xref: /linux/arch/x86/xen/setup.c (revision 55f1b540d893da740a81200450014c45a8103f54)
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 #include <linux/acpi.h>
19 
20 #include <asm/elf.h>
21 #include <asm/vdso.h>
22 #include <asm/e820/api.h>
23 #include <asm/setup.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 
38 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
39 
40 /* Number of pages released from the initial allocation. */
41 unsigned long xen_released_pages;
42 
43 /* Memory map would allow PCI passthrough. */
44 bool xen_pv_pci_possible;
45 
46 /* E820 map used during setting up memory. */
47 static struct e820_table xen_e820_table __initdata;
48 
49 /* Number of initially usable memory pages. */
50 static unsigned long ini_nr_pages __initdata;
51 
52 /*
53  * Buffer used to remap identity mapped pages. We only need the virtual space.
54  * The physical page behind this address is remapped as needed to different
55  * buffer pages.
56  */
57 #define REMAP_SIZE	(P2M_PER_PAGE - 3)
58 static struct {
59 	unsigned long	next_area_mfn;
60 	unsigned long	target_pfn;
61 	unsigned long	size;
62 	unsigned long	mfns[REMAP_SIZE];
63 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
64 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
65 
66 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
67 
68 static void __init xen_parse_512gb(void)
69 {
70 	bool val = false;
71 	char *arg;
72 
73 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
74 	if (!arg)
75 		return;
76 
77 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
78 	if (!arg)
79 		val = true;
80 	else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
81 		return;
82 
83 	xen_512gb_limit = val;
84 }
85 
86 static void __init xen_del_extra_mem(unsigned long start_pfn,
87 				     unsigned long n_pfns)
88 {
89 	int i;
90 	unsigned long start_r, size_r;
91 
92 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
93 		start_r = xen_extra_mem[i].start_pfn;
94 		size_r = xen_extra_mem[i].n_pfns;
95 
96 		/* Start of region. */
97 		if (start_r == start_pfn) {
98 			BUG_ON(n_pfns > size_r);
99 			xen_extra_mem[i].start_pfn += n_pfns;
100 			xen_extra_mem[i].n_pfns -= n_pfns;
101 			break;
102 		}
103 		/* End of region. */
104 		if (start_r + size_r == start_pfn + n_pfns) {
105 			BUG_ON(n_pfns > size_r);
106 			xen_extra_mem[i].n_pfns -= n_pfns;
107 			break;
108 		}
109 		/* Mid of region. */
110 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
111 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
112 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
113 			/* Calling memblock_reserve() again is okay. */
114 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
115 					  (start_pfn + n_pfns));
116 			break;
117 		}
118 	}
119 	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
120 }
121 
122 /*
123  * Called during boot before the p2m list can take entries beyond the
124  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
125  * invalid.
126  */
127 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
128 {
129 	int i;
130 
131 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
132 		if (pfn >= xen_extra_mem[i].start_pfn &&
133 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
134 			return INVALID_P2M_ENTRY;
135 	}
136 
137 	return IDENTITY_FRAME(pfn);
138 }
139 
140 /*
141  * Mark all pfns of extra mem as invalid in p2m list.
142  */
143 void __init xen_inv_extra_mem(void)
144 {
145 	unsigned long pfn, pfn_s, pfn_e;
146 	int i;
147 
148 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
149 		if (!xen_extra_mem[i].n_pfns)
150 			continue;
151 		pfn_s = xen_extra_mem[i].start_pfn;
152 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
153 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
154 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
155 	}
156 }
157 
158 /*
159  * Finds the next RAM pfn available in the E820 map after min_pfn.
160  * This function updates min_pfn with the pfn found and returns
161  * the size of that range or zero if not found.
162  */
163 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
164 {
165 	const struct e820_entry *entry = xen_e820_table.entries;
166 	unsigned int i;
167 	unsigned long done = 0;
168 
169 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
170 		unsigned long s_pfn;
171 		unsigned long e_pfn;
172 
173 		if (entry->type != E820_TYPE_RAM)
174 			continue;
175 
176 		e_pfn = PFN_DOWN(entry->addr + entry->size);
177 
178 		/* We only care about E820 after this */
179 		if (e_pfn <= *min_pfn)
180 			continue;
181 
182 		s_pfn = PFN_UP(entry->addr);
183 
184 		/* If min_pfn falls within the E820 entry, we want to start
185 		 * at the min_pfn PFN.
