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/debugreg.h>
  30 #include <asm/desc.h>
  31 #include <asm/setup.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 #include <asm/special_insns.h>
  42
  43 /*
  44  * nop stub, which must not clobber anything *including the stack* to
  45  * avoid confusing the entry prologues.
  46  */
  47 extern void _paravirt_nop(void);
  48 asm (".pushsection .entry.text, \"ax\"\n"
  49      ".global _paravirt_nop\n"
  50      "_paravirt_nop:\n\t"
  51      "ret\n\t"
  52      ".size _paravirt_nop, . - _paravirt_nop\n\t"
  53      ".type _paravirt_nop, @function\n\t"
  54      ".popsection");
  55
  56 /* identity function, which can be inlined */
  57 u32 notrace _paravirt_ident_32(u32 x)
  58 {
  59         return x;
  60 }
  61
  62 u64 notrace _paravirt_ident_64(u64 x)
  63 {
  64         return x;
  65 }
  66
  67 void __init default_banner(void)
  68 {
  69         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  70                pv_info.name);
  71 }
  72
  73 /* Undefined instruction for dealing with missing ops pointers. */
  74 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  75
  76 unsigned paravirt_patch_nop(void)
  77 {
  78         return 0;
  79 }
  80
  81 unsigned paravirt_patch_ignore(unsigned len)
  82 {
  83         return len;
  84 }
  85
  86 struct branch {
  87         unsigned char opcode;
  88         u32 delta;
  89 } __attribute__((packed));
  90
  91 unsigned paravirt_patch_call(void *insnbuf,
  92                              const void *target, u16 tgt_clobbers,
  93                              unsigned long addr, u16 site_clobbers,
  94                              unsigned len)
  95 {
  96         struct branch *b = insnbuf;
  97         unsigned long delta = (unsigned long)target - (addr+5);
  98
  99         if (tgt_clobbers & ~site_clobbers)
 100                 return len;     /* target would clobber too much for this site */
 101         if (len < 5)
 102                 return len;     /* call too long for patch site */
 103
 104         b->opcode = 0xe8; /* call */
 105         b->delta = delta;
 106         BUILD_BUG_ON(sizeof(*b) != 5);
 107
 108         return 5;
 109 }
 110
 111 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 112                             unsigned long addr, unsigned len)
 113 {
 114         struct branch *b = insnbuf;
 115         unsigned long delta = (unsigned long)target - (addr+5);
 116
 117         if (len < 5)
 118                 return len;     /* call too long for patch site */
 119
 120         b->opcode = 0xe9;       /* jmp */
 121         b->delta = delta;
 122
 123         return 5;
 124 }
 125
 126 /* Neat trick to map patch type back to the call within the
 127  * corresponding structure. */
 128 static void *get_call_destination(u8 type)
 129 {
 130         struct paravirt_patch_template tmpl = {
 131                 .pv_init_ops = pv_init_ops,
 132                 .pv_time_ops = pv_time_ops,
 133                 .pv_cpu_ops = pv_cpu_ops,
 134                 .pv_irq_ops = pv_irq_ops,
 135                 .pv_apic_ops = pv_apic_ops,
 136                 .pv_mmu_ops = pv_mmu_ops,
 137 #ifdef CONFIG_PARAVIRT_SPINLOCKS
 138                 .pv_lock_ops = pv_lock_ops,
 139 #endif
 140         };
 141         return *((void **)&tmpl + type);
 142 }
 143
 144 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 145                                 unsigned long addr, unsigned len)
 146 {
 147         void *opfunc = get_call_destination(type);
 148         unsigned ret;
 149
 150         if (opfunc == NULL)
 151                 /* If there's no function, patch it with a ud2a (BUG) */
 152                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 153         else if (opfunc == _paravirt_nop)
 154                 /* If the operation is a nop, then nop the callsite */
 155                 ret = paravirt_patch_nop();
 156
 157         /* identity functions just return their single argument */
 158         else if (opfunc == _paravirt_ident_32)
 159                 ret = paravirt_patch_ident_32(insnbuf, len);
 160         else if (opfunc == _paravirt_ident_64)
 161                 ret = paravirt_patch_ident_64(insnbuf, len);
 162
 163         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
 164                  type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
 165                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
 166                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
 167                 /* If operation requires a jmp, then jmp */
 168                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 169         else
 170                 /* Otherwise call the function; assume target could
 171                    clobber any caller-save reg */
 172                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 173                                           addr, clobbers, len);
 174
 175         return ret;
 176 }
 177
 178 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 179                               const char *start, const char *end)
 180 {
 181         unsigned insn_len = end - start;
 182
 183         if (insn_len > len || start == NULL)
 184                 insn_len = len;
 185         else
 186                 memcpy(insnbuf, start, insn_len);
 187
 188         return insn_len;
 189 }
 190
 191 static void native_flush_tlb(void)
 192 {
 193         __native_flush_tlb();
 194 }
 195
 196 /*
 197  * Global pages have to be flushed a bit differently. Not a real
 198  * performance problem because this does not happen often.
