xref: /openbmc/linux/arch/x86/xen/setup.c (revision da1d9caf)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Machine specific setup for xen
4  *
5  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
6  */
7 
8 #include <linux/init.h>
9 #include <linux/sched.h>
10 #include <linux/mm.h>
11 #include <linux/pm.h>
12 #include <linux/memblock.h>
13 #include <linux/cpuidle.h>
14 #include <linux/cpufreq.h>
15 #include <linux/memory_hotplug.h>
16 
17 #include <asm/elf.h>
18 #include <asm/vdso.h>
19 #include <asm/e820/api.h>
20 #include <asm/setup.h>
21 #include <asm/acpi.h>
22 #include <asm/numa.h>
23 #include <asm/idtentry.h>
24 #include <asm/xen/hypervisor.h>
25 #include <asm/xen/hypercall.h>
26 
27 #include <xen/xen.h>
28 #include <xen/page.h>
29 #include <xen/interface/callback.h>
30 #include <xen/interface/memory.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/features.h>
33 #include <xen/hvc-console.h>
34 #include "xen-ops.h"
35 #include "mmu.h"
36 
37 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
38 
39 /* Amount of extra memory space we add to the e820 ranges */
40 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
41 
42 /* Number of pages released from the initial allocation. */
43 unsigned long xen_released_pages;
44 
45 /* E820 map used during setting up memory. */
46 static struct e820_table xen_e820_table __initdata;
47 
48 /*
49  * Buffer used to remap identity mapped pages. We only need the virtual space.
50  * The physical page behind this address is remapped as needed to different
51  * buffer pages.
52  */
53 #define REMAP_SIZE	(P2M_PER_PAGE - 3)
54 static struct {
55 	unsigned long	next_area_mfn;
56 	unsigned long	target_pfn;
57 	unsigned long	size;
58 	unsigned long	mfns[REMAP_SIZE];
59 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
60 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
61 
62 /*
63  * The maximum amount of extra memory compared to the base size.  The
64  * main scaling factor is the size of struct page.  At extreme ratios
65  * of base:extra, all the base memory can be filled with page
66  * structures for the extra memory, leaving no space for anything
67  * else.
68  *
69  * 10x seems like a reasonable balance between scaling flexibility and
70  * leaving a practically usable system.
71  */
72 #define EXTRA_MEM_RATIO		(10)
73 
74 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
75 
76 static void __init xen_parse_512gb(void)
77 {
78 	bool val = false;
79 	char *arg;
80 
81 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
82 	if (!arg)
83 		return;
84 
85 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
86 	if (!arg)
87 		val = true;
88 	else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
89 		return;
90 
91 	xen_512gb_limit = val;
92 }
93 
94 static void __init xen_add_extra_mem(unsigned long start_pfn,
95 				     unsigned long n_pfns)
96 {
97 	int i;
98 
99 	/*
100 	 * No need to check for zero size, should happen rarely and will only
101 	 * write a new entry regarded to be unused due to zero size.
102 	 */
103 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
104 		/* Add new region. */
105 		if (xen_extra_mem[i].n_pfns == 0) {
106 			xen_extra_mem[i].start_pfn = start_pfn;
107 			xen_extra_mem[i].n_pfns = n_pfns;
108 			break;
109 		}
110 		/* Append to existing region. */
111 		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
112 		    start_pfn) {
113 			xen_extra_mem[i].n_pfns += n_pfns;
114 			break;
115 		}
116 	}
117 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
118 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
119 
120 	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
121 }
122 
123 static void __init xen_del_extra_mem(unsigned long start_pfn,
124 				     unsigned long n_pfns)
125 {
126 	int i;
127 	unsigned long start_r, size_r;
128 
129 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
130 		start_r = xen_extra_mem[i].start_pfn;
131 		size_r = xen_extra_mem[i].n_pfns;
132 
133 		/* Start of region. */
134 		if (start_r == start_pfn) {
135 			BUG_ON(n_pfns > size_r);
136 			xen_extra_mem[i].start_pfn += n_pfns;
137 			xen_extra_mem[i].n_pfns -= n_pfns;
138 			break;
139 		}
140 		/* End of region. */
141 		if (start_r + size_r == start_pfn + n_pfns) {
142 			BUG_ON(n_pfns > size_r);
143 			xen_extra_mem[i].n_pfns -= n_pfns;
144 			break;
145 		}
146 		/* Mid of region. */
147 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
148 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
149 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
150 			/* Calling memblock_reserve() again is okay. */
151 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
152 					  (start_pfn + n_pfns));
153 			break;
154 		}
155 	}
156 	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
157 }
158 
159 /*
160  * Called during boot before the p2m list can take entries beyond the
161  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
162  * invalid.
