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
  45 /*
  46  * nop stub, which must not clobber anything *including the stack* to
  47  * avoid confusing the entry prologues.
  48  */
  49 extern void _paravirt_nop(void);
  50 asm (".pushsection .entry.text, \"ax\"\n"
  51      ".global _paravirt_nop\n"
  52      "_paravirt_nop:\n\t"
  53      "ret\n\t"
  54      ".size _paravirt_nop, . - _paravirt_nop\n\t"
  55      ".type _paravirt_nop, @function\n\t"
  56      ".popsection");
  57
  58 /* identity function, which can be inlined */
  59 u32 notrace _paravirt_ident_32(u32 x)
  60 {
  61         return x;
  62 }
  63
  64 u64 notrace _paravirt_ident_64(u64 x)
  65 {
  66         return x;
  67 }
  68
  69 void __init default_banner(void)
  70 {
  71         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  72                pv_info.name);
  73 }
  74
  75 /* Undefined instruction for dealing with missing ops pointers. */
  76 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  77
  78 struct branch {
  79         unsigned char opcode;
  80         u32 delta;
  81 } __attribute__((packed));
  82
  83 unsigned paravirt_patch_call(void *insnbuf,
  84                              const void *target, u16 tgt_clobbers,
  85                              unsigned long addr, u16 site_clobbers,
  86                              unsigned len)
  87 {
  88         struct branch *b = insnbuf;
  89         unsigned long delta = (unsigned long)target - (addr+5);
  90
  91         if (len < 5) {
  92 #ifdef CONFIG_RETPOLINE
  93                 WARN_ONCE(1, "Failing to patch indirect CALL in %ps\n", (void *)addr);
  94 #endif
  95                 return len;     /* call too long for patch site */
  96         }
  97
  98         b->opcode = 0xe8; /* call */
  99         b->delta = delta;
 100         BUILD_BUG_ON(sizeof(*b) != 5);
 101
 102         return 5;
 103 }
 104
 105 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 106                             unsigned long addr, unsigned len)
 107 {
 108         struct branch *b = insnbuf;
 109         unsigned long delta = (unsigned long)target - (addr+5);
 110
 111         if (len < 5) {
 112 #ifdef CONFIG_RETPOLINE
 113                 WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr);
 114 #endif
 115                 return len;     /* call too long for patch site */
 116         }
 117
 118         b->opcode = 0xe9;       /* jmp */
 119         b->delta = delta;
 120
 121         return 5;
 122 }
 123
 124 /* Neat trick to map patch type back to the call within the
 125  * corresponding structure. */
 126 static void *get_call_destination(u8 type)
 127 {
 128         struct paravirt_patch_template tmpl = {
 129                 .pv_init_ops = pv_init_ops,
 130                 .pv_time_ops = pv_time_ops,
 131                 .pv_cpu_ops = pv_cpu_ops,
 132                 .pv_irq_ops = pv_irq_ops,
 133                 .pv_mmu_ops = pv_mmu_ops,
 134 #ifdef CONFIG_PARAVIRT_SPINLOCKS
 135                 .pv_lock_ops = pv_lock_ops,
 136 #endif
 137         };
 138         return *((void **)&tmpl + type);
 139 }
 140
 141 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 142                                 unsigned long addr, unsigned len)
 143 {
 144         void *opfunc = get_call_destination(type);
 145         unsigned ret;
 146
 147         if (opfunc == NULL)
 148                 /* If there's no function, patch it with a ud2a (BUG) */
 149                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 150         else if (opfunc == _paravirt_nop)
 151                 ret = 0;
 152
 153         /* identity functions just return their single argument */
 154         else if (opfunc == _paravirt_ident_32)
 155                 ret = paravirt_patch_ident_32(insnbuf, len);
 156         else if (opfunc == _paravirt_ident_64)
 157                 ret = paravirt_patch_ident_64(insnbuf, len);
 158
 159         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
 160                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
 161                 /* If operation requires a jmp, then jmp */
 162                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 163         else
 164                 /* Otherwise call the function; assume target could
 165                    clobber any caller-save reg */
 166                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 167                                           addr, clobbers, len);
 168
 169         return ret;
 170 }
 171
 172 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 173                               const char *start, const char *end)
 174 {
 175         unsigned insn_len = end - start;
 176
 177         if (insn_len > len || start == NULL)
 178                 insn_len = len;
 179         else
 180                 memcpy(insnbuf, start, insn_len);
 181
 182         return insn_len;
 183 }
 184
 185 static void native_flush_tlb(void)
 186 {
 187         __native_flush_tlb();
 188 }
 189
 190 /*
 191  * Global pages have to be flushed a bit differently. Not a real
 192  * performance problem because this does not happen often.
