Linux Audio

Check our new training course

Loading...
v6.8
  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 */
 42struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
 43
 44/* Number of pages released from the initial allocation. */
 45unsigned long xen_released_pages;
 46
 47/* Memory map would allow PCI passthrough. */
 48bool xen_pv_pci_possible;
 49
 50/* E820 map used during setting up memory. */
 51static struct e820_table xen_e820_table __initdata;
 52
 53/*
 54 * Buffer used to remap identity mapped pages. We only need the virtual space.
 55 * The physical page behind this address is remapped as needed to different
 56 * buffer pages.
 57 */
 58#define REMAP_SIZE	(P2M_PER_PAGE - 3)
 59static struct {
 60	unsigned long	next_area_mfn;
 61	unsigned long	target_pfn;
 62	unsigned long	size;
 63	unsigned long	mfns[REMAP_SIZE];
 64} xen_remap_buf __initdata __aligned(PAGE_SIZE);
 65static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
 66
 67/*
 68 * The maximum amount of extra memory compared to the base size.  The
 69 * main scaling factor is the size of struct page.  At extreme ratios
 70 * of base:extra, all the base memory can be filled with page
 71 * structures for the extra memory, leaving no space for anything
 72 * else.
 73 *
 74 * 10x seems like a reasonable balance between scaling flexibility and
 75 * leaving a practically usable system.
 76 */
 77#define EXTRA_MEM_RATIO		(10)
 78
 79static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
 80
 81static void __init xen_parse_512gb(void)
 82{
 83	bool val = false;
 84	char *arg;
 85
 86	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
 87	if (!arg)
 88		return;
 89
 90	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
 91	if (!arg)
 92		val = true;
 93	else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
 94		return;
 95
 96	xen_512gb_limit = val;
 97}
 98
 99static void __init xen_add_extra_mem(unsigned long start_pfn,
100				     unsigned long n_pfns)
101{
102	int i;
103
104	/*
105	 * No need to check for zero size, should happen rarely and will only
106	 * write a new entry regarded to be unused due to zero size.
107	 */
108	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
109		/* Add new region. */
110		if (xen_extra_mem[i].n_pfns == 0) {
111			xen_extra_mem[i].start_pfn = start_pfn;
112			xen_extra_mem[i].n_pfns = n_pfns;
113			break;
114		}
115		/* Append to existing region. */
116		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
117		    start_pfn) {
118			xen_extra_mem[i].n_pfns += n_pfns;
119			break;
120		}
121	}
122	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
123		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
124
125	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
126}
127
128static void __init xen_del_extra_mem(unsigned long start_pfn,
129				     unsigned long n_pfns)
130{
131	int i;
132	unsigned long start_r, size_r;
133
134	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
135		start_r = xen_extra_mem[i].start_pfn;
136		size_r = xen_extra_mem[i].n_pfns;
137
138		/* Start of region. */
139		if (start_r == start_pfn) {
140			BUG_ON(n_pfns > size_r);
141			xen_extra_mem[i].start_pfn += n_pfns;
142			xen_extra_mem[i].n_pfns -= n_pfns;
143			break;
144		}
145		/* End of region. */
146		if (start_r + size_r == start_pfn + n_pfns) {
147			BUG_ON(n_pfns > size_r);
148			xen_extra_mem[i].n_pfns -= n_pfns;
149			break;
150		}
151		/* Mid of region. */
152		if (start_pfn > start_r && start_pfn < start_r + size_r) {
153			BUG_ON(start_pfn + n_pfns > start_r + size_r);
154			xen_extra_mem[i].n_pfns = start_pfn - start_r;
155			/* Calling memblock_reserve() again is okay. */
156			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
157					  (start_pfn + n_pfns));
158			break;
159		}
160	}
161	memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
162}
163
164/*
165 * Called during boot before the p2m list can take entries beyond the
166 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
167 * invalid.
168 */
169unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
170{
171	int i;
172
173	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
174		if (pfn >= xen_extra_mem[i].start_pfn &&
175		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
176			return INVALID_P2M_ENTRY;
177	}
178
179	return IDENTITY_FRAME(pfn);
180}
181
182/*
183 * Mark all pfns of extra mem as invalid in p2m list.
184 */
185void __init xen_inv_extra_mem(void)
186{
187	unsigned long pfn, pfn_s, pfn_e;
188	int i;
189
190	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
191		if (!xen_extra_mem[i].n_pfns)
192			continue;
193		pfn_s = xen_extra_mem[i].start_pfn;
194		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
195		for (pfn = pfn_s; pfn < pfn_e; pfn++)
196			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
197	}
198}
199
200/*
201 * Finds the next RAM pfn available in the E820 map after min_pfn.
202 * This function updates min_pfn with the pfn found and returns
203 * the size of that range or zero if not found.
204 */
205static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
206{
207	const struct e820_entry *entry = xen_e820_table.entries;
208	unsigned int i;
209	unsigned long done = 0;
210
211	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
212		unsigned long s_pfn;
213		unsigned long e_pfn;
214
215		if (entry->type != E820_TYPE_RAM)
216			continue;
217
218		e_pfn = PFN_DOWN(entry->addr + entry->size);
219
220		/* We only care about E820 after this */
221		if (e_pfn <= *min_pfn)
222			continue;
223
224		s_pfn = PFN_UP(entry->addr);
225
226		/* If min_pfn falls within the E820 entry, we want to start
227		 * at the min_pfn PFN.
228		 */
229		if (s_pfn <= *min_pfn) {
230			done = e_pfn - *min_pfn;
231		} else {
232			done = e_pfn - s_pfn;
233			*min_pfn = s_pfn;
234		}
235		break;
236	}
237
238	return done;
239}
240
241static int __init xen_free_mfn(unsigned long mfn)
242{
243	struct xen_memory_reservation reservation = {
244		.address_bits = 0,
245		.extent_order = 0,
246		.domid        = DOMID_SELF
247	};
248
249	set_xen_guest_handle(reservation.extent_start, &mfn);
250	reservation.nr_extents = 1;
251
252	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
253}
254
255/*
256 * This releases a chunk of memory and then does the identity map. It's used
257 * as a fallback if the remapping fails.
258 */
259static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
260			unsigned long end_pfn, unsigned long nr_pages)
261{
262	unsigned long pfn, end;
263	int ret;
264
265	WARN_ON(start_pfn > end_pfn);
266
267	/* Release pages first. */
268	end = min(end_pfn, nr_pages);
269	for (pfn = start_pfn; pfn < end; pfn++) {
270		unsigned long mfn = pfn_to_mfn(pfn);
271
272		/* Make sure pfn exists to start with */
273		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
274			continue;
275
276		ret = xen_free_mfn(mfn);
277		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
278
279		if (ret == 1) {
280			xen_released_pages++;
281			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
282				break;
283		} else
284			break;
285	}
286
287	set_phys_range_identity(start_pfn, end_pfn);
288}
289
290/*
291 * Helper function to update the p2m and m2p tables and kernel mapping.
292 */
293static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
294{
295	struct mmu_update update = {
296		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
297		.val = pfn
298	};
299
300	/* Update p2m */
301	if (!set_phys_to_machine(pfn, mfn)) {
302		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
303		     pfn, mfn);
304		BUG();
305	}
306
307	/* Update m2p */
308	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
309		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
310		     mfn, pfn);
311		BUG();
312	}
313
 
