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/init.h>
  23 #include <linux/export.h>
  24 #include <linux/efi.h>
  25 #include <linux/bcd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/kprobes.h>
  28
  29 #include <asm/bug.h>
  30 #include <asm/paravirt.h>
  31 #include <asm/debugreg.h>
  32 #include <asm/desc.h>
  33 #include <asm/setup.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/time.h>
  36 #include <asm/pgalloc.h>
  37 #include <asm/irq.h>
  38 #include <asm/delay.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/apic.h>
  41 #include <asm/tlbflush.h>
  42 #include <asm/timer.h>
  43 #include <asm/special_insns.h>
  44 #include <asm/tlb.h>
  45
  46 /*
  47  * nop stub, which must not clobber anything *including the stack* to
  48  * avoid confusing the entry prologues.
  49  */
  50 extern void _paravirt_nop(void);
  51 asm (".pushsection .entry.text, \"ax\"\n"
  52      ".global _paravirt_nop\n"
  53      "_paravirt_nop:\n\t"
  54      "ret\n\t"
  55      ".size _paravirt_nop, . - _paravirt_nop\n\t"
  56      ".type _paravirt_nop, @function\n\t"
  57      ".popsection");
  58
  59 /* identity function, which can be inlined */
  60 u32 notrace _paravirt_ident_32(u32 x)
  61 {
  62         return x;
  63 }
  64
  65 u64 notrace _paravirt_ident_64(u64 x)
  66 {
  67         return x;
  68 }
  69
  70 void __init default_banner(void)
  71 {
  72         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  73                pv_info.name);
  74 }
  75
  76 /* Undefined instruction for dealing with missing ops pointers. */
  77 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  78
  79 struct branch {
  80         unsigned char opcode;
  81         u32 delta;
  82 } __attribute__((packed));
  83
  84 static unsigned paravirt_patch_call(void *insnbuf, const void *target,
  85                                     unsigned long addr, unsigned len)
  86 {
  87         struct branch *b = insnbuf;
  88         unsigned long delta = (unsigned long)target - (addr+5);
  89
  90         if (len < 5) {
  91 #ifdef CONFIG_RETPOLINE
  92                 WARN_ONCE("Failing to patch indirect CALL in %ps\n", (void *)addr);
  93 #endif
  94                 return len;     /* call too long for patch site */
  95         }
  96
  97         b->opcode = 0xe8; /* call */
  98         b->delta = delta;
  99         BUILD_BUG_ON(sizeof(*b) != 5);
 100
 101         return 5;
 102 }
 103
 104 static unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 105                                    unsigned long addr, unsigned len)
 106 {
 107         struct branch *b = insnbuf;
 108         unsigned long delta = (unsigned long)target - (addr+5);
 109
 110         if (len < 5) {
 111 #ifdef CONFIG_RETPOLINE
 112                 WARN_ONCE("Failing to patch indirect JMP in %ps\n", (void *)addr);
 113 #endif
 114                 return len;     /* call too long for patch site */
 115         }
 116
 117         b->opcode = 0xe9;       /* jmp */
 118         b->delta = delta;
 119
 120         return 5;
 121 }
 122
 123 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
 124
 125 void __init native_pv_lock_init(void)
 126 {
 127         if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
 128                 static_branch_disable(&virt_spin_lock_key);
 129 }
 130
 131 unsigned paravirt_patch_default(u8 type, void *insnbuf,
 132                                 unsigned long addr, unsigned len)
 133 {
 134         /*
 135          * Neat trick to map patch type back to the call within the
 136          * corresponding structure.
