p2m.c

  1// SPDX-License-Identifier: GPL-2.0
  2
  3/*
  4 * Xen leaves the responsibility for maintaining p2m mappings to the
  5 * guests themselves, but it must also access and update the p2m array
  6 * during suspend/resume when all the pages are reallocated.
  7 *
  8 * The logical flat p2m table is mapped to a linear kernel memory area.
  9 * For accesses by Xen a three-level tree linked via mfns only is set up to
 10 * allow the address space to be sparse.
 11 *
 12 *               Xen
 13 *                |
 14 *          p2m_top_mfn
 15 *              /   \
 16 * p2m_mid_mfn p2m_mid_mfn
 17 *         /           /
 18 *  p2m p2m p2m ...
 19 *
 20 * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
 21 *
 22 * The p2m_top_mfn level is limited to 1 page, so the maximum representable
 23 * pseudo-physical address space is:
 24 *  P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
 25 *
 26 * P2M_PER_PAGE depends on the architecture, as a mfn is always
 27 * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
 28 * 512 and 1024 entries respectively.
 29 *
 30 * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
 31 *
 32 * However not all entries are filled with MFNs. Specifically for all other
 33 * leaf entries, or for the top  root, or middle one, for which there is a void
 34 * entry, we assume it is  "missing". So (for example)
 35 *  pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
 36 * We have a dedicated page p2m_missing with all entries being
 37 * INVALID_P2M_ENTRY. This page may be referenced multiple times in the p2m
 38 * list/tree in case there are multiple areas with P2M_PER_PAGE invalid pfns.
 39 *
 40 * We also have the possibility of setting 1-1 mappings on certain regions, so
 41 * that:
 42 *  pfn_to_mfn(0xc0000)=0xc0000
 43 *
 44 * The benefit of this is, that we can assume for non-RAM regions (think
 45 * PCI BARs, or ACPI spaces), we can create mappings easily because we
 46 * get the PFN value to match the MFN.
 47 *
 48 * For this to work efficiently we have one new page p2m_identity. All entries
 49 * in p2m_identity are set to INVALID_P2M_ENTRY type (Xen toolstack only
 50 * recognizes that and MFNs, no other fancy value).
 51 *
 52 * On lookup we spot that the entry points to p2m_identity and return the
 53 * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
 54 * If the entry points to an allocated page, we just proceed as before and
 55 * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
 56 * appropriate functions (pfn_to_mfn).
 57 *
 58 * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
 59 * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
 60 * non-identity pfn. To protect ourselves against we elect to set (and get) the
 61 * IDENTITY_FRAME_BIT on all identity mapped PFNs.
 62 */
 63
 64#include <linux/init.h>
 65#include <linux/export.h>
 66#include <linux/list.h>
 67#include <linux/hash.h>
 68#include <linux/sched.h>
 69#include <linux/seq_file.h>
 70#include <linux/memblock.h>
 71#include <linux/slab.h>
 72#include <linux/vmalloc.h>
 73#include <linux/acpi.h>
 74
 75#include <asm/cache.h>
 76#include <asm/setup.h>
 77#include <linux/uaccess.h>
 78
 79#include <asm/xen/page.h>
 80#include <asm/xen/hypercall.h>
 81#include <asm/xen/hypervisor.h>
 82#include <xen/balloon.h>
 83#include <xen/grant_table.h>
 84#include <xen/hvc-console.h>
 85
 86#include "xen-ops.h"
 87
 88#define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
 89#define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
 90
 91#define MAX_P2M_PFN	(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
 92
 93#define PMDS_PER_MID_PAGE	(P2M_MID_PER_PAGE / PTRS_PER_PTE)
 94
 95unsigned long *xen_p2m_addr __read_mostly;
 96EXPORT_SYMBOL_GPL(xen_p2m_addr);
 97unsigned long xen_p2m_size __read_mostly;
 98EXPORT_SYMBOL_GPL(xen_p2m_size);
 99unsigned long xen_max_p2m_pfn __read_mostly;
100EXPORT_SYMBOL_GPL(xen_max_p2m_pfn);
101
102#ifdef CONFIG_XEN_MEMORY_HOTPLUG_LIMIT
103#define P2M_LIMIT CONFIG_XEN_MEMORY_HOTPLUG_LIMIT
104#else
105#define P2M_LIMIT 0
106#endif
107
108static DEFINE_SPINLOCK(p2m_update_lock);
109
110static unsigned long *p2m_mid_missing_mfn;
111static unsigned long *p2m_top_mfn;
112static unsigned long **p2m_top_mfn_p;
113static unsigned long *p2m_missing;
114static unsigned long *p2m_identity;
115static pte_t *p2m_missing_pte;
116static pte_t *p2m_identity_pte;
117
118/*
119 * Hint at last populated PFN.
