arch/x86/xen/setup.c

   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
  14 #include <asm/elf.h>
  15 #include <asm/vdso.h>
  16 #include <asm/e820.h>
  17 #include <asm/setup.h>
  18 #include <asm/acpi.h>
  19 #include <asm/xen/hypervisor.h>
  20 #include <asm/xen/hypercall.h>
  21
  22 #include <xen/xen.h>
  23 #include <xen/page.h>
  24 #include <xen/interface/callback.h>
  25 #include <xen/interface/memory.h>
  26 #include <xen/interface/physdev.h>
  27 #include <xen/features.h>
  28
  29 #include "xen-ops.h"
  30 #include "vdso.h"
  31
  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);
  38
  39 /* Amount of extra memory space we add to the e820 ranges */
  40 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
  41
  42 /*
  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
  47  * else.
  48  *
  49  * 10x seems like a reasonable balance between scaling flexibility and
  50  * leaving a practically usable system.
  51  */
  52 #define EXTRA_MEM_RATIO         (10)
  53
  54 static void __init xen_add_extra_mem(unsigned long pages)
  55 {
  56         unsigned long pfn;
  57
  58         u64 size = (u64)pages * PAGE_SIZE;
  59         u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
  60
  61         if (!pages)
  62                 return;
  63
  64         e820_add_region(extra_start, size, E820_RAM);
  65         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  66
  67         memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
  68
  69         xen_extra_mem_size += size;
  70
  71         xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
  72
  73         for (pfn = PFN_DOWN(extra_start); pfn <= xen_max_p2m_pfn; pfn++)
  74                 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  75 }
  76
  77 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  78                                               phys_addr_t end_addr)
  79 {
  80         struct xen_memory_reservation reservation = {
  81                 .address_bits = 0,
  82                 .extent_order = 0,
  83                 .domid        = DOMID_SELF
  84         };
  85         unsigned long start, end;
  86         unsigned long len = 0;
  87         unsigned long pfn;
  88         int ret;
  89
  90         start = PFN_UP(start_addr);
  91         end = PFN_DOWN(end_addr);
  92
  93         if (end <= start)
  94                 return 0;
  95
  96         for(pfn = start; pfn < end; pfn++) {
  97                 unsigned long mfn = pfn_to_mfn(pfn);
  98
  99                 /* Make sure pfn exists to start with */
 100                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
 101                         continue;
 102
 103                 set_xen_guest_handle(reservation.extent_start, &mfn);
 104                 reservation.nr_extents = 1;
 105
 106                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 107                                            &reservation);
 108                 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
 109                 if (ret == 1) {
 110                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 111                         len++;
 112                 }
 113         }
 114         printk(KERN_INFO "Freeing  %lx-%lx pfn range: %lu pages freed\n",
 115                start, end, len);
 116
 117         return len;
 118 }
 119
 120 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
 121                                                      const struct e820map *e820)
 122 {
 123         phys_addr_t max_addr = PFN_PHYS(max_pfn);
 124         phys_addr_t last_end = ISA_END_ADDRESS;
 125         unsigned long released = 0;
 126         int i;
 127
 128         /* Free any unused memory above the low 1Mbyte. */
 129         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
 130                 phys_addr_t end = e820->map[i].addr;
 131                 end = min(max_addr, end);
 132
 133                 if (last_end < end)
 134                         released += xen_release_chunk(last_end, end);
 135                 last_end = max(last_end, e820->map[i].addr + e820->map[i].size);
 136         }
 137
 138         if (last_end < max_addr)
 139                 released += xen_release_chunk(last_end, max_addr);
 140
 141         printk(KERN_INFO "released %lu pages of unused memory\n", released);
 142         return released;
 143 }
 144
 145 static unsigned long __init xen_set_identity(const struct e820entry *list,
 146                                              ssize_t map_size)
 147 {
 148         phys_addr_t last = xen_initial_domain() ? 0 : ISA_END_ADDRESS;
 149         phys_addr_t start_pci = last;
 150         const struct e820entry *entry;
 151         unsigned long identity = 0;
 152         int i;
 153
 154         for (i = 0, entry = list; i < map_size; i++, entry++) {
 155                 phys_addr_t start = entry->addr;
 156                 phys_addr_t end = start + entry->size;
 157
 158                 if (start < last)
 159                         start = last;
 160
 161                 if (end <= start)
 162                         continue;
 163
 164                 /* Skip over the 1MB region. */
 165                 if (last > end)
 166                         continue;
 167
 168                 if ((entry->type == E820_RAM) || (entry->type == E820_UNUSABLE)) {
 169                         if (start > start_pci)
 170                                 identity += set_phys_range_identity(
 171                                                 PFN_UP(start_pci), PFN_DOWN(start));
 172
 173                         /* Without saving 'last' we would gooble RAM too
 174                          * at the end of the loop. */
 175                         last = end;
 176                         start_pci = end;
 177                         continue;
 178                 }
 179                 start_pci = min(start, start_pci);
 180                 last = end;
 181         }
 182         if (last > start_pci)
 183                 identity += set_phys_range_identity(
 184                                         PFN_UP(start_pci), PFN_DOWN(last));
 185         return identity;
 186 }
 187 /**
 188  * machine_specific_memory_setup - Hook for machine specific memory setup.
