Loading...
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/kstrtox.h>
11#include <linux/mm.h>
12#include <linux/pm.h>
13#include <linux/memblock.h>
14#include <linux/cpuidle.h>
15#include <linux/cpufreq.h>
16#include <linux/memory_hotplug.h>
17
18#include <asm/elf.h>
19#include <asm/vdso.h>
20#include <asm/e820/api.h>
21#include <asm/setup.h>
22#include <asm/acpi.h>
23#include <asm/numa.h>
24#include <asm/idtentry.h>
25#include <asm/xen/hypervisor.h>
26#include <asm/xen/hypercall.h>
27
28#include <xen/xen.h>
29#include <xen/page.h>
30#include <xen/interface/callback.h>
31#include <xen/interface/memory.h>
32#include <xen/interface/physdev.h>
33#include <xen/features.h>
34#include <xen/hvc-console.h>
35#include "xen-ops.h"
36#include "mmu.h"
37
38#define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
39
40/* Amount of extra memory space we add to the e820 ranges */
41struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
42
43/* Number of pages released from the initial allocation. */
44unsigned long xen_released_pages;
45
46/* E820 map used during setting up memory. */
47static struct e820_table xen_e820_table __initdata;
48
49/*
50 * Buffer used to remap identity mapped pages. We only need the virtual space.
51 * The physical page behind this address is remapped as needed to different
52 * buffer pages.
53 */
54#define REMAP_SIZE (P2M_PER_PAGE - 3)
55static struct {
56 unsigned long next_area_mfn;
57 unsigned long target_pfn;
58 unsigned long size;
59 unsigned long mfns[REMAP_SIZE];
60} xen_remap_buf __initdata __aligned(PAGE_SIZE);
61static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
62
63/*
64 * The maximum amount of extra memory compared to the base size. The
65 * main scaling factor is the size of struct page. At extreme ratios
66 * of base:extra, all the base memory can be filled with page
67 * structures for the extra memory, leaving no space for anything
68 * else.
69 *
70 * 10x seems like a reasonable balance between scaling flexibility and
71 * leaving a practically usable system.
72 */
73#define EXTRA_MEM_RATIO (10)
74
75static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
76
77static void __init xen_parse_512gb(void)
78{
79 bool val = false;
80 char *arg;
81
82 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
83 if (!arg)
84 return;
85
86 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
87 if (!arg)
88 val = true;
89 else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
90 return;
91
92 xen_512gb_limit = val;
93}
94
95static void __init xen_add_extra_mem(unsigned long start_pfn,
96 unsigned long n_pfns)
97{
98 int i;
99
100 /*
101 * No need to check for zero size, should happen rarely and will only
102 * write a new entry regarded to be unused due to zero size.
103 */
104 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
105 /* Add new region. */
106 if (xen_extra_mem[i].n_pfns == 0) {
107 xen_extra_mem[i].start_pfn = start_pfn;
108 xen_extra_mem[i].n_pfns = n_pfns;
109 break;
110 }
111 /* Append to existing region. */
112 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
113 start_pfn) {
114 xen_extra_mem[i].n_pfns += n_pfns;
115 break;
116 }
117 }
118 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
119 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
120
121 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
122}
123
124static void __init xen_del_extra_mem(unsigned long start_pfn,
125 unsigned long n_pfns)
126{
127 int i;
128 unsigned long start_r, size_r;
129
130 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
131 start_r = xen_extra_mem[i].start_pfn;
132 size_r = xen_extra_mem[i].n_pfns;
133
134 /* Start of region. */
135 if (start_r == start_pfn) {
136 BUG_ON(n_pfns > size_r);
137 xen_extra_mem[i].start_pfn += n_pfns;
138 xen_extra_mem[i].n_pfns -= n_pfns;
139 break;
140 }
141 /* End of region. */
142 if (start_r + size_r == start_pfn + n_pfns) {
143 BUG_ON(n_pfns > size_r);
144 xen_extra_mem[i].n_pfns -= n_pfns;
145 break;
146 }
147 /* Mid of region. */
148 if (start_pfn > start_r && start_pfn < start_r + size_r) {
149 BUG_ON(start_pfn + n_pfns > start_r + size_r);
150 xen_extra_mem[i].n_pfns = start_pfn - start_r;
151 /* Calling memblock_reserve() again is okay. */
152 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
153 (start_pfn + n_pfns));
154 break;
155 }
156 }
157 memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
158}
159
160/*
161 * Called during boot before the p2m list can take entries beyond the
162 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
163 * invalid.
164 */
165unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
166{
167 int i;
168
169 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
170 if (pfn >= xen_extra_mem[i].start_pfn &&
171 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
172 return INVALID_P2M_ENTRY;
173 }
174
175 return IDENTITY_FRAME(pfn);
176}
177
178/*
179 * Mark all pfns of extra mem as invalid in p2m list.
180 */
181void __init xen_inv_extra_mem(void)
182{
183 unsigned long pfn, pfn_s, pfn_e;
184 int i;
185
186 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
187 if (!xen_extra_mem[i].n_pfns)
188 continue;
189 pfn_s = xen_extra_mem[i].start_pfn;
190 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
191 for (pfn = pfn_s; pfn < pfn_e; pfn++)
192 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
193 }
194}
195
196/*
197 * Finds the next RAM pfn available in the E820 map after min_pfn.
198 * This function updates min_pfn with the pfn found and returns
199 * the size of that range or zero if not found.
200 */
201static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
202{
203 const struct e820_entry *entry = xen_e820_table.entries;
204 unsigned int i;
205 unsigned long done = 0;
206
207 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
208 unsigned long s_pfn;
209 unsigned long e_pfn;
210
211 if (entry->type != E820_TYPE_RAM)
212 continue;
213
214 e_pfn = PFN_DOWN(entry->addr + entry->size);
215
216 /* We only care about E820 after this */
217 if (e_pfn <= *min_pfn)
218 continue;
219
220 s_pfn = PFN_UP(entry->addr);
221
222 /* If min_pfn falls within the E820 entry, we want to start
223 * at the min_pfn PFN.
224 */
225 if (s_pfn <= *min_pfn) {
226 done = e_pfn - *min_pfn;
227 } else {
228 done = e_pfn - s_pfn;
229 *min_pfn = s_pfn;
230 }
231 break;
232 }
233
234 return done;
235}
236
237static int __init xen_free_mfn(unsigned long mfn)
238{
239 struct xen_memory_reservation reservation = {
240 .address_bits = 0,
241 .extent_order = 0,
242 .domid = DOMID_SELF
243 };
244
245 set_xen_guest_handle(reservation.extent_start, &mfn);
246 reservation.nr_extents = 1;
247
248 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
249}
250
251/*
252 * This releases a chunk of memory and then does the identity map. It's used
253 * as a fallback if the remapping fails.
