2 * Xen leaves the responsibility for maintaining p2m mappings to the
3 * guests themselves, but it must also access and update the p2m array
4 * during suspend/resume when all the pages are reallocated.
6 * The p2m table is logically a flat array, but we implement it as a
7 * three-level tree to allow the address space to be sparse.
13 * p2m_mid p2m_mid p2m_mid_mfn p2m_mid_mfn
15 * p2m p2m p2m p2m p2m p2m p2m ...
17 * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
19 * The p2m_top and p2m_top_mfn levels are limited to 1 page, so the
20 * maximum representable pseudo-physical address space is:
21 * P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
23 * P2M_PER_PAGE depends on the architecture, as a mfn is always
24 * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
25 * 512 and 1024 entries respectively.
27 * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
29 * However not all entries are filled with MFNs. Specifically for all other
30 * leaf entries, or for the top root, or middle one, for which there is a void
31 * entry, we assume it is "missing". So (for example)
32 * pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
34 * We also have the possibility of setting 1-1 mappings on certain regions, so
36 * pfn_to_mfn(0xc0000)=0xc0000
38 * The benefit of this is, that we can assume for non-RAM regions (think
39 * PCI BARs, or ACPI spaces), we can create mappings easily because we
40 * get the PFN value to match the MFN.
42 * For this to work efficiently we have one new page p2m_identity and
43 * allocate (via reserved_brk) any other pages we need to cover the sides
44 * (1GB or 4MB boundary violations). All entries in p2m_identity are set to
45 * INVALID_P2M_ENTRY type (Xen toolstack only recognizes that and MFNs,
46 * no other fancy value).
48 * On lookup we spot that the entry points to p2m_identity and return the
49 * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
50 * If the entry points to an allocated page, we just proceed as before and
51 * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
52 * appropriate functions (pfn_to_mfn).
54 * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
55 * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
56 * non-identity pfn. To protect ourselves against we elect to set (and get) the
57 * IDENTITY_FRAME_BIT on all identity mapped PFNs.
59 * This simplistic diagram is used to explain the more subtle piece of code.
60 * There is also a digram of the P2M at the end that can help.
61 * Imagine your E820 looking as so:
64 * /-------------------+---------\/----\ /----------\ /---+-----\
65 * | System RAM | Sys RAM ||ACPI| | reserved | | Sys RAM |
66 * \-------------------+---------/\----/ \----------/ \---+-----/
69 * [1029MB = 263424 (0x40500), 2001MB = 512256 (0x7D100),
70 * 2048MB = 524288 (0x80000)]
72 * And dom0_mem=max:3GB,1GB is passed in to the guest, meaning memory past 1GB
73 * is actually not present (would have to kick the balloon driver to put it in).
75 * When we are told to set the PFNs for identity mapping (see patch: "xen/setup:
76 * Set identity mapping for non-RAM E820 and E820 gaps.") we pass in the start
77 * of the PFN and the end PFN (263424 and 512256 respectively). The first step
78 * is to reserve_brk a top leaf page if the p2m[1] is missing. The top leaf page
79 * covers 512^2 of page estate (1GB) and in case the start or end PFN is not
80 * aligned on 512^2*PAGE_SIZE (1GB) we reserve_brk new middle and leaf pages as
81 * required to split any existing p2m_mid_missing middle pages.
83 * With the E820 example above, 263424 is not 1GB aligned so we allocate a
84 * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000.
85 * Each entry in the allocate page is "missing" (points to p2m_missing).
87 * Next stage is to determine if we need to do a more granular boundary check
88 * on the 4MB (or 2MB depending on architecture) off the start and end pfn's.
89 * We check if the start pfn and end pfn violate that boundary check, and if
90 * so reserve_brk a (p2m[x][y]) leaf page. This way we have a much finer
91 * granularity of setting which PFNs are missing and which ones are identity.
92 * In our example 263424 and 512256 both fail the check so we reserve_brk two
93 * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing"
94 * values) and assign them to p2m[1][2] and p2m[1][488] respectively.
96 * At this point we would at minimum reserve_brk one page, but could be up to
97 * three. Each call to set_phys_range_identity has at maximum a three page
98 * cost. If we were to query the P2M at this stage, all those entries from
99 * start PFN through end PFN (so 1029MB -> 2001MB) would return
100 * INVALID_P2M_ENTRY ("missing").
102 * The next step is to walk from the start pfn to the end pfn setting
103 * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity.
