xref: /openbmc/linux/arch/x86/xen/setup.c (revision b737eecd)
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 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
63 
64 static void __init xen_parse_512gb(void)
65 {
66 	bool val = false;
67 	char *arg;
68 
69 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
70 	if (!arg)
71 		return;
72 
73 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
74 	if (!arg)
75 		val = true;
76 	else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
77 		return;
78 
79 	xen_512gb_limit = val;
80 }
81 
82 static void __init xen_add_extra_mem(unsigned long start_pfn,
83 				     unsigned long n_pfns)
84 {
85 	int i;
86 
87 	/*
88 	 * No need to check for zero size, should happen rarely and will only
89 	 * write a new entry regarded to be unused due to zero size.
90 	 */
91 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
92 		/* Add new region. */
93 		if (xen_extra_mem[i].n_pfns == 0) {
94 			xen_extra_mem[i].start_pfn = start_pfn;
95 			xen_extra_mem[i].n_pfns = n_pfns;
96 			break;
97 		}
98 		/* Append to existing region. */
99 		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
100 		    start_pfn) {
101 			xen_extra_mem[i].n_pfns += n_pfns;
102 			break;
103 		}
104 	}
105 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
106 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
107 
108 	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
109 }
110 
111 static void __init xen_del_extra_mem(unsigned long start_pfn,
112 				     unsigned long n_pfns)
113 {
114 	int i;
115 	unsigned long start_r, size_r;
116 
117 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
118 		start_r = xen_extra_mem[i].start_pfn;
119 		size_r = xen_extra_mem[i].n_pfns;
120 
121 		/* Start of region. */
122 		if (start_r == start_pfn) {
123 			BUG_ON(n_pfns > size_r);
124 			xen_extra_mem[i].start_pfn += n_pfns;
125 			xen_extra_mem[i].n_pfns -= n_pfns;
126 			break;
127 		}
128 		/* End of region. */
129 		if (start_r + size_r == start_pfn + n_pfns) {
130 			BUG_ON(n_pfns > size_r);
131 			xen_extra_mem[i].n_pfns -= n_pfns;
132 			break;
133 		}
134 		/* Mid of region. */
135 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
136 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
137 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
138 			/* Calling memblock_reserve() again is okay. */
139 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
140 					  (start_pfn + n_pfns));
141 			break;
142 		}
143 	}
144 	memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
145 }
146 
147 /*
148  * Called during boot before the p2m list can take entries beyond the
149  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
150  * invalid.
151  */
152 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
153 {
154 	int i;
155 
156 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
157 		if (pfn >= xen_extra_mem[i].start_pfn &&
158 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
159 			return INVALID_P2M_ENTRY;
160 	}
161 
162 	return IDENTITY_FRAME(pfn);
163 }
164 
165 /*
166  * Mark all pfns of extra mem as invalid in p2m list.
167  */
168 void __init xen_inv_extra_mem(void)
169 {
170 	unsigned long pfn, pfn_s, pfn_e;
171 	int i;
172 
173 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
174 		if (!xen_extra_mem[i].n_pfns)
175 			continue;
176 		pfn_s = xen_extra_mem[i].start_pfn;
177 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
178 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
179 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
180 	}
181 }
182 
183 /*
184  * Finds the next RAM pfn available in the E820 map after min_pfn.
185  * This function updates min_pfn with the pfn found and returns
186  * the size of that range or zero if not found.
187  */
188 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
189 {
190 	const struct e820_entry *entry = xen_e820_table.entries;
191 	unsigned int i;
192 	unsigned long done = 0;
193 
194 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
195 		unsigned long s_pfn;
196 		unsigned long e_pfn;
197 
198 		if (entry->type != E820_TYPE_RAM)
199 			continue;
200 
201 		e_pfn = PFN_DOWN(entry->addr + entry->size);
202 
203 		/* We only care about E820 after this */
204 		if (e_pfn <= *min_pfn)
205 			continue;
206 
207 		s_pfn = PFN_UP(entry->addr);
208 
209 		/* If min_pfn falls within the E820 entry, we want to start
210 		 * at the min_pfn PFN.
