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 (tgt_clobbers & ~site_clobbers)
  92                 return len;     /* target would clobber too much for this site */
  93         if (len < 5)
  94                 return len;     /* call too long for patch site */
  95
  96         b->opcode = 0xe8; /* call */
  97         b->delta = delta;
  98         BUILD_BUG_ON(sizeof(*b) != 5);
  99
 100         return 5;
 101 }
 102
 103 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 104                             unsigned long addr, unsigned len)
 105 {
 106         struct branch *b = insnbuf;
 107         unsigned long delta = (unsigned long)target - (addr+5);
 108
 109         if (len < 5)
 110                 return len;     /* call too long for patch site */
 111
 112         b->opcode = 0xe9;       /* jmp */
 113         b->delta = delta;
 114
 115         return 5;
 116 }
 117
 118 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
 119
 120 void __init native_pv_lock_init(void)
 121 {
 122         if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
 123                 static_branch_disable(&virt_spin_lock_key);
 124 }
 125
 126 /*
 127  * Neat trick to map patch type back to the call within the
 128  * corresponding structure.
 129  */
 130 static void *get_call_destination(u8 type)
 131 {
 132         struct paravirt_patch_template tmpl = {
 133                 .pv_init_ops = pv_init_ops,
 134                 .pv_time_ops = pv_time_ops,
 135                 .pv_cpu_ops = pv_cpu_ops,
 136                 .pv_irq_ops = pv_irq_ops,
 137                 .pv_mmu_ops = pv_mmu_ops,
 138 #ifdef CONFIG_PARAVIRT_SPINLOCKS
 139                 .pv_lock_ops = pv_lock_ops,
 140 #endif
 141         };
 142         return *((void **)&tmpl + type);
 143 }
 144
 145 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 146                                 unsigned long addr, unsigned len)
 147 {
 148         void *opfunc = get_call_destination(type);
 149         unsigned ret;
 150
 151         if (opfunc == NULL)
 152                 /* If there's no function, patch it with a ud2a (BUG) */
 153                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 154         else if (opfunc == _paravirt_nop)
 155                 ret = 0;
 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.usergs_sysret64))
 165                 /* If operation requires a jmp, then jmp */
 166                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 167         else
 168                 /* Otherwise call the function; assume target could
 169                    clobber any caller-save reg */
 170                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 171                                           addr, clobbers, len);
 172
 173         return ret;
 174 }
 175
 176 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 177                               const char *start, const char *end)
 178 {
 179         unsigned insn_len = end - start;
 180
 181         if (insn_len > len || start == NULL)
 182                 insn_len = len;
 183         else
 184                 memcpy(insnbuf, start, insn_len);
 185
 186         return insn_len;
 187 }
 188
 189 static void native_flush_tlb(void)
 190 {
 191         __native_flush_tlb();
 192 }
 193
 194 /*
 195  * Global pages have to be flushed a bit differently. Not a real
 196  * performance problem because this does not happen often.
 197  */
 198 static void native_flush_tlb_global(void)
 199 {
 200         __native_flush_tlb_global();
 201 }
 202
 203 static void native_flush_tlb_single(unsigned long addr)
 204 {
 205         __native_flush_tlb_single(addr);
 206 }
 207
 208 struct static_key paravirt_steal_enabled;
 209 struct static_key paravirt_steal_rq_enabled;
 210
 211 static u64 native_steal_clock(int cpu)
 212 {
 213         return 0;
 214 }
 215
 216 /* These are in entry.S */
 217 extern void native_iret(void);
 218 extern void native_usergs_sysret64(void);
 219
 220 static struct resource reserve_ioports = {
 221         .start = 0,
 222         .end = IO_SPACE_LIMIT,
 223         .name = "paravirt-ioport",
 224         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 225 };
 226
 227 /*
 228  * Reserve the whole legacy IO space to prevent any legacy drivers
 229  * from wasting time probing for their hardware.  This is a fairly
 230  * brute-force approach to disabling all non-virtual drivers.
 231  *
 232  * Note that this must be called very early to have any effect.