 193  */
 194 static void native_flush_tlb_global(void)
 195 {
 196         __native_flush_tlb_global();
 197 }
 198
 199 static void native_flush_tlb_one_user(unsigned long addr)
 200 {
 201         __native_flush_tlb_one_user(addr);
 202 }
 203
 204 struct static_key paravirt_steal_enabled;
 205 struct static_key paravirt_steal_rq_enabled;
 206
 207 static u64 native_steal_clock(int cpu)
 208 {
 209         return 0;
 210 }
 211
 212 /* These are in entry.S */
 213 extern void native_iret(void);
 214 extern void native_usergs_sysret64(void);
 215
 216 static struct resource reserve_ioports = {
 217         .start = 0,
 218         .end = IO_SPACE_LIMIT,
 219         .name = "paravirt-ioport",
 220         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 221 };
 222
 223 /*
 224  * Reserve the whole legacy IO space to prevent any legacy drivers
 225  * from wasting time probing for their hardware.  This is a fairly
 226  * brute-force approach to disabling all non-virtual drivers.
 227  *
 228  * Note that this must be called very early to have any effect.
 229  */
 230 int paravirt_disable_iospace(void)
 231 {
 232         return request_resource(&ioport_resource, &reserve_ioports);
 233 }
 234
 235 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 236
 237 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 238 {
 239         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 240
 241         this_cpu_write(paravirt_lazy_mode, mode);
 242 }
 243
 244 static void leave_lazy(enum paravirt_lazy_mode mode)
 245 {
 246         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 247
 248         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 249 }
 250
 251 void paravirt_enter_lazy_mmu(void)
 252 {
 253         enter_lazy(PARAVIRT_LAZY_MMU);
 254 }
 255
 256 void paravirt_leave_lazy_mmu(void)
 257 {
 258         leave_lazy(PARAVIRT_LAZY_MMU);
 259 }
 260
 261 void paravirt_flush_lazy_mmu(void)
 262 {
 263         preempt_disable();
 264
 265         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 266                 arch_leave_lazy_mmu_mode();
 267                 arch_enter_lazy_mmu_mode();
 268         }
 269
 270         preempt_enable();
 271 }
 272
 273 void paravirt_start_context_switch(struct task_struct *prev)
 274 {
 275         BUG_ON(preemptible());
 276
 277         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 278                 arch_leave_lazy_mmu_mode();
 279                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 280         }
 281         enter_lazy(PARAVIRT_LAZY_CPU);
 282 }
 283
 284 void paravirt_end_context_switch(struct task_struct *next)
 285 {
 286         BUG_ON(preemptible());
 287
 288         leave_lazy(PARAVIRT_LAZY_CPU);
 289
 290         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 291                 arch_enter_lazy_mmu_mode();
 292 }
 293
 294 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 295 {
 296         if (in_interrupt())
 297                 return PARAVIRT_LAZY_NONE;
 298
 299         return this_cpu_read(paravirt_lazy_mode);
 300 }
 301
 302 struct pv_info pv_info = {
 303         .name = "bare hardware",
 304         .kernel_rpl = 0,
 305         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 306
 307 #ifdef CONFIG_X86_64
 308         .extra_user_64bit_cs = __USER_CS,
 309 #endif
 310 };
 311
 312 struct pv_init_ops pv_init_ops = {
 313         .patch = native_patch,
 314 };
 315
 316 struct pv_time_ops pv_time_ops = {
 317         .sched_clock = native_sched_clock,
 318         .steal_clock = native_steal_clock,
 319 };
 320
 321 __visible struct pv_irq_ops pv_irq_ops = {
 322         .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
 323         .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
 324         .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
 325         .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
 326         .safe_halt = native_safe_halt,
 327         .halt = native_halt,
 328 };
 329
 330 __visible struct pv_cpu_ops pv_cpu_ops = {
 331         .cpuid = native_cpuid,
 332         .get_debugreg = native_get_debugreg,
 333         .set_debugreg = native_set_debugreg,
 334         .read_cr0 = native_read_cr0,
 335         .write_cr0 = native_write_cr0,
 336         .write_cr4 = native_write_cr4,
 337 #ifdef CONFIG_X86_64
 338         .read_cr8 = native_read_cr8,
 339         .write_cr8 = native_write_cr8,
 340 #endif