104 * If we find that the middle entry is pointing to p2m_missing we can swap it
105 * over to p2m_identity - this way covering 4MB (or 2MB) PFN space (and
106 * similarly swapping p2m_mid_missing for p2m_mid_identity for larger regions).
107 * At this point we do not need to worry about boundary aligment (so no need to
108 * reserve_brk a middle page, figure out which PFNs are "missing" and which
109 * ones are identity), as that has been done earlier. If we find that the
110 * middle leaf is not occupied by p2m_identity or p2m_missing, we dereference
111 * that page (which covers 512 PFNs) and set the appropriate PFN with
112 * IDENTITY_FRAME_BIT. In our example 263424 and 512256 end up there, and we
113 * set from p2m[1][2][256->511] and p2m[1][488][0->256] with
114 * IDENTITY_FRAME_BIT set.
116 * All other regions that are void (or not filled) either point to p2m_missing
117 * (considered missing) or have the default value of INVALID_P2M_ENTRY (also
118 * considered missing). In our case, p2m[1][2][0->255] and p2m[1][488][257->511]
119 * contain the INVALID_P2M_ENTRY value and are considered "missing."
121 * Finally, the region beyond the end of of the E820 (4 GB in this example)
122 * is set to be identity (in case there are MMIO regions placed here).
124 * This is what the p2m ends up looking (for the E820 above) with this
127 * p2m /--------------\
128 * /-----\ | &mfn_list[0],| /-----------------\
129 * | 0 |------>| &mfn_list[1],| /---------------\ | ~0, ~0, .. |
130 * |-----| | ..., ~0, ~0 | | ~0, ~0, [x]---+----->| IDENTITY [@256] |
131 * | 1 |---\ \--------------/ | [p2m_identity]+\ | IDENTITY [@257] |
132 * |-----| \ | [p2m_identity]+\\ | .... |
133 * | 2 |--\ \-------------------->| ... | \\ \----------------/
134 * |-----| \ \---------------/ \\
135 * | 3 |-\ \ \\ p2m_identity [1]
136 * |-----| \ \-------------------->/---------------\ /-----------------\
137 * | .. |\ | | [p2m_identity]+-->| ~0, ~0, ~0, ... |
138 * \-----/ | | | [p2m_identity]+-->| ..., ~0 |
139 * | | | .... | \-----------------/
140 * | | +-[x], ~0, ~0.. +\
141 * | | \---------------/ \
142 * | | \-> /---------------\
143 * | V p2m_mid_missing p2m_missing | IDENTITY[@0] |
144 * | /-----------------\ /------------\ | IDENTITY[@256]|
145 * | | [p2m_missing] +---->| ~0, ~0, ...| | ~0, ~0, .... |
146 * | | [p2m_missing] +---->| ..., ~0 | \---------------/
147 * | | ... | \------------/
148 * | \-----------------/
151 * | /-----------------\
152 * \-->| [p2m_identity] +---->[1]
153 * | [p2m_identity] +---->[1]
155 * \-----------------/
157 * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT)
160 #include <linux/init.h>
161 #include <linux/module.h>
162 #include <linux/list.h>
163 #include <linux/hash.h>
164 #include <linux/sched.h>
165 #include <linux/seq_file.h>
166 #include <linux/bootmem.h>
167 #include <linux/slab.h>
169 #include <asm/cache.h>
170 #include <asm/setup.h>
172 #include <asm/xen/page.h>
173 #include <asm/xen/hypercall.h>
174 #include <asm/xen/hypervisor.h>
175 #include <xen/balloon.h>
176 #include <xen/grant_table.h>
179 #include "multicalls.h"
182 static void __init m2p_override_init(void);
184 unsigned long xen_max_p2m_pfn __read_mostly;
186 static unsigned long *p2m_mid_missing_mfn;
187 static unsigned long *p2m_top_mfn;
188 static unsigned long **p2m_top_mfn_p;
190 /* Placeholders for holes in the address space */
191 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE);
192 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE);
194 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE);
196 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE);
197 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_identity, P2M_MID_PER_PAGE);
199 RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
201 /* For each I/O range remapped we may lose up to two leaf pages for the boundary
202 * violations and three mid pages to cover up to 3GB. With
203 * early_can_reuse_p2m_middle() most of the leaf pages will be reused by the
206 RESERVE_BRK(p2m_identity_remap, PAGE_SIZE * 2 * 3 * MAX_REMAP_RANGES);
208 static int use_brk = 1;
210 static inline unsigned p2m_top_index(unsigned long pfn)
212 BUG_ON(pfn >= MAX_P2M_PFN);
213 return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
216 static inline unsigned p2m_mid_index(unsigned long pfn)
218 return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
221 static inline unsigned p2m_index(unsigned long pfn)
