1 /* 2 * Machine specific setup for xen 3 * 4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 5 */ 6 7 #include <linux/module.h> 8 #include <linux/sched.h> 9 #include <linux/mm.h> 10 #include <linux/pm.h> 11 #include <linux/memblock.h> 12 #include <linux/cpuidle.h> 13 #include <linux/cpufreq.h> 14 15 #include <asm/elf.h> 16 #include <asm/vdso.h> 17 #include <asm/e820.h> 18 #include <asm/setup.h> 19 #include <asm/acpi.h> 20 #include <asm/xen/hypervisor.h> 21 #include <asm/xen/hypercall.h> 22 23 #include <xen/xen.h> 24 #include <xen/page.h> 25 #include <xen/interface/callback.h> 26 #include <xen/interface/memory.h> 27 #include <xen/interface/physdev.h> 28 #include <xen/features.h> 29 30 #include "xen-ops.h" 31 #include "vdso.h" 32 33 /* These are code, but not functions. Defined in entry.S */ 34 extern const char xen_hypervisor_callback[]; 35 extern const char xen_failsafe_callback[]; 36 extern void xen_sysenter_target(void); 37 extern void xen_syscall_target(void); 38 extern void xen_syscall32_target(void); 39 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; 42 43 /* Number of pages released from the initial allocation. */ 44 unsigned long xen_released_pages; 45 46 /* 47 * The maximum amount of extra memory compared to the base size. The 48 * main scaling factor is the size of struct page. At extreme ratios 49 * of base:extra, all the base memory can be filled with page 50 * structures for the extra memory, leaving no space for anything 51 * else. 52 * 53 * 10x seems like a reasonable balance between scaling flexibility and 54 * leaving a practically usable system. 55 */ 56 #define EXTRA_MEM_RATIO (10) 57 58 static void __init xen_add_extra_mem(u64 start, u64 size) 59 { 60 unsigned long pfn; 61 int i; 62 63 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { 64 /* Add new region. */ 65 if (xen_extra_mem[i].size == 0) { 66 xen_extra_mem[i].start = start; 67 xen_extra_mem[i].size = size; 68 break; 69 } 70 /* Append to existing region. */ 71 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) { 72 xen_extra_mem[i].size += size; 73 break; 74 } 75 } 76 if (i == XEN_EXTRA_MEM_MAX_REGIONS) 77 printk(KERN_WARNING "Warning: not enough extra memory regions\n"); 78 79 memblock_reserve(start, size); 80 81 xen_max_p2m_pfn = PFN_DOWN(start + size); 82 83 for (pfn = PFN_DOWN(start); pfn <= xen_max_p2m_pfn; pfn++) 84 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 85 } 86 87 static unsigned long __init xen_release_chunk(unsigned long start, 88 unsigned long end) 89 { 90 struct xen_memory_reservation reservation = { 91 .address_bits = 0, 92 .extent_order = 0, 93 .domid = DOMID_SELF 94 }; 95 unsigned long len = 0; 96 unsigned long pfn; 97 int ret; 98 99 for(pfn = start; pfn < end; pfn++) { 100 unsigned long mfn = pfn_to_mfn(pfn); 101 102 /* Make sure pfn exists to start with */ 103 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn) 104 continue; 105 106 set_xen_guest_handle(reservation.extent_start, &mfn); 107 reservation.nr_extents = 1; 108 109 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, 110 &reservation); 111 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret); 112 if (ret == 1) { 113 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 114 len++; 115 } 116 } 117 printk(KERN_INFO "Freeing %lx-%lx pfn range: %lu pages freed\n", 118 start, end, len); 119 120 return len; 121 } 122 123 static unsigned long __init xen_set_identity_and_release( 124 const struct e820entry *list, size_t map_size, unsigned long nr_pages) 125 { 126 phys_addr_t start = 0; 127 unsigned long released = 0; 128 unsigned long identity = 0; 129 const struct e820entry *entry; 130 int i; 131 132 /* 133 * Combine non-RAM regions and gaps until a RAM region (or the 134 * end of the map) is reached, then set the 1:1 map and 135 * release the pages (if available) in those non-RAM regions. 136 * 137 * The combined non-RAM regions are rounded to a whole number 138 * of pages so any partial pages are accessible via the 1:1 139 * mapping. This is needed for some BIOSes that put (for 140 * example) the DMI tables in a reserved region that begins on 141 * a non-page boundary. 