xref: /openbmc/linux/arch/x86/xen/setup.c (revision 1b36955c)
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 /* Amount of extra memory space we add to the e820 ranges */
42 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
43 
44 /* Number of pages released from the initial allocation. */
45 unsigned long xen_released_pages;
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 /*
65  * The maximum amount of extra memory compared to the base size.  The
66  * main scaling factor is the size of struct page.  At extreme ratios
67  * of base:extra, all the base memory can be filled with page
68  * structures for the extra memory, leaving no space for anything
69  * else.
70  *
71  * 10x seems like a reasonable balance between scaling flexibility and
72  * leaving a practically usable system.
73  */
74 #define EXTRA_MEM_RATIO		(10)
75 
76 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
77 
78 static void __init xen_parse_512gb(void)
79 {
80 	bool val = false;
81 	char *arg;
82 
83 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
84 	if (!arg)
85 		return;
86 
87 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
88 	if (!arg)
89 		val = true;
90 	else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
91 		return;
92 
93 	xen_512gb_limit = val;
94 }
95 
96 static void __init xen_add_extra_mem(unsigned long start_pfn,
97 				     unsigned long n_pfns)
98 {
99 	int i;
100 
101 	/*
102 	 * No need to check for zero size, should happen rarely and will only
103 	 * write a new entry regarded to be unused due to zero size.
104 	 */
105 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
106 		/* Add new region. */
107 		if (xen_extra_mem[i].n_pfns == 0) {
108 			xen_extra_mem[i].start_pfn = start_pfn;
109 			xen_extra_mem[i].n_pfns = n_pfns;
110 			break;
111 		}
112 		/* Append to existing region. */
113 		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
114 		    start_pfn) {
115 			xen_extra_mem[i].n_pfns += n_pfns;
116 			break;
117 		}
118 	}
119 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
120 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
121 
122 	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
123 }
124 
125 static void __init xen_del_extra_mem(unsigned long start_pfn,
126 				     unsigned long n_pfns)
127 {
128 	int i;
129 	unsigned long start_r, size_r;
130 
131 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
132 		start_r = xen_extra_mem[i].start_pfn;
133 		size_r = xen_extra_mem[i].n_pfns;
134 
135 		/* Start of region. */
136 		if (start_r == start_pfn) {
137 			BUG_ON(n_pfns > size_r);
138 			xen_extra_mem[i].start_pfn += n_pfns;
139 			xen_extra_mem[i].n_pfns -= n_pfns;
140 			break;
141 		}
142 		/* End of region. */
143 		if (start_r + size_r == start_pfn + n_pfns) {
144 			BUG_ON(n_pfns > size_r);
145 			xen_extra_mem[i].n_pfns -= n_pfns;
146 			break;
147 		}
148 		/* Mid of region. */
149 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
150 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
151 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
152 			/* Calling memblock_reserve() again is okay. */
153 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
154 					  (start_pfn + n_pfns));
155 			break;
156 		}
157 	}
158 	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
159 }
160 
161 /*
162  * Called during boot before the p2m list can take entries beyond the
163  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
164  * invalid.
165  */
166 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
167 {
168 	int i;
169 
170 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
171 		if (pfn >= xen_extra_mem[i].start_pfn &&
172 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
173 			return INVALID_P2M_ENTRY;
174 	}
175 
176 	return IDENTITY_FRAME(pfn);
177 }
178 
179 /*
180  * Mark all pfns of extra mem as invalid in p2m list.
181  */
182 void __init xen_inv_extra_mem(void)
183 {
184 	unsigned long pfn, pfn_s, pfn_e;
185 	int i;
186 
187 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
188 		if (!xen_extra_mem[i].n_pfns)
189 			continue;
190 		pfn_s = xen_extra_mem[i].start_pfn;
191 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
192 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
193 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
194 	}
195 }
196 
197 /*
198  * Finds the next RAM pfn available in the E820 map after min_pfn.