 
 
 
314	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
315					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
316		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
317		      mfn, pfn);
318		BUG();
319	}
320}
321
322/*
323 * This function updates the p2m and m2p tables with an identity map from
324 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
325 * original allocation at remap_pfn. The information needed for remapping is
326 * saved in the memory itself to avoid the need for allocating buffers. The
327 * complete remap information is contained in a list of MFNs each containing
328 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
329 * This enables us to preserve the original mfn sequence while doing the
330 * remapping at a time when the memory management is capable of allocating
331 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
332 * its callers.
333 */
334static void __init xen_do_set_identity_and_remap_chunk(
335        unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
336{
337	unsigned long buf = (unsigned long)&xen_remap_buf;
338	unsigned long mfn_save, mfn;
339	unsigned long ident_pfn_iter, remap_pfn_iter;
340	unsigned long ident_end_pfn = start_pfn + size;
341	unsigned long left = size;
342	unsigned int i, chunk;
343
344	WARN_ON(size == 0);
345
346	mfn_save = virt_to_mfn((void *)buf);
347
348	for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
349	     ident_pfn_iter < ident_end_pfn;
350	     ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
351		chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
352
353		/* Map first pfn to xen_remap_buf */
354		mfn = pfn_to_mfn(ident_pfn_iter);
355		set_pte_mfn(buf, mfn, PAGE_KERNEL);
356
357		/* Save mapping information in page */
358		xen_remap_buf.next_area_mfn = xen_remap_mfn;
359		xen_remap_buf.target_pfn = remap_pfn_iter;
360		xen_remap_buf.size = chunk;
361		for (i = 0; i < chunk; i++)
362			xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
363
364		/* Put remap buf into list. */
365		xen_remap_mfn = mfn;
366
367		/* Set identity map */
368		set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
369
370		left -= chunk;
371	}
372
373	/* Restore old xen_remap_buf mapping */
374	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
375}
376
377/*
378 * This function takes a contiguous pfn range that needs to be identity mapped
379 * and:
380 *
381 *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
382 *  2) Calls the do_ function to actually do the mapping/remapping work.
383 *
384 * The goal is to not allocate additional memory but to remap the existing
385 * pages. In the case of an error the underlying memory is simply released back
386 * to Xen and not remapped.
387 */
388static unsigned long __init xen_set_identity_and_remap_chunk(
389	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
390	unsigned long remap_pfn)
391{
392	unsigned long pfn;
393	unsigned long i = 0;
394	unsigned long n = end_pfn - start_pfn;
395
396	if (remap_pfn == 0)
397		remap_pfn = nr_pages;
398
399	while (i < n) {
400		unsigned long cur_pfn = start_pfn + i;
401		unsigned long left = n - i;
402		unsigned long size = left;
403		unsigned long remap_range_size;
404
405		/* Do not remap pages beyond the current allocation */
406		if (cur_pfn >= nr_pages) {
407			/* Identity map remaining pages */
408			set_phys_range_identity(cur_pfn, cur_pfn + size);
409			break;
410		}
411		if (cur_pfn + size > nr_pages)
412			size = nr_pages - cur_pfn;
413
414		remap_range_size = xen_find_pfn_range(&remap_pfn);
415		if (!remap_range_size) {
416			pr_warn("Unable to find available pfn range, not remapping identity pages\n");
417			xen_set_identity_and_release_chunk(cur_pfn,
418						cur_pfn + left, nr_pages);
419			break;
420		}
421		/* Adjust size to fit in current e820 RAM region */
422		if (size > remap_range_size)
423			size = remap_range_size;
424
425		xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
426
427		/* Update variables to reflect new mappings. */
428		i += size;
429		remap_pfn += size;
430	}
431
432	/*
433	 * If the PFNs are currently mapped, their VA mappings need to be
434	 * zapped.
435	 */
436	for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
437		(void)HYPERVISOR_update_va_mapping(
438			(unsigned long)__va(pfn << PAGE_SHIFT),
439			native_make_pte(0), 0);
440
441	return remap_pfn;
442}
443
444static unsigned long __init xen_count_remap_pages(
445	unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
446	unsigned long remap_pages)
447{
448	if (start_pfn >= nr_pages)
449		return remap_pages;
450
451	return remap_pages + min(end_pfn, nr_pages) - start_pfn;
452}
453
454static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
455	unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
456			      unsigned long nr_pages, unsigned long last_val))
457{
458	phys_addr_t start = 0;
459	unsigned long ret_val = 0;
460	const struct e820_entry *entry = xen_e820_table.entries;
461	int i;
462
463	/*
464	 * Combine non-RAM regions and gaps until a RAM region (or the
465	 * end of the map) is reached, then call the provided function
466	 * to perform its duty on the non-RAM region.
467	 *
468	 * The combined non-RAM regions are rounded to a whole number
469	 * of pages so any partial pages are accessible via the 1:1
470	 * mapping.  This is needed for some BIOSes that put (for
471	 * example) the DMI tables in a reserved region that begins on
472	 * a non-page boundary.
473	 */
474	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
475		phys_addr_t end = entry->addr + entry->size;
476		if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
477			unsigned long start_pfn = PFN_DOWN(start);
478			unsigned long end_pfn = PFN_UP(end);
479
480			if (entry->type == E820_TYPE_RAM)
481				end_pfn = PFN_UP(entry->addr);
482
483			if (start_pfn < end_pfn)
484				ret_val = func(start_pfn, end_pfn, nr_pages,
485					       ret_val);
486			start = end;
487		}
488	}
489
490	return ret_val;
491}
492
493/*
494 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
495 * The remap information (which mfn remap to which pfn) is contained in the
496 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
497 * This scheme allows to remap the different chunks in arbitrary order while
498 * the resulting mapping will be independent from the order.
499 */
500void __init xen_remap_memory(void)
501{
502	unsigned long buf = (unsigned long)&xen_remap_buf;
503	unsigned long mfn_save, pfn;
504	unsigned long remapped = 0;
505	unsigned int i;
506	unsigned long pfn_s = ~0UL;
507	unsigned long len = 0;
508
509	mfn_save = virt_to_mfn((void *)buf);
510
511	while (xen_remap_mfn != INVALID_P2M_ENTRY) {
512		/* Map the remap information */
513		set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
514
515		BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
516
517		pfn = xen_remap_buf.target_pfn;
518		for (i = 0; i < xen_remap_buf.size; i++) {
519			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
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		xen_remap_mfn = xen_remap_buf.next_area_mfn;
534	}
535
536	if (pfn_s != ~0UL && len)
537		xen_del_extra_mem(pfn_s, len);
538
539	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
540
541	pr_info("Remapped %ld page(s)\n", remapped);
542}
543
544static unsigned long __init xen_get_pages_limit(void)
545{
546	unsigned long limit;
547
548	limit = MAXMEM / PAGE_SIZE;
