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
  12 #include <asm/elf.h>
  13 #include <asm/vdso.h>
  14 #include <asm/e820.h>
  15 #include <asm/setup.h>
  16 #include <asm/acpi.h>
  17 #include <asm/xen/hypervisor.h>
  18 #include <asm/xen/hypercall.h>
  19
  20 #include <xen/page.h>
  21 #include <xen/interface/callback.h>
  22 #include <xen/interface/memory.h>
  23 #include <xen/interface/physdev.h>
  24 #include <xen/interface/memory.h>
  25 #include <xen/features.h>
  26
  27 #include "xen-ops.h"
  28 #include "vdso.h"
  29
  30 /* These are code, but not functions.  Defined in entry.S */
  31 extern const char xen_hypervisor_callback[];
  32 extern const char xen_failsafe_callback[];
  33 extern void xen_sysenter_target(void);
  34 extern void xen_syscall_target(void);
  35 extern void xen_syscall32_target(void);
  36
  37 /* Amount of extra memory space we add to the e820 ranges */
  38 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
  39
  40 static __init void xen_add_extra_mem(unsigned long pages)
  41 {
  42         u64 size = (u64)pages * PAGE_SIZE;
  43
  44         if (!pages)
  45                 return;
  46
  47         e820_add_region(xen_extra_mem_start + xen_extra_mem_size, size, E820_RAM);
  48         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  49
  50         reserve_early(xen_extra_mem_start + xen_extra_mem_size,
  51                       xen_extra_mem_start + xen_extra_mem_size + size,
  52                       "XEN EXTRA");
  53
  54         xen_extra_mem_size += size;
  55 }
  56
  57 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  58                                               phys_addr_t end_addr)
  59 {
  60         struct xen_memory_reservation reservation = {
  61                 .address_bits = 0,
  62                 .extent_order = 0,
  63                 .domid        = DOMID_SELF
  64         };
  65         unsigned long start, end;
  66         unsigned long len = 0;
  67         unsigned long pfn;
  68         int ret;
  69
  70         start = PFN_UP(start_addr);
  71         end = PFN_DOWN(end_addr);
  72
  73         if (end <= start)
  74                 return 0;
  75
  76         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  77                start, end);
  78         for(pfn = start; pfn < end; pfn++) {
  79                 unsigned long mfn = pfn_to_mfn(pfn);
  80
  81                 /* Make sure pfn exists to start with */
  82                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
  83                         continue;
  84
  85                 set_xen_guest_handle(reservation.extent_start, &mfn);
  86                 reservation.nr_extents = 1;
  87
  88                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
  89                                            &reservation);
  90                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
  91                      start, end, ret);
  92                 if (ret == 1) {
  93                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  94                         len++;
  95                 }
  96         }
  97         printk(KERN_CONT "%ld pages freed\n", len);
  98
  99         return len;
 100 }
 101
 102 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
 103                                                      const struct e820map *e820)
 104 {
 105         phys_addr_t max_addr = PFN_PHYS(max_pfn);
 106         phys_addr_t last_end = 0;
 107         unsigned long released = 0;
 108         int i;
 109
 110         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
 111                 phys_addr_t end = e820->map[i].addr;
 112                 end = min(max_addr, end);
 113
 114                 released += xen_release_chunk(last_end, end);
 115                 last_end = e820->map[i].addr + e820->map[i].size;
 116         }
 117
 118         if (last_end < max_addr)
 119                 released += xen_release_chunk(last_end, max_addr);
 120
 121         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 122         return released;
 123 }
 124
 125 /**
 126  * machine_specific_memory_setup - Hook for machine specific memory setup.
 127  **/
 128 char * __init xen_memory_setup(void)
 129 {
 130         static struct e820entry map[E820MAX] __initdata;
 131
 132         unsigned long max_pfn = xen_start_info->nr_pages;
 133         unsigned long long mem_end;
 134         int rc;
 135         struct xen_memory_map memmap;
 136         unsigned long extra_pages = 0;
 137         int i;
 138
 139         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 140         mem_end = PFN_PHYS(max_pfn);
 141
 142         memmap.nr_entries = E820MAX;
 143         set_xen_guest_handle(memmap.buffer, map);
 144
 145         rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
 146         if (rc == -ENOSYS) {
 147                 memmap.nr_entries = 1;
 148                 map[0].addr = 0ULL;
 149                 map[0].size = mem_end;
 150                 /* 8MB slack (to balance backend allocations). */
 151                 map[0].size += 8ULL << 20;
 152                 map[0].type = E820_RAM;
 153                 rc = 0;
 154         }
 155         BUG_ON(rc);
 156
 157         e820.nr_map = 0;
 158         xen_extra_mem_start = mem_end;
 159         for (i = 0; i < memmap.nr_entries; i++) {
 160                 unsigned long long end = map[i].addr + map[i].size;
 161                 if (map[i].type == E820_RAM) {
 162                         if (map[i].addr > mem_end)
 163                                 continue;
 164                         if (end > mem_end) {
 165                                 /* Truncate region to max_mem. */
 166                                 map[i].size -= end - mem_end;
 167
 168                                 extra_pages += PFN_DOWN(end - mem_end);
 169                         }
 170                 }
 171                 if (map[i].size > 0)
 172                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 173         }
 174
 175         /*
 176          * Even though this is normal, usable memory under Xen, reserve
 177          * ISA memory anyway because too many things think they can poke
 178          * about in there.
