arch/x86/kernel/paravirt.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*  Paravirtualization interfaces
   3     Copyright (C) 2006 Rusty Russell IBM Corporation
   4
   5
   6     2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
   7 */
   8
   9 #include <linux/errno.h>
  10 #include <linux/init.h>
  11 #include <linux/export.h>
  12 #include <linux/efi.h>
  13 #include <linux/bcd.h>
  14 #include <linux/highmem.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/pgtable.h>
  17
  18 #include <asm/bug.h>
  19 #include <asm/paravirt.h>
  20 #include <asm/debugreg.h>
  21 #include <asm/desc.h>
  22 #include <asm/setup.h>
  23 #include <asm/time.h>
  24 #include <asm/pgalloc.h>
  25 #include <asm/irq.h>
  26 #include <asm/delay.h>
  27 #include <asm/fixmap.h>
  28 #include <asm/apic.h>
  29 #include <asm/tlbflush.h>
  30 #include <asm/timer.h>
  31 #include <asm/special_insns.h>
  32 #include <asm/tlb.h>
  33 #include <asm/io_bitmap.h>
  34 #include <asm/text-patching.h>
  35
  36 /*
  37  * nop stub, which must not clobber anything *including the stack* to
  38  * avoid confusing the entry prologues.
  39  */
  40 extern void _paravirt_nop(void);
  41 asm (".pushsection .entry.text, \"ax\"\n"
  42      ".global _paravirt_nop\n"
  43      "_paravirt_nop:\n\t"
  44      ASM_RET
  45      ".size _paravirt_nop, . - _paravirt_nop\n\t"
  46      ".type _paravirt_nop, @function\n\t"
  47      ".popsection");
  48
  49 void __init default_banner(void)
  50 {
  51         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  52                pv_info.name);
  53 }
  54
  55 /* Undefined instruction for dealing with missing ops pointers. */
  56 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  57
  58 struct branch {
  59        unsigned char opcode;
  60        u32 delta;
  61 } __attribute__((packed));
  62
  63 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
  64                                     unsigned long addr, unsigned len)
  65 {
  66         __text_gen_insn(insn_buff, CALL_INSN_OPCODE,
  67                         (void *)addr, target, CALL_INSN_SIZE);
  68         return CALL_INSN_SIZE;
  69 }
  70
  71 #ifdef CONFIG_PARAVIRT_XXL
  72 /* identity function, which can be inlined */
  73 u64 notrace _paravirt_ident_64(u64 x)
  74 {
  75         return x;
  76 }
  77
  78 static unsigned paravirt_patch_jmp(void *insn_buff, const void *target,
  79                                    unsigned long addr, unsigned len)
  80 {
  81         struct branch *b = insn_buff;
  82         unsigned long delta = (unsigned long)target - (addr+5);
  83
  84         if (len < 5) {
  85 #ifdef CONFIG_RETPOLINE
  86                 WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr);
  87 #endif
  88                 return len;     /* call too long for patch site */
  89         }
  90
  91         b->opcode = 0xe9;       /* jmp */
  92         b->delta = delta;
  93
  94         return 5;
  95 }
  96 #endif
  97
  98 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
  99
 100 void __init native_pv_lock_init(void)
 101 {
 102         if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
 103                 static_branch_disable(&virt_spin_lock_key);
 104 }
 105
 106 unsigned paravirt_patch_default(u8 type, void *insn_buff,
 107                                 unsigned long addr, unsigned len)
 108 {
 109         /*
 110          * Neat trick to map patch type back to the call within the
 111          * corresponding structure.
 112          */
 113         void *opfunc = *((void **)&pv_ops + type);
 114         unsigned ret;
 115
 116         if (opfunc == NULL)
 117                 /* If there's no function, patch it with a ud2a (BUG) */
 118                 ret = paravirt_patch_insns(insn_buff, len, ud2a, ud2a+sizeof(ud2a));
 119         else if (opfunc == _paravirt_nop)
 120                 ret = 0;
 121
 122 #ifdef CONFIG_PARAVIRT_XXL
 123         /* identity functions just return their single argument */
 124         else if (opfunc == _paravirt_ident_64)
 125                 ret = paravirt_patch_ident_64(insn_buff, len);
 126
 127         else if (type == PARAVIRT_PATCH(cpu.iret) ||
 128                  type == PARAVIRT_PATCH(cpu.usergs_sysret64))
 129                 /* If operation requires a jmp, then jmp */
 130                 ret = paravirt_patch_jmp(insn_buff, opfunc, addr, len);
 131 #endif
 132         else
 133                 /* Otherwise call the function. */
 134                 ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
 135
 136         return ret;
 137 }
 138
 139 unsigned paravirt_patch_insns(void *insn_buff, unsigned len,
 140                               const char *start, const char *end)
 141 {
 142         unsigned insn_len = end - start;
 143
 144         /* Alternative instruction is too large for the patch site and we cannot continue: */
 145         BUG_ON(insn_len > len || start == NULL);
 146
 147         memcpy(insn_buff, start, insn_len);
 148
 149         return insn_len;
 150 }
 151
 152 struct static_key paravirt_steal_enabled;
 153 struct static_key paravirt_steal_rq_enabled;
 154
 155 static u64 native_steal_clock(int cpu)
 156 {
 157         return 0;
 158 }
 159
 160 /* These are in entry.S */
 161 extern void native_iret(void);
 162 extern void native_usergs_sysret64(void);
 163
 164 static struct resource reserve_ioports = {
 165         .start = 0,
 166         .end = IO_SPACE_LIMIT,
 167         .name = "paravirt-ioport",
 168         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 169 };
 170
 171 /*
 172  * Reserve the whole legacy IO space to prevent any legacy drivers
 173  * from wasting time probing for their hardware.  This is a fairly
 174  * brute-force approach to disabling all non-virtual drivers.