549	if (!xen_initial_domain() && xen_512gb_limit)
550		limit = GB(512) / PAGE_SIZE;
551
552	return limit;
553}
554
555static unsigned long __init xen_get_max_pages(void)
556{
557	unsigned long max_pages, limit;
558	domid_t domid = DOMID_SELF;
559	long ret;
560
561	limit = xen_get_pages_limit();
562	max_pages = limit;
563
564	/*
565	 * For the initial domain we use the maximum reservation as
566	 * the maximum page.
567	 *
568	 * For guest domains the current maximum reservation reflects
569	 * the current maximum rather than the static maximum. In this
570	 * case the e820 map provided to us will cover the static
571	 * maximum region.
572	 */
573	if (xen_initial_domain()) {
574		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
575		if (ret > 0)
576			max_pages = ret;
577	}
578
579	return min(max_pages, limit);
580}
581
582static void __init xen_align_and_add_e820_region(phys_addr_t start,
583						 phys_addr_t size, int type)
584{
585	phys_addr_t end = start + size;
586
587	/* Align RAM regions to page boundaries. */
588	if (type == E820_TYPE_RAM) {
589		start = PAGE_ALIGN(start);
590		end &= ~((phys_addr_t)PAGE_SIZE - 1);
591#ifdef CONFIG_MEMORY_HOTPLUG
592		/*
593		 * Don't allow adding memory not in E820 map while booting the
594		 * system. Once the balloon driver is up it will remove that
595		 * restriction again.
596		 */
597		max_mem_size = end;
598#endif
599	}
600
601	e820__range_add(start, end - start, type);
602}
603
604static void __init xen_ignore_unusable(void)
605{
606	struct e820_entry *entry = xen_e820_table.entries;
607	unsigned int i;
608
609	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
610		if (entry->type == E820_TYPE_UNUSABLE)
611			entry->type = E820_TYPE_RAM;
612	}
613}
614
615bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
616{
617	struct e820_entry *entry;
618	unsigned mapcnt;
619	phys_addr_t end;
620
621	if (!size)
622		return false;
623
624	end = start + size;
625	entry = xen_e820_table.entries;
626
627	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
628		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
629		    (entry->addr + entry->size) >= end)
630			return false;
631
632		entry++;
633	}
634
635	return true;
636}
637
638/*
639 * Find a free area in physical memory not yet reserved and compliant with
640 * E820 map.
641 * Used to relocate pre-allocated areas like initrd or p2m list which are in
642 * conflict with the to be used E820 map.
643 * In case no area is found, return 0. Otherwise return the physical address
644 * of the area which is already reserved for convenience.
645 */
646phys_addr_t __init xen_find_free_area(phys_addr_t size)
647{
648	unsigned mapcnt;
649	phys_addr_t addr, start;
650	struct e820_entry *entry = xen_e820_table.entries;
651
652	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
653		if (entry->type != E820_TYPE_RAM || entry->size < size)
654			continue;
655		start = entry->addr;
656		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
657			if (!memblock_is_reserved(addr))
658				continue;
659			start = addr + PAGE_SIZE;
660			if (start + size > entry->addr + entry->size)
661				break;
662		}
663		if (addr >= start + size) {
664			memblock_reserve(start, size);
665			return start;
666		}
667	}
668
669	return 0;
670}
671
672/*
673 * Like memcpy, but with physical addresses for dest and src.
674 */
675static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
676				   phys_addr_t n)
677{
678	phys_addr_t dest_off, src_off, dest_len, src_len, len;
679	void *from, *to;
680
681	while (n) {
682		dest_off = dest & ~PAGE_MASK;
683		src_off = src & ~PAGE_MASK;
684		dest_len = n;
685		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
686			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
687		src_len = n;
688		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
689			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
690		len = min(dest_len, src_len);
691		to = early_memremap(dest - dest_off, dest_len + dest_off);
692		from = early_memremap(src - src_off, src_len + src_off);
693		memcpy(to, from, len);
694		early_memunmap(to, dest_len + dest_off);
695		early_memunmap(from, src_len + src_off);
696		n -= len;
697		dest += len;
698		src += len;
699	}
700}
701
702/*
703 * Reserve Xen mfn_list.
704 */
705static void __init xen_reserve_xen_mfnlist(void)
706{
707	phys_addr_t start, size;
708
709	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
710		start = __pa(xen_start_info->mfn_list);
711		size = PFN_ALIGN(xen_start_info->nr_pages *
712				 sizeof(unsigned long));
713	} else {
714		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
715		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
716	}
717
718	memblock_reserve(start, size);
719	if (!xen_is_e820_reserved(start, size))
720		return;
721
722	xen_relocate_p2m();
723	memblock_phys_free(start, size);
724}
725
726/**
727 * xen_memory_setup - Hook for machine specific memory setup.
728 **/
729char * __init xen_memory_setup(void)
730{
731	unsigned long max_pfn, pfn_s, n_pfns;
732	phys_addr_t mem_end, addr, size, chunk_size;
733	u32 type;
734	int rc;
735	struct xen_memory_map memmap;
736	unsigned long max_pages;
737	unsigned long extra_pages = 0;
738	int i;
739	int op;
740
741	xen_parse_512gb();
742	max_pfn = xen_get_pages_limit();
743	max_pfn = min(max_pfn, xen_start_info->nr_pages);
744	mem_end = PFN_PHYS(max_pfn);
745
746	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
747	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
748
749#if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
750	xen_saved_max_mem_size = max_mem_size;
751#endif
752
753	op = xen_initial_domain() ?
754		XENMEM_machine_memory_map :
755		XENMEM_memory_map;
756	rc = HYPERVISOR_memory_op(op, &memmap);
757	if (rc == -ENOSYS) {
758		BUG_ON(xen_initial_domain());
759		memmap.nr_entries = 1;
760		xen_e820_table.entries[0].addr = 0ULL;
761		xen_e820_table.entries[0].size = mem_end;
762		/* 8MB slack (to balance backend allocations). */
763		xen_e820_table.entries[0].size += 8ULL << 20;
764		xen_e820_table.entries[0].type = E820_TYPE_RAM;
765		rc = 0;
766	}
767	BUG_ON(rc);
768	BUG_ON(memmap.nr_entries == 0);
769	xen_e820_table.nr_entries = memmap.nr_entries;
770
771	if (xen_initial_domain()) {
772		/*
773		 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
774		 * regions, so if we're using the machine memory map leave the
775		 * region as RAM as it is in the pseudo-physical map.
776		 *
777		 * UNUSABLE regions in domUs are not handled and will need
778		 * a patch in the future.
779		 */
780		xen_ignore_unusable();
781
782#ifdef CONFIG_ISCSI_IBFT_FIND
783		/* Reserve 0.5 MiB to 1 MiB region so iBFT can be found */
784		xen_e820_table.entries[xen_e820_table.nr_entries].addr = IBFT_START;
785		xen_e820_table.entries[xen_e820_table.nr_entries].size = IBFT_END - IBFT_START;
786		xen_e820_table.entries[xen_e820_table.nr_entries].type = E820_TYPE_RESERVED;
787		xen_e820_table.nr_entries++;
788#endif
789	}
790
791	/* Make sure the Xen-supplied memory map is well-ordered. */
792	e820__update_table(&xen_e820_table);
793
794	max_pages = xen_get_max_pages();
795
796	/* How many extra pages do we need due to remapping? */
797	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
798
799	if (max_pages > max_pfn)
800		extra_pages += max_pages - max_pfn;
801
802	/*
803	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
804	 * factor the base size.
 