 199  */
 200 static void native_flush_tlb_global(void)
 201 {
 202         __native_flush_tlb_global();
 203 }
 204
 205 static void native_flush_tlb_single(unsigned long addr)
 206 {
 207         __native_flush_tlb_single(addr);
 208 }
 209
 210 struct static_key paravirt_steal_enabled;
 211 struct static_key paravirt_steal_rq_enabled;
 212
 213 static u64 native_steal_clock(int cpu)
 214 {
 215         return 0;
 216 }
 217
 218 /* These are in entry.S */
 219 extern void native_iret(void);
 220 extern void native_irq_enable_sysexit(void);
 221 extern void native_usergs_sysret32(void);
 222 extern void native_usergs_sysret64(void);
 223
 224 static struct resource reserve_ioports = {
 225         .start = 0,
 226         .end = IO_SPACE_LIMIT,
 227         .name = "paravirt-ioport",
 228         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 229 };
 230
 231 /*
 232  * Reserve the whole legacy IO space to prevent any legacy drivers
 233  * from wasting time probing for their hardware.  This is a fairly
 234  * brute-force approach to disabling all non-virtual drivers.
 235  *
 236  * Note that this must be called very early to have any effect.
 237  */
 238 int paravirt_disable_iospace(void)
 239 {
 240         return request_resource(&ioport_resource, &reserve_ioports);
 241 }
 242
 243 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 244
 245 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 246 {
 247         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 248
 249         this_cpu_write(paravirt_lazy_mode, mode);
 250 }
 251
 252 static void leave_lazy(enum paravirt_lazy_mode mode)
 253 {
 254         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 255
 256         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 257 }
 258
 259 void paravirt_enter_lazy_mmu(void)
 260 {
 261         enter_lazy(PARAVIRT_LAZY_MMU);
 262 }
 263
 264 void paravirt_leave_lazy_mmu(void)
 265 {
 266         leave_lazy(PARAVIRT_LAZY_MMU);
 267 }
 268
 269 void paravirt_flush_lazy_mmu(void)
 270 {
 271         preempt_disable();
 272
 273         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 274                 arch_leave_lazy_mmu_mode();
 275                 arch_enter_lazy_mmu_mode();
 276         }
 277
 278         preempt_enable();
 279 }
 280
 281 void paravirt_start_context_switch(struct task_struct *prev)
 282 {
 283         BUG_ON(preemptible());
 284
 285         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 286                 arch_leave_lazy_mmu_mode();
 287                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 288         }
 289         enter_lazy(PARAVIRT_LAZY_CPU);
 290 }
 291
 292 void paravirt_end_context_switch(struct task_struct *next)
 293 {
 294         BUG_ON(preemptible());
 295
 296         leave_lazy(PARAVIRT_LAZY_CPU);
 297
 298         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 299                 arch_enter_lazy_mmu_mode();
 300 }
 301
 302 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 303 {
 304         if (in_interrupt())
 305                 return PARAVIRT_LAZY_NONE;
 306
 307         return this_cpu_read(paravirt_lazy_mode);
 308 }
 309
 310 struct pv_info pv_info = {
 311         .name = "bare hardware",
 312         .paravirt_enabled = 0,
 313         .kernel_rpl = 0,
 314         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 315
 316 #ifdef CONFIG_X86_64
 317         .extra_user_64bit_cs = __USER_CS,
 318 #endif
 319 };
 320
 321 struct pv_init_ops pv_init_ops = {
 322         .patch = native_patch,
 323 };
 324
 325 struct pv_time_ops pv_time_ops = {
 326         .sched_clock = native_sched_clock,
 327         .steal_clock = native_steal_clock,
 328 };
 329
 330 __visible struct pv_irq_ops pv_irq_ops = {
 331         .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
 332         .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
 333         .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
 334         .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
 335         .safe_halt = native_safe_halt,
 336         .halt = native_halt,
 337 #ifdef CONFIG_X86_64
 338         .adjust_exception_frame = paravirt_nop,
 339 #endif
 340 };
 341
 342 __visible struct pv_cpu_ops pv_cpu_ops = {
 343         .cpuid = native_cpuid,
 344         .get_debugreg = native_get_debugreg,
 345         .set_debugreg = native_set_debugreg,
 346         .clts = native_clts,
 347         .read_cr0 = native_read_cr0,
 348         .write_cr0 = native_write_cr0,
 349         .read_cr4 = native_read_cr4,
 350         .read_cr4_safe = native_read_cr4_safe,
 351         .write_cr4 = native_write_cr4,
