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/iscsi_ibft.h>
10 #include <linux/sched.h>
11 #include <linux/kstrtox.h>
14 #include <linux/memblock.h>
15 #include <linux/cpuidle.h>
16 #include <linux/cpufreq.h>
17 #include <linux/memory_hotplug.h>
18 #include <linux/acpi.h>
22 #include <asm/e820/api.h>
23 #include <asm/setup.h>
25 #include <asm/idtentry.h>
26 #include <asm/xen/hypervisor.h>
27 #include <asm/xen/hypercall.h>
31 #include <xen/interface/callback.h>
32 #include <xen/interface/memory.h>
33 #include <xen/interface/physdev.h>
34 #include <xen/features.h>
35 #include <xen/hvc-console.h>
38 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
40 /* Memory map would allow PCI passthrough. */
41 bool xen_pv_pci_possible;
43 /* E820 map used during setting up memory. */
44 static struct e820_table xen_e820_table __initdata;
46 /* Number of initially usable memory pages. */
47 static unsigned long ini_nr_pages __initdata;
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
54 #define REMAP_SIZE (P2M_PER_PAGE - 3)
56 unsigned long next_area_mfn;
57 unsigned long target_pfn;
59 unsigned long mfns[REMAP_SIZE];
60 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
61 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
63 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
65 static void __init xen_parse_512gb(void)
70 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
74 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
77 else if (kstrtobool(arg + strlen("xen_512gb_limit="), &val))
80 xen_512gb_limit = val;
83 static void __init xen_del_extra_mem(unsigned long start_pfn,
87 unsigned long start_r, size_r;
89 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
90 start_r = xen_extra_mem[i].start_pfn;
91 size_r = xen_extra_mem[i].n_pfns;
93 /* Start of region. */
94 if (start_r == start_pfn) {
95 BUG_ON(n_pfns > size_r);
96 xen_extra_mem[i].start_pfn += n_pfns;
97 xen_extra_mem[i].n_pfns -= n_pfns;
101 if (start_r + size_r == start_pfn + n_pfns) {
102 BUG_ON(n_pfns > size_r);
103 xen_extra_mem[i].n_pfns -= n_pfns;
107 if (start_pfn > start_r && start_pfn < start_r + size_r) {
108 BUG_ON(start_pfn + n_pfns > start_r + size_r);
109 xen_extra_mem[i].n_pfns = start_pfn - start_r;
110 /* Calling memblock_reserve() again is okay. */
111 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
112 (start_pfn + n_pfns));
116 memblock_phys_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
120 * Called during boot before the p2m list can take entries beyond the
121 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
124 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
128 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
129 if (pfn >= xen_extra_mem[i].start_pfn &&
130 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
131 return INVALID_P2M_ENTRY;
134 return IDENTITY_FRAME(pfn);
138 * Mark all pfns of extra mem as invalid in p2m list.
140 void __init xen_inv_extra_mem(void)
142 unsigned long pfn, pfn_s, pfn_e;
145 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
146 if (!xen_extra_mem[i].n_pfns)
148 pfn_s = xen_extra_mem[i].start_pfn;
149 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
150 for (pfn = pfn_s; pfn < pfn_e; pfn++)
151 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
156 * Finds the next RAM pfn available in the E820 map after min_pfn.
157 * This function updates min_pfn with the pfn found and returns
158 * the size of that range or zero if not found.
160 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
162 const struct e820_entry *entry = xen_e820_table.entries;
164 unsigned long done = 0;
166 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
170 if (entry->type != E820_TYPE_RAM)
173 e_pfn = PFN_DOWN(entry->addr + entry->size);
175 /* We only care about E820 after this */
176 if (e_pfn <= *min_pfn)
179 s_pfn = PFN_UP(entry->addr);
181 /* If min_pfn falls within the E820 entry, we want to start
182 * at the min_pfn PFN.
184 if (s_pfn <= *min_pfn) {
185 done = e_pfn - *min_pfn;
187 done = e_pfn - s_pfn;
196 static int __init xen_free_mfn(unsigned long mfn)
198 struct xen_memory_reservation reservation = {
204 set_xen_guest_handle(reservation.extent_start, &mfn);
205 reservation.nr_extents = 1;
207 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
211 * This releases a chunk of memory and then does the identity map. It's used
212 * as a fallback if the remapping fails.