 137          */
 138         void *opfunc = *((void **)&pv_ops + type);
 139         unsigned ret;
 140
 141         if (opfunc == NULL)
 142                 /* If there's no function, patch it with a ud2a (BUG) */
 143                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 144         else if (opfunc == _paravirt_nop)
 145                 ret = 0;
 146
 147         /* identity functions just return their single argument */
 148         else if (opfunc == _paravirt_ident_32)
 149                 ret = paravirt_patch_ident_32(insnbuf, len);
 150         else if (opfunc == _paravirt_ident_64)
 151                 ret = paravirt_patch_ident_64(insnbuf, len);
 152
 153         else if (type == PARAVIRT_PATCH(cpu.iret) ||
 154                  type == PARAVIRT_PATCH(cpu.usergs_sysret64))
 155                 /* If operation requires a jmp, then jmp */
 156                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 157         else
 158                 /* Otherwise call the function. */
 159                 ret = paravirt_patch_call(insnbuf, opfunc, addr, len);
 160
 161         return ret;
 162 }
 163
 164 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 165                               const char *start, const char *end)
 166 {
 167         unsigned insn_len = end - start;
 168
 169         if (insn_len > len || start == NULL)
 170                 insn_len = len;
 171         else
 172                 memcpy(insnbuf, start, insn_len);
 173
 174         return insn_len;
 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_one_user(unsigned long addr)
 192 {
 193         __native_flush_tlb_one_user(addr);
 194 }
 195
 196 struct static_key paravirt_steal_enabled;
 197 struct static_key paravirt_steal_rq_enabled;
 198
 199 static u64 native_steal_clock(int cpu)
 200 {
 201         return 0;
 202 }
 203
 204 /* These are in entry.S */
 205 extern void native_iret(void);
 206 extern void native_usergs_sysret64(void);
 207
 208 static struct resource reserve_ioports = {
 209         .start = 0,
 210         .end = IO_SPACE_LIMIT,
 211         .name = "paravirt-ioport",
 212         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 213 };
 214
 215 /*
 216  * Reserve the whole legacy IO space to prevent any legacy drivers
 217  * from wasting time probing for their hardware.  This is a fairly
 218  * brute-force approach to disabling all non-virtual drivers.
 219  *
 220  * Note that this must be called very early to have any effect.
 221  */
 222 int paravirt_disable_iospace(void)
 223 {
 224         return request_resource(&ioport_resource, &reserve_ioports);
 225 }
 226
 227 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 228
 229 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 230 {
 231         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 232
 233         this_cpu_write(paravirt_lazy_mode, mode);
 234 }
 235
 236 static void leave_lazy(enum paravirt_lazy_mode mode)
 237 {
 238         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 239
 240         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 241 }
 242
 243 void paravirt_enter_lazy_mmu(void)
 244 {
 245         enter_lazy(PARAVIRT_LAZY_MMU);
 246 }
 247
 248 void paravirt_leave_lazy_mmu(void)
 249 {
 250         leave_lazy(PARAVIRT_LAZY_MMU);
 251 }
 252
 253 void paravirt_flush_lazy_mmu(void)
 254 {
 255         preempt_disable();
 256
 257         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 258                 arch_leave_lazy_mmu_mode();
 259                 arch_enter_lazy_mmu_mode();
 260         }
 261
 262         preempt_enable();
 263 }
 264
 265 void paravirt_start_context_switch(struct task_struct *prev)
 266 {
 267         BUG_ON(preemptible());
 268
 269         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 270                 arch_leave_lazy_mmu_mode();
 271                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 272         }
 273         enter_lazy(PARAVIRT_LAZY_CPU);
 274 }
