arch/x86/kernel/paravirt.c

   1 /*  Paravirtualization interfaces
   2     Copyright (C) 2006 Rusty Russell IBM Corporation
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17
  18     2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
  19 */
  20
  21 #include <linux/errno.h>
  22 #include <linux/module.h>
  23 #include <linux/efi.h>
  24 #include <linux/bcd.h>
  25 #include <linux/highmem.h>
  26
  27 #include <asm/bug.h>
  28 #include <asm/paravirt.h>
  29 #include <asm/desc.h>
  30 #include <asm/setup.h>
  31 #include <asm/arch_hooks.h>
  32 #include <asm/pgtable.h>
  33 #include <asm/time.h>
  34 #include <asm/pgalloc.h>
  35 #include <asm/irq.h>
  36 #include <asm/delay.h>
  37 #include <asm/fixmap.h>
  38 #include <asm/apic.h>
  39 #include <asm/tlbflush.h>
  40 #include <asm/timer.h>
  41
  42 /* nop stub */
  43 void _paravirt_nop(void)
  44 {
  45 }
  46
  47 static void __init default_banner(void)
  48 {
  49         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  50                pv_info.name);
  51 }
  52
  53 char *memory_setup(void)
  54 {
  55         return pv_init_ops.memory_setup();
  56 }
  57
  58 /* Simple instruction patching code. */
  59 #define DEF_NATIVE(ops, name, code)                                     \
  60         extern const char start_##ops##_##name[], end_##ops##_##name[]; \
  61         asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
  62
  63 /* Undefined instruction for dealing with missing ops pointers. */
  64 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  65
  66 unsigned paravirt_patch_nop(void)
  67 {
  68         return 0;
  69 }
  70
  71 unsigned paravirt_patch_ignore(unsigned len)
  72 {
  73         return len;
  74 }
  75
  76 struct branch {
  77         unsigned char opcode;
  78         u32 delta;
  79 } __attribute__((packed));
  80
  81 unsigned paravirt_patch_call(void *insnbuf,
  82                              const void *target, u16 tgt_clobbers,
  83                              unsigned long addr, u16 site_clobbers,
  84                              unsigned len)
  85 {
  86         struct branch *b = insnbuf;
  87         unsigned long delta = (unsigned long)target - (addr+5);
  88
  89         if (tgt_clobbers & ~site_clobbers)
  90                 return len;     /* target would clobber too much for this site */
  91         if (len < 5)
  92                 return len;     /* call too long for patch site */
  93
  94         b->opcode = 0xe8; /* call */
  95         b->delta = delta;
  96         BUILD_BUG_ON(sizeof(*b) != 5);
  97
  98         return 5;
  99 }
 100
 101 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 102                             unsigned long addr, unsigned len)
 103 {
 104         struct branch *b = insnbuf;
 105         unsigned long delta = (unsigned long)target - (addr+5);
 106
 107         if (len < 5)
 108                 return len;     /* call too long for patch site */
 109
 110         b->opcode = 0xe9;       /* jmp */
 111         b->delta = delta;
 112
 113         return 5;
 114 }
 115
 116 /* Neat trick to map patch type back to the call within the
 117  * corresponding structure. */
 118 static void *get_call_destination(u8 type)
 119 {
 120         struct paravirt_patch_template tmpl = {
 121                 .pv_init_ops = pv_init_ops,
 122                 .pv_time_ops = pv_time_ops,
 123                 .pv_cpu_ops = pv_cpu_ops,
 124                 .pv_irq_ops = pv_irq_ops,
 125                 .pv_apic_ops = pv_apic_ops,
 126                 .pv_mmu_ops = pv_mmu_ops,
 127                 .pv_lock_ops = pv_lock_ops,
 128         };
 129         return *((void **)&tmpl + type);
 130 }
 131
 132 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 133                                 unsigned long addr, unsigned len)
 134 {
 135         void *opfunc = get_call_destination(type);
 136         unsigned ret;
 137
 138         if (opfunc == NULL)
 139                 /* If there's no function, patch it with a ud2a (BUG) */
 140                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 141         else if (opfunc == paravirt_nop)
 142                 /* If the operation is a nop, then nop the callsite */
 143                 ret = paravirt_patch_nop();
 144         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
 145                  type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
 146                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
 147                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
 148                 /* If operation requires a jmp, then jmp */
 149                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 150         else
 151                 /* Otherwise call the function; assume target could
 152                    clobber any caller-save reg */
 153                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 154                                           addr, clobbers, len);
 155
 156         return ret;
 157 }
 158
 159 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 160                               const char *start, const char *end)
 161 {
 162         unsigned insn_len = end - start;
 163
 164         if (insn_len > len || start == NULL)
 165                 insn_len = len;
 166         else
 167                 memcpy(insnbuf, start, insn_len);
 168
 169         return insn_len;
 170 }
 171
 172 void init_IRQ(void)
 173 {
 174         pv_irq_ops.init_IRQ();
 175 }
 176
 177 static void native_flush_tlb(void)
 178 {
 179         __native_flush_tlb();
 180 }
 181
 182 /*
 183  * Global pages have to be flushed a bit differently. Not a real
 184  * performance problem because this does not happen often.