142 */ 143 for (i = 0, entry = list; i < map_size; i++, entry++) { 144 phys_addr_t end = entry->addr + entry->size; 145 146 if (entry->type == E820_RAM || i == map_size - 1) { 147 unsigned long start_pfn = PFN_DOWN(start); 148 unsigned long end_pfn = PFN_UP(end); 149 150 if (entry->type == E820_RAM) 151 end_pfn = PFN_UP(entry->addr); 152 153 if (start_pfn < end_pfn) { 154 if (start_pfn < nr_pages) 155 released += xen_release_chunk( 156 start_pfn, min(end_pfn, nr_pages)); 157 158 identity += set_phys_range_identity( 159 start_pfn, end_pfn); 160 } 161 start = end; 162 } 163 } 164 165 printk(KERN_INFO "Released %lu pages of unused memory\n", released); 166 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity); 167 168 return released; 169 } 170 171 static unsigned long __init xen_get_max_pages(void) 172 { 173 unsigned long max_pages = MAX_DOMAIN_PAGES; 174 domid_t domid = DOMID_SELF; 175 int ret; 176 177 /* 178 * For the initial domain we use the maximum reservation as 179 * the maximum page. 180 * 181 * For guest domains the current maximum reservation reflects 182 * the current maximum rather than the static maximum. In this 183 * case the e820 map provided to us will cover the static 184 * maximum region. 185 */ 186 if (xen_initial_domain()) { 187 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); 188 if (ret > 0) 189 max_pages = ret; 190 } 191 192 return min(max_pages, MAX_DOMAIN_PAGES); 193 } 194 195 static void xen_align_and_add_e820_region(u64 start, u64 size, int type) 196 { 197 u64 end = start + size; 198 199 /* Align RAM regions to page boundaries. */ 200 if (type == E820_RAM) { 201 start = PAGE_ALIGN(start); 202 end &= ~((u64)PAGE_SIZE - 1); 203 } 204 205 e820_add_region(start, end - start, type); 206 } 207 208 /** 209 * machine_specific_memory_setup - Hook for machine specific memory setup. 210 **/ 211 char * __init xen_memory_setup(void) 212 { 213 static struct e820entry map[E820MAX] __initdata; 214 215 unsigned long max_pfn = xen_start_info->nr_pages; 216 unsigned long long mem_end; 217 int rc; 218 struct xen_memory_map memmap; 219 unsigned long max_pages; 220 unsigned long extra_pages = 0; 221 int i; 222 int op; 223 224 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn); 225 mem_end = PFN_PHYS(max_pfn); 226 227 memmap.nr_entries = E820MAX; 228 set_xen_guest_handle(memmap.buffer, map); 229 230 op = xen_initial_domain() ? 231 XENMEM_machine_memory_map : 232 XENMEM_memory_map; 233 rc = HYPERVISOR_memory_op(op, &memmap); 234 if (rc == -ENOSYS) { 235 BUG_ON(xen_initial_domain()); 236 memmap.nr_entries = 1; 237 map[0].addr = 0ULL; 238 map[0].size = mem_end; 239 /* 8MB slack (to balance backend allocations). */ 240 map[0].size += 8ULL << 20; 241 map[0].type = E820_RAM; 242 rc = 0; 243 } 244 BUG_ON(rc); 245 246 /* Make sure the Xen-supplied memory map is well-ordered. */ 247 sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries); 248 249 max_pages = xen_get_max_pages(); 250 if (max_pages > max_pfn) 251 extra_pages += max_pages - max_pfn; 252 253 /* 254 * Set P2M for all non-RAM pages and E820 gaps to be identity 255 * type PFNs. Any RAM pages that would be made inaccesible by 256 * this are first released. 257 */ 258 xen_released_pages = xen_set_identity_and_release( 259 map, memmap.nr_entries, max_pfn); 260 extra_pages += xen_released_pages; 261 262 /* 263 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO 264 * factor the base size. On non-highmem systems, the base 265 * size is the full initial memory allocation; on highmem it 266 * is limited to the max size of lowmem, so that it doesn't 267 * get completely filled. 