 189  **/
 190 char * __init xen_memory_setup(void)
 191 {
 192         static struct e820entry map[E820MAX] __initdata;
 193         static struct e820entry map_raw[E820MAX] __initdata;
 194
 195         unsigned long max_pfn = xen_start_info->nr_pages;
 196         unsigned long long mem_end;
 197         int rc;
 198         struct xen_memory_map memmap;
 199         unsigned long extra_pages = 0;
 200         unsigned long extra_limit;
 201         unsigned long identity_pages = 0;
 202         int i;
 203         int op;
 204
 205         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 206         mem_end = PFN_PHYS(max_pfn);
 207
 208         memmap.nr_entries = E820MAX;
 209         set_xen_guest_handle(memmap.buffer, map);
 210
 211         op = xen_initial_domain() ?
 212                 XENMEM_machine_memory_map :
 213                 XENMEM_memory_map;
 214         rc = HYPERVISOR_memory_op(op, &memmap);
 215         if (rc == -ENOSYS) {
 216                 BUG_ON(xen_initial_domain());
 217                 memmap.nr_entries = 1;
 218                 map[0].addr = 0ULL;
 219                 map[0].size = mem_end;
 220                 /* 8MB slack (to balance backend allocations). */
 221                 map[0].size += 8ULL << 20;
 222                 map[0].type = E820_RAM;
 223                 rc = 0;
 224         }
 225         BUG_ON(rc);
 226
 227         memcpy(map_raw, map, sizeof(map));
 228         e820.nr_map = 0;
 229         xen_extra_mem_start = mem_end;
 230         for (i = 0; i < memmap.nr_entries; i++) {
 231                 unsigned long long end;
 232
 233                 /* Guard against non-page aligned E820 entries. */
 234                 if (map[i].type == E820_RAM)
 235                         map[i].size -= (map[i].size + map[i].addr) % PAGE_SIZE;
 236
 237                 end = map[i].addr + map[i].size;
 238                 if (map[i].type == E820_RAM && end > mem_end) {
 239                         /* RAM off the end - may be partially included */
 240                         u64 delta = min(map[i].size, end - mem_end);
 241
 242                         map[i].size -= delta;
 243                         end -= delta;
 244
 245                         extra_pages += PFN_DOWN(delta);
 246                         /*
 247                          * Set RAM below 4GB that is not for us to be unusable.
 248                          * This prevents "System RAM" address space from being
 249                          * used as potential resource for I/O address (happens
 250                          * when 'allocate_resource' is called).
 251                          */
 252                         if (delta &&
 253                                 (xen_initial_domain() && end < 0x100000000ULL))
 254                                 e820_add_region(end, delta, E820_UNUSABLE);
 255                 }
 256
 257                 if (map[i].size > 0 && end > xen_extra_mem_start)
 258                         xen_extra_mem_start = end;
 259
 260                 /* Add region if any remains */
 261                 if (map[i].size > 0)
 262                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 263         }
 264         /* Align the balloon area so that max_low_pfn does not get set
 265          * to be at the _end_ of the PCI gap at the far end (fee01000).
 266          * Note that xen_extra_mem_start gets set in the loop above to be
 267          * past the last E820 region. */
 268         if (xen_initial_domain() && (xen_extra_mem_start < (1ULL<<32)))
 269                 xen_extra_mem_start = (1ULL<<32);
 270
 271         /*
 272          * In domU, the ISA region is normal, usable memory, but we
 273          * reserve ISA memory anyway because too many things poke
 274          * about in there.