254 */
255static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
256 unsigned long end_pfn, unsigned long nr_pages)
257{
258 unsigned long pfn, end;
259 int ret;
260
261 WARN_ON(start_pfn > end_pfn);
262
263 /* Release pages first. */
264 end = min(end_pfn, nr_pages);
265 for (pfn = start_pfn; pfn < end; pfn++) {
266 unsigned long mfn = pfn_to_mfn(pfn);
267
268 /* Make sure pfn exists to start with */
269 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
270 continue;
271
272 ret = xen_free_mfn(mfn);
273 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
274
275 if (ret == 1) {
276 xen_released_pages++;
277 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
278 break;
279 } else
280 break;
281 }
282
283 set_phys_range_identity(start_pfn, end_pfn);
284}
285
286/*
287 * Helper function to update the p2m and m2p tables and kernel mapping.
288 */
289static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
290{
291 struct mmu_update update = {
292 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
293 .val = pfn
294 };
295
296 /* Update p2m */
297 if (!set_phys_to_machine(pfn, mfn)) {
298 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
299 pfn, mfn);
300 BUG();
301 }
302
303 /* Update m2p */
304 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
305 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
306 mfn, pfn);
307 BUG();
308 }
309
310 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
311 mfn_pte(mfn, PAGE_KERNEL), 0)) {
312 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
313 mfn, pfn);
314 BUG();
315 }
316}
317
318/*
319 * This function updates the p2m and m2p tables with an identity map from
320 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
321 * original allocation at remap_pfn. The information needed for remapping is
322 * saved in the memory itself to avoid the need for allocating buffers. The
323 * complete remap information is contained in a list of MFNs each containing
324 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
325 * This enables us to preserve the original mfn sequence while doing the
326 * remapping at a time when the memory management is capable of allocating
327 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
328 * its callers.
329 */
330static void __init xen_do_set_identity_and_remap_chunk(
331 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
332{
333 unsigned long buf = (unsigned long)&xen_remap_buf;
334 unsigned long mfn_save, mfn;
335 unsigned long ident_pfn_iter, remap_pfn_iter;
336 unsigned long ident_end_pfn = start_pfn + size;
337 unsigned long left = size;
338 unsigned int i, chunk;
339
340 WARN_ON(size == 0);
341
342 mfn_save = virt_to_mfn(buf);
343
344 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
345 ident_pfn_iter < ident_end_pfn;
346 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
347 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
348
349 /* Map first pfn to xen_remap_buf */
350 mfn = pfn_to_mfn(ident_pfn_iter);
351 set_pte_mfn(buf, mfn, PAGE_KERNEL);
352
353 /* Save mapping information in page */
354 xen_remap_buf.next_area_mfn = xen_remap_mfn;
355 xen_remap_buf.target_pfn = remap_pfn_iter;
356 xen_remap_buf.size = chunk;
357 for (i = 0; i < chunk; i++)
358 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
359
360 /* Put remap buf into list. */
361 xen_remap_mfn = mfn;
362
363 /* Set identity map */
364 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
365
366 left -= chunk;
367 }
368
369 /* Restore old xen_remap_buf mapping */
370 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
371}
372
373/*
374 * This function takes a contiguous pfn range that needs to be identity mapped
375 * and:
376 *
377 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
378 * 2) Calls the do_ function to actually do the mapping/remapping work.
379 *
380 * The goal is to not allocate additional memory but to remap the existing
381 * pages. In the case of an error the underlying memory is simply released back
382 * to Xen and not remapped.
383 */
384static unsigned long __init xen_set_identity_and_remap_chunk(
385 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
386 unsigned long remap_pfn)
387{
388 unsigned long pfn;
389 unsigned long i = 0;
390 unsigned long n = end_pfn - start_pfn;
391
392 if (remap_pfn == 0)
393 remap_pfn = nr_pages;
394
395 while (i < n) {
396 unsigned long cur_pfn = start_pfn + i;
397 unsigned long left = n - i;
398 unsigned long size = left;
399 unsigned long remap_range_size;
400
401 /* Do not remap pages beyond the current allocation */
402 if (cur_pfn >= nr_pages) {
403 /* Identity map remaining pages */
404 set_phys_range_identity(cur_pfn, cur_pfn + size);
405 break;
406 }
407 if (cur_pfn + size > nr_pages)
408 size = nr_pages - cur_pfn;
409
410 remap_range_size = xen_find_pfn_range(&remap_pfn);
411 if (!remap_range_size) {
412 pr_warn("Unable to find available pfn range, not remapping identity pages\n");
413 xen_set_identity_and_release_chunk(cur_pfn,
414 cur_pfn + left, nr_pages);
415 break;
416 }
417 /* Adjust size to fit in current e820 RAM region */
418 if (size > remap_range_size)
419 size = remap_range_size;
420
421 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
422
423 /* Update variables to reflect new mappings. */
424 i += size;
425 remap_pfn += size;
426 }
427
428 /*
429 * If the PFNs are currently mapped, their VA mappings need to be
430 * zapped.
431 */
432 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
433 (void)HYPERVISOR_update_va_mapping(
434 (unsigned long)__va(pfn << PAGE_SHIFT),
435 native_make_pte(0), 0);
436
437 return remap_pfn;
438}
439
440static unsigned long __init xen_count_remap_pages(
441 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
442 unsigned long remap_pages)
443{
444 if (start_pfn >= nr_pages)
445 return remap_pages;
446
447 return remap_pages + min(end_pfn, nr_pages) - start_pfn;
448}
449
450static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
451 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
452 unsigned long nr_pages, unsigned long last_val))
453{
454 phys_addr_t start = 0;
455 unsigned long ret_val = 0;
456 const struct e820_entry *entry = xen_e820_table.entries;
457 int i;
458
459 /*
460 * Combine non-RAM regions and gaps until a RAM region (or the
461 * end of the map) is reached, then call the provided function
462 * to perform its duty on the non-RAM region.
463 *
464 * The combined non-RAM regions are rounded to a whole number
465 * of pages so any partial pages are accessible via the 1:1
466 * mapping. This is needed for some BIOSes that put (for
467 * example) the DMI tables in a reserved region that begins on
468 * a non-page boundary.
469 */
470 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
471 phys_addr_t end = entry->addr + entry->size;
472 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
473 unsigned long start_pfn = PFN_DOWN(start);
474 unsigned long end_pfn = PFN_UP(end);
475
476 if (entry->type == E820_TYPE_RAM)
477 end_pfn = PFN_UP(entry->addr);
478
479 if (start_pfn < end_pfn)
480 ret_val = func(start_pfn, end_pfn, nr_pages,
481 ret_val);
482 start = end;
483 }
484 }
485
486 return ret_val;
487}
488
489/*
490 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
491 * The remap information (which mfn remap to which pfn) is contained in the
492 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
493 * This scheme allows to remap the different chunks in arbitrary order while
494 * the resulting mapping will be independent from the order.
