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>
17 #include <asm/setup.h>
19 #include <asm/xen/hypervisor.h>
20 #include <asm/xen/hypercall.h>
24 #include <xen/interface/callback.h>
25 #include <xen/interface/memory.h>
26 #include <xen/interface/physdev.h>
27 #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 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
43 * The maximum amount of extra memory compared to the base size. The
44 * main scaling factor is the size of struct page. At extreme ratios
45 * of base:extra, all the base memory can be filled with page
46 * structures for the extra memory, leaving no space for anything
49 * 10x seems like a reasonable balance between scaling flexibility and
50 * leaving a practically usable system.
52 #define EXTRA_MEM_RATIO (10)
54 static void __init xen_add_extra_mem(unsigned long pages)
58 u64 size = (u64)pages * PAGE_SIZE;
59 u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
64 e820_add_region(extra_start, size, E820_RAM);
65 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
67 memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
69 xen_extra_mem_size += size;
71 xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
73 for (pfn = PFN_DOWN(extra_start); pfn <= xen_max_p2m_pfn; pfn++)
74 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
77 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
80 struct xen_memory_reservation reservation = {
85 unsigned long start, end;
86 unsigned long len = 0;
90 start = PFN_UP(start_addr);
91 end = PFN_DOWN(end_addr);
96 printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
98 for(pfn = start; pfn < end; pfn++) {
99 unsigned long mfn = pfn_to_mfn(pfn);
101 /* Make sure pfn exists to start with */
102 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
105 set_xen_guest_handle(reservation.extent_start, &mfn);
106 reservation.nr_extents = 1;
108 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
110 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
113 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
117 printk(KERN_CONT "%ld pages freed\n", len);
122 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
123 const struct e820map *e820)
125 phys_addr_t max_addr = PFN_PHYS(max_pfn);
126 phys_addr_t last_end = ISA_END_ADDRESS;
127 unsigned long released = 0;
130 /* Free any unused memory above the low 1Mbyte. */
131 for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
132 phys_addr_t end = e820->map[i].addr;
133 end = min(max_addr, end);
136 released += xen_release_chunk(last_end, end);
137 last_end = max(last_end, e820->map[i].addr + e820->map[i].size);
140 if (last_end < max_addr)
141 released += xen_release_chunk(last_end, max_addr);
143 printk(KERN_INFO "released %ld pages of unused memory\n", released);
147 static unsigned long __init xen_set_identity(const struct e820entry *list,
150 phys_addr_t last = xen_initial_domain() ? 0 : ISA_END_ADDRESS;
151 phys_addr_t start_pci = last;
152 const struct e820entry *entry;
153 unsigned long identity = 0;
156 for (i = 0, entry = list; i < map_size; i++, entry++) {
157 phys_addr_t start = entry->addr;
158 phys_addr_t end = start + entry->size;
166 /* Skip over the 1MB region. */
170 if ((entry->type == E820_RAM) || (entry->type == E820_UNUSABLE)) {
171 if (start > start_pci)
172 identity += set_phys_range_identity(
173 PFN_UP(start_pci), PFN_DOWN(start));
175 /* Without saving 'last' we would gooble RAM too
176 * at the end of the loop. */
181 start_pci = min(start, start_pci);
184 if (last > start_pci)
185 identity += set_phys_range_identity(
186 PFN_UP(start_pci), PFN_DOWN(last));
190 * machine_specific_memory_setup - Hook for machine specific memory setup.
192 char * __init xen_memory_setup(void)
194 static struct e820entry map[E820MAX] __initdata;
195 static struct e820entry map_raw[E820MAX] __initdata;
197 unsigned long max_pfn = xen_start_info->nr_pages;
198 unsigned long long mem_end;
200 struct xen_memory_map memmap;
201 unsigned long extra_pages = 0;
202 unsigned long extra_limit;
203 unsigned long identity_pages = 0;
207 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
208 mem_end = PFN_PHYS(max_pfn);
210 memmap.nr_entries = E820MAX;
211 set_xen_guest_handle(memmap.buffer, map);
213 op = xen_initial_domain() ?
