1 // SPDX-License-Identifier: GPL-2.0
3 * Machine specific setup for xen
5 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
8 #include <linux/init.h>
9 #include <linux/sched.h>
12 #include <linux/memblock.h>
13 #include <linux/cpuidle.h>
14 #include <linux/cpufreq.h>
15 #include <linux/memory_hotplug.h>
19 #include <asm/e820/api.h>
20 #include <asm/setup.h>
23 #include <asm/idtentry.h>
24 #include <asm/xen/hypervisor.h>
25 #include <asm/xen/hypercall.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>
37 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
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;
42 /* Number of pages released from the initial allocation. */
43 unsigned long xen_released_pages;
45 /* E820 map used during setting up memory. */
46 static struct e820_table xen_e820_table __initdata;
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
53 #define REMAP_SIZE (P2M_PER_PAGE - 3)
55 unsigned long next_area_mfn;
56 unsigned long target_pfn;
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;
62 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
64 static void __init xen_parse_512gb(void)
69 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
73 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
76 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
79 xen_512gb_limit = val;
82 static void __init xen_add_extra_mem(unsigned long start_pfn,
88 * No need to check for zero size, should happen rarely and will only
89 * write a new entry regarded to be unused due to zero size.
91 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
93 if (xen_extra_mem[i].n_pfns == 0) {
94 xen_extra_mem[i].start_pfn = start_pfn;
95 xen_extra_mem[i].n_pfns = n_pfns;
98 /* Append to existing region. */
99 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
101 xen_extra_mem[i].n_pfns += n_pfns;
105 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
106 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
108 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
111 static void __init xen_del_extra_mem(unsigned long start_pfn,
112 unsigned long n_pfns)
115 unsigned long start_r, size_r;
117 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
118 start_r = xen_extra_mem[i].start_pfn;
119 size_r = xen_extra_mem[i].n_pfns;
121 /* Start of region. */
122 if (start_r == start_pfn) {
123 BUG_ON(n_pfns > size_r);
124 xen_extra_mem[i].start_pfn += n_pfns;
125 xen_extra_mem[i].n_pfns -= n_pfns;
129 if (start_r + size_r == start_pfn + n_pfns) {
130 BUG_ON(n_pfns > size_r);
131 xen_extra_mem[i].n_pfns -= n_pfns;
135 if (start_pfn > start_r && start_pfn < start_r + size_r) {
136 BUG_ON(start_pfn + n_pfns > start_r + size_r);
137 xen_extra_mem[i].n_pfns = start_pfn - start_r;
138 /* Calling memblock_reserve() again is okay. */
139 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
140 (start_pfn + n_pfns));
144 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
148 * Called during boot before the p2m list can take entries beyond the
149 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
152 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
156 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
157 if (pfn >= xen_extra_mem[i].start_pfn &&
158 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
159 return INVALID_P2M_ENTRY;
162 return IDENTITY_FRAME(pfn);
166 * Mark all pfns of extra mem as invalid in p2m list.
168 void __init xen_inv_extra_mem(void)
170 unsigned long pfn, pfn_s, pfn_e;
173 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
174 if (!xen_extra_mem[i].n_pfns)
176 pfn_s = xen_extra_mem[i].start_pfn;
177 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
178 for (pfn = pfn_s; pfn < pfn_e; pfn++)
179 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
184 * Finds the next RAM pfn available in the E820 map after min_pfn.
185 * This function updates min_pfn with the pfn found and returns
186 * the size of that range or zero if not found.
188 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
190 const struct e820_entry *entry = xen_e820_table.entries;
192 unsigned long done = 0;
194 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
198 if (entry->type != E820_TYPE_RAM)
201 e_pfn = PFN_DOWN(entry->addr + entry->size);
203 /* We only care about E820 after this */
204 if (e_pfn <= *min_pfn)
207 s_pfn = PFN_UP(entry->addr);
209 /* If min_pfn falls within the E820 entry, we want to start
210 * at the min_pfn PFN.
212 if (s_pfn <= *min_pfn) {
213 done = e_pfn - *min_pfn;
215 done = e_pfn - s_pfn;
224 static int __init xen_free_mfn(unsigned long mfn)
226 struct xen_memory_reservation reservation = {
232 set_xen_guest_handle(reservation.extent_start, &mfn);
233 reservation.nr_extents = 1;
235 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
239 * This releases a chunk of memory and then does the identity map. It's used
240 * as a fallback if the remapping fails.
