest page boundary */
470 	sva &= ~PAGE_MASK;
471 	eva &= ~PAGE_MASK;
472 
473 	for ( ; sva <= eva; sva += PAGE_SIZE) {
474 		xen_bcast_invlpg(sva);
475 	}
476 
477 	return;
478 }
479 
480 int
481 xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom)
482 {
483 	mmu_update_t op;
484 	int ok;
485 
486 	xpq_flush_queue();
487 
488 	op.ptr = ptr;
489 	op.val = val;
490 	if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0)
491 		return EFAULT;
492 	return (0);
493 }
494 
495 #ifdef XENDEBUG
496 void
497 xpq_debug_dump(void)
498 {
499 	int i;
500 
501 	mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid];
502 	int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid];
503 
504 	XENPRINTK2(("idx: %d\n", xpq_idx));
505 	for (i = 0; i < xpq_idx; i++) {
506 		snprintf(XBUF, sizeof(XBUF), "%" PRIx64 " %08" PRIx64,
507 		    xpq_queue[i].ptr, xpq_queue[i].val);
508 		if (++i < xpq_idx)
509 			snprintf(XBUF + strlen(XBUF),
510 			    sizeof(XBUF) - strlen(XBUF),
511 			    "%" PRIx64 " %08" PRIx64,
512 			    xpq_queue[i].ptr, xpq_queue[i].val);
513 		if (++i < xpq_idx)
514 			snprintf(XBUF + strlen(XBUF),
515 			    sizeof(XBUF) - strlen(XBUF),
516 			    "%" PRIx64 " %08" PRIx64,
517 			    xpq_queue[i].ptr, xpq_queue[i].val);
518 		if (++i < xpq_idx)
519 			snprintf(XBUF + strlen(XBUF),
520 			    sizeof(XBUF) - strlen(XBUF),
521 			    "%" PRIx64 " %08" PRIx64,
522 			    xpq_queue[i].ptr, xpq_queue[i].val);
523 		XENPRINTK2(("%d: %s\n", xpq_idx, XBUF));
524 	}
525 }
526 #endif
527 
528 
529 extern volatile struct xencons_interface *xencons_interface; /* XXX */
530 extern struct xenstore_domain_interface *xenstore_interface; /* XXX */
531 
532 static void xen_bt_set_readonly (vaddr_t);
533 static void xen_bootstrap_tables (vaddr_t, vaddr_t, int, int, int);
534 
535 /* How many PDEs ? */
536 #if L2_SLOT_KERNBASE > 0
537 #define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1))
538 #else
539 #define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1)
540 #endif
541 
542 /*
543  * Construct and switch to new pagetables
544  * first_avail is the first vaddr we can use after
545  * we get rid of Xen pagetables
546  */
547 
548 vaddr_t xen_pmap_bootstrap (void);
549 
550 /*
551  * Function to get rid of Xen bootstrap tables
552  */
553 
554 /* How many PDP do we need: */
555 #ifdef PAE
556 /*
557  * For PAE, we consider a single contigous L2 "superpage" of 4 pages,
558  * all of them mapped by the L3 page. We also need a shadow page
559  * for L3[3].
