xref: /linux/arch/x86/xen/setup.c (revision fcc8487d477a3452a1d0ccbdd4c5e0e1e3cb8bed)
1 /*
2  * Machine specific setup for xen
3  *
4  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
5  */
6 
7 #include <linux/init.h>
8 #include <linux/sched.h>
9 #include <linux/mm.h>
10 #include <linux/pm.h>
11 #include <linux/memblock.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpufreq.h>
14 
15 #include <asm/elf.h>
16 #include <asm/vdso.h>
17 #include <asm/e820/api.h>
18 #include <asm/setup.h>
19 #include <asm/acpi.h>
20 #include <asm/numa.h>
21 #include <asm/xen/hypervisor.h>
22 #include <asm/xen/hypercall.h>
23 
24 #include <xen/xen.h>
25 #include <xen/page.h>
26 #include <xen/interface/callback.h>
27 #include <xen/interface/memory.h>
28 #include <xen/interface/physdev.h>
29 #include <xen/features.h>
30 #include <xen/hvc-console.h>
31 #include "xen-ops.h"
32 #include "vdso.h"
33 #include "mmu.h"
34 
35 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
36 
37 /* Amount of extra memory space we add to the e820 ranges */
38 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
39 
40 /* Number of pages released from the initial allocation. */
41 unsigned long xen_released_pages;
42 
43 /* E820 map used during setting up memory. */
44 static struct e820_table xen_e820_table __initdata;
45 
46 /*
47  * Buffer used to remap identity mapped pages. We only need the virtual space.
48  * The physical page behind this address is remapped as needed to different
49  * buffer pages.
50  */
51 #define REMAP_SIZE	(P2M_PER_PAGE - 3)
52 static struct {
53 	unsigned long	next_area_mfn;
54 	unsigned long	target_pfn;
55 	unsigned long	size;
56 	unsigned long	mfns[REMAP_SIZE];
57 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
58 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
59 
60 /*
61  * The maximum amount of extra memory compared to the base size.  The
62  * main scaling factor is the size of struct page.  At extreme ratios
63  * of base:extra, all the base memory can be filled with page
64  * structures for the extra memory, leaving no space for anything
65  * else.
66  *
67  * 10x seems like a reasonable balance between scaling flexibility and
68  * leaving a practically usable system.
69  */
70 #define EXTRA_MEM_RATIO		(10)
71 
72 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
73 
74 static void __init xen_parse_512gb(void)
75 {
76 	bool val = false;
77 	char *arg;
78 
79 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
80 	if (!arg)
81 		return;
82 
83 	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
84 	if (!arg)
85 		val = true;
86 	else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
87 		return;
88 
89 	xen_512gb_limit = val;
90 }
91 
92 static void __init xen_add_extra_mem(unsigned long start_pfn,
93 				     unsigned long n_pfns)
94 {
95 	int i;
96 
97 	/*
98 	 * No need to check for zero size, should happen rarely and will only
99 	 * write a new entry regarded to be unused due to zero size.
100 	 */
101 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
102 		/* Add new region. */
103 		if (xen_extra_mem[i].n_pfns == 0) {
104 			xen_extra_mem[i].start_pfn = start_pfn;
105 			xen_extra_mem[i].n_pfns = n_pfns;
106 			break;
107 		}
108 		/* Append to existing region. */
109 		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
110 		    start_pfn) {
111 			xen_extra_mem[i].n_pfns += n_pfns;
112 			break;
113 		}
114 	}
115 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
116 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
117 
118 	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
119 }
120 
121 static void __init xen_del_extra_mem(unsigned long start_pfn,
122 				     unsigned long n_pfns)
123 {
124 	int i;
125 	unsigned long start_r, size_r;
126 
127 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
128 		start_r = xen_extra_mem[i].start_pfn;
129 		size_r = xen_extra_mem[i].n_pfns;
130 
131 		/* Start of region. */
132 		if (start_r == start_pfn) {
133 			BUG_ON(n_pfns > size_r);
134 			xen_extra_mem[i].start_pfn += n_pfns;
135 			xen_extra_mem[i].n_pfns -= n_pfns;
136 			break;
137 		}
138 		/* End of region. */
139 		if (start_r + size_r == start_pfn + n_pfns) {
140 			BUG_ON(n_pfns > size_r);
141 			xen_extra_mem[i].n_pfns -= n_pfns;
142 			break;
143 		}
144 		/* Mid of region. */
145 		if (start_pfn > start_r && start_pfn < start_r + size_r) {
146 			BUG_ON(start_pfn + n_pfns > start_r + size_r);
147 			xen_extra_mem[i].n_pfns = start_pfn - start_r;
148 			/* Calling memblock_reserve() again is okay. */
149 			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
150 					  (start_pfn + n_pfns));
151 			break;
152 		}
153 	}
154 	memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
155 }
156 
157 /*
158  * Called during boot before the p2m list can take entries beyond the
159  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
160  * invalid.