242 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
243 unsigned long end_pfn, unsigned long nr_pages)
245 unsigned long pfn, end;
248 WARN_ON(start_pfn > end_pfn);
250 /* Release pages first. */
251 end = min(end_pfn, nr_pages);
252 for (pfn = start_pfn; pfn < end; pfn++) {
253 unsigned long mfn = pfn_to_mfn(pfn);
255 /* Make sure pfn exists to start with */
256 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
259 ret = xen_free_mfn(mfn);
260 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
263 xen_released_pages++;
264 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
270 set_phys_range_identity(start_pfn, end_pfn);
274 * Helper function to update the p2m and m2p tables and kernel mapping.
276 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
278 struct mmu_update update = {
279 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
284 if (!set_phys_to_machine(pfn, mfn)) {
285 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
291 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
292 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
297 /* Update kernel mapping, but not for highmem. */
298 if (pfn >= PFN_UP(__pa(high_memory - 1)))
301 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
302 mfn_pte(mfn, PAGE_KERNEL), 0)) {
303 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
310 * This function updates the p2m and m2p tables with an identity map from
311 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
312 * original allocation at remap_pfn. The information needed for remapping is
313 * saved in the memory itself to avoid the need for allocating buffers. The
314 * complete remap information is contained in a list of MFNs each containing
315 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
316 * This enables us to preserve the original mfn sequence while doing the
317 * remapping at a time when the memory management is capable of allocating
318 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
321 static void __init xen_do_set_identity_and_remap_chunk(
322 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
324 unsigned long buf = (unsigned long)&xen_remap_buf;
325 unsigned long mfn_save, mfn;
326 unsigned long ident_pfn_iter, remap_pfn_iter;
327 unsigned long ident_end_pfn = start_pfn + size;
328 unsigned long left = size;
329 unsigned int i, chunk;
333 mfn_save = virt_to_mfn(buf);
335 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
336 ident_pfn_iter < ident_end_pfn;
337 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
338 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
340 /* Map first pfn to xen_remap_buf */
341 mfn = pfn_to_mfn(ident_pfn_iter);
342 set_pte_mfn(buf, mfn, PAGE_KERNEL);
344 /* Save mapping information in page */
345 xen_remap_buf.next_area_mfn = xen_remap_mfn;
346 xen_remap_buf.target_pfn = remap_pfn_iter;
347 xen_remap_buf.size = chunk;
348 for (i = 0; i < chunk; i++)
349 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
351 /* Put remap buf into list. */
354 /* Set identity map */
355 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
360 /* Restore old xen_remap_buf mapping */
361 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
365 * This function takes a contiguous pfn range that needs to be identity mapped
368 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
369 * 2) Calls the do_ function to actually do the mapping/remapping work.
371 * The goal is to not allocate additional memory but to remap the existing
372 * pages. In the case of an error the underlying memory is simply released back
373 * to Xen and not remapped.
375 static unsigned long __init xen_set_identity_and_remap_chunk(
376 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
377 unsigned long remap_pfn)
381 unsigned long n = end_pfn - start_pfn;
384 remap_pfn = nr_pages;
387 unsigned long cur_pfn = start_pfn + i;
388 unsigned long left = n - i;
389 unsigned long size = left;
390 unsigned long remap_range_size;
392 /* Do not remap pages beyond the current allocation */
393 if (cur_pfn >= nr_pages) {
394 /* Identity map remaining pages */
395 set_phys_range_identity(cur_pfn, cur_pfn + size);
398 if (cur_pfn + size > nr_pages)
399 size = nr_pages - cur_pfn;
401 remap_range_size = xen_find_pfn_range(&remap_pfn);
402 if (!remap_range_size) {
403 pr_warn("Unable to find available pfn range, not remapping identity pages\n");
404 xen_set_identity_and_release_chunk(cur_pfn,
405 cur_pfn + left, nr_pages);
408 /* Adjust size to fit in current e820 RAM region */
409 if (size > remap_range_size)
410 size = remap_range_size;
412 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
414 /* Update variables to reflect new mappings. */
420 * If the PFNs are currently mapped, the VA mapping also needs
421 * to be updated to be 1:1.