 275          *
 276          * In Dom0, the host E820 information can leave gaps in the
 277          * ISA range, which would cause us to release those pages.  To
 278          * avoid this, we unconditionally reserve them here.
 279          */
 280         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 281                         E820_RESERVED);
 282
 283         /*
 284          * Reserve Xen bits:
 285          *  - mfn_list
 286          *  - xen_start_info
 287          * See comment above "struct start_info" in <xen/interface/xen.h>
 288          */
 289         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 290                       __pa(xen_start_info->pt_base),
 291                         "XEN START INFO");
 292
 293         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 294
 295         extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 296
 297         /*
 298          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 299          * factor the base size.  On non-highmem systems, the base
 300          * size is the full initial memory allocation; on highmem it
 301          * is limited to the max size of lowmem, so that it doesn't
 302          * get completely filled.
 303          *
 304          * In principle there could be a problem in lowmem systems if
 305          * the initial memory is also very large with respect to
 306          * lowmem, but we won't try to deal with that here.
 307          */
 308         extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 309                           max_pfn + extra_pages);
 310
 311         if (extra_limit >= max_pfn)
 312                 extra_pages = extra_limit - max_pfn;
 313         else
 314                 extra_pages = 0;
 315
 316         xen_add_extra_mem(extra_pages);
 317
 318         /*
 319          * Set P2M for all non-RAM pages and E820 gaps to be identity
 320          * type PFNs. We supply it with the non-sanitized version
 321          * of the E820.
 322          */
 323         identity_pages = xen_set_identity(map_raw, memmap.nr_entries);
 324         printk(KERN_INFO "Set %ld page(s) to 1-1 mapping.\n", identity_pages);
 325         return "Xen";
 326 }
 327
 328 /*
 329  * Set the bit indicating "nosegneg" library variants should be used.
 330  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 331  * can have un-truncated segments, so wrapping around is allowed.
 332  */
 333 static void __init fiddle_vdso(void)
 334 {
 335 #ifdef CONFIG_X86_32
 336         u32 *mask;
 337         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 338         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 339         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 340         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 341 #endif
 342 }
 343
 344 static int __cpuinit register_callback(unsigned type, const void *func)
 345 {
 346         struct callback_register callback = {
 347                 .type = type,
 348                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 349                 .flags = CALLBACKF_mask_events,
 350         };
 351
 352         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 353 }
 354
 355 void __cpuinit xen_enable_sysenter(void)
 356 {
 357         int ret;
 358         unsigned sysenter_feature;
 359
 360 #ifdef CONFIG_X86_32
 361         sysenter_feature = X86_FEATURE_SEP;
 362 #else
 363         sysenter_feature = X86_FEATURE_SYSENTER32;
 364 #endif
 365
 366         if (!boot_cpu_has(sysenter_feature))
 367                 return;
 368
 369         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 370         if(ret != 0)
 371                 setup_clear_cpu_cap(sysenter_feature);
 372 }
 373
 374 void __cpuinit xen_enable_syscall(void)
 375 {
 376 #ifdef CONFIG_X86_64
 377         int ret;
 378
 379         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 380         if (ret != 0) {
 381                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 382                 /* Pretty fatal; 64-bit userspace has no other
 383                    mechanism for syscalls. */
 384         }
 385
 386         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 387                 ret = register_callback(CALLBACKTYPE_syscall32,
 388                                         xen_syscall32_target);
 389                 if (ret != 0)
 390                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 391         }
 392 #endif /* CONFIG_X86_64 */
 393 }
 394
 395 void __init xen_arch_setup(void)
 396 {
 397         xen_panic_handler_init();
 398
 399         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 400         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 401
 402         if (!xen_feature(XENFEAT_auto_translated_physmap))
 403                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 404                                      VMASST_TYPE_pae_extended_cr3);
 405
 406         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 407             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 408                 BUG();
 409
 410         xen_enable_sysenter();
 411         xen_enable_syscall();
 412
 413 #ifdef CONFIG_ACPI
 414         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 415                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 416                 disable_acpi();
 417         }
 418 #endif
 419
 420         memcpy(boot_command_line, xen_start_info->cmd_line,
 421                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 422                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 423
 424         /* Set up idle, making sure it calls safe_halt() pvop */
 425 #ifdef CONFIG_X86_32
 426         boot_cpu_data.hlt_works_ok = 1;
 427 #endif
 428         disable_cpuidle();
 429         boot_option_idle_override = IDLE_HALT;
 430
 431         fiddle_vdso();
 432 }