2 * Machine specific setup for xen
4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
7 #include <linux/module.h>
8 #include <linux/sched.h>
11 #include <linux/memblock.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpufreq.h>
18 #include <asm/setup.h>
20 #include <asm/xen/hypervisor.h>
21 #include <asm/xen/hypercall.h>
25 #include <xen/interface/callback.h>
26 #include <xen/interface/memory.h>
27 #include <xen/interface/physdev.h>
28 #include <xen/features.h>
32 /* These are code, but not functions. Defined in entry.S */
33 extern const char xen_hypervisor_callback[];
34 extern const char xen_failsafe_callback[];
35 extern void xen_sysenter_target(void);
36 extern void xen_syscall_target(void);
37 extern void xen_syscall32_target(void);
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;
46 * The maximum amount of extra memory compared to the base size. The
47 * main scaling factor is the size of struct page. At extreme ratios
48 * of base:extra, all the base memory can be filled with page
49 * structures for the extra memory, leaving no space for anything
52 * 10x seems like a reasonable balance between scaling flexibility and
53 * leaving a practically usable system.
55 #define EXTRA_MEM_RATIO (10)
57 static void __init xen_add_extra_mem(u64 start, u64 size)
62 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
64 if (xen_extra_mem[i].size == 0) {
65 xen_extra_mem[i].start = start;
66 xen_extra_mem[i].size = size;
69 /* Append to existing region. */
70 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
71 xen_extra_mem[i].size += size;
75 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
76 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
78 memblock_reserve(start, size);
80 xen_max_p2m_pfn = PFN_DOWN(start + size);
82 for (pfn = PFN_DOWN(start); pfn <= xen_max_p2m_pfn; pfn++)
83 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
86 static unsigned long __init xen_do_chunk(unsigned long start,
87 unsigned long end, bool release)
89 struct xen_memory_reservation reservation = {
94 unsigned long len = 0;
98 for (pfn = start; pfn < end; pfn++) {
100 unsigned long mfn = pfn_to_mfn(pfn);
103 /* Make sure pfn exists to start with */
104 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
108 if (mfn != INVALID_P2M_ENTRY)
112 set_xen_guest_handle(reservation.extent_start, &frame);
113 reservation.nr_extents = 1;
115 ret = HYPERVISOR_memory_op(release ? XENMEM_decrease_reservation : XENMEM_populate_physmap,
117 WARN(ret != 1, "Failed to %s pfn %lx err=%d\n",
118 release ? "release" : "populate", pfn, ret);
121 if (!early_set_phys_to_machine(pfn, release ? INVALID_P2M_ENTRY : frame)) {
124 set_xen_guest_handle(reservation.extent_start, &frame);
125 reservation.nr_extents = 1;
126 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
135 printk(KERN_INFO "%s %lx-%lx pfn range: %lu pages %s\n",
136 release ? "Freeing" : "Populating",
138 release ? "freed" : "added");
143 static unsigned long __init xen_release_chunk(unsigned long start,
146 return xen_do_chunk(start, end, true);
149 static unsigned long __init xen_populate_chunk(
150 const struct e820entry *list, size_t map_size,
151 unsigned long max_pfn, unsigned long *last_pfn,
152 unsigned long credits_left)
154 const struct e820entry *entry;
156 unsigned long done = 0;
157 unsigned long dest_pfn;
159 for (i = 0, entry = list; i < map_size; i++, entry++) {
165 if (credits_left <= 0)
168 if (entry->type != E820_RAM)
171 e_pfn = PFN_DOWN(entry->addr + entry->size);
173 /* We only care about E820 after the xen_start_info->nr_pages */
174 if (e_pfn <= max_pfn)
177 s_pfn = PFN_UP(entry->addr);
178 /* If the E820 falls within the nr_pages, we want to start
179 * at the nr_pages PFN.