186 		 */
187 		if (s_pfn <= *min_pfn) {
188 			done = e_pfn - *min_pfn;
189 		} else {
190 			done = e_pfn - s_pfn;
191 			*min_pfn = s_pfn;
192 		}
193 		break;
194 	}
195 
196 	return done;
197 }
198 
199 static int __init xen_free_mfn(unsigned long mfn)
200 {
201 	struct xen_memory_reservation reservation = {
202 		.address_bits = 0,
203 		.extent_order = 0,
204 		.domid        = DOMID_SELF
205 	};
206 
207 	set_xen_guest_handle(reservation.extent_start, &mfn);
208 	reservation.nr_extents = 1;
209 
210 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
211 }
212 
213 /*
214  * This releases a chunk of memory and then does the identity map. It's used
215  * as a fallback if the remapping fails.
216  */
217 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
218 						      unsigned long end_pfn)
219 {
220 	unsigned long pfn, end;
221 	int ret;
222 
223 	WARN_ON(start_pfn > end_pfn);
224 
225 	/* Release pages first. */
226 	end = min(end_pfn, ini_nr_pages);
227 	for (pfn = start_pfn; pfn < end; pfn++) {
228 		unsigned long mfn = pfn_to_mfn(pfn);
229 
230 		/* Make sure pfn exists to start with */
231 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
232 			continue;
233 
234 		ret = xen_free_mfn(mfn);
235 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
236 
237 		if (ret == 1) {
238 			xen_released_pages++;
239 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
240 				break;
241 		} else
242 			break;
243 	}
244 
245 	set_phys_range_identity(start_pfn, end_pfn);
246 }
247 
248 /*
249  * Helper function to update the p2m and m2p tables and kernel mapping.
250  */
251 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
252 {
253 	struct mmu_update update = {
254 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
255 		.val = pfn
256 	};
257 
258 	/* Update p2m */
259 	if (!set_phys_to_machine(pfn, mfn)) {
260 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
261 		     pfn, mfn);
262 		BUG();
263 	}
264 
265 	/* Update m2p */
266 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
267 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
268 		     mfn, pfn);
269 		BUG();
270 	}
271 
272 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
273 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
274 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
275 		      mfn, pfn);
276 		BUG();
277 	}
278 }
279 
280 /*
281  * This function updates the p2m and m2p tables with an identity map from
282  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
283  * original allocation at remap_pfn. The information needed for remapping is
284  * saved in the memory itself to avoid the need for allocating buffers. The
285  * complete remap information is contained in a list of MFNs each containing
286  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
287  * This enables us to preserve the original mfn sequence while doing the
288  * remapping at a time when the memory management is capable of allocating
289  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
290  * its callers.
291  */
292 static void __init xen_do_set_identity_and_remap_chunk(
293         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
294 {
295 	unsigned long buf = (unsigned long)&xen_remap_buf;
296 	unsigned long mfn_save, mfn;
297 	unsigned long ident_pfn_iter, remap_pfn_iter;
298 	unsigned long ident_end_pfn = start_pfn + size;
299 	unsigned long left = size;
300 	unsigned int i, chunk;
301 
302 	WARN_ON(size == 0);
303 
304 	mfn_save = virt_to_mfn((void *)buf);
305 
306 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
307 	     ident_pfn_iter < ident_end_pfn;
308 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
309 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
310 
311 		/* Map first pfn to xen_remap_buf */
312 		mfn = pfn_to_mfn(ident_pfn_iter);
313 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
314 
315 		/* Save mapping information in page */
316 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
317 		xen_remap_buf.target_pfn = remap_pfn_iter;
318 		xen_remap_buf.size = chunk;
319 		for (i = 0; i < chunk; i++)
320 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
321 
322 		/* Put remap buf into list. */
323 		xen_remap_mfn = mfn;
324 
325 		/* Set identity map */
326 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
327 
328 		left -= chunk;
329 	}
330 
331 	/* Restore old xen_remap_buf mapping */
332 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
333 }
334 
335 /*
336  * This function takes a contiguous pfn range that needs to be identity mapped
337  * and:
338  *
339  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
340  *  2) Calls the do_ function to actually do the mapping/remapping work.
341  *
342  * The goal is to not allocate additional memory but to remap the existing
343  * pages. In the case of an error the underlying memory is simply released back
344  * to Xen and not remapped.
