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>
38 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
40 /* Amount of extra memory space we add to the e820 ranges */
41 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
43 /* Number of pages released from the initial allocation. */
44 unsigned long xen_released_pages;
46 /* E820 map used during setting up memory. */
47 static struct e820_table xen_e820_table __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;
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
70 * 10x seems like a reasonable balance between scaling flexibility and
71 * leaving a practically usable system.
73 #define EXTRA_MEM_RATIO (10)
75 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
77 static void __init xen_parse_512gb(void)
82 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
86 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
89 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
92 xen_512gb_limit = val;
95 static void __init xen_add_extra_mem(unsigned long start_pfn,
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.
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;
111 /* Append to existing region. */
112 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
114 xen_extra_mem[i].n_pfns += n_pfns;
118 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
119 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
121 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
124 static void __init xen_del_extra_mem(unsigned long start_pfn,
125 unsigned long n_pfns)
128 unsigned long start_r, size_r;
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;
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;
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;
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));
157 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
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
165 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
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;
175 return IDENTITY_FRAME(pfn);
179 * Mark all pfns of extra mem as invalid in p2m list.
181 void __init xen_inv_extra_mem(void)
183 unsigned long pfn, pfn_s, pfn_e;
186 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
187 if (!xen_extra_mem[i].n_pfns)
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);
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.
201 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
203 const struct e820_entry *entry = xen_e820_table.entries;
205 unsigned long done = 0;
207 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
211 if (entry->type != E820_TYPE_RAM)
214 e_pfn = PFN_DOWN(entry->addr + entry->size);
216 /* We only care about E820 after this */
217 if (e_pfn <= *min_pfn)
220 s_pfn = PFN_UP(entry->addr);
222 /* If min_pfn falls within the E820 entry, we want to start
223 * at the min_pfn PFN.
225 if (s_pfn <= *min_pfn) {
226 done = e_pfn - *min_pfn;
228 done = e_pfn - s_pfn;
237 static int __init xen_free_mfn(unsigned long mfn)
239 struct xen_memory_reservation reservation = {
245 set_xen_guest_handle(reservation.extent_start, &mfn);
246 reservation.nr_extents = 1;
248 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
252 * This releases a chunk of memory and then does the identity map. It's used
253 * as a fallback if the remapping fails.
255 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
256 unsigned long end_pfn, unsigned long nr_pages)
258 unsigned long pfn, end;
261 WARN_ON(start_pfn > end_pfn);
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);
268 /* Make sure pfn exists to start with */
269 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
272 ret = xen_free_mfn(mfn);
273 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
276 xen_released_pages++;
277 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
283 set_phys_range_identity(start_pfn, end_pfn);
287 * Helper function to update the p2m and m2p tables and kernel mapping.
289 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
291 struct mmu_update update = {
292 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
297 if (!set_phys_to_machine(pfn, mfn)) {
298 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
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",
310 /* Update kernel mapping, but not for highmem. */
311 if (pfn >= PFN_UP(__pa(high_memory - 1)))
314 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
315 mfn_pte(mfn, PAGE_KERNEL), 0)) {
316 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
323 * This function updates the p2m and m2p tables with an identity map from
324 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
325 * original allocation at remap_pfn. The information needed for remapping is
326 * saved in the memory itself to avoid the need for allocating buffers. The
327 * complete remap information is contained in a list of MFNs each containing
328 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
329 * This enables us to preserve the original mfn sequence while doing the
330 * remapping at a time when the memory management is capable of allocating
331 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
334 static void __init xen_do_set_identity_and_remap_chunk(
335 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
337 unsigned long buf = (unsigned long)&xen_remap_buf;
338 unsigned long mfn_save, mfn;
339 unsigned long ident_pfn_iter, remap_pfn_iter;
340 unsigned long ident_end_pfn = start_pfn + size;
341 unsigned long left = size;
342 unsigned int i, chunk;
346 mfn_save = virt_to_mfn(buf);
348 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
349 ident_pfn_iter < ident_end_pfn;
350 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
351 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
353 /* Map first pfn to xen_remap_buf */
354 mfn = pfn_to_mfn(ident_pfn_iter);
355 set_pte_mfn(buf, mfn, PAGE_KERNEL);
357 /* Save mapping information in page */
358 xen_remap_buf.next_area_mfn = xen_remap_mfn;
359 xen_remap_buf.target_pfn = remap_pfn_iter;
360 xen_remap_buf.size = chunk;
361 for (i = 0; i < chunk; i++)
362 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
364 /* Put remap buf into list. */
367 /* Set identity map */
368 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
373 /* Restore old xen_remap_buf mapping */
374 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
378 * This function takes a contiguous pfn range that needs to be identity mapped
381 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
382 * 2) Calls the do_ function to actually do the mapping/remapping work.
