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 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
  41
  42 /* Number of pages released from the initial allocation. */
  43 unsigned long xen_released_pages;
  44
  45 /*
  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
  50  * else.
  51  *
  52  * 10x seems like a reasonable balance between scaling flexibility and
  53  * leaving a practically usable system.
  54  */
  55 #define EXTRA_MEM_RATIO         (10)
  56
  57 static void __init xen_add_extra_mem(u64 start, u64 size)
  58 {
  59         unsigned long pfn;
  60         int i;
  61
  62         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  63                 /* Add new region. */
  64                 if (xen_extra_mem[i].size == 0) {
  65                         xen_extra_mem[i].start = start;
  66                         xen_extra_mem[i].size  = size;
  67                         break;
  68                 }
  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;
  72                         break;
  73                 }
  74         }
  75         if (i == XEN_EXTRA_MEM_MAX_REGIONS)
  76                 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
  77
  78         memblock_x86_reserve_range(start, start + size, "XEN EXTRA");
  79
  80         xen_max_p2m_pfn = PFN_DOWN(start + size);
  81
  82         for (pfn = PFN_DOWN(start); pfn <= xen_max_p2m_pfn; pfn++)
  83                 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  84 }
  85
  86 static unsigned long __init xen_release_chunk(unsigned long start,
  87                                               unsigned long end)
  88 {
  89         struct xen_memory_reservation reservation = {
  90                 .address_bits = 0,
  91                 .extent_order = 0,
  92                 .domid        = DOMID_SELF
  93         };
  94         unsigned long len = 0;
  95         unsigned long pfn;
  96         int ret;
  97
  98         for(pfn = start; pfn < end; pfn++) {
  99                 unsigned long mfn = pfn_to_mfn(pfn);
 100
 101                 /* Make sure pfn exists to start with */
 102                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
 103                         continue;
 104
 105                 set_xen_guest_handle(reservation.extent_start, &mfn);
 106                 reservation.nr_extents = 1;
 107
 108                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 109                                            &reservation);
 110                 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
 111                 if (ret == 1) {
 112                         __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 113                         len++;
 114                 }
 115         }
 116         printk(KERN_INFO "Freeing  %lx-%lx pfn range: %lu pages freed\n",
 117                start, end, len);
 118
 119         return len;
 120 }
 121
 122 static unsigned long __init xen_set_identity_and_release(
 123         const struct e820entry *list, size_t map_size, unsigned long nr_pages)
 124 {
 125         phys_addr_t start = 0;
 126         unsigned long released = 0;
 127         unsigned long identity = 0;
 128         const struct e820entry *entry;
 129         int i;
 130
 131         /*
 132          * Combine non-RAM regions and gaps until a RAM region (or the
 133          * end of the map) is reached, then set the 1:1 map and
 134          * release the pages (if available) in those non-RAM regions.
 135          *
 136          * The combined non-RAM regions are rounded to a whole number
 137          * of pages so any partial pages are accessible via the 1:1
 138          * mapping.  This is needed for some BIOSes that put (for
 139          * example) the DMI tables in a reserved region that begins on
 140          * a non-page boundary.