 185  */
 186 static void native_flush_tlb_global(void)
 187 {
 188         __native_flush_tlb_global();
 189 }
 190
 191 static void native_flush_tlb_single(unsigned long addr)
 192 {
 193         __native_flush_tlb_single(addr);
 194 }
 195
 196 /* These are in entry.S */
 197 extern void native_iret(void);
 198 extern void native_irq_enable_sysexit(void);
 199 extern void native_usergs_sysret32(void);
 200 extern void native_usergs_sysret64(void);
 201
 202 static int __init print_banner(void)
 203 {
 204         pv_init_ops.banner();
 205         return 0;
 206 }
 207 core_initcall(print_banner);
 208
 209 static struct resource reserve_ioports = {
 210         .start = 0,
 211         .end = IO_SPACE_LIMIT,
 212         .name = "paravirt-ioport",
 213         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 214 };
 215
 216 /*
 217  * Reserve the whole legacy IO space to prevent any legacy drivers
 218  * from wasting time probing for their hardware.  This is a fairly
 219  * brute-force approach to disabling all non-virtual drivers.
 220  *
 221  * Note that this must be called very early to have any effect.
 222  */
 223 int paravirt_disable_iospace(void)
 224 {
 225         return request_resource(&ioport_resource, &reserve_ioports);
 226 }
 227
 228 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 229
 230 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 231 {
 232         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 233         BUG_ON(preemptible());
 234
 235         __get_cpu_var(paravirt_lazy_mode) = mode;
 236 }
 237
 238 void paravirt_leave_lazy(enum paravirt_lazy_mode mode)
 239 {
 240         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
 241         BUG_ON(preemptible());
 242
 243         __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 244 }
 245
 246 void paravirt_enter_lazy_mmu(void)
 247 {
 248         enter_lazy(PARAVIRT_LAZY_MMU);
 249 }
 250
 251 void paravirt_leave_lazy_mmu(void)
 252 {
 253         paravirt_leave_lazy(PARAVIRT_LAZY_MMU);
 254 }
 255
 256 void paravirt_enter_lazy_cpu(void)
 257 {
 258         enter_lazy(PARAVIRT_LAZY_CPU);
 259 }
 260
 261 void paravirt_leave_lazy_cpu(void)
 262 {
 263         paravirt_leave_lazy(PARAVIRT_LAZY_CPU);
 264 }
 265
 266 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 267 {
 268         return __get_cpu_var(paravirt_lazy_mode);
 269 }
 270
 271 void arch_flush_lazy_mmu_mode(void)
 272 {
 273         preempt_disable();
 274
 275         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 276                 WARN_ON(preempt_count() == 1);
 277                 arch_leave_lazy_mmu_mode();
 278                 arch_enter_lazy_mmu_mode();
 279         }
 280
 281         preempt_enable();
 282 }
 283
 284 void arch_flush_lazy_cpu_mode(void)
 285 {
 286         preempt_disable();
 287
 288         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
 289                 WARN_ON(preempt_count() == 1);
 290                 arch_leave_lazy_cpu_mode();
 291                 arch_enter_lazy_cpu_mode();
 292         }
 293
 294         preempt_enable();
 295 }
 296
 297 struct pv_info pv_info = {
 298         .name = "bare hardware",
 299         .paravirt_enabled = 0,
 300         .kernel_rpl = 0,