214 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
215 unsigned long end_pfn)
217 unsigned long pfn, end;
220 WARN_ON(start_pfn > end_pfn);
222 /* Release pages first. */
223 end = min(end_pfn, ini_nr_pages);
224 for (pfn = start_pfn; pfn < end; pfn++) {
225 unsigned long mfn = pfn_to_mfn(pfn);
227 /* Make sure pfn exists to start with */
228 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
231 ret = xen_free_mfn(mfn);
232 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
235 xen_released_pages++;
236 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
242 set_phys_range_identity(start_pfn, end_pfn);
246 * Helper function to update the p2m and m2p tables and kernel mapping.
248 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
250 struct mmu_update update = {
251 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
256 if (!set_phys_to_machine(pfn, mfn)) {
257 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
263 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
264 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
269 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
270 mfn_pte(mfn, PAGE_KERNEL), 0)) {
271 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
278 * This function updates the p2m and m2p tables with an identity map from
279 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
280 * original allocation at remap_pfn. The information needed for remapping is
281 * saved in the memory itself to avoid the need for allocating buffers. The
282 * complete remap information is contained in a list of MFNs each containing
283 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
284 * This enables us to preserve the original mfn sequence while doing the
285 * remapping at a time when the memory management is capable of allocating
286 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
289 static void __init xen_do_set_identity_and_remap_chunk(
290 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
292 unsigned long buf = (unsigned long)&xen_remap_buf;
293 unsigned long mfn_save, mfn;
294 unsigned long ident_pfn_iter, remap_pfn_iter;
295 unsigned long ident_end_pfn = start_pfn + size;
296 unsigned long left = size;
297 unsigned int i, chunk;
301 mfn_save = virt_to_mfn((void *)buf);
303 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
304 ident_pfn_iter < ident_end_pfn;
305 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
306 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
308 /* Map first pfn to xen_remap_buf */
309 mfn = pfn_to_mfn(ident_pfn_iter);
310 set_pte_mfn(buf, mfn, PAGE_KERNEL);
312 /* Save mapping information in page */
313 xen_remap_buf.next_area_mfn = xen_remap_mfn;
314 xen_remap_buf.target_pfn = remap_pfn_iter;
315 xen_remap_buf.size = chunk;
316 for (i = 0; i < chunk; i++)
317 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
319 /* Put remap buf into list. */
322 /* Set identity map */
323 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
328 /* Restore old xen_remap_buf mapping */
329 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
333 * This function takes a contiguous pfn range that needs to be identity mapped
336 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
337 * 2) Calls the do_ function to actually do the mapping/remapping work.
339 * The goal is to not allocate additional memory but to remap the existing
340 * pages. In the case of an error the underlying memory is simply released back
341 * to Xen and not remapped.
343 static unsigned long __init xen_set_identity_and_remap_chunk(
344 unsigned long start_pfn, unsigned long end_pfn, unsigned long remap_pfn)
348 unsigned long n = end_pfn - start_pfn;
351 remap_pfn = ini_nr_pages;
354 unsigned long cur_pfn = start_pfn + i;
355 unsigned long left = n - i;
356 unsigned long size = left;
357 unsigned long remap_range_size;
359 /* Do not remap pages beyond the current allocation */
360 if (cur_pfn >= ini_nr_pages) {
361 /* Identity map remaining pages */
362 set_phys_range_identity(cur_pfn, cur_pfn + size);
365 if (cur_pfn + size > ini_nr_pages)
366 size = ini_nr_pages - cur_pfn;
368 remap_range_size = xen_find_pfn_range(&remap_pfn);
369 if (!remap_range_size) {
370 pr_warn("Unable to find available pfn range, not remapping identity pages\n");
371 xen_set_identity_and_release_chunk(cur_pfn,
375 /* Adjust size to fit in current e820 RAM region */
376 if (size > remap_range_size)
377 size = remap_range_size;
379 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
381 /* Update variables to reflect new mappings. */
387 * If the PFNs are currently mapped, their VA mappings need to be
390 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
391 (void)HYPERVISOR_update_va_mapping(
392 (unsigned long)__va(pfn << PAGE_SHIFT),
393 native_make_pte(0), 0);
398 static unsigned long __init xen_count_remap_pages(
399 unsigned long start_pfn, unsigned long end_pfn,
400 unsigned long remap_pages)
402 if (start_pfn >= ini_nr_pages)
405 return remap_pages + min(end_pfn, ini_nr_pages) - start_pfn;
408 static unsigned long __init xen_foreach_remap_area(
409 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
410 unsigned long last_val))
412 phys_addr_t start = 0;
413 unsigned long ret_val = 0;
414 const struct e820_entry *entry = xen_e820_table.entries;
418 * Combine non-RAM regions and gaps until a RAM region (or the
419 * end of the map) is reached, then call the provided function
420 * to perform its duty on the non-RAM region.