384 * The goal is to not allocate additional memory but to remap the existing
385 * pages. In the case of an error the underlying memory is simply released back
386 * to Xen and not remapped.
388 static unsigned long __init xen_set_identity_and_remap_chunk(
389 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
390 unsigned long remap_pfn)
394 unsigned long n = end_pfn - start_pfn;
397 remap_pfn = nr_pages;
400 unsigned long cur_pfn = start_pfn + i;
401 unsigned long left = n - i;
402 unsigned long size = left;
403 unsigned long remap_range_size;
405 /* Do not remap pages beyond the current allocation */
406 if (cur_pfn >= nr_pages) {
407 /* Identity map remaining pages */
408 set_phys_range_identity(cur_pfn, cur_pfn + size);
411 if (cur_pfn + size > nr_pages)
412 size = nr_pages - cur_pfn;
414 remap_range_size = xen_find_pfn_range(&remap_pfn);
415 if (!remap_range_size) {
416 pr_warn("Unable to find available pfn range, not remapping identity pages\n");
417 xen_set_identity_and_release_chunk(cur_pfn,
418 cur_pfn + left, nr_pages);
421 /* Adjust size to fit in current e820 RAM region */
422 if (size > remap_range_size)
423 size = remap_range_size;
425 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
427 /* Update variables to reflect new mappings. */
433 * If the PFNs are currently mapped, the VA mapping also needs
434 * to be updated to be 1:1.
436 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
437 (void)HYPERVISOR_update_va_mapping(
438 (unsigned long)__va(pfn << PAGE_SHIFT),
439 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
444 static unsigned long __init xen_count_remap_pages(
445 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
446 unsigned long remap_pages)
448 if (start_pfn >= nr_pages)
451 return remap_pages + min(end_pfn, nr_pages) - start_pfn;
454 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
455 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
456 unsigned long nr_pages, unsigned long last_val))
458 phys_addr_t start = 0;
459 unsigned long ret_val = 0;
460 const struct e820_entry *entry = xen_e820_table.entries;
464 * Combine non-RAM regions and gaps until a RAM region (or the
465 * end of the map) is reached, then call the provided function
466 * to perform its duty on the non-RAM region.
468 * The combined non-RAM regions are rounded to a whole number
469 * of pages so any partial pages are accessible via the 1:1
470 * mapping. This is needed for some BIOSes that put (for
471 * example) the DMI tables in a reserved region that begins on
472 * a non-page boundary.
474 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
475 phys_addr_t end = entry->addr + entry->size;
476 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
477 unsigned long start_pfn = PFN_DOWN(start);
478 unsigned long end_pfn = PFN_UP(end);
480 if (entry->type == E820_TYPE_RAM)
481 end_pfn = PFN_UP(entry->addr);
483 if (start_pfn < end_pfn)
484 ret_val = func(start_pfn, end_pfn, nr_pages,
494 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
495 * The remap information (which mfn remap to which pfn) is contained in the
496 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
497 * This scheme allows to remap the different chunks in arbitrary order while
498 * the resulting mapping will be independent from the order.