 
 
805	 *
806	 * Make sure we have no memory above max_pages, as this area
807	 * isn't handled by the p2m management.
 
 
 
 
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_RESERVED)
821			xen_pv_pci_possible = true;
822
823		if (type == E820_TYPE_RAM) {
824			if (addr < mem_end) {
825				chunk_size = min(size, mem_end - addr);
826			} else if (extra_pages) {
827				chunk_size = min(size, PFN_PHYS(extra_pages));
828				pfn_s = PFN_UP(addr);
829				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
830				extra_pages -= n_pfns;
831				xen_add_extra_mem(pfn_s, n_pfns);
832				xen_max_p2m_pfn = pfn_s + n_pfns;
833			} else
834				discard = true;
835		}
836
837		if (!discard)
838			xen_align_and_add_e820_region(addr, chunk_size, type);
839
840		addr += chunk_size;
841		size -= chunk_size;
842		if (size == 0) {
843			i++;
844			if (i < xen_e820_table.nr_entries) {
845				addr = xen_e820_table.entries[i].addr;
846				size = xen_e820_table.entries[i].size;
847			}
848		}
849	}
850
851	/*
852	 * Set the rest as identity mapped, in case PCI BARs are
853	 * located here.
854	 */
855	set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
856
857	/*
858	 * In domU, the ISA region is normal, usable memory, but we
859	 * reserve ISA memory anyway because too many things poke
860	 * about in there.
861	 */
862	e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
863
864	e820__update_table(e820_table);
865
866	/*
867	 * Check whether the kernel itself conflicts with the target E820 map.
868	 * Failing now is better than running into weird problems later due
869	 * to relocating (and even reusing) pages with kernel text or data.
870	 */
871	if (xen_is_e820_reserved(__pa_symbol(_text),
872			__pa_symbol(__bss_stop) - __pa_symbol(_text))) {
873		xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
874		BUG();
875	}
876
877	/*
878	 * Check for a conflict of the hypervisor supplied page tables with
879	 * the target E820 map.
880	 */
881	xen_pt_check_e820();
882
883	xen_reserve_xen_mfnlist();
884
885	/* Check for a conflict of the initrd with the target E820 map. */
886	if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
887				 boot_params.hdr.ramdisk_size)) {
888		phys_addr_t new_area, start, size;
889
890		new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
891		if (!new_area) {
892			xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
893			BUG();
894		}
895
896		start = boot_params.hdr.ramdisk_image;
897		size = boot_params.hdr.ramdisk_size;
898		xen_phys_memcpy(new_area, start, size);
899		pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
900			start, start + size, new_area, new_area + size);
901		memblock_phys_free(start, size);
902		boot_params.hdr.ramdisk_image = new_area;
903		boot_params.ext_ramdisk_image = new_area >> 32;
904	}
905
906	/*
907	 * Set identity map on non-RAM pages and prepare remapping the
908	 * underlying RAM.
909	 */
910	xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
911
912	pr_info("Released %ld page(s)\n", xen_released_pages);
913
914	return "Xen";
915}
916
917static int register_callback(unsigned type, const void *func)
918{
919	struct callback_register callback = {
920		.type = type,
921		.address = XEN_CALLBACK(__KERNEL_CS, func),
922		.flags = CALLBACKF_mask_events,
923	};
924
925	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
926}
927
928void xen_enable_sysenter(void)
929{
930	if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
931	    register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
932		setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
 
 
 
 
 
 
 
 
933}
934
935void xen_enable_syscall(void)
936{
937	int ret;
938
939	ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
940	if (ret != 0) {
941		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
942		/* Pretty fatal; 64-bit userspace has no other
943		   mechanism for syscalls. */
944	}
945
946	if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
947	    register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
948		setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 
 
 
949}
950
951static void __init xen_pvmmu_arch_setup(void)
952{
 
953	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 
 
 
954
955	if (register_callback(CALLBACKTYPE_event,
956			      xen_asm_exc_xen_hypervisor_callback) ||
957	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
958		BUG();
959
960	xen_enable_sysenter();
961	xen_enable_syscall();
962}
963
964/* This function is not called for HVM domains */
965void __init xen_arch_setup(void)
966{
967	xen_panic_handler_init();
968	xen_pvmmu_arch_setup();
969
970#ifdef CONFIG_ACPI
971	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
972		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
973		disable_acpi();
974	}
975#endif
976
977	memcpy(boot_command_line, xen_start_info->cmd_line,
978	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
979	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
980
981	/* Set up idle, making sure it calls safe_halt() pvop */
982	disable_cpuidle();
983	disable_cpufreq();
984	WARN_ON(xen_set_default_idle());
985#ifdef CONFIG_NUMA
986	numa_off = 1;
987#endif
988}
v5.9
  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/sched.h>
 
 10#include <linux/mm.h>
 11#include <linux/pm.h>
 12#include <linux/memblock.h>
 13#include <linux/cpuidle.h>
 14#include <linux/cpufreq.h>
 15#include <linux/memory_hotplug.h>
 16
 17#include <asm/elf.h>
 18#include <asm/vdso.h>
 19#include <asm/e820/api.h>
 20#include <asm/setup.h>
 21#include <asm/acpi.h>
 22#include <asm/numa.h>
 23#include <asm/idtentry.h>
 24#include <asm/xen/hypervisor.h>
 25#include <asm/xen/hypercall.h>
 26
 27#include <xen/xen.h>
 28#include <xen/page.h>
 29#include <xen/interface/callback.h>
 30#include <xen/interface/memory.h>
 31#include <xen/interface/physdev.h>
 32#include <xen/features.h>
 33#include <xen/hvc-console.h>
 34#include "xen-ops.h"
 35#include "mmu.h"
 36
 37#define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
 38
 39/* Amount of extra memory space we add to the e820 ranges */
 40struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
 41
 42/* Number of pages released from the initial allocation. */
 43unsigned long xen_released_pages;
 44
 
 
 