560  */
561 static const int l2_4_count = 6;
562 #elif defined(__x86_64__)
563 static const int l2_4_count = PTP_LEVELS;
564 #else
565 static const int l2_4_count = PTP_LEVELS - 1;
566 #endif
567 
568 vaddr_t
569 xen_pmap_bootstrap(void)
570 {
571 	int count, oldcount;
572 	long mapsize;
573 	vaddr_t bootstrap_tables, init_tables;
574 
575 	memset(xpq_idx_array, 0, sizeof xpq_idx_array);
576 
577 	xpmap_phys_to_machine_mapping =
578 	    (unsigned long *)xen_start_info.mfn_list;
579 	init_tables = xen_start_info.pt_base;
580 	__PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables));
581 
582 	/* Space after Xen boostrap tables should be free */
583 	bootstrap_tables = xen_start_info.pt_base +
584 		(xen_start_info.nr_pt_frames * PAGE_SIZE);
585 
586 	/*
587 	 * Calculate how many space we need
588 	 * first everything mapped before the Xen bootstrap tables
589 	 */
590 	mapsize = init_tables - KERNTEXTOFF;
591 	/* after the tables we'll have:
592 	 *  - UAREA
593 	 *  - dummy user PGD (x86_64)
594 	 *  - HYPERVISOR_shared_info
595 	 *  - early_zerop
596 	 *  - ISA I/O mem (if needed)
597 	 */
598 	mapsize += UPAGES * NBPG;
599 #ifdef __x86_64__
600 	mapsize += NBPG;
601 #endif
602 	mapsize += NBPG;
603 	mapsize += NBPG;
604 
605 #ifdef DOM0OPS
606 	if (xendomain_is_dom0()) {
607 		/* space for ISA I/O mem */
608 		mapsize += IOM_SIZE;
609 	}
610 #endif
611 	/* at this point mapsize doens't include the table size */
612 
613 #ifdef __x86_64__
614 	count = TABLE_L2_ENTRIES;
615 #else
616 	count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT;
617 #endif /* __x86_64__ */
618 
619 	/* now compute how many L2 pages we need exactly */
620 	XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count));
621 	while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF >
622 	    ((long)count << L2_SHIFT) + KERNBASE) {
623 		count++;
624 	}
625 #ifndef __x86_64__
626 	/*
627 	 * one more L2 page: we'll alocate several pages after kva_start
628 	 * in pmap_bootstrap() before pmap_growkernel(), which have not been
629 	 * counted here. It's not a big issue to allocate one more L2 as
630 	 * pmap_growkernel() will be called anyway.
631 	 */
632 	count++;
633 	nkptp[1] = count;
634 #endif
635 
636 	/*
637 	 * install bootstrap pages. We may need more L2 pages than will
638 	 * have the final table here, as it's installed after the final table
639 	 */
640 	oldcount = count;
641 
642 bootstrap_again:
643 	XENPRINTK(("bootstrap_again oldcount %d\n", oldcount));
644 	/*
645 	 * Xen space we'll reclaim may not be enough for our new page tables,
646 	 * move bootstrap tables if necessary
647 	 */
648 	if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE))
649 		bootstrap_tables = init_tables +
650 					((count + l2_4_count) * PAGE_SIZE);
651 	/* make sure we have enough to map the bootstrap_tables */
652 	if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) >
653 	    ((long)oldcount << L2_SHIFT) + KERNBASE) {
654 		oldcount++;
655 		goto bootstrap_again;
656 	}
657 
658 	/* Create temporary tables */
659 	xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables,
660 		xen_start_info.nr_pt_frames, oldcount, 0);
661 
662 	/* Create final tables */
663 	xen_bootstrap_tables(bootstrap_tables, init_tables,
664 	    oldcount + l2_4_count, count, 1);
665 
666 	/* zero out free space after tables */
667 	memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0,
668 	    (UPAGES + 1) * NBPG);
669 
670 	/* Finally, flush TLB. */
671 	xpq_queue_tlb_flush();
672 
673 	return (init_tables + ((count + l2_4_count) * PAGE_SIZE));
674 }
675 
676 /*
677  * Build a new table and switch to it
678  * old_count is # of old tables (including PGD, PDTPE and PDE)
679  * new_count is # of new tables (PTE only)
680  * we assume areas don't overlap
681  */
682 static void
683 xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd,
684 	int old_count, int new_count, int final)
685 {
686 	pd_entry_t *pdtpe, *pde, *pte;
687 	pd_entry_t *cur_pgd, *bt_pgd;
688 	paddr_t addr;
689 	vaddr_t page, avail, text_end, map_end;
690 	int i;
691 	extern char __data_start;
692 	extern char *early_zerop; /* from pmap.c */
693 
694 	__PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR ","
695 	    " %d, %d)\n",
696 	    old_pgd, new_pgd, old_count, new_count));
697 	text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK;
698 	/*
699 	 * size of R/W area after kernel text:
700 	 *  xencons_interface (if present)
701 	 *  xenstore_interface (if present)
702 	 *  table pages (new_count + l2_4_count entries)
703 	 * extra mappings (only when final is true):
704 	 *  UAREA
705 	 *  dummy user PGD (x86_64 only)/gdt page (i386 only)
706 	 *  HYPERVISOR_shared_info
707 	 *  early_zerop
708 	 *  ISA I/O mem (if needed)
709 	 */
710 	map_end = new_pgd + ((new_count + l2_4_count) * NBPG);
711 	if (final) {
712 		map_end += (UPAGES + 1) * NBPG;
713 		HYPERVISOR_shared_info = (shared_info_t *)map_end;
714 		map_end += NBPG;
715 		early_zerop = (char *)map_end;
716 		map_end += NBPG;
717 	}
718 	/*
719 	 * we always set atdevbase, as it's used by init386 to find the first
720 	 * available VA. map_end is updated only if we are dom0, so
721 	 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in
722 	 * this case.