345  */
346 static unsigned long __init xen_set_identity_and_remap_chunk(
347 	unsigned long start_pfn, unsigned long end_pfn, unsigned long remap_pfn)
348 {
349 	unsigned long pfn;
350 	unsigned long i = 0;
351 	unsigned long n = end_pfn - start_pfn;
352 
353 	if (remap_pfn == 0)
354 		remap_pfn = ini_nr_pages;
355 
356 	while (i < n) {
357 		unsigned long cur_pfn = start_pfn + i;
358 		unsigned long left = n - i;
359 		unsigned long size = left;
360 		unsigned long remap_range_size;
361 
362 		/* Do not remap pages beyond the current allocation */
363 		if (cur_pfn >= ini_nr_pages) {
364 			/* Identity map remaining pages */
365 			set_phys_range_identity(cur_pfn, cur_pfn + size);
366 			break;
367 		}
368 		if (cur_pfn + size > ini_nr_pages)
369 			size = ini_nr_pages - cur_pfn;
370 
371 		remap_range_size = xen_find_pfn_range(&remap_pfn);
372 		if (!remap_range_size) {
373 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
374 			xen_set_identity_and_release_chunk(cur_pfn,
375 							   cur_pfn + left);
376 			break;
377 		}
378 		/* Adjust size to fit in current e820 RAM region */
379 		if (size > remap_range_size)
380 			size = remap_range_size;
381 
382 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
383 
384 		/* Update variables to reflect new mappings. */
385 		i += size;
386 		remap_pfn += size;
387 	}
388 
389 	/*
390 	 * If the PFNs are currently mapped, their VA mappings need to be
391 	 * zapped.
392 	 */
393 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
394 		(void)HYPERVISOR_update_va_mapping(
395 			(unsigned long)__va(pfn << PAGE_SHIFT),
396 			native_make_pte(0), 0);
397 
398 	return remap_pfn;
399 }
400 
401 static unsigned long __init xen_count_remap_pages(
402 	unsigned long start_pfn, unsigned long end_pfn,
403 	unsigned long remap_pages)
404 {
405 	if (start_pfn >= ini_nr_pages)
406 		return remap_pages;
407 
408 	return remap_pages + min(end_pfn, ini_nr_pages) - start_pfn;
409 }
410 
411 static unsigned long __init xen_foreach_remap_area(
412 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
413 			      unsigned long last_val))
414 {
415 	phys_addr_t start = 0;
416 	unsigned long ret_val = 0;
417 	const struct e820_entry *entry = xen_e820_table.entries;
418 	int i;
419 
420 	/*
421 	 * Combine non-RAM regions and gaps until a RAM region (or the
422 	 * end of the map) is reached, then call the provided function
423 	 * to perform its duty on the non-RAM region.
424 	 *
425 	 * The combined non-RAM regions are rounded to a whole number
426 	 * of pages so any partial pages are accessible via the 1:1
427 	 * mapping.  This is needed for some BIOSes that put (for
428 	 * example) the DMI tables in a reserved region that begins on
429 	 * a non-page boundary.
430 	 */
431 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
432 		phys_addr_t end = entry->addr + entry->size;
433 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
434 			unsigned long start_pfn = PFN_DOWN(start);
435 			unsigned long end_pfn = PFN_UP(end);
436 
437 			if (entry->type == E820_TYPE_RAM)
438 				end_pfn = PFN_UP(entry->addr);
439 
440 			if (start_pfn < end_pfn)
441 				ret_val = func(start_pfn, end_pfn, 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 	xen_do_remap_nonram();
500 }
501 
502 static unsigned long __init xen_get_pages_limit(void)
503 {
504 	unsigned long limit;
505 
506 	limit = MAXMEM / PAGE_SIZE;
507 	if (!xen_initial_domain() && xen_512gb_limit)
508 		limit = GB(512) / PAGE_SIZE;
509 
510 	return limit;
511 }
512 
513 static unsigned long __init xen_get_max_pages(void)
514 {
515 	unsigned long max_pages, limit;
516 	domid_t domid = DOMID_SELF;
517 	long ret;
518 
519 	limit = xen_get_pages_limit();
520 	max_pages = limit;
521 
522 	/*
523 	 * For the initial domain we use the maximum reservation as
524 	 * the maximum page.
