xref: /linux/arch/x86/xen/setup.c (revision 300a90b2cb5d442879e6398920c49aebbd5c8e40)
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 
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 /*
50  * Buffer used to remap identity mapped pages. We only need the virtual space.
51  * The physical page behind this address is remapped as needed to different
52  * buffer pages.
53  */
54 #define REMAP_SIZE	(P2M_PER_PAGE - 3)
55 static struct {
56 	unsigned long	next_area_mfn;
57 	unsigned long	target_pfn;
58 	unsigned long	size;
59 	unsigned long	mfns[REMAP_SIZE];
60 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
61 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
62 
63 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
64 
65 static void __init xen_parse_512gb(void)
66 {
67 	bool val = false;
68 	char *arg;
69 
70 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
71 	if (!arg)
72 		return;
73 
74 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
75 	if (!arg)
76 		val = true;
77 	else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
78 		return;
79 
80 	xen_512gb_limit = val;
81 }
82 
83 static void __init xen_del_extra_mem(unsigned long start_pfn,
84 				     unsigned long n_pfns)
85 {
86 	int i;
87 	unsigned long start_r, size_r;
88 
89 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
90 		start_r = xen_extra_mem[i].start_pfn;
91 		size_r = xen_extra_mem[i].n_pfns;
92 
93 		/* Start of region. */
94 		if (start_r == start_pfn) {
95 			BUG_ON(n_pfns > size_r);
96 			xen_extra_mem[i].start_pfn += n_pfns;
97 			xen_extra_mem[i].n_pfns -= n_pfns;
98 			break;
99 		}
100 		/* End of region. */
101 		if (start_r + size_r == start_pfn + n_pfns) {
102 			BUG_ON(n_pfns > size_r);
103 			xen_extra_mem[i].n_pfns -= n_pfns;
104 			break;
105 		}
106 		/* Mid of region. */
107 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
108 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
109 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
110 			/* Calling memblock_reserve() again is okay. */
111 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
112 					  (start_pfn + n_pfns));
113 			break;
114 		}
115 	}
116 	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
117 }
118 
119 /*
120  * Called during boot before the p2m list can take entries beyond the
121  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
122  * invalid.
123  */
124 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
125 {
126 	int i;
127 
128 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
129 		if (pfn >= xen_extra_mem[i].start_pfn &&
130 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
131 			return INVALID_P2M_ENTRY;
132 	}
133 
134 	return IDENTITY_FRAME(pfn);
135 }
136 
137 /*
138  * Mark all pfns of extra mem as invalid in p2m list.
139  */
140 void __init xen_inv_extra_mem(void)
141 {
142 	unsigned long pfn, pfn_s, pfn_e;
143 	int i;
144 
145 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
146 		if (!xen_extra_mem[i].n_pfns)
147 			continue;
148 		pfn_s = xen_extra_mem[i].start_pfn;
149 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
150 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
151 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
152 	}
153 }
154 
155 /*
156  * Finds the next RAM pfn available in the E820 map after min_pfn.
157  * This function updates min_pfn with the pfn found and returns
158  * the size of that range or zero if not found.
159  */
160 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
161 {
162 	const struct e820_entry *entry = xen_e820_table.entries;
163 	unsigned int i;
164 	unsigned long done = 0;
165 
166 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
167 		unsigned long s_pfn;
168 		unsigned long e_pfn;
169 
170 		if (entry->type != E820_TYPE_RAM)
171 			continue;
172 
173 		e_pfn = PFN_DOWN(entry->addr + entry->size);
174 
175 		/* We only care about E820 after this */
176 		if (e_pfn <= *min_pfn)
177 			continue;
178 
179 		s_pfn = PFN_UP(entry->addr);
180 
181 		/* If min_pfn falls within the E820 entry, we want to start
182 		 * at the min_pfn PFN.
183 		 */
184 		if (s_pfn <= *min_pfn) {
185 			done = e_pfn - *min_pfn;
186 		} else {
187 			done = e_pfn - s_pfn;
188 			*min_pfn = s_pfn;
189 		}
190 		break;
191 	}
192 
193 	return done;
194 }
195 
196 static int __init xen_free_mfn(unsigned long mfn)
197 {
198 	struct xen_memory_reservation reservation = {
199 		.address_bits = 0,
200 		.extent_order = 0,
201 		.domid        = DOMID_SELF
202 	};
203 
204 	set_xen_guest_handle(reservation.extent_start, &mfn);
205 	reservation.nr_extents = 1;
206 
207 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
208 }
209 
210 /*
211  * This releases a chunk of memory and then does the identity map. It's used
212  * as a fallback if the remapping fails.