422 * The combined non-RAM regions are rounded to a whole number
423 * of pages so any partial pages are accessible via the 1:1
424 * mapping. This is needed for some BIOSes that put (for
425 * example) the DMI tables in a reserved region that begins on
426 * a non-page boundary.
428 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
429 phys_addr_t end = entry->addr + entry->size;
430 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
431 unsigned long start_pfn = PFN_DOWN(start);
432 unsigned long end_pfn = PFN_UP(end);
434 if (entry->type == E820_TYPE_RAM)
435 end_pfn = PFN_UP(entry->addr);
437 if (start_pfn < end_pfn)
438 ret_val = func(start_pfn, end_pfn, ret_val);
447 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
448 * The remap information (which mfn remap to which pfn) is contained in the
449 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
450 * This scheme allows to remap the different chunks in arbitrary order while
451 * the resulting mapping will be independent from the order.
453 void __init xen_remap_memory(void)
455 unsigned long buf = (unsigned long)&xen_remap_buf;
456 unsigned long mfn_save, pfn;
457 unsigned long remapped = 0;
459 unsigned long pfn_s = ~0UL;
460 unsigned long len = 0;
462 mfn_save = virt_to_mfn((void *)buf);
464 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
465 /* Map the remap information */
466 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
468 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
470 pfn = xen_remap_buf.target_pfn;
471 for (i = 0; i < xen_remap_buf.size; i++) {
472 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
476 if (pfn_s == ~0UL || pfn == pfn_s) {
477 pfn_s = xen_remap_buf.target_pfn;
478 len += xen_remap_buf.size;
479 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
480 len += xen_remap_buf.size;
482 xen_del_extra_mem(pfn_s, len);
483 pfn_s = xen_remap_buf.target_pfn;
484 len = xen_remap_buf.size;
486 xen_remap_mfn = xen_remap_buf.next_area_mfn;
489 if (pfn_s != ~0UL && len)
490 xen_del_extra_mem(pfn_s, len);
492 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
494 pr_info("Remapped %ld page(s)\n", remapped);
496 xen_do_remap_nonram();
499 static unsigned long __init xen_get_pages_limit(void)
503 limit = MAXMEM / PAGE_SIZE;
504 if (!xen_initial_domain() && xen_512gb_limit)
505 limit = GB(512) / PAGE_SIZE;
510 static unsigned long __init xen_get_max_pages(void)
512 unsigned long max_pages, limit;
513 domid_t domid = DOMID_SELF;
516 limit = xen_get_pages_limit();
520 * For the initial domain we use the maximum reservation as
523 * For guest domains the current maximum reservation reflects
524 * the current maximum rather than the static maximum. In this
525 * case the e820 map provided to us will cover the static
528 if (xen_initial_domain()) {
529 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
534 return min(max_pages, limit);
537 static void __init xen_align_and_add_e820_region(phys_addr_t start,
538 phys_addr_t size, int type)
540 phys_addr_t end = start + size;
542 /* Align RAM regions to page boundaries. */
543 if (type == E820_TYPE_RAM) {
544 start = PAGE_ALIGN(start);
545 end &= ~((phys_addr_t)PAGE_SIZE - 1);
546 #ifdef CONFIG_MEMORY_HOTPLUG
548 * Don't allow adding memory not in E820 map while booting the
549 * system. Once the balloon driver is up it will remove that
556 e820__range_add(start, end - start, type);
559 static void __init xen_ignore_unusable(void)
561 struct e820_entry *entry = xen_e820_table.entries;
564 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
565 if (entry->type == E820_TYPE_UNUSABLE)
566 entry->type = E820_TYPE_RAM;
570 static bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
572 struct e820_entry *entry;
580 entry = xen_e820_table.entries;
582 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
583 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
584 (entry->addr + entry->size) >= end)
594 * Find a free area in physical memory not yet reserved and compliant with
596 * Used to relocate pre-allocated areas like initrd or p2m list which are in
597 * conflict with the to be used E820 map.