161  */
162 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
163 {
164 	int i;
165 
166 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
167 		if (pfn >= xen_extra_mem[i].start_pfn &&
168 		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
169 			return INVALID_P2M_ENTRY;
170 	}
171 
172 	return IDENTITY_FRAME(pfn);
173 }
174 
175 /*
176  * Mark all pfns of extra mem as invalid in p2m list.
177  */
178 void __init xen_inv_extra_mem(void)
179 {
180 	unsigned long pfn, pfn_s, pfn_e;
181 	int i;
182 
183 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
184 		if (!xen_extra_mem[i].n_pfns)
185 			continue;
186 		pfn_s = xen_extra_mem[i].start_pfn;
187 		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
188 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
189 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
190 	}
191 }
192 
193 /*
194  * Finds the next RAM pfn available in the E820 map after min_pfn.
195  * This function updates min_pfn with the pfn found and returns
196  * the size of that range or zero if not found.
197  */
198 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
199 {
200 	const struct e820_entry *entry = xen_e820_table.entries;
201 	unsigned int i;
202 	unsigned long done = 0;
203 
204 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
205 		unsigned long s_pfn;
206 		unsigned long e_pfn;
207 
208 		if (entry->type != E820_TYPE_RAM)
209 			continue;
210 
211 		e_pfn = PFN_DOWN(entry->addr + entry->size);
212 
213 		/* We only care about E820 after this */
214 		if (e_pfn <= *min_pfn)
215 			continue;
216 
217 		s_pfn = PFN_UP(entry->addr);
218 
219 		/* If min_pfn falls within the E820 entry, we want to start
220 		 * at the min_pfn PFN.
221 		 */
222 		if (s_pfn <= *min_pfn) {
223 			done = e_pfn - *min_pfn;
224 		} else {
225 			done = e_pfn - s_pfn;
226 			*min_pfn = s_pfn;
227 		}
228 		break;
229 	}
230 
231 	return done;
232 }
233 
234 static int __init xen_free_mfn(unsigned long mfn)
235 {
236 	struct xen_memory_reservation reservation = {
237 		.address_bits = 0,
238 		.extent_order = 0,
239 		.domid        = DOMID_SELF
240 	};
241 
242 	set_xen_guest_handle(reservation.extent_start, &mfn);
243 	reservation.nr_extents = 1;
244 
245 	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
246 }
247 
248 /*
249  * This releases a chunk of memory and then does the identity map. It's used
250  * as a fallback if the remapping fails.
251  */
252 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
253 			unsigned long end_pfn, unsigned long nr_pages)
254 {
255 	unsigned long pfn, end;
256 	int ret;
257 
258 	WARN_ON(start_pfn > end_pfn);
259 
260 	/* Release pages first. */
261 	end = min(end_pfn, nr_pages);
262 	for (pfn = start_pfn; pfn < end; pfn++) {
263 		unsigned long mfn = pfn_to_mfn(pfn);
264 
265 		/* Make sure pfn exists to start with */
266 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
267 			continue;
268 
269 		ret = xen_free_mfn(mfn);
270 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
271 
272 		if (ret == 1) {
273 			xen_released_pages++;
274 			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
275 				break;
276 		} else
277 			break;
278 	}
279 
280 	set_phys_range_identity(start_pfn, end_pfn);
281 }
282 
283 /*
284  * Helper function to update the p2m and m2p tables and kernel mapping.
