xref: /linux/arch/x86/xen/setup.c (revision 0340dc4c82590d8735c58cf904a8aa1173273ab5)
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/iscsi_ibft.h>
10 #include <linux/sched.h>
11 #include <linux/kstrtox.h>
12 #include <linux/mm.h>
13 #include <linux/pm.h>
14 #include <linux/memblock.h>
15 #include <linux/cpuidle.h>
16 #include <linux/cpufreq.h>
17 #include <linux/memory_hotplug.h>
18 
19 #include <asm/elf.h>
20 #include <asm/vdso.h>
21 #include <asm/e820/api.h>
22 #include <asm/setup.h>
23 #include <asm/acpi.h>
24 #include <asm/numa.h>
25 #include <asm/idtentry.h>
26 #include <asm/xen/hypervisor.h>
27 #include <asm/xen/hypercall.h>
28 
29 #include <xen/xen.h>
30 #include <xen/page.h>
31 #include <xen/interface/callback.h>
32 #include <xen/interface/memory.h>
33 #include <xen/interface/physdev.h>
34 #include <xen/features.h>
35 #include <xen/hvc-console.h>
36 #include "xen-ops.h"
37 #include "mmu.h"
38 
39 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
40 
41 /* Number of pages released from the initial allocation. */
42 unsigned long xen_released_pages;
43 
44 /* Memory map would allow PCI passthrough. */
45 bool xen_pv_pci_possible;
46 
47 /* E820 map used during setting up memory. */
48 static struct e820_table xen_e820_table __initdata;
49 
50 /*
51  * Buffer used to remap identity mapped pages. We only need the virtual space.
52  * The physical page behind this address is remapped as needed to different
53  * buffer pages.
54  */
55 #define REMAP_SIZE	(P2M_PER_PAGE - 3)
56 static struct {
57 	unsigned long	next_area_mfn;
58 	unsigned long	target_pfn;
59 	unsigned long	size;
60 	unsigned long	mfns[REMAP_SIZE];
61 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
62 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
63 
64 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
65 
66 static void __init xen_parse_512gb(void)
67 {
68 	bool val = false;
69 	char *arg;
70 
71 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
72 	if (!arg)
73 		return;
74 
75 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
76 	if (!arg)
77 		val = true;
78 	else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
79 		return;
80 
81 	xen_512gb_limit = val;
82 }
83 
84 static void __init xen_del_extra_mem(unsigned long start_pfn,
85 				     unsigned long n_pfns)
86 {
87 	int i;
88 	unsigned long start_r, size_r;
89 
90 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
91 		start_r = xen_extra_mem[i].start_pfn;
92 		size_r = xen_extra_mem[i].n_pfns;
93 
94 		/* Start of region. */
95 		if (start_r == start_pfn) {
96 			BUG_ON(n_pfns > size_r);
97 			xen_extra_mem[i].start_pfn += n_pfns;
98 			xen_extra_mem[i].n_pfns -= n_pfns;
99 			break;
100 		}
101 		/* End of region. */
102 		if (start_r + size_r == start_pfn + n_pfns) {
103 			BUG_ON(n_pfns > size_r);
104 			xen_extra_mem[i].n_pfns -= n_pfns;
105 			break;
106 		}
107 		/* Mid of region. */
108 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
109 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
110 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
111 			/* Calling memblock_reserve() again is okay. */
112 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
113 					  (start_pfn + n_pfns));
114 			break;
115 		}
116 	}
117 	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
118 }
119 
120 /*
121  * Called during boot before the p2m list can take entries beyond the
122  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
123  * invalid.
124  */
125 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
126 {
127 	int i;
128 
129 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
130 		if (pfn >= xen_extra_mem[i].start_pfn &&
131 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
132 			return INVALID_P2M_ENTRY;
133 	}
134 
135 	return IDENTITY_FRAME(pfn);
136 }
137 
138 /*
139  * Mark all pfns of extra mem as invalid in p2m list.