211 		 */
212 		if (s_pfn <= *min_pfn) {
213 			done = e_pfn - *min_pfn;
214 		} else {
215 			done = e_pfn - s_pfn;
216 			*min_pfn = s_pfn;
217 		}
218 		break;
219 	}
220 
221 	return done;
222 }
223 
224 static int __init xen_free_mfn(unsigned long mfn)
225 {
226 	struct xen_memory_reservation reservation = {
227 		.address_bits = 0,
228 		.extent_order = 0,
229 		.domid        = DOMID_SELF
230 	};
231 
232 	set_xen_guest_handle(reservation.extent_start, &mfn);
233 	reservation.nr_extents = 1;
234 
235 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
236 }
237 
238 /*
239  * This releases a chunk of memory and then does the identity map. It's used
240  * as a fallback if the remapping fails.
241  */
242 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
243 			unsigned long end_pfn, unsigned long nr_pages)
244 {
245 	unsigned long pfn, end;
246 	int ret;
247 
248 	WARN_ON(start_pfn > end_pfn);
249 
250 	/* Release pages first. */
251 	end = min(end_pfn, nr_pages);
252 	for (pfn = start_pfn; pfn < end; pfn++) {
253 		unsigned long mfn = pfn_to_mfn(pfn);
254 
255 		/* Make sure pfn exists to start with */
256 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
257 			continue;
258 
259 		ret = xen_free_mfn(mfn);
260 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
261 
262 		if (ret == 1) {
263 			xen_released_pages++;
264 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
265 				break;
266 		} else
267 			break;
268 	}
269 
270 	set_phys_range_identity(start_pfn, end_pfn);
271 }
272 
273 /*
274  * Helper function to update the p2m and m2p tables and kernel mapping.
275  */
276 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
277 {
278 	struct mmu_update update = {
279 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
280 		.val = pfn
281 	};
282 
283 	/* Update p2m */
284 	if (!set_phys_to_machine(pfn, mfn)) {
285 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
286 		     pfn, mfn);
287 		BUG();
288 	}
289 
290 	/* Update m2p */
291 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
292 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
293 		     mfn, pfn);
294 		BUG();
295 	}
296 
297 	/* Update kernel mapping, but not for highmem. */
298 	if (pfn >= PFN_UP(__pa(high_memory - 1)))
299 		return;
300 
301 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
302 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
303 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
304 		      mfn, pfn);
305 		BUG();
306 	}
307 }
308 
309 /*
310  * This function updates the p2m and m2p tables with an identity map from
311  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
312  * original allocation at remap_pfn. The information needed for remapping is
313  * saved in the memory itself to avoid the need for allocating buffers. The
314  * complete remap information is contained in a list of MFNs each containing
315  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
316  * This enables us to preserve the original mfn sequence while doing the
317  * remapping at a time when the memory management is capable of allocating
318  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
319  * its callers.
320  */
321 static void __init xen_do_set_identity_and_remap_chunk(
322         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
323 {
324 	unsigned long buf = (unsigned long)&xen_remap_buf;
325 	unsigned long mfn_save, mfn;
326 	unsigned long ident_pfn_iter, remap_pfn_iter;
327 	unsigned long ident_end_pfn = start_pfn + size;
328 	unsigned long left = size;
329 	unsigned int i, chunk;
330 
331 	WARN_ON(size == 0);
332 
333 	mfn_save = virt_to_mfn(buf);
334 
335 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
336 	     ident_pfn_iter < ident_end_pfn;
337 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
338 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
339 
340 		/* Map first pfn to xen_remap_buf */
341 		mfn = pfn_to_mfn(ident_pfn_iter);
342 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
343 
344 		/* Save mapping information in page */
345 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
346 		xen_remap_buf.target_pfn = remap_pfn_iter;
347 		xen_remap_buf.size = chunk;
348 		for (i = 0; i < chunk; i++)
349 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
350 
351 		/* Put remap buf into list. */
352 		xen_remap_mfn = mfn;
353 
354 		/* Set identity map */
355 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
356 
357 		left -= chunk;
358 	}
359 
360 	/* Restore old xen_remap_buf mapping */
361 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
362 }
363 
364 /*
365  * This function takes a contiguous pfn range that needs to be identity mapped
366  * and:
367  *
368  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
369  *  2) Calls the do_ function to actually do the mapping/remapping work.