285  */
286 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
287 {
288 	struct mmu_update update = {
289 		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
290 		.val = pfn
291 	};
292 
293 	/* Update p2m */
294 	if (!set_phys_to_machine(pfn, mfn)) {
295 		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
296 		     pfn, mfn);
297 		BUG();
298 	}
299 
300 	/* Update m2p */
301 	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
302 		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
303 		     mfn, pfn);
304 		BUG();
305 	}
306 
307 	/* Update kernel mapping, but not for highmem. */
308 	if (pfn >= PFN_UP(__pa(high_memory - 1)))
309 		return;
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 	BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
344 
345 	mfn_save = virt_to_mfn(buf);
346 
347 	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
348 	     ident_pfn_iter < ident_end_pfn;
349 	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
350 		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
351 
352 		/* Map first pfn to xen_remap_buf */
353 		mfn = pfn_to_mfn(ident_pfn_iter);
354 		set_pte_mfn(buf, mfn, PAGE_KERNEL);
355 
356 		/* Save mapping information in page */
357 		xen_remap_buf.next_area_mfn = xen_remap_mfn;
358 		xen_remap_buf.target_pfn = remap_pfn_iter;
359 		xen_remap_buf.size = chunk;
360 		for (i = 0; i < chunk; i++)
361 			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
362 
363 		/* Put remap buf into list. */
364 		xen_remap_mfn = mfn;
365 
366 		/* Set identity map */
367 		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
368 
369 		left -= chunk;
370 	}
371 
372 	/* Restore old xen_remap_buf mapping */
373 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
374 }
375 
376 /*
377  * This function takes a contiguous pfn range that needs to be identity mapped
378  * and:
379  *
380  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
381  *  2) Calls the do_ function to actually do the mapping/remapping work.
382  *
383  * The goal is to not allocate additional memory but to remap the existing
384  * pages. In the case of an error the underlying memory is simply released back
385  * to Xen and not remapped.
386  */
387 static unsigned long __init xen_set_identity_and_remap_chunk(
388 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
389 	unsigned long remap_pfn)
390 {
391 	unsigned long pfn;
392 	unsigned long i = 0;
393 	unsigned long n = end_pfn - start_pfn;
394 
395 	if (remap_pfn == 0)
396 		remap_pfn = nr_pages;
397 
398 	while (i < n) {
399 		unsigned long cur_pfn = start_pfn + i;
400 		unsigned long left = n - i;
401 		unsigned long size = left;
402 		unsigned long remap_range_size;
403 
404 		/* Do not remap pages beyond the current allocation */
405 		if (cur_pfn >= nr_pages) {
406 			/* Identity map remaining pages */
407 			set_phys_range_identity(cur_pfn, cur_pfn + size);
408 			break;
409 		}
410 		if (cur_pfn + size > nr_pages)
411 			size = nr_pages - cur_pfn;
412 
413 		remap_range_size = xen_find_pfn_range(&remap_pfn);
414 		if (!remap_range_size) {
415 			pr_warning("Unable to find available pfn range, not remapping identity pages\n");
416 			xen_set_identity_and_release_chunk(cur_pfn,
417 						cur_pfn + left, nr_pages);
418 			break;
419 		}
420 		/* Adjust size to fit in current e820 RAM region */
421 		if (size > remap_range_size)
422 			size = remap_range_size;
423 
424 		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
425 
426 		/* Update variables to reflect new mappings. */
427 		i += size;
428 		remap_pfn += size;
429 	}
430 
431 	/*
432 	 * If the PFNs are currently mapped, the VA mapping also needs
433 	 * to be updated to be 1:1.