140  */
141 void __init xen_inv_extra_mem(void)
142 {
143 	unsigned long pfn, pfn_s, pfn_e;
144 	int i;
145 
146 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
147 		if (!xen_extra_mem[i].n_pfns)
148 			continue;
149 		pfn_s = xen_extra_mem[i].start_pfn;
150 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
151 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
152 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
153 	}
154 }
155 
156 /*
157  * Finds the next RAM pfn available in the E820 map after min_pfn.
158  * This function updates min_pfn with the pfn found and returns
159  * the size of that range or zero if not found.
160  */
161 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
162 {
163 	const struct e820_entry *entry = xen_e820_table.entries;
164 	unsigned int i;
165 	unsigned long done = 0;
166 
167 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
168 		unsigned long s_pfn;
169 		unsigned long e_pfn;
170 
171 		if (entry->type != E820_TYPE_RAM)
172 			continue;
173 
174 		e_pfn = PFN_DOWN(entry->addr + entry->size);
175 
176 		/* We only care about E820 after this */
177 		if (e_pfn <= *min_pfn)
178 			continue;
179 
180 		s_pfn = PFN_UP(entry->addr);
181 
182 		/* If min_pfn falls within the E820 entry, we want to start
183 		 * at the min_pfn PFN.
184 		 */
185 		if (s_pfn <= *min_pfn) {
186 			done = e_pfn - *min_pfn;
187 		} else {
188 			done = e_pfn - s_pfn;
189 			*min_pfn = s_pfn;
190 		}
191 		break;
192 	}
193 
194 	return done;
195 }
196 
197 static int __init xen_free_mfn(unsigned long mfn)
198 {
199 	struct xen_memory_reservation reservation = {
200 		.address_bits = 0,
201 		.extent_order = 0,
202 		.domid        = DOMID_SELF
203 	};
204 
205 	set_xen_guest_handle(reservation.extent_start, &mfn);
206 	reservation.nr_extents = 1;
207 
208 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
209 }
210 
211 /*
212  * This releases a chunk of memory and then does the identity map. It's used
213  * as a fallback if the remapping fails.
214  */
215 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
216 			unsigned long end_pfn, unsigned long nr_pages)
217 {
218 	unsigned long pfn, end;
219 	int ret;
220 
221 	WARN_ON(start_pfn > end_pfn);
222 
223 	/* Release pages first. */
224 	end = min(end_pfn, nr_pages);
225 	for (pfn = start_pfn; pfn < end; pfn++) {
226 		unsigned long mfn = pfn_to_mfn(pfn);
227 
228 		/* Make sure pfn exists to start with */
229 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
230 			continue;
231 
232 		ret = xen_free_mfn(mfn);
233 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
234 
235 		if (ret == 1) {
236 			xen_released_pages++;
237 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
238 				break;
239 		} else
240 			break;
241 	}
242 
243 	set_phys_range_identity(start_pfn, end_pfn);
244 }
245 
246 /*
247  * Helper function to update the p2m and m2p tables and kernel mapping.
248  */
249 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
250 {
251 	struct mmu_update update = {
252 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
253 		.val = pfn
254 	};
255 
256 	/* Update p2m */
257 	if (!set_phys_to_machine(pfn, mfn)) {
258 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
259 		     pfn, mfn);
260 		BUG();
261 	}
262 
263 	/* Update m2p */
264 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
265 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
266 		     mfn, pfn);
267 		BUG();
268 	}
269 
270 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
271 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
272 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
273 		      mfn, pfn);
274 		BUG();
275 	}
276 }
277 
278 /*
279  * This function updates the p2m and m2p tables with an identity map from
280  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
281  * original allocation at remap_pfn. The information needed for remapping is
282  * saved in the memory itself to avoid the need for allocating buffers. The
283  * complete remap information is contained in a list of MFNs each containing
284  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
285  * This enables us to preserve the original mfn sequence while doing the
286  * remapping at a time when the memory management is capable of allocating
287  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
288  * its callers.