120 *
121 * Used to set HYPERVISOR_shared_info->arch.max_pfn so the toolstack
122 * can avoid scanning the whole P2M (which may be sized to account for
123 * hotplugged memory).
124 */
125static unsigned long xen_p2m_last_pfn;
126
127static inline unsigned p2m_top_index(unsigned long pfn)
128{
129	BUG_ON(pfn >= MAX_P2M_PFN);
130	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
131}
132
133static inline unsigned p2m_mid_index(unsigned long pfn)
134{
135	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
136}
137
138static void p2m_top_mfn_init(unsigned long *top)
139{
140	unsigned i;
141
142	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
143		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
144}
145
146static void p2m_top_mfn_p_init(unsigned long **top)
147{
148	unsigned i;
149
150	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
151		top[i] = p2m_mid_missing_mfn;
152}
153
154static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
155{
156	unsigned i;
157
158	for (i = 0; i < P2M_MID_PER_PAGE; i++)
159		mid[i] = virt_to_mfn(leaf);
160}
161
162static void p2m_init(unsigned long *p2m)
163{
164	unsigned i;
165
166	for (i = 0; i < P2M_PER_PAGE; i++)
167		p2m[i] = INVALID_P2M_ENTRY;
168}
169
170static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
171{
172	unsigned i;
173
174	for (i = 0; i < P2M_PER_PAGE; i++)
175		p2m[i] = IDENTITY_FRAME(pfn + i);
176}
177
178static void * __ref alloc_p2m_page(void)
179{
180	if (unlikely(!slab_is_available())) {
181		return memblock_alloc_or_panic(PAGE_SIZE, PAGE_SIZE);
182	}
183
184	return (void *)__get_free_page(GFP_KERNEL);
185}
186
187static void __ref free_p2m_page(void *p)
188{
189	if (unlikely(!slab_is_available())) {
190		memblock_free(p, PAGE_SIZE);
191		return;
192	}
193
194	free_page((unsigned long)p);
195}
196
197/*
198 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
199 *
200 * This is called both at boot time, and after resuming from suspend:
201 * - At boot time we're called rather early, and must use alloc_bootmem*()
202 *   to allocate memory.
203 *
204 * - After resume we're called from within stop_machine, but the mfn
205 *   tree should already be completely allocated.
206 */
207void __ref xen_build_mfn_list_list(void)
208{
209	unsigned long pfn, mfn;
210	pte_t *ptep;
211	unsigned int level, topidx, mididx;
212	unsigned long *mid_mfn_p;
213
214	if (xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
215		return;
216
217	/* Pre-initialize p2m_top_mfn to be completely missing */
218	if (p2m_top_mfn == NULL) {
219		p2m_mid_missing_mfn = alloc_p2m_page();
220		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
221
222		p2m_top_mfn_p = alloc_p2m_page();
223		p2m_top_mfn_p_init(p2m_top_mfn_p);
224
225		p2m_top_mfn = alloc_p2m_page();
226		p2m_top_mfn_init(p2m_top_mfn);
227	} else {
228		/* Reinitialise, mfn's all change after migration */
229		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
230	}
231
232	for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
233	     pfn += P2M_PER_PAGE) {
234		topidx = p2m_top_index(pfn);
235		mididx = p2m_mid_index(pfn);
236
237		mid_mfn_p = p2m_top_mfn_p[topidx];
238		ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
239				      &level);
240		BUG_ON(!ptep || level != PG_LEVEL_4K);
241		mfn = pte_mfn(*ptep);
242		ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
243
244		/* Don't bother allocating any mfn mid levels if
245		 * they're just missing, just update the stored mfn,
246		 * since all could have changed over a migrate.