500 void __init xen_remap_memory(void)
502 unsigned long buf = (unsigned long)&xen_remap_buf;
503 unsigned long mfn_save, pfn;
504 unsigned long remapped = 0;
506 unsigned long pfn_s = ~0UL;
507 unsigned long len = 0;
509 mfn_save = virt_to_mfn(buf);
511 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
512 /* Map the remap information */
513 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
515 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
517 pfn = xen_remap_buf.target_pfn;
518 for (i = 0; i < xen_remap_buf.size; i++) {
519 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
523 if (pfn_s == ~0UL || pfn == pfn_s) {
524 pfn_s = xen_remap_buf.target_pfn;
525 len += xen_remap_buf.size;
526 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
527 len += xen_remap_buf.size;
529 xen_del_extra_mem(pfn_s, len);
530 pfn_s = xen_remap_buf.target_pfn;
531 len = xen_remap_buf.size;
533 xen_remap_mfn = xen_remap_buf.next_area_mfn;
536 if (pfn_s != ~0UL && len)
537 xen_del_extra_mem(pfn_s, len);
539 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
541 pr_info("Remapped %ld page(s)\n", remapped);
544 static unsigned long __init xen_get_pages_limit(void)
549 limit = GB(64) / PAGE_SIZE;
551 limit = MAXMEM / PAGE_SIZE;
552 if (!xen_initial_domain() && xen_512gb_limit)
553 limit = GB(512) / PAGE_SIZE;
558 static unsigned long __init xen_get_max_pages(void)
560 unsigned long max_pages, limit;
561 domid_t domid = DOMID_SELF;
564 limit = xen_get_pages_limit();
568 * For the initial domain we use the maximum reservation as
571 * For guest domains the current maximum reservation reflects
572 * the current maximum rather than the static maximum. In this
573 * case the e820 map provided to us will cover the static
576 if (xen_initial_domain()) {
577 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
582 return min(max_pages, limit);
585 static void __init xen_align_and_add_e820_region(phys_addr_t start,
586 phys_addr_t size, int type)
588 phys_addr_t end = start + size;
590 /* Align RAM regions to page boundaries. */
591 if (type == E820_TYPE_RAM) {
592 start = PAGE_ALIGN(start);
593 end &= ~((phys_addr_t)PAGE_SIZE - 1);
594 #ifdef CONFIG_MEMORY_HOTPLUG
596 * Don't allow adding memory not in E820 map while booting the
597 * system. Once the balloon driver is up it will remove that
604 e820__range_add(start, end - start, type);
607 static void __init xen_ignore_unusable(void)
609 struct e820_entry *entry = xen_e820_table.entries;
612 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
613 if (entry->type == E820_TYPE_UNUSABLE)
614 entry->type = E820_TYPE_RAM;
618 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
620 struct e820_entry *entry;
628 entry = xen_e820_table.entries;
630 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
631 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
632 (entry->addr + entry->size) >= end)
642 * Find a free area in physical memory not yet reserved and compliant with
644 * Used to relocate pre-allocated areas like initrd or p2m list which are in
645 * conflict with the to be used E820 map.
646 * In case no area is found, return 0. Otherwise return the physical address
647 * of the area which is already reserved for convenience.
649 phys_addr_t __init xen_find_free_area(phys_addr_t size)
652 phys_addr_t addr, start;
653 struct e820_entry *entry = xen_e820_table.entries;
655 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
656 if (entry->type != E820_TYPE_RAM || entry->size < size)
659 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
660 if (!memblock_is_reserved(addr))
662 start = addr + PAGE_SIZE;
663 if (start + size > entry->addr + entry->size)
666 if (addr >= start + size) {
667 memblock_reserve(start, size);
676 * Like memcpy, but with physical addresses for dest and src.