434 	 */
435 	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
436 		(void)HYPERVISOR_update_va_mapping(
437 			(unsigned long)__va(pfn << PAGE_SHIFT),
438 			mfn_pte(pfn, PAGE_KERNEL_IO), 0);
439 
440 	return remap_pfn;
441 }
442 
443 static unsigned long __init xen_count_remap_pages(
444 	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
445 	unsigned long remap_pages)
446 {
447 	if (start_pfn >= nr_pages)
448 		return remap_pages;
449 
450 	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
451 }
452 
453 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
454 	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
455 			      unsigned long nr_pages, unsigned long last_val))
456 {
457 	phys_addr_t start = 0;
458 	unsigned long ret_val = 0;
459 	const struct e820_entry *entry = xen_e820_table.entries;
460 	int i;
461 
462 	/*
463 	 * Combine non-RAM regions and gaps until a RAM region (or the
464 	 * end of the map) is reached, then call the provided function
465 	 * to perform its duty on the non-RAM region.
466 	 *
467 	 * The combined non-RAM regions are rounded to a whole number
468 	 * of pages so any partial pages are accessible via the 1:1
469 	 * mapping.  This is needed for some BIOSes that put (for
470 	 * example) the DMI tables in a reserved region that begins on
471 	 * a non-page boundary.
472 	 */
473 	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
474 		phys_addr_t end = entry->addr + entry->size;
475 		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
476 			unsigned long start_pfn = PFN_DOWN(start);
477 			unsigned long end_pfn = PFN_UP(end);
478 
479 			if (entry->type == E820_TYPE_RAM)
480 				end_pfn = PFN_UP(entry->addr);
481 
482 			if (start_pfn < end_pfn)
483 				ret_val = func(start_pfn, end_pfn, nr_pages,
484 					       ret_val);
485 			start = end;
486 		}
487 	}
488 
489 	return ret_val;
490 }
491 
492 /*
493  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
494  * The remap information (which mfn remap to which pfn) is contained in the
495  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
496  * This scheme allows to remap the different chunks in arbitrary order while
497  * the resulting mapping will be independant from the order.
498  */
499 void __init xen_remap_memory(void)
500 {
501 	unsigned long buf = (unsigned long)&xen_remap_buf;
502 	unsigned long mfn_save, mfn, pfn;
503 	unsigned long remapped = 0;
504 	unsigned int i;
505 	unsigned long pfn_s = ~0UL;
506 	unsigned long len = 0;
507 
508 	mfn_save = virt_to_mfn(buf);
509 
510 	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
511 		/* Map the remap information */
512 		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
513 
514 		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
515 
516 		pfn = xen_remap_buf.target_pfn;
517 		for (i = 0; i < xen_remap_buf.size; i++) {
518 			mfn = xen_remap_buf.mfns[i];
519 			xen_update_mem_tables(pfn, mfn);
520 			remapped++;
521 			pfn++;
522 		}
523 		if (pfn_s == ~0UL || pfn == pfn_s) {
524 			pfn_s = xen_remap_buf.target_pfn;
525 			len += xen_remap_buf.size;
526 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
527 			len += xen_remap_buf.size;
528 		} else {
529 			xen_del_extra_mem(pfn_s, len);
530 			pfn_s = xen_remap_buf.target_pfn;
531 			len = xen_remap_buf.size;
532 		}
533 
534 		mfn = xen_remap_mfn;
535 		xen_remap_mfn = xen_remap_buf.next_area_mfn;
536 	}
537 
538 	if (pfn_s != ~0UL && len)
539 		xen_del_extra_mem(pfn_s, len);
540 
541 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
542 
543 	pr_info("Remapped %ld page(s)\n", remapped);
544 }
545 
546 static unsigned long __init xen_get_pages_limit(void)
547 {
548 	unsigned long limit;
549 
550 #ifdef CONFIG_X86_32
551 	limit = GB(64) / PAGE_SIZE;
552 #else
553 	limit = MAXMEM / PAGE_SIZE;
554 	if (!xen_initial_domain() && xen_512gb_limit)
555 		limit = GB(512) / PAGE_SIZE;
556 #endif
557 	return limit;
558 }
559 
560 static unsigned long __init xen_get_max_pages(void)
561 {
562 	unsigned long max_pages, limit;
563 	domid_t domid = DOMID_SELF;
564 	long ret;
565 
566 	limit = xen_get_pages_limit();
567 	max_pages = limit;
568 
569 	/*
570 	 * For the initial domain we use the maximum reservation as
571 	 * the maximum page.