163  */
164 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
165 {
166 	int i;
167 
168 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
169 		if (pfn >= xen_extra_mem[i].start_pfn &&
170 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
171 			return INVALID_P2M_ENTRY;
172 	}
173 
174 	return IDENTITY_FRAME(pfn);
175 }
176 
177 /*
178  * Mark all pfns of extra mem as invalid in p2m list.
179  */
180 void __init xen_inv_extra_mem(void)
181 {
182 	unsigned long pfn, pfn_s, pfn_e;
183 	int i;
184 
185 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
186 		if (!xen_extra_mem[i].n_pfns)
187 			continue;
188 		pfn_s = xen_extra_mem[i].start_pfn;
189 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
190 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
191 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
192 	}
193 }
194 
195 /*
196  * Finds the next RAM pfn available in the E820 map after min_pfn.
197  * This function updates min_pfn with the pfn found and returns
198  * the size of that range or zero if not found.
199  */
200 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
201 {
202 	const struct e820_entry *entry = xen_e820_table.entries;
203 	unsigned int i;
204 	unsigned long done = 0;
205 
206 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
207 		unsigned long s_pfn;
208 		unsigned long e_pfn;
209 
210 		if (entry->type != E820_TYPE_RAM)
211 			continue;
212 
213 		e_pfn = PFN_DOWN(entry->addr + entry->size);
214 
215 		/* We only care about E820 after this */
216 		if (e_pfn <= *min_pfn)
217 			continue;
218 
219 		s_pfn = PFN_UP(entry->addr);
220 
221 		/* If min_pfn falls within the E820 entry, we want to start
222 		 * at the min_pfn PFN.
223 		 */
224 		if (s_pfn <= *min_pfn) {
225 			done = e_pfn - *min_pfn;
226 		} else {
227 			done = e_pfn - s_pfn;
228 			*min_pfn = s_pfn;
229 		}
230 		break;
231 	}
232 
233 	return done;
234 }
235 
236 static int __init xen_free_mfn(unsigned long mfn)
237 {
238 	struct xen_memory_reservation reservation = {
239 		.address_bits = 0,
240 		.extent_order = 0,
241 		.domid        = DOMID_SELF
242 	};
243 
244 	set_xen_guest_handle(reservation.extent_start, &mfn);
245 	reservation.nr_extents = 1;
246 
247 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
248 }
249 
250 /*
251  * This releases a chunk of memory and then does the identity map. It's used
252  * as a fallback if the remapping fails.
253  */
254 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
255 			unsigned long end_pfn, unsigned long nr_pages)
256 {
257 	unsigned long pfn, end;
258 	int ret;
259 
260 	WARN_ON(start_pfn > end_pfn);
261 
262 	/* Release pages first. */
263 	end = min(end_pfn, nr_pages);
264 	for (pfn = start_pfn; pfn < end; pfn++) {
265 		unsigned long mfn = pfn_to_mfn(pfn);
266 
267 		/* Make sure pfn exists to start with */
268 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
269 			continue;
270 
271 		ret = xen_free_mfn(mfn);
272 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
273 
274 		if (ret == 1) {
275 			xen_released_pages++;
276 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
277 				break;
278 		} else
279 			break;
280 	}
281 
282 	set_phys_range_identity(start_pfn, end_pfn);
283 }
284 
285 /*
286  * Helper function to update the p2m and m2p tables and kernel mapping.