678 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
681 phys_addr_t dest_off, src_off, dest_len, src_len, len;
685 dest_off = dest & ~PAGE_MASK;
686 src_off = src & ~PAGE_MASK;
688 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
689 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
691 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
692 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
693 len = min(dest_len, src_len);
694 to = early_memremap(dest - dest_off, dest_len + dest_off);
695 from = early_memremap(src - src_off, src_len + src_off);
696 memcpy(to, from, len);
697 early_memunmap(to, dest_len + dest_off);
698 early_memunmap(from, src_len + src_off);
706 * Reserve Xen mfn_list.
708 static void __init xen_reserve_xen_mfnlist(void)
710 phys_addr_t start, size;
712 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
713 start = __pa(xen_start_info->mfn_list);
714 size = PFN_ALIGN(xen_start_info->nr_pages *
715 sizeof(unsigned long));
717 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
718 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
721 memblock_reserve(start, size);
722 if (!xen_is_e820_reserved(start, size))
727 * Relocating the p2m on 32 bit system to an arbitrary virtual address
728 * is not supported, so just give up.
730 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
734 memblock_free(start, size);
739 * machine_specific_memory_setup - Hook for machine specific memory setup.
741 char * __init xen_memory_setup(void)
743 unsigned long max_pfn, pfn_s, n_pfns;
744 phys_addr_t mem_end, addr, size, chunk_size;
747 struct xen_memory_map memmap;
748 unsigned long max_pages;
749 unsigned long extra_pages = 0;
754 max_pfn = xen_get_pages_limit();
755 max_pfn = min(max_pfn, xen_start_info->nr_pages);
756 mem_end = PFN_PHYS(max_pfn);
758 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
759 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
761 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
762 xen_saved_max_mem_size = max_mem_size;
765 op = xen_initial_domain() ?
766 XENMEM_machine_memory_map :
768 rc = HYPERVISOR_memory_op(op, &memmap);
770 BUG_ON(xen_initial_domain());
771 memmap.nr_entries = 1;
772 xen_e820_table.entries[0].addr = 0ULL;
773 xen_e820_table.entries[0].size = mem_end;
774 /* 8MB slack (to balance backend allocations). */
775 xen_e820_table.entries[0].size += 8ULL << 20;
776 xen_e820_table.entries[0].type = E820_TYPE_RAM;
780 BUG_ON(memmap.nr_entries == 0);
781 xen_e820_table.nr_entries = memmap.nr_entries;
784 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
785 * regions, so if we're using the machine memory map leave the
786 * region as RAM as it is in the pseudo-physical map.
788 * UNUSABLE regions in domUs are not handled and will need
789 * a patch in the future.
791 if (xen_initial_domain())
792 xen_ignore_unusable();
794 /* Make sure the Xen-supplied memory map is well-ordered. */
795 e820__update_table(&xen_e820_table);
797 max_pages = xen_get_max_pages();
799 /* How many extra pages do we need due to remapping? */
800 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
802 if (max_pages > max_pfn)
803 extra_pages += max_pages - max_pfn;
806 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
807 * factor the base size. On non-highmem systems, the base
808 * size is the full initial memory allocation; on highmem it
809 * is limited to the max size of lowmem, so that it doesn't
810 * get completely filled.
812 * Make sure we have no memory above max_pages, as this area
813 * isn't handled by the p2m management.
815 * In principle there could be a problem in lowmem systems if
816 * the initial memory is also very large with respect to
817 * lowmem, but we won't try to deal with that here.