213  */
214 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
215 			unsigned long end_pfn, unsigned long nr_pages)
216 {
217 	unsigned long pfn, end;
218 	int ret;
219 
220 	WARN_ON(start_pfn > end_pfn);
221 
222 	/* Release pages first. */
223 	end = min(end_pfn, nr_pages);
224 	for (pfn = start_pfn; pfn < end; pfn++) {
225 		unsigned long mfn = pfn_to_mfn(pfn);
226 
227 		/* Make sure pfn exists to start with */
228 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
229 			continue;
230 
231 		ret = xen_free_mfn(mfn);
232 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
233 
234 		if (ret == 1) {
235 			xen_released_pages++;
236 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
237 				break;
238 		} else
239 			break;
240 	}
241 
242 	set_phys_range_identity(start_pfn, end_pfn);
243 }
244 
245 /*
246  * Helper function to update the p2m and m2p tables and kernel mapping.
247  */
248 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
249 {
250 	struct mmu_update update = {
251 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
252 		.val = pfn
253 	};
254 
255 	/* Update p2m */
256 	if (!set_phys_to_machine(pfn, mfn)) {
257 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
258 		     pfn, mfn);
259 		BUG();
260 	}
261 
262 	/* Update m2p */
263 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
264 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
265 		     mfn, pfn);
266 		BUG();
267 	}
268 
269 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
270 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
271 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
272 		      mfn, pfn);
273 		BUG();
274 	}
275 }
276 
277 /*
278  * This function updates the p2m and m2p tables with an identity map from
279  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
280  * original allocation at remap_pfn. The information needed for remapping is
281  * saved in the memory itself to avoid the need for allocating buffers. The
282  * complete remap information is contained in a list of MFNs each containing
283  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
284  * This enables us to preserve the original mfn sequence while doing the
285  * remapping at a time when the memory management is capable of allocating
286  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
287  * its callers.
288  */
289 static void __init xen_do_set_identity_and_remap_chunk(
290         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
291 {
292 	unsigned long buf = (unsigned long)&xen_remap_buf;
293 	unsigned long mfn_save, mfn;
294 	unsigned long ident_pfn_iter, remap_pfn_iter;
295 	unsigned long ident_end_pfn = start_pfn + size;
296 	unsigned long left = size;
297 	unsigned int i, chunk;
298 
299 	WARN_ON(size == 0);
300 
301 	mfn_save = virt_to_mfn((void *)buf);
302 
303 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
304 	     ident_pfn_iter < ident_end_pfn;
305 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
306 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
307 
308 		/* Map first pfn to xen_remap_buf */
309 		mfn = pfn_to_mfn(ident_pfn_iter);
310 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
311 
312 		/* Save mapping information in page */
313 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
314 		xen_remap_buf.target_pfn = remap_pfn_iter;
315 		xen_remap_buf.size = chunk;
316 		for (i = 0; i < chunk; i++)
317 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
318 
319 		/* Put remap buf into list. */
320 		xen_remap_mfn = mfn;
321 
322 		/* Set identity map */
323 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
324 
325 		left -= chunk;
326 	}
327 
328 	/* Restore old xen_remap_buf mapping */
329 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
330 }
331 
332 /*
333  * This function takes a contiguous pfn range that needs to be identity mapped
334  * and:
335  *
336  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
337  *  2) Calls the do_ function to actually do the mapping/remapping work.
338  *
339  * The goal is to not allocate additional memory but to remap the existing
340  * pages. In the case of an error the underlying memory is simply released back
341  * to Xen and not remapped.
342  */
343 static unsigned long __init xen_set_identity_and_remap_chunk(
344 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
345 	unsigned long remap_pfn)
346 {
347 	unsigned long pfn;
348 	unsigned long i = 0;
349 	unsigned long n = end_pfn - start_pfn;
350 
351 	if (remap_pfn == 0)
352 		remap_pfn = nr_pages;
353 
354 	while (i < n) {
355 		unsigned long cur_pfn = start_pfn + i;
356 		unsigned long left = n - i;
357 		unsigned long size = left;
358 		unsigned long remap_range_size;
359 
360 		/* Do not remap pages beyond the current allocation */
361 		if (cur_pfn >= nr_pages) {
362 			/* Identity map remaining pages */
363 			set_phys_range_identity(cur_pfn, cur_pfn + size);
364 			break;
365 		}
366 		if (cur_pfn + size > nr_pages)
367 			size = nr_pages - cur_pfn;
368 
369 		remap_range_size = xen_find_pfn_range(&remap_pfn);
370 		if (!remap_range_size) {
371 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
372 			xen_set_identity_and_release_chunk(cur_pfn,
373 						cur_pfn + left, nr_pages);
374 			break;
375 		}
376 		/* Adjust size to fit in current e820 RAM region */
377 		if (size > remap_range_size)
378 			size = remap_range_size;
379 
380 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
381 
382 		/* Update variables to reflect new mappings. */
383 		i += size;
384 		remap_pfn += size;
385 	}
386 
387 	/*
388 	 * If the PFNs are currently mapped, their VA mappings need to be
389 	 * zapped.
