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
  13 #include <asm/elf.h>
  14 #include <asm/vdso.h>
  15 #include <asm/e820.h>
  16 #include <asm/setup.h>
  17 #include <asm/acpi.h>
  18 #include <asm/xen/hypervisor.h>
  19 #include <asm/xen/hypercall.h>
  20
  21 #include <xen/xen.h>
  22 #include <xen/page.h>
  23 #include <xen/interface/callback.h>
  24 #include <xen/interface/memory.h>
  25 #include <xen/interface/physdev.h>
  26 #include <xen/features.h>
  27
  28 #include "xen-ops.h"
  29 #include "vdso.h"
  30
  31 /* These are code, but not functions.  Defined in entry.S */
  32 extern const char xen_hypervisor_callback[];
  33 extern const char xen_failsafe_callback[];
  34 extern void xen_sysenter_target(void);
  35 extern void xen_syscall_target(void);
  36 extern void xen_syscall32_target(void);
  37
  38 /* Amount of extra memory space we add to the e820 ranges */
  39 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
  40
  41 /*
  42  * The maximum amount of extra memory compared to the base size.  The
  43  * main scaling factor is the size of struct page.  At extreme ratios
  44  * of base:extra, all the base memory can be filled with page
  45  * structures for the extra memory, leaving no space for anything
  46  * else.
  47  *
  48  * 10x seems like a reasonable balance between scaling flexibility and
  49  * leaving a practically usable system.
  50  */
  51 #define EXTRA_MEM_RATIO         (10)
  52
  53 static void __init xen_add_extra_mem(unsigned long pages)
  54 {
  55         unsigned long pfn;
  56
  57         u64 size = (u64)pages * PAGE_SIZE;
  58         u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
  59
  60         if (!pages)
  61                 return;
  62
  63         e820_add_region(extra_start, size, E820_RAM);
  64         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  65
  66         memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
  67
  68         xen_extra_mem_size += size;
  69
  70         xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
  71
  72         for (pfn = PFN_DOWN(extra_start); pfn <= xen_max_p2m_pfn; pfn++)
  73                 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  74 }
  75
  76 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  77                                               phys_addr_t end_addr)
  78 {
  79         struct xen_memory_reservation reservation = {
  80                 .address_bits = 0,
  81                 .extent_order = 0,
  82                 .domid        = DOMID_SELF
  83         };
  84         unsigned long start, end;
  85         unsigned long len = 0;
  86         unsigned long pfn;
  87         int ret;
  88
  89         start = PFN_UP(start_addr);
  90         end = PFN_DOWN(end_addr);
  91
  92         if (end <= start)
  93                 return 0;
  94
  95         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  96                start, end);
  97         for(pfn = start; pfn < end; pfn++) {
  98                 unsigned long mfn = pfn_to_mfn(pfn);
  99
 100                 /* Make sure pfn exists to start with */
 101                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
 102                         continue;
 103
 104                 set_xen_guest_handle(reservation.extent_start, &mfn);
 105                 reservation.nr_extents = 1;
 106
 107                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 108                                            &reservation);
 109                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
 110                      start, end, ret);
 111                 if (ret == 1) {
 112                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 113                         len++;
 114                 }
 115         }
 116         printk(KERN_CONT "%ld pages freed\n", len);
 117
 118         return len;
 119 }
 120
 121 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
 122                                                      const struct e820map *e820)
 123 {
 124         phys_addr_t max_addr = PFN_PHYS(max_pfn);
 125         phys_addr_t last_end = ISA_END_ADDRESS;
 126         unsigned long released = 0;
 127         int i;
 128
 129         /* Free any unused memory above the low 1Mbyte. */
 130         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
 131                 phys_addr_t end = e820->map[i].addr;
 132                 end = min(max_addr, end);
 133
 134                 if (last_end < end)
 135                         released += xen_release_chunk(last_end, end);
 136                 last_end = max(last_end, e820->map[i].addr + e820->map[i].size);
 137         }
 138
 139         if (last_end < max_addr)
 140                 released += xen_release_chunk(last_end, max_addr);
 141
 142         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 143         return released;
 144 }
 145
 146 static unsigned long __init xen_set_identity(const struct e820entry *list,
 147                                              ssize_t map_size)
 148 {
 149         phys_addr_t last = xen_initial_domain() ? 0 : ISA_END_ADDRESS;
 150         phys_addr_t start_pci = last;
 151         const struct e820entry *entry;
 152         unsigned long identity = 0;
 153         int i;
 154
 155         for (i = 0, entry = list; i < map_size; i++, entry++) {
 156                 phys_addr_t start = entry->addr;
 157                 phys_addr_t end = start + entry->size;
 158
 159                 if (start < last)
 160                         start = last;
 161
 162                 if (end <= start)
 163                         continue;
 164
 165                 /* Skip over the 1MB region. */
 166                 if (last > end)
 167                         continue;
 168
 169                 if (entry->type == E820_RAM) {
 170                         if (start > start_pci)
 171                                 identity += set_phys_range_identity(
 172                                                 PFN_UP(start_pci), PFN_DOWN(start));
 173
 174                         /* Without saving 'last' we would gooble RAM too
 175                          * at the end of the loop. */
 176                         last = end;
 177                         start_pci = end;
 178                         continue;
 179                 }
 180                 start_pci = min(start, start_pci);
 181                 last = end;
 182         }
 183         if (last > start_pci)
 184                 identity += set_phys_range_identity(
 185                                         PFN_UP(start_pci), PFN_DOWN(last));
 186         return identity;
 187 }
 188 /**
 189  * machine_specific_memory_setup - Hook for machine specific memory setup.