819 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
820 extra_pages, max_pages - max_pfn);
822 addr = xen_e820_table.entries[0].addr;
823 size = xen_e820_table.entries[0].size;
824 while (i < xen_e820_table.nr_entries) {
825 bool discard = false;
828 type = xen_e820_table.entries[i].type;
830 if (type == E820_TYPE_RAM) {
831 if (addr < mem_end) {
832 chunk_size = min(size, mem_end - addr);
833 } else if (extra_pages) {
834 chunk_size = min(size, PFN_PHYS(extra_pages));
835 pfn_s = PFN_UP(addr);
836 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
837 extra_pages -= n_pfns;
838 xen_add_extra_mem(pfn_s, n_pfns);
839 xen_max_p2m_pfn = pfn_s + n_pfns;
845 xen_align_and_add_e820_region(addr, chunk_size, type);
851 if (i < xen_e820_table.nr_entries) {
852 addr = xen_e820_table.entries[i].addr;
853 size = xen_e820_table.entries[i].size;
859 * Set the rest as identity mapped, in case PCI BARs are
862 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
865 * In domU, the ISA region is normal, usable memory, but we
866 * reserve ISA memory anyway because too many things poke
869 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
871 e820__update_table(e820_table);
874 * Check whether the kernel itself conflicts with the target E820 map.
875 * Failing now is better than running into weird problems later due
876 * to relocating (and even reusing) pages with kernel text or data.
878 if (xen_is_e820_reserved(__pa_symbol(_text),
879 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
880 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
885 * Check for a conflict of the hypervisor supplied page tables with
886 * the target E820 map.
890 xen_reserve_xen_mfnlist();
892 /* Check for a conflict of the initrd with the target E820 map. */
893 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
894 boot_params.hdr.ramdisk_size)) {
895 phys_addr_t new_area, start, size;
897 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
899 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
903 start = boot_params.hdr.ramdisk_image;
904 size = boot_params.hdr.ramdisk_size;
905 xen_phys_memcpy(new_area, start, size);
906 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
907 start, start + size, new_area, new_area + size);
908 memblock_free(start, size);
909 boot_params.hdr.ramdisk_image = new_area;
910 boot_params.ext_ramdisk_image = new_area >> 32;
914 * Set identity map on non-RAM pages and prepare remapping the
917 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
919 pr_info("Released %ld page(s)\n", xen_released_pages);
925 * Set the bit indicating "nosegneg" library variants should be used.
926 * We only need to bother in pure 32-bit mode; compat 32-bit processes
927 * can have un-truncated segments, so wrapping around is allowed.
929 static void __init fiddle_vdso(void)
932 u32 *mask = vdso_image_32.data +
933 vdso_image_32.sym_VDSO32_NOTE_MASK;
934 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
938 static int register_callback(unsigned type, const void *func)
940 struct callback_register callback = {
942 .address = XEN_CALLBACK(__KERNEL_CS, func),
943 .flags = CALLBACKF_mask_events,
946 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
949 void xen_enable_sysenter(void)
952 unsigned sysenter_feature;
955 sysenter_feature = X86_FEATURE_SEP;
957 sysenter_feature = X86_FEATURE_SYSENTER32;
960 if (!boot_cpu_has(sysenter_feature))
963 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
965 setup_clear_cpu_cap(sysenter_feature);
968 void xen_enable_syscall(void)
973 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
975 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
976 /* Pretty fatal; 64-bit userspace has no other
977 mechanism for syscalls. */
980 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
981 ret = register_callback(CALLBACKTYPE_syscall32,
982 xen_syscall32_target);
984 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
986 #endif /* CONFIG_X86_64 */
989 static void __init xen_pvmmu_arch_setup(void)
991 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
992 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
994 HYPERVISOR_vm_assist(VMASST_CMD_enable,
995 VMASST_TYPE_pae_extended_cr3);
997 if (register_callback(CALLBACKTYPE_event,
998 xen_asm_exc_xen_hypervisor_callback) ||
999 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1002 xen_enable_sysenter();
1003 xen_enable_syscall();
1006 /* This function is not called for HVM domains */
1007 void __init xen_arch_setup(void)
1009 xen_panic_handler_init();
1010 xen_pvmmu_arch_setup();
1013 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1014 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1019 memcpy(boot_command_line, xen_start_info->cmd_line,
1020 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1021 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1023 /* Set up idle, making sure it calls safe_halt() pvop */
1026 WARN_ON(xen_set_default_idle());