268 * 269 * In principle there could be a problem in lowmem systems if 270 * the initial memory is also very large with respect to 271 * lowmem, but we won't try to deal with that here. 272 */ 273 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), 274 extra_pages); 275 276 i = 0; 277 while (i < memmap.nr_entries) { 278 u64 addr = map[i].addr; 279 u64 size = map[i].size; 280 u32 type = map[i].type; 281 282 if (type == E820_RAM) { 283 if (addr < mem_end) { 284 size = min(size, mem_end - addr); 285 } else if (extra_pages) { 286 size = min(size, (u64)extra_pages * PAGE_SIZE); 287 extra_pages -= size / PAGE_SIZE; 288 xen_add_extra_mem(addr, size); 289 } else 290 type = E820_UNUSABLE; 291 } 292 293 xen_align_and_add_e820_region(addr, size, type); 294 295 map[i].addr += size; 296 map[i].size -= size; 297 if (map[i].size == 0) 298 i++; 299 } 300 301 /* 302 * In domU, the ISA region is normal, usable memory, but we 303 * reserve ISA memory anyway because too many things poke 304 * about in there. 305 */ 306 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, 307 E820_RESERVED); 308 309 /* 310 * Reserve Xen bits: 311 * - mfn_list 312 * - xen_start_info 313 * See comment above "struct start_info" in <xen/interface/xen.h> 314 */ 315 memblock_reserve(__pa(xen_start_info->mfn_list), 316 xen_start_info->pt_base - xen_start_info->mfn_list); 317 318 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 319 320 return "Xen"; 321 } 322 323 /* 324 * Set the bit indicating "nosegneg" library variants should be used. 325 * We only need to bother in pure 32-bit mode; compat 32-bit processes 326 * can have un-truncated segments, so wrapping around is allowed. 327 */ 328 static void __init fiddle_vdso(void) 329 { 330 #ifdef CONFIG_X86_32 331 u32 *mask; 332 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK); 333 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; 334 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK); 335 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; 336 #endif 337 } 338 339 static int __cpuinit register_callback(unsigned type, const void *func) 340 { 341 struct callback_register callback = { 342 .type = type, 343 .address = XEN_CALLBACK(__KERNEL_CS, func), 344 .flags = CALLBACKF_mask_events, 345 }; 346 347 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); 348 } 349 350 void __cpuinit xen_enable_sysenter(void) 351 { 352 int ret; 353 unsigned sysenter_feature; 354 355 #ifdef CONFIG_X86_32 356 sysenter_feature = X86_FEATURE_SEP; 357 #else 358 sysenter_feature = X86_FEATURE_SYSENTER32; 359 #endif 360 361 if (!boot_cpu_has(sysenter_feature)) 362 return; 363 364 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); 365 if(ret != 0) 366 setup_clear_cpu_cap(sysenter_feature); 367 } 368 369 void __cpuinit xen_enable_syscall(void) 370 { 371 #ifdef CONFIG_X86_64 372 int ret; 373 374 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); 375 if (ret != 0) { 376 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); 377 /* Pretty fatal; 64-bit userspace has no other 378 mechanism for syscalls. */ 379 } 380 381 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { 382 ret = register_callback(CALLBACKTYPE_syscall32, 383 xen_syscall32_target); 384 if (ret != 0) 385 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); 386 } 387 #endif /* CONFIG_X86_64 */ 388 } 389 390 void __init xen_arch_setup(void) 391 { 392 xen_panic_handler_init(); 393 394 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); 395 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); 396 397 if (!xen_feature(XENFEAT_auto_translated_physmap)) 398 HYPERVISOR_vm_assist(VMASST_CMD_enable, 399 VMASST_TYPE_pae_extended_cr3); 400 401 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || 402 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) 403 BUG(); 404 405 xen_enable_sysenter(); 406 xen_enable_syscall(); 407 408 #ifdef CONFIG_ACPI 409 if (!(xen_start_info->flags & SIF_INITDOMAIN)) { 410 printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); 411 disable_acpi(); 412 } 413 #endif 414 415 memcpy(boot_command_line, xen_start_info->cmd_line, 416 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? 417 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); 418 419 /* Set up idle, making sure it calls safe_halt() pvop */ 420 #ifdef CONFIG_X86_32 421 boot_cpu_data.hlt_works_ok = 1; 422 #endif 423 disable_cpuidle(); 424 disable_cpufreq(); 425 WARN_ON(set_pm_idle_to_default()); 426 fiddle_vdso(); 427 } 428