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                 arch_leave_lazy_mmu_mode();
 277                 arch_enter_lazy_mmu_mode();
 278         }
 279
 280         preempt_enable();
 281 }
 282
 283 void arch_flush_lazy_cpu_mode(void)
 284 {
 285         preempt_disable();
 286
 287         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
 288                 arch_leave_lazy_cpu_mode();
 289                 arch_enter_lazy_cpu_mode();
 290         }
 291
 292         preempt_enable();
 293 }
 294
 295 struct pv_info pv_info = {
 296         .name = "bare hardware",
 297         .paravirt_enabled = 0,
 298         .kernel_rpl = 0,
 299         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 300 };
 301
 302 struct pv_init_ops pv_init_ops = {
 303         .patch = native_patch,
 304         .banner = default_banner,
 305         .arch_setup = paravirt_nop,
 306         .memory_setup = machine_specific_memory_setup,
 307 };
 308
 309 struct pv_time_ops pv_time_ops = {
 310         .time_init = hpet_time_init,
 311         .get_wallclock = native_get_wallclock,
 312         .set_wallclock = native_set_wallclock,
 313         .sched_clock = native_sched_clock,
 314         .get_tsc_khz = native_calibrate_tsc,
 315 };
 316
 317 struct pv_irq_ops pv_irq_ops = {
 318         .init_IRQ = native_init_IRQ,
 319         .save_fl = native_save_fl,
 320         .restore_fl = native_restore_fl,
 321         .irq_disable = native_irq_disable,
 322         .irq_enable = native_irq_enable,
 323         .safe_halt = native_safe_halt,
 324         .halt = native_halt,
 325 #ifdef CONFIG_X86_64
 326         .adjust_exception_frame = paravirt_nop,
 327 #endif
 328 };
 329
 330 struct pv_cpu_ops pv_cpu_ops = {
 331         .cpuid = native_cpuid,
 332         .get_debugreg = native_get_debugreg,
 333         .set_debugreg = native_set_debugreg,
 334         .clts = native_clts,
 335         .read_cr0 = native_read_cr0,
 336         .write_cr0 = native_write_cr0,
 337         .read_cr4 = native_read_cr4,
 338         .read_cr4_safe = native_read_cr4_safe,
 339         .write_cr4 = native_write_cr4,
 340 #ifdef CONFIG_X86_64
 341         .read_cr8 = native_read_cr8,
 342         .write_cr8 = native_write_cr8,
 343 #endif
 344         .wbinvd = native_wbinvd,
 345         .read_msr = native_read_msr_safe,
 346         .read_msr_amd = native_read_msr_amd_safe,
 347         .write_msr = native_write_msr_safe,
 348         .read_tsc = native_read_tsc,
 349         .read_pmc = native_read_pmc,
 350         .read_tscp = native_read_tscp,
 351         .load_tr_desc = native_load_tr_desc,
 352         .set_ldt = native_set_ldt,
 353         .load_gdt = native_load_gdt,
 354         .load_idt = native_load_idt,
 355         .store_gdt = native_store_gdt,
 356         .store_idt = native_store_idt,
 357         .store_tr = native_store_tr,
 358         .load_tls = native_load_tls,
 359 #ifdef CONFIG_X86_64
 360         .load_gs_index = native_load_gs_index,
 361 #endif
 362         .write_ldt_entry = native_write_ldt_entry,
 363         .write_gdt_entry = native_write_gdt_entry,
 364         .write_idt_entry = native_write_idt_entry,
 365
 366         .alloc_ldt = paravirt_nop,
 367         .free_ldt = paravirt_nop,
 368
 369         .load_sp0 = native_load_sp0,
 370
 371 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 372         .irq_enable_sysexit = native_irq_enable_sysexit,
 373 #endif
 374 #ifdef CONFIG_X86_64
 375 #ifdef CONFIG_IA32_EMULATION
 376         .usergs_sysret32 = native_usergs_sysret32,
 377 #endif