495 */
496void __init xen_remap_memory(void)
497{
498 unsigned long buf = (unsigned long)&xen_remap_buf;
499 unsigned long mfn_save, pfn;
500 unsigned long remapped = 0;
501 unsigned int i;
502 unsigned long pfn_s = ~0UL;
503 unsigned long len = 0;
504
505 mfn_save = virt_to_mfn(buf);
506
507 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
508 /* Map the remap information */
509 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
510
511 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
512
513 pfn = xen_remap_buf.target_pfn;
514 for (i = 0; i < xen_remap_buf.size; i++) {
515 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
516 remapped++;
517 pfn++;
518 }
519 if (pfn_s == ~0UL || pfn == pfn_s) {
520 pfn_s = xen_remap_buf.target_pfn;
521 len += xen_remap_buf.size;
522 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
523 len += xen_remap_buf.size;
524 } else {
525 xen_del_extra_mem(pfn_s, len);
526 pfn_s = xen_remap_buf.target_pfn;
527 len = xen_remap_buf.size;
528 }
529 xen_remap_mfn = xen_remap_buf.next_area_mfn;
530 }
531
532 if (pfn_s != ~0UL && len)
533 xen_del_extra_mem(pfn_s, len);
534
535 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
536
537 pr_info("Remapped %ld page(s)\n", remapped);
538}
539
540static unsigned long __init xen_get_pages_limit(void)
541{
542 unsigned long limit;
543
544 limit = MAXMEM / PAGE_SIZE;
545 if (!xen_initial_domain() && xen_512gb_limit)
546 limit = GB(512) / PAGE_SIZE;
547
548 return limit;
549}
550
551static unsigned long __init xen_get_max_pages(void)
552{
553 unsigned long max_pages, limit;
554 domid_t domid = DOMID_SELF;
555 long ret;
556
557 limit = xen_get_pages_limit();
558 max_pages = limit;
559
560 /*
561 * For the initial domain we use the maximum reservation as
562 * the maximum page.
563 *
564 * For guest domains the current maximum reservation reflects
565 * the current maximum rather than the static maximum. In this
566 * case the e820 map provided to us will cover the static
567 * maximum region.
568 */
569 if (xen_initial_domain()) {
570 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
571 if (ret > 0)
572 max_pages = ret;
573 }
574
575 return min(max_pages, limit);
576}
577
578static void __init xen_align_and_add_e820_region(phys_addr_t start,
579 phys_addr_t size, int type)
580{
581 phys_addr_t end = start + size;
582
583 /* Align RAM regions to page boundaries. */
584 if (type == E820_TYPE_RAM) {
585 start = PAGE_ALIGN(start);
586 end &= ~((phys_addr_t)PAGE_SIZE - 1);
587#ifdef CONFIG_MEMORY_HOTPLUG
588 /*
589 * Don't allow adding memory not in E820 map while booting the
590 * system. Once the balloon driver is up it will remove that
591 * restriction again.
592 */
593 max_mem_size = end;
594#endif
595 }
596
597 e820__range_add(start, end - start, type);
598}
599
600static void __init xen_ignore_unusable(void)
601{
602 struct e820_entry *entry = xen_e820_table.entries;
603 unsigned int i;
604
605 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
606 if (entry->type == E820_TYPE_UNUSABLE)
607 entry->type = E820_TYPE_RAM;
608 }
609}
610
611bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
612{
613 struct e820_entry *entry;
614 unsigned mapcnt;
615 phys_addr_t end;
616
617 if (!size)
618 return false;
619
620 end = start + size;
621 entry = xen_e820_table.entries;
622
623 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
624 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
625 (entry->addr + entry->size) >= end)
626 return false;
627
628 entry++;
629 }
630
631 return true;
632}
633
634/*
635 * Find a free area in physical memory not yet reserved and compliant with
636 * E820 map.
637 * Used to relocate pre-allocated areas like initrd or p2m list which are in
638 * conflict with the to be used E820 map.
639 * In case no area is found, return 0. Otherwise return the physical address
640 * of the area which is already reserved for convenience.
641 */
642phys_addr_t __init xen_find_free_area(phys_addr_t size)
643{
644 unsigned mapcnt;
645 phys_addr_t addr, start;
646 struct e820_entry *entry = xen_e820_table.entries;
647
648 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
649 if (entry->type != E820_TYPE_RAM || entry->size < size)
650 continue;
651 start = entry->addr;
652 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
653 if (!memblock_is_reserved(addr))
654 continue;
655 start = addr + PAGE_SIZE;
656 if (start + size > entry->addr + entry->size)
657 break;
658 }
659 if (addr >= start + size) {
660 memblock_reserve(start, size);
661 return start;
662 }
663 }
664
665 return 0;
666}
667
668/*
669 * Like memcpy, but with physical addresses for dest and src.
670 */
671static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
672 phys_addr_t n)
673{
674 phys_addr_t dest_off, src_off, dest_len, src_len, len;
675 void *from, *to;
676
677 while (n) {
678 dest_off = dest & ~PAGE_MASK;
679 src_off = src & ~PAGE_MASK;
680 dest_len = n;
681 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
682 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
683 src_len = n;
684 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
685 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
686 len = min(dest_len, src_len);
687 to = early_memremap(dest - dest_off, dest_len + dest_off);
688 from = early_memremap(src - src_off, src_len + src_off);
689 memcpy(to, from, len);
690 early_memunmap(to, dest_len + dest_off);
691 early_memunmap(from, src_len + src_off);
692 n -= len;
693 dest += len;
694 src += len;
695 }
696}
697
698/*
699 * Reserve Xen mfn_list.
700 */
701static void __init xen_reserve_xen_mfnlist(void)
702{
703 phys_addr_t start, size;
704
705 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
706 start = __pa(xen_start_info->mfn_list);
707 size = PFN_ALIGN(xen_start_info->nr_pages *
708 sizeof(unsigned long));
709 } else {
710 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
711 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
712 }
713
714 memblock_reserve(start, size);
715 if (!xen_is_e820_reserved(start, size))
716 return;
717
718 xen_relocate_p2m();
719 memblock_phys_free(start, size);
720}
721
722/**
723 * xen_memory_setup - Hook for machine specific memory setup.
724 **/
725char * __init xen_memory_setup(void)
726{
727 unsigned long max_pfn, pfn_s, n_pfns;
728 phys_addr_t mem_end, addr, size, chunk_size;
729 u32 type;
730 int rc;
731 struct xen_memory_map memmap;
732 unsigned long max_pages;
733 unsigned long extra_pages = 0;
734 int i;
735 int op;
736
737 xen_parse_512gb();
738 max_pfn = xen_get_pages_limit();
739 max_pfn = min(max_pfn, xen_start_info->nr_pages);
740 mem_end = PFN_PHYS(max_pfn);
741
742 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
743 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
744
745#if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
746 xen_saved_max_mem_size = max_mem_size;
747#endif
748
749 op = xen_initial_domain() ?