 141          */
 142         for (i = 0, entry = list; i < map_size; i++, entry++) {
 143                 phys_addr_t end = entry->addr + entry->size;
 144
 145                 if (entry->type == E820_RAM || i == map_size - 1) {
 146                         unsigned long start_pfn = PFN_DOWN(start);
 147                         unsigned long end_pfn = PFN_UP(end);
 148
 149                         if (entry->type == E820_RAM)
 150                                 end_pfn = PFN_UP(entry->addr);
 151
 152                         if (start_pfn < end_pfn) {
 153                                 if (start_pfn < nr_pages)
 154                                         released += xen_release_chunk(
 155                                                 start_pfn, min(end_pfn, nr_pages));
 156
 157                                 identity += set_phys_range_identity(
 158                                         start_pfn, end_pfn);
 159                         }
 160                         start = end;
 161                 }
 162         }
 163
 164         printk(KERN_INFO "Released %lu pages of unused memory\n", released);
 165         printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
 166
 167         return released;
 168 }
 169
 170 static unsigned long __init xen_get_max_pages(void)
 171 {
 172         unsigned long max_pages = MAX_DOMAIN_PAGES;
 173         domid_t domid = DOMID_SELF;
 174         int ret;
 175
 176         ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
 177         if (ret > 0)
 178                 max_pages = ret;
 179         return min(max_pages, MAX_DOMAIN_PAGES);
 180 }
 181
 182 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
 183 {
 184         u64 end = start + size;
 185
 186         /* Align RAM regions to page boundaries. */
 187         if (type == E820_RAM) {
 188                 start = PAGE_ALIGN(start);
 189                 end &= ~((u64)PAGE_SIZE - 1);
 190         }
 191
 192         e820_add_region(start, end - start, type);
 193 }
 194
 195 /**
 196  * machine_specific_memory_setup - Hook for machine specific memory setup.
 197  **/
 198 char * __init xen_memory_setup(void)
 199 {
 200         static struct e820entry map[E820MAX] __initdata;
 201
 202         unsigned long max_pfn = xen_start_info->nr_pages;
 203         unsigned long long mem_end;
 204         int rc;
 205         struct xen_memory_map memmap;
 206         unsigned long max_pages;
 207         unsigned long extra_pages = 0;
 208         int i;
 209         int op;
 210
 211         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 212         mem_end = PFN_PHYS(max_pfn);
 213
 214         memmap.nr_entries = E820MAX;
 215         set_xen_guest_handle(memmap.buffer, map);
 216
 217         op = xen_initial_domain() ?
 218                 XENMEM_machine_memory_map :
 219                 XENMEM_memory_map;
 220         rc = HYPERVISOR_memory_op(op, &memmap);
 221         if (rc == -ENOSYS) {
 222                 BUG_ON(xen_initial_domain());
 223                 memmap.nr_entries = 1;
 224                 map[0].addr = 0ULL;
 225                 map[0].size = mem_end;
 226                 /* 8MB slack (to balance backend allocations). */
 227                 map[0].size += 8ULL << 20;
 228                 map[0].type = E820_RAM;
 229                 rc = 0;
 230         }
 231         BUG_ON(rc);
 232
 233         /* Make sure the Xen-supplied memory map is well-ordered. */
 234         sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
 235
 236         max_pages = xen_get_max_pages();
 237         if (max_pages > max_pfn)
 238                 extra_pages += max_pages - max_pfn;
 239
 240         /*
 241          * Set P2M for all non-RAM pages and E820 gaps to be identity
 242          * type PFNs.  Any RAM pages that would be made inaccesible by
 243          * this are first released.
 244          */
 245         xen_released_pages = xen_set_identity_and_release(
 246                 map, memmap.nr_entries, max_pfn);
 247         extra_pages += xen_released_pages;
 248
 249         /*
 250          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 251          * factor the base size.  On non-highmem systems, the base
 252          * size is the full initial memory allocation; on highmem it
 253          * is limited to the max size of lowmem, so that it doesn't
 254          * get completely filled.
 255          *
 256          * In principle there could be a problem in lowmem systems if
 257          * the initial memory is also very large with respect to
 258          * lowmem, but we won't try to deal with that here.