 45/* E820 map used during setting up memory. */
 46static struct e820_table xen_e820_table __initdata;
 47
 48/*
 49 * Buffer used to remap identity mapped pages. We only need the virtual space.
 50 * The physical page behind this address is remapped as needed to different
 51 * buffer pages.
 52 */
 53#define REMAP_SIZE	(P2M_PER_PAGE - 3)
 54static struct {
 55	unsigned long	next_area_mfn;
 56	unsigned long	target_pfn;
 57	unsigned long	size;
 58	unsigned long	mfns[REMAP_SIZE];
 59} xen_remap_buf __initdata __aligned(PAGE_SIZE);
 60static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
 61
 62/* 
 63 * The maximum amount of extra memory compared to the base size.  The
 64 * main scaling factor is the size of struct page.  At extreme ratios
 65 * of base:extra, all the base memory can be filled with page
 66 * structures for the extra memory, leaving no space for anything
 67 * else.
 68 * 
 69 * 10x seems like a reasonable balance between scaling flexibility and
 70 * leaving a practically usable system.
 71 */
 72#define EXTRA_MEM_RATIO		(10)
 73
 74static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
 75
 76static void __init xen_parse_512gb(void)
 77{
 78	bool val = false;
 79	char *arg;
 80
 81	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
 82	if (!arg)
 83		return;
 84
 85	arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
 86	if (!arg)
 87		val = true;
 88	else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
 89		return;
 90
 91	xen_512gb_limit = val;
 92}
 93
 94static void __init xen_add_extra_mem(unsigned long start_pfn,
 95				     unsigned long n_pfns)
 96{
 97	int i;
 98
 99	/*
100	 * No need to check for zero size, should happen rarely and will only
101	 * write a new entry regarded to be unused due to zero size.
102	 */
103	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
104		/* Add new region. */
105		if (xen_extra_mem[i].n_pfns == 0) {
106			xen_extra_mem[i].start_pfn = start_pfn;
107			xen_extra_mem[i].n_pfns = n_pfns;
108			break;
109		}
110		/* Append to existing region. */
111		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
112		    start_pfn) {
113			xen_extra_mem[i].n_pfns += n_pfns;
114			break;
115		}
116	}
117	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
118		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
119
120	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
121}
122
123static void __init xen_del_extra_mem(unsigned long start_pfn,
124				     unsigned long n_pfns)
125{
126	int i;
127	unsigned long start_r, size_r;
128
129	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
130		start_r = xen_extra_mem[i].start_pfn;
131		size_r = xen_extra_mem[i].n_pfns;
132
133		/* Start of region. */
134		if (start_r == start_pfn) {
135			BUG_ON(n_pfns > size_r);
136			xen_extra_mem[i].start_pfn += n_pfns;
137			xen_extra_mem[i].n_pfns -= n_pfns;
138			break;
139		}
140		/* End of region. */
141		if (start_r + size_r == start_pfn + n_pfns) {
142			BUG_ON(n_pfns > size_r);
143			xen_extra_mem[i].n_pfns -= n_pfns;
144			break;
145		}
146		/* Mid of region. */
147		if (start_pfn > start_r && start_pfn < start_r + size_r) {
148			BUG_ON(start_pfn + n_pfns > start_r + size_r);
149			xen_extra_mem[i].n_pfns = start_pfn - start_r;
150			/* Calling memblock_reserve() again is okay. */
151			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
152					  (start_pfn + n_pfns));
153			break;
154		}
155	}
156	memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
157}
158
159/*
160 * Called during boot before the p2m list can take entries beyond the
161 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
162 * invalid.
163 */
164unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
165{
166	int i;
167
168	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
169		if (pfn >= xen_extra_mem[i].start_pfn &&
170		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
171			return INVALID_P2M_ENTRY;
172	}
173
174	return IDENTITY_FRAME(pfn);
175}
176
177/*
178 * Mark all pfns of extra mem as invalid in p2m list.
179 */
180void __init xen_inv_extra_mem(void)
181{
182	unsigned long pfn, pfn_s, pfn_e;
183	int i;
184
185	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
186		if (!xen_extra_mem[i].n_pfns)
187			continue;
188		pfn_s = xen_extra_mem[i].start_pfn;
189		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
190		for (pfn = pfn_s; pfn < pfn_e; pfn++)
191			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
192	}
193}
194
195/*
196 * Finds the next RAM pfn available in the E820 map after min_pfn.
197 * This function updates min_pfn with the pfn found and returns
198 * the size of that range or zero if not found.
199 */
200static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
201{
202	const struct e820_entry *entry = xen_e820_table.entries;
203	unsigned int i;
204	unsigned long done = 0;
205
206	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
207		unsigned long s_pfn;
208		unsigned long e_pfn;
209
210		if (entry->type != E820_TYPE_RAM)
211			continue;
212
213		e_pfn = PFN_DOWN(entry->addr + entry->size);
214
215		/* We only care about E820 after this */
216		if (e_pfn <= *min_pfn)
217			continue;
218
219		s_pfn = PFN_UP(entry->addr);
220
221		/* If min_pfn falls within the E820 entry, we want to start
222		 * at the min_pfn PFN.
223		 */
224		if (s_pfn <= *min_pfn) {
225			done = e_pfn - *min_pfn;
226		} else {
227			done = e_pfn - s_pfn;
228			*min_pfn = s_pfn;
229		}
230		break;
231	}
232
233	return done;
234}
235
236static int __init xen_free_mfn(unsigned long mfn)
237{
238	struct xen_memory_reservation reservation = {