598 * In case no area is found, return 0. Otherwise return the physical address
599 * of the area which is already reserved for convenience.
601 phys_addr_t __init xen_find_free_area(phys_addr_t size)
604 phys_addr_t addr, start;
605 struct e820_entry *entry = xen_e820_table.entries;
607 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
608 if (entry->type != E820_TYPE_RAM || entry->size < size)
611 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
612 if (!memblock_is_reserved(addr))
614 start = addr + PAGE_SIZE;
615 if (start + size > entry->addr + entry->size)
618 if (addr >= start + size) {
619 memblock_reserve(start, size);
628 * Swap a non-RAM E820 map entry with RAM above ini_nr_pages.
629 * Note that the E820 map is modified accordingly, but the P2M map isn't yet.
630 * The adaption of the P2M must be deferred until page allocation is possible.
632 static void __init xen_e820_swap_entry_with_ram(struct e820_entry *swap_entry)
634 struct e820_entry *entry;
636 phys_addr_t mem_end = PFN_PHYS(ini_nr_pages);
637 phys_addr_t swap_addr, swap_size, entry_end;
639 swap_addr = PAGE_ALIGN_DOWN(swap_entry->addr);
640 swap_size = PAGE_ALIGN(swap_entry->addr - swap_addr + swap_entry->size);
641 entry = xen_e820_table.entries;
643 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
644 entry_end = entry->addr + entry->size;
645 if (entry->type == E820_TYPE_RAM && entry->size >= swap_size &&
646 entry_end - swap_size >= mem_end) {
647 /* Reduce RAM entry by needed space (whole pages). */
648 entry->size -= swap_size;
650 /* Add new entry at the end of E820 map. */
651 entry = xen_e820_table.entries +
652 xen_e820_table.nr_entries;
653 xen_e820_table.nr_entries++;
655 /* Fill new entry (keep size and page offset). */
656 entry->type = swap_entry->type;
657 entry->addr = entry_end - swap_size +
658 swap_addr - swap_entry->addr;
659 entry->size = swap_entry->size;
661 /* Convert old entry to RAM, align to pages. */
662 swap_entry->type = E820_TYPE_RAM;
663 swap_entry->addr = swap_addr;
664 swap_entry->size = swap_size;
666 /* Remember PFN<->MFN relation for P2M update. */
667 xen_add_remap_nonram(swap_addr, entry_end - swap_size,
670 /* Order E820 table and merge entries. */
671 e820__update_table(&xen_e820_table);
679 xen_raw_console_write("No suitable area found for required E820 entry remapping action\n");
684 * Look for non-RAM memory types in a specific guest physical area and move
685 * those away if possible (ACPI NVS only for now).
687 static void __init xen_e820_resolve_conflicts(phys_addr_t start,
690 struct e820_entry *entry;
698 entry = xen_e820_table.entries;
700 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
701 if (entry->addr >= end)
704 if (entry->addr + entry->size > start &&
705 entry->type == E820_TYPE_NVS)
706 xen_e820_swap_entry_with_ram(entry);
713 * Check for an area in physical memory to be usable for non-movable purposes.
714 * An area is considered to usable if the used E820 map lists it to be RAM or
715 * some other type which can be moved to higher PFNs while keeping the MFNs.
716 * In case the area is not usable, crash the system with an error message.
718 void __init xen_chk_is_e820_usable(phys_addr_t start, phys_addr_t size,
719 const char *component)
721 xen_e820_resolve_conflicts(start, size);
723 if (!xen_is_e820_reserved(start, size))
726 xen_raw_console_write("Xen hypervisor allocated ");
727 xen_raw_console_write(component);
728 xen_raw_console_write(" memory conflicts with E820 map\n");
733 * Like memcpy, but with physical addresses for dest and src.