750 XENMEM_machine_memory_map :
751 XENMEM_memory_map;
752 rc = HYPERVISOR_memory_op(op, &memmap);
753 if (rc == -ENOSYS) {
754 BUG_ON(xen_initial_domain());
755 memmap.nr_entries = 1;
756 xen_e820_table.entries[0].addr = 0ULL;
757 xen_e820_table.entries[0].size = mem_end;
758 /* 8MB slack (to balance backend allocations). */
759 xen_e820_table.entries[0].size += 8ULL << 20;
760 xen_e820_table.entries[0].type = E820_TYPE_RAM;
761 rc = 0;
762 }
763 BUG_ON(rc);
764 BUG_ON(memmap.nr_entries == 0);
765 xen_e820_table.nr_entries = memmap.nr_entries;
766
767 /*
768 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
769 * regions, so if we're using the machine memory map leave the
770 * region as RAM as it is in the pseudo-physical map.
771 *
772 * UNUSABLE regions in domUs are not handled and will need
773 * a patch in the future.
774 */
775 if (xen_initial_domain())
776 xen_ignore_unusable();
777
778 /* Make sure the Xen-supplied memory map is well-ordered. */
779 e820__update_table(&xen_e820_table);
780
781 max_pages = xen_get_max_pages();
782
783 /* How many extra pages do we need due to remapping? */
784 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
785
786 if (max_pages > max_pfn)
787 extra_pages += max_pages - max_pfn;
788
789 /*
790 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
791 * factor the base size.
792 *
793 * Make sure we have no memory above max_pages, as this area
794 * isn't handled by the p2m management.
795 */
796 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
797 extra_pages, max_pages - max_pfn);
798 i = 0;
799 addr = xen_e820_table.entries[0].addr;
800 size = xen_e820_table.entries[0].size;
801 while (i < xen_e820_table.nr_entries) {
802 bool discard = false;
803
804 chunk_size = size;
805 type = xen_e820_table.entries[i].type;
806
807 if (type == E820_TYPE_RAM) {
808 if (addr < mem_end) {
809 chunk_size = min(size, mem_end - addr);
810 } else if (extra_pages) {
811 chunk_size = min(size, PFN_PHYS(extra_pages));
812 pfn_s = PFN_UP(addr);
813 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
814 extra_pages -= n_pfns;
815 xen_add_extra_mem(pfn_s, n_pfns);
816 xen_max_p2m_pfn = pfn_s + n_pfns;
817 } else
818 discard = true;
819 }
820
821 if (!discard)
822 xen_align_and_add_e820_region(addr, chunk_size, type);
823
824 addr += chunk_size;
825 size -= chunk_size;
826 if (size == 0) {
827 i++;
828 if (i < xen_e820_table.nr_entries) {
829 addr = xen_e820_table.entries[i].addr;
830 size = xen_e820_table.entries[i].size;
831 }
832 }
833 }
834
835 /*
836 * Set the rest as identity mapped, in case PCI BARs are
837 * located here.
838 */
839 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
840
841 /*
842 * In domU, the ISA region is normal, usable memory, but we
843 * reserve ISA memory anyway because too many things poke
844 * about in there.
845 */
846 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
847
848 e820__update_table(e820_table);
849
850 /*
851 * Check whether the kernel itself conflicts with the target E820 map.
852 * Failing now is better than running into weird problems later due
853 * to relocating (and even reusing) pages with kernel text or data.
854 */
855 if (xen_is_e820_reserved(__pa_symbol(_text),
856 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
857 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
858 BUG();
859 }
860
861 /*
862 * Check for a conflict of the hypervisor supplied page tables with
863 * the target E820 map.
864 */
865 xen_pt_check_e820();
866
867 xen_reserve_xen_mfnlist();
868
869 /* Check for a conflict of the initrd with the target E820 map. */
870 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
871 boot_params.hdr.ramdisk_size)) {
872 phys_addr_t new_area, start, size;
873
874 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
875 if (!new_area) {
876 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
877 BUG();
878 }
879
880 start = boot_params.hdr.ramdisk_image;
881 size = boot_params.hdr.ramdisk_size;
882 xen_phys_memcpy(new_area, start, size);
883 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
884 start, start + size, new_area, new_area + size);
885 memblock_phys_free(start, size);
886 boot_params.hdr.ramdisk_image = new_area;
887 boot_params.ext_ramdisk_image = new_area >> 32;
888 }
889
890 /*
891 * Set identity map on non-RAM pages and prepare remapping the
892 * underlying RAM.
893 */
894 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
895
896 pr_info("Released %ld page(s)\n", xen_released_pages);
897
898 return "Xen";
899}
900
901static int register_callback(unsigned type, const void *func)
902{
903 struct callback_register callback = {
904 .type = type,
905 .address = XEN_CALLBACK(__KERNEL_CS, func),
906 .flags = CALLBACKF_mask_events,
907 };
908
909 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
910}
911
912void xen_enable_sysenter(void)
913{
914 if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
915 register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
916 setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
917}
918
919void xen_enable_syscall(void)
920{
921 int ret;
922
923 ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
924 if (ret != 0) {
925 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
926 /* Pretty fatal; 64-bit userspace has no other
927 mechanism for syscalls. */
928 }
929
930 if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
931 register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
932 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
933}
934
935static void __init xen_pvmmu_arch_setup(void)
936{
937 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
938 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
939
940 HYPERVISOR_vm_assist(VMASST_CMD_enable,
941 VMASST_TYPE_pae_extended_cr3);
942
943 if (register_callback(CALLBACKTYPE_event,
944 xen_asm_exc_xen_hypervisor_callback) ||
945 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
946 BUG();
947
948 xen_enable_sysenter();
949 xen_enable_syscall();
950}
951
952/* This function is not called for HVM domains */
953void __init xen_arch_setup(void)
954{
955 xen_panic_handler_init();
956 xen_pvmmu_arch_setup();
957
958#ifdef CONFIG_ACPI
959 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
960 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
961 disable_acpi();
962 }
963#endif
964
965 memcpy(boot_command_line, xen_start_info->cmd_line,
966 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
967 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
968
969 /* Set up idle, making sure it calls safe_halt() pvop */
970 disable_cpuidle();
971 disable_cpufreq();
972 WARN_ON(xen_set_default_idle());
973#ifdef CONFIG_NUMA
974 numa_off = 1;
975#endif
976}
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
16#include <asm/elf.h>
17#include <asm/vdso.h>
18#include <asm/e820/api.h>
19#include <asm/setup.h>
20#include <asm/acpi.h>
21#include <asm/numa.h>
22#include <asm/xen/hypervisor.h>
23#include <asm/xen/hypercall.h>
24
25#include <xen/xen.h>
26#include <xen/page.h>
27#include <xen/interface/callback.h>
28#include <xen/interface/memory.h>
29#include <xen/interface/physdev.h>
30#include <xen/features.h>
31#include <xen/hvc-console.h>
32#include "xen-ops.h"
33#include "vdso.h"
34#include "mmu.h"
35
36#define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
37
38/* Amount of extra memory space we add to the e820 ranges */
39struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
40
41/* Number of pages released from the initial allocation. */
42unsigned long xen_released_pages;
43
44/* E820 map used during setting up memory. */
45static struct e820_table xen_e820_table __initdata;
46
47/*
48 * Buffer used to remap identity mapped pages. We only need the virtual space.
49 * The physical page behind this address is remapped as needed to different
50 * buffer pages.