199  * This function updates min_pfn with the pfn found and returns
200  * the size of that range or zero if not found.
201  */
202 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
203 {
204 	const struct e820_entry *entry = xen_e820_table.entries;
205 	unsigned int i;
206 	unsigned long done = 0;
207 
208 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
209 		unsigned long s_pfn;
210 		unsigned long e_pfn;
211 
212 		if (entry->type != E820_TYPE_RAM)
213 			continue;
214 
215 		e_pfn = PFN_DOWN(entry->addr + entry->size);
216 
217 		/* We only care about E820 after this */
218 		if (e_pfn <= *min_pfn)
219 			continue;
220 
221 		s_pfn = PFN_UP(entry->addr);
222 
223 		/* If min_pfn falls within the E820 entry, we want to start
224 		 * at the min_pfn PFN.
225 		 */
226 		if (s_pfn <= *min_pfn) {
227 			done = e_pfn - *min_pfn;
228 		} else {
229 			done = e_pfn - s_pfn;
230 			*min_pfn = s_pfn;
231 		}
232 		break;
233 	}
234 
235 	return done;
236 }
237 
238 static int __init xen_free_mfn(unsigned long mfn)
239 {
240 	struct xen_memory_reservation reservation = {
241 		.address_bits = 0,
242 		.extent_order = 0,
243 		.domid        = DOMID_SELF
244 	};
245 
246 	set_xen_guest_handle(reservation.extent_start, &mfn);
247 	reservation.nr_extents = 1;
248 
249 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
250 }
251 
252 /*
253  * This releases a chunk of memory and then does the identity map. It's used
254  * as a fallback if the remapping fails.
255  */
256 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
257 			unsigned long end_pfn, unsigned long nr_pages)
258 {
259 	unsigned long pfn, end;
260 	int ret;
261 
262 	WARN_ON(start_pfn > end_pfn);
263 
264 	/* Release pages first. */
265 	end = min(end_pfn, nr_pages);
266 	for (pfn = start_pfn; pfn < end; pfn++) {
267 		unsigned long mfn = pfn_to_mfn(pfn);
268 
269 		/* Make sure pfn exists to start with */
270 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
271 			continue;
272 
273 		ret = xen_free_mfn(mfn);
274 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
275 
276 		if (ret == 1) {
277 			xen_released_pages++;
278 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
279 				break;
280 		} else
281 			break;
282 	}
283 
284 	set_phys_range_identity(start_pfn, end_pfn);
285 }
286 
287 /*
288  * Helper function to update the p2m and m2p tables and kernel mapping.
289  */
290 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
291 {
292 	struct mmu_update update = {
293 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
294 		.val = pfn
295 	};
296 
297 	/* Update p2m */
298 	if (!set_phys_to_machine(pfn, mfn)) {
299 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
300 		     pfn, mfn);
301 		BUG();
302 	}
303 
304 	/* Update m2p */
305 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
306 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
307 		     mfn, pfn);
308 		BUG();
309 	}
310 
311 	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
312 					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
313 		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
314 		      mfn, pfn);
315 		BUG();
316 	}
317 }
318 
319 /*
320  * This function updates the p2m and m2p tables with an identity map from
321  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
322  * original allocation at remap_pfn. The information needed for remapping is
323  * saved in the memory itself to avoid the need for allocating buffers. The
324  * complete remap information is contained in a list of MFNs each containing
325  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
326  * This enables us to preserve the original mfn sequence while doing the
327  * remapping at a time when the memory management is capable of allocating
328  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
329  * its callers.