 179          */
 180         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 181                         E820_RESERVED);
 182
 183         /*
 184          * Reserve Xen bits:
 185          *  - mfn_list
 186          *  - xen_start_info
 187          * See comment above "struct start_info" in <xen/interface/xen.h>
 188          */
 189         reserve_early(__pa(xen_start_info->mfn_list),
 190                       __pa(xen_start_info->pt_base),
 191                         "XEN START INFO");
 192
 193         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 194
 195         extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 196
 197         xen_add_extra_mem(extra_pages);
 198
 199         return "Xen";
 200 }
 201
 202 static void xen_idle(void)
 203 {
 204         local_irq_disable();
 205
 206         if (need_resched())
 207                 local_irq_enable();
 208         else {
 209                 current_thread_info()->status &= ~TS_POLLING;
 210                 smp_mb__after_clear_bit();
 211                 safe_halt();
 212                 current_thread_info()->status |= TS_POLLING;
 213         }
 214 }
 215
 216 /*
 217  * Set the bit indicating "nosegneg" library variants should be used.
 218  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 219  * can have un-truncated segments, so wrapping around is allowed.
 220  */
 221 static void __init fiddle_vdso(void)
 222 {
 223 #ifdef CONFIG_X86_32
 224         u32 *mask;
 225         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 226         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 227         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 228         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 229 #endif
 230 }
 231
 232 static __cpuinit int register_callback(unsigned type, const void *func)
 233 {
 234         struct callback_register callback = {
 235                 .type = type,
 236                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 237                 .flags = CALLBACKF_mask_events,
 238         };
 239
 240         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 241 }
 242
 243 void __cpuinit xen_enable_sysenter(void)
 244 {
 245         int ret;
 246         unsigned sysenter_feature;
 247
 248 #ifdef CONFIG_X86_32
 249         sysenter_feature = X86_FEATURE_SEP;
 250 #else
 251         sysenter_feature = X86_FEATURE_SYSENTER32;
 252 #endif
 253
 254         if (!boot_cpu_has(sysenter_feature))
 255                 return;
 256
 257         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 258         if(ret != 0)
 259                 setup_clear_cpu_cap(sysenter_feature);
 260 }
 261
 262 void __cpuinit xen_enable_syscall(void)
 263 {
 264 #ifdef CONFIG_X86_64
 265         int ret;
 266
 267         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 268         if (ret != 0) {
 269                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 270                 /* Pretty fatal; 64-bit userspace has no other
 271                    mechanism for syscalls. */
 272         }
 273
 274         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 275                 ret = register_callback(CALLBACKTYPE_syscall32,
 276                                         xen_syscall32_target);
 277                 if (ret != 0)
 278                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 279         }
 280 #endif /* CONFIG_X86_64 */
 281 }
 282
 283 void __init xen_arch_setup(void)
 284 {
 285         struct physdev_set_iopl set_iopl;
 286         int rc;
 287
 288         xen_panic_handler_init();
 289
 290         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 291         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 292
 293         if (!xen_feature(XENFEAT_auto_translated_physmap))
 294                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 295                                      VMASST_TYPE_pae_extended_cr3);
 296
 297         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 298             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 299                 BUG();
 300
 301         xen_enable_sysenter();
 302         xen_enable_syscall();
 303
 304         set_iopl.iopl = 1;
 305         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 306         if (rc != 0)
 307                 printk(KERN_INFO "physdev_op failed %d\n", rc);
 308
 309 #ifdef CONFIG_ACPI
 310         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 311                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 312                 disable_acpi();
 313         }
 314 #endif
 315
 316         memcpy(boot_command_line, xen_start_info->cmd_line,
 317                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 318                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 319
 320         pm_idle = xen_idle;
 321
 322         paravirt_disable_iospace();
 323
 324         fiddle_vdso();
 325 }