180 * If that would mean going past the E820 entry, skip it
182 if (s_pfn <= max_pfn) {
183 capacity = e_pfn - max_pfn;
186 capacity = e_pfn - s_pfn;
190 if (credits_left < capacity)
191 capacity = credits_left;
193 pfns = xen_do_chunk(dest_pfn, dest_pfn + capacity, false);
195 *last_pfn = (dest_pfn + pfns);
198 credits_left -= pfns;
203 static void __init xen_set_identity_and_release_chunk(
204 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
205 unsigned long *released, unsigned long *identity)
210 * If the PFNs are currently mapped, the VA mapping also needs
211 * to be updated to be 1:1.
213 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
214 (void)HYPERVISOR_update_va_mapping(
215 (unsigned long)__va(pfn << PAGE_SHIFT),
216 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
218 if (start_pfn < nr_pages)
219 *released += xen_release_chunk(
220 start_pfn, min(end_pfn, nr_pages));
222 *identity += set_phys_range_identity(start_pfn, end_pfn);
225 static unsigned long __init xen_set_identity_and_release(
226 const struct e820entry *list, size_t map_size, unsigned long nr_pages)
228 phys_addr_t start = 0;
229 unsigned long released = 0;
230 unsigned long identity = 0;
231 const struct e820entry *entry;
235 * Combine non-RAM regions and gaps until a RAM region (or the
236 * end of the map) is reached, then set the 1:1 map and
237 * release the pages (if available) in those non-RAM regions.
239 * The combined non-RAM regions are rounded to a whole number
240 * of pages so any partial pages are accessible via the 1:1
241 * mapping. This is needed for some BIOSes that put (for
242 * example) the DMI tables in a reserved region that begins on
243 * a non-page boundary.
245 for (i = 0, entry = list; i < map_size; i++, entry++) {
246 phys_addr_t end = entry->addr + entry->size;
247 if (entry->type == E820_RAM || i == map_size - 1) {
248 unsigned long start_pfn = PFN_DOWN(start);
249 unsigned long end_pfn = PFN_UP(end);
251 if (entry->type == E820_RAM)
252 end_pfn = PFN_UP(entry->addr);
254 if (start_pfn < end_pfn)
255 xen_set_identity_and_release_chunk(
256 start_pfn, end_pfn, nr_pages,
257 &released, &identity);
264 printk(KERN_INFO "Released %lu pages of unused memory\n", released);
266 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
271 static unsigned long __init xen_get_max_pages(void)
273 unsigned long max_pages = MAX_DOMAIN_PAGES;
274 domid_t domid = DOMID_SELF;
278 * For the initial domain we use the maximum reservation as
281 * For guest domains the current maximum reservation reflects
282 * the current maximum rather than the static maximum. In this
283 * case the e820 map provided to us will cover the static
286 if (xen_initial_domain()) {
287 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
292 return min(max_pages, MAX_DOMAIN_PAGES);
295 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
297 u64 end = start + size;
299 /* Align RAM regions to page boundaries. */
300 if (type == E820_RAM) {
301 start = PAGE_ALIGN(start);
302 end &= ~((u64)PAGE_SIZE - 1);
305 e820_add_region(start, end - start, type);
309 * machine_specific_memory_setup - Hook for machine specific memory setup.
311 char * __init xen_memory_setup(void)
313 static struct e820entry map[E820MAX] __initdata;
315 unsigned long max_pfn = xen_start_info->nr_pages;
316 unsigned long long mem_end;
318 struct xen_memory_map memmap;
319 unsigned long max_pages;
320 unsigned long last_pfn = 0;
321 unsigned long extra_pages = 0;
322 unsigned long populated;
326 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
327 mem_end = PFN_PHYS(max_pfn);
329 memmap.nr_entries = E820MAX;
330 set_xen_guest_handle(memmap.buffer, map);
332 op = xen_initial_domain() ?
333 XENMEM_machine_memory_map :
335 rc = HYPERVISOR_memory_op(op, &memmap);
337 BUG_ON(xen_initial_domain());
338 memmap.nr_entries = 1;
340 map[0].size = mem_end;
341 /* 8MB slack (to balance backend allocations). */
342 map[0].size += 8ULL << 20;
343 map[0].type = E820_RAM;
348 /* Make sure the Xen-supplied memory map is well-ordered. */
349 sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
351 max_pages = xen_get_max_pages();
352 if (max_pages > max_pfn)
353 extra_pages += max_pages - max_pfn;
356 * Set P2M for all non-RAM pages and E820 gaps to be identity
357 * type PFNs. Any RAM pages that would be made inaccesible by
358 * this are first released.