214 XENMEM_machine_memory_map :
216 rc = HYPERVISOR_memory_op(op, &memmap);
218 BUG_ON(xen_initial_domain());
219 memmap.nr_entries = 1;
221 map[0].size = mem_end;
222 /* 8MB slack (to balance backend allocations). */
223 map[0].size += 8ULL << 20;
224 map[0].type = E820_RAM;
229 memcpy(map_raw, map, sizeof(map));
231 xen_extra_mem_start = mem_end;
232 for (i = 0; i < memmap.nr_entries; i++) {
233 unsigned long long end;
235 /* Guard against non-page aligned E820 entries. */
236 if (map[i].type == E820_RAM)
237 map[i].size -= (map[i].size + map[i].addr) % PAGE_SIZE;
239 end = map[i].addr + map[i].size;
240 if (map[i].type == E820_RAM && end > mem_end) {
241 /* RAM off the end - may be partially included */
242 u64 delta = min(map[i].size, end - mem_end);
244 map[i].size -= delta;
247 extra_pages += PFN_DOWN(delta);
249 * Set RAM below 4GB that is not for us to be unusable.
250 * This prevents "System RAM" address space from being
251 * used as potential resource for I/O address (happens
252 * when 'allocate_resource' is called).
255 (xen_initial_domain() && end < 0x100000000ULL))
256 e820_add_region(end, delta, E820_UNUSABLE);
259 if (map[i].size > 0 && end > xen_extra_mem_start)
260 xen_extra_mem_start = end;
262 /* Add region if any remains */
264 e820_add_region(map[i].addr, map[i].size, map[i].type);
266 /* Align the balloon area so that max_low_pfn does not get set
267 * to be at the _end_ of the PCI gap at the far end (fee01000).
268 * Note that xen_extra_mem_start gets set in the loop above to be
269 * past the last E820 region. */
270 if (xen_initial_domain() && (xen_extra_mem_start < (1ULL<<32)))
271 xen_extra_mem_start = (1ULL<<32);
274 * In domU, the ISA region is normal, usable memory, but we
275 * reserve ISA memory anyway because too many things poke
278 * In Dom0, the host E820 information can leave gaps in the
279 * ISA range, which would cause us to release those pages. To
280 * avoid this, we unconditionally reserve them here.
282 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
289 * See comment above "struct start_info" in <xen/interface/xen.h>
291 memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
292 __pa(xen_start_info->pt_base),
295 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
297 extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
300 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
301 * factor the base size. On non-highmem systems, the base
302 * size is the full initial memory allocation; on highmem it
303 * is limited to the max size of lowmem, so that it doesn't
304 * get completely filled.
306 * In principle there could be a problem in lowmem systems if
307 * the initial memory is also very large with respect to
308 * lowmem, but we won't try to deal with that here.
310 extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
311 max_pfn + extra_pages);
313 if (extra_limit >= max_pfn)
314 extra_pages = extra_limit - max_pfn;
318 xen_add_extra_mem(extra_pages);
321 * Set P2M for all non-RAM pages and E820 gaps to be identity
322 * type PFNs. We supply it with the non-sanitized version
325 identity_pages = xen_set_identity(map_raw, memmap.nr_entries);
326 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping.\n", identity_pages);
331 * Set the bit indicating "nosegneg" library variants should be used.
332 * We only need to bother in pure 32-bit mode; compat 32-bit processes
333 * can have un-truncated segments, so wrapping around is allowed.
335 static void __init fiddle_vdso(void)
339 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
340 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
341 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
342 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
346 static int __cpuinit register_callback(unsigned type, const void *func)
348 struct callback_register callback = {
350 .address = XEN_CALLBACK(__KERNEL_CS, func),
351 .flags = CALLBACKF_mask_events,
354 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
357 void __cpuinit xen_enable_sysenter(void)
360 unsigned sysenter_feature;
363 sysenter_feature = X86_FEATURE_SEP;
365 sysenter_feature = X86_FEATURE_SYSENTER32;
368 if (!boot_cpu_has(sysenter_feature))
371 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
373 setup_clear_cpu_cap(sysenter_feature);
376 void __cpuinit xen_enable_syscall(void)
381 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
383 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
384 /* Pretty fatal; 64-bit userspace has no other
385 mechanism for syscalls. */
388 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
389 ret = register_callback(CALLBACKTYPE_syscall32,
390 xen_syscall32_target);
392 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
394 #endif /* CONFIG_X86_64 */
397 void __init xen_arch_setup(void)
399 xen_panic_handler_init();
401 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
402 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
404 if (!xen_feature(XENFEAT_auto_translated_physmap))
405 HYPERVISOR_vm_assist(VMASST_CMD_enable,
406 VMASST_TYPE_pae_extended_cr3);
408 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
409 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
412 xen_enable_sysenter();
413 xen_enable_syscall();
416 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
417 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
422 memcpy(boot_command_line, xen_start_info->cmd_line,
423 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
424 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
426 /* Set up idle, making sure it calls safe_halt() pvop */
428 boot_cpu_data.hlt_works_ok = 1;
431 boot_option_idle_override = IDLE_HALT;