239		.address_bits = 0,
240		.extent_order = 0,
241		.domid        = DOMID_SELF
242	};
243
244	set_xen_guest_handle(reservation.extent_start, &mfn);
245	reservation.nr_extents = 1;
246
247	return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
248}
249
250/*
251 * This releases a chunk of memory and then does the identity map. It's used
252 * as a fallback if the remapping fails.
253 */
254static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
255			unsigned long end_pfn, unsigned long nr_pages)
256{
257	unsigned long pfn, end;
258	int ret;
259
260	WARN_ON(start_pfn > end_pfn);
261
262	/* Release pages first. */
263	end = min(end_pfn, nr_pages);
264	for (pfn = start_pfn; pfn < end; pfn++) {
265		unsigned long mfn = pfn_to_mfn(pfn);
266
267		/* Make sure pfn exists to start with */
268		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
269			continue;
270
271		ret = xen_free_mfn(mfn);
272		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
273
274		if (ret == 1) {
275			xen_released_pages++;
276			if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
277				break;
278		} else
279			break;
280	}
281
282	set_phys_range_identity(start_pfn, end_pfn);
283}
284
285/*
286 * Helper function to update the p2m and m2p tables and kernel mapping.
287 */
288static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
289{
290	struct mmu_update update = {
291		.ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
292		.val = pfn
293	};
294
295	/* Update p2m */
296	if (!set_phys_to_machine(pfn, mfn)) {
297		WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
298		     pfn, mfn);
299		BUG();
300	}
301
302	/* Update m2p */
303	if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
304		WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
305		     mfn, pfn);
306		BUG();
307	}
308
309	/* Update kernel mapping, but not for highmem. */
310	if (pfn >= PFN_UP(__pa(high_memory - 1)))
311		return;
312
313	if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
314					 mfn_pte(mfn, PAGE_KERNEL), 0)) {
315		WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
316		      mfn, pfn);
317		BUG();
318	}
319}
320
321/*
322 * This function updates the p2m and m2p tables with an identity map from
323 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
324 * original allocation at remap_pfn. The information needed for remapping is
325 * saved in the memory itself to avoid the need for allocating buffers. The
326 * complete remap information is contained in a list of MFNs each containing
327 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
328 * This enables us to preserve the original mfn sequence while doing the
329 * remapping at a time when the memory management is capable of allocating
330 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
331 * its callers.
332 */
333static void __init xen_do_set_identity_and_remap_chunk(
334        unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
335{
336	unsigned long buf = (unsigned long)&xen_remap_buf;
337	unsigned long mfn_save, mfn;
338	unsigned long ident_pfn_iter, remap_pfn_iter;
339	unsigned long ident_end_pfn = start_pfn + size;
340	unsigned long left = size;
341	unsigned int i, chunk;
342
343	WARN_ON(size == 0);
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 */
387static 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_warn("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
443static 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
453static 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 independent from the order.
498 */
499void __init xen_remap_memory(void)
500{
501	unsigned long buf = (unsigned long)&xen_remap_buf;
502	unsigned long mfn_save, 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			xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
519			remapped++;
520			pfn++;
521		}
522		if (pfn_s == ~0UL || pfn == pfn_s) {
523			pfn_s = xen_remap_buf.target_pfn;
524			len += xen_remap_buf.size;
525		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
526			len += xen_remap_buf.size;
527		} else {
528			xen_del_extra_mem(pfn_s, len);
529			pfn_s = xen_remap_buf.target_pfn;
530			len = xen_remap_buf.size;
531		}
532		xen_remap_mfn = xen_remap_buf.next_area_mfn;
533	}
534
535	if (pfn_s != ~0UL && len)
536		xen_del_extra_mem(pfn_s, len);
537
538	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
539
540	pr_info("Remapped %ld page(s)\n", remapped);
541}
542
543static unsigned long __init xen_get_pages_limit(void)
544{
545	unsigned long limit;
546
547	limit = MAXMEM / PAGE_SIZE;
548	if (!xen_initial_domain() && xen_512gb_limit)
549		limit = GB(512) / PAGE_SIZE;
550
551	return limit;
552}
553
554static unsigned long __init xen_get_max_pages(void)
555{
556	unsigned long max_pages, limit;
557	domid_t domid = DOMID_SELF;
558	long ret;
559
560	limit = xen_get_pages_limit();
561	max_pages = limit;
562
563	/*
564	 * For the initial domain we use the maximum reservation as
565	 * the maximum page.
566	 *
567	 * For guest domains the current maximum reservation reflects
568	 * the current maximum rather than the static maximum. In this
569	 * case the e820 map provided to us will cover the static
570	 * maximum region.
571	 */
572	if (xen_initial_domain()) {
573		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
574		if (ret > 0)
575			max_pages = ret;
576	}
577
578	return min(max_pages, limit);
579}
580
581static void __init xen_align_and_add_e820_region(phys_addr_t start,
582						 phys_addr_t size, int type)
583{
584	phys_addr_t end = start + size;
585