287  */
288 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
289 {
290 	struct mmu_update update = {
291 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
292 		.val = pfn
293 	};
294 
295 	/* Update p2m */
296 	if (!set_phys_to_machine(pfn, mfn)) {
297 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
298 		     pfn, mfn);
299 		BUG();
300 	}
301 
302 	/* Update m2p */
303 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
304 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
305 		     mfn, pfn);
306 		BUG();
307 	}
308 
309 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
310 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
311 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
312 		      mfn, pfn);
313 		BUG();
314 	}
315 }
316 
317 /*
318  * This function updates the p2m and m2p tables with an identity map from
319  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
320  * original allocation at remap_pfn. The information needed for remapping is
321  * saved in the memory itself to avoid the need for allocating buffers. The
322  * complete remap information is contained in a list of MFNs each containing
323  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
324  * This enables us to preserve the original mfn sequence while doing the
325  * remapping at a time when the memory management is capable of allocating
326  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
327  * its callers.
328  */
329 static void __init xen_do_set_identity_and_remap_chunk(
330         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
331 {
332 	unsigned long buf = (unsigned long)&xen_remap_buf;
333 	unsigned long mfn_save, mfn;
334 	unsigned long ident_pfn_iter, remap_pfn_iter;
335 	unsigned long ident_end_pfn = start_pfn + size;
336 	unsigned long left = size;
337 	unsigned int i, chunk;
338 
339 	WARN_ON(size == 0);
340 
341 	mfn_save = virt_to_mfn(buf);
342 
343 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
344 	     ident_pfn_iter < ident_end_pfn;
345 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
346 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
347 
348 		/* Map first pfn to xen_remap_buf */
349 		mfn = pfn_to_mfn(ident_pfn_iter);
350 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
351 
352 		/* Save mapping information in page */
353 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
354 		xen_remap_buf.target_pfn = remap_pfn_iter;
355 		xen_remap_buf.size = chunk;
356 		for (i = 0; i < chunk; i++)
357 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
358 
359 		/* Put remap buf into list. */
360 		xen_remap_mfn = mfn;
361 
362 		/* Set identity map */
363 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
364 
365 		left -= chunk;
366 	}
367 
368 	/* Restore old xen_remap_buf mapping */
369 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
370 }
371 
372 /*
373  * This function takes a contiguous pfn range that needs to be identity mapped
374  * and:
375  *
376  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
377  *  2) Calls the do_ function to actually do the mapping/remapping work.
378  *
379  * The goal is to not allocate additional memory but to remap the existing
380  * pages. In the case of an error the underlying memory is simply released back
381  * to Xen and not remapped.
382  */
383 static unsigned long __init xen_set_identity_and_remap_chunk(
384 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
385 	unsigned long remap_pfn)
386 {
387 	unsigned long pfn;
388 	unsigned long i = 0;
389 	unsigned long n = end_pfn - start_pfn;
390 
391 	if (remap_pfn == 0)
392 		remap_pfn = nr_pages;
393 
394 	while (i < n) {
395 		unsigned long cur_pfn = start_pfn + i;
396 		unsigned long left = n - i;
397 		unsigned long size = left;
398 		unsigned long remap_range_size;
399 
400 		/* Do not remap pages beyond the current allocation */
401 		if (cur_pfn >= nr_pages) {
402 			/* Identity map remaining pages */
403 			set_phys_range_identity(cur_pfn, cur_pfn + size);
404 			break;
405 		}
406 		if (cur_pfn + size > nr_pages)
407 			size = nr_pages - cur_pfn;
408 
409 		remap_range_size = xen_find_pfn_range(&remap_pfn);
410 		if (!remap_range_size) {
411 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
412 			xen_set_identity_and_release_chunk(cur_pfn,
413 						cur_pfn + left, nr_pages);
414 			break;
415 		}
416 		/* Adjust size to fit in current e820 RAM region */
417 		if (size > remap_range_size)
418 			size = remap_range_size;
419 
420 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
421 
422 		/* Update variables to reflect new mappings. */
423 		i += size;
424 		remap_pfn += size;
425 	}
426 
427 	/*
428 	 * If the PFNs are currently mapped, their VA mappings need to be
429 	 * zapped.