289  */
290 static void __init xen_do_set_identity_and_remap_chunk(
291         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
292 {
293 	unsigned long buf = (unsigned long)&xen_remap_buf;
294 	unsigned long mfn_save, mfn;
295 	unsigned long ident_pfn_iter, remap_pfn_iter;
296 	unsigned long ident_end_pfn = start_pfn + size;
297 	unsigned long left = size;
298 	unsigned int i, chunk;
299 
300 	WARN_ON(size == 0);
301 
302 	mfn_save = virt_to_mfn((void *)buf);
303 
304 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
305 	     ident_pfn_iter < ident_end_pfn;
306 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
307 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
308 
309 		/* Map first pfn to xen_remap_buf */
310 		mfn = pfn_to_mfn(ident_pfn_iter);
311 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
312 
313 		/* Save mapping information in page */
314 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
315 		xen_remap_buf.target_pfn = remap_pfn_iter;
316 		xen_remap_buf.size = chunk;
317 		for (i = 0; i < chunk; i++)
318 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
319 
320 		/* Put remap buf into list. */
321 		xen_remap_mfn = mfn;
322 
323 		/* Set identity map */
324 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
325 
326 		left -= chunk;
327 	}
328 
329 	/* Restore old xen_remap_buf mapping */
330 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
331 }
332 
333 /*
334  * This function takes a contiguous pfn range that needs to be identity mapped
335  * and:
336  *
337  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
338  *  2) Calls the do_ function to actually do the mapping/remapping work.
339  *
340  * The goal is to not allocate additional memory but to remap the existing
341  * pages. In the case of an error the underlying memory is simply released back
342  * to Xen and not remapped.
343  */
344 static unsigned long __init xen_set_identity_and_remap_chunk(
345 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
346 	unsigned long remap_pfn)
347 {
348 	unsigned long pfn;
349 	unsigned long i = 0;
350 	unsigned long n = end_pfn - start_pfn;
351 
352 	if (remap_pfn == 0)
353 		remap_pfn = nr_pages;
354 
355 	while (i < n) {
356 		unsigned long cur_pfn = start_pfn + i;
357 		unsigned long left = n - i;
358 		unsigned long size = left;
359 		unsigned long remap_range_size;
360 
361 		/* Do not remap pages beyond the current allocation */
362 		if (cur_pfn >= nr_pages) {
363 			/* Identity map remaining pages */
364 			set_phys_range_identity(cur_pfn, cur_pfn + size);
365 			break;
366 		}
367 		if (cur_pfn + size > nr_pages)
368 			size = nr_pages - cur_pfn;
369 
370 		remap_range_size = xen_find_pfn_range(&remap_pfn);
371 		if (!remap_range_size) {
372 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
373 			xen_set_identity_and_release_chunk(cur_pfn,
374 						cur_pfn + left, nr_pages);
375 			break;
376 		}
377 		/* Adjust size to fit in current e820 RAM region */
378 		if (size > remap_range_size)
379 			size = remap_range_size;
380 
381 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
382 
383 		/* Update variables to reflect new mappings. */
384 		i += size;
385 		remap_pfn += size;
386 	}
387 
388 	/*
389 	 * If the PFNs are currently mapped, their VA mappings need to be
390 	 * zapped.
391 	 */
392 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
393 		(void)HYPERVISOR_update_va_mapping(
394 			(unsigned long)__va(pfn << PAGE_SHIFT),
395 			native_make_pte(0), 0);
396 
397 	return remap_pfn;
398 }
399 
400 static unsigned long __init xen_count_remap_pages(
401 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
402 	unsigned long remap_pages)
403 {
404 	if (start_pfn >= nr_pages)
405 		return remap_pages;
406 
407 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
408 }
409 
410 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
411 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
412 			      unsigned long nr_pages, unsigned long last_val))
413 {
414 	phys_addr_t start = 0;
415 	unsigned long ret_val = 0;
416 	const struct e820_entry *entry = xen_e820_table.entries;
417 	int i;
418 
419 	/*
420 	 * Combine non-RAM regions and gaps until a RAM region (or the
421 	 * end of the map) is reached, then call the provided function
422 	 * to perform its duty on the non-RAM region.