 275
 276 void paravirt_end_context_switch(struct task_struct *next)
 277 {
 278         BUG_ON(preemptible());
 279
 280         leave_lazy(PARAVIRT_LAZY_CPU);
 281
 282         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 283                 arch_enter_lazy_mmu_mode();
 284 }
 285
 286 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 287 {
 288         if (in_interrupt())
 289                 return PARAVIRT_LAZY_NONE;
 290
 291         return this_cpu_read(paravirt_lazy_mode);
 292 }
 293
 294 struct pv_info pv_info = {
 295         .name = "bare hardware",
 296 #ifdef CONFIG_PARAVIRT_XXL
 297         .kernel_rpl = 0,
 298         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 299
 300 #ifdef CONFIG_X86_64
 301         .extra_user_64bit_cs = __USER_CS,
 302 #endif
 303 #endif
 304 };
 305
 306 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
 307 /* 32-bit pagetable entries */
 308 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
 309 #else
 310 /* 64-bit pagetable entries */
 311 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 312 #endif
 313
 314 struct paravirt_patch_template pv_ops = {
 315         /* Init ops. */
 316         .init.patch             = native_patch,
 317
 318         /* Time ops. */
 319         .time.sched_clock       = native_sched_clock,
 320         .time.steal_clock       = native_steal_clock,
 321
 322         /* Cpu ops. */
 323         .cpu.cpuid              = native_cpuid,
 324         .cpu.get_debugreg       = native_get_debugreg,
 325         .cpu.set_debugreg       = native_set_debugreg,
 326         .cpu.read_cr0           = native_read_cr0,
 327         .cpu.write_cr0          = native_write_cr0,
 328         .cpu.write_cr4          = native_write_cr4,
 329 #ifdef CONFIG_X86_64
 330         .cpu.read_cr8           = native_read_cr8,
 331         .cpu.write_cr8          = native_write_cr8,
 332 #endif
 333         .cpu.wbinvd             = native_wbinvd,
 334         .cpu.read_msr           = native_read_msr,
 335         .cpu.write_msr          = native_write_msr,
 336         .cpu.read_msr_safe      = native_read_msr_safe,
 337         .cpu.write_msr_safe     = native_write_msr_safe,
 338         .cpu.read_pmc           = native_read_pmc,
 339         .cpu.load_tr_desc       = native_load_tr_desc,
 340         .cpu.set_ldt            = native_set_ldt,
 341         .cpu.load_gdt           = native_load_gdt,
 342         .cpu.load_idt           = native_load_idt,
 343         .cpu.store_tr           = native_store_tr,
 344         .cpu.load_tls           = native_load_tls,
 345 #ifdef CONFIG_X86_64
 346         .cpu.load_gs_index      = native_load_gs_index,
 347 #endif
 348         .cpu.write_ldt_entry    = native_write_ldt_entry,
 349         .cpu.write_gdt_entry    = native_write_gdt_entry,
 350         .cpu.write_idt_entry    = native_write_idt_entry,
 351
 352         .cpu.alloc_ldt          = paravirt_nop,
 353         .cpu.free_ldt           = paravirt_nop,
 354
 355         .cpu.load_sp0           = native_load_sp0,
 356
 357 #ifdef CONFIG_X86_64
 358         .cpu.usergs_sysret64    = native_usergs_sysret64,
 359 #endif
 360         .cpu.iret               = native_iret,
 361         .cpu.swapgs             = native_swapgs,
 362
 363         .cpu.set_iopl_mask      = native_set_iopl_mask,
 364         .cpu.io_delay           = native_io_delay,
 365
 366         .cpu.start_context_switch       = paravirt_nop,
 367         .cpu.end_context_switch         = paravirt_nop,
 368
 369         /* Irq ops. */
 370         .irq.save_fl            = __PV_IS_CALLEE_SAVE(native_save_fl),
 371         .irq.restore_fl         = __PV_IS_CALLEE_SAVE(native_restore_fl),
 372         .irq.irq_disable        = __PV_IS_CALLEE_SAVE(native_irq_disable),
 373         .irq.irq_enable         = __PV_IS_CALLEE_SAVE(native_irq_enable),
 374         .irq.safe_halt          = native_safe_halt,
 375         .irq.halt               = native_halt,
 376
 377         /* Mmu ops. */