586	/* Align RAM regions to page boundaries. */
587	if (type == E820_TYPE_RAM) {
588		start = PAGE_ALIGN(start);
589		end &= ~((phys_addr_t)PAGE_SIZE - 1);
590#ifdef CONFIG_MEMORY_HOTPLUG
591		/*
592		 * Don't allow adding memory not in E820 map while booting the
593		 * system. Once the balloon driver is up it will remove that
594		 * restriction again.
595		 */
596		max_mem_size = end;
597#endif
598	}
599
600	e820__range_add(start, end - start, type);
601}
602
603static void __init xen_ignore_unusable(void)
604{
605	struct e820_entry *entry = xen_e820_table.entries;
606	unsigned int i;
607
608	for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
609		if (entry->type == E820_TYPE_UNUSABLE)
610			entry->type = E820_TYPE_RAM;
611	}
612}
613
614bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
615{
616	struct e820_entry *entry;
617	unsigned mapcnt;
618	phys_addr_t end;
619
620	if (!size)
621		return false;
622
623	end = start + size;
624	entry = xen_e820_table.entries;
625
626	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
627		if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
628		    (entry->addr + entry->size) >= end)
629			return false;
630
631		entry++;
632	}
633
634	return true;
635}
636
637/*
638 * Find a free area in physical memory not yet reserved and compliant with
639 * E820 map.
640 * Used to relocate pre-allocated areas like initrd or p2m list which are in
641 * conflict with the to be used E820 map.
642 * In case no area is found, return 0. Otherwise return the physical address
643 * of the area which is already reserved for convenience.
644 */
645phys_addr_t __init xen_find_free_area(phys_addr_t size)
646{
647	unsigned mapcnt;
648	phys_addr_t addr, start;
649	struct e820_entry *entry = xen_e820_table.entries;
650
651	for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
652		if (entry->type != E820_TYPE_RAM || entry->size < size)
653			continue;
654		start = entry->addr;
655		for (addr = start; addr < start + size; addr += PAGE_SIZE) {
656			if (!memblock_is_reserved(addr))
657				continue;
658			start = addr + PAGE_SIZE;
659			if (start + size > entry->addr + entry->size)
660				break;
661		}
662		if (addr >= start + size) {
663			memblock_reserve(start, size);
664			return start;
665		}
666	}
667
668	return 0;
669}
670
671/*
672 * Like memcpy, but with physical addresses for dest and src.
673 */
674static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
675				   phys_addr_t n)
676{
677	phys_addr_t dest_off, src_off, dest_len, src_len, len;
678	void *from, *to;
679
680	while (n) {
681		dest_off = dest & ~PAGE_MASK;
682		src_off = src & ~PAGE_MASK;
683		dest_len = n;
684		if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
685			dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
686		src_len = n;
687		if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
688			src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
689		len = min(dest_len, src_len);
690		to = early_memremap(dest - dest_off, dest_len + dest_off);
691		from = early_memremap(src - src_off, src_len + src_off);
692		memcpy(to, from, len);
693		early_memunmap(to, dest_len + dest_off);
694		early_memunmap(from, src_len + src_off);
695		n -= len;
696		dest += len;
697		src += len;
698	}
699}
700
701/*
702 * Reserve Xen mfn_list.
703 */
704static void __init xen_reserve_xen_mfnlist(void)
705{
706	phys_addr_t start, size;
707
708	if (xen_start_info->mfn_list >= __START_KERNEL_map) {
709		start = __pa(xen_start_info->mfn_list);
710		size = PFN_ALIGN(xen_start_info->nr_pages *
711				 sizeof(unsigned long));
712	} else {
713		start = PFN_PHYS(xen_start_info->first_p2m_pfn);
714		size = PFN_PHYS(xen_start_info->nr_p2m_frames);
715	}
716
717	memblock_reserve(start, size);
718	if (!xen_is_e820_reserved(start, size))
719		return;
720
721	xen_relocate_p2m();
722	memblock_free(start, size);
723}
724
725/**
726 * machine_specific_memory_setup - Hook for machine specific memory setup.
727 **/
728char * __init xen_memory_setup(void)
729{
730	unsigned long max_pfn, pfn_s, n_pfns;
731	phys_addr_t mem_end, addr, size, chunk_size;
732	u32 type;
733	int rc;
734	struct xen_memory_map memmap;
735	unsigned long max_pages;
736	unsigned long extra_pages = 0;
737	int i;
738	int op;
739
740	xen_parse_512gb();
741	max_pfn = xen_get_pages_limit();
742	max_pfn = min(max_pfn, xen_start_info->nr_pages);
743	mem_end = PFN_PHYS(max_pfn);
744
745	memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
746	set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
747
748#if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
749	xen_saved_max_mem_size = max_mem_size;
750#endif
751
752	op = xen_initial_domain() ?
753		XENMEM_machine_memory_map :
754		XENMEM_memory_map;
755	rc = HYPERVISOR_memory_op(op, &memmap);
756	if (rc == -ENOSYS) {
757		BUG_ON(xen_initial_domain());
758		memmap.nr_entries = 1;
759		xen_e820_table.entries[0].addr = 0ULL;
760		xen_e820_table.entries[0].size = mem_end;
761		/* 8MB slack (to balance backend allocations). */
762		xen_e820_table.entries[0].size += 8ULL << 20;
763		xen_e820_table.entries[0].type = E820_TYPE_RAM;
764		rc = 0;
765	}
766	BUG_ON(rc);
767	BUG_ON(memmap.nr_entries == 0);
768	xen_e820_table.nr_entries = memmap.nr_entries;
769
770	/*
771	 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
772	 * regions, so if we're using the machine memory map leave the
773	 * region as RAM as it is in the pseudo-physical map.
774	 *
775	 * UNUSABLE regions in domUs are not handled and will need
776	 * a patch in the future.
777	 */
778	if (xen_initial_domain())
779		xen_ignore_unusable();
780
 