525 	 *
526 	 * For guest domains the current maximum reservation reflects
527 	 * the current maximum rather than the static maximum. In this
528 	 * case the e820 map provided to us will cover the static
529 	 * maximum region.
530 	 */
531 	if (xen_initial_domain()) {
532 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
533 		if (ret > 0)
534 			max_pages = ret;
535 	}
536 
537 	return min(max_pages, limit);
538 }
539 
540 static void __init xen_align_and_add_e820_region(phys_addr_t start,
541 						 phys_addr_t size, int type)
542 {
543 	phys_addr_t end = start + size;
544 
545 	/* Align RAM regions to page boundaries. */
546 	if (type == E820_TYPE_RAM) {
547 		start = PAGE_ALIGN(start);
548 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
549 #ifdef CONFIG_MEMORY_HOTPLUG
550 		/*
551 		 * Don't allow adding memory not in E820 map while booting the
552 		 * system. Once the balloon driver is up it will remove that
553 		 * restriction again.
554 		 */
555 		max_mem_size = end;
556 #endif
557 	}
558 
559 	e820__range_add(start, end - start, type);
560 }
561 
562 static void __init xen_ignore_unusable(void)
563 {
564 	struct e820_entry *entry = xen_e820_table.entries;
565 	unsigned int i;
566 
567 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
568 		if (entry->type == E820_TYPE_UNUSABLE)
569 			entry->type = E820_TYPE_RAM;
570 	}
571 }
572 
573 static bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
574 {
575 	struct e820_entry *entry;
576 	unsigned mapcnt;
577 	phys_addr_t end;
578 
579 	if (!size)
580 		return false;
581 
582 	end = start + size;
583 	entry = xen_e820_table.entries;
584 
585 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
586 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
587 		    (entry->addr + entry->size) >= end)
588 			return false;
589 
590 		entry++;
591 	}
592 
593 	return true;
594 }
595 
596 /*
597  * Find a free area in physical memory not yet reserved and compliant with
598  * E820 map.
599  * Used to relocate pre-allocated areas like initrd or p2m list which are in
600  * conflict with the to be used E820 map.
601  * In case no area is found, return 0. Otherwise return the physical address
602  * of the area which is already reserved for convenience.
603  */
604 phys_addr_t __init xen_find_free_area(phys_addr_t size)
605 {
606 	unsigned mapcnt;
607 	phys_addr_t addr, start;
608 	struct e820_entry *entry = xen_e820_table.entries;
609 
610 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
611 		if (entry->type != E820_TYPE_RAM || entry->size < size)
612 			continue;
613 		start = entry->addr;
614 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
615 			if (!memblock_is_reserved(addr))
616 				continue;
617 			start = addr + PAGE_SIZE;
618 			if (start + size > entry->addr + entry->size)
619 				break;
620 		}
621 		if (addr >= start + size) {
622 			memblock_reserve(start, size);
623 			return start;
624 		}
625 	}
626 
627 	return 0;
628 }
629 
630 /*
631  * Swap a non-RAM E820 map entry with RAM above ini_nr_pages.
632  * Note that the E820 map is modified accordingly, but the P2M map isn't yet.
633  * The adaption of the P2M must be deferred until page allocation is possible.
634  */
635 static void __init xen_e820_swap_entry_with_ram(struct e820_entry *swap_entry)
636 {
637 	struct e820_entry *entry;
638 	unsigned int mapcnt;
639 	phys_addr_t mem_end = PFN_PHYS(ini_nr_pages);
640 	phys_addr_t swap_addr, swap_size, entry_end;
641 
642 	swap_addr = PAGE_ALIGN_DOWN(swap_entry->addr);
643 	swap_size = PAGE_ALIGN(swap_entry->addr - swap_addr + swap_entry->size);
644 	entry = xen_e820_table.entries;
645 
646 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
647 		entry_end = entry->addr + entry->size;
648 		if (entry->type == E820_TYPE_RAM && entry->size >= swap_size &&
649 		    entry_end - swap_size >= mem_end) {
650 			/* Reduce RAM entry by needed space (whole pages). */
651 			entry->size -= swap_size;
652 
653 			/* Add new entry at the end of E820 map. */
654 			entry = xen_e820_table.entries +
655 				xen_e820_table.nr_entries;
656 			xen_e820_table.nr_entries++;
657 
658 			/* Fill new entry (keep size and page offset). */
659 			entry->type = swap_entry->type;
660 			entry->addr = entry_end - swap_size +
661 				      swap_addr - swap_entry->addr;
662 			entry->size = swap_entry->size;
663 
664 			/* Convert old entry to RAM, align to pages. */
665 			swap_entry->type = E820_TYPE_RAM;
666 			swap_entry->addr = swap_addr;
667 			swap_entry->size = swap_size;
668 
669 			/* Remember PFN<->MFN relation for P2M update. */
670 			xen_add_remap_nonram(swap_addr, entry_end - swap_size,
671 					     swap_size);
672 
673 			/* Order E820 table and merge entries. */
674 			e820__update_table(&xen_e820_table);
675 
676 			return;
677 		}
678 
679 		entry++;
680 	}
681 
682 	xen_raw_console_write("No suitable area found for required E820 entry remapping action\n");
683 	BUG();
684 }
685 
686 /*
687  * Look for non-RAM memory types in a specific guest physical area and move
688  * those away if possible (ACPI NVS only for now).