723 	 */
724 	if (final)
725 		atdevbase = map_end;
726 #ifdef DOM0OPS
727 	if (final && xendomain_is_dom0()) {
728 		/* ISA I/O mem */
729 		map_end += IOM_SIZE;
730 	}
731 #endif /* DOM0OPS */
732 
733 	__PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n",
734 	    text_end, map_end));
735 	__PRINTK(("console %#lx ", xen_start_info.console_mfn));
736 	__PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn));
737 
738 	/*
739 	 * Create bootstrap page tables
740 	 * What we need:
741 	 * - a PGD (level 4)
742 	 * - a PDTPE (level 3)
743 	 * - a PDE (level2)
744 	 * - some PTEs (level 1)
745 	 */
746 
747 	cur_pgd = (pd_entry_t *) old_pgd;
748 	bt_pgd = (pd_entry_t *) new_pgd;
749 	memset (bt_pgd, 0, PAGE_SIZE);
750 	avail = new_pgd + PAGE_SIZE;
751 #if PTP_LEVELS > 3
752 	/* per-cpu L4 PD */
753 	pd_entry_t *bt_cpu_pgd = bt_pgd;
754 	/* pmap_kernel() "shadow" L4 PD */
755 	bt_pgd = (pd_entry_t *) avail;
756 	memset(bt_pgd, 0, PAGE_SIZE);
757 	avail += PAGE_SIZE;
758 
759 	/* Install level 3 */
760 	pdtpe = (pd_entry_t *) avail;
761 	memset (pdtpe, 0, PAGE_SIZE);
762 	avail += PAGE_SIZE;
763 
764 	addr = ((u_long) pdtpe) - KERNBASE;
765 	bt_pgd[pl4_pi(KERNTEXTOFF)] = bt_cpu_pgd[pl4_pi(KERNTEXTOFF)] =
766 	    xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V;
767 
768 	__PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR
769 	    " -> L4[%#x]\n",
770 	    pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF)));
771 #else
772 	pdtpe = bt_pgd;
773 #endif /* PTP_LEVELS > 3 */
774 
775 #if PTP_LEVELS > 2
776 	/* Level 2 */
777 	pde = (pd_entry_t *) avail;
778 	memset(pde, 0, PAGE_SIZE);
779 	avail += PAGE_SIZE;
780 
781 	addr = ((u_long) pde) - KERNBASE;
782 	pdtpe[pl3_pi(KERNTEXTOFF)] =
783 	    xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW;
784 	__PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR
785 	    " -> L3[%#x]\n",
786 	    pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF)));
787 #elif defined(PAE)
788 	/* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */
789 	pde = (pd_entry_t *) avail;
790 	memset(pde, 0, PAGE_SIZE * 5);
791 	avail += PAGE_SIZE * 5;
792 	addr = ((u_long) pde) - KERNBASE;
793 	/*
794 	 * enter L2 pages in the L3.
795 	 * The real L2 kernel PD will be the last one (so that
796 	 * pde[L2_SLOT_KERN] always point to the shadow).