223 return pfn % P2M_PER_PAGE;
226 static void p2m_top_init(unsigned long ***top)
230 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
231 top[i] = p2m_mid_missing;
234 static void p2m_top_mfn_init(unsigned long *top)
238 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
239 top[i] = virt_to_mfn(p2m_mid_missing_mfn);
242 static void p2m_top_mfn_p_init(unsigned long **top)
246 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
247 top[i] = p2m_mid_missing_mfn;
250 static void p2m_mid_init(unsigned long **mid, unsigned long *leaf)
254 for (i = 0; i < P2M_MID_PER_PAGE; i++)
258 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
262 for (i = 0; i < P2M_MID_PER_PAGE; i++)
263 mid[i] = virt_to_mfn(leaf);
266 static void p2m_init(unsigned long *p2m)
270 for (i = 0; i < P2M_MID_PER_PAGE; i++)
271 p2m[i] = INVALID_P2M_ENTRY;
274 static void * __ref alloc_p2m_page(void)
276 if (unlikely(use_brk))
277 return extend_brk(PAGE_SIZE, PAGE_SIZE);
279 if (unlikely(!slab_is_available()))
280 return alloc_bootmem_align(PAGE_SIZE, PAGE_SIZE);
282 return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
285 /* Only to be called in case of a race for a page just allocated! */
286 static void free_p2m_page(void *p)
288 BUG_ON(!slab_is_available());
289 free_page((unsigned long)p);
293 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
295 * This is called both at boot time, and after resuming from suspend:
296 * - At boot time we're called rather early, and must use alloc_bootmem*()
297 * to allocate memory.
299 * - After resume we're called from within stop_machine, but the mfn
300 * tree should already be completely allocated.
302 void __ref xen_build_mfn_list_list(void)
306 if (xen_feature(XENFEAT_auto_translated_physmap))
309 /* Pre-initialize p2m_top_mfn to be completely missing */
310 if (p2m_top_mfn == NULL) {
311 p2m_mid_missing_mfn = alloc_p2m_page();
312 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
314 p2m_top_mfn_p = alloc_p2m_page();
315 p2m_top_mfn_p_init(p2m_top_mfn_p);
317 p2m_top_mfn = alloc_p2m_page();
318 p2m_top_mfn_init(p2m_top_mfn);
320 /* Reinitialise, mfn's all change after migration */
321 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
324 for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) {
325 unsigned topidx = p2m_top_index(pfn);
326 unsigned mididx = p2m_mid_index(pfn);
328 unsigned long *mid_mfn_p;
330 mid = p2m_top[topidx];
331 mid_mfn_p = p2m_top_mfn_p[topidx];
333 /* Don't bother allocating any mfn mid levels if
334 * they're just missing, just update the stored mfn,
335 * since all could have changed over a migrate.
337 if (mid == p2m_mid_missing) {
339 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
340 p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
341 pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
345 if (mid_mfn_p == p2m_mid_missing_mfn) {
347 * XXX boot-time only! We should never find
348 * missing parts of the mfn tree after
351 mid_mfn_p = alloc_p2m_page();
352 p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
354 p2m_top_mfn_p[topidx] = mid_mfn_p;
357 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
358 mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]);
362 void xen_setup_mfn_list_list(void)
364 if (xen_feature(XENFEAT_auto_translated_physmap))
367 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
369 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
370 virt_to_mfn(p2m_top_mfn);
371 HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
374 /* Set up p2m_top to point to the domain-builder provided p2m pages */
375 void __init xen_build_dynamic_phys_to_machine(void)
377 unsigned long *mfn_list;
378 unsigned long max_pfn;
381 if (xen_feature(XENFEAT_auto_translated_physmap))
384 mfn_list = (unsigned long *)xen_start_info->mfn_list;
385 max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
386 xen_max_p2m_pfn = max_pfn;
388 p2m_missing = alloc_p2m_page();
389 p2m_init(p2m_missing);
390 p2m_identity = alloc_p2m_page();
391 p2m_init(p2m_identity);
393 p2m_mid_missing = alloc_p2m_page();
394 p2m_mid_init(p2m_mid_missing, p2m_missing);
395 p2m_mid_identity = alloc_p2m_page();
396 p2m_mid_init(p2m_mid_identity, p2m_identity);
398 p2m_top = alloc_p2m_page();
399 p2m_top_init(p2m_top);
402 * The domain builder gives us a pre-constructed p2m array in
403 * mfn_list for all the pages initially given to us, so we just
404 * need to graft that into our tree structure.