423 	 *
424 	 * The combined non-RAM regions are rounded to a whole number
425 	 * of pages so any partial pages are accessible via the 1:1
426 	 * mapping.  This is needed for some BIOSes that put (for
427 	 * example) the DMI tables in a reserved region that begins on
428 	 * a non-page boundary.
429 	 */
430 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
431 		phys_addr_t end = entry->addr + entry->size;
432 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
433 			unsigned long start_pfn = PFN_DOWN(start);
434 			unsigned long end_pfn = PFN_UP(end);
435 
436 			if (entry->type == E820_TYPE_RAM)
437 				end_pfn = PFN_UP(entry->addr);
438 
439 			if (start_pfn < end_pfn)
440 				ret_val = func(start_pfn, end_pfn, nr_pages,
441 					       ret_val);
442 			start = end;
443 		}
444 	}
445 
446 	return ret_val;
447 }
448 
449 /*
450  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
451  * The remap information (which mfn remap to which pfn) is contained in the
452  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
453  * This scheme allows to remap the different chunks in arbitrary order while
454  * the resulting mapping will be independent from the order.
455  */
456 void __init xen_remap_memory(void)
457 {
458 	unsigned long buf = (unsigned long)&xen_remap_buf;
459 	unsigned long mfn_save, pfn;
460 	unsigned long remapped = 0;
461 	unsigned int i;
462 	unsigned long pfn_s = ~0UL;
463 	unsigned long len = 0;
464 
465 	mfn_save = virt_to_mfn((void *)buf);
466 
467 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
468 		/* Map the remap information */
469 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
470 
471 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
472 
473 		pfn = xen_remap_buf.target_pfn;
474 		for (i = 0; i < xen_remap_buf.size; i++) {
475 			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
476 			remapped++;
477 			pfn++;
478 		}
479 		if (pfn_s == ~0UL || pfn == pfn_s) {
480 			pfn_s = xen_remap_buf.target_pfn;
481 			len += xen_remap_buf.size;
482 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
483 			len += xen_remap_buf.size;
484 		} else {
485 			xen_del_extra_mem(pfn_s, len);
486 			pfn_s = xen_remap_buf.target_pfn;
487 			len = xen_remap_buf.size;
488 		}
489 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
490 	}
491 
492 	if (pfn_s != ~0UL && len)
493 		xen_del_extra_mem(pfn_s, len);
494 
495 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
496 
497 	pr_info("Remapped %ld page(s)\n", remapped);
498 }
499 
500 static unsigned long __init xen_get_pages_limit(void)
501 {
502 	unsigned long limit;
503 
504 	limit = MAXMEM / PAGE_SIZE;
505 	if (!xen_initial_domain() && xen_512gb_limit)
506 		limit = GB(512) / PAGE_SIZE;
507 
508 	return limit;
509 }
510 
511 static unsigned long __init xen_get_max_pages(void)
512 {
513 	unsigned long max_pages, limit;
514 	domid_t domid = DOMID_SELF;
515 	long ret;
516 
517 	limit = xen_get_pages_limit();
518 	max_pages = limit;
519 
520 	/*
521 	 * For the initial domain we use the maximum reservation as
522 	 * the maximum page.
523 	 *
524 	 * For guest domains the current maximum reservation reflects
525 	 * the current maximum rather than the static maximum. In this
526 	 * case the e820 map provided to us will cover the static
527 	 * maximum region.
528 	 */
529 	if (xen_initial_domain()) {
530 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
531 		if (ret > 0)
532 			max_pages = ret;
533 	}
534 
535 	return min(max_pages, limit);
536 }
537 
538 static void __init xen_align_and_add_e820_region(phys_addr_t start,
539 						 phys_addr_t size, int type)
540 {
541 	phys_addr_t end = start + size;
542 
543 	/* Align RAM regions to page boundaries. */
544 	if (type == E820_TYPE_RAM) {
545 		start = PAGE_ALIGN(start);
546 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
547 #ifdef CONFIG_MEMORY_HOTPLUG
548 		/*
549 		 * Don't allow adding memory not in E820 map while booting the
550 		 * system. Once the balloon driver is up it will remove that
551 		 * restriction again.