247		 */
248		if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
249			BUG_ON(mididx);
250			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
251			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
252			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
253			continue;
254		}
255
256		if (mid_mfn_p == p2m_mid_missing_mfn) {
257			mid_mfn_p = alloc_p2m_page();
258			p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
259
260			p2m_top_mfn_p[topidx] = mid_mfn_p;
261		}
262
263		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
264		mid_mfn_p[mididx] = mfn;
265	}
266}
267
268void xen_setup_mfn_list_list(void)
269{
270	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
271
272	if (xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
273		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = ~0UL;
274	else
275		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
276			virt_to_mfn(p2m_top_mfn);
277	HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
278	HYPERVISOR_shared_info->arch.p2m_generation = 0;
279	HYPERVISOR_shared_info->arch.p2m_vaddr = (unsigned long)xen_p2m_addr;
280	HYPERVISOR_shared_info->arch.p2m_cr3 =
281		xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
282}
283
284/* Set up p2m_top to point to the domain-builder provided p2m pages */
285void __init xen_build_dynamic_phys_to_machine(void)
286{
287	unsigned long pfn;
288
289	xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
290	xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
291
292	for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
293		xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
294
295	xen_max_p2m_pfn = xen_p2m_size;
296}
297
298#define P2M_TYPE_IDENTITY	0
299#define P2M_TYPE_MISSING	1
300#define P2M_TYPE_PFN		2
301#define P2M_TYPE_UNKNOWN	3
302
303static int xen_p2m_elem_type(unsigned long pfn)
304{
305	unsigned long mfn;
306
307	if (pfn >= xen_p2m_size)
308		return P2M_TYPE_IDENTITY;
309
310	mfn = xen_p2m_addr[pfn];
311
312	if (mfn == INVALID_P2M_ENTRY)
313		return P2M_TYPE_MISSING;
314
315	if (mfn & IDENTITY_FRAME_BIT)
316		return P2M_TYPE_IDENTITY;
317
318	return P2M_TYPE_PFN;
319}
320
321static void __init xen_rebuild_p2m_list(unsigned long *p2m)
322{
323	unsigned int i, chunk;
324	unsigned long pfn;
325	unsigned long *mfns;
326	pte_t *ptep;
327	pmd_t *pmdp;
328	int type;
329
330	p2m_missing = alloc_p2m_page();
331	p2m_init(p2m_missing);
332	p2m_identity = alloc_p2m_page();
333	p2m_init(p2m_identity);
334
335	p2m_missing_pte = alloc_p2m_page();
336	paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
337	p2m_identity_pte = alloc_p2m_page();
338	paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
339	for (i = 0; i < PTRS_PER_PTE; i++) {
340		set_pte(p2m_missing_pte + i,
341			pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
342		set_pte(p2m_identity_pte + i,
343			pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
344	}
345
346	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
347		/*
348		 * Try to map missing/identity PMDs or p2m-pages if possible.
349		 * We have to respect the structure of the mfn_list_list
350		 * which will be built just afterwards.
351		 * Chunk size to test is one p2m page if we are in the middle
352		 * of a mfn_list_list mid page and the complete mid page area
353		 * if we are at index 0 of the mid page. Please note that a
354		 * mid page might cover more than one PMD, e.g. on 32 bit PAE
355		 * kernels.
356		 */
357		chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
358			P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
359
360		type = xen_p2m_elem_type(pfn);
361		i = 0;
362		if (type != P2M_TYPE_PFN)
363			for (i = 1; i < chunk; i++)
364				if (xen_p2m_elem_type(pfn + i) != type)
365					break;
366		if (i < chunk)
367			/* Reset to minimal chunk size. */
368			chunk = P2M_PER_PAGE;
369
370		if (type == P2M_TYPE_PFN || i < chunk) {
371			/* Use initial p2m page contents. */
372			mfns = alloc_p2m_page();
373			copy_page(mfns, xen_p2m_addr + pfn);
374			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
375			set_pte(ptep,
376				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
377			continue;
378		}
379
380		if (chunk == P2M_PER_PAGE) {
381			/* Map complete missing or identity p2m-page. */
382			mfns = (type == P2M_TYPE_MISSING) ?
383				p2m_missing : p2m_identity;
384			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
385			set_pte(ptep,
386				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
387			continue;
388		}
389
390		/* Complete missing or identity PMD(s) can be mapped. */
391		ptep = (type == P2M_TYPE_MISSING) ?