330  */
331 static void __init xen_do_set_identity_and_remap_chunk(
332         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
333 {
334 	unsigned long buf = (unsigned long)&xen_remap_buf;
335 	unsigned long mfn_save, mfn;
336 	unsigned long ident_pfn_iter, remap_pfn_iter;
337 	unsigned long ident_end_pfn = start_pfn + size;
338 	unsigned long left = size;
339 	unsigned int i, chunk;
340 
341 	WARN_ON(size == 0);
342 
343 	mfn_save = virt_to_mfn((void *)buf);
344 
345 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
346 	     ident_pfn_iter < ident_end_pfn;
347 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
348 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
349 
350 		/* Map first pfn to xen_remap_buf */
351 		mfn = pfn_to_mfn(ident_pfn_iter);
352 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
353 
354 		/* Save mapping information in page */
355 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
356 		xen_remap_buf.target_pfn = remap_pfn_iter;
357 		xen_remap_buf.size = chunk;
358 		for (i = 0; i < chunk; i++)
359 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
360 
361 		/* Put remap buf into list. */
362 		xen_remap_mfn = mfn;
363 
364 		/* Set identity map */
365 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
366 
367 		left -= chunk;
368 	}
369 
370 	/* Restore old xen_remap_buf mapping */
371 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
372 }
373 
374 /*
375  * This function takes a contiguous pfn range that needs to be identity mapped
376  * and:
377  *
378  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
379  *  2) Calls the do_ function to actually do the mapping/remapping work.
380  *
381  * The goal is to not allocate additional memory but to remap the existing
382  * pages. In the case of an error the underlying memory is simply released back
383  * to Xen and not remapped.
384  */
385 static unsigned long __init xen_set_identity_and_remap_chunk(
386 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
387 	unsigned long remap_pfn)
388 {
389 	unsigned long pfn;
390 	unsigned long i = 0;
391 	unsigned long n = end_pfn - start_pfn;
392 
393 	if (remap_pfn == 0)
394 		remap_pfn = nr_pages;
395 
396 	while (i < n) {
397 		unsigned long cur_pfn = start_pfn + i;
398 		unsigned long left = n - i;
399 		unsigned long size = left;
400 		unsigned long remap_range_size;
401 
402 		/* Do not remap pages beyond the current allocation */
403 		if (cur_pfn >= nr_pages) {
404 			/* Identity map remaining pages */
405 			set_phys_range_identity(cur_pfn, cur_pfn + size);
406 			break;
407 		}
408 		if (cur_pfn + size > nr_pages)
409 			size = nr_pages - cur_pfn;
410 
411 		remap_range_size = xen_find_pfn_range(&remap_pfn);
412 		if (!remap_range_size) {
413 			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
414 			xen_set_identity_and_release_chunk(cur_pfn,
415 						cur_pfn + left, nr_pages);
416 			break;
417 		}
418 		/* Adjust size to fit in current e820 RAM region */
419 		if (size > remap_range_size)
420 			size = remap_range_size;
421 
422 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
423 
424 		/* Update variables to reflect new mappings. */
425 		i += size;
426 		remap_pfn += size;
427 	}
428 
429 	/*
430 	 * If the PFNs are currently mapped, their VA mappings need to be
431 	 * zapped.
432 	 */
433 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
434 		(void)HYPERVISOR_update_va_mapping(
435 			(unsigned long)__va(pfn << PAGE_SHIFT),
436 			native_make_pte(0), 0);
437 
438 	return remap_pfn;
439 }
440 
441 static unsigned long __init xen_count_remap_pages(
442 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
443 	unsigned long remap_pages)
444 {
445 	if (start_pfn >= nr_pages)
446 		return remap_pages;
447 
448 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
449 }
450 
451 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
452 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
453 			      unsigned long nr_pages, unsigned long last_val))
454 {
455 	phys_addr_t start = 0;
456 	unsigned long ret_val = 0;
457 	const struct e820_entry *entry = xen_e820_table.entries;
458 	int i;
459 
460 	/*
461 	 * Combine non-RAM regions and gaps until a RAM region (or the
462 	 * end of the map) is reached, then call the provided function
463 	 * to perform its duty on the non-RAM region.
464 	 *
465 	 * The combined non-RAM regions are rounded to a whole number
466 	 * of pages so any partial pages are accessible via the 1:1
467 	 * mapping.  This is needed for some BIOSes that put (for
468 	 * example) the DMI tables in a reserved region that begins on
469 	 * a non-page boundary.