 341         .wbinvd = native_wbinvd,
 342         .read_msr = native_read_msr,
 343         .write_msr = native_write_msr,
 344         .read_msr_safe = native_read_msr_safe,
 345         .write_msr_safe = native_write_msr_safe,
 346         .read_pmc = native_read_pmc,
 347         .load_tr_desc = native_load_tr_desc,
 348         .set_ldt = native_set_ldt,
 349         .load_gdt = native_load_gdt,
 350         .load_idt = native_load_idt,
 351         .store_tr = native_store_tr,
 352         .load_tls = native_load_tls,
 353 #ifdef CONFIG_X86_64
 354         .load_gs_index = native_load_gs_index,
 355 #endif
 356         .write_ldt_entry = native_write_ldt_entry,
 357         .write_gdt_entry = native_write_gdt_entry,
 358         .write_idt_entry = native_write_idt_entry,
 359
 360         .alloc_ldt = paravirt_nop,
 361         .free_ldt = paravirt_nop,
 362
 363         .load_sp0 = native_load_sp0,
 364
 365 #ifdef CONFIG_X86_64
 366         .usergs_sysret64 = native_usergs_sysret64,
 367 #endif
 368         .iret = native_iret,
 369         .swapgs = native_swapgs,
 370
 371         .set_iopl_mask = native_set_iopl_mask,
 372         .io_delay = native_io_delay,
 373
 374         .start_context_switch = paravirt_nop,
 375         .end_context_switch = paravirt_nop,
 376 };
 377
 378 /* At this point, native_get/set_debugreg has real function entries */
 379 NOKPROBE_SYMBOL(native_get_debugreg);
 380 NOKPROBE_SYMBOL(native_set_debugreg);
 381 NOKPROBE_SYMBOL(native_load_idt);
 382
 383 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
 384 /* 32-bit pagetable entries */
 385 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
 386 #else
 387 /* 64-bit pagetable entries */
 388 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 389 #endif
 390
 391 struct pv_mmu_ops pv_mmu_ops __ro_after_init = {
 392
 393         .read_cr2 = native_read_cr2,
 394         .write_cr2 = native_write_cr2,
 395         .read_cr3 = __native_read_cr3,
 396         .write_cr3 = native_write_cr3,
 397
 398         .flush_tlb_user = native_flush_tlb,
 399         .flush_tlb_kernel = native_flush_tlb_global,
 400         .flush_tlb_one_user = native_flush_tlb_one_user,
 401         .flush_tlb_others = native_flush_tlb_others,
 402
 403         .pgd_alloc = __paravirt_pgd_alloc,
 404         .pgd_free = paravirt_nop,
 405
 406         .alloc_pte = paravirt_nop,
 407         .alloc_pmd = paravirt_nop,
 408         .alloc_pud = paravirt_nop,
 409         .alloc_p4d = paravirt_nop,
 410         .release_pte = paravirt_nop,
 411         .release_pmd = paravirt_nop,
 412         .release_pud = paravirt_nop,
 413         .release_p4d = paravirt_nop,
 414
 415         .set_pte = native_set_pte,
 416         .set_pte_at = native_set_pte_at,
 417         .set_pmd = native_set_pmd,
 418
 419         .ptep_modify_prot_start = __ptep_modify_prot_start,
 420         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 421
 422 #if CONFIG_PGTABLE_LEVELS >= 3
 423 #ifdef CONFIG_X86_PAE
 424         .set_pte_atomic = native_set_pte_atomic,
 425         .pte_clear = native_pte_clear,
 426         .pmd_clear = native_pmd_clear,
 427 #endif
 428         .set_pud = native_set_pud,
 429
 430         .pmd_val = PTE_IDENT,
 431         .make_pmd = PTE_IDENT,
 432
 433 #if CONFIG_PGTABLE_LEVELS >= 4
 434         .pud_val = PTE_IDENT,
 435         .make_pud = PTE_IDENT,
 436
 437         .set_p4d = native_set_p4d,
 438
 439 #if CONFIG_PGTABLE_LEVELS >= 5
 440         .p4d_val = PTE_IDENT,
 441         .make_p4d = PTE_IDENT,
 442
 443         .set_pgd = native_set_pgd,
 444 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
 445 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */
 446 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
 447
 448         .pte_val = PTE_IDENT,
 449         .pgd_val = PTE_IDENT,
 450
 451         .make_pte = PTE_IDENT,
 452         .make_pgd = PTE_IDENT,
 453
 454         .dup_mmap = paravirt_nop,
 455         .exit_mmap = paravirt_nop,
 456         .activate_mm = paravirt_nop,
 457
 458         .lazy_mode = {
 459                 .enter = paravirt_nop,
 460                 .leave = paravirt_nop,
 461                 .flush = paravirt_nop,
 462         },
 463
 464         .set_fixmap = native_set_fixmap,
 465 };
 466
 467 EXPORT_SYMBOL_GPL(pv_time_ops);
 468 EXPORT_SYMBOL    (pv_cpu_ops);
 469 EXPORT_SYMBOL    (pv_mmu_ops);
 470 EXPORT_SYMBOL_GPL(pv_info);
 471 EXPORT_SYMBOL    (pv_irq_ops);