392			p2m_missing_pte : p2m_identity_pte;
393		for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
394			pmdp = populate_extra_pmd(
395				(unsigned long)(p2m + pfn) + i * PMD_SIZE);
396			set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
397		}
398	}
399}
400
401void __init xen_vmalloc_p2m_tree(void)
402{
403	static struct vm_struct vm;
404	unsigned long p2m_limit;
405
406	xen_p2m_last_pfn = xen_max_p2m_pfn;
407
408	p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
409	vm.flags = VM_ALLOC;
410	vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
411			PMD_SIZE * PMDS_PER_MID_PAGE);
412	vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
413	pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
414
415	xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
416
417	xen_rebuild_p2m_list(vm.addr);
418
419	xen_p2m_addr = vm.addr;
420	xen_p2m_size = xen_max_p2m_pfn;
421
422	xen_inv_extra_mem();
423}
424
425unsigned long get_phys_to_machine(unsigned long pfn)
426{
427	pte_t *ptep;
428	unsigned int level;
429
430	if (unlikely(pfn >= xen_p2m_size)) {
431		if (pfn < xen_max_p2m_pfn)
432			return xen_chk_extra_mem(pfn);
433
434		return IDENTITY_FRAME(pfn);
435	}
436
437	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
438	BUG_ON(!ptep || level != PG_LEVEL_4K);
439
440	/*
441	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
442	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
443	 * would be wrong.
444	 */
445	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
446		return IDENTITY_FRAME(pfn);
447
448	return xen_p2m_addr[pfn];
449}
450EXPORT_SYMBOL_GPL(get_phys_to_machine);
451
452/*
453 * Allocate new pmd(s). It is checked whether the old pmd is still in place.
454 * If not, nothing is changed. This is okay as the only reason for allocating
455 * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
456 * pmd.
457 */
458static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
459{
460	pte_t *ptechk;
461	pte_t *pte_newpg[PMDS_PER_MID_PAGE];
462	pmd_t *pmdp;
463	unsigned int level;
464	unsigned long flags;
465	unsigned long vaddr;
466	int i;
467
468	/* Do all allocations first to bail out in error case. */
469	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
470		pte_newpg[i] = alloc_p2m_page();
471		if (!pte_newpg[i]) {
472			for (i--; i >= 0; i--)
473				free_p2m_page(pte_newpg[i]);
474
475			return NULL;
476		}
477	}
478
479	vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
480
481	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
482		copy_page(pte_newpg[i], pte_pg);
483		paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
484
485		pmdp = lookup_pmd_address(vaddr);
486		BUG_ON(!pmdp);
487
488		spin_lock_irqsave(&p2m_update_lock, flags);
489
490		ptechk = lookup_address(vaddr, &level);
491		if (ptechk == pte_pg) {
492			HYPERVISOR_shared_info->arch.p2m_generation++;
493			wmb(); /* Tools are synchronizing via p2m_generation. */
494			set_pmd(pmdp,
495				__pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
496			wmb(); /* Tools are synchronizing via p2m_generation. */
497			HYPERVISOR_shared_info->arch.p2m_generation++;
498			pte_newpg[i] = NULL;
499		}
500
501		spin_unlock_irqrestore(&p2m_update_lock, flags);
502
503		if (pte_newpg[i]) {
504			paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
505			free_p2m_page(pte_newpg[i]);
506		}
507
508		vaddr += PMD_SIZE;
509	}
510
511	return lookup_address(addr, &level);
512}
513
514/*
515 * Fully allocate the p2m structure for a given pfn.  We need to check
516 * that both the top and mid levels are allocated, and make sure the
517 * parallel mfn tree is kept in sync.  We may race with other cpus, so
518 * the new pages are installed with cmpxchg; if we lose the race then
519 * simply free the page we allocated and use the one that's there.