51 */
52#define REMAP_SIZE (P2M_PER_PAGE - 3)
53static struct {
54 unsigned long next_area_mfn;
55 unsigned long target_pfn;
56 unsigned long size;
57 unsigned long mfns[REMAP_SIZE];
58} xen_remap_buf __initdata __aligned(PAGE_SIZE);
59static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
60
61/*
62 * The maximum amount of extra memory compared to the base size. The
63 * main scaling factor is the size of struct page. At extreme ratios
64 * of base:extra, all the base memory can be filled with page
65 * structures for the extra memory, leaving no space for anything
66 * else.
67 *
68 * 10x seems like a reasonable balance between scaling flexibility and
69 * leaving a practically usable system.
70 */
71#define EXTRA_MEM_RATIO (10)
72
73static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
74
75static void __init xen_parse_512gb(void)
76{
77 bool val = false;
78 char *arg;
79
80 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
81 if (!arg)
82 return;
83
84 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
85 if (!arg)
86 val = true;
87 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
88 return;
89
90 xen_512gb_limit = val;
91}
92
93static void __init xen_add_extra_mem(unsigned long start_pfn,
94 unsigned long n_pfns)
95{
96 int i;
97
98 /*
99 * No need to check for zero size, should happen rarely and will only
100 * write a new entry regarded to be unused due to zero size.
101 */
102 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
103 /* Add new region. */
104 if (xen_extra_mem[i].n_pfns == 0) {
105 xen_extra_mem[i].start_pfn = start_pfn;
106 xen_extra_mem[i].n_pfns = n_pfns;
107 break;
108 }
109 /* Append to existing region. */
110 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
111 start_pfn) {
112 xen_extra_mem[i].n_pfns += n_pfns;
113 break;
114 }
115 }
116 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
117 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
118
119 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
120}
121
122static void __init xen_del_extra_mem(unsigned long start_pfn,
123 unsigned long n_pfns)
124{
125 int i;
126 unsigned long start_r, size_r;
127
128 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
129 start_r = xen_extra_mem[i].start_pfn;
130 size_r = xen_extra_mem[i].n_pfns;
131
132 /* Start of region. */
133 if (start_r == start_pfn) {
134 BUG_ON(n_pfns > size_r);
135 xen_extra_mem[i].start_pfn += n_pfns;
136 xen_extra_mem[i].n_pfns -= n_pfns;
137 break;
138 }
139 /* End of region. */
140 if (start_r + size_r == start_pfn + n_pfns) {
141 BUG_ON(n_pfns > size_r);
142 xen_extra_mem[i].n_pfns -= n_pfns;
143 break;
144 }
145 /* Mid of region. */
146 if (start_pfn > start_r && start_pfn < start_r + size_r) {
147 BUG_ON(start_pfn + n_pfns > start_r + size_r);
148 xen_extra_mem[i].n_pfns = start_pfn - start_r;
149 /* Calling memblock_reserve() again is okay. */
150 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
151 (start_pfn + n_pfns));
152 break;
153 }
154 }
155 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
156}
157
158/*
159 * Called during boot before the p2m list can take entries beyond the
160 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
161 * invalid.
162 */
163unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
164{
165 int i;
166
167 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
168 if (pfn >= xen_extra_mem[i].start_pfn &&
169 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
170 return INVALID_P2M_ENTRY;
171 }
172
173 return IDENTITY_FRAME(pfn);
174}
175
176/*
177 * Mark all pfns of extra mem as invalid in p2m list.
178 */
179void __init xen_inv_extra_mem(void)
180{
181 unsigned long pfn, pfn_s, pfn_e;
182 int i;
183
184 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
185 if (!xen_extra_mem[i].n_pfns)
186 continue;
187 pfn_s = xen_extra_mem[i].start_pfn;
188 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
189 for (pfn = pfn_s; pfn < pfn_e; pfn++)
190 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
191 }
192}
193
194/*
195 * Finds the next RAM pfn available in the E820 map after min_pfn.
196 * This function updates min_pfn with the pfn found and returns
197 * the size of that range or zero if not found.
198 */
199static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
200{
201 const struct e820_entry *entry = xen_e820_table.entries;
202 unsigned int i;
203 unsigned long done = 0;
204
205 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
206 unsigned long s_pfn;
207 unsigned long e_pfn;
208
209 if (entry->type != E820_TYPE_RAM)
210 continue;
211
212 e_pfn = PFN_DOWN(entry->addr + entry->size);
213
214 /* We only care about E820 after this */
215 if (e_pfn <= *min_pfn)
216 continue;
217
218 s_pfn = PFN_UP(entry->addr);
219
220 /* If min_pfn falls within the E820 entry, we want to start
221 * at the min_pfn PFN.
222 */
223 if (s_pfn <= *min_pfn) {
224 done = e_pfn - *min_pfn;
225 } else {
226 done = e_pfn - s_pfn;
227 *min_pfn = s_pfn;
228 }
229 break;
230 }
231
232 return done;
233}
234
235static int __init xen_free_mfn(unsigned long mfn)
236{
237 struct xen_memory_reservation reservation = {
238 .address_bits = 0,
239 .extent_order = 0,
240 .domid = DOMID_SELF
241 };
242
243 set_xen_guest_handle(reservation.extent_start, &mfn);
244 reservation.nr_extents = 1;
245
246 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
247}
248
249/*
250 * This releases a chunk of memory and then does the identity map. It's used
251 * as a fallback if the remapping fails.
252 */
253static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
254 unsigned long end_pfn, unsigned long nr_pages)
255{
256 unsigned long pfn, end;
257 int ret;
258
259 WARN_ON(start_pfn > end_pfn);
260
261 /* Release pages first. */
262 end = min(end_pfn, nr_pages);
263 for (pfn = start_pfn; pfn < end; pfn++) {
264 unsigned long mfn = pfn_to_mfn(pfn);
265
266 /* Make sure pfn exists to start with */
267 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
268 continue;
269
270 ret = xen_free_mfn(mfn);
271 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
272
273 if (ret == 1) {
274 xen_released_pages++;
275 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
276 break;
277 } else
278 break;
279 }
280
281 set_phys_range_identity(start_pfn, end_pfn);
282}
283
284/*
285 * Helper function to update the p2m and m2p tables and kernel mapping.
286 */
287static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
288{
289 struct mmu_update update = {
290 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
291 .val = pfn
292 };
293
294 /* Update p2m */
295 if (!set_phys_to_machine(pfn, mfn)) {
296 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
297 pfn, mfn);
298 BUG();
299 }
300
301 /* Update m2p */
302 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
303 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
304 mfn, pfn);
305 BUG();
306 }
307
308 /* Update kernel mapping, but not for highmem. */
309 if (pfn >= PFN_UP(__pa(high_memory - 1)))
310 return;
311
312 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
313 mfn_pte(mfn, PAGE_KERNEL), 0)) {
314 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
315 mfn, pfn);
316 BUG();
317 }
318}
319
320/*
321 * This function updates the p2m and m2p tables with an identity map from
322 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
323 * original allocation at remap_pfn. The information needed for remapping is
324 * saved in the memory itself to avoid the need for allocating buffers. The
325 * complete remap information is contained in a list of MFNs each containing
326 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
327 * This enables us to preserve the original mfn sequence while doing the
328 * remapping at a time when the memory management is capable of allocating
329 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
330 * its callers.