 352 #ifdef CONFIG_X86_64
 353         .read_cr8 = native_read_cr8,
 354         .write_cr8 = native_write_cr8,
 355 #endif
 356         .wbinvd = native_wbinvd,
 357         .read_msr = native_read_msr_safe,
 358         .write_msr = native_write_msr_safe,
 359         .read_tsc = native_read_tsc,
 360         .read_pmc = native_read_pmc,
 361         .read_tscp = native_read_tscp,
 362         .load_tr_desc = native_load_tr_desc,
 363         .set_ldt = native_set_ldt,
 364         .load_gdt = native_load_gdt,
 365         .load_idt = native_load_idt,
 366         .store_idt = native_store_idt,
 367         .store_tr = native_store_tr,
 368         .load_tls = native_load_tls,
 369 #ifdef CONFIG_X86_64
 370         .load_gs_index = native_load_gs_index,
 371 #endif
 372         .write_ldt_entry = native_write_ldt_entry,
 373         .write_gdt_entry = native_write_gdt_entry,
 374         .write_idt_entry = native_write_idt_entry,
 375
 376         .alloc_ldt = paravirt_nop,
 377         .free_ldt = paravirt_nop,
 378
 379         .load_sp0 = native_load_sp0,
 380
 381 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 382         .irq_enable_sysexit = native_irq_enable_sysexit,
 383 #endif
 384 #ifdef CONFIG_X86_64
 385 #ifdef CONFIG_IA32_EMULATION
 386         .usergs_sysret32 = native_usergs_sysret32,
 387 #endif
 388         .usergs_sysret64 = native_usergs_sysret64,
 389 #endif
 390         .iret = native_iret,
 391         .swapgs = native_swapgs,
 392
 393         .set_iopl_mask = native_set_iopl_mask,
 394         .io_delay = native_io_delay,
 395
 396         .start_context_switch = paravirt_nop,
 397         .end_context_switch = paravirt_nop,
 398 };
 399
 400 struct pv_apic_ops pv_apic_ops = {
 401 #ifdef CONFIG_X86_LOCAL_APIC
 402         .startup_ipi_hook = paravirt_nop,
 403 #endif
 404 };
 405
 406 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
 407 /* 32-bit pagetable entries */
 408 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
 409 #else
 410 /* 64-bit pagetable entries */
 411 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 412 #endif
 413
 414 struct pv_mmu_ops pv_mmu_ops = {
 415
 416         .read_cr2 = native_read_cr2,
 417         .write_cr2 = native_write_cr2,
 418         .read_cr3 = native_read_cr3,
 419         .write_cr3 = native_write_cr3,
 420
 421         .flush_tlb_user = native_flush_tlb,
 422         .flush_tlb_kernel = native_flush_tlb_global,
 423         .flush_tlb_single = native_flush_tlb_single,
 424         .flush_tlb_others = native_flush_tlb_others,
 425
 426         .pgd_alloc = __paravirt_pgd_alloc,
 427         .pgd_free = paravirt_nop,
 428
 429         .alloc_pte = paravirt_nop,
 430         .alloc_pmd = paravirt_nop,
 431         .alloc_pud = paravirt_nop,
 432         .release_pte = paravirt_nop,
 433         .release_pmd = paravirt_nop,
 434         .release_pud = paravirt_nop,
 435
 436         .set_pte = native_set_pte,
 437         .set_pte_at = native_set_pte_at,
 438         .set_pmd = native_set_pmd,
 439         .set_pmd_at = native_set_pmd_at,
 440         .pte_update = paravirt_nop,
 441         .pte_update_defer = paravirt_nop,
 442         .pmd_update = paravirt_nop,
 443         .pmd_update_defer = paravirt_nop,
 444
 445         .ptep_modify_prot_start = __ptep_modify_prot_start,
 446         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 447
 448 #if PAGETABLE_LEVELS >= 3
 449 #ifdef CONFIG_X86_PAE
 450         .set_pte_atomic = native_set_pte_atomic,
 451         .pte_clear = native_pte_clear,
 452         .pmd_clear = native_pmd_clear,
 453 #endif
 454         .set_pud = native_set_pud,
 455
 456         .pmd_val = PTE_IDENT,
 457         .make_pmd = PTE_IDENT,
 458
 459 #if PAGETABLE_LEVELS == 4
 460         .pud_val = PTE_IDENT,
 461         .make_pud = PTE_IDENT,
 462
 463         .set_pgd = native_set_pgd,
 464 #endif
 465 #endif /* PAGETABLE_LEVELS >= 3 */
 466
 467         .pte_val = PTE_IDENT,
 468         .pgd_val = PTE_IDENT,
 469
 470         .make_pte = PTE_IDENT,
 471         .make_pgd = PTE_IDENT,
 472
 473         .dup_mmap = paravirt_nop,
 474         .exit_mmap = paravirt_nop,
 475         .activate_mm = paravirt_nop,
 476
 477         .lazy_mode = {
 478                 .enter = paravirt_nop,
 479                 .leave = paravirt_nop,
 480                 .flush = paravirt_nop,
 481         },
 482
 483         .set_fixmap = native_set_fixmap,
 484 };
 485
 486 EXPORT_SYMBOL_GPL(pv_time_ops);
 487 EXPORT_SYMBOL    (pv_cpu_ops);
 488 EXPORT_SYMBOL    (pv_mmu_ops);
 489 EXPORT_SYMBOL_GPL(pv_apic_ops);
 490 EXPORT_SYMBOL_GPL(pv_info);
 491 EXPORT_SYMBOL    (pv_irq_ops);