520 */
521int xen_alloc_p2m_entry(unsigned long pfn)
522{
523	unsigned topidx;
524	unsigned long *top_mfn_p, *mid_mfn;
525	pte_t *ptep, *pte_pg;
526	unsigned int level;
527	unsigned long flags;
528	unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
529	unsigned long p2m_pfn;
530
531	ptep = lookup_address(addr, &level);
532	BUG_ON(!ptep || level != PG_LEVEL_4K);
533	pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
534
535	if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
536		/* PMD level is missing, allocate a new one */
537		ptep = alloc_p2m_pmd(addr, pte_pg);
538		if (!ptep)
539			return -ENOMEM;
540	}
541
542	if (p2m_top_mfn && pfn < MAX_P2M_PFN) {
543		topidx = p2m_top_index(pfn);
544		top_mfn_p = &p2m_top_mfn[topidx];
545		mid_mfn = READ_ONCE(p2m_top_mfn_p[topidx]);
546
547		BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
548
549		if (mid_mfn == p2m_mid_missing_mfn) {
550			/* Separately check the mid mfn level */
551			unsigned long missing_mfn;
552			unsigned long mid_mfn_mfn;
553
554			mid_mfn = alloc_p2m_page();
555			if (!mid_mfn)
556				return -ENOMEM;
557
558			p2m_mid_mfn_init(mid_mfn, p2m_missing);
559
560			missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
561			mid_mfn_mfn = virt_to_mfn(mid_mfn);
562			/* try_cmpxchg() updates missing_mfn on failure. */
563			if (try_cmpxchg(top_mfn_p, &missing_mfn, mid_mfn_mfn)) {
564				p2m_top_mfn_p[topidx] = mid_mfn;
565			} else {
566				free_p2m_page(mid_mfn);
567				mid_mfn = mfn_to_virt(missing_mfn);
568			}
569		}
570	} else {
571		mid_mfn = NULL;
572	}
573
574	p2m_pfn = pte_pfn(READ_ONCE(*ptep));
575	if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
576	    p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
577		/* p2m leaf page is missing */
578		unsigned long *p2m;
579
580		p2m = alloc_p2m_page();
581		if (!p2m)
582			return -ENOMEM;
583
584		if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
585			p2m_init(p2m);
586		else
587			p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
588
589		spin_lock_irqsave(&p2m_update_lock, flags);
590
591		if (pte_pfn(*ptep) == p2m_pfn) {
592			HYPERVISOR_shared_info->arch.p2m_generation++;
593			wmb(); /* Tools are synchronizing via p2m_generation. */
594			set_pte(ptep,
595				pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
596			wmb(); /* Tools are synchronizing via p2m_generation. */
597			HYPERVISOR_shared_info->arch.p2m_generation++;
598			if (mid_mfn)
599				mid_mfn[p2m_mid_index(pfn)] = virt_to_mfn(p2m);
600			p2m = NULL;
601		}
602
603		spin_unlock_irqrestore(&p2m_update_lock, flags);
604
605		if (p2m)
606			free_p2m_page(p2m);
607	}
608
609	/* Expanded the p2m? */
610	if (pfn >= xen_p2m_last_pfn) {
611		xen_p2m_last_pfn = ALIGN(pfn + 1, P2M_PER_PAGE);
612		HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
613	}
614
615	return 0;
616}
617EXPORT_SYMBOL(xen_alloc_p2m_entry);
618
619unsigned long __init set_phys_range_identity(unsigned long pfn_s,
620				      unsigned long pfn_e)
621{
622	unsigned long pfn;
623
624	if (unlikely(pfn_s >= xen_p2m_size))
625		return 0;
626
627	if (pfn_s > pfn_e)
628		return 0;
629
630	if (pfn_e > xen_p2m_size)
631		pfn_e = xen_p2m_size;
632
633	for (pfn = pfn_s; pfn < pfn_e; pfn++)
634		xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
635
636	return pfn - pfn_s;
637}
638
639bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
640{
641	pte_t *ptep;
642	unsigned int level;
643
644	/* Only invalid entries allowed above the highest p2m covered frame. */
645	if (unlikely(pfn >= xen_p2m_size))
646		return mfn == INVALID_P2M_ENTRY;
647
648	/*
649	 * The interface requires atomic updates on p2m elements.
650	 * xen_safe_write_ulong() is using an atomic store via asm().