370  *
371  * The goal is to not allocate additional memory but to remap the existing
372  * pages. In the case of an error the underlying memory is simply released back
373  * to Xen and not remapped.
374  */
375 static unsigned long __init xen_set_identity_and_remap_chunk(
376 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
377 	unsigned long remap_pfn)
378 {
379 	unsigned long pfn;
380 	unsigned long i = 0;
381 	unsigned long n = end_pfn - start_pfn;
382 
383 	if (remap_pfn == 0)
384 		remap_pfn = nr_pages;
385 
386 	while (i < n) {
387 		unsigned long cur_pfn = start_pfn + i;
388 		unsigned long left = n - i;
389 		unsigned long size = left;
390 		unsigned long remap_range_size;
391 
392 		/* Do not remap pages beyond the current allocation */
393 		if (cur_pfn >= nr_pages) {
394 			/* Identity map remaining pages */
395 			set_phys_range_identity(cur_pfn, cur_pfn + size);
396 			break;
397 		}
398 		if (cur_pfn + size > nr_pages)
399 			size = nr_pages - cur_pfn;
400 
401 		remap_range_size = xen_find_pfn_range(&remap_pfn);
402 		if (!remap_range_size) {
403 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
404 			xen_set_identity_and_release_chunk(cur_pfn,
405 						cur_pfn + left, nr_pages);
406 			break;
407 		}
408 		/* Adjust size to fit in current e820 RAM region */
409 		if (size > remap_range_size)
410 			size = remap_range_size;
411 
412 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
413 
414 		/* Update variables to reflect new mappings. */
415 		i += size;
416 		remap_pfn += size;
417 	}
418 
419 	/*
420 	 * If the PFNs are currently mapped, the VA mapping also needs
421 	 * to be updated to be 1:1.
422 	 */
423 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
424 		(void)HYPERVISOR_update_va_mapping(
425 			(unsigned long)__va(pfn << PAGE_SHIFT),
426 			mfn_pte(pfn, PAGE_KERNEL_IO), 0);
427 
428 	return remap_pfn;
429 }
430 
431 static unsigned long __init xen_count_remap_pages(
432 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
433 	unsigned long remap_pages)
434 {
435 	if (start_pfn >= nr_pages)
436 		return remap_pages;
437 
438 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
439 }
440 
441 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
442 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
443 			      unsigned long nr_pages, unsigned long last_val))
444 {
445 	phys_addr_t start = 0;
446 	unsigned long ret_val = 0;
447 	const struct e820_entry *entry = xen_e820_table.entries;
448 	int i;
449 
450 	/*
451 	 * Combine non-RAM regions and gaps until a RAM region (or the
452 	 * end of the map) is reached, then call the provided function
453 	 * to perform its duty on the non-RAM region.
454 	 *
455 	 * The combined non-RAM regions are rounded to a whole number
456 	 * of pages so any partial pages are accessible via the 1:1
457 	 * mapping.  This is needed for some BIOSes that put (for
458 	 * example) the DMI tables in a reserved region that begins on
459 	 * a non-page boundary.