470 	 */
471 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
472 		phys_addr_t end = entry->addr + entry->size;
473 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
474 			unsigned long start_pfn = PFN_DOWN(start);
475 			unsigned long end_pfn = PFN_UP(end);
476 
477 			if (entry->type == E820_TYPE_RAM)
478 				end_pfn = PFN_UP(entry->addr);
479 
480 			if (start_pfn < end_pfn)
481 				ret_val = func(start_pfn, end_pfn, nr_pages,
482 					       ret_val);
483 			start = end;
484 		}
485 	}
486 
487 	return ret_val;
488 }
489 
490 /*
491  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
492  * The remap information (which mfn remap to which pfn) is contained in the
493  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
494  * This scheme allows to remap the different chunks in arbitrary order while
495  * the resulting mapping will be independent from the order.
496  */
497 void __init xen_remap_memory(void)
498 {
499 	unsigned long buf = (unsigned long)&xen_remap_buf;
500 	unsigned long mfn_save, pfn;
501 	unsigned long remapped = 0;
502 	unsigned int i;
503 	unsigned long pfn_s = ~0UL;
504 	unsigned long len = 0;
505 
506 	mfn_save = virt_to_mfn((void *)buf);
507 
508 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
509 		/* Map the remap information */
510 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
511 
512 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
513 
514 		pfn = xen_remap_buf.target_pfn;
515 		for (i = 0; i < xen_remap_buf.size; i++) {
516 			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
517 			remapped++;
518 			pfn++;
519 		}
520 		if (pfn_s == ~0UL || pfn == pfn_s) {
521 			pfn_s = xen_remap_buf.target_pfn;
522 			len += xen_remap_buf.size;
523 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
524 			len += xen_remap_buf.size;
525 		} else {
526 			xen_del_extra_mem(pfn_s, len);
527 			pfn_s = xen_remap_buf.target_pfn;
528 			len = xen_remap_buf.size;
529 		}
530 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
531 	}
532 
533 	if (pfn_s != ~0UL && len)
534 		xen_del_extra_mem(pfn_s, len);
535 
536 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
537 
538 	pr_info("Remapped %ld page(s)\n", remapped);
539 }
540 
541 static unsigned long __init xen_get_pages_limit(void)
542 {
543 	unsigned long limit;
544 
545 	limit = MAXMEM / PAGE_SIZE;
546 	if (!xen_initial_domain() && xen_512gb_limit)
547 		limit = GB(512) / PAGE_SIZE;
548 
549 	return limit;
550 }
551 
552 static unsigned long __init xen_get_max_pages(void)
553 {
554 	unsigned long max_pages, limit;
555 	domid_t domid = DOMID_SELF;
556 	long ret;
557 
558 	limit = xen_get_pages_limit();
559 	max_pages = limit;
560 
561 	/*
562 	 * For the initial domain we use the maximum reservation as
563 	 * the maximum page.
564 	 *
565 	 * For guest domains the current maximum reservation reflects
566 	 * the current maximum rather than the static maximum. In this
567 	 * case the e820 map provided to us will cover the static
568 	 * maximum region.
569 	 */
570 	if (xen_initial_domain()) {
571 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
572 		if (ret > 0)
573 			max_pages = ret;
574 	}
575 
576 	return min(max_pages, limit);
577 }
578 
579 static void __init xen_align_and_add_e820_region(phys_addr_t start,
580 						 phys_addr_t size, int type)
581 {
582 	phys_addr_t end = start + size;
583 
584 	/* Align RAM regions to page boundaries. */
585 	if (type == E820_TYPE_RAM) {
586 		start = PAGE_ALIGN(start);
587 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
588 #ifdef CONFIG_MEMORY_HOTPLUG
589 		/*
590 		 * Don't allow adding memory not in E820 map while booting the
591 		 * system. Once the balloon driver is up it will remove that
592 		 * restriction again.