423 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
424 (void)HYPERVISOR_update_va_mapping(
425 (unsigned long)__va(pfn << PAGE_SHIFT),
426 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
431 static unsigned long __init xen_count_remap_pages(
432 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
433 unsigned long remap_pages)
435 if (start_pfn >= nr_pages)
438 return remap_pages + min(end_pfn, nr_pages) - start_pfn;
441 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
442 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
443 unsigned long nr_pages, unsigned long last_val))
445 phys_addr_t start = 0;
446 unsigned long ret_val = 0;
447 const struct e820_entry *entry = xen_e820_table.entries;
451 * Combine non-RAM regions and gaps until a RAM region (or the
452 * end of the map) is reached, then call the provided function
453 * to perform its duty on the non-RAM region.
455 * The combined non-RAM regions are rounded to a whole number
456 * of pages so any partial pages are accessible via the 1:1
457 * mapping. This is needed for some BIOSes that put (for
458 * example) the DMI tables in a reserved region that begins on
459 * a non-page boundary.
461 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
462 phys_addr_t end = entry->addr + entry->size;
463 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
464 unsigned long start_pfn = PFN_DOWN(start);
465 unsigned long end_pfn = PFN_UP(end);
467 if (entry->type == E820_TYPE_RAM)
468 end_pfn = PFN_UP(entry->addr);
470 if (start_pfn < end_pfn)
471 ret_val = func(start_pfn, end_pfn, nr_pages,
481 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
482 * The remap information (which mfn remap to which pfn) is contained in the
483 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
484 * This scheme allows to remap the different chunks in arbitrary order while
485 * the resulting mapping will be independent from the order.
487 void __init xen_remap_memory(void)
489 unsigned long buf = (unsigned long)&xen_remap_buf;
490 unsigned long mfn_save, pfn;
491 unsigned long remapped = 0;
493 unsigned long pfn_s = ~0UL;
494 unsigned long len = 0;
496 mfn_save = virt_to_mfn(buf);
498 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
499 /* Map the remap information */
500 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
502 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
504 pfn = xen_remap_buf.target_pfn;
505 for (i = 0; i < xen_remap_buf.size; i++) {
506 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
510 if (pfn_s == ~0UL || pfn == pfn_s) {
511 pfn_s = xen_remap_buf.target_pfn;
512 len += xen_remap_buf.size;
513 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
514 len += xen_remap_buf.size;
516 xen_del_extra_mem(pfn_s, len);
517 pfn_s = xen_remap_buf.target_pfn;
518 len = xen_remap_buf.size;
520 xen_remap_mfn = xen_remap_buf.next_area_mfn;
523 if (pfn_s != ~0UL && len)
524 xen_del_extra_mem(pfn_s, len);
526 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
528 pr_info("Remapped %ld page(s)\n", remapped);
531 static unsigned long __init xen_get_pages_limit(void)
535 limit = MAXMEM / PAGE_SIZE;
536 if (!xen_initial_domain() && xen_512gb_limit)
537 limit = GB(512) / PAGE_SIZE;
542 static unsigned long __init xen_get_max_pages(void)
544 unsigned long max_pages, limit;
545 domid_t domid = DOMID_SELF;
548 limit = xen_get_pages_limit();
552 * For the initial domain we use the maximum reservation as
555 * For guest domains the current maximum reservation reflects
556 * the current maximum rather than the static maximum. In this
557 * case the e820 map provided to us will cover the static
560 if (xen_initial_domain()) {
561 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
566 return min(max_pages, limit);
569 static void __init xen_align_and_add_e820_region(phys_addr_t start,
570 phys_addr_t size, int type)
572 phys_addr_t end = start + size;
574 /* Align RAM regions to page boundaries. */
575 if (type == E820_TYPE_RAM) {
576 start = PAGE_ALIGN(start);
577 end &= ~((phys_addr_t)PAGE_SIZE - 1);
578 #ifdef CONFIG_MEMORY_HOTPLUG
580 * Don't allow adding memory not in E820 map while booting the
581 * system. Once the balloon driver is up it will remove that
588 e820__range_add(start, end - start, type);
591 static void __init xen_ignore_unusable(void)
593 struct e820_entry *entry = xen_e820_table.entries;
596 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
597 if (entry->type == E820_TYPE_UNUSABLE)
598 entry->type = E820_TYPE_RAM;
602 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
604 struct e820_entry *entry;
612 entry = xen_e820_table.entries;
614 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
615 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
616 (entry->addr + entry->size) >= end)
626 * Find a free area in physical memory not yet reserved and compliant with
628 * Used to relocate pre-allocated areas like initrd or p2m list which are in
629 * conflict with the to be used E820 map.