406 for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) {
407 unsigned topidx = p2m_top_index(pfn);
408 unsigned mididx = p2m_mid_index(pfn);
410 if (p2m_top[topidx] == p2m_mid_missing) {
411 unsigned long **mid = alloc_p2m_page();
412 p2m_mid_init(mid, p2m_missing);
414 p2m_top[topidx] = mid;
418 * As long as the mfn_list has enough entries to completely
419 * fill a p2m page, pointing into the array is ok. But if
420 * not the entries beyond the last pfn will be undefined.
422 if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) {
423 unsigned long p2midx;
425 p2midx = max_pfn % P2M_PER_PAGE;
426 for ( ; p2midx < P2M_PER_PAGE; p2midx++)
427 mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY;
429 p2m_top[topidx][mididx] = &mfn_list[pfn];
433 unsigned long __init xen_revector_p2m_tree(void)
435 unsigned long va_start;
436 unsigned long va_end;
438 unsigned long pfn_free = 0;
439 unsigned long *mfn_list = NULL;
443 va_start = xen_start_info->mfn_list;
444 /*We copy in increments of P2M_PER_PAGE * sizeof(unsigned long),
445 * so make sure it is rounded up to that */
446 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
447 va_end = va_start + size;
449 /* If we were revectored already, don't do it again. */
450 if (va_start <= __START_KERNEL_map && va_start >= __PAGE_OFFSET)
453 mfn_list = alloc_bootmem_align(size, PAGE_SIZE);
455 pr_warn("Could not allocate space for a new P2M tree!\n");
456 return xen_start_info->mfn_list;
458 /* Fill it out with INVALID_P2M_ENTRY value */
459 memset(mfn_list, 0xFF, size);
461 for (pfn = 0; pfn < ALIGN(MAX_DOMAIN_PAGES, P2M_PER_PAGE); pfn += P2M_PER_PAGE) {
462 unsigned topidx = p2m_top_index(pfn);
464 unsigned long *mid_p;
466 if (!p2m_top[topidx])
469 if (p2m_top[topidx] == p2m_mid_missing)
472 mididx = p2m_mid_index(pfn);
473 mid_p = p2m_top[topidx][mididx];
476 if ((mid_p == p2m_missing) || (mid_p == p2m_identity))
479 if ((unsigned long)mid_p == INVALID_P2M_ENTRY)
482 /* The old va. Rebase it on mfn_list */
483 if (mid_p >= (unsigned long *)va_start && mid_p <= (unsigned long *)va_end) {
486 if (pfn_free > (size / sizeof(unsigned long))) {
487 WARN(1, "Only allocated for %ld pages, but we want %ld!\n",
488 size / sizeof(unsigned long), pfn_free);
491 new = &mfn_list[pfn_free];
493 copy_page(new, mid_p);
494 p2m_top[topidx][mididx] = &mfn_list[pfn_free];
496 pfn_free += P2M_PER_PAGE;
499 /* This should be the leafs allocated for identity from _brk. */
503 return (unsigned long)mfn_list;
506 unsigned long __init xen_revector_p2m_tree(void)
513 unsigned long get_phys_to_machine(unsigned long pfn)
515 unsigned topidx, mididx, idx;
517 if (unlikely(pfn >= MAX_P2M_PFN))
518 return IDENTITY_FRAME(pfn);
520 topidx = p2m_top_index(pfn);
521 mididx = p2m_mid_index(pfn);
522 idx = p2m_index(pfn);
525 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
526 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
529 if (p2m_top[topidx][mididx] == p2m_identity)
530 return IDENTITY_FRAME(pfn);
532 return p2m_top[topidx][mididx][idx];
534 EXPORT_SYMBOL_GPL(get_phys_to_machine);
537 * Fully allocate the p2m structure for a given pfn. We need to check
538 * that both the top and mid levels are allocated, and make sure the
539 * parallel mfn tree is kept in sync. We may race with other cpus, so
540 * the new pages are installed with cmpxchg; if we lose the race then
541 * simply free the page we allocated and use the one that's there.