797 	 */
798 	for (i = 0; i < 3; i++, addr += PAGE_SIZE) {
799 		/*
800 		 * Xen doesn't want R/W mappings in L3 entries, it'll add it
801 		 * itself.
802 		 */
803 		pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V;
804 		__PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR
805 		    " -> L3[%#x]\n",
806 		    (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i));
807 	}
808 	addr += PAGE_SIZE;
809 	pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V;
810 	__PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR
811 	    " -> L3[%#x]\n",
812 	    (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3));
813 
814 #else /* PAE */
815 	pde = bt_pgd;
816 #endif /* PTP_LEVELS > 2 */
817 
818 	/* Level 1 */
819 	page = KERNTEXTOFF;
820 	for (i = 0; i < new_count; i ++) {
821 		vaddr_t cur_page = page;
822 
823 		pte = (pd_entry_t *) avail;
824 		avail += PAGE_SIZE;
825 
826 		memset(pte, 0, PAGE_SIZE);
827 		while (pl2_pi(page) == pl2_pi (cur_page)) {
828 			if (page >= map_end) {
829 				/* not mapped at all */
830 				pte[pl1_pi(page)] = 0;
831 				page += PAGE_SIZE;
832 				continue;
833 			}
834 			pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE);
835 			if (page == (vaddr_t)HYPERVISOR_shared_info) {
836 				pte[pl1_pi(page)] = xen_start_info.shared_info;
837 				__PRINTK(("HYPERVISOR_shared_info "
838 				    "va %#lx pte %#" PRIxPADDR "\n",
839 				    HYPERVISOR_shared_info, pte[pl1_pi(page)]));
840 			}
841 			if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT)
842 			    == xen_start_info.console.domU.mfn) {
843 				xencons_interface = (void *)page;
844 				pte[pl1_pi(page)] = xen_start_info.console_mfn;
845 				pte[pl1_pi(page)] <<= PAGE_SHIFT;
846 				__PRINTK(("xencons_interface "
847 				    "va %#lx pte %#" PRIxPADDR "\n",
848 				    xencons_interface, pte[pl1_pi(page)]));
849 			}
850 			if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT)
851 			    == xen_start_info.store_mfn) {
852 				xenstore_interface = (void *)page;
853 				pte[pl1_pi(page)] = xen_start_info.store_mfn;
854 				pte[pl1_pi(page)] <<= PAGE_SHIFT;
855 				__PRINTK(("xenstore_interface "
856 				    "va %#lx pte %#" PRIxPADDR "\n",
857 				    xenstore_interface, pte[pl1_pi(page)]));
858 			}
859 #ifdef DOM0OPS
860 			if (page >= (vaddr_t)atdevbase &&
861 			    page < (vaddr_t)atdevbase + IOM_SIZE) {
862 				pte[pl1_pi(page)] =
863 				    IOM_BEGIN + (page - (vaddr_t)atdevbase);
864 			}
865 #endif
866 			pte[pl1_pi(page)] |= PG_k | PG_V;
867 			if (page < text_end) {
868 				/* map kernel text RO */
869 				pte[pl1_pi(page)] |= 0;
870 			} else if (page >= old_pgd
871 			    && page < old_pgd + (old_count * PAGE_SIZE)) {
872 				/* map old page tables RO */
873 				pte[pl1_pi(page)] |= 0;
874 			} else if (page >= new_pgd &&
875 			    page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) {
876 				/* map new page tables RO */
877 				pte[pl1_pi(page)] |= 0;
878 #ifdef i386
879 			} else if (page == (vaddr_t)tmpgdt) {
880 				/*
881 				 * Map bootstrap gdt R/O. Later, we
882 				 * will re-add this to page to uvm
883 				 * after making it writable.