593 		 */
594 		max_mem_size = end;
595 #endif
596 	}
597 
598 	e820__range_add(start, end - start, type);
599 }
600 
601 static void __init xen_ignore_unusable(void)
602 {
603 	struct e820_entry *entry = xen_e820_table.entries;
604 	unsigned int i;
605 
606 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
607 		if (entry->type == E820_TYPE_UNUSABLE)
608 			entry->type = E820_TYPE_RAM;
609 	}
610 }
611 
612 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
613 {
614 	struct e820_entry *entry;
615 	unsigned mapcnt;
616 	phys_addr_t end;
617 
618 	if (!size)
619 		return false;
620 
621 	end = start + size;
622 	entry = xen_e820_table.entries;
623 
624 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
625 		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
626 		    (entry->addr + entry->size) >= end)
627 			return false;
628 
629 		entry++;
630 	}
631 
632 	return true;
633 }
634 
635 /*
636  * Find a free area in physical memory not yet reserved and compliant with
637  * E820 map.
638  * Used to relocate pre-allocated areas like initrd or p2m list which are in
639  * conflict with the to be used E820 map.
640  * In case no area is found, return 0. Otherwise return the physical address
641  * of the area which is already reserved for convenience.
642  */
643 phys_addr_t __init xen_find_free_area(phys_addr_t size)
644 {
645 	unsigned mapcnt;
646 	phys_addr_t addr, start;
647 	struct e820_entry *entry = xen_e820_table.entries;
648 
649 	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
650 		if (entry->type != E820_TYPE_RAM || entry->size < size)
651 			continue;
652 		start = entry->addr;
653 		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
654 			if (!memblock_is_reserved(addr))
655 				continue;
656 			start = addr + PAGE_SIZE;
657 			if (start + size > entry->addr + entry->size)
658 				break;
659 		}
660 		if (addr >= start + size) {
661 			memblock_reserve(start, size);
662 			return start;
663 		}
664 	}
665 
666 	return 0;
667 }
668 
669 /*
670  * Like memcpy, but with physical addresses for dest and src.
671  */
672 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
673 				   phys_addr_t n)
674 {
675 	phys_addr_t dest_off, src_off, dest_len, src_len, len;
676 	void *from, *to;
677 
678 	while (n) {
679 		dest_off = dest & ~PAGE_MASK;
680 		src_off = src & ~PAGE_MASK;
681 		dest_len = n;
682 		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
683 			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
684 		src_len = n;
685 		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
686 			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
687 		len = min(dest_len, src_len);
688 		to = early_memremap(dest - dest_off, dest_len + dest_off);
689 		from = early_memremap(src - src_off, src_len + src_off);
690 		memcpy(to, from, len);
691 		early_memunmap(to, dest_len + dest_off);
692 		early_memunmap(from, src_len + src_off);
693 		n -= len;
694 		dest += len;
695 		src += len;
696 	}
697 }
698 
699 /*
700  * Reserve Xen mfn_list.
701  */
702 static void __init xen_reserve_xen_mfnlist(void)
703 {
704 	phys_addr_t start, size;
705 
706 	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
707 		start = __pa(xen_start_info->mfn_list);
708 		size = PFN_ALIGN(xen_start_info->nr_pages *
709 				 sizeof(unsigned long));
710 	} else {
711 		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
712 		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
713 	}
714 
715 	memblock_reserve(start, size);
716 	if (!xen_is_e820_reserved(start, size))
717 		return;
718 
719 	xen_relocate_p2m();
720 	memblock_phys_free(start, size);
721 }
722 
723 /**
724  * xen_memory_setup - Hook for machine specific memory setup.