543 static bool alloc_p2m(unsigned long pfn)
545 unsigned topidx, mididx;
546 unsigned long ***top_p, **mid;
547 unsigned long *top_mfn_p, *mid_mfn;
548 unsigned long *p2m_orig;
550 topidx = p2m_top_index(pfn);
551 mididx = p2m_mid_index(pfn);
553 top_p = &p2m_top[topidx];
554 mid = ACCESS_ONCE(*top_p);
556 if (mid == p2m_mid_missing) {
557 /* Mid level is missing, allocate a new one */
558 mid = alloc_p2m_page();
562 p2m_mid_init(mid, p2m_missing);
564 if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing)
568 top_mfn_p = &p2m_top_mfn[topidx];
569 mid_mfn = ACCESS_ONCE(p2m_top_mfn_p[topidx]);
571 BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
573 if (mid_mfn == p2m_mid_missing_mfn) {
574 /* Separately check the mid mfn level */
575 unsigned long missing_mfn;
576 unsigned long mid_mfn_mfn;
577 unsigned long old_mfn;
579 mid_mfn = alloc_p2m_page();
583 p2m_mid_mfn_init(mid_mfn, p2m_missing);
585 missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
586 mid_mfn_mfn = virt_to_mfn(mid_mfn);
587 old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
588 if (old_mfn != missing_mfn) {
589 free_p2m_page(mid_mfn);
590 mid_mfn = mfn_to_virt(old_mfn);
592 p2m_top_mfn_p[topidx] = mid_mfn;
596 p2m_orig = ACCESS_ONCE(p2m_top[topidx][mididx]);
597 if (p2m_orig == p2m_identity || p2m_orig == p2m_missing) {
598 /* p2m leaf page is missing */
601 p2m = alloc_p2m_page();
607 if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig)
610 mid_mfn[mididx] = virt_to_mfn(p2m);
616 static bool __init early_alloc_p2m(unsigned long pfn, bool check_boundary)
618 unsigned topidx, mididx, idx;
621 topidx = p2m_top_index(pfn);
622 mididx = p2m_mid_index(pfn);
623 idx = p2m_index(pfn);
625 /* Pfff.. No boundary cross-over, lets get out. */
626 if (!idx && check_boundary)
629 WARN(p2m_top[topidx][mididx] == p2m_identity,
630 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
634 * Could be done by xen_build_dynamic_phys_to_machine..
636 if (p2m_top[topidx][mididx] != p2m_missing)
639 /* Boundary cross-over for the edges: */
640 p2m = alloc_p2m_page();
644 p2m_top[topidx][mididx] = p2m;
649 static bool __init early_alloc_p2m_middle(unsigned long pfn)
651 unsigned topidx = p2m_top_index(pfn);
654 mid = p2m_top[topidx];
655 if (mid == p2m_mid_missing) {
656 mid = alloc_p2m_page();
658 p2m_mid_init(mid, p2m_missing);
660 p2m_top[topidx] = mid;
665 static void __init early_split_p2m(unsigned long pfn)
667 unsigned long mididx, idx;
669 mididx = p2m_mid_index(pfn);
670 idx = p2m_index(pfn);
673 * Allocate new middle and leaf pages if this pfn lies in the
677 early_alloc_p2m_middle(pfn);
679 early_alloc_p2m(pfn, false);
682 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
687 if (unlikely(pfn_s >= MAX_P2M_PFN))
690 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
691 return pfn_e - pfn_s;
696 if (pfn_e > MAX_P2M_PFN)
699 early_split_p2m(pfn_s);
700 early_split_p2m(pfn_e);
702 for (pfn = pfn_s; pfn < pfn_e;) {
703 unsigned topidx = p2m_top_index(pfn);
704 unsigned mididx = p2m_mid_index(pfn);
706 if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
711 * If the PFN was set to a middle or leaf identity
712 * page the remainder must also be identity, so skip
713 * ahead to the next middle or leaf entry.
715 if (p2m_top[topidx] == p2m_mid_identity)
716 pfn = ALIGN(pfn, P2M_MID_PER_PAGE * P2M_PER_PAGE);
717 else if (p2m_top[topidx][mididx] == p2m_identity)
718 pfn = ALIGN(pfn, P2M_PER_PAGE);
721 WARN((pfn - pfn_s) != (pfn_e - pfn_s),
722 "Identity mapping failed. We are %ld short of 1-1 mappings!\n",
723 (pfn_e - pfn_s) - (pfn - pfn_s));
728 /* Try to install p2m mapping; fail if intermediate bits missing */
729 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
731 unsigned topidx, mididx, idx;
733 /* don't track P2M changes in autotranslate guests */
734 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
737 if (unlikely(pfn >= MAX_P2M_PFN)) {
738 BUG_ON(mfn != INVALID_P2M_ENTRY);
742 topidx = p2m_top_index(pfn);
743 mididx = p2m_mid_index(pfn);
744 idx = p2m_index(pfn);
746 /* For sparse holes were the p2m leaf has real PFN along with
747 * PCI holes, stick in the PFN as the MFN value.