331 */
332static void __init xen_do_set_identity_and_remap_chunk(
333 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
334{
335 unsigned long buf = (unsigned long)&xen_remap_buf;
336 unsigned long mfn_save, mfn;
337 unsigned long ident_pfn_iter, remap_pfn_iter;
338 unsigned long ident_end_pfn = start_pfn + size;
339 unsigned long left = size;
340 unsigned int i, chunk;
341
342 WARN_ON(size == 0);
343
344 mfn_save = virt_to_mfn(buf);
345
346 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
347 ident_pfn_iter < ident_end_pfn;
348 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
349 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
350
351 /* Map first pfn to xen_remap_buf */
352 mfn = pfn_to_mfn(ident_pfn_iter);
353 set_pte_mfn(buf, mfn, PAGE_KERNEL);
354
355 /* Save mapping information in page */
356 xen_remap_buf.next_area_mfn = xen_remap_mfn;
357 xen_remap_buf.target_pfn = remap_pfn_iter;
358 xen_remap_buf.size = chunk;
359 for (i = 0; i < chunk; i++)
360 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
361
362 /* Put remap buf into list. */
363 xen_remap_mfn = mfn;
364
365 /* Set identity map */
366 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
367
368 left -= chunk;
369 }
370
371 /* Restore old xen_remap_buf mapping */
372 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
373}
374
375/*
376 * This function takes a contiguous pfn range that needs to be identity mapped
377 * and:
378 *
379 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
380 * 2) Calls the do_ function to actually do the mapping/remapping work.
381 *
382 * The goal is to not allocate additional memory but to remap the existing
383 * pages. In the case of an error the underlying memory is simply released back
384 * to Xen and not remapped.
385 */
386static unsigned long __init xen_set_identity_and_remap_chunk(
387 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
388 unsigned long remap_pfn)
389{
390 unsigned long pfn;
391 unsigned long i = 0;
392 unsigned long n = end_pfn - start_pfn;
393
394 if (remap_pfn == 0)
395 remap_pfn = nr_pages;
396
397 while (i < n) {
398 unsigned long cur_pfn = start_pfn + i;
399 unsigned long left = n - i;
400 unsigned long size = left;
401 unsigned long remap_range_size;
402
403 /* Do not remap pages beyond the current allocation */
404 if (cur_pfn >= nr_pages) {
405 /* Identity map remaining pages */
406 set_phys_range_identity(cur_pfn, cur_pfn + size);
407 break;
408 }
409 if (cur_pfn + size > nr_pages)
410 size = nr_pages - cur_pfn;
411
412 remap_range_size = xen_find_pfn_range(&remap_pfn);
413 if (!remap_range_size) {
414 pr_warning("Unable to find available pfn range, not remapping identity pages\n");
415 xen_set_identity_and_release_chunk(cur_pfn,
416 cur_pfn + left, nr_pages);
417 break;
418 }
419 /* Adjust size to fit in current e820 RAM region */
420 if (size > remap_range_size)
421 size = remap_range_size;
422
423 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
424
425 /* Update variables to reflect new mappings. */
426 i += size;
427 remap_pfn += size;
428 }
429
430 /*
431 * If the PFNs are currently mapped, the VA mapping also needs
432 * to be updated to be 1:1.
433 */
434 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
435 (void)HYPERVISOR_update_va_mapping(
436 (unsigned long)__va(pfn << PAGE_SHIFT),
437 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
438
439 return remap_pfn;
440}
441
442static unsigned long __init xen_count_remap_pages(
443 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
444 unsigned long remap_pages)
445{
446 if (start_pfn >= nr_pages)
447 return remap_pages;
448
449 return remap_pages + min(end_pfn, nr_pages) - start_pfn;
450}
451
452static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
453 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
454 unsigned long nr_pages, unsigned long last_val))
455{
456 phys_addr_t start = 0;
457 unsigned long ret_val = 0;
458 const struct e820_entry *entry = xen_e820_table.entries;
459 int i;
460
461 /*
462 * Combine non-RAM regions and gaps until a RAM region (or the
463 * end of the map) is reached, then call the provided function
464 * to perform its duty on the non-RAM region.
465 *
466 * The combined non-RAM regions are rounded to a whole number
467 * of pages so any partial pages are accessible via the 1:1
468 * mapping. This is needed for some BIOSes that put (for
469 * example) the DMI tables in a reserved region that begins on
470 * a non-page boundary.
471 */
472 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
473 phys_addr_t end = entry->addr + entry->size;
474 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
475 unsigned long start_pfn = PFN_DOWN(start);
476 unsigned long end_pfn = PFN_UP(end);
477
478 if (entry->type == E820_TYPE_RAM)
479 end_pfn = PFN_UP(entry->addr);
480
481 if (start_pfn < end_pfn)
482 ret_val = func(start_pfn, end_pfn, nr_pages,
483 ret_val);
484 start = end;
485 }
486 }
487
488 return ret_val;
489}
490
491/*
492 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
493 * The remap information (which mfn remap to which pfn) is contained in the
494 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
495 * This scheme allows to remap the different chunks in arbitrary order while
496 * the resulting mapping will be independant from the order.
497 */
498void __init xen_remap_memory(void)
499{
500 unsigned long buf = (unsigned long)&xen_remap_buf;
501 unsigned long mfn_save, pfn;
502 unsigned long remapped = 0;
503 unsigned int i;
504 unsigned long pfn_s = ~0UL;
505 unsigned long len = 0;
506
507 mfn_save = virt_to_mfn(buf);
508
509 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
510 /* Map the remap information */
511 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
512
513 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
514
515 pfn = xen_remap_buf.target_pfn;
516 for (i = 0; i < xen_remap_buf.size; i++) {
517 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
518 remapped++;
519 pfn++;
520 }
521 if (pfn_s == ~0UL || pfn == pfn_s) {
522 pfn_s = xen_remap_buf.target_pfn;
523 len += xen_remap_buf.size;
524 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
525 len += xen_remap_buf.size;
526 } else {
527 xen_del_extra_mem(pfn_s, len);
528 pfn_s = xen_remap_buf.target_pfn;
529 len = xen_remap_buf.size;
530 }
531 xen_remap_mfn = xen_remap_buf.next_area_mfn;
532 }
533
534 if (pfn_s != ~0UL && len)
535 xen_del_extra_mem(pfn_s, len);
536
537 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
538
539 pr_info("Remapped %ld page(s)\n", remapped);
540}
541
542static unsigned long __init xen_get_pages_limit(void)
543{
544 unsigned long limit;
545
546#ifdef CONFIG_X86_32
547 limit = GB(64) / PAGE_SIZE;
548#else
549 limit = MAXMEM / PAGE_SIZE;
550 if (!xen_initial_domain() && xen_512gb_limit)
551 limit = GB(512) / PAGE_SIZE;
552#endif
553 return limit;
554}
555
556static unsigned long __init xen_get_max_pages(void)
557{
558 unsigned long max_pages, limit;
559 domid_t domid = DOMID_SELF;
560 long ret;
561
562 limit = xen_get_pages_limit();
563 max_pages = limit;
564
565 /*
566 * For the initial domain we use the maximum reservation as
567 * the maximum page.