735 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
738 phys_addr_t dest_off, src_off, dest_len, src_len, len;
742 dest_off = dest & ~PAGE_MASK;
743 src_off = src & ~PAGE_MASK;
745 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
746 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
748 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
749 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
750 len = min(dest_len, src_len);
751 to = early_memremap(dest - dest_off, dest_len + dest_off);
752 from = early_memremap(src - src_off, src_len + src_off);
753 memcpy(to, from, len);
754 early_memunmap(to, dest_len + dest_off);
755 early_memunmap(from, src_len + src_off);
763 * Reserve Xen mfn_list.
765 static void __init xen_reserve_xen_mfnlist(void)
767 phys_addr_t start, size;
769 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
770 start = __pa(xen_start_info->mfn_list);
771 size = PFN_ALIGN(xen_start_info->nr_pages *
772 sizeof(unsigned long));
774 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
775 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
778 memblock_reserve(start, size);
779 if (!xen_is_e820_reserved(start, size))
783 memblock_phys_free(start, size);
787 * xen_memory_setup - Hook for machine specific memory setup.
789 char * __init xen_memory_setup(void)
791 unsigned long pfn_s, n_pfns;
792 phys_addr_t mem_end, addr, size, chunk_size;
795 struct xen_memory_map memmap;
796 unsigned long max_pages;
797 unsigned long extra_pages = 0;
798 unsigned long maxmem_pages;
803 ini_nr_pages = min(xen_get_pages_limit(), xen_start_info->nr_pages);
804 mem_end = PFN_PHYS(ini_nr_pages);
806 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
807 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
809 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
810 xen_saved_max_mem_size = max_mem_size;
813 op = xen_initial_domain() ?
814 XENMEM_machine_memory_map :
816 rc = HYPERVISOR_memory_op(op, &memmap);
818 BUG_ON(xen_initial_domain());
819 memmap.nr_entries = 1;
820 xen_e820_table.entries[0].addr = 0ULL;
821 xen_e820_table.entries[0].size = mem_end;
822 /* 8MB slack (to balance backend allocations). */
823 xen_e820_table.entries[0].size += 8ULL << 20;
824 xen_e820_table.entries[0].type = E820_TYPE_RAM;
828 BUG_ON(memmap.nr_entries == 0);
829 xen_e820_table.nr_entries = memmap.nr_entries;
831 if (xen_initial_domain()) {
833 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
834 * regions, so if we're using the machine memory map leave the
835 * region as RAM as it is in the pseudo-physical map.
837 * UNUSABLE regions in domUs are not handled and will need
838 * a patch in the future.
840 xen_ignore_unusable();
842 #ifdef CONFIG_ISCSI_IBFT_FIND
843 /* Reserve 0.5 MiB to 1 MiB region so iBFT can be found */
844 xen_e820_table.entries[xen_e820_table.nr_entries].addr = IBFT_START;
845 xen_e820_table.entries[xen_e820_table.nr_entries].size = IBFT_END - IBFT_START;
846 xen_e820_table.entries[xen_e820_table.nr_entries].type = E820_TYPE_RESERVED;
847 xen_e820_table.nr_entries++;
851 /* Make sure the Xen-supplied memory map is well-ordered. */
852 e820__update_table(&xen_e820_table);
855 * Check whether the kernel itself conflicts with the target E820 map.
856 * Failing now is better than running into weird problems later due
857 * to relocating (and even reusing) pages with kernel text or data.
859 xen_chk_is_e820_usable(__pa_symbol(_text),
860 __pa_symbol(_end) - __pa_symbol(_text),
864 * Check for a conflict of the xen_start_info memory with the target
867 xen_chk_is_e820_usable(__pa(xen_start_info), sizeof(*xen_start_info),
871 * Check for a conflict of the hypervisor supplied page tables with
872 * the target E820 map.
876 max_pages = xen_get_max_pages();
878 /* How many extra pages do we need due to remapping? */
879 max_pages += xen_foreach_remap_area(xen_count_remap_pages);
881 if (max_pages > ini_nr_pages)
882 extra_pages += max_pages - ini_nr_pages;
885 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
886 * factor the base size.