689  */
690 static void __init xen_e820_resolve_conflicts(phys_addr_t start,
691 					      phys_addr_t size)
692 {
693 	struct e820_entry *entry;
694 	unsigned int mapcnt;
695 	phys_addr_t end;
696 
697 	if (!size)
698 		return;
699 
700 	end = start + size;
701 	entry = xen_e820_table.entries;
702 
703 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
704 		if (entry->addr >= end)
705 			return;
706 
707 		if (entry->addr + entry->size > start &&
708 		    entry->type == E820_TYPE_NVS)
709 			xen_e820_swap_entry_with_ram(entry);
710 
711 		entry++;
712 	}
713 }
714 
715 /*
716  * Check for an area in physical memory to be usable for non-movable purposes.
717  * An area is considered to usable if the used E820 map lists it to be RAM or
718  * some other type which can be moved to higher PFNs while keeping the MFNs.
719  * In case the area is not usable, crash the system with an error message.
720  */
721 void __init xen_chk_is_e820_usable(phys_addr_t start, phys_addr_t size,
722 				   const char *component)
723 {
724 	xen_e820_resolve_conflicts(start, size);
725 
726 	if (!xen_is_e820_reserved(start, size))
727 		return;
728 
729 	xen_raw_console_write("Xen hypervisor allocated ");
730 	xen_raw_console_write(component);
731 	xen_raw_console_write(" memory conflicts with E820 map\n");
732 	BUG();
733 }
734 
735 /*
736  * Like memcpy, but with physical addresses for dest and src.
737  */
738 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
739 				   phys_addr_t n)
740 {
741 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
742 	void *from, *to;
743 
744 	while (n) {
745 		dest_off = dest & ~PAGE_MASK;
746 		src_off = src & ~PAGE_MASK;
747 		dest_len = n;
748 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
749 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
750 		src_len = n;
751 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
752 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
753 		len = min(dest_len, src_len);
754 		to = early_memremap(dest - dest_off, dest_len + dest_off);
755 		from = early_memremap(src - src_off, src_len + src_off);
756 		memcpy(to, from, len);
757 		early_memunmap(to, dest_len + dest_off);
758 		early_memunmap(from, src_len + src_off);
759 		n -= len;
760 		dest += len;
761 		src += len;
762 	}
763 }
764 
765 /*
766  * Reserve Xen mfn_list.
767  */
768 static void __init xen_reserve_xen_mfnlist(void)
769 {
770 	phys_addr_t start, size;
771 
772 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
773 		start = __pa(xen_start_info->mfn_list);
774 		size = PFN_ALIGN(xen_start_info->nr_pages *
775 				 sizeof(unsigned long));
776 	} else {
777 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
778 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
779 	}
780 
781 	memblock_reserve(start, size);
782 	if (!xen_is_e820_reserved(start, size))
783 		return;
784 
785 	xen_relocate_p2m();
786 	memblock_phys_free(start, size);
787 }
788 
789 /**
790  * xen_memory_setup - Hook for machine specific memory setup.