572 	 *
573 	 * For guest domains the current maximum reservation reflects
574 	 * the current maximum rather than the static maximum. In this
575 	 * case the e820 map provided to us will cover the static
576 	 * maximum region.
577 	 */
578 	if (xen_initial_domain()) {
579 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
580 		if (ret > 0)
581 			max_pages = ret;
582 	}
583 
584 	return min(max_pages, limit);
585 }
586 
587 static void __init xen_align_and_add_e820_region(phys_addr_t start,
588 						 phys_addr_t size, int type)
589 {
590 	phys_addr_t end = start + size;
591 
592 	/* Align RAM regions to page boundaries. */
593 	if (type == E820_TYPE_RAM) {
594 		start = PAGE_ALIGN(start);
595 		end &= ~((phys_addr_t)PAGE_SIZE - 1);
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 #ifdef CONFIG_X86_32
720 	/*
721 	 * Relocating the p2m on 32 bit system to an arbitrary virtual address
722 	 * is not supported, so just give up.
723 	 */
724 	xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
725 	BUG();
726 #else
727 	xen_relocate_p2m();
728 	memblock_free(start, size);
729 #endif
730 }
731 
732 /**
733  * machine_specific_memory_setup - Hook for machine specific memory setup.
734  **/
735 char * __init xen_memory_setup(void)
736 {
737 	unsigned long max_pfn, pfn_s, n_pfns;
738 	phys_addr_t mem_end, addr, size, chunk_size;
739 	u32 type;
740 	int rc;
741 	struct xen_memory_map memmap;
742 	unsigned long max_pages;
743 	unsigned long extra_pages = 0;
744 	int i;
745 	int op;
746 
747 	xen_parse_512gb();
748 	max_pfn = xen_get_pages_limit();
749 	max_pfn = min(max_pfn, xen_start_info->nr_pages);
750 	mem_end = PFN_PHYS(max_pfn);
751 
752 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
753 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
754 
755 	op = xen_initial_domain() ?
756 		XENMEM_machine_memory_map :
757 		XENMEM_memory_map;
758 	rc = HYPERVISOR_memory_op(op, &memmap);
759 	if (rc == -ENOSYS) {
760 		BUG_ON(xen_initial_domain());
761 		memmap.nr_entries = 1;
762 		xen_e820_table.entries[0].addr = 0ULL;
763 		xen_e820_table.entries[0].size = mem_end;
764 		/* 8MB slack (to balance backend allocations). */
765 		xen_e820_table.entries[0].size += 8ULL << 20;
766 		xen_e820_table.entries[0].type = E820_TYPE_RAM;
767 		rc = 0;
768 	}
769 	BUG_ON(rc);
770 	BUG_ON(memmap.nr_entries == 0);
771 	xen_e820_table.nr_entries = memmap.nr_entries;
772 
773 	/*
774 	 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
775 	 * regions, so if we're using the machine memory map leave the
776 	 * region as RAM as it is in the pseudo-physical map.
777 	 *
778 	 * UNUSABLE regions in domUs are not handled and will need
779 	 * a patch in the future.
780 	 */
781 	if (xen_initial_domain())
782 		xen_ignore_unusable();
783 
784 	/* Make sure the Xen-supplied memory map is well-ordered. */
785 	e820__update_table(&xen_e820_table);
786 
787 	max_pages = xen_get_max_pages();
788 
789 	/* How many extra pages do we need due to remapping? */
790 	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
791 
792 	if (max_pages > max_pfn)
793 		extra_pages += max_pages - max_pfn;
794 
795 	/*
796 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
797 	 * factor the base size.  On non-highmem systems, the base
798 	 * size is the full initial memory allocation; on highmem it
799 	 * is limited to the max size of lowmem, so that it doesn't
800 	 * get completely filled.