360 xen_released_pages = xen_set_identity_and_release(
361 map, memmap.nr_entries, max_pfn);
364 * Populate back the non-RAM pages and E820 gaps that had been
366 populated = xen_populate_chunk(map, memmap.nr_entries,
367 max_pfn, &last_pfn, xen_released_pages);
369 xen_released_pages -= populated;
370 extra_pages += xen_released_pages;
372 if (last_pfn > max_pfn) {
373 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
374 mem_end = PFN_PHYS(max_pfn);
377 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
378 * factor the base size. On non-highmem systems, the base
379 * size is the full initial memory allocation; on highmem it
380 * is limited to the max size of lowmem, so that it doesn't
381 * get completely filled.
383 * In principle there could be a problem in lowmem systems if
384 * the initial memory is also very large with respect to
385 * lowmem, but we won't try to deal with that here.
387 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
390 while (i < memmap.nr_entries) {
391 u64 addr = map[i].addr;
392 u64 size = map[i].size;
393 u32 type = map[i].type;
395 if (type == E820_RAM) {
396 if (addr < mem_end) {
397 size = min(size, mem_end - addr);
398 } else if (extra_pages) {
399 size = min(size, (u64)extra_pages * PAGE_SIZE);
400 extra_pages -= size / PAGE_SIZE;
401 xen_add_extra_mem(addr, size);
403 type = E820_UNUSABLE;
406 xen_align_and_add_e820_region(addr, size, type);
410 if (map[i].size == 0)
415 * In domU, the ISA region is normal, usable memory, but we
416 * reserve ISA memory anyway because too many things poke
419 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
426 * See comment above "struct start_info" in <xen/interface/xen.h>
428 memblock_reserve(__pa(xen_start_info->mfn_list),
429 xen_start_info->pt_base - xen_start_info->mfn_list);
431 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
437 * Set the bit indicating "nosegneg" library variants should be used.
438 * We only need to bother in pure 32-bit mode; compat 32-bit processes
439 * can have un-truncated segments, so wrapping around is allowed.
441 static void __init fiddle_vdso(void)
445 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
446 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
447 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
448 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
452 static int __cpuinit register_callback(unsigned type, const void *func)
454 struct callback_register callback = {
456 .address = XEN_CALLBACK(__KERNEL_CS, func),
457 .flags = CALLBACKF_mask_events,
460 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
463 void __cpuinit xen_enable_sysenter(void)
466 unsigned sysenter_feature;
469 sysenter_feature = X86_FEATURE_SEP;
471 sysenter_feature = X86_FEATURE_SYSENTER32;
474 if (!boot_cpu_has(sysenter_feature))
477 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
479 setup_clear_cpu_cap(sysenter_feature);
482 void __cpuinit xen_enable_syscall(void)
487 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
489 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
490 /* Pretty fatal; 64-bit userspace has no other
491 mechanism for syscalls. */
494 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
495 ret = register_callback(CALLBACKTYPE_syscall32,
496 xen_syscall32_target);
498 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
500 #endif /* CONFIG_X86_64 */
503 void __init xen_arch_setup(void)
505 xen_panic_handler_init();
507 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
508 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
510 if (!xen_feature(XENFEAT_auto_translated_physmap))
511 HYPERVISOR_vm_assist(VMASST_CMD_enable,
512 VMASST_TYPE_pae_extended_cr3);
514 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
515 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
518 xen_enable_sysenter();
519 xen_enable_syscall();
522 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
523 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
528 memcpy(boot_command_line, xen_start_info->cmd_line,
529 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
530 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
532 /* Set up idle, making sure it calls safe_halt() pvop */
534 boot_cpu_data.hlt_works_ok = 1;
538 WARN_ON(set_pm_idle_to_default());