460 	 */
461 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
462 		phys_addr_t end = entry->addr + entry->size;
463 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
464 			unsigned long start_pfn = PFN_DOWN(start);
465 			unsigned long end_pfn = PFN_UP(end);
466 
467 			if (entry->type == E820_TYPE_RAM)
468 				end_pfn = PFN_UP(entry->addr);
469 
470 			if (start_pfn < end_pfn)
471 				ret_val = func(start_pfn, end_pfn, nr_pages,
472 					       ret_val);
473 			start = end;
474 		}
475 	}
476 
477 	return ret_val;
478 }
479 
480 /*
481  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
482  * The remap information (which mfn remap to which pfn) is contained in the
483  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
484  * This scheme allows to remap the different chunks in arbitrary order while
485  * the resulting mapping will be independent from the order.
486  */
487 void __init xen_remap_memory(void)
488 {
489 	unsigned long buf = (unsigned long)&xen_remap_buf;
490 	unsigned long mfn_save, pfn;
491 	unsigned long remapped = 0;
492 	unsigned int i;
493 	unsigned long pfn_s = ~0UL;
494 	unsigned long len = 0;
495 
496 	mfn_save = virt_to_mfn(buf);
497 
498 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
499 		/* Map the remap information */
500 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
501 
502 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
503 
504 		pfn = xen_remap_buf.target_pfn;
505 		for (i = 0; i < xen_remap_buf.size; i++) {
506 			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
507 			remapped++;
508 			pfn++;
509 		}
510 		if (pfn_s == ~0UL || pfn == pfn_s) {
511 			pfn_s = xen_remap_buf.target_pfn;
512 			len += xen_remap_buf.size;
513 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
514 			len += xen_remap_buf.size;
515 		} else {
516 			xen_del_extra_mem(pfn_s, len);
517 			pfn_s = xen_remap_buf.target_pfn;
518 			len = xen_remap_buf.size;
519 		}
520 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
521 	}
522 
523 	if (pfn_s != ~0UL && len)
524 		xen_del_extra_mem(pfn_s, len);
525 
526 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
527 
528 	pr_info("Remapped %ld page(s)\n", remapped);
529 }
530 
531 static unsigned long __init xen_get_pages_limit(void)
532 {
533 	unsigned long limit;
534 
535 	limit = MAXMEM / PAGE_SIZE;
536 	if (!xen_initial_domain() && xen_512gb_limit)
537 		limit = GB(512) / PAGE_SIZE;
538 
539 	return limit;
540 }
541 
542 static unsigned long __init xen_get_max_pages(void)
543 {
544 	unsigned long max_pages, limit;
545 	domid_t domid = DOMID_SELF;
546 	long ret;
547 
548 	limit = xen_get_pages_limit();
549 	max_pages = limit;
550 
551 	/*
552 	 * For the initial domain we use the maximum reservation as
553 	 * the maximum page.
554 	 *
555 	 * For guest domains the current maximum reservation reflects
556 	 * the current maximum rather than the static maximum. In this
557 	 * case the e820 map provided to us will cover the static
558 	 * maximum region.
559 	 */
560 	if (xen_initial_domain()) {
561 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
562 		if (ret > 0)
563 			max_pages = ret;
564 	}
565 
566 	return min(max_pages, limit);
567 }
568 
569 static void __init xen_align_and_add_e820_region(phys_addr_t start,
570 						 phys_addr_t size, int type)
571 {
572 	phys_addr_t end = start + size;
573 
574 	/* Align RAM regions to page boundaries. */
575 	if (type == E820_TYPE_RAM) {
576 		start = PAGE_ALIGN(start);
577 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
578 #ifdef CONFIG_MEMORY_HOTPLUG
579 		/*
580 		 * Don't allow adding memory not in E820 map while booting the
581 		 * system. Once the balloon driver is up it will remove that
582 		 * restriction again.