884 				 */
885 
886 				pte[pl1_pi(page)] = 0;
887 				page += PAGE_SIZE;
888 				continue;
889 #endif /* i386 */
890 			} else {
891 				/* map page RW */
892 				pte[pl1_pi(page)] |= PG_RW;
893 			}
894 
895 			if ((page  >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE))
896 			    || page >= new_pgd) {
897 				__PRINTK(("va %#lx pa %#lx "
898 				    "entry 0x%" PRIxPADDR " -> L1[%#x]\n",
899 				    page, page - KERNBASE,
900 				    pte[pl1_pi(page)], pl1_pi(page)));
901 			}
902 			page += PAGE_SIZE;
903 		}
904 
905 		addr = ((u_long) pte) - KERNBASE;
906 		pde[pl2_pi(cur_page)] =
907 		    xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V;
908 		__PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR
909 		    " -> L2[%#x]\n",
910 		    pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page)));
911 		/* Mark readonly */
912 		xen_bt_set_readonly((vaddr_t) pte);
913 	}
914 
915 	/* Install recursive page tables mapping */
916 #ifdef PAE
917 	/*
918 	 * we need a shadow page for the kernel's L2 page
919 	 * The real L2 kernel PD will be the last one (so that
920 	 * pde[L2_SLOT_KERN] always point to the shadow.
921 	 */
922 	memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE);
923 	cpu_info_primary.ci_kpm_pdir = &pde[L2_SLOT_KERN + NPDPG];
924 	cpu_info_primary.ci_kpm_pdirpa =
925 	    (vaddr_t) cpu_info_primary.ci_kpm_pdir - KERNBASE;
926 
927 	/*
928 	 * We don't enter a recursive entry from the L3 PD. Instead,
929 	 * we enter the first 4 L2 pages, which includes the kernel's L2
930 	 * shadow. But we have to entrer the shadow after switching
931 	 * %cr3, or Xen will refcount some PTE with the wrong type.
932 	 */
933 	addr = (u_long)pde - KERNBASE;
934 	for (i = 0; i < 3; i++, addr += PAGE_SIZE) {
935 		pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V;
936 		__PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR
937 		    " entry %#" PRIxPADDR "\n",
938 		    (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i,
939 		    addr, pde[PDIR_SLOT_PTE + i]));
940 	}
941 #if 0
942 	addr += PAGE_SIZE; /* point to shadow L2 */
943 	pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V;
944 	__PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n",
945 	    (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr,
946 	    (int64_t)pde[PDIR_SLOT_PTE + 3]));
947 #endif
948 	/* Mark tables RO, and pin the kernel's shadow as L2 */
949 	addr = (u_long)pde - KERNBASE;
950 	for (i = 0; i < 5; i++, addr += PAGE_SIZE) {
951 		xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i);
952 		if (i == 2 || i == 3)
953 			continue;
954 #if 0
955 		__PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr));
956 		xpq_queue_pin_l2_table(xpmap_ptom_masked(addr));
957 #endif
958 	}
959 	if (final) {
960 		addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE;
961 		__PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr));
962 		xpq_queue_pin_l2_table(xpmap_ptom_masked(addr));
963 	}
964 #if 0
965 	addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE;
966 	__PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr));
967 	xpq_queue_pin_l2_table(xpmap_ptom_masked(addr));
968 #endif
969 #else /* PAE */
970 	/* recursive entry in higher-level per-cpu PD and pmap_kernel() */
971 	bt_pgd[PDIR_SLOT_PTE] = xpmap_ptom_masked((paddr_t)bt_pgd - KERNBASE) | PG_k | PG_V;
972 #ifdef __x86_64__
973 	   bt_cpu_pgd[PDIR_SLOT_PTE] =
974 		   xpmap_ptom_masked((paddr_t)bt_cpu_pgd - KERNBASE) | PG_k | PG_V;
975 #endif /* __x86_64__ */
976 	__PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR
977 	    " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE,
978 	    bt_pgd[PDIR_SLOT_PTE]));