725  **/
726 char * __init xen_memory_setup(void)
727 {
728 	unsigned long max_pfn, pfn_s, n_pfns;
729 	phys_addr_t mem_end, addr, size, chunk_size;
730 	u32 type;
731 	int rc;
732 	struct xen_memory_map memmap;
733 	unsigned long max_pages;
734 	unsigned long extra_pages = 0;
735 	int i;
736 	int op;
737 
738 	xen_parse_512gb();
739 	max_pfn = xen_get_pages_limit();
740 	max_pfn = min(max_pfn, xen_start_info->nr_pages);
741 	mem_end = PFN_PHYS(max_pfn);
742 
743 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
744 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
745 
746 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
747 	xen_saved_max_mem_size = max_mem_size;
748 #endif
749 
750 	op = xen_initial_domain() ?
751 		XENMEM_machine_memory_map :
752 		XENMEM_memory_map;
753 	rc = HYPERVISOR_memory_op(op, &memmap);
754 	if (rc == -ENOSYS) {
755 		BUG_ON(xen_initial_domain());
756 		memmap.nr_entries = 1;
757 		xen_e820_table.entries[0].addr = 0ULL;
758 		xen_e820_table.entries[0].size = mem_end;
759 		/* 8MB slack (to balance backend allocations). */
760 		xen_e820_table.entries[0].size += 8ULL << 20;
761 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
762 		rc = 0;
763 	}
764 	BUG_ON(rc);
765 	BUG_ON(memmap.nr_entries == 0);
766 	xen_e820_table.nr_entries = memmap.nr_entries;
767 
768 	if (xen_initial_domain()) {
769 		/*
770 		 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
771 		 * regions, so if we're using the machine memory map leave the
772 		 * region as RAM as it is in the pseudo-physical map.
773 		 *
774 		 * UNUSABLE regions in domUs are not handled and will need
775 		 * a patch in the future.
776 		 */
777 		xen_ignore_unusable();
778 
779 #ifdef CONFIG_ISCSI_IBFT_FIND
780 		/* Reserve 0.5 MiB to 1 MiB region so iBFT can be found */
781 		xen_e820_table.entries[xen_e820_table.nr_entries].addr = IBFT_START;
782 		xen_e820_table.entries[xen_e820_table.nr_entries].size = IBFT_END - IBFT_START;
783 		xen_e820_table.entries[xen_e820_table.nr_entries].type = E820_TYPE_RESERVED;
784 		xen_e820_table.nr_entries++;
785 #endif
786 	}
787 
788 	/* Make sure the Xen-supplied memory map is well-ordered. */
789 	e820__update_table(&xen_e820_table);
790 
791 	max_pages = xen_get_max_pages();
792 
793 	/* How many extra pages do we need due to remapping? */
794 	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
795 
796 	if (max_pages > max_pfn)
797 		extra_pages += max_pages - max_pfn;
798 
799 	/*
800 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
801 	 * factor the base size.
802 	 *
803 	 * Make sure we have no memory above max_pages, as this area
804 	 * isn't handled by the p2m management.
805 	 */
806 	extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
807 			   extra_pages, max_pages - max_pfn);
808 	i = 0;
809 	addr = xen_e820_table.entries[0].addr;
810 	size = xen_e820_table.entries[0].size;
811 	while (i < xen_e820_table.nr_entries) {
812 		bool discard = false;
813 
814 		chunk_size = size;
815 		type = xen_e820_table.entries[i].type;
816 
817 		if (type == E820_TYPE_RAM) {
818 			if (addr < mem_end) {
819 				chunk_size = min(size, mem_end - addr);
820 			} else if (extra_pages) {
821 				chunk_size = min(size, PFN_PHYS(extra_pages));
822 				pfn_s = PFN_UP(addr);
823 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
824 				extra_pages -= n_pfns;
825 				xen_add_extra_mem(pfn_s, n_pfns);
826 				xen_max_p2m_pfn = pfn_s + n_pfns;
827 			} else
828 				discard = true;
829 		}
830 
831 		if (!discard)
832 			xen_align_and_add_e820_region(addr, chunk_size, type);
833 
834 		addr += chunk_size;
835 		size -= chunk_size;
836 		if (size == 0) {
837 			i++;
838 			if (i < xen_e820_table.nr_entries) {
839 				addr = xen_e820_table.entries[i].addr;
840 				size = xen_e820_table.entries[i].size;
841 			}
842 		}
843 	}
844 
845 	/*
846 	 * Set the rest as identity mapped, in case PCI BARs are
847 	 * located here.