630 * In case no area is found, return 0. Otherwise return the physical address
631 * of the area which is already reserved for convenience.
633 phys_addr_t __init xen_find_free_area(phys_addr_t size)
636 phys_addr_t addr, start;
637 struct e820_entry *entry = xen_e820_table.entries;
639 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
640 if (entry->type != E820_TYPE_RAM || entry->size < size)
643 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
644 if (!memblock_is_reserved(addr))
646 start = addr + PAGE_SIZE;
647 if (start + size > entry->addr + entry->size)
650 if (addr >= start + size) {
651 memblock_reserve(start, size);
660 * Like memcpy, but with physical addresses for dest and src.
662 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
665 phys_addr_t dest_off, src_off, dest_len, src_len, len;
669 dest_off = dest & ~PAGE_MASK;
670 src_off = src & ~PAGE_MASK;
672 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
673 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
675 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
676 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
677 len = min(dest_len, src_len);
678 to = early_memremap(dest - dest_off, dest_len + dest_off);
679 from = early_memremap(src - src_off, src_len + src_off);
680 memcpy(to, from, len);
681 early_memunmap(to, dest_len + dest_off);
682 early_memunmap(from, src_len + src_off);
690 * Reserve Xen mfn_list.
692 static void __init xen_reserve_xen_mfnlist(void)
694 phys_addr_t start, size;
696 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
697 start = __pa(xen_start_info->mfn_list);
698 size = PFN_ALIGN(xen_start_info->nr_pages *
699 sizeof(unsigned long));
701 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
702 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
705 memblock_reserve(start, size);
706 if (!xen_is_e820_reserved(start, size))
710 memblock_free(start, size);
714 * machine_specific_memory_setup - Hook for machine specific memory setup.
716 char * __init xen_memory_setup(void)
718 unsigned long max_pfn, pfn_s, n_pfns;
719 phys_addr_t mem_end, addr, size, chunk_size;
722 struct xen_memory_map memmap;
723 unsigned long max_pages;
724 unsigned long extra_pages = 0;
729 max_pfn = xen_get_pages_limit();
730 max_pfn = min(max_pfn, xen_start_info->nr_pages);
731 mem_end = PFN_PHYS(max_pfn);
733 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
734 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
736 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
737 xen_saved_max_mem_size = max_mem_size;
740 op = xen_initial_domain() ?
741 XENMEM_machine_memory_map :
743 rc = HYPERVISOR_memory_op(op, &memmap);
745 BUG_ON(xen_initial_domain());
746 memmap.nr_entries = 1;
747 xen_e820_table.entries[0].addr = 0ULL;
748 xen_e820_table.entries[0].size = mem_end;
749 /* 8MB slack (to balance backend allocations). */
750 xen_e820_table.entries[0].size += 8ULL << 20;
751 xen_e820_table.entries[0].type = E820_TYPE_RAM;
755 BUG_ON(memmap.nr_entries == 0);
756 xen_e820_table.nr_entries = memmap.nr_entries;
759 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
760 * regions, so if we're using the machine memory map leave the
761 * region as RAM as it is in the pseudo-physical map.
763 * UNUSABLE regions in domUs are not handled and will need
764 * a patch in the future.
766 if (xen_initial_domain())
767 xen_ignore_unusable();
769 /* Make sure the Xen-supplied memory map is well-ordered. */
770 e820__update_table(&xen_e820_table);
772 max_pages = xen_get_max_pages();
774 /* How many extra pages do we need due to remapping? */
775 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
777 if (max_pages > max_pfn)
778 extra_pages += max_pages - max_pfn;
781 * Clamp the amount of extra memory to a XEN_EXTRA_MEM_RATIO
782 * factor the base size.