979 	/* Mark tables RO */
980 	xen_bt_set_readonly((vaddr_t) pde);
981 #endif
982 #if PTP_LEVELS > 2 || defined(PAE)
983 	xen_bt_set_readonly((vaddr_t) pdtpe);
984 #endif
985 #if PTP_LEVELS > 3
986 	xen_bt_set_readonly(new_pgd);
987 #endif
988 	/* Pin the PGD */
989 	__PRINTK(("pin PGD: %"PRIxVADDR"\n", new_pgd - KERNBASE));
990 #ifdef __x86_64__
991 	xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE));
992 #elif PAE
993 	xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE));
994 #else
995 	xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE));
996 #endif
997 
998 	/* Save phys. addr of PDP, for libkvm. */
999 #ifdef PAE
1000 	PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */
1001 #else
1002 	PDPpaddr = (u_long)bt_pgd - KERNBASE;
1003 #endif
1004 
1005 	/* Switch to new tables */
1006 	__PRINTK(("switch to PGD\n"));
1007 	xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE));
1008 	__PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n",
1009 	    bt_pgd[PDIR_SLOT_PTE]));
1010 
1011 #ifdef PAE
1012 	if (final) {
1013 		/* save the address of the L3 page */
1014 		cpu_info_primary.ci_pae_l3_pdir = pdtpe;
1015 		cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE);
1016 
1017 		/* now enter kernel's PTE mappings */
1018 		addr =  (u_long)pde - KERNBASE + PAGE_SIZE * 3;
1019 		xpq_queue_pte_update(
1020 		    xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE),
1021 		    xpmap_ptom_masked(addr) | PG_k | PG_V);
1022 		xpq_flush_queue();
1023 	}
1024 #elif defined(__x86_64__)
1025 	if (final) {
1026 		/* save the address of the real per-cpu L4 pgd page */
1027 		cpu_info_primary.ci_kpm_pdir = bt_cpu_pgd;
1028 		cpu_info_primary.ci_kpm_pdirpa = ((paddr_t) bt_cpu_pgd - KERNBASE);
1029 	}
1030 #endif
1031 
1032 	/* Now we can safely reclaim space taken by old tables */
1033 
1034 	__PRINTK(("unpin old PGD\n"));
1035 	/* Unpin old PGD */
1036 	xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE));
1037 	/* Mark old tables RW */
1038 	page = old_pgd;
1039 	addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME;
1040 	addr = xpmap_mtop(addr);
1041 	pte = (pd_entry_t *) ((u_long)addr + KERNBASE);
1042 	pte += pl1_pi(page);
1043 	__PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n",
1044 	    pde[pl2_pi(page)], addr, (long)pte));
1045 	while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) {
1046 		addr = xpmap_ptom(((u_long) pte) - KERNBASE);
1047 		XENPRINTK(("addr %#" PRIxPADDR " pte %#lx "
1048 		   "*pte %#" PRIxPADDR "\n",
1049 		   addr, (long)pte, *pte));
1050 		xpq_queue_pte_update(addr, *pte | PG_RW);
1051 		page += PAGE_SIZE;
1052 		/*
1053 		 * Our ptes are contiguous
1054 		 * so it's safe to just "++" here
1055 		 */
1056 		pte++;
1057 	}
1058 	xpq_flush_queue();
1059 }
1060 
1061 
1062 /*
1063  * Bootstrap helper functions
1064  */
1065 
1066 /*
1067  * Mark a page readonly
1068  * XXX: assuming vaddr = paddr + KERNBASE
1069  */
1070 
1071 static void
1072 xen_bt_set_readonly (vaddr_t page)
1073 {
1074 	pt_entry_t entry;
1075 
1076 	entry = xpmap_ptom_masked(page - KERNBASE);
1077 	entry |= PG_k | PG_V;
1078 
1079 	HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG);
1080 }
1081 
1082 #ifdef __x86_64__
1083 void
1084 xen_set_user_pgd(paddr_t page)
1085 {
1086 	struct mmuext_op op;
1087 	int s = splvm();
1088 
1089 	xpq_flush_queue();
1090 	op.cmd = MMUEXT_NEW_USER_BASEPTR;
1091 	op.arg1.mfn = pfn_to_mfn(page >> PAGE_SHIFT);
1092         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0)
1093 		panic("xen_set_user_pgd: failed to install new user page"
1094 			" directory %#" PRIxPADDR, page);
1095 	splx(s);
1096 }
1097 #endif /* __x86_64__ */
1098