 301         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 302 };
 303
 304 struct pv_init_ops pv_init_ops = {
 305         .patch = native_patch,
 306         .banner = default_banner,
 307         .arch_setup = paravirt_nop,
 308         .memory_setup = machine_specific_memory_setup,
 309 };
 310
 311 struct pv_time_ops pv_time_ops = {
 312         .time_init = hpet_time_init,
 313         .get_wallclock = native_get_wallclock,
 314         .set_wallclock = native_set_wallclock,
 315         .sched_clock = native_sched_clock,
 316         .get_tsc_khz = native_calibrate_tsc,
 317 };
 318
 319 struct pv_irq_ops pv_irq_ops = {
 320         .init_IRQ = native_init_IRQ,
 321         .save_fl = native_save_fl,
 322         .restore_fl = native_restore_fl,
 323         .irq_disable = native_irq_disable,
 324         .irq_enable = native_irq_enable,
 325         .safe_halt = native_safe_halt,
 326         .halt = native_halt,
 327 #ifdef CONFIG_X86_64
 328         .adjust_exception_frame = paravirt_nop,
 329 #endif
 330 };
 331
 332 struct pv_cpu_ops pv_cpu_ops = {
 333         .cpuid = native_cpuid,
 334         .get_debugreg = native_get_debugreg,
 335         .set_debugreg = native_set_debugreg,
 336         .clts = native_clts,
 337         .read_cr0 = native_read_cr0,
 338         .write_cr0 = native_write_cr0,
 339         .read_cr4 = native_read_cr4,
 340         .read_cr4_safe = native_read_cr4_safe,
 341         .write_cr4 = native_write_cr4,
 342 #ifdef CONFIG_X86_64
 343         .read_cr8 = native_read_cr8,
 344         .write_cr8 = native_write_cr8,
 345 #endif
 346         .wbinvd = native_wbinvd,
 347         .read_msr = native_read_msr_safe,
 348         .read_msr_amd = native_read_msr_amd_safe,
 349         .write_msr = native_write_msr_safe,
 350         .read_tsc = native_read_tsc,
 351         .read_pmc = native_read_pmc,
 352         .read_tscp = native_read_tscp,
 353         .load_tr_desc = native_load_tr_desc,
 354         .set_ldt = native_set_ldt,
 355         .load_gdt = native_load_gdt,
 356         .load_idt = native_load_idt,
 357         .store_gdt = native_store_gdt,
 358         .store_idt = native_store_idt,
 359         .store_tr = native_store_tr,
 360         .load_tls = native_load_tls,
 361 #ifdef CONFIG_X86_64
 362         .load_gs_index = native_load_gs_index,
 363 #endif
 364         .write_ldt_entry = native_write_ldt_entry,
 365         .write_gdt_entry = native_write_gdt_entry,
 366         .write_idt_entry = native_write_idt_entry,
 367
 368         .alloc_ldt = paravirt_nop,
 369         .free_ldt = paravirt_nop,
 370
 371         .load_sp0 = native_load_sp0,
 372
 373 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 374         .irq_enable_sysexit = native_irq_enable_sysexit,
 375 #endif
 376 #ifdef CONFIG_X86_64
 377 #ifdef CONFIG_IA32_EMULATION
 378         .usergs_sysret32 = native_usergs_sysret32,
 379 #endif
 380         .usergs_sysret64 = native_usergs_sysret64,
 381 #endif
 382         .iret = native_iret,
 383         .swapgs = native_swapgs,
 384
 385         .set_iopl_mask = native_set_iopl_mask,
 386         .io_delay = native_io_delay,
 387
 388         .lazy_mode = {
 389                 .enter = paravirt_nop,