784 * Make sure we have no memory above max_pages, as this area
785 * isn't handled by the p2m management.
787 extra_pages = min3(XEN_EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
788 extra_pages, max_pages - max_pfn);
790 addr = xen_e820_table.entries[0].addr;
791 size = xen_e820_table.entries[0].size;
792 while (i < xen_e820_table.nr_entries) {
793 bool discard = false;
796 type = xen_e820_table.entries[i].type;
798 if (type == E820_TYPE_RAM) {
799 if (addr < mem_end) {
800 chunk_size = min(size, mem_end - addr);
801 } else if (extra_pages) {
802 chunk_size = min(size, PFN_PHYS(extra_pages));
803 pfn_s = PFN_UP(addr);
804 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
805 extra_pages -= n_pfns;
806 xen_add_extra_mem(pfn_s, n_pfns);
807 xen_max_p2m_pfn = pfn_s + n_pfns;
813 xen_align_and_add_e820_region(addr, chunk_size, type);
819 if (i < xen_e820_table.nr_entries) {
820 addr = xen_e820_table.entries[i].addr;
821 size = xen_e820_table.entries[i].size;
827 * Set the rest as identity mapped, in case PCI BARs are
830 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
833 * In domU, the ISA region is normal, usable memory, but we
834 * reserve ISA memory anyway because too many things poke
837 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
839 e820__update_table(e820_table);
842 * Check whether the kernel itself conflicts with the target E820 map.
843 * Failing now is better than running into weird problems later due
844 * to relocating (and even reusing) pages with kernel text or data.
846 if (xen_is_e820_reserved(__pa_symbol(_text),
847 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
848 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
853 * Check for a conflict of the hypervisor supplied page tables with
854 * the target E820 map.
858 xen_reserve_xen_mfnlist();
860 /* Check for a conflict of the initrd with the target E820 map. */
861 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
862 boot_params.hdr.ramdisk_size)) {
863 phys_addr_t new_area, start, size;
865 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
867 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
871 start = boot_params.hdr.ramdisk_image;
872 size = boot_params.hdr.ramdisk_size;
873 xen_phys_memcpy(new_area, start, size);
874 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
875 start, start + size, new_area, new_area + size);
876 memblock_free(start, size);
877 boot_params.hdr.ramdisk_image = new_area;
878 boot_params.ext_ramdisk_image = new_area >> 32;
882 * Set identity map on non-RAM pages and prepare remapping the
885 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
887 pr_info("Released %ld page(s)\n", xen_released_pages);
892 static int register_callback(unsigned type, const void *func)
894 struct callback_register callback = {
896 .address = XEN_CALLBACK(__KERNEL_CS, func),
897 .flags = CALLBACKF_mask_events,
900 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
903 void xen_enable_sysenter(void)
906 unsigned sysenter_feature;
908 sysenter_feature = X86_FEATURE_SYSENTER32;
910 if (!boot_cpu_has(sysenter_feature))
913 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
915 setup_clear_cpu_cap(sysenter_feature);
918 void xen_enable_syscall(void)
922 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
924 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
925 /* Pretty fatal; 64-bit userspace has no other
926 mechanism for syscalls. */
929 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
930 ret = register_callback(CALLBACKTYPE_syscall32,
931 xen_syscall32_target);
933 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
937 static void __init xen_pvmmu_arch_setup(void)
939 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
940 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
942 HYPERVISOR_vm_assist(VMASST_CMD_enable,
943 VMASST_TYPE_pae_extended_cr3);
945 if (register_callback(CALLBACKTYPE_event,
946 xen_asm_exc_xen_hypervisor_callback) ||
947 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
950 xen_enable_sysenter();
951 xen_enable_syscall();
954 /* This function is not called for HVM domains */
955 void __init xen_arch_setup(void)
957 xen_panic_handler_init();
958 xen_pvmmu_arch_setup();
961 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
962 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
967 memcpy(boot_command_line, xen_start_info->cmd_line,
968 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
969 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
971 /* Set up idle, making sure it calls safe_halt() pvop */
974 WARN_ON(xen_set_default_idle());