749 * set_phys_range_identity() will have allocated new middle
750 * and leaf pages as required so an existing p2m_mid_missing
751 * or p2m_missing mean that whole range will be identity so
752 * these can be switched to p2m_mid_identity or p2m_identity.
754 if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
755 if (p2m_top[topidx] == p2m_mid_identity)
758 if (p2m_top[topidx] == p2m_mid_missing) {
759 WARN_ON(cmpxchg(&p2m_top[topidx], p2m_mid_missing,
760 p2m_mid_identity) != p2m_mid_missing);
764 if (p2m_top[topidx][mididx] == p2m_identity)
767 /* Swap over from MISSING to IDENTITY if needed. */
768 if (p2m_top[topidx][mididx] == p2m_missing) {
769 WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
770 p2m_identity) != p2m_missing);
775 if (p2m_top[topidx][mididx] == p2m_missing)
776 return mfn == INVALID_P2M_ENTRY;
778 p2m_top[topidx][mididx][idx] = mfn;
783 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
785 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
789 if (!__set_phys_to_machine(pfn, mfn))
796 #define M2P_OVERRIDE_HASH_SHIFT 10
797 #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT)
799 static struct list_head *m2p_overrides;
800 static DEFINE_SPINLOCK(m2p_override_lock);
802 static void __init m2p_override_init(void)
806 m2p_overrides = alloc_bootmem_align(
807 sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
808 sizeof(unsigned long));
810 for (i = 0; i < M2P_OVERRIDE_HASH; i++)
811 INIT_LIST_HEAD(&m2p_overrides[i]);
814 static unsigned long mfn_hash(unsigned long mfn)
816 return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
819 /* Add an MFN override for a particular page */
820 static int m2p_add_override(unsigned long mfn, struct page *page,
821 struct gnttab_map_grant_ref *kmap_op)
825 unsigned long uninitialized_var(address);
829 pfn = page_to_pfn(page);
830 if (!PageHighMem(page)) {
831 address = (unsigned long)__va(pfn << PAGE_SHIFT);
832 ptep = lookup_address(address, &level);
833 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
834 "m2p_add_override: pfn %lx not mapped", pfn))
838 if (kmap_op != NULL) {
839 if (!PageHighMem(page)) {
840 struct multicall_space mcs =
841 xen_mc_entry(sizeof(*kmap_op));
843 MULTI_grant_table_op(mcs.mc,
844 GNTTABOP_map_grant_ref, kmap_op, 1);
846 xen_mc_issue(PARAVIRT_LAZY_MMU);
849 spin_lock_irqsave(&m2p_override_lock, flags);
850 list_add(&page->lru, &m2p_overrides[mfn_hash(mfn)]);
851 spin_unlock_irqrestore(&m2p_override_lock, flags);
853 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
854 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
855 * pfn so that the following mfn_to_pfn(mfn) calls will return the
856 * pfn from the m2p_override (the backend pfn) instead.
857 * We need to do this because the pages shared by the frontend
858 * (xen-blkfront) can be already locked (lock_page, called by
859 * do_read_cache_page); when the userspace backend tries to use them
860 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
861 * do_blockdev_direct_IO is going to try to lock the same pages
862 * again resulting in a deadlock.