552 		 */
553 		max_mem_size = end;
554 #endif
555 	}
556 
557 	e820__range_add(start, end - start, type);
558 }
559 
560 static void __init xen_ignore_unusable(void)
561 {
562 	struct e820_entry *entry = xen_e820_table.entries;
563 	unsigned int i;
564 
565 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
566 		if (entry->type == E820_TYPE_UNUSABLE)
567 			entry->type = E820_TYPE_RAM;
568 	}
569 }
570 
571 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
572 {
573 	struct e820_entry *entry;
574 	unsigned mapcnt;
575 	phys_addr_t end;
576 
577 	if (!size)
578 		return false;
579 
580 	end = start + size;
581 	entry = xen_e820_table.entries;
582 
583 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
584 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
585 		    (entry->addr + entry->size) >= end)
586 			return false;
587 
588 		entry++;
589 	}
590 
591 	return true;
592 }
593 
594 /*
595  * Find a free area in physical memory not yet reserved and compliant with
596  * E820 map.
597  * Used to relocate pre-allocated areas like initrd or p2m list which are in
598  * conflict with the to be used E820 map.
599  * In case no area is found, return 0. Otherwise return the physical address
600  * of the area which is already reserved for convenience.
601  */
602 phys_addr_t __init xen_find_free_area(phys_addr_t size)
603 {
604 	unsigned mapcnt;
605 	phys_addr_t addr, start;
606 	struct e820_entry *entry = xen_e820_table.entries;
607 
608 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
609 		if (entry->type != E820_TYPE_RAM || entry->size < size)
610 			continue;
611 		start = entry->addr;
612 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
613 			if (!memblock_is_reserved(addr))
614 				continue;
615 			start = addr + PAGE_SIZE;
616 			if (start + size > entry->addr + entry->size)
617 				break;
618 		}
619 		if (addr >= start + size) {
620 			memblock_reserve(start, size);
621 			return start;
622 		}
623 	}
624 
625 	return 0;
626 }
627 
628 /*
629  * Like memcpy, but with physical addresses for dest and src.
630  */
631 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
632 				   phys_addr_t n)
633 {
634 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
635 	void *from, *to;
636 
637 	while (n) {
638 		dest_off = dest & ~PAGE_MASK;
639 		src_off = src & ~PAGE_MASK;
640 		dest_len = n;
641 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
642 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
643 		src_len = n;
644 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
645 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
646 		len = min(dest_len, src_len);
647 		to = early_memremap(dest - dest_off, dest_len + dest_off);
648 		from = early_memremap(src - src_off, src_len + src_off);
649 		memcpy(to, from, len);
650 		early_memunmap(to, dest_len + dest_off);
651 		early_memunmap(from, src_len + src_off);
652 		n -= len;
653 		dest += len;
654 		src += len;
655 	}
656 }
657 
658 /*
659  * Reserve Xen mfn_list.
660  */
661 static void __init xen_reserve_xen_mfnlist(void)
662 {
663 	phys_addr_t start, size;
664 
665 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
666 		start = __pa(xen_start_info->mfn_list);
667 		size = PFN_ALIGN(xen_start_info->nr_pages *
668 				 sizeof(unsigned long));
669 	} else {
670 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
671 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
672 	}
673 
674 	memblock_reserve(start, size);
675 	if (!xen_is_e820_reserved(start, size))
676 		return;
677 
678 	xen_relocate_p2m();
679 	memblock_phys_free(start, size);
680 }
681 
682 /**
683  * xen_memory_setup - Hook for machine specific memory setup.