430 	 */
431 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
432 		(void)HYPERVISOR_update_va_mapping(
433 			(unsigned long)__va(pfn << PAGE_SHIFT),
434 			native_make_pte(0), 0);
435 
436 	return remap_pfn;
437 }
438 
439 static unsigned long __init xen_count_remap_pages(
440 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
441 	unsigned long remap_pages)
442 {
443 	if (start_pfn >= nr_pages)
444 		return remap_pages;
445 
446 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
447 }
448 
449 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
450 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
451 			      unsigned long nr_pages, unsigned long last_val))
452 {
453 	phys_addr_t start = 0;
454 	unsigned long ret_val = 0;
455 	const struct e820_entry *entry = xen_e820_table.entries;
456 	int i;
457 
458 	/*
459 	 * Combine non-RAM regions and gaps until a RAM region (or the
460 	 * end of the map) is reached, then call the provided function
461 	 * to perform its duty on the non-RAM region.
462 	 *
463 	 * The combined non-RAM regions are rounded to a whole number
464 	 * of pages so any partial pages are accessible via the 1:1
465 	 * mapping.  This is needed for some BIOSes that put (for
466 	 * example) the DMI tables in a reserved region that begins on
467 	 * a non-page boundary.
468 	 */
469 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
470 		phys_addr_t end = entry->addr + entry->size;
471 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
472 			unsigned long start_pfn = PFN_DOWN(start);
473 			unsigned long end_pfn = PFN_UP(end);
474 
475 			if (entry->type == E820_TYPE_RAM)
476 				end_pfn = PFN_UP(entry->addr);
477 
478 			if (start_pfn < end_pfn)
479 				ret_val = func(start_pfn, end_pfn, nr_pages,
480 					       ret_val);
481 			start = end;
482 		}
483 	}
484 
485 	return ret_val;
486 }
487 
488 /*
489  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
490  * The remap information (which mfn remap to which pfn) is contained in the
491  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
492  * This scheme allows to remap the different chunks in arbitrary order while
493  * the resulting mapping will be independent from the order.
494  */
495 void __init xen_remap_memory(void)
496 {
497 	unsigned long buf = (unsigned long)&xen_remap_buf;
498 	unsigned long mfn_save, pfn;
499 	unsigned long remapped = 0;
500 	unsigned int i;
501 	unsigned long pfn_s = ~0UL;
502 	unsigned long len = 0;
503 
504 	mfn_save = virt_to_mfn(buf);
505 
506 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
507 		/* Map the remap information */
508 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
509 
510 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
511 
512 		pfn = xen_remap_buf.target_pfn;
513 		for (i = 0; i < xen_remap_buf.size; i++) {
514 			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
515 			remapped++;
516 			pfn++;
517 		}
518 		if (pfn_s == ~0UL || pfn == pfn_s) {
519 			pfn_s = xen_remap_buf.target_pfn;
520 			len += xen_remap_buf.size;
521 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
522 			len += xen_remap_buf.size;
523 		} else {
524 			xen_del_extra_mem(pfn_s, len);
525 			pfn_s = xen_remap_buf.target_pfn;
526 			len = xen_remap_buf.size;
527 		}
528 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
529 	}
530 
531 	if (pfn_s != ~0UL && len)
532 		xen_del_extra_mem(pfn_s, len);
533 
534 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
535 
536 	pr_info("Remapped %ld page(s)\n", remapped);
537 }
538 
539 static unsigned long __init xen_get_pages_limit(void)
540 {
541 	unsigned long limit;
542 
543 	limit = MAXMEM / PAGE_SIZE;
544 	if (!xen_initial_domain() && xen_512gb_limit)
545 		limit = GB(512) / PAGE_SIZE;
546 
547 	return limit;
548 }
549 
550 static unsigned long __init xen_get_max_pages(void)
551 {
552 	unsigned long max_pages, limit;
553 	domid_t domid = DOMID_SELF;
554 	long ret;
555 
556 	limit = xen_get_pages_limit();
557 	max_pages = limit;
558 
559 	/*
560 	 * For the initial domain we use the maximum reservation as
561 	 * the maximum page.
562 	 *
563 	 * For guest domains the current maximum reservation reflects
564 	 * the current maximum rather than the static maximum. In this
565 	 * case the e820 map provided to us will cover the static
566 	 * maximum region.