791  **/
792 char * __init xen_memory_setup(void)
793 {
794 	unsigned long pfn_s, n_pfns;
795 	phys_addr_t mem_end, addr, size, chunk_size;
796 	u32 type;
797 	int rc;
798 	struct xen_memory_map memmap;
799 	unsigned long max_pages;
800 	unsigned long extra_pages = 0;
801 	unsigned long maxmem_pages;
802 	int i;
803 	int op;
804 
805 	xen_parse_512gb();
806 	ini_nr_pages = min(xen_get_pages_limit(), xen_start_info->nr_pages);
807 	mem_end = PFN_PHYS(ini_nr_pages);
808 
809 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
810 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
811 
812 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
813 	xen_saved_max_mem_size = max_mem_size;
814 #endif
815 
816 	op = xen_initial_domain() ?
817 		XENMEM_machine_memory_map :
818 		XENMEM_memory_map;
819 	rc = HYPERVISOR_memory_op(op, &memmap);
820 	if (rc == -ENOSYS) {
821 		BUG_ON(xen_initial_domain());
822 		memmap.nr_entries = 1;
823 		xen_e820_table.entries[0].addr = 0ULL;
824 		xen_e820_table.entries[0].size = mem_end;
825 		/* 8MB slack (to balance backend allocations). */
826 		xen_e820_table.entries[0].size += 8ULL << 20;
827 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
828 		rc = 0;
829 	}
830 	BUG_ON(rc);
831 	BUG_ON(memmap.nr_entries == 0);
832 	xen_e820_table.nr_entries = memmap.nr_entries;
833 
834 	if (xen_initial_domain()) {
835 		/*
836 		 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
837 		 * regions, so if we're using the machine memory map leave the
838 		 * region as RAM as it is in the pseudo-physical map.
839 		 *
840 		 * UNUSABLE regions in domUs are not handled and will need
841 		 * a patch in the future.
842 		 */
843 		xen_ignore_unusable();
844 
845 #ifdef CONFIG_ISCSI_IBFT_FIND
846 		/* Reserve 0.5 MiB to 1 MiB region so iBFT can be found */
847 		xen_e820_table.entries[xen_e820_table.nr_entries].addr = IBFT_START;
848 		xen_e820_table.entries[xen_e820_table.nr_entries].size = IBFT_END - IBFT_START;
849 		xen_e820_table.entries[xen_e820_table.nr_entries].type = E820_TYPE_RESERVED;
850 		xen_e820_table.nr_entries++;
851 #endif
852 	}
853 
854 	/* Make sure the Xen-supplied memory map is well-ordered. */
855 	e820__update_table(&xen_e820_table);
856 
857 	/*
858 	 * Check whether the kernel itself conflicts with the target E820 map.
859 	 * Failing now is better than running into weird problems later due
860 	 * to relocating (and even reusing) pages with kernel text or data.
861 	 */
862 	xen_chk_is_e820_usable(__pa_symbol(_text),
863 			       __pa_symbol(_end) - __pa_symbol(_text),
864 			       "kernel");
865 
866 	/*
867 	 * Check for a conflict of the xen_start_info memory with the target
868 	 * E820 map.
869 	 */
870 	xen_chk_is_e820_usable(__pa(xen_start_info), sizeof(*xen_start_info),
871 			       "xen_start_info");
872 
873 	/*
874 	 * Check for a conflict of the hypervisor supplied page tables with
875 	 * the target E820 map.
876 	 */
877 	xen_pt_check_e820();
878 
879 	max_pages = xen_get_max_pages();
880 
881 	/* How many extra pages do we need due to remapping? */
882 	max_pages += xen_foreach_remap_area(xen_count_remap_pages);
883 
884 	if (max_pages > ini_nr_pages)
885 		extra_pages += max_pages - ini_nr_pages;
886 
887 	/*
888 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
889 	 * factor the base size.
890 	 *
891 	 * Make sure we have no memory above max_pages, as this area
892 	 * isn't handled by the p2m management.