 378         .mmu.read_cr2           = native_read_cr2,
 379         .mmu.write_cr2          = native_write_cr2,
 380         .mmu.read_cr3           = __native_read_cr3,
 381         .mmu.write_cr3          = native_write_cr3,
 382
 383         .mmu.flush_tlb_user     = native_flush_tlb,
 384         .mmu.flush_tlb_kernel   = native_flush_tlb_global,
 385         .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
 386         .mmu.flush_tlb_others   = native_flush_tlb_others,
 387         .mmu.tlb_remove_table   =
 388                         (void (*)(struct mmu_gather *, void *))tlb_remove_page,
 389
 390         .mmu.pgd_alloc          = __paravirt_pgd_alloc,
 391         .mmu.pgd_free           = paravirt_nop,
 392
 393         .mmu.alloc_pte          = paravirt_nop,
 394         .mmu.alloc_pmd          = paravirt_nop,
 395         .mmu.alloc_pud          = paravirt_nop,
 396         .mmu.alloc_p4d          = paravirt_nop,
 397         .mmu.release_pte        = paravirt_nop,
 398         .mmu.release_pmd        = paravirt_nop,
 399         .mmu.release_pud        = paravirt_nop,
 400         .mmu.release_p4d        = paravirt_nop,
 401
 402         .mmu.set_pte            = native_set_pte,
 403         .mmu.set_pte_at         = native_set_pte_at,
 404         .mmu.set_pmd            = native_set_pmd,
 405
 406         .mmu.ptep_modify_prot_start     = __ptep_modify_prot_start,
 407         .mmu.ptep_modify_prot_commit    = __ptep_modify_prot_commit,
 408
 409 #if CONFIG_PGTABLE_LEVELS >= 3
 410 #ifdef CONFIG_X86_PAE
 411         .mmu.set_pte_atomic     = native_set_pte_atomic,
 412         .mmu.pte_clear          = native_pte_clear,
 413         .mmu.pmd_clear          = native_pmd_clear,
 414 #endif
 415         .mmu.set_pud            = native_set_pud,
 416
 417         .mmu.pmd_val            = PTE_IDENT,
 418         .mmu.make_pmd           = PTE_IDENT,
 419
 420 #if CONFIG_PGTABLE_LEVELS >= 4
 421         .mmu.pud_val            = PTE_IDENT,
 422         .mmu.make_pud           = PTE_IDENT,
 423
 424         .mmu.set_p4d            = native_set_p4d,
 425
 426 #if CONFIG_PGTABLE_LEVELS >= 5
 427         .mmu.p4d_val            = PTE_IDENT,
 428         .mmu.make_p4d           = PTE_IDENT,
 429
 430         .mmu.set_pgd            = native_set_pgd,
 431 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
 432 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */
 433 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
 434
 435         .mmu.pte_val            = PTE_IDENT,
 436         .mmu.pgd_val            = PTE_IDENT,
 437
 438         .mmu.make_pte           = PTE_IDENT,
 439         .mmu.make_pgd           = PTE_IDENT,
 440
 441         .mmu.dup_mmap           = paravirt_nop,
 442         .mmu.exit_mmap          = paravirt_nop,
 443         .mmu.activate_mm        = paravirt_nop,
 444
 445         .mmu.lazy_mode = {
 446                 .enter          = paravirt_nop,
 447                 .leave          = paravirt_nop,
 448                 .flush          = paravirt_nop,
 449         },
 450
 451         .mmu.set_fixmap         = native_set_fixmap,
 452
 453 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
 454         /* Lock ops. */
 455 #ifdef CONFIG_SMP
 456         .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
 457         .lock.queued_spin_unlock        =
 458                                 PV_CALLEE_SAVE(__native_queued_spin_unlock),
 459         .lock.wait                      = paravirt_nop,
 460         .lock.kick                      = paravirt_nop,
 461         .lock.vcpu_is_preempted         =
 462                                 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
 463 #endif /* SMP */
 464 #endif
 465 };
 466
 467 /* At this point, native_get/set_debugreg has real function entries */
 468 NOKPROBE_SYMBOL(native_get_debugreg);
 469 NOKPROBE_SYMBOL(native_set_debugreg);
 470 NOKPROBE_SYMBOL(native_load_idt);
 471
 472 EXPORT_SYMBOL_GPL(pv_ops);
 473 EXPORT_SYMBOL_GPL(pv_info);