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 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  38                                               phys_addr_t end_addr)
  39 {
  40         struct xen_memory_reservation reservation = {
  41                 .address_bits = 0,
  42                 .extent_order = 0,
  43                 .domid        = DOMID_SELF
  44         };
  45         unsigned long start, end;
  46         unsigned long len = 0;
  47         unsigned long pfn;
  48         int ret;
  49
  50         start = PFN_UP(start_addr);
  51         end = PFN_DOWN(end_addr);
  52
  53         if (end <= start)
  54                 return 0;
  55
  56         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  57                start, end);
  58         for(pfn = start; pfn < end; pfn++) {
  59                 unsigned long mfn = pfn_to_mfn(pfn);
  60
  61                 /* Make sure pfn exists to start with */
  62                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
  63                         continue;
  64
  65                 set_xen_guest_handle(reservation.extent_start, &mfn);
  66                 reservation.nr_extents = 1;
  67
  68                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
  69                                            &reservation);
  70                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
  71                      start, end, ret);
  72                 if (ret == 1) {
  73                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  74                         len++;
  75                 }
  76         }
  77         printk(KERN_CONT "%ld pages freed\n", len);
  78
  79         return len;
  80 }
  81
  82 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
  83                                                      const struct e820map *e820)
  84 {
  85         phys_addr_t max_addr = PFN_PHYS(max_pfn);
  86         phys_addr_t last_end = 0;
  87         unsigned long released = 0;
  88         int i;
  89
  90         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
  91                 phys_addr_t end = e820->map[i].addr;
  92                 end = min(max_addr, end);
  93
  94                 released += xen_release_chunk(last_end, end);
  95                 last_end = e820->map[i].addr + e820->map[i].size;
  96         }
  97
  98         if (last_end < max_addr)
  99                 released += xen_release_chunk(last_end, max_addr);
 100
 101         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 102         return released;
 103 }
 104
 105 /**
 106  * machine_specific_memory_setup - Hook for machine specific memory setup.
 107  **/
 108
 109 char * __init xen_memory_setup(void)
 110 {
 111         static struct e820entry map[E820MAX] __initdata;
 112
 113         unsigned long max_pfn = xen_start_info->nr_pages;
 114         unsigned long long mem_end;
 115         int rc;
 116         struct xen_memory_map memmap;
 117         int i;
 118
 119         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 120         mem_end = PFN_PHYS(max_pfn);
 121
 122         memmap.nr_entries = E820MAX;
 123         set_xen_guest_handle(memmap.buffer, map);
 124
 125         rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
 126         if (rc == -ENOSYS) {
 127                 memmap.nr_entries = 1;
 128                 map[0].addr = 0ULL;
 129                 map[0].size = mem_end;
 130                 /* 8MB slack (to balance backend allocations). */
 131                 map[0].size += 8ULL << 20;
 132                 map[0].type = E820_RAM;
 133                 rc = 0;
 134         }
 135         BUG_ON(rc);
 136
 137         e820.nr_map = 0;
 138         for (i = 0; i < memmap.nr_entries; i++) {
 139                 unsigned long long end = map[i].addr + map[i].size;
 140                 if (map[i].type == E820_RAM) {
 141                         if (map[i].addr > mem_end)
 142                                 continue;
 143                         if (end > mem_end) {
 144                                 /* Truncate region to max_mem. */
 145                                 map[i].size -= end - mem_end;
 146                         }
 147                 }
 148                 if (map[i].size > 0)
 149                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 150         }
 151
 152         /*
 153          * Even though this is normal, usable memory under Xen, reserve
 154          * ISA memory anyway because too many things think they can poke
 155          * about in there.
 156          */
 157         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 158                         E820_RESERVED);
 159
 160         /*
 161          * Reserve Xen bits:
 162          *  - mfn_list
 163          *  - xen_start_info
 164          * See comment above "struct start_info" in <xen/interface/xen.h>
 165          */
 166         reserve_early(__pa(xen_start_info->mfn_list),
 167                       __pa(xen_start_info->pt_base),
 168                         "XEN START INFO");
 169
 170         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 171
 172         xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 173
 174         return "Xen";
 175 }
 176
 177 static void xen_idle(void)
 178 {
 179         local_irq_disable();
 180
 181         if (need_resched())
 182                 local_irq_enable();
 183         else {
 184                 current_thread_info()->status &= ~TS_POLLING;
 185                 smp_mb__after_clear_bit();
 186                 safe_halt();
 187                 current_thread_info()->status |= TS_POLLING;
 188         }
 189 }
 190
 191 /*
 192  * Set the bit indicating "nosegneg" library variants should be used.
 193  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 194  * can have un-truncated segments, so wrapping around is allowed.
 195  */
 196 static void __init fiddle_vdso(void)
 197 {
 198 #ifdef CONFIG_X86_32
 199         u32 *mask;
 200         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 201         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 202         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 203         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 204 #endif
 205 }
 206
 207 static __cpuinit int register_callback(unsigned type, const void *func)
 208 {
 209         struct callback_register callback = {
 210                 .type = type,
 211                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 212                 .flags = CALLBACKF_mask_events,
 213         };
 214
 215         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 216 }
 217
 218 void __cpuinit xen_enable_sysenter(void)
 219 {
 220         int ret;
 221         unsigned sysenter_feature;
 222
 223 #ifdef CONFIG_X86_32
 224         sysenter_feature = X86_FEATURE_SEP;
 225 #else
 226         sysenter_feature = X86_FEATURE_SYSENTER32;
 227 #endif
 228
 229         if (!boot_cpu_has(sysenter_feature))
 230                 return;
 231
 232         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 233         if(ret != 0)
 234                 setup_clear_cpu_cap(sysenter_feature);
 235 }
 236
 237 void __cpuinit xen_enable_syscall(void)
 238 {
 239 #ifdef CONFIG_X86_64
 240         int ret;
 241
 242         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 243         if (ret != 0) {
 244                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 245                 /* Pretty fatal; 64-bit userspace has no other
 246                    mechanism for syscalls. */
 247         }
 248
 249         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 250                 ret = register_callback(CALLBACKTYPE_syscall32,
 251                                         xen_syscall32_target);
 252                 if (ret != 0)
 253                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 254         }
 255 #endif /* CONFIG_X86_64 */
 256 }
 257
 258 void __init xen_arch_setup(void)
 259 {
 260         struct physdev_set_iopl set_iopl;
 261         int rc;
 262
 263         xen_panic_handler_init();
 264
 265         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 266         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 267
 268         if (!xen_feature(XENFEAT_auto_translated_physmap))
 269                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 270                                      VMASST_TYPE_pae_extended_cr3);
 271
 272         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 273             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 274                 BUG();
 275
 276         xen_enable_sysenter();
 277         xen_enable_syscall();
 278
 279         set_iopl.iopl = 1;
 280         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 281         if (rc != 0)
 282                 printk(KERN_INFO "physdev_op failed %d\n", rc);
 283
 284 #ifdef CONFIG_ACPI
 285         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 286                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 287                 disable_acpi();
 288         }
 289 #endif
 290
 291         memcpy(boot_command_line, xen_start_info->cmd_line,
 292                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 293                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 294
 295         pm_idle = xen_idle;
 296
 297         paravirt_disable_iospace();
 298
 299         fiddle_vdso();
 300 }