583 		 */
584 		max_mem_size = end;
585 #endif
586 	}
587 
588 	e820__range_add(start, end - start, type);
589 }
590 
591 static void __init xen_ignore_unusable(void)
592 {
593 	struct e820_entry *entry = xen_e820_table.entries;
594 	unsigned int i;
595 
596 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
597 		if (entry->type == E820_TYPE_UNUSABLE)
598 			entry->type = E820_TYPE_RAM;
599 	}
600 }
601 
602 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
603 {
604 	struct e820_entry *entry;
605 	unsigned mapcnt;
606 	phys_addr_t end;
607 
608 	if (!size)
609 		return false;
610 
611 	end = start + size;
612 	entry = xen_e820_table.entries;
613 
614 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
615 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
616 		    (entry->addr + entry->size) >= end)
617 			return false;
618 
619 		entry++;
620 	}
621 
622 	return true;
623 }
624 
625 /*
626  * Find a free area in physical memory not yet reserved and compliant with
627  * E820 map.
628  * Used to relocate pre-allocated areas like initrd or p2m list which are in
629  * conflict with the to be used E820 map.
630  * In case no area is found, return 0. Otherwise return the physical address
631  * of the area which is already reserved for convenience.
632  */
633 phys_addr_t __init xen_find_free_area(phys_addr_t size)
634 {
635 	unsigned mapcnt;
636 	phys_addr_t addr, start;
637 	struct e820_entry *entry = xen_e820_table.entries;
638 
639 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
640 		if (entry->type != E820_TYPE_RAM || entry->size < size)
641 			continue;
642 		start = entry->addr;
643 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
644 			if (!memblock_is_reserved(addr))
645 				continue;
646 			start = addr + PAGE_SIZE;
647 			if (start + size > entry->addr + entry->size)
648 				break;
649 		}
650 		if (addr >= start + size) {
651 			memblock_reserve(start, size);
652 			return start;
653 		}
654 	}
655 
656 	return 0;
657 }
658 
659 /*
660  * Like memcpy, but with physical addresses for dest and src.
661  */
662 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
663 				   phys_addr_t n)
664 {
665 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
666 	void *from, *to;
667 
668 	while (n) {
669 		dest_off = dest & ~PAGE_MASK;
670 		src_off = src & ~PAGE_MASK;
671 		dest_len = n;
672 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
673 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
674 		src_len = n;
675 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
676 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
677 		len = min(dest_len, src_len);
678 		to = early_memremap(dest - dest_off, dest_len + dest_off);
679 		from = early_memremap(src - src_off, src_len + src_off);
680 		memcpy(to, from, len);
681 		early_memunmap(to, dest_len + dest_off);
682 		early_memunmap(from, src_len + src_off);
683 		n -= len;
684 		dest += len;
685 		src += len;
686 	}
687 }
688 
689 /*
690  * Reserve Xen mfn_list.
691  */
692 static void __init xen_reserve_xen_mfnlist(void)
693 {
694 	phys_addr_t start, size;
695 
696 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
697 		start = __pa(xen_start_info->mfn_list);
698 		size = PFN_ALIGN(xen_start_info->nr_pages *
699 				 sizeof(unsigned long));
700 	} else {
701 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
702 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
703 	}
704 
705 	memblock_reserve(start, size);
706 	if (!xen_is_e820_reserved(start, size))
707 		return;
708 
709 	xen_relocate_p2m();
710 	memblock_free(start, size);
711 }
712 
713 /**
714  * machine_specific_memory_setup - Hook for machine specific memory setup.
715  **/
716 char * __init xen_memory_setup(void)
717 {
718 	unsigned long max_pfn, pfn_s, n_pfns;
719 	phys_addr_t mem_end, addr, size, chunk_size;
720 	u32 type;
721 	int rc;
722 	struct xen_memory_map memmap;
723 	unsigned long max_pages;
724 	unsigned long extra_pages = 0;
725 	int i;
726 	int op;
727 
728 	xen_parse_512gb();
729 	max_pfn = xen_get_pages_limit();
730 	max_pfn = min(max_pfn, xen_start_info->nr_pages);
731 	mem_end = PFN_PHYS(max_pfn);
732 
733 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
734 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
735 
736 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
737 	xen_saved_max_mem_size = max_mem_size;
738 #endif
739 
740 	op = xen_initial_domain() ?