567 	 */
568 	if (xen_initial_domain()) {
569 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
570 		if (ret > 0)
571 			max_pages = ret;
572 	}
573 
574 	return min(max_pages, limit);
575 }
576 
577 static void __init xen_align_and_add_e820_region(phys_addr_t start,
578 						 phys_addr_t size, int type)
579 {
580 	phys_addr_t end = start + size;
581 
582 	/* Align RAM regions to page boundaries. */
583 	if (type == E820_TYPE_RAM) {
584 		start = PAGE_ALIGN(start);
585 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
586 #ifdef CONFIG_MEMORY_HOTPLUG
587 		/*
588 		 * Don't allow adding memory not in E820 map while booting the
589 		 * system. Once the balloon driver is up it will remove that
590 		 * restriction again.
591 		 */
592 		max_mem_size = end;
593 #endif
594 	}
595 
596 	e820__range_add(start, end - start, type);
597 }
598 
599 static void __init xen_ignore_unusable(void)
600 {
601 	struct e820_entry *entry = xen_e820_table.entries;
602 	unsigned int i;
603 
604 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
605 		if (entry->type == E820_TYPE_UNUSABLE)
606 			entry->type = E820_TYPE_RAM;
607 	}
608 }
609 
610 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
611 {
612 	struct e820_entry *entry;
613 	unsigned mapcnt;
614 	phys_addr_t end;
615 
616 	if (!size)
617 		return false;
618 
619 	end = start + size;
620 	entry = xen_e820_table.entries;
621 
622 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
623 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
624 		    (entry->addr + entry->size) >= end)
625 			return false;
626 
627 		entry++;
628 	}
629 
630 	return true;
631 }
632 
633 /*
634  * Find a free area in physical memory not yet reserved and compliant with
635  * E820 map.
636  * Used to relocate pre-allocated areas like initrd or p2m list which are in
637  * conflict with the to be used E820 map.
638  * In case no area is found, return 0. Otherwise return the physical address
639  * of the area which is already reserved for convenience.
640  */
641 phys_addr_t __init xen_find_free_area(phys_addr_t size)
642 {
643 	unsigned mapcnt;
644 	phys_addr_t addr, start;
645 	struct e820_entry *entry = xen_e820_table.entries;
646 
647 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
648 		if (entry->type != E820_TYPE_RAM || entry->size < size)
649 			continue;
650 		start = entry->addr;
651 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
652 			if (!memblock_is_reserved(addr))
653 				continue;
654 			start = addr + PAGE_SIZE;
655 			if (start + size > entry->addr + entry->size)
656 				break;
657 		}
658 		if (addr >= start + size) {
659 			memblock_reserve(start, size);
660 			return start;
661 		}
662 	}
663 
664 	return 0;
665 }
666 
667 /*
668  * Like memcpy, but with physical addresses for dest and src.
669  */
670 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
671 				   phys_addr_t n)
672 {
673 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
674 	void *from, *to;
675 
676 	while (n) {
677 		dest_off = dest & ~PAGE_MASK;
678 		src_off = src & ~PAGE_MASK;
679 		dest_len = n;
680 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
681 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
682 		src_len = n;
683 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
684 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
685 		len = min(dest_len, src_len);
686 		to = early_memremap(dest - dest_off, dest_len + dest_off);
687 		from = early_memremap(src - src_off, src_len + src_off);
688 		memcpy(to, from, len);
689 		early_memunmap(to, dest_len + dest_off);
690 		early_memunmap(from, src_len + src_off);
691 		n -= len;
692 		dest += len;
693 		src += len;
694 	}
695 }
696 
697 /*
698  * Reserve Xen mfn_list.
699  */
700 static void __init xen_reserve_xen_mfnlist(void)
701 {
702 	phys_addr_t start, size;
703 
704 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
705 		start = __pa(xen_start_info->mfn_list);
706 		size = PFN_ALIGN(xen_start_info->nr_pages *
707 				 sizeof(unsigned long));
708 	} else {
709 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
710 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
711 	}
712 
713 	memblock_reserve(start, size);
714 	if (!xen_is_e820_reserved(start, size))
715 		return;
716 
717 	xen_relocate_p2m();
718 	memblock_phys_free(start, size);
719 }
720 
721 /**
722  * xen_memory_setup - Hook for machine specific memory setup.