 390                 .leave = paravirt_nop,
 391         },
 392 };
 393
 394 struct pv_apic_ops pv_apic_ops = {
 395 #ifdef CONFIG_X86_LOCAL_APIC
 396         .setup_boot_clock = setup_boot_APIC_clock,
 397         .setup_secondary_clock = setup_secondary_APIC_clock,
 398         .startup_ipi_hook = paravirt_nop,
 399 #endif
 400 };
 401
 402 struct pv_mmu_ops pv_mmu_ops = {
 403 #ifndef CONFIG_X86_64
 404         .pagetable_setup_start = native_pagetable_setup_start,
 405         .pagetable_setup_done = native_pagetable_setup_done,
 406 #else
 407         .pagetable_setup_start = paravirt_nop,
 408         .pagetable_setup_done = paravirt_nop,
 409 #endif
 410
 411         .read_cr2 = native_read_cr2,
 412         .write_cr2 = native_write_cr2,
 413         .read_cr3 = native_read_cr3,
 414         .write_cr3 = native_write_cr3,
 415
 416         .flush_tlb_user = native_flush_tlb,
 417         .flush_tlb_kernel = native_flush_tlb_global,
 418         .flush_tlb_single = native_flush_tlb_single,
 419         .flush_tlb_others = native_flush_tlb_others,
 420
 421         .pgd_alloc = __paravirt_pgd_alloc,
 422         .pgd_free = paravirt_nop,
 423
 424         .alloc_pte = paravirt_nop,
 425         .alloc_pmd = paravirt_nop,
 426         .alloc_pmd_clone = paravirt_nop,
 427         .alloc_pud = paravirt_nop,
 428         .release_pte = paravirt_nop,
 429         .release_pmd = paravirt_nop,
 430         .release_pud = paravirt_nop,
 431
 432         .set_pte = native_set_pte,
 433         .set_pte_at = native_set_pte_at,
 434         .set_pmd = native_set_pmd,
 435         .pte_update = paravirt_nop,
 436         .pte_update_defer = paravirt_nop,
 437
 438         .ptep_modify_prot_start = __ptep_modify_prot_start,
 439         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 440
 441 #ifdef CONFIG_HIGHPTE
 442         .kmap_atomic_pte = kmap_atomic,
 443 #endif
 444
 445 #if PAGETABLE_LEVELS >= 3
 446 #ifdef CONFIG_X86_PAE
 447         .set_pte_atomic = native_set_pte_atomic,
 448         .set_pte_present = native_set_pte_present,
 449         .pte_clear = native_pte_clear,
 450         .pmd_clear = native_pmd_clear,
 451 #endif
 452         .set_pud = native_set_pud,
 453         .pmd_val = native_pmd_val,
 454         .make_pmd = native_make_pmd,
 455
 456 #if PAGETABLE_LEVELS == 4
 457         .pud_val = native_pud_val,
 458         .make_pud = native_make_pud,
 459         .set_pgd = native_set_pgd,
 460 #endif
 461 #endif /* PAGETABLE_LEVELS >= 3 */
 462
 463         .pte_val = native_pte_val,
 464         .pte_flags = native_pte_flags,
 465         .pgd_val = native_pgd_val,
 466
 467         .make_pte = native_make_pte,
 468         .make_pgd = native_make_pgd,
 469
 470         .dup_mmap = paravirt_nop,
 471         .exit_mmap = paravirt_nop,
 472         .activate_mm = paravirt_nop,
 473
 474         .lazy_mode = {
 475                 .enter = paravirt_nop,
 476                 .leave = paravirt_nop,
 477         },
 478
 479         .set_fixmap = native_set_fixmap,
 480 };
 481
 482 EXPORT_SYMBOL_GPL(pv_time_ops);
 483 EXPORT_SYMBOL    (pv_cpu_ops);
 484 EXPORT_SYMBOL    (pv_mmu_ops);
 485 EXPORT_SYMBOL_GPL(pv_apic_ops);
 486 EXPORT_SYMBOL_GPL(pv_info);
 487 EXPORT_SYMBOL    (pv_irq_ops);