848 	 */
849 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
850 
851 	/*
852 	 * In domU, the ISA region is normal, usable memory, but we
853 	 * reserve ISA memory anyway because too many things poke
854 	 * about in there.
855 	 */
856 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
857 
858 	e820__update_table(e820_table);
859 
860 	/*
861 	 * Check whether the kernel itself conflicts with the target E820 map.
862 	 * Failing now is better than running into weird problems later due
863 	 * to relocating (and even reusing) pages with kernel text or data.
864 	 */
865 	if (xen_is_e820_reserved(__pa_symbol(_text),
866 			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
867 		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
868 		BUG();
869 	}
870 
871 	/*
872 	 * Check for a conflict of the hypervisor supplied page tables with
873 	 * the target E820 map.
874 	 */
875 	xen_pt_check_e820();
876 
877 	xen_reserve_xen_mfnlist();
878 
879 	/* Check for a conflict of the initrd with the target E820 map. */
880 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
881 				 boot_params.hdr.ramdisk_size)) {
882 		phys_addr_t new_area, start, size;
883 
884 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
885 		if (!new_area) {
886 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
887 			BUG();
888 		}
889 
890 		start = boot_params.hdr.ramdisk_image;
891 		size = boot_params.hdr.ramdisk_size;
892 		xen_phys_memcpy(new_area, start, size);
893 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
894 			start, start + size, new_area, new_area + size);
895 		memblock_phys_free(start, size);
896 		boot_params.hdr.ramdisk_image = new_area;
897 		boot_params.ext_ramdisk_image = new_area >> 32;
898 	}
899 
900 	/*
901 	 * Set identity map on non-RAM pages and prepare remapping the
902 	 * underlying RAM.
903 	 */
904 	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
905 
906 	pr_info("Released %ld page(s)\n", xen_released_pages);
907 
908 	return "Xen";
909 }
910 
911 static int register_callback(unsigned type, const void *func)
912 {
913 	struct callback_register callback = {
914 		.type = type,
915 		.address = XEN_CALLBACK(__KERNEL_CS, func),
916 		.flags = CALLBACKF_mask_events,
917 	};
918 
919 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
920 }
921 
922 void xen_enable_sysenter(void)
923 {
924 	if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
925 	    register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
926 		setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
927 }
928 
929 void xen_enable_syscall(void)
930 {
931 	int ret;
932 
933 	ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
934 	if (ret != 0) {
935 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
936 		/* Pretty fatal; 64-bit userspace has no other
937 		   mechanism for syscalls. */
938 	}
939 
940 	if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
941 	    register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
942 		setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
943 }
944 
945 static void __init xen_pvmmu_arch_setup(void)
946 {
947 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
948 
949 	if (register_callback(CALLBACKTYPE_event,
950 			      xen_asm_exc_xen_hypervisor_callback) ||
951 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
952 		BUG();
953 
954 	xen_enable_sysenter();
955 	xen_enable_syscall();
956 }
957 
958 /* This function is not called for HVM domains */
959 void __init xen_arch_setup(void)
960 {
961 	xen_panic_handler_init();
962 	xen_pvmmu_arch_setup();
963 
964 #ifdef CONFIG_ACPI
965 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
966 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
967 		disable_acpi();
968 	}
969 #endif
970 
971 	memcpy(boot_command_line, xen_start_info->cmd_line,
972 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
973 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
974 
975 	/* Set up idle, making sure it calls safe_halt() pvop */
976 	disable_cpuidle();
977 	disable_cpufreq();
978 	WARN_ON(xen_set_default_idle());
979 #ifdef CONFIG_NUMA
980 	numa_off = 1;
981 #endif
982 }
983