723  **/
724 char * __init xen_memory_setup(void)
725 {
726 	unsigned long max_pfn, pfn_s, n_pfns;
727 	phys_addr_t mem_end, addr, size, chunk_size;
728 	u32 type;
729 	int rc;
730 	struct xen_memory_map memmap;
731 	unsigned long max_pages;
732 	unsigned long extra_pages = 0;
733 	int i;
734 	int op;
735 
736 	xen_parse_512gb();
737 	max_pfn = xen_get_pages_limit();
738 	max_pfn = min(max_pfn, xen_start_info->nr_pages);
739 	mem_end = PFN_PHYS(max_pfn);
740 
741 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
742 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
743 
744 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
745 	xen_saved_max_mem_size = max_mem_size;
746 #endif
747 
748 	op = xen_initial_domain() ?
749 		XENMEM_machine_memory_map :
750 		XENMEM_memory_map;
751 	rc = HYPERVISOR_memory_op(op, &memmap);
752 	if (rc == -ENOSYS) {
753 		BUG_ON(xen_initial_domain());
754 		memmap.nr_entries = 1;
755 		xen_e820_table.entries[0].addr = 0ULL;
756 		xen_e820_table.entries[0].size = mem_end;
757 		/* 8MB slack (to balance backend allocations). */
758 		xen_e820_table.entries[0].size += 8ULL << 20;
759 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
760 		rc = 0;
761 	}
762 	BUG_ON(rc);
763 	BUG_ON(memmap.nr_entries == 0);
764 	xen_e820_table.nr_entries = memmap.nr_entries;
765 
766 	/*
767 	 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
768 	 * regions, so if we're using the machine memory map leave the
769 	 * region as RAM as it is in the pseudo-physical map.
770 	 *
771 	 * UNUSABLE regions in domUs are not handled and will need
772 	 * a patch in the future.
773 	 */
774 	if (xen_initial_domain())
775 		xen_ignore_unusable();
776 
777 	/* Make sure the Xen-supplied memory map is well-ordered. */
778 	e820__update_table(&xen_e820_table);
779 
780 	max_pages = xen_get_max_pages();
781 
782 	/* How many extra pages do we need due to remapping? */
783 	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
784 
785 	if (max_pages > max_pfn)
786 		extra_pages += max_pages - max_pfn;
787 
788 	/*
789 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
790 	 * factor the base size.
791 	 *
792 	 * Make sure we have no memory above max_pages, as this area
793 	 * isn't handled by the p2m management.
794 	 */
795 	extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
796 			   extra_pages, max_pages - max_pfn);
797 	i = 0;
798 	addr = xen_e820_table.entries[0].addr;
799 	size = xen_e820_table.entries[0].size;
800 	while (i < xen_e820_table.nr_entries) {
801 		bool discard = false;
802 
803 		chunk_size = size;
804 		type = xen_e820_table.entries[i].type;
805 
806 		if (type == E820_TYPE_RAM) {
807 			if (addr < mem_end) {
808 				chunk_size = min(size, mem_end - addr);
809 			} else if (extra_pages) {
810 				chunk_size = min(size, PFN_PHYS(extra_pages));
811 				pfn_s = PFN_UP(addr);
812 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
813 				extra_pages -= n_pfns;
814 				xen_add_extra_mem(pfn_s, n_pfns);
815 				xen_max_p2m_pfn = pfn_s + n_pfns;
816 			} else
817 				discard = true;
818 		}
819 
820 		if (!discard)
821 			xen_align_and_add_e820_region(addr, chunk_size, type);
822 
823 		addr += chunk_size;
824 		size -= chunk_size;
825 		if (size == 0) {
826 			i++;
827 			if (i < xen_e820_table.nr_entries) {
828 				addr = xen_e820_table.entries[i].addr;
829 				size = xen_e820_table.entries[i].size;
830 			}
831 		}
832 	}
833 
834 	/*
835 	 * Set the rest as identity mapped, in case PCI BARs are
836 	 * located here.