 175  *
 176  * Note that this must be called very early to have any effect.
 177  */
 178 int paravirt_disable_iospace(void)
 179 {
 180         return request_resource(&ioport_resource, &reserve_ioports);
 181 }
 182
 183 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 184
 185 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 186 {
 187         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 188
 189         this_cpu_write(paravirt_lazy_mode, mode);
 190 }
 191
 192 static void leave_lazy(enum paravirt_lazy_mode mode)
 193 {
 194         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 195
 196         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 197 }
 198
 199 void paravirt_enter_lazy_mmu(void)
 200 {
 201         enter_lazy(PARAVIRT_LAZY_MMU);
 202 }
 203
 204 void paravirt_leave_lazy_mmu(void)
 205 {
 206         leave_lazy(PARAVIRT_LAZY_MMU);
 207 }
 208
 209 void paravirt_flush_lazy_mmu(void)
 210 {
 211         preempt_disable();
 212
 213         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 214                 arch_leave_lazy_mmu_mode();
 215                 arch_enter_lazy_mmu_mode();
 216         }
 217
 218         preempt_enable();
 219 }
 220
 221 #ifdef CONFIG_PARAVIRT_XXL
 222 void paravirt_start_context_switch(struct task_struct *prev)
 223 {
 224         BUG_ON(preemptible());
 225
 226         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 227                 arch_leave_lazy_mmu_mode();
 228                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 229         }
 230         enter_lazy(PARAVIRT_LAZY_CPU);
 231 }
 232
 233 void paravirt_end_context_switch(struct task_struct *next)
 234 {
 235         BUG_ON(preemptible());
 236
 237         leave_lazy(PARAVIRT_LAZY_CPU);
 238
 239         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 240                 arch_enter_lazy_mmu_mode();
 241 }
 242 #endif
 243
 244 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 245 {
 246         if (in_interrupt())
 247                 return PARAVIRT_LAZY_NONE;
 248
 249         return this_cpu_read(paravirt_lazy_mode);
 250 }
 251
 252 struct pv_info pv_info = {
 253         .name = "bare hardware",
 254 #ifdef CONFIG_PARAVIRT_XXL
 255         .extra_user_64bit_cs = __USER_CS,
 256 #endif
 257 };
 258
 259 /* 64-bit pagetable entries */
 260 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 261
 262 struct paravirt_patch_template pv_ops = {
 263         /* Init ops. */
 264         .init.patch             = native_patch,
 265
 266         /* Time ops. */
 267         .time.sched_clock       = native_sched_clock,
 268         .time.steal_clock       = native_steal_clock,
 269
 270         /* Cpu ops. */
 271         .cpu.io_delay           = native_io_delay,
 272
 273 #ifdef CONFIG_PARAVIRT_XXL
 274         .cpu.cpuid              = native_cpuid,
 275         .cpu.get_debugreg       = native_get_debugreg,
 276         .cpu.set_debugreg       = native_set_debugreg,
 277         .cpu.read_cr0           = native_read_cr0,
 278         .cpu.write_cr0          = native_write_cr0,
 279         .cpu.write_cr4          = native_write_cr4,
 280         .cpu.wbinvd             = native_wbinvd,
 281         .cpu.read_msr           = native_read_msr,
 282         .cpu.write_msr          = native_write_msr,
 283         .cpu.read_msr_safe      = native_read_msr_safe,
 284         .cpu.write_msr_safe     = native_write_msr_safe,
 285         .cpu.read_pmc           = native_read_pmc,
 286         .cpu.load_tr_desc       = native_load_tr_desc,
 287         .cpu.set_ldt            = native_set_ldt,
 288         .cpu.load_gdt           = native_load_gdt,
 289         .cpu.load_idt           = native_load_idt,
 290         .cpu.store_tr           = native_store_tr,
 291         .cpu.load_tls           = native_load_tls,
 292         .cpu.load_gs_index      = native_load_gs_index,
 293         .cpu.write_ldt_entry    = native_write_ldt_entry,
 294         .cpu.write_gdt_entry    = native_write_gdt_entry,
 295         .cpu.write_idt_entry    = native_write_idt_entry,
 296
 297         .cpu.alloc_ldt          = paravirt_nop,
 298         .cpu.free_ldt           = paravirt_nop,
 299
 300         .cpu.load_sp0           = native_load_sp0,
 301
 302         .cpu.usergs_sysret64    = native_usergs_sysret64,
 303         .cpu.iret               = native_iret,
 304
 305 #ifdef CONFIG_X86_IOPL_IOPERM
 306         .cpu.invalidate_io_bitmap       = native_tss_invalidate_io_bitmap,
 307         .cpu.update_io_bitmap           = native_tss_update_io_bitmap,
 308 #endif
 309
 310         .cpu.start_context_switch       = paravirt_nop,
 311         .cpu.end_context_switch         = paravirt_nop,
 312
 313         /* Irq ops. */
 314         .irq.save_fl            = __PV_IS_CALLEE_SAVE(native_save_fl),
 315         .irq.restore_fl         = __PV_IS_CALLEE_SAVE(native_restore_fl),
 316         .irq.irq_disable        = __PV_IS_CALLEE_SAVE(native_irq_disable),
 317         .irq.irq_enable