651	 */
652	if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
653		return true;
654
655	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
656	BUG_ON(!ptep || level != PG_LEVEL_4K);
657
658	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
659		return mfn == INVALID_P2M_ENTRY;
660
661	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
662		return mfn == IDENTITY_FRAME(pfn);
663
664	return false;
665}
666
667bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
668{
669	if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
670		int ret;
671
672		ret = xen_alloc_p2m_entry(pfn);
673		if (ret < 0)
674			return false;
675
676		return __set_phys_to_machine(pfn, mfn);
677	}
678
679	return true;
680}
681
682int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
683			    struct gnttab_map_grant_ref *kmap_ops,
684			    struct page **pages, unsigned int count)
685{
686	int i, ret = 0;
687	pte_t *pte;
688
689	if (!xen_pv_domain())
690		return 0;
691
692	if (kmap_ops) {
693		ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
694						kmap_ops, count);
695		if (ret)
696			goto out;
697	}
698
699	for (i = 0; i < count; i++) {
700		unsigned long mfn, pfn;
701		struct gnttab_unmap_grant_ref unmap[2];
702		int rc;
703
704		/* Do not add to override if the map failed. */
705		if (map_ops[i].status != GNTST_okay ||
706		    (kmap_ops && kmap_ops[i].status != GNTST_okay))
707			continue;
708
709		if (map_ops[i].flags & GNTMAP_contains_pte) {
710			pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
711				(map_ops[i].host_addr & ~PAGE_MASK));
712			mfn = pte_mfn(*pte);
713		} else {
714			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
715		}
716		pfn = page_to_pfn(pages[i]);
717
718		WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
719
720		if (likely(set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
721			continue;
722
723		/*
724		 * Signal an error for this slot. This in turn requires
725		 * immediate unmapping.
726		 */
727		map_ops[i].status = GNTST_general_error;
728		unmap[0].host_addr = map_ops[i].host_addr;
729		unmap[0].handle = map_ops[i].handle;
730		map_ops[i].handle = INVALID_GRANT_HANDLE;
731		if (map_ops[i].flags & GNTMAP_device_map)
732			unmap[0].dev_bus_addr = map_ops[i].dev_bus_addr;
733		else
734			unmap[0].dev_bus_addr = 0;
735
736		if (kmap_ops) {
737			kmap_ops[i].status = GNTST_general_error;
738			unmap[1].host_addr = kmap_ops[i].host_addr;
739			unmap[1].handle = kmap_ops[i].handle;
740			kmap_ops[i].handle = INVALID_GRANT_HANDLE;
741			if (kmap_ops[i].flags & GNTMAP_device_map)
742				unmap[1].dev_bus_addr = kmap_ops[i].dev_bus_addr;
743			else
744				unmap[1].dev_bus_addr = 0;
745		}
746
747		/*
748		 * Pre-populate both status fields, to be recognizable in
749		 * the log message below.
750		 */
751		unmap[0].status = 1;
752		unmap[1].status = 1;
753
754		rc = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
755					       unmap, 1 + !!kmap_ops);
756		if (rc || unmap[0].status != GNTST_okay ||
757		    unmap[1].status != GNTST_okay)
758			pr_err_once("gnttab unmap failed: rc=%d st0=%d st1=%d\n",
759				    rc, unmap[0].status, unmap[1].status);
760	}
761
762out:
763	return ret;
764}
765
766int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
767			      struct gnttab_unmap_grant_ref *kunmap_ops,
768			      struct page **pages, unsigned int count)
769{
770	int i, ret = 0;
771
772	if (!xen_pv_domain())
773		return 0;
774
775	for (i = 0; i < count; i++) {
776		unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
777		unsigned long pfn = page_to_pfn(pages[i]);
778
779		if (mfn != INVALID_P2M_ENTRY && (mfn & FOREIGN_FRAME_BIT))
780			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
781		else
782			ret = -EINVAL;
783	}
784	if (kunmap_ops)
785		ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
786						kunmap_ops, count) ?: ret;
787
788	return ret;
789}
790
791/* Remapped non-RAM areas */
792#define NR_NONRAM_REMAP 4
793static struct nonram_remap {
794	phys_addr_t maddr;
795	phys_addr_t paddr;
796	size_t size;
797} xen_nonram_remap[NR_NONRAM_REMAP] __ro_after_init;
798static unsigned int nr_nonram_remap __ro_after_init;
799
800/*
801 * Do the real remapping of non-RAM regions as specified in the
802 * xen_nonram_remap[] array.
803 * In case of an error just crash the system.