 190  **/
 191 char * __init xen_memory_setup(void)
 192 {
 193         static struct e820entry map[E820MAX] __initdata;
 194         static struct e820entry map_raw[E820MAX] __initdata;
 195
 196         unsigned long max_pfn = xen_start_info->nr_pages;
 197         unsigned long long mem_end;
 198         int rc;
 199         struct xen_memory_map memmap;
 200         unsigned long extra_pages = 0;
 201         unsigned long extra_limit;
 202         unsigned long identity_pages = 0;
 203         int i;
 204         int op;
 205
 206         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 207         mem_end = PFN_PHYS(max_pfn);
 208
 209         memmap.nr_entries = E820MAX;
 210         set_xen_guest_handle(memmap.buffer, map);
 211
 212         op = xen_initial_domain() ?
 213                 XENMEM_machine_memory_map :
 214                 XENMEM_memory_map;
 215         rc = HYPERVISOR_memory_op(op, &memmap);
 216         if (rc == -ENOSYS) {
 217                 BUG_ON(xen_initial_domain());
 218                 memmap.nr_entries = 1;
 219                 map[0].addr = 0ULL;
 220                 map[0].size = mem_end;
 221                 /* 8MB slack (to balance backend allocations). */
 222                 map[0].size += 8ULL << 20;
 223                 map[0].type = E820_RAM;
 224                 rc = 0;
 225         }
 226         BUG_ON(rc);
 227
 228         memcpy(map_raw, map, sizeof(map));
 229         e820.nr_map = 0;
 230         xen_extra_mem_start = max((1ULL << 32), mem_end);
 231         for (i = 0; i < memmap.nr_entries; i++) {
 232                 unsigned long long end;
 233
 234                 /* Guard against non-page aligned E820 entries. */
 235                 if (map[i].type == E820_RAM)
 236                         map[i].size -= (map[i].size + map[i].addr) % PAGE_SIZE;
 237
 238                 end = map[i].addr + map[i].size;
 239                 if (map[i].type == E820_RAM && end > mem_end) {
 240                         /* RAM off the end - may be partially included */
 241                         u64 delta = min(map[i].size, end - mem_end);
 242
 243                         map[i].size -= delta;
 244                         end -= delta;
 245
 246                         extra_pages += PFN_DOWN(delta);
 247                         /*
 248                          * Set RAM below 4GB that is not for us to be unusable.
 249                          * This prevents "System RAM" address space from being
 250                          * used as potential resource for I/O address (happens
 251                          * when 'allocate_resource' is called).
 252                          */
 253                         if (delta &&
 254                                 (xen_initial_domain() && end < 0x100000000ULL))
 255                                 e820_add_region(end, delta, E820_UNUSABLE);
 256                 }
 257
 258                 if (map[i].size > 0 && end > xen_extra_mem_start)
 259                         xen_extra_mem_start = end;
 260
 261                 /* Add region if any remains */
 262                 if (map[i].size > 0)
 263                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 264         }
 265
 266         /*
 267          * In domU, the ISA region is normal, usable memory, but we
 268          * reserve ISA memory anyway because too many things poke
 269          * about in there.
 270          *
 271          * In Dom0, the host E820 information can leave gaps in the
 272          * ISA range, which would cause us to release those pages.  To
 273          * avoid this, we unconditionally reserve them here.
 274          */
 275         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 276                         E820_RESERVED);
 277
 278         /*
 279          * Reserve Xen bits:
 280          *  - mfn_list
 281          *  - xen_start_info
 282          * See comment above "struct start_info" in <xen/interface/xen.h>
 283          */
 284         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 285                       __pa(xen_start_info->pt_base),
 286                         "XEN START INFO");
 287
 288         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 289
 290         extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 291
 292         /*
 293          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 294          * factor the base size.  On non-highmem systems, the base
 295          * size is the full initial memory allocation; on highmem it
 296          * is limited to the max size of lowmem, so that it doesn't
 297          * get completely filled.
 298          *
 299          * In principle there could be a problem in lowmem systems if
 300          * the initial memory is also very large with respect to
 301          * lowmem, but we won't try to deal with that here.
 302          */
 303         extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 304                           max_pfn + extra_pages);
 305
 306         if (extra_limit >= max_pfn)
 307                 extra_pages = extra_limit - max_pfn;
 308         else
 309                 extra_pages = 0;
 310
 311         xen_add_extra_mem(extra_pages);
 312
 313         /*
 314          * Set P2M for all non-RAM pages and E820 gaps to be identity
 315          * type PFNs. We supply it with the non-sanitized version
 316          * of the E820.
 317          */
 318         identity_pages = xen_set_identity(map_raw, memmap.nr_entries);
 319         printk(KERN_INFO "Set %ld page(s) to 1-1 mapping.\n", identity_pages);
 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         pm_idle = default_idle;
 424         boot_option_idle_override = IDLE_HALT;
 425
 426         fiddle_vdso();
 427 }