390 	 */
391 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
392 		(void)HYPERVISOR_update_va_mapping(
393 			(unsigned long)__va(pfn << PAGE_SHIFT),
394 			native_make_pte(0), 0);
395 
396 	return remap_pfn;
397 }
398 
399 static unsigned long __init xen_count_remap_pages(
400 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
401 	unsigned long remap_pages)
402 {
403 	if (start_pfn >= nr_pages)
404 		return remap_pages;
405 
406 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
407 }
408 
409 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
410 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
411 			      unsigned long nr_pages, unsigned long last_val))
412 {
413 	phys_addr_t start = 0;
414 	unsigned long ret_val = 0;
415 	const struct e820_entry *entry = xen_e820_table.entries;
416 	int i;
417 
418 	/*
419 	 * Combine non-RAM regions and gaps until a RAM region (or the
420 	 * end of the map) is reached, then call the provided function
421 	 * to perform its duty on the non-RAM region.
422 	 *
423 	 * The combined non-RAM regions are rounded to a whole number
424 	 * of pages so any partial pages are accessible via the 1:1
425 	 * mapping.  This is needed for some BIOSes that put (for
426 	 * example) the DMI tables in a reserved region that begins on
427 	 * a non-page boundary.
428 	 */
429 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
430 		phys_addr_t end = entry->addr + entry->size;
431 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
432 			unsigned long start_pfn = PFN_DOWN(start);
433 			unsigned long end_pfn = PFN_UP(end);
434 
435 			if (entry->type == E820_TYPE_RAM)
436 				end_pfn = PFN_UP(entry->addr);
437 
438 			if (start_pfn < end_pfn)
439 				ret_val = func(start_pfn, end_pfn, nr_pages,
440 					       ret_val);
441 			start = end;
442 		}
443 	}
444 
445 	return ret_val;
446 }
447 
448 /*
449  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
450  * The remap information (which mfn remap to which pfn) is contained in the
451  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
452  * This scheme allows to remap the different chunks in arbitrary order while
453  * the resulting mapping will be independent from the order.
454  */
455 void __init xen_remap_memory(void)
456 {
457 	unsigned long buf = (unsigned long)&xen_remap_buf;
458 	unsigned long mfn_save, pfn;
459 	unsigned long remapped = 0;
460 	unsigned int i;
461 	unsigned long pfn_s = ~0UL;
462 	unsigned long len = 0;
463 
464 	mfn_save = virt_to_mfn((void *)buf);
465 
466 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
467 		/* Map the remap information */
468 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
469 
470 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
471 
472 		pfn = xen_remap_buf.target_pfn;
473 		for (i = 0; i < xen_remap_buf.size; i++) {
474 			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
475 			remapped++;
476 			pfn++;
477 		}
478 		if (pfn_s == ~0UL || pfn == pfn_s) {
479 			pfn_s = xen_remap_buf.target_pfn;
480 			len += xen_remap_buf.size;
481 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
482 			len += xen_remap_buf.size;
483 		} else {
484 			xen_del_extra_mem(pfn_s, len);
485 			pfn_s = xen_remap_buf.target_pfn;
486 			len = xen_remap_buf.size;
487 		}
488 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
489 	}
490 
491 	if (pfn_s != ~0UL && len)
492 		xen_del_extra_mem(pfn_s, len);
493 
494 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
495 
496 	pr_info("Remapped %ld page(s)\n", remapped);
497 }
498 
499 static unsigned long __init xen_get_pages_limit(void)
500 {
501 	unsigned long limit;
502 
503 	limit = MAXMEM / PAGE_SIZE;
504 	if (!xen_initial_domain() && xen_512gb_limit)
505 		limit = GB(512) / PAGE_SIZE;
506 
507 	return limit;
508 }
509 
510 static unsigned long __init xen_get_max_pages(void)
511 {
512 	unsigned long max_pages, limit;
513 	domid_t domid = DOMID_SELF;
514 	long ret;
515 
516 	limit = xen_get_pages_limit();
517 	max_pages = limit;
518 
519 	/*
520 	 * For the initial domain we use the maximum reservation as
521 	 * the maximum page.
522 	 *
523 	 * For guest domains the current maximum reservation reflects
524 	 * the current maximum rather than the static maximum. In this
525 	 * case the e820 map provided to us will cover the static
526 	 * maximum region.