804 */
805void __init xen_do_remap_nonram(void)
806{
807	unsigned int i;
808	unsigned int remapped = 0;
809	const struct nonram_remap *remap = xen_nonram_remap;
810	unsigned long pfn, mfn, end_pfn;
811
812	for (i = 0; i < nr_nonram_remap; i++) {
813		end_pfn = PFN_UP(remap->paddr + remap->size);
814		pfn = PFN_DOWN(remap->paddr);
815		mfn = PFN_DOWN(remap->maddr);
816		while (pfn < end_pfn) {
817			if (!set_phys_to_machine(pfn, mfn))
818				panic("Failed to set p2m mapping for pfn=%lx mfn=%lx\n",
819				       pfn, mfn);
820
821			pfn++;
822			mfn++;
823			remapped++;
824		}
825
826		remap++;
827	}
828
829	pr_info("Remapped %u non-RAM page(s)\n", remapped);
830}
831
832#ifdef CONFIG_ACPI
833/*
834 * Xen variant of acpi_os_ioremap() taking potentially remapped non-RAM
835 * regions into account.
836 * Any attempt to map an area crossing a remap boundary will produce a
837 * WARN() splat.
838 * phys is related to remap->maddr on input and will be rebased to remap->paddr.
839 */
840static void __iomem *xen_acpi_os_ioremap(acpi_physical_address phys,
841					 acpi_size size)
842{
843	unsigned int i;
844	const struct nonram_remap *remap = xen_nonram_remap;
845
846	for (i = 0; i < nr_nonram_remap; i++) {
847		if (phys + size > remap->maddr &&
848		    phys < remap->maddr + remap->size) {
849			WARN_ON(phys < remap->maddr ||
850				phys + size > remap->maddr + remap->size);
851			phys += remap->paddr - remap->maddr;
852			break;
853		}
854	}
855
856	return x86_acpi_os_ioremap(phys, size);
857}
858#endif /* CONFIG_ACPI */
859
860/*
861 * Add a new non-RAM remap entry.
862 * In case of no free entry found, just crash the system.
863 */
864void __init xen_add_remap_nonram(phys_addr_t maddr, phys_addr_t paddr,
865				 unsigned long size)
866{
867	BUG_ON((maddr & ~PAGE_MASK) != (paddr & ~PAGE_MASK));
868
869	if (nr_nonram_remap == NR_NONRAM_REMAP) {
870		xen_raw_console_write("Number of required E820 entry remapping actions exceed maximum value\n");
871		BUG();
872	}
873
874#ifdef CONFIG_ACPI
875	/* Switch to the Xen acpi_os_ioremap() variant. */
876	if (nr_nonram_remap == 0)
877		acpi_os_ioremap = xen_acpi_os_ioremap;
878#endif
879
880	xen_nonram_remap[nr_nonram_remap].maddr = maddr;
881	xen_nonram_remap[nr_nonram_remap].paddr = paddr;
882	xen_nonram_remap[nr_nonram_remap].size = size;
883
884	nr_nonram_remap++;
885}
886
887#ifdef CONFIG_XEN_DEBUG_FS
888#include <linux/debugfs.h>
889static int p2m_dump_show(struct seq_file *m, void *v)
890{
891	static const char * const type_name[] = {
892				[P2M_TYPE_IDENTITY] = "identity",
893				[P2M_TYPE_MISSING] = "missing",
894				[P2M_TYPE_PFN] = "pfn",
895				[P2M_TYPE_UNKNOWN] = "abnormal"};
896	unsigned long pfn, first_pfn;
897	int type, prev_type;
898
899	prev_type = xen_p2m_elem_type(0);
900	first_pfn = 0;
901
902	for (pfn = 0; pfn < xen_p2m_size; pfn++) {
903		type = xen_p2m_elem_type(pfn);
904		if (type != prev_type) {
905			seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
906				   type_name[prev_type]);
907			prev_type = type;
908			first_pfn = pfn;
909		}
910	}
911	seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
912		   type_name[prev_type]);
913	return 0;
914}
915
916DEFINE_SHOW_ATTRIBUTE(p2m_dump);
917
918static struct dentry *d_mmu_debug;
919
920static int __init xen_p2m_debugfs(void)
921{
922	struct dentry *d_xen = xen_init_debugfs();
923
924	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
925
926	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
927	return 0;
928}
929fs_initcall(xen_p2m_debugfs);
930#endif /* CONFIG_XEN_DEBUG_FS */