 378         .usergs_sysret64 = native_usergs_sysret64,
 379 #endif
 380         .iret = native_iret,
 381         .swapgs = native_swapgs,
 382
 383         .set_iopl_mask = native_set_iopl_mask,
 384         .io_delay = native_io_delay,
 385
 386         .lazy_mode = {
 387                 .enter = paravirt_nop,
 388                 .leave = paravirt_nop,
 389         },
 390 };
 391
 392 struct pv_apic_ops pv_apic_ops = {
 393 #ifdef CONFIG_X86_LOCAL_APIC
 394         .setup_boot_clock = setup_boot_APIC_clock,
 395         .setup_secondary_clock = setup_secondary_APIC_clock,
 396         .startup_ipi_hook = paravirt_nop,
 397 #endif
 398 };
 399
 400 struct pv_mmu_ops pv_mmu_ops = {
 401 #ifndef CONFIG_X86_64
 402         .pagetable_setup_start = native_pagetable_setup_start,
 403         .pagetable_setup_done = native_pagetable_setup_done,
 404 #else
 405         .pagetable_setup_start = paravirt_nop,
 406         .pagetable_setup_done = paravirt_nop,
 407 #endif
 408
 409         .read_cr2 = native_read_cr2,
 410         .write_cr2 = native_write_cr2,
 411         .read_cr3 = native_read_cr3,
 412         .write_cr3 = native_write_cr3,
 413
 414         .flush_tlb_user = native_flush_tlb,
 415         .flush_tlb_kernel = native_flush_tlb_global,
 416         .flush_tlb_single = native_flush_tlb_single,
 417         .flush_tlb_others = native_flush_tlb_others,
 418
 419         .pgd_alloc = __paravirt_pgd_alloc,
 420         .pgd_free = paravirt_nop,
 421
 422         .alloc_pte = paravirt_nop,
 423         .alloc_pmd = paravirt_nop,
 424         .alloc_pmd_clone = paravirt_nop,
 425         .alloc_pud = paravirt_nop,
 426         .release_pte = paravirt_nop,
 427         .release_pmd = paravirt_nop,
 428         .release_pud = paravirt_nop,
 429
 430         .set_pte = native_set_pte,
 431         .set_pte_at = native_set_pte_at,
 432         .set_pmd = native_set_pmd,
 433         .pte_update = paravirt_nop,
 434         .pte_update_defer = paravirt_nop,
 435
 436         .ptep_modify_prot_start = __ptep_modify_prot_start,
 437         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 438
 439 #ifdef CONFIG_HIGHPTE
 440         .kmap_atomic_pte = kmap_atomic,
 441 #endif
 442
 443 #if PAGETABLE_LEVELS >= 3
 444 #ifdef CONFIG_X86_PAE
 445         .set_pte_atomic = native_set_pte_atomic,
 446         .set_pte_present = native_set_pte_present,
 447         .pte_clear = native_pte_clear,
 448         .pmd_clear = native_pmd_clear,
 449 #endif
 450         .set_pud = native_set_pud,
 451         .pmd_val = native_pmd_val,
 452         .make_pmd = native_make_pmd,
 453
 454 #if PAGETABLE_LEVELS == 4
 455         .pud_val = native_pud_val,
 456         .make_pud = native_make_pud,
 457         .set_pgd = native_set_pgd,
 458 #endif
 459 #endif /* PAGETABLE_LEVELS >= 3 */
 460
 461         .pte_val = native_pte_val,
 462         .pte_flags = native_pte_flags,
 463         .pgd_val = native_pgd_val,
 464
 465         .make_pte = native_make_pte,
 466         .make_pgd = native_make_pgd,
 467
 468         .dup_mmap = paravirt_nop,
 469         .exit_mmap = paravirt_nop,
 470         .activate_mm = paravirt_nop,
 471
 472         .lazy_mode = {
 473                 .enter = paravirt_nop,
 474                 .leave = paravirt_nop,
 475         },
 476
 477         .set_fixmap = native_set_fixmap,
 478 };
 479
 480 EXPORT_SYMBOL_GPL(pv_time_ops);
 481 EXPORT_SYMBOL    (pv_cpu_ops);
 482 EXPORT_SYMBOL    (pv_mmu_ops);
 483 EXPORT_SYMBOL_GPL(pv_apic_ops);
 484 EXPORT_SYMBOL_GPL(pv_info);
 485 EXPORT_SYMBOL    (pv_irq_ops);