 259          */
 260         extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 261                           extra_pages);
 262
 263         i = 0;
 264         while (i < memmap.nr_entries) {
 265                 u64 addr = map[i].addr;
 266                 u64 size = map[i].size;
 267                 u32 type = map[i].type;
 268
 269                 if (type == E820_RAM) {
 270                         if (addr < mem_end) {
 271                                 size = min(size, mem_end - addr);
 272                         } else if (extra_pages) {
 273                                 size = min(size, (u64)extra_pages * PAGE_SIZE);
 274                                 extra_pages -= size / PAGE_SIZE;
 275                                 xen_add_extra_mem(addr, size);
 276                         } else
 277                                 type = E820_UNUSABLE;
 278                 }
 279
 280                 xen_align_and_add_e820_region(addr, size, type);
 281
 282                 map[i].addr += size;
 283                 map[i].size -= size;
 284                 if (map[i].size == 0)
 285                         i++;
 286         }
 287
 288         /*
 289          * In domU, the ISA region is normal, usable memory, but we
 290          * reserve ISA memory anyway because too many things poke
 291          * about in there.
 292          */
 293         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 294                         E820_RESERVED);
 295
 296         /*
 297          * Reserve Xen bits:
 298          *  - mfn_list
 299          *  - xen_start_info
 300          * See comment above "struct start_info" in <xen/interface/xen.h>
 301          */
 302         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 303                       __pa(xen_start_info->pt_base),
 304                         "XEN START INFO");
 305
 306         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 307
 308         return "Xen";
 309 }
 310
 311 /*
 312  * Set the bit indicating "nosegneg" library variants should be used.
 313  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 314  * can have un-truncated segments, so wrapping around is allowed.
 315  */
 316 static void __init fiddle_vdso(void)
 317 {
 318 #ifdef CONFIG_X86_32
 319         u32 *mask;
 320         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 321         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 322         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 323         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 324 #endif
 325 }
 326
 327 static int __cpuinit register_callback(unsigned type, const void *func)
 328 {
 329         struct callback_register callback = {
 330                 .type = type,
 331                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 332                 .flags = CALLBACKF_mask_events,
 333         };
 334
 335         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 336 }
 337
 338 void __cpuinit xen_enable_sysenter(void)
 339 {
 340         int ret;
 341         unsigned sysenter_feature;
 342
 343 #ifdef CONFIG_X86_32
 344         sysenter_feature = X86_FEATURE_SEP;
 345 #else
 346         sysenter_feature = X86_FEATURE_SYSENTER32;
 347 #endif
 348
 349         if (!boot_cpu_has(sysenter_feature))
 350                 return;
 351
 352         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 353         if(ret != 0)
 354                 setup_clear_cpu_cap(sysenter_feature);
 355 }
 356
 357 void __cpuinit xen_enable_syscall(void)
 358 {
 359 #ifdef CONFIG_X86_64
 360         int ret;
 361
 362         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 363         if (ret != 0) {
 364                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 365                 /* Pretty fatal; 64-bit userspace has no other
 366                    mechanism for syscalls. */
 367         }
 368
 369         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 370                 ret = register_callback(CALLBACKTYPE_syscall32,
 371                                         xen_syscall32_target);
 372                 if (ret != 0)
 373                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 374         }
 375 #endif /* CONFIG_X86_64 */
 376 }
 377
 378 void __init xen_arch_setup(void)
 379 {
 380         xen_panic_handler_init();
 381
 382         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 383         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 384
 385         if (!xen_feature(XENFEAT_auto_translated_physmap))
 386                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 387                                      VMASST_TYPE_pae_extended_cr3);
 388
 389         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 390             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 391                 BUG();
 392
 393         xen_enable_sysenter();
 394         xen_enable_syscall();
 395
 396 #ifdef CONFIG_ACPI
 397         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 398                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 399                 disable_acpi();
 400         }
 401 #endif
 402
 403         memcpy(boot_command_line, xen_start_info->cmd_line,
 404                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 405                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 406
 407         /* Set up idle, making sure it calls safe_halt() pvop */
 408 #ifdef CONFIG_X86_32
 409         boot_cpu_data.hlt_works_ok = 1;
 410 #endif
 411         disable_cpuidle();
 412         boot_option_idle_override = IDLE_HALT;
 413
 414         fiddle_vdso();
 415 }