684  **/
685 char * __init xen_memory_setup(void)
686 {
687 	unsigned long max_pfn, pfn_s, n_pfns;
688 	phys_addr_t mem_end, addr, size, chunk_size;
689 	u32 type;
690 	int rc;
691 	struct xen_memory_map memmap;
692 	unsigned long max_pages;
693 	unsigned long extra_pages = 0;
694 	int i;
695 	int op;
696 
697 	xen_parse_512gb();
698 	max_pfn = xen_get_pages_limit();
699 	max_pfn = min(max_pfn, xen_start_info->nr_pages);
700 	mem_end = PFN_PHYS(max_pfn);
701 
702 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
703 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
704 
705 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
706 	xen_saved_max_mem_size = max_mem_size;
707 #endif
708 
709 	op = xen_initial_domain() ?
710 		XENMEM_machine_memory_map :
711 		XENMEM_memory_map;
712 	rc = HYPERVISOR_memory_op(op, &memmap);
713 	if (rc == -ENOSYS) {
714 		BUG_ON(xen_initial_domain());
715 		memmap.nr_entries = 1;
716 		xen_e820_table.entries[0].addr = 0ULL;
717 		xen_e820_table.entries[0].size = mem_end;
718 		/* 8MB slack (to balance backend allocations). */
719 		xen_e820_table.entries[0].size += 8ULL << 20;
720 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
721 		rc = 0;
722 	}
723 	BUG_ON(rc);
724 	BUG_ON(memmap.nr_entries == 0);
725 	xen_e820_table.nr_entries = memmap.nr_entries;
726 
727 	if (xen_initial_domain()) {
728 		/*
729 		 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
730 		 * regions, so if we're using the machine memory map leave the
731 		 * region as RAM as it is in the pseudo-physical map.
732 		 *
733 		 * UNUSABLE regions in domUs are not handled and will need
734 		 * a patch in the future.
735 		 */
736 		xen_ignore_unusable();
737 
738 #ifdef CONFIG_ISCSI_IBFT_FIND
739 		/* Reserve 0.5 MiB to 1 MiB region so iBFT can be found */
740 		xen_e820_table.entries[xen_e820_table.nr_entries].addr = IBFT_START;
741 		xen_e820_table.entries[xen_e820_table.nr_entries].size = IBFT_END - IBFT_START;
742 		xen_e820_table.entries[xen_e820_table.nr_entries].type = E820_TYPE_RESERVED;
743 		xen_e820_table.nr_entries++;
744 #endif
745 	}
746 
747 	/* Make sure the Xen-supplied memory map is well-ordered. */
748 	e820__update_table(&xen_e820_table);
749 
750 	max_pages = xen_get_max_pages();
751 
752 	/* How many extra pages do we need due to remapping? */
753 	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
754 
755 	if (max_pages > max_pfn)
756 		extra_pages += max_pages - max_pfn;
757 
758 	/*
759 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
760 	 * factor the base size.
761 	 *
762 	 * Make sure we have no memory above max_pages, as this area
763 	 * isn't handled by the p2m management.
764 	 */
765 	extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
766 			   extra_pages, max_pages - max_pfn);
767 	i = 0;
768 	addr = xen_e820_table.entries[0].addr;
769 	size = xen_e820_table.entries[0].size;
770 	while (i < xen_e820_table.nr_entries) {
771 		bool discard = false;
772 
773 		chunk_size = size;
774 		type = xen_e820_table.entries[i].type;
775 
776 		if (type == E820_TYPE_RESERVED)
777 			xen_pv_pci_possible = true;
778 
779 		if (type == E820_TYPE_RAM) {
780 			if (addr < mem_end) {
781 				chunk_size = min(size, mem_end - addr);
782 			} else if (extra_pages) {
783 				chunk_size = min(size, PFN_PHYS(extra_pages));
784 				pfn_s = PFN_UP(addr);
785 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
786 				extra_pages -= n_pfns;
787 				xen_add_extra_mem(pfn_s, n_pfns);
788 				xen_max_p2m_pfn = pfn_s + n_pfns;
789 			} else
790 				discard = true;
791 		}
792 
793 		if (!discard)
794 			xen_align_and_add_e820_region(addr, chunk_size, type);
795 
796 		addr += chunk_size;
797 		size -= chunk_size;
798 		if (size == 0) {
799 			i++;
800 			if (i < xen_e820_table.nr_entries) {
801 				addr = xen_e820_table.entries[i].addr;
802 				size = xen_e820_table.entries[i].size;
803 			}
804 		}
805 	}
806 
807 	/*
808 	 * Set the rest as identity mapped, in case PCI BARs are
809 	 * located here.