741 		XENMEM_machine_memory_map :
742 		XENMEM_memory_map;
743 	rc = HYPERVISOR_memory_op(op, &memmap);
744 	if (rc == -ENOSYS) {
745 		BUG_ON(xen_initial_domain());
746 		memmap.nr_entries = 1;
747 		xen_e820_table.entries[0].addr = 0ULL;
748 		xen_e820_table.entries[0].size = mem_end;
749 		/* 8MB slack (to balance backend allocations). */
750 		xen_e820_table.entries[0].size += 8ULL << 20;
751 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
752 		rc = 0;
753 	}
754 	BUG_ON(rc);
755 	BUG_ON(memmap.nr_entries == 0);
756 	xen_e820_table.nr_entries = memmap.nr_entries;
757 
758 	/*
759 	 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
760 	 * regions, so if we're using the machine memory map leave the
761 	 * region as RAM as it is in the pseudo-physical map.
762 	 *
763 	 * UNUSABLE regions in domUs are not handled and will need
764 	 * a patch in the future.
765 	 */
766 	if (xen_initial_domain())
767 		xen_ignore_unusable();
768 
769 	/* Make sure the Xen-supplied memory map is well-ordered. */
770 	e820__update_table(&xen_e820_table);
771 
772 	max_pages = xen_get_max_pages();
773 
774 	/* How many extra pages do we need due to remapping? */
775 	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
776 
777 	if (max_pages > max_pfn)
778 		extra_pages += max_pages - max_pfn;
779 
780 	/*
781 	 * Clamp the amount of extra memory to a XEN_EXTRA_MEM_RATIO
782 	 * factor the base size.
783 	 *
784 	 * Make sure we have no memory above max_pages, as this area
785 	 * isn't handled by the p2m management.
786 	 */
787 	extra_pages = min3(XEN_EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
788 			   extra_pages, max_pages - max_pfn);
789 	i = 0;
790 	addr = xen_e820_table.entries[0].addr;
791 	size = xen_e820_table.entries[0].size;
792 	while (i < xen_e820_table.nr_entries) {
793 		bool discard = false;
794 
795 		chunk_size = size;
796 		type = xen_e820_table.entries[i].type;
797 
798 		if (type == E820_TYPE_RAM) {
799 			if (addr < mem_end) {
800 				chunk_size = min(size, mem_end - addr);
801 			} else if (extra_pages) {
802 				chunk_size = min(size, PFN_PHYS(extra_pages));
803 				pfn_s = PFN_UP(addr);
804 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
805 				extra_pages -= n_pfns;
806 				xen_add_extra_mem(pfn_s, n_pfns);
807 				xen_max_p2m_pfn = pfn_s + n_pfns;
808 			} else
809 				discard = true;
810 		}
811 
812 		if (!discard)
813 			xen_align_and_add_e820_region(addr, chunk_size, type);
814 
815 		addr += chunk_size;
816 		size -= chunk_size;
817 		if (size == 0) {
818 			i++;
819 			if (i < xen_e820_table.nr_entries) {
820 				addr = xen_e820_table.entries[i].addr;
821 				size = xen_e820_table.entries[i].size;
822 			}
823 		}
824 	}
825 
826 	/*
827 	 * Set the rest as identity mapped, in case PCI BARs are
828 	 * located here.
829 	 */
830 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
831 
832 	/*
833 	 * In domU, the ISA region is normal, usable memory, but we
834 	 * reserve ISA memory anyway because too many things poke
835 	 * about in there.
836 	 */
837 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
838 
839 	e820__update_table(e820_table);
840 
841 	/*
842 	 * Check whether the kernel itself conflicts with the target E820 map.