801 	 *
802 	 * Make sure we have no memory above max_pages, as this area
803 	 * isn't handled by the p2m management.
804 	 *
805 	 * In principle there could be a problem in lowmem systems if
806 	 * the initial memory is also very large with respect to
807 	 * lowmem, but we won't try to deal with that here.
808 	 */
809 	extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
810 			   extra_pages, max_pages - max_pfn);
811 	i = 0;
812 	addr = xen_e820_table.entries[0].addr;
813 	size = xen_e820_table.entries[0].size;
814 	while (i < xen_e820_table.nr_entries) {
815 		bool discard = false;
816 
817 		chunk_size = size;
818 		type = xen_e820_table.entries[i].type;
819 
820 		if (type == E820_TYPE_RAM) {
821 			if (addr < mem_end) {
822 				chunk_size = min(size, mem_end - addr);
823 			} else if (extra_pages) {
824 				chunk_size = min(size, PFN_PHYS(extra_pages));
825 				pfn_s = PFN_UP(addr);
826 				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
827 				extra_pages -= n_pfns;
828 				xen_add_extra_mem(pfn_s, n_pfns);
829 				xen_max_p2m_pfn = pfn_s + n_pfns;
830 			} else
831 				discard = true;
832 		}
833 
834 		if (!discard)
835 			xen_align_and_add_e820_region(addr, chunk_size, type);
836 
837 		addr += chunk_size;
838 		size -= chunk_size;
839 		if (size == 0) {
840 			i++;
841 			if (i < xen_e820_table.nr_entries) {
842 				addr = xen_e820_table.entries[i].addr;
843 				size = xen_e820_table.entries[i].size;
844 			}
845 		}
846 	}
847 
848 	/*
849 	 * Set the rest as identity mapped, in case PCI BARs are
850 	 * located here.
851 	 */
852 	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
853 
854 	/*
855 	 * In domU, the ISA region is normal, usable memory, but we
856 	 * reserve ISA memory anyway because too many things poke
857 	 * about in there.
858 	 */
859 	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
860 
861 	e820__update_table(e820_table);
862 
863 	/*
864 	 * Check whether the kernel itself conflicts with the target E820 map.
865 	 * Failing now is better than running into weird problems later due
866 	 * to relocating (and even reusing) pages with kernel text or data.
867 	 */
868 	if (xen_is_e820_reserved(__pa_symbol(_text),
869 			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
870 		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
871 		BUG();
872 	}
873 
874 	/*
875 	 * Check for a conflict of the hypervisor supplied page tables with
876 	 * the target E820 map.
877 	 */
878 	xen_pt_check_e820();
879 
880 	xen_reserve_xen_mfnlist();
881 
882 	/* Check for a conflict of the initrd with the target E820 map. */
883 	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
884 				 boot_params.hdr.ramdisk_size)) {
885 		phys_addr_t new_area, start, size;
886 
887 		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
888 		if (!new_area) {
889 			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
890 			BUG();
891 		}
892 
893 		start = boot_params.hdr.ramdisk_image;
894 		size = boot_params.hdr.ramdisk_size;
895 		xen_phys_memcpy(new_area, start, size);
896 		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
897 			start, start + size, new_area, new_area + size);
898 		memblock_free(start, size);
899 		boot_params.hdr.ramdisk_image = new_area;
900 		boot_params.ext_ramdisk_image = new_area >> 32;
901 	}
902 
903 	/*
904 	 * Set identity map on non-RAM pages and prepare remapping the
905 	 * underlying RAM.
906 	 */
907 	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
908 
909 	pr_info("Released %ld page(s)\n", xen_released_pages);
910 
911 	return "Xen";
912 }
913 
914 /*
915  * Machine specific memory setup for auto-translated guests.