 233  */
 234 int paravirt_disable_iospace(void)
 235 {
 236         return request_resource(&ioport_resource, &reserve_ioports);
 237 }
 238
 239 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 240
 241 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 242 {
 243         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 244
 245         this_cpu_write(paravirt_lazy_mode, mode);
 246 }
 247
 248 static void leave_lazy(enum paravirt_lazy_mode mode)
 249 {
 250         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 251
 252         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 253 }
 254
 255 void paravirt_enter_lazy_mmu(void)
 256 {
 257         enter_lazy(PARAVIRT_LAZY_MMU);
 258 }
 259
 260 void paravirt_leave_lazy_mmu(void)
 261 {
 262         leave_lazy(PARAVIRT_LAZY_MMU);
 263 }
 264
 265 void paravirt_flush_lazy_mmu(void)
 266 {
 267         preempt_disable();
 268
 269         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 270                 arch_leave_lazy_mmu_mode();
 271                 arch_enter_lazy_mmu_mode();
 272         }
 273
 274         preempt_enable();
 275 }
 276
 277 void paravirt_start_context_switch(struct task_struct *prev)
 278 {
 279         BUG_ON(preemptible());
 280
 281         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 282                 arch_leave_lazy_mmu_mode();
 283                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 284         }
 285         enter_lazy(PARAVIRT_LAZY_CPU);
 286 }
 287
 288 void paravirt_end_context_switch(struct task_struct *next)
 289 {
 290         BUG_ON(preemptible());
 291
 292         leave_lazy(PARAVIRT_LAZY_CPU);
 293
 294         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 295                 arch_enter_lazy_mmu_mode();
 296 }
 297
 298 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 299 {
 300         if (in_interrupt())
 301                 return PARAVIRT_LAZY_NONE;
 302
 303         return this_cpu_read(paravirt_lazy_mode);
 304 }
 305
 306 struct pv_info pv_info = {
 307         .name = "bare hardware",
 308         .kernel_rpl = 0,
 309         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 310
 311 #ifdef CONFIG_X86_64
 312         .extra_user_64bit_cs = __USER_CS,
 313 #endif
 314 };
 315
 316 struct pv_init_ops pv_init_ops = {
 317         .patch = native_patch,
 318 };
 319
 320 struct pv_time_ops pv_time_ops = {
 321         .sched_clock = native_sched_clock,
 322         .steal_clock = native_steal_clock,
 323 };
 324
 325 __visible struct pv_irq_ops pv_irq_ops = {
 326         .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
 327         .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
 328         .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
 329         .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
 330         .safe_halt = native_safe_halt,
 331         .halt = native_halt,
 332 };
 333
 334 __visible struct pv_cpu_ops pv_cpu_ops = {
 335         .cpuid = native_cpuid,
 336         .get_debugreg = native_get_debugreg,
 337         .set_debugreg = native_set_debugreg,
 338         .read_cr0 = native_read_cr0,
 339         .write_cr0 = native_write_cr0,
 340         .write_cr4 = native_write_cr4,
 341 #ifdef CONFIG_X86_64
 342         .read_cr8 = native_read_cr8,
 343         .write_cr8 = native_write_cr8,
 344 #endif
 345         .wbinvd = native_wbinvd,
 346         .read_msr = native_read_msr,
 347         .write_msr = native_write_msr,
 348         .read_msr_safe = native_read_msr_safe,
 349         .write_msr_safe = native_write_msr_safe,
 350         .read_pmc = native_read_pmc,
 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_tr = native_store_tr,
 356         .load_tls = native_load_tls,
 357 #ifdef CONFIG_X86_64
 358         .load_gs_index = native_load_gs_index,
 359 #endif