568 *
569 * For guest domains the current maximum reservation reflects
570 * the current maximum rather than the static maximum. In this
571 * case the e820 map provided to us will cover the static
572 * maximum region.
573 */
574 if (xen_initial_domain()) {
575 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
576 if (ret > 0)
577 max_pages = ret;
578 }
579
580 return min(max_pages, limit);
581}
582
583static void __init xen_align_and_add_e820_region(phys_addr_t start,
584 phys_addr_t size, int type)
585{
586 phys_addr_t end = start + size;
587
588 /* Align RAM regions to page boundaries. */
589 if (type == E820_TYPE_RAM) {
590 start = PAGE_ALIGN(start);
591 end &= ~((phys_addr_t)PAGE_SIZE - 1);
592 }
593
594 e820__range_add(start, end - start, type);
595}
596
597static void __init xen_ignore_unusable(void)
598{
599 struct e820_entry *entry = xen_e820_table.entries;
600 unsigned int i;
601
602 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
603 if (entry->type == E820_TYPE_UNUSABLE)
604 entry->type = E820_TYPE_RAM;
605 }
606}
607
608bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
609{
610 struct e820_entry *entry;
611 unsigned mapcnt;
612 phys_addr_t end;
613
614 if (!size)
615 return false;
616
617 end = start + size;
618 entry = xen_e820_table.entries;
619
620 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
621 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
622 (entry->addr + entry->size) >= end)
623 return false;
624
625 entry++;
626 }
627
628 return true;
629}
630
631/*
632 * Find a free area in physical memory not yet reserved and compliant with
633 * E820 map.
634 * Used to relocate pre-allocated areas like initrd or p2m list which are in
635 * conflict with the to be used E820 map.
636 * In case no area is found, return 0. Otherwise return the physical address
637 * of the area which is already reserved for convenience.
638 */
639phys_addr_t __init xen_find_free_area(phys_addr_t size)
640{
641 unsigned mapcnt;
642 phys_addr_t addr, start;
643 struct e820_entry *entry = xen_e820_table.entries;
644
645 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
646 if (entry->type != E820_TYPE_RAM || entry->size < size)
647 continue;
648 start = entry->addr;
649 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
650 if (!memblock_is_reserved(addr))
651 continue;
652 start = addr + PAGE_SIZE;
653 if (start + size > entry->addr + entry->size)
654 break;
655 }
656 if (addr >= start + size) {
657 memblock_reserve(start, size);
658 return start;
659 }
660 }
661
662 return 0;
663}
664
665/*
666 * Like memcpy, but with physical addresses for dest and src.
667 */
668static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
669 phys_addr_t n)
670{
671 phys_addr_t dest_off, src_off, dest_len, src_len, len;
672 void *from, *to;
673
674 while (n) {
675 dest_off = dest & ~PAGE_MASK;
676 src_off = src & ~PAGE_MASK;
677 dest_len = n;
678 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
679 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
680 src_len = n;
681 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
682 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
683 len = min(dest_len, src_len);
684 to = early_memremap(dest - dest_off, dest_len + dest_off);
685 from = early_memremap(src - src_off, src_len + src_off);
686 memcpy(to, from, len);
687 early_memunmap(to, dest_len + dest_off);
688 early_memunmap(from, src_len + src_off);
689 n -= len;
690 dest += len;
691 src += len;
692 }
693}
694
695/*
696 * Reserve Xen mfn_list.
697 */
698static void __init xen_reserve_xen_mfnlist(void)
699{
700 phys_addr_t start, size;
701
702 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
703 start = __pa(xen_start_info->mfn_list);
704 size = PFN_ALIGN(xen_start_info->nr_pages *
705 sizeof(unsigned long));
706 } else {
707 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
708 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
709 }
710
711 memblock_reserve(start, size);
712 if (!xen_is_e820_reserved(start, size))
713 return;
714
715#ifdef CONFIG_X86_32
716 /*
717 * Relocating the p2m on 32 bit system to an arbitrary virtual address
718 * is not supported, so just give up.
719 */
720 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
721 BUG();
722#else
723 xen_relocate_p2m();
724 memblock_free(start, size);
725#endif
726}
727
728/**
729 * machine_specific_memory_setup - Hook for machine specific memory setup.
730 **/
731char * __init xen_memory_setup(void)
732{
733 unsigned long max_pfn, pfn_s, n_pfns;
734 phys_addr_t mem_end, addr, size, chunk_size;
735 u32 type;
736 int rc;
737 struct xen_memory_map memmap;
738 unsigned long max_pages;
739 unsigned long extra_pages = 0;
740 int i;
741 int op;
742
743 xen_parse_512gb();
744 max_pfn = xen_get_pages_limit();
745 max_pfn = min(max_pfn, xen_start_info->nr_pages);
746 mem_end = PFN_PHYS(max_pfn);
747
748 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
749 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
750
751 op = xen_initial_domain() ?
752 XENMEM_machine_memory_map :
753 XENMEM_memory_map;
754 rc = HYPERVISOR_memory_op(op, &memmap);
755 if (rc == -ENOSYS) {
756 BUG_ON(xen_initial_domain());
757 memmap.nr_entries = 1;
758 xen_e820_table.entries[0].addr = 0ULL;
759 xen_e820_table.entries[0].size = mem_end;
760 /* 8MB slack (to balance backend allocations). */
761 xen_e820_table.entries[0].size += 8ULL << 20;
762 xen_e820_table.entries[0].type = E820_TYPE_RAM;
763 rc = 0;
764 }
765 BUG_ON(rc);
766 BUG_ON(memmap.nr_entries == 0);
767 xen_e820_table.nr_entries = memmap.nr_entries;
768
769 /*
770 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
771 * regions, so if we're using the machine memory map leave the
772 * region as RAM as it is in the pseudo-physical map.
773 *
774 * UNUSABLE regions in domUs are not handled and will need
775 * a patch in the future.
776 */
777 if (xen_initial_domain())
778 xen_ignore_unusable();
779
780 /* Make sure the Xen-supplied memory map is well-ordered. */
781 e820__update_table(&xen_e820_table);
782
783 max_pages = xen_get_max_pages();
784
785 /* How many extra pages do we need due to remapping? */
786 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
787
788 if (max_pages > max_pfn)
789 extra_pages += max_pages - max_pfn;
790
791 /*
792 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
793 * factor the base size. On non-highmem systems, the base
794 * size is the full initial memory allocation; on highmem it
795 * is limited to the max size of lowmem, so that it doesn't
796 * get completely filled.
797 *
798 * Make sure we have no memory above max_pages, as this area
799 * isn't handled by the p2m management.
800 *
801 * In principle there could be a problem in lowmem systems if
802 * the initial memory is also very large with respect to
803 * lowmem, but we won't try to deal with that here.
804 */
805 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
806 extra_pages, max_pages - max_pfn);
807 i = 0;
808 addr = xen_e820_table.entries[0].addr;
809 size = xen_e820_table.entries[0].size;
810 while (i < xen_e820_table.nr_entries) {
811
812 chunk_size = size;
813 type = xen_e820_table.entries[i].type;
814
815 if (type == E820_TYPE_RAM) {
816 if (addr < mem_end) {
817 chunk_size = min(size, mem_end - addr);
818 } else if (extra_pages) {
819 chunk_size = min(size, PFN_PHYS(extra_pages));
820 pfn_s = PFN_UP(addr);
821 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
822 extra_pages -= n_pfns;
823 xen_add_extra_mem(pfn_s, n_pfns);
824 xen_max_p2m_pfn = pfn_s + n_pfns;
825 } else
826 type = E820_TYPE_UNUSABLE;
827 }
828
829 xen_align_and_add_e820_region(addr, chunk_size, type);
830
831 addr += chunk_size;
832 size -= chunk_size;
833 if (size == 0) {
834 i++;
835 if (i < xen_e820_table.nr_entries) {
836 addr = xen_e820_table.entries[i].addr;
837 size = xen_e820_table.entries[i].size;
838 }
839 }
840 }
841
842 /*
843 * Set the rest as identity mapped, in case PCI BARs are
844 * located here.
845 */
846 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
847
848 /*
849 * In domU, the ISA region is normal, usable memory, but we
850 * reserve ISA memory anyway because too many things poke
851 * about in there.
852 */
853 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
854
855 e820__update_table(e820_table);
856
857 /*
858 * Check whether the kernel itself conflicts with the target E820 map.
859 * Failing now is better than running into weird problems later due
860 * to relocating (and even reusing) pages with kernel text or data.
861 */
862 if (xen_is_e820_reserved(__pa_symbol(_text),
863 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
864 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
865 BUG();
866 }
867
868 /*
869 * Check for a conflict of the hypervisor supplied page tables with
870 * the target E820 map.
871 */
872 xen_pt_check_e820();
873
874 xen_reserve_xen_mfnlist();
875
876 /* Check for a conflict of the initrd with the target E820 map. */
877 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
878 boot_params.hdr.ramdisk_size)) {
879 phys_addr_t new_area, start, size;
880
881 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
882 if (!new_area) {
883 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
884 BUG();
885 }
886
887 start = boot_params.hdr.ramdisk_image;
888 size = boot_params.hdr.ramdisk_size;
889 xen_phys_memcpy(new_area, start, size);
890 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
891 start, start + size, new_area, new_area + size);
892 memblock_free(start, size);
893 boot_params.hdr.ramdisk_image = new_area;
894 boot_params.ext_ramdisk_image = new_area >> 32;
895 }
896
897 /*
898 * Set identity map on non-RAM pages and prepare remapping the
899 * underlying RAM.
900 */
901 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
902
903 pr_info("Released %ld page(s)\n", xen_released_pages);
904
905 return "Xen";
906}
907
908/*
909 * Machine specific memory setup for auto-translated guests.
910 */
911char * __init xen_auto_xlated_memory_setup(void)
912{
913 struct xen_memory_map memmap;
914 int i;
915 int rc;
916
917 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
918 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
919
920 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
921 if (rc < 0)
922 panic("No memory map (%d)\n", rc);
923
924 xen_e820_table.nr_entries = memmap.nr_entries;
925
926 e820__update_table(&xen_e820_table);
927
928 for (i = 0; i < xen_e820_table.nr_entries; i++)
929 e820__range_add(xen_e820_table.entries[i].addr, xen_e820_table.entries[i].size, xen_e820_table.entries[i].type);
930
931 /* Remove p2m info, it is not needed. */
932 xen_start_info->mfn_list = 0;
933 xen_start_info->first_p2m_pfn = 0;
934 xen_start_info->nr_p2m_frames = 0;
935
936 return "Xen";
937}
938
939/*
940 * Set the bit indicating "nosegneg" library variants should be used.
941 * We only need to bother in pure 32-bit mode; compat 32-bit processes
942 * can have un-truncated segments, so wrapping around is allowed.
943 */
944static void __init fiddle_vdso(void)
945{
946#ifdef CONFIG_X86_32
947 u32 *mask = vdso_image_32.data +
948 vdso_image_32.sym_VDSO32_NOTE_MASK;
949 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
950#endif
951}
952
953static int register_callback(unsigned type, const void *func)
954{
955 struct callback_register callback = {
956 .type = type,
957 .address = XEN_CALLBACK(__KERNEL_CS, func),
958 .flags = CALLBACKF_mask_events,
959 };
960
961 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
962}
963
964void xen_enable_sysenter(void)
965{
966 int ret;
967 unsigned sysenter_feature;
968
969#ifdef CONFIG_X86_32
970 sysenter_feature = X86_FEATURE_SEP;
971#else
972 sysenter_feature = X86_FEATURE_SYSENTER32;
973#endif
974
975 if (!boot_cpu_has(sysenter_feature))
976 return;
977
978 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
979 if(ret != 0)
980 setup_clear_cpu_cap(sysenter_feature);
981}
982
983void xen_enable_syscall(void)
984{
985#ifdef CONFIG_X86_64
986 int ret;
987
988 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
989 if (ret != 0) {
990 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
991 /* Pretty fatal; 64-bit userspace has no other
992 mechanism for syscalls. */
993 }
994
995 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
996 ret = register_callback(CALLBACKTYPE_syscall32,
997 xen_syscall32_target);
998 if (ret != 0)
999 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1000 }
1001#endif /* CONFIG_X86_64 */
1002}
1003
1004void __init xen_pvmmu_arch_setup(void)
1005{
1006 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
1007 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1008
1009 HYPERVISOR_vm_assist(VMASST_CMD_enable,
1010 VMASST_TYPE_pae_extended_cr3);
1011
1012 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
1013 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1014 BUG();
1015
1016 xen_enable_sysenter();
1017 xen_enable_syscall();
1018}
1019
1020/* This function is not called for HVM domains */
1021void __init xen_arch_setup(void)
1022{
1023 xen_panic_handler_init();
1024 xen_pvmmu_arch_setup();
1025
1026#ifdef CONFIG_ACPI
1027 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1028 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1029 disable_acpi();
1030 }
1031#endif
1032
1033 memcpy(boot_command_line, xen_start_info->cmd_line,
1034 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1035 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1036
1037 /* Set up idle, making sure it calls safe_halt() pvop */
1038 disable_cpuidle();
1039 disable_cpufreq();
1040 WARN_ON(xen_set_default_idle());
1041 fiddle_vdso();
1042#ifdef CONFIG_NUMA
1043 numa_off = 1;
1044#endif
1045}