 
 
 
 
 
 
 
 
781	/* Make sure the Xen-supplied memory map is well-ordered. */
782	e820__update_table(&xen_e820_table);
783
784	max_pages = xen_get_max_pages();
785
786	/* How many extra pages do we need due to remapping? */
787	max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
788
789	if (max_pages > max_pfn)
790		extra_pages += max_pages - max_pfn;
791
792	/*
793	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
794	 * factor the base size.  On non-highmem systems, the base
795	 * size is the full initial memory allocation; on highmem it
796	 * is limited to the max size of lowmem, so that it doesn't
797	 * get completely filled.
798	 *
799	 * Make sure we have no memory above max_pages, as this area
800	 * isn't handled by the p2m management.
801	 *
802	 * In principle there could be a problem in lowmem systems if
803	 * the initial memory is also very large with respect to
804	 * lowmem, but we won't try to deal with that here.
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_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
911static 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
922void xen_enable_sysenter(void)
923{
924	int ret;
925	unsigned sysenter_feature;
926
927	sysenter_feature = X86_FEATURE_SYSENTER32;
928
929	if (!boot_cpu_has(sysenter_feature))
930		return;
931
932	ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
933	if(ret != 0)
934		setup_clear_cpu_cap(sysenter_feature);
935}
936
937void xen_enable_syscall(void)
938{
939	int ret;
940
941	ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
942	if (ret != 0) {
943		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
944		/* Pretty fatal; 64-bit userspace has no other
945		   mechanism for syscalls. */
946	}
947
948	if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
949		ret = register_callback(CALLBACKTYPE_syscall32,
950					xen_syscall32_target);
951		if (ret != 0)
952			setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
953	}
954}
955
956static void __init xen_pvmmu_arch_setup(void)
957{
958	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
959	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
960
961	HYPERVISOR_vm_assist(VMASST_CMD_enable,
962			     VMASST_TYPE_pae_extended_cr3);
963
964	if (register_callback(CALLBACKTYPE_event,
965			      xen_asm_exc_xen_hypervisor_callback) ||
966	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
967		BUG();
968
969	xen_enable_sysenter();
970	xen_enable_syscall();
971}
972
973/* This function is not called for HVM domains */
974void __init xen_arch_setup(void)
975{
976	xen_panic_handler_init();
977	xen_pvmmu_arch_setup();
978
979#ifdef CONFIG_ACPI
980	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
981		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
982		disable_acpi();
983	}
984#endif
985
986	memcpy(boot_command_line, xen_start_info->cmd_line,
987	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
988	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
989
990	/* Set up idle, making sure it calls safe_halt() pvop */
991	disable_cpuidle();
992	disable_cpufreq();
993	WARN_ON(xen_set_default_idle());
994#ifdef CONFIG_NUMA
995	numa_off = 1;
996#endif
997}