527 	 */
528 	if (xen_initial_domain()) {
529 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
530 		if (ret > 0)
531 			max_pages = ret;
532 	}
533 
534 	return min(max_pages, limit);
535 }
536 
537 static void __init xen_align_and_add_e820_region(phys_addr_t start,
538 						 phys_addr_t size, int type)
539 {
540 	phys_addr_t end = start + size;
541 
542 	/* Align RAM regions to page boundaries. */
543 	if (type == E820_TYPE_RAM) {
544 		start = PAGE_ALIGN(start);
545 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
546 #ifdef CONFIG_MEMORY_HOTPLUG
547 		/*
548 		 * Don't allow adding memory not in E820 map while booting the
549 		 * system. Once the balloon driver is up it will remove that
550 		 * restriction again.
551 		 */
552 		max_mem_size = end;
553 #endif
554 	}
555 
556 	e820__range_add(start, end - start, type);
557 }
558 
559 static void __init xen_ignore_unusable(void)
560 {
561 	struct e820_entry *entry = xen_e820_table.entries;
562 	unsigned int i;
563 
564 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
565 		if (entry->type == E820_TYPE_UNUSABLE)
566 			entry->type = E820_TYPE_RAM;
567 	}
568 }
569 
570 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
571 {
572 	struct e820_entry *entry;
573 	unsigned mapcnt;
574 	phys_addr_t end;
575 
576 	if (!size)
577 		return false;
578 
579 	end = start + size;
580 	entry = xen_e820_table.entries;
581 
582 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
583 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
584 		    (entry->addr + entry->size) >= end)
585 			return false;
586 
587 		entry++;
588 	}
589 
590 	return true;
591 }
592 
593 /*
594  * Find a free area in physical memory not yet reserved and compliant with
595  * E820 map.
596  * Used to relocate pre-allocated areas like initrd or p2m list which are in
597  * conflict with the to be used E820 map.
598  * In case no area is found, return 0. Otherwise return the physical address
599  * of the area which is already reserved for convenience.
600  */
601 phys_addr_t __init xen_find_free_area(phys_addr_t size)
602 {
603 	unsigned mapcnt;
604 	phys_addr_t addr, start;
605 	struct e820_entry *entry = xen_e820_table.entries;
606 
607 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
608 		if (entry->type != E820_TYPE_RAM || entry->size < size)
609 			continue;
610 		start = entry->addr;
611 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
612 			if (!memblock_is_reserved(addr))
613 				continue;
614 			start = addr + PAGE_SIZE;
615 			if (start + size > entry->addr + entry->size)
616 				break;
617 		}
618 		if (addr >= start + size) {
619 			memblock_reserve(start, size);
620 			return start;
621 		}
622 	}
623 
624 	return 0;
625 }
626 
627 /*
628  * Like memcpy, but with physical addresses for dest and src.
629  */
630 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
631 				   phys_addr_t n)
632 {
633 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
634 	void *from, *to;
635 
636 	while (n) {
637 		dest_off = dest & ~PAGE_MASK;
638 		src_off = src & ~PAGE_MASK;
639 		dest_len = n;
640 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
641 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
642 		src_len = n;
643 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
644 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
645 		len = min(dest_len, src_len);
646 		to = early_memremap(dest - dest_off, dest_len + dest_off);
647 		from = early_memremap(src - src_off, src_len + src_off);
648 		memcpy(to, from, len);
649 		early_memunmap(to, dest_len + dest_off);
650 		early_memunmap(from, src_len + src_off);
651 		n -= len;
652 		dest += len;
653 		src += len;
654 	}
655 }
656 
657 /*
658  * Reserve Xen mfn_list.
659  */
660 static void __init xen_reserve_xen_mfnlist(void)
661 {
662 	phys_addr_t start, size;
663 
664 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
665 		start = __pa(xen_start_info->mfn_list);
666 		size = PFN_ALIGN(xen_start_info->nr_pages *
667 				 sizeof(unsigned long));
668 	} else {
669 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
670 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
671 	}
672 
673 	memblock_reserve(start, size);
674 	if (!xen_is_e820_reserved(start, size))
675 		return;
676 
677 	xen_relocate_p2m();
678 	memblock_phys_free(start, size);
679 }
680 
681 /**
682  * xen_memory_setup - Hook for machine specific memory setup.
683  **/
684 char * __init xen_memory_setup(void)
685 {
686 	unsigned long max_pfn, pfn_s, n_pfns;
687 	phys_addr_t mem_end, addr, size, chunk_size;
688 	u32 type;
689 	int rc;
690 	struct xen_memory_map memmap;
691 	unsigned long max_pages;
692 	unsigned long extra_pages = 0;
693 	unsigned long maxmem_pages;
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 	maxmem_pages = EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM));
766 	extra_pages = min3(maxmem_pages, 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