 360         .write_ldt_entry = native_write_ldt_entry,
 361         .write_gdt_entry = native_write_gdt_entry,
 362         .write_idt_entry = native_write_idt_entry,
 363
 364         .alloc_ldt = paravirt_nop,
 365         .free_ldt = paravirt_nop,
 366
 367         .load_sp0 = native_load_sp0,
 368
 369 #ifdef CONFIG_X86_64
 370         .usergs_sysret64 = native_usergs_sysret64,
 371 #endif
 372         .iret = native_iret,
 373         .swapgs = native_swapgs,
 374
 375         .set_iopl_mask = native_set_iopl_mask,
 376         .io_delay = native_io_delay,
 377
 378         .start_context_switch = paravirt_nop,
 379         .end_context_switch = paravirt_nop,
 380 };
 381
 382 /* At this point, native_get/set_debugreg has real function entries */
 383 NOKPROBE_SYMBOL(native_get_debugreg);
 384 NOKPROBE_SYMBOL(native_set_debugreg);
 385 NOKPROBE_SYMBOL(native_load_idt);
 386
 387 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
 388 /* 32-bit pagetable entries */
 389 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
 390 #else
 391 /* 64-bit pagetable entries */
 392 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 393 #endif
 394
 395 struct pv_mmu_ops pv_mmu_ops __ro_after_init = {
 396
 397         .read_cr2 = native_read_cr2,
 398         .write_cr2 = native_write_cr2,
 399         .read_cr3 = __native_read_cr3,
 400         .write_cr3 = native_write_cr3,
 401
 402         .flush_tlb_user = native_flush_tlb,
 403         .flush_tlb_kernel = native_flush_tlb_global,
 404         .flush_tlb_single = native_flush_tlb_single,
 405         .flush_tlb_others = native_flush_tlb_others,
 406
 407         .pgd_alloc = __paravirt_pgd_alloc,
 408         .pgd_free = paravirt_nop,
 409
 410         .alloc_pte = paravirt_nop,
 411         .alloc_pmd = paravirt_nop,
 412         .alloc_pud = paravirt_nop,
 413         .alloc_p4d = paravirt_nop,
 414         .release_pte = paravirt_nop,
 415         .release_pmd = paravirt_nop,
 416         .release_pud = paravirt_nop,
 417         .release_p4d = paravirt_nop,
 418
 419         .set_pte = native_set_pte,
 420         .set_pte_at = native_set_pte_at,
 421         .set_pmd = native_set_pmd,
 422
 423         .ptep_modify_prot_start = __ptep_modify_prot_start,
 424         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 425
 426 #if CONFIG_PGTABLE_LEVELS >= 3
 427 #ifdef CONFIG_X86_PAE
 428         .set_pte_atomic = native_set_pte_atomic,
 429         .pte_clear = native_pte_clear,
 430         .pmd_clear = native_pmd_clear,
 431 #endif
 432         .set_pud = native_set_pud,
 433
 434         .pmd_val = PTE_IDENT,
 435         .make_pmd = PTE_IDENT,
 436
 437 #if CONFIG_PGTABLE_LEVELS >= 4
 438         .pud_val = PTE_IDENT,
 439         .make_pud = PTE_IDENT,
 440
 441         .set_p4d = native_set_p4d,
 442
 443 #if CONFIG_PGTABLE_LEVELS >= 5
 444         .p4d_val = PTE_IDENT,
 445         .make_p4d = PTE_IDENT,
 446
 447         .set_pgd = native_set_pgd,
 448 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
 449 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */
 450 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
 451
 452         .pte_val = PTE_IDENT,
 453         .pgd_val = PTE_IDENT,
 454
 455         .make_pte = PTE_IDENT,
 456         .make_pgd = PTE_IDENT,
 457
 458         .dup_mmap = paravirt_nop,
 459         .exit_mmap = paravirt_nop,
 460         .activate_mm = paravirt_nop,
 461
 462         .lazy_mode = {
 463                 .enter = paravirt_nop,
 464                 .leave = paravirt_nop,
 465                 .flush = paravirt_nop,
 466         },
 467
 468         .set_fixmap = native_set_fixmap,
 469 };
 470
 471 EXPORT_SYMBOL_GPL(pv_time_ops);
 472 EXPORT_SYMBOL    (pv_cpu_ops);
 473 EXPORT_SYMBOL    (pv_mmu_ops);
 474 EXPORT_SYMBOL_GPL(pv_info);
 475 EXPORT_SYMBOL    (pv_irq_ops);