810 	 */
811 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
812 
813 	/*
814 	 * In domU, the ISA region is normal, usable memory, but we
815 	 * reserve ISA memory anyway because too many things poke
816 	 * about in there.
817 	 */
818 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
819 
820 	e820__update_table(e820_table);
821 
822 	/*
823 	 * Check whether the kernel itself conflicts with the target E820 map.
824 	 * Failing now is better than running into weird problems later due
825 	 * to relocating (and even reusing) pages with kernel text or data.
826 	 */
827 	if (xen_is_e820_reserved(__pa_symbol(_text),
828 			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
829 		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
830 		BUG();
831 	}
832 
833 	/*
834 	 * Check for a conflict of the hypervisor supplied page tables with
835 	 * the target E820 map.
836 	 */
837 	xen_pt_check_e820();
838 
839 	xen_reserve_xen_mfnlist();
840 
841 	/* Check for a conflict of the initrd with the target E820 map. */
842 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
843 				 boot_params.hdr.ramdisk_size)) {
844 		phys_addr_t new_area, start, size;
845 
846 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
847 		if (!new_area) {
848 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
849 			BUG();
850 		}
851 
852 		start = boot_params.hdr.ramdisk_image;
853 		size = boot_params.hdr.ramdisk_size;
854 		xen_phys_memcpy(new_area, start, size);
855 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
856 			start, start + size, new_area, new_area + size);
857 		memblock_phys_free(start, size);
858 		boot_params.hdr.ramdisk_image = new_area;
859 		boot_params.ext_ramdisk_image = new_area >> 32;
860 	}
861 
862 	/*
863 	 * Set identity map on non-RAM pages and prepare remapping the
864 	 * underlying RAM.
865 	 */
866 	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
867 
868 	pr_info("Released %ld page(s)\n", xen_released_pages);
869 
870 	return "Xen";
871 }
872 
873 static int register_callback(unsigned type, const void *func)
874 {
875 	struct callback_register callback = {
876 		.type = type,
877 		.address = XEN_CALLBACK(__KERNEL_CS, func),
878 		.flags = CALLBACKF_mask_events,
879 	};
880 
881 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
882 }
883 
884 void xen_enable_sysenter(void)
885 {
886 	if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
887 	    register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
888 		setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
889 }
890 
891 void xen_enable_syscall(void)
892 {
893 	int ret;
894 
895 	ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
896 	if (ret != 0) {
897 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
898 		/* Pretty fatal; 64-bit userspace has no other
899 		   mechanism for syscalls. */
900 	}
901 
902 	if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
903 	    register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
904 		setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
905 }
906 
907 static void __init xen_pvmmu_arch_setup(void)
908 {
909 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
910 
911 	if (register_callback(CALLBACKTYPE_event,
912 			      xen_asm_exc_xen_hypervisor_callback) ||
913 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
914 		BUG();
915 
916 	xen_enable_sysenter();
917 	xen_enable_syscall();
918 }
919 
920 /* This function is not called for HVM domains */
921 void __init xen_arch_setup(void)
922 {
923 	xen_panic_handler_init();
924 	xen_pvmmu_arch_setup();
925 
926 #ifdef CONFIG_ACPI
927 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
928 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
929 		disable_acpi();
930 	}
931 #endif
932 
933 	memcpy(boot_command_line, xen_start_info->cmd_line,
934 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
935 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
936 
937 	/* Set up idle, making sure it calls safe_halt() pvop */
938 	disable_cpuidle();
939 	disable_cpufreq();
940 	WARN_ON(xen_set_default_idle());
941 #ifdef CONFIG_NUMA
942 	numa_off = 1;
943 #endif
944 }
945