863 * As a side effect get_user_pages_fast might not be safe on the
864 * frontend pages while they are being shared with the backend,
865 * because mfn_to_pfn (that ends up being called by GUPF) will
866 * return the backend pfn rather than the frontend pfn. */
867 pfn = mfn_to_pfn_no_overrides(mfn);
868 if (get_phys_to_machine(pfn) == mfn)
869 set_phys_to_machine(pfn, FOREIGN_FRAME(mfn));
874 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
875 struct gnttab_map_grant_ref *kmap_ops,
876 struct page **pages, unsigned int count)
882 if (xen_feature(XENFEAT_auto_translated_physmap))
887 paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE) {
888 arch_enter_lazy_mmu_mode();
892 for (i = 0; i < count; i++) {
893 unsigned long mfn, pfn;
895 /* Do not add to override if the map failed. */
896 if (map_ops[i].status)
899 if (map_ops[i].flags & GNTMAP_contains_pte) {
900 pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
901 (map_ops[i].host_addr & ~PAGE_MASK));
904 mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
906 pfn = page_to_pfn(pages[i]);
908 WARN_ON(PagePrivate(pages[i]));
909 SetPagePrivate(pages[i]);
910 set_page_private(pages[i], mfn);
911 pages[i]->index = pfn_to_mfn(pfn);
913 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
919 ret = m2p_add_override(mfn, pages[i], &kmap_ops[i]);
927 arch_leave_lazy_mmu_mode();
931 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
933 static struct page *m2p_find_override(unsigned long mfn)
936 struct list_head *bucket;
937 struct page *p, *ret;
939 if (unlikely(!m2p_overrides))
943 bucket = &m2p_overrides[mfn_hash(mfn)];
945 spin_lock_irqsave(&m2p_override_lock, flags);
947 list_for_each_entry(p, bucket, lru) {
948 if (page_private(p) == mfn) {
954 spin_unlock_irqrestore(&m2p_override_lock, flags);
959 static int m2p_remove_override(struct page *page,
960 struct gnttab_map_grant_ref *kmap_op,
965 unsigned long uninitialized_var(address);
969 pfn = page_to_pfn(page);
971 if (!PageHighMem(page)) {
972 address = (unsigned long)__va(pfn << PAGE_SHIFT);
973 ptep = lookup_address(address, &level);
975 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
976 "m2p_remove_override: pfn %lx not mapped", pfn))
980 spin_lock_irqsave(&m2p_override_lock, flags);
981 list_del(&page->lru);
982 spin_unlock_irqrestore(&m2p_override_lock, flags);
984 if (kmap_op != NULL) {
985 if (!PageHighMem(page)) {
986 struct multicall_space mcs;
987 struct gnttab_unmap_and_replace *unmap_op;
988 struct page *scratch_page = get_balloon_scratch_page();
989 unsigned long scratch_page_address = (unsigned long)
990 __va(page_to_pfn(scratch_page) << PAGE_SHIFT);
993 * It might be that we queued all the m2p grant table
994 * hypercalls in a multicall, then m2p_remove_override
995 * get called before the multicall has actually been
996 * issued. In this case handle is going to -1 because
997 * it hasn't been modified yet.
999 if (kmap_op->handle == -1)
1002 * Now if kmap_op->handle is negative it means that the
1003 * hypercall actually returned an error.
1005 if (kmap_op->handle == GNTST_general_error) {
1006 pr_warn("m2p_remove_override: pfn %lx mfn %lx, failed to modify kernel mappings",
1008 put_balloon_scratch_page();
1014 mcs = __xen_mc_entry(
1015 sizeof(struct gnttab_unmap_and_replace));
1016 unmap_op = mcs.args;
1017 unmap_op->host_addr = kmap_op->host_addr;
1018 unmap_op->new_addr = scratch_page_address;
1019 unmap_op->handle = kmap_op->handle;
1021 MULTI_grant_table_op(mcs.mc,
1022 GNTTABOP_unmap_and_replace, unmap_op, 1);
1024 mcs = __xen_mc_entry(0);
1025 MULTI_update_va_mapping(mcs.mc, scratch_page_address,
1026 pfn_pte(page_to_pfn(scratch_page),
1027 PAGE_KERNEL_RO), 0);
1029 xen_mc_issue(PARAVIRT_LAZY_MMU);
1031 kmap_op->host_addr = 0;
1032 put_balloon_scratch_page();
1036 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
1037 * somewhere in this domain, even before being added to the
1038 * m2p_override (see comment above in m2p_add_override).