916  */
917 char * __init xen_auto_xlated_memory_setup(void)
918 {
919 	struct xen_memory_map memmap;
920 	int i;
921 	int rc;
922 
923 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
924 	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
925 
926 	rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
927 	if (rc < 0)
928 		panic("No memory map (%d)\n", rc);
929 
930 	xen_e820_table.nr_entries = memmap.nr_entries;
931 
932 	e820__update_table(&xen_e820_table);
933 
934 	for (i = 0; i < xen_e820_table.nr_entries; i++)
935 		e820__range_add(xen_e820_table.entries[i].addr, xen_e820_table.entries[i].size, xen_e820_table.entries[i].type);
936 
937 	/* Remove p2m info, it is not needed. */
938 	xen_start_info->mfn_list = 0;
939 	xen_start_info->first_p2m_pfn = 0;
940 	xen_start_info->nr_p2m_frames = 0;
941 
942 	return "Xen";
943 }
944 
945 /*
946  * Set the bit indicating "nosegneg" library variants should be used.
947  * We only need to bother in pure 32-bit mode; compat 32-bit processes
948  * can have un-truncated segments, so wrapping around is allowed.
949  */
950 static void __init fiddle_vdso(void)
951 {
952 #ifdef CONFIG_X86_32
953 	u32 *mask = vdso_image_32.data +
954 		vdso_image_32.sym_VDSO32_NOTE_MASK;
955 	*mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
956 #endif
957 }
958 
959 static int register_callback(unsigned type, const void *func)
960 {
961 	struct callback_register callback = {
962 		.type = type,
963 		.address = XEN_CALLBACK(__KERNEL_CS, func),
964 		.flags = CALLBACKF_mask_events,
965 	};
966 
967 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
968 }
969 
970 void xen_enable_sysenter(void)
971 {
972 	int ret;
973 	unsigned sysenter_feature;
974 
975 #ifdef CONFIG_X86_32
976 	sysenter_feature = X86_FEATURE_SEP;
977 #else
978 	sysenter_feature = X86_FEATURE_SYSENTER32;
979 #endif
980 
981 	if (!boot_cpu_has(sysenter_feature))
982 		return;
983 
984 	ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
985 	if(ret != 0)
986 		setup_clear_cpu_cap(sysenter_feature);
987 }
988 
989 void xen_enable_syscall(void)
990 {
991 #ifdef CONFIG_X86_64
992 	int ret;
993 
994 	ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
995 	if (ret != 0) {
996 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
997 		/* Pretty fatal; 64-bit userspace has no other
998 		   mechanism for syscalls. */
999 	}
1000 
1001 	if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
1002 		ret = register_callback(CALLBACKTYPE_syscall32,
1003 					xen_syscall32_target);
1004 		if (ret != 0)
1005 			setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1006 	}
1007 #endif /* CONFIG_X86_64 */
1008 }
1009 
1010 void __init xen_pvmmu_arch_setup(void)
1011 {
1012 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
1013 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1014 
1015 	HYPERVISOR_vm_assist(VMASST_CMD_enable,
1016 			     VMASST_TYPE_pae_extended_cr3);
1017 
1018 	if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
1019 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1020 		BUG();
1021 
1022 	xen_enable_sysenter();
1023 	xen_enable_syscall();
1024 }
1025 
1026 /* This function is not called for HVM domains */
1027 void __init xen_arch_setup(void)
1028 {
1029 	xen_panic_handler_init();
1030 	if (!xen_feature(XENFEAT_auto_translated_physmap))
1031 		xen_pvmmu_arch_setup();
1032 
1033 #ifdef CONFIG_ACPI
1034 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1035 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1036 		disable_acpi();
1037 	}
1038 #endif
1039 
1040 	memcpy(boot_command_line, xen_start_info->cmd_line,
1041 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1042 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1043 
1044 	/* Set up idle, making sure it calls safe_halt() pvop */
1045 	disable_cpuidle();
1046 	disable_cpufreq();
1047 	WARN_ON(xen_set_default_idle());
1048 	fiddle_vdso();
1049 #ifdef CONFIG_NUMA
1050 	numa_off = 1;
1051 #endif
1052 }
1053