837 	 */
838 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
839 
840 	/*
841 	 * In domU, the ISA region is normal, usable memory, but we
842 	 * reserve ISA memory anyway because too many things poke
843 	 * about in there.
844 	 */
845 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
846 
847 	e820__update_table(e820_table);
848 
849 	/*
850 	 * Check whether the kernel itself conflicts with the target E820 map.
851 	 * Failing now is better than running into weird problems later due
852 	 * to relocating (and even reusing) pages with kernel text or data.
853 	 */
854 	if (xen_is_e820_reserved(__pa_symbol(_text),
855 			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
856 		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
857 		BUG();
858 	}
859 
860 	/*
861 	 * Check for a conflict of the hypervisor supplied page tables with
862 	 * the target E820 map.
863 	 */
864 	xen_pt_check_e820();
865 
866 	xen_reserve_xen_mfnlist();
867 
868 	/* Check for a conflict of the initrd with the target E820 map. */
869 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
870 				 boot_params.hdr.ramdisk_size)) {
871 		phys_addr_t new_area, start, size;
872 
873 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
874 		if (!new_area) {
875 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
876 			BUG();
877 		}
878 
879 		start = boot_params.hdr.ramdisk_image;
880 		size = boot_params.hdr.ramdisk_size;
881 		xen_phys_memcpy(new_area, start, size);
882 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
883 			start, start + size, new_area, new_area + size);
884 		memblock_phys_free(start, size);
885 		boot_params.hdr.ramdisk_image = new_area;
886 		boot_params.ext_ramdisk_image = new_area >> 32;
887 	}
888 
889 	/*
890 	 * Set identity map on non-RAM pages and prepare remapping the
891 	 * underlying RAM.
892 	 */
893 	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
894 
895 	pr_info("Released %ld page(s)\n", xen_released_pages);
896 
897 	return "Xen";
898 }
899 
900 static int register_callback(unsigned type, const void *func)
901 {
902 	struct callback_register callback = {
903 		.type = type,
904 		.address = XEN_CALLBACK(__KERNEL_CS, func),
905 		.flags = CALLBACKF_mask_events,
906 	};
907 
908 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
909 }
910 
911 void xen_enable_sysenter(void)
912 {
913 	int ret;
914 	unsigned sysenter_feature;
915 
916 	sysenter_feature = X86_FEATURE_SYSENTER32;
917 
918 	if (!boot_cpu_has(sysenter_feature))
919 		return;
920 
921 	ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
922 	if(ret != 0)
923 		setup_clear_cpu_cap(sysenter_feature);
924 }
925 
926 void xen_enable_syscall(void)
927 {
928 	int ret;
929 
930 	ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
931 	if (ret != 0) {
932 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
933 		/* Pretty fatal; 64-bit userspace has no other
934 		   mechanism for syscalls. */
935 	}
936 
937 	if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
938 		ret = register_callback(CALLBACKTYPE_syscall32,
939 					xen_syscall32_target);
940 		if (ret != 0)
941 			setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
942 	}
943 }
944 
945 static void __init xen_pvmmu_arch_setup(void)
946 {
947 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
948 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
949 
950 	HYPERVISOR_vm_assist(VMASST_CMD_enable,
951 			     VMASST_TYPE_pae_extended_cr3);
952 
953 	if (register_callback(CALLBACKTYPE_event,
954 			      xen_asm_exc_xen_hypervisor_callback) ||
955 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
956 		BUG();
957 
958 	xen_enable_sysenter();
959 	xen_enable_syscall();
960 }
961 
962 /* This function is not called for HVM domains */
963 void __init xen_arch_setup(void)
964 {
965 	xen_panic_handler_init();
966 	xen_pvmmu_arch_setup();
967 
968 #ifdef CONFIG_ACPI
969 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
970 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
971 		disable_acpi();
972 	}
973 #endif
974 
975 	memcpy(boot_command_line, xen_start_info->cmd_line,
976 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
977 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
978 
979 	/* Set up idle, making sure it calls safe_halt() pvop */
980 	disable_cpuidle();
981 	disable_cpufreq();
982 	WARN_ON(xen_set_default_idle());
983 #ifdef CONFIG_NUMA
984 	numa_off = 1;
985 #endif
986 }
987