1039 * If there are no other entries in the m2p_override corresponding
1040 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
1041 * the original pfn (the one shared by the frontend): the backend
1042 * cannot do any IO on this page anymore because it has been
1043 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
1044 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
1046 mfn &= ~FOREIGN_FRAME_BIT;
1047 pfn = mfn_to_pfn_no_overrides(mfn);
1048 if (get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) &&
1049 m2p_find_override(mfn) == NULL)
1050 set_phys_to_machine(pfn, mfn);
1055 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
1056 struct gnttab_map_grant_ref *kmap_ops,
1057 struct page **pages, unsigned int count)
1062 if (xen_feature(XENFEAT_auto_translated_physmap))
1067 paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE) {
1068 arch_enter_lazy_mmu_mode();
1072 for (i = 0; i < count; i++) {
1073 unsigned long mfn = get_phys_to_machine(page_to_pfn(pages[i]));
1074 unsigned long pfn = page_to_pfn(pages[i]);
1076 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
1081 set_page_private(pages[i], INVALID_P2M_ENTRY);
1082 WARN_ON(!PagePrivate(pages[i]));
1083 ClearPagePrivate(pages[i]);
1084 set_phys_to_machine(pfn, pages[i]->index);
1087 ret = m2p_remove_override(pages[i], &kmap_ops[i], mfn);
1094 arch_leave_lazy_mmu_mode();
1097 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
1099 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
1101 struct page *p = m2p_find_override(mfn);
1102 unsigned long ret = pfn;
1105 ret = page_to_pfn(p);
1109 EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
1111 #ifdef CONFIG_XEN_DEBUG_FS
1112 #include <linux/debugfs.h>
1113 #include "debugfs.h"
1114 static int p2m_dump_show(struct seq_file *m, void *v)
1116 static const char * const level_name[] = { "top", "middle",
1117 "entry", "abnormal", "error"};
1118 #define TYPE_IDENTITY 0
1119 #define TYPE_MISSING 1
1121 #define TYPE_UNKNOWN 3
1122 static const char * const type_name[] = {
1123 [TYPE_IDENTITY] = "identity",
1124 [TYPE_MISSING] = "missing",
1126 [TYPE_UNKNOWN] = "abnormal"};
1127 unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
1128 unsigned int uninitialized_var(prev_level);
1129 unsigned int uninitialized_var(prev_type);
1134 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
1135 unsigned topidx = p2m_top_index(pfn);
1136 unsigned mididx = p2m_mid_index(pfn);
1137 unsigned idx = p2m_index(pfn);
1141 type = TYPE_UNKNOWN;
1142 if (p2m_top[topidx] == p2m_mid_missing) {
1143 lvl = 0; type = TYPE_MISSING;
1144 } else if (p2m_top[topidx] == NULL) {
1145 lvl = 0; type = TYPE_UNKNOWN;
1146 } else if (p2m_top[topidx][mididx] == NULL) {
1147 lvl = 1; type = TYPE_UNKNOWN;
1148 } else if (p2m_top[topidx][mididx] == p2m_identity) {
1149 lvl = 1; type = TYPE_IDENTITY;
1150 } else if (p2m_top[topidx][mididx] == p2m_missing) {
1151 lvl = 1; type = TYPE_MISSING;
1152 } else if (p2m_top[topidx][mididx][idx] == 0) {
1153 lvl = 2; type = TYPE_UNKNOWN;
1154 } else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
1155 lvl = 2; type = TYPE_IDENTITY;
1156 } else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
1157 lvl = 2; type = TYPE_MISSING;
1158 } else if (p2m_top[topidx][mididx][idx] == pfn) {
1159 lvl = 2; type = TYPE_PFN;
1160 } else if (p2m_top[topidx][mididx][idx] != pfn) {
1161 lvl = 2; type = TYPE_PFN;
1167 if (pfn == MAX_DOMAIN_PAGES-1) {
1169 type = TYPE_UNKNOWN;
1171 if (prev_type != type) {
1172 seq_printf(m, " [0x%lx->0x%lx] %s\n",
1173 prev_pfn_type, pfn, type_name[prev_type]);
1174 prev_pfn_type = pfn;
1177 if (prev_level != lvl) {
1178 seq_printf(m, " [0x%lx->0x%lx] level %s\n",
1179 prev_pfn_level, pfn, level_name[prev_level]);
1180 prev_pfn_level = pfn;
1185 #undef TYPE_IDENTITY
1191 static int p2m_dump_open(struct inode *inode, struct file *filp)
1193 return single_open(filp, p2m_dump_show, NULL);
1196 static const struct file_operations p2m_dump_fops = {
1197 .open = p2m_dump_open,
1199 .llseek = seq_lseek,
1200 .release = single_release,
1203 static struct dentry *d_mmu_debug;
1205 static int __init xen_p2m_debugfs(void)
1207 struct dentry *d_xen = xen_init_debugfs();
1212 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1214 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
1217 fs_initcall(xen_p2m_debugfs);
1218 #endif /* CONFIG_XEN_DEBUG_FS */