893 	 */
894 	maxmem_pages = EXTRA_MEM_RATIO * min(ini_nr_pages, PFN_DOWN(MAXMEM));
895 	extra_pages = min3(maxmem_pages, extra_pages, max_pages - ini_nr_pages);
896 	i = 0;
897 	addr = xen_e820_table.entries[0].addr;
898 	size = xen_e820_table.entries[0].size;
899 	while (i < xen_e820_table.nr_entries) {
900 		bool discard = false;
901 
902 		chunk_size = size;
903 		type = xen_e820_table.entries[i].type;
904 
905 		if (type == E820_TYPE_RESERVED)
906 			xen_pv_pci_possible = true;
907 
908 		if (type == E820_TYPE_RAM) {
909 			if (addr < mem_end) {
910 				chunk_size = min(size, mem_end - addr);
911 			} else if (extra_pages) {
912 				chunk_size = min(size, PFN_PHYS(extra_pages));
913 				pfn_s = PFN_UP(addr);
914 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
915 				extra_pages -= n_pfns;
916 				xen_add_extra_mem(pfn_s, n_pfns);
917 				xen_max_p2m_pfn = pfn_s + n_pfns;
918 			} else
919 				discard = true;
920 		}
921 
922 		if (!discard)
923 			xen_align_and_add_e820_region(addr, chunk_size, type);
924 
925 		addr += chunk_size;
926 		size -= chunk_size;
927 		if (size == 0) {
928 			i++;
929 			if (i < xen_e820_table.nr_entries) {
930 				addr = xen_e820_table.entries[i].addr;
931 				size = xen_e820_table.entries[i].size;
932 			}
933 		}
934 	}
935 
936 	/*
937 	 * Set the rest as identity mapped, in case PCI BARs are
938 	 * located here.
939 	 */
940 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
941 
942 	/*
943 	 * In domU, the ISA region is normal, usable memory, but we
944 	 * reserve ISA memory anyway because too many things poke
945 	 * about in there.
946 	 */
947 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
948 
949 	e820__update_table(e820_table);
950 
951 	xen_reserve_xen_mfnlist();
952 
953 	/* Check for a conflict of the initrd with the target E820 map. */
954 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
955 				 boot_params.hdr.ramdisk_size)) {
956 		phys_addr_t new_area, start, size;
957 
958 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
959 		if (!new_area) {
960 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
961 			BUG();
962 		}
963 
964 		start = boot_params.hdr.ramdisk_image;
965 		size = boot_params.hdr.ramdisk_size;
966 		xen_phys_memcpy(new_area, start, size);
967 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
968 			start, start + size, new_area, new_area + size);
969 		memblock_phys_free(start, size);
970 		boot_params.hdr.ramdisk_image = new_area;
971 		boot_params.ext_ramdisk_image = new_area >> 32;
972 	}
973 
974 	/*
975 	 * Set identity map on non-RAM pages and prepare remapping the
976 	 * underlying RAM.
977 	 */
978 	xen_foreach_remap_area(xen_set_identity_and_remap_chunk);
979 
980 	pr_info("Released %ld page(s)\n", xen_released_pages);
981 
982 	return "Xen";
983 }
984 
985 static int register_callback(unsigned type, const void *func)
986 {
987 	struct callback_register callback = {
988 		.type = type,
989 		.address = XEN_CALLBACK(__KERNEL_CS, func),
990 		.flags = CALLBACKF_mask_events,
991 	};
992 
993 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
994 }
995 
996 void xen_enable_sysenter(void)
997 {
998 	if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
999 	    register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
1000 		setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
1001 }
1002 
1003 void xen_enable_syscall(void)
1004 {
1005 	int ret;
1006 
1007 	ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
1008 	if (ret != 0) {
1009 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
1010 		/* Pretty fatal; 64-bit userspace has no other
1011 		   mechanism for syscalls. */
1012 	}
1013 
1014 	if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
1015 	    register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
1016 		setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1017 }
1018 
1019 static void __init xen_pvmmu_arch_setup(void)
1020 {
1021 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1022 
1023 	if (register_callback(CALLBACKTYPE_event,
1024 			      xen_asm_exc_xen_hypervisor_callback) ||
1025 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1026 		BUG();
1027 
1028 	xen_enable_sysenter();
1029 	xen_enable_syscall();
1030 }
1031 
1032 /* This function is not called for HVM domains */
1033 void __init xen_arch_setup(void)
1034 {
1035 	xen_panic_handler_init();
1036 	xen_pvmmu_arch_setup();
1037 
1038 #ifdef CONFIG_ACPI
1039 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1040 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1041 		disable_acpi();
1042 	}
1043 #endif
1044 
1045 	memcpy(boot_command_line, xen_start_info->cmd_line,
1046 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1047 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1048 
1049 	/* Set up idle, making sure it calls safe_halt() pvop */
1050 	disable_cpuidle();
1051 	disable_cpufreq();
1052 	WARN_ON(xen_set_default_idle());
1053 #ifdef CONFIG_NUMA
1054 	numa_off = 1;
1055 #endif
1056 }
1057