888 * Make sure we have no memory above max_pages, as this area
889 * isn't handled by the p2m management.
891 maxmem_pages = EXTRA_MEM_RATIO * min(ini_nr_pages, PFN_DOWN(MAXMEM));
892 extra_pages = min3(maxmem_pages, extra_pages, max_pages - ini_nr_pages);
894 addr = xen_e820_table.entries[0].addr;
895 size = xen_e820_table.entries[0].size;
896 while (i < xen_e820_table.nr_entries) {
897 bool discard = false;
900 type = xen_e820_table.entries[i].type;
902 if (type == E820_TYPE_RESERVED)
903 xen_pv_pci_possible = true;
905 if (type == E820_TYPE_RAM) {
906 if (addr < mem_end) {
907 chunk_size = min(size, mem_end - addr);
908 } else if (extra_pages) {
909 chunk_size = min(size, PFN_PHYS(extra_pages));
910 pfn_s = PFN_UP(addr);
911 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
912 extra_pages -= n_pfns;
913 xen_add_extra_mem(pfn_s, n_pfns);
914 xen_max_p2m_pfn = pfn_s + n_pfns;
920 xen_align_and_add_e820_region(addr, chunk_size, type);
926 if (i < xen_e820_table.nr_entries) {
927 addr = xen_e820_table.entries[i].addr;
928 size = xen_e820_table.entries[i].size;
934 * Set the rest as identity mapped, in case PCI BARs are
937 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
940 * In domU, the ISA region is normal, usable memory, but we
941 * reserve ISA memory anyway because too many things poke
944 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
946 e820__update_table(e820_table);
948 xen_reserve_xen_mfnlist();
950 /* Check for a conflict of the initrd with the target E820 map. */
951 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
952 boot_params.hdr.ramdisk_size)) {
953 phys_addr_t new_area, start, size;
955 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
957 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
961 start = boot_params.hdr.ramdisk_image;
962 size = boot_params.hdr.ramdisk_size;
963 xen_phys_memcpy(new_area, start, size);
964 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
965 start, start + size, new_area, new_area + size);
966 memblock_phys_free(start, size);
967 boot_params.hdr.ramdisk_image = new_area;
968 boot_params.ext_ramdisk_image = new_area >> 32;
972 * Set identity map on non-RAM pages and prepare remapping the
975 xen_foreach_remap_area(xen_set_identity_and_remap_chunk);
977 pr_info("Released %ld page(s)\n", xen_released_pages);
982 static int register_callback(unsigned type, const void *func)
984 struct callback_register callback = {
986 .address = XEN_CALLBACK(__KERNEL_CS, func),
987 .flags = CALLBACKF_mask_events,
990 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
993 void xen_enable_sysenter(void)
995 if (cpu_feature_enabled(X86_FEATURE_SYSENTER32) &&
996 register_callback(CALLBACKTYPE_sysenter, xen_entry_SYSENTER_compat))
997 setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
1000 void xen_enable_syscall(void)
1004 ret = register_callback(CALLBACKTYPE_syscall, xen_entry_SYSCALL_64);
1006 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
1007 /* Pretty fatal; 64-bit userspace has no other
1008 mechanism for syscalls. */
1011 if (cpu_feature_enabled(X86_FEATURE_SYSCALL32) &&
1012 register_callback(CALLBACKTYPE_syscall32, xen_entry_SYSCALL_compat))
1013 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1016 static void __init xen_pvmmu_arch_setup(void)
1018 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1020 if (register_callback(CALLBACKTYPE_event,
1021 xen_asm_exc_xen_hypervisor_callback) ||
1022 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1025 xen_enable_sysenter();
1026 xen_enable_syscall();
1029 /* This function is not called for HVM domains */
1030 void __init xen_arch_setup(void)
1032 xen_panic_handler_init();
1033 xen_pvmmu_arch_setup();
1036 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1037 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1042 memcpy(boot_command_line, xen_start_info->cmd_line,
1043 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1044 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1046 /* Set up idle, making sure it calls safe_halt() pvop */
1049 WARN_ON(xen_set_default_idle());