843 	 * Failing now is better than running into weird problems later due
844 	 * to relocating (and even reusing) pages with kernel text or data.
845 	 */
846 	if (xen_is_e820_reserved(__pa_symbol(_text),
847 			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
848 		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
849 		BUG();
850 	}
851 
852 	/*
853 	 * Check for a conflict of the hypervisor supplied page tables with
854 	 * the target E820 map.
855 	 */
856 	xen_pt_check_e820();
857 
858 	xen_reserve_xen_mfnlist();
859 
860 	/* Check for a conflict of the initrd with the target E820 map. */
861 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
862 				 boot_params.hdr.ramdisk_size)) {
863 		phys_addr_t new_area, start, size;
864 
865 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
866 		if (!new_area) {
867 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
868 			BUG();
869 		}
870 
871 		start = boot_params.hdr.ramdisk_image;
872 		size = boot_params.hdr.ramdisk_size;
873 		xen_phys_memcpy(new_area, start, size);
874 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
875 			start, start + size, new_area, new_area + size);
876 		memblock_free(start, size);
877 		boot_params.hdr.ramdisk_image = new_area;
878 		boot_params.ext_ramdisk_image = new_area >> 32;
879 	}
880 
881 	/*
882 	 * Set identity map on non-RAM pages and prepare remapping the
883 	 * underlying RAM.
884 	 */
885 	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
886 
887 	pr_info("Released %ld page(s)\n", xen_released_pages);
888 
889 	return "Xen";
890 }
891 
892 static int register_callback(unsigned type, const void *func)
893 {
894 	struct callback_register callback = {
895 		.type = type,
896 		.address = XEN_CALLBACK(__KERNEL_CS, func),
897 		.flags = CALLBACKF_mask_events,
898 	};
899 
900 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
901 }
902 
903 void xen_enable_sysenter(void)
904 {
905 	int ret;
906 	unsigned sysenter_feature;
907 
908 	sysenter_feature = X86_FEATURE_SYSENTER32;
909 
910 	if (!boot_cpu_has(sysenter_feature))
911 		return;
912 
913 	ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
914 	if(ret != 0)
915 		setup_clear_cpu_cap(sysenter_feature);
916 }
917 
918 void xen_enable_syscall(void)
919 {
920 	int ret;
921 
922 	ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
923 	if (ret != 0) {
924 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
925 		/* Pretty fatal; 64-bit userspace has no other
926 		   mechanism for syscalls. */
927 	}
928 
929 	if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
930 		ret = register_callback(CALLBACKTYPE_syscall32,
931 					xen_syscall32_target);
932 		if (ret != 0)
933 			setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
934 	}
935 }
936 
937 static void __init xen_pvmmu_arch_setup(void)
938 {
939 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
940 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
941 
942 	HYPERVISOR_vm_assist(VMASST_CMD_enable,
943 			     VMASST_TYPE_pae_extended_cr3);
944 
945 	if (register_callback(CALLBACKTYPE_event,
946 			      xen_asm_exc_xen_hypervisor_callback) ||
947 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
948 		BUG();
949 
950 	xen_enable_sysenter();
951 	xen_enable_syscall();
952 }
953 
954 /* This function is not called for HVM domains */
955 void __init xen_arch_setup(void)
956 {
957 	xen_panic_handler_init();
958 	xen_pvmmu_arch_setup();
959 
960 #ifdef CONFIG_ACPI
961 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
962 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
963 		disable_acpi();
964 	}
965 #endif
966 
967 	memcpy(boot_command_line, xen_start_info->cmd_line,
968 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
969 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
970 
971 	/* Set up idle, making sure it calls safe_halt() pvop */
972 	disable_cpuidle();
973 	disable_cpufreq();
974 	WARN_ON(xen_set_default_idle());
975 #ifdef CONFIG_NUMA
976 	numa_off = 1;
977 #endif
978 }
979