1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Machine specific setup for xen
4 *
5 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
6 */
7
8 #include <linux/init.h>
9 #include <linux/sched.h>
10 #include <linux/mm.h>
11 #include <linux/pm.h>
12 #include <linux/memblock.h>
13 #include <linux/cpuidle.h>
14 #include <linux/cpufreq.h>
15 #include <linux/memory_hotplug.h>
16
17 #include <asm/elf.h>
18 #include <asm/vdso.h>
19 #include <asm/e820/api.h>
20 #include <asm/setup.h>
21 #include <asm/acpi.h>
22 #include <asm/numa.h>
23 #include <asm/idtentry.h>
24 #include <asm/xen/hypervisor.h>
25 #include <asm/xen/hypercall.h>
26
27 #include <xen/xen.h>
28 #include <xen/page.h>
29 #include <xen/interface/callback.h>
30 #include <xen/interface/memory.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/features.h>
33 #include <xen/hvc-console.h>
34 #include "xen-ops.h"
35 #include "mmu.h"
36
37 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
38
39 /* Amount of extra memory space we add to the e820 ranges */
40 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
41
42 /* Number of pages released from the initial allocation. */
43 unsigned long xen_released_pages;
44
45 /* E820 map used during setting up memory. */
46 static struct e820_table xen_e820_table __initdata;
47
48 /*
49 * Buffer used to remap identity mapped pages. We only need the virtual space.
50 * The physical page behind this address is remapped as needed to different
51 * buffer pages.
52 */
53 #define REMAP_SIZE (P2M_PER_PAGE - 3)
54 static struct {
55 unsigned long next_area_mfn;
56 unsigned long target_pfn;
57 unsigned long size;
58 unsigned long mfns[REMAP_SIZE];
59 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
60 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
61
62 /*
63 * The maximum amount of extra memory compared to the base size. The
64 * main scaling factor is the size of struct page. At extreme ratios
65 * of base:extra, all the base memory can be filled with page
66 * structures for the extra memory, leaving no space for anything
67 * else.
68 *
69 * 10x seems like a reasonable balance between scaling flexibility and
70 * leaving a practically usable system.
71 */
72 #define EXTRA_MEM_RATIO (10)
73
74 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
75
xen_parse_512gb(void)76 static void __init xen_parse_512gb(void)
77 {
78 bool val = false;
79 char *arg;
80
81 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
82 if (!arg)
83 return;
84
85 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
86 if (!arg)
87 val = true;
88 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
89 return;
90
91 xen_512gb_limit = val;
92 }
93
xen_add_extra_mem(unsigned long start_pfn,unsigned long n_pfns)94 static void __init xen_add_extra_mem(unsigned long start_pfn,
95 unsigned long n_pfns)
96 {
97 int i;
98
99 /*
100 * No need to check for zero size, should happen rarely and will only
101 * write a new entry regarded to be unused due to zero size.
102 */
103 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
104 /* Add new region. */
105 if (xen_extra_mem[i].n_pfns == 0) {
106 xen_extra_mem[i].start_pfn = start_pfn;
107 xen_extra_mem[i].n_pfns = n_pfns;
108 break;
109 }
110 /* Append to existing region. */
111 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
112 start_pfn) {
113 xen_extra_mem[i].n_pfns += n_pfns;
114 break;
115 }
116 }
117 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
118 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
119
120 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
121 }
122
xen_del_extra_mem(unsigned long start_pfn,unsigned long n_pfns)123 static void __init xen_del_extra_mem(unsigned long start_pfn,
124 unsigned long n_pfns)
125 {
126 int i;
127 unsigned long start_r, size_r;
128
129 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
130 start_r = xen_extra_mem[i].start_pfn;
131 size_r = xen_extra_mem[i].n_pfns;
132
133 /* Start of region. */
134 if (start_r == start_pfn) {
135 BUG_ON(n_pfns > size_r);
136 xen_extra_mem[i].start_pfn += n_pfns;
137 xen_extra_mem[i].n_pfns -= n_pfns;
138 break;
139 }
140 /* End of region. */
141 if (start_r + size_r == start_pfn + n_pfns) {
142 BUG_ON(n_pfns > size_r);
143 xen_extra_mem[i].n_pfns -= n_pfns;
144 break;
145 }
146 /* Mid of region. */
147 if (start_pfn > start_r && start_pfn < start_r + size_r) {
148 BUG_ON(start_pfn + n_pfns > start_r + size_r);
149 xen_extra_mem[i].n_pfns = start_pfn - start_r;
150 /* Calling memblock_reserve() again is okay. */
151 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
152 (start_pfn + n_pfns));
153 break;
154 }
155 }
156 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
157 }
158
159 /*
160 * Called during boot before the p2m list can take entries beyond the
161 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
162 * invalid.
163 */
xen_chk_extra_mem(unsigned long pfn)164 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
165 {
166 int i;
167
168 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
169 if (pfn >= xen_extra_mem[i].start_pfn &&
170 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
171 return INVALID_P2M_ENTRY;
172 }
173
174 return IDENTITY_FRAME(pfn);
175 }
176
177 /*
178 * Mark all pfns of extra mem as invalid in p2m list.
179 */
xen_inv_extra_mem(void)180 void __init xen_inv_extra_mem(void)
181 {
182 unsigned long pfn, pfn_s, pfn_e;
183 int i;
184
185 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
186 if (!xen_extra_mem[i].n_pfns)
187 continue;
188 pfn_s = xen_extra_mem[i].start_pfn;
189 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
190 for (pfn = pfn_s; pfn < pfn_e; pfn++)
191 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
192 }
193 }
194
195 /*
196 * Finds the next RAM pfn available in the E820 map after min_pfn.
197 * This function updates min_pfn with the pfn found and returns
198 * the size of that range or zero if not found.
199 */
xen_find_pfn_range(unsigned long * min_pfn)200 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
201 {
202 const struct e820_entry *entry = xen_e820_table.entries;
203 unsigned int i;
204 unsigned long done = 0;
205
206 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
207 unsigned long s_pfn;
208 unsigned long e_pfn;
209
210 if (entry->type != E820_TYPE_RAM)
211 continue;
212
213 e_pfn = PFN_DOWN(entry->addr + entry->size);
214
215 /* We only care about E820 after this */
216 if (e_pfn <= *min_pfn)
217 continue;
218
219 s_pfn = PFN_UP(entry->addr);
220
221 /* If min_pfn falls within the E820 entry, we want to start
222 * at the min_pfn PFN.
223 */
224 if (s_pfn <= *min_pfn) {
225 done = e_pfn - *min_pfn;
226 } else {
227 done = e_pfn - s_pfn;
228 *min_pfn = s_pfn;
229 }
230 break;
231 }
232
233 return done;
234 }
235
xen_free_mfn(unsigned long mfn)236 static int __init xen_free_mfn(unsigned long mfn)
237 {
238 struct xen_memory_reservation reservation = {
239 .address_bits = 0,
240 .extent_order = 0,
241 .domid = DOMID_SELF
242 };
243
244 set_xen_guest_handle(reservation.extent_start, &mfn);
245 reservation.nr_extents = 1;
246
247 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
248 }
249
250 /*
251 * This releases a chunk of memory and then does the identity map. It's used
252 * as a fallback if the remapping fails.
253 */
xen_set_identity_and_release_chunk(unsigned long start_pfn,unsigned long end_pfn,unsigned long nr_pages)254 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
255 unsigned long end_pfn, unsigned long nr_pages)
256 {
257 unsigned long pfn, end;
258 int ret;
259
260 WARN_ON(start_pfn > end_pfn);
261
262 /* Release pages first. */
263 end = min(end_pfn, nr_pages);
264 for (pfn = start_pfn; pfn < end; pfn++) {
265 unsigned long mfn = pfn_to_mfn(pfn);
266
267 /* Make sure pfn exists to start with */
268 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
269 continue;
270
271 ret = xen_free_mfn(mfn);
272 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
273
274 if (ret == 1) {
275 xen_released_pages++;
276 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
277 break;
278 } else
279 break;
280 }
281
282 set_phys_range_identity(start_pfn, end_pfn);
283 }
284
285 /*
286 * Helper function to update the p2m and m2p tables and kernel mapping.
287 */
xen_update_mem_tables(unsigned long pfn,unsigned long mfn)288 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
289 {
290 struct mmu_update update = {
291 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
292 .val = pfn
293 };
294
295 /* Update p2m */
296 if (!set_phys_to_machine(pfn, mfn)) {
297 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
298 pfn, mfn);
299 BUG();
300 }
301
302 /* Update m2p */
303 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
304 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
305 mfn, pfn);
306 BUG();
307 }
308
309 /* Update kernel mapping, but not for highmem. */
310 if (pfn >= PFN_UP(__pa(high_memory - 1)))
311 return;
312
313 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
314 mfn_pte(mfn, PAGE_KERNEL), 0)) {
315 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
316 mfn, pfn);
317 BUG();
318 }
319 }
320
321 /*
322 * This function updates the p2m and m2p tables with an identity map from
323 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
324 * original allocation at remap_pfn. The information needed for remapping is
325 * saved in the memory itself to avoid the need for allocating buffers. The
326 * complete remap information is contained in a list of MFNs each containing
327 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
328 * This enables us to preserve the original mfn sequence while doing the
329 * remapping at a time when the memory management is capable of allocating
330 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
331 * its callers.
332 */
xen_do_set_identity_and_remap_chunk(unsigned long start_pfn,unsigned long size,unsigned long remap_pfn)333 static void __init xen_do_set_identity_and_remap_chunk(
334 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
335 {
336 unsigned long buf = (unsigned long)&xen_remap_buf;
337 unsigned long mfn_save, mfn;
338 unsigned long ident_pfn_iter, remap_pfn_iter;
339 unsigned long ident_end_pfn = start_pfn + size;
340 unsigned long left = size;
341 unsigned int i, chunk;
342
343 WARN_ON(size == 0);
344
345 mfn_save = virt_to_mfn(buf);
346
347 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
348 ident_pfn_iter < ident_end_pfn;
349 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
350 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
351
352 /* Map first pfn to xen_remap_buf */
353 mfn = pfn_to_mfn(ident_pfn_iter);
354 set_pte_mfn(buf, mfn, PAGE_KERNEL);
355
356 /* Save mapping information in page */
357 xen_remap_buf.next_area_mfn = xen_remap_mfn;
358 xen_remap_buf.target_pfn = remap_pfn_iter;
359 xen_remap_buf.size = chunk;
360 for (i = 0; i < chunk; i++)
361 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
362
363 /* Put remap buf into list. */
364 xen_remap_mfn = mfn;
365
366 /* Set identity map */
367 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
368
369 left -= chunk;
370 }
371
372 /* Restore old xen_remap_buf mapping */
373 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
374 }
375
376 /*
377 * This function takes a contiguous pfn range that needs to be identity mapped
378 * and:
379 *
380 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
381 * 2) Calls the do_ function to actually do the mapping/remapping work.
382 *
383 * The goal is to not allocate additional memory but to remap the existing
384 * pages. In the case of an error the underlying memory is simply released back
385 * to Xen and not remapped.
386 */
xen_set_identity_and_remap_chunk(unsigned long start_pfn,unsigned long end_pfn,unsigned long nr_pages,unsigned long remap_pfn)387 static unsigned long __init xen_set_identity_and_remap_chunk(
388 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
389 unsigned long remap_pfn)
390 {
391 unsigned long pfn;
392 unsigned long i = 0;
393 unsigned long n = end_pfn - start_pfn;
394
395 if (remap_pfn == 0)
396 remap_pfn = nr_pages;
397
398 while (i < n) {
399 unsigned long cur_pfn = start_pfn + i;
400 unsigned long left = n - i;
401 unsigned long size = left;
402 unsigned long remap_range_size;
403
404 /* Do not remap pages beyond the current allocation */
405 if (cur_pfn >= nr_pages) {
406 /* Identity map remaining pages */
407 set_phys_range_identity(cur_pfn, cur_pfn + size);
408 break;
409 }
410 if (cur_pfn + size > nr_pages)
411 size = nr_pages - cur_pfn;
412
413 remap_range_size = xen_find_pfn_range(&remap_pfn);
414 if (!remap_range_size) {
415 pr_warn("Unable to find available pfn range, not remapping identity pages\n");
416 xen_set_identity_and_release_chunk(cur_pfn,
417 cur_pfn + left, nr_pages);
418 break;
419 }
420 /* Adjust size to fit in current e820 RAM region */
421 if (size > remap_range_size)
422 size = remap_range_size;
423
424 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
425
426 /* Update variables to reflect new mappings. */
427 i += size;
428 remap_pfn += size;
429 }
430
431 /*
432 * If the PFNs are currently mapped, the VA mapping also needs
433 * to be updated to be 1:1.
434 */
435 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
436 (void)HYPERVISOR_update_va_mapping(
437 (unsigned long)__va(pfn << PAGE_SHIFT),
438 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
439
440 return remap_pfn;
441 }
442
xen_count_remap_pages(unsigned long start_pfn,unsigned long end_pfn,unsigned long nr_pages,unsigned long remap_pages)443 static unsigned long __init xen_count_remap_pages(
444 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
445 unsigned long remap_pages)
446 {
447 if (start_pfn >= nr_pages)
448 return remap_pages;
449
450 return remap_pages + min(end_pfn, nr_pages) - start_pfn;
451 }
452
xen_foreach_remap_area(unsigned long nr_pages,unsigned long (* func)(unsigned long start_pfn,unsigned long end_pfn,unsigned long nr_pages,unsigned long last_val))453 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
454 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
455 unsigned long nr_pages, unsigned long last_val))
456 {
457 phys_addr_t start = 0;
458 unsigned long ret_val = 0;
459 const struct e820_entry *entry = xen_e820_table.entries;
460 int i;
461
462 /*
463 * Combine non-RAM regions and gaps until a RAM region (or the
464 * end of the map) is reached, then call the provided function
465 * to perform its duty on the non-RAM region.
466 *
467 * The combined non-RAM regions are rounded to a whole number
468 * of pages so any partial pages are accessible via the 1:1
469 * mapping. This is needed for some BIOSes that put (for
470 * example) the DMI tables in a reserved region that begins on
471 * a non-page boundary.
472 */
473 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
474 phys_addr_t end = entry->addr + entry->size;
475 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
476 unsigned long start_pfn = PFN_DOWN(start);
477 unsigned long end_pfn = PFN_UP(end);
478
479 if (entry->type == E820_TYPE_RAM)
480 end_pfn = PFN_UP(entry->addr);
481
482 if (start_pfn < end_pfn)
483 ret_val = func(start_pfn, end_pfn, nr_pages,
484 ret_val);
485 start = end;
486 }
487 }
488
489 return ret_val;
490 }
491
492 /*
493 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
494 * The remap information (which mfn remap to which pfn) is contained in the
495 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
496 * This scheme allows to remap the different chunks in arbitrary order while
497 * the resulting mapping will be independent from the order.
498 */
xen_remap_memory(void)499 void __init xen_remap_memory(void)
500 {
501 unsigned long buf = (unsigned long)&xen_remap_buf;
502 unsigned long mfn_save, pfn;
503 unsigned long remapped = 0;
504 unsigned int i;
505 unsigned long pfn_s = ~0UL;
506 unsigned long len = 0;
507
508 mfn_save = virt_to_mfn(buf);
509
510 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
511 /* Map the remap information */
512 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
513
514 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
515
516 pfn = xen_remap_buf.target_pfn;
517 for (i = 0; i < xen_remap_buf.size; i++) {
518 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
519 remapped++;
520 pfn++;
521 }
522 if (pfn_s == ~0UL || pfn == pfn_s) {
523 pfn_s = xen_remap_buf.target_pfn;
524 len += xen_remap_buf.size;
525 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
526 len += xen_remap_buf.size;
527 } else {
528 xen_del_extra_mem(pfn_s, len);
529 pfn_s = xen_remap_buf.target_pfn;
530 len = xen_remap_buf.size;
531 }
532 xen_remap_mfn = xen_remap_buf.next_area_mfn;
533 }
534
535 if (pfn_s != ~0UL && len)
536 xen_del_extra_mem(pfn_s, len);
537
538 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
539
540 pr_info("Remapped %ld page(s)\n", remapped);
541 }
542
xen_get_pages_limit(void)543 static unsigned long __init xen_get_pages_limit(void)
544 {
545 unsigned long limit;
546
547 limit = MAXMEM / PAGE_SIZE;
548 if (!xen_initial_domain() && xen_512gb_limit)
549 limit = GB(512) / PAGE_SIZE;
550
551 return limit;
552 }
553
xen_get_max_pages(void)554 static unsigned long __init xen_get_max_pages(void)
555 {
556 unsigned long max_pages, limit;
557 domid_t domid = DOMID_SELF;
558 long ret;
559
560 limit = xen_get_pages_limit();
561 max_pages = limit;
562
563 /*
564 * For the initial domain we use the maximum reservation as
565 * the maximum page.
566 *
567 * For guest domains the current maximum reservation reflects
568 * the current maximum rather than the static maximum. In this
569 * case the e820 map provided to us will cover the static
570 * maximum region.
571 */
572 if (xen_initial_domain()) {
573 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
574 if (ret > 0)
575 max_pages = ret;
576 }
577
578 return min(max_pages, limit);
579 }
580
xen_align_and_add_e820_region(phys_addr_t start,phys_addr_t size,int type)581 static void __init xen_align_and_add_e820_region(phys_addr_t start,
582 phys_addr_t size, int type)
583 {
584 phys_addr_t end = start + size;
585
586 /* Align RAM regions to page boundaries. */
587 if (type == E820_TYPE_RAM) {
588 start = PAGE_ALIGN(start);
589 end &= ~((phys_addr_t)PAGE_SIZE - 1);
590 #ifdef CONFIG_MEMORY_HOTPLUG
591 /*
592 * Don't allow adding memory not in E820 map while booting the
593 * system. Once the balloon driver is up it will remove that
594 * restriction again.
595 */
596 max_mem_size = end;
597 #endif
598 }
599
600 e820__range_add(start, end - start, type);
601 }
602
xen_ignore_unusable(void)603 static void __init xen_ignore_unusable(void)
604 {
605 struct e820_entry *entry = xen_e820_table.entries;
606 unsigned int i;
607
608 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
609 if (entry->type == E820_TYPE_UNUSABLE)
610 entry->type = E820_TYPE_RAM;
611 }
612 }
613
xen_is_e820_reserved(phys_addr_t start,phys_addr_t size)614 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
615 {
616 struct e820_entry *entry;
617 unsigned mapcnt;
618 phys_addr_t end;
619
620 if (!size)
621 return false;
622
623 end = start + size;
624 entry = xen_e820_table.entries;
625
626 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
627 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
628 (entry->addr + entry->size) >= end)
629 return false;
630
631 entry++;
632 }
633
634 return true;
635 }
636
637 /*
638 * Find a free area in physical memory not yet reserved and compliant with
639 * E820 map.
640 * Used to relocate pre-allocated areas like initrd or p2m list which are in
641 * conflict with the to be used E820 map.
642 * In case no area is found, return 0. Otherwise return the physical address
643 * of the area which is already reserved for convenience.
644 */
xen_find_free_area(phys_addr_t size)645 phys_addr_t __init xen_find_free_area(phys_addr_t size)
646 {
647 unsigned mapcnt;
648 phys_addr_t addr, start;
649 struct e820_entry *entry = xen_e820_table.entries;
650
651 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
652 if (entry->type != E820_TYPE_RAM || entry->size < size)
653 continue;
654 start = entry->addr;
655 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
656 if (!memblock_is_reserved(addr))
657 continue;
658 start = addr + PAGE_SIZE;
659 if (start + size > entry->addr + entry->size)
660 break;
661 }
662 if (addr >= start + size) {
663 memblock_reserve(start, size);
664 return start;
665 }
666 }
667
668 return 0;
669 }
670
671 /*
672 * Like memcpy, but with physical addresses for dest and src.
673 */
xen_phys_memcpy(phys_addr_t dest,phys_addr_t src,phys_addr_t n)674 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
675 phys_addr_t n)
676 {
677 phys_addr_t dest_off, src_off, dest_len, src_len, len;
678 void *from, *to;
679
680 while (n) {
681 dest_off = dest & ~PAGE_MASK;
682 src_off = src & ~PAGE_MASK;
683 dest_len = n;
684 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
685 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
686 src_len = n;
687 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
688 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
689 len = min(dest_len, src_len);
690 to = early_memremap(dest - dest_off, dest_len + dest_off);
691 from = early_memremap(src - src_off, src_len + src_off);
692 memcpy(to, from, len);
693 early_memunmap(to, dest_len + dest_off);
694 early_memunmap(from, src_len + src_off);
695 n -= len;
696 dest += len;
697 src += len;
698 }
699 }
700
701 /*
702 * Reserve Xen mfn_list.
703 */
xen_reserve_xen_mfnlist(void)704 static void __init xen_reserve_xen_mfnlist(void)
705 {
706 phys_addr_t start, size;
707
708 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
709 start = __pa(xen_start_info->mfn_list);
710 size = PFN_ALIGN(xen_start_info->nr_pages *
711 sizeof(unsigned long));
712 } else {
713 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
714 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
715 }
716
717 memblock_reserve(start, size);
718 if (!xen_is_e820_reserved(start, size))
719 return;
720
721 xen_relocate_p2m();
722 memblock_free(start, size);
723 }
724
725 /**
726 * machine_specific_memory_setup - Hook for machine specific memory setup.
727 **/
xen_memory_setup(void)728 char * __init xen_memory_setup(void)
729 {
730 unsigned long max_pfn, pfn_s, n_pfns;
731 phys_addr_t mem_end, addr, size, chunk_size;
732 u32 type;
733 int rc;
734 struct xen_memory_map memmap;
735 unsigned long max_pages;
736 unsigned long extra_pages = 0;
737 int i;
738 int op;
739
740 xen_parse_512gb();
741 max_pfn = xen_get_pages_limit();
742 max_pfn = min(max_pfn, xen_start_info->nr_pages);
743 mem_end = PFN_PHYS(max_pfn);
744
745 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
746 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
747
748 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
749 xen_saved_max_mem_size = max_mem_size;
750 #endif
751
752 op = xen_initial_domain() ?
753 XENMEM_machine_memory_map :
754 XENMEM_memory_map;
755 rc = HYPERVISOR_memory_op(op, &memmap);
756 if (rc == -ENOSYS) {
757 BUG_ON(xen_initial_domain());
758 memmap.nr_entries = 1;
759 xen_e820_table.entries[0].addr = 0ULL;
760 xen_e820_table.entries[0].size = mem_end;
761 /* 8MB slack (to balance backend allocations). */
762 xen_e820_table.entries[0].size += 8ULL << 20;
763 xen_e820_table.entries[0].type = E820_TYPE_RAM;
764 rc = 0;
765 }
766 BUG_ON(rc);
767 BUG_ON(memmap.nr_entries == 0);
768 xen_e820_table.nr_entries = memmap.nr_entries;
769
770 /*
771 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
772 * regions, so if we're using the machine memory map leave the
773 * region as RAM as it is in the pseudo-physical map.
774 *
775 * UNUSABLE regions in domUs are not handled and will need
776 * a patch in the future.
777 */
778 if (xen_initial_domain())
779 xen_ignore_unusable();
780
781 /* Make sure the Xen-supplied memory map is well-ordered. */
782 e820__update_table(&xen_e820_table);
783
784 max_pages = xen_get_max_pages();
785
786 /* How many extra pages do we need due to remapping? */
787 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
788
789 if (max_pages > max_pfn)
790 extra_pages += max_pages - max_pfn;
791
792 /*
793 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
794 * factor the base size.
795 *
796 * Make sure we have no memory above max_pages, as this area
797 * isn't handled by the p2m management.
798 */
799 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
800 extra_pages, max_pages - max_pfn);
801 i = 0;
802 addr = xen_e820_table.entries[0].addr;
803 size = xen_e820_table.entries[0].size;
804 while (i < xen_e820_table.nr_entries) {
805 bool discard = false;
806
807 chunk_size = size;
808 type = xen_e820_table.entries[i].type;
809
810 if (type == E820_TYPE_RAM) {
811 if (addr < mem_end) {
812 chunk_size = min(size, mem_end - addr);
813 } else if (extra_pages) {
814 chunk_size = min(size, PFN_PHYS(extra_pages));
815 pfn_s = PFN_UP(addr);
816 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
817 extra_pages -= n_pfns;
818 xen_add_extra_mem(pfn_s, n_pfns);
819 xen_max_p2m_pfn = pfn_s + n_pfns;
820 } else
821 discard = true;
822 }
823
824 if (!discard)
825 xen_align_and_add_e820_region(addr, chunk_size, type);
826
827 addr += chunk_size;
828 size -= chunk_size;
829 if (size == 0) {
830 i++;
831 if (i < xen_e820_table.nr_entries) {
832 addr = xen_e820_table.entries[i].addr;
833 size = xen_e820_table.entries[i].size;
834 }
835 }
836 }
837
838 /*
839 * Set the rest as identity mapped, in case PCI BARs are
840 * located here.
841 */
842 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
843
844 /*
845 * In domU, the ISA region is normal, usable memory, but we
846 * reserve ISA memory anyway because too many things poke
847 * about in there.
848 */
849 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
850
851 e820__update_table(e820_table);
852
853 /*
854 * Check whether the kernel itself conflicts with the target E820 map.
855 * Failing now is better than running into weird problems later due
856 * to relocating (and even reusing) pages with kernel text or data.
857 */
858 if (xen_is_e820_reserved(__pa_symbol(_text),
859 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
860 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
861 BUG();
862 }
863
864 /*
865 * Check for a conflict of the hypervisor supplied page tables with
866 * the target E820 map.
867 */
868 xen_pt_check_e820();
869
870 xen_reserve_xen_mfnlist();
871
872 /* Check for a conflict of the initrd with the target E820 map. */
873 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
874 boot_params.hdr.ramdisk_size)) {
875 phys_addr_t new_area, start, size;
876
877 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
878 if (!new_area) {
879 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
880 BUG();
881 }
882
883 start = boot_params.hdr.ramdisk_image;
884 size = boot_params.hdr.ramdisk_size;
885 xen_phys_memcpy(new_area, start, size);
886 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
887 start, start + size, new_area, new_area + size);
888 memblock_free(start, size);
889 boot_params.hdr.ramdisk_image = new_area;
890 boot_params.ext_ramdisk_image = new_area >> 32;
891 }
892
893 /*
894 * Set identity map on non-RAM pages and prepare remapping the
895 * underlying RAM.
896 */
897 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
898
899 pr_info("Released %ld page(s)\n", xen_released_pages);
900
901 return "Xen";
902 }
903
register_callback(unsigned type,const void * func)904 static int register_callback(unsigned type, const void *func)
905 {
906 struct callback_register callback = {
907 .type = type,
908 .address = XEN_CALLBACK(__KERNEL_CS, func),
909 .flags = CALLBACKF_mask_events,
910 };
911
912 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
913 }
914
xen_enable_sysenter(void)915 void xen_enable_sysenter(void)
916 {
917 int ret;
918 unsigned sysenter_feature;
919
920 sysenter_feature = X86_FEATURE_SYSENTER32;
921
922 if (!boot_cpu_has(sysenter_feature))
923 return;
924
925 ret = register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat);
926 if(ret != 0)
927 setup_clear_cpu_cap(sysenter_feature);
928 }
929
xen_enable_syscall(void)930 void xen_enable_syscall(void)
931 {
932 int ret;
933
934 ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
935 if (ret != 0) {
936 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
937 /* Pretty fatal; 64-bit userspace has no other
938 mechanism for syscalls. */
939 }
940
941 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
942 ret = register_callback(CALLBACKTYPE_syscall32,
943 xen_entry_SYSCALL_compat);
944 if (ret != 0)
945 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
946 }
947 }
948
xen_pvmmu_arch_setup(void)949 static void __init xen_pvmmu_arch_setup(void)
950 {
951 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
952 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
953
954 HYPERVISOR_vm_assist(VMASST_CMD_enable,
955 VMASST_TYPE_pae_extended_cr3);
956
957 if (register_callback(CALLBACKTYPE_event,
958 xen_asm_exc_xen_hypervisor_callback) ||
959 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
960 BUG();
961
962 xen_enable_sysenter();
963 xen_enable_syscall();
964 }
965
966 /* This function is not called for HVM domains */
xen_arch_setup(void)967 void __init xen_arch_setup(void)
968 {
969 xen_panic_handler_init();
970 xen_pvmmu_arch_setup();
971
972 #ifdef CONFIG_ACPI
973 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
974 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
975 disable_acpi();
976 }
977 #endif
978
979 memcpy(boot_command_line, xen_start_info->cmd_line,
980 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
981 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
982
983 /* Set up idle, making sure it calls safe_halt() pvop */
984 disable_cpuidle();
985 disable_cpufreq();
986 WARN_ON(xen_set_default_idle());
987 #ifdef CONFIG_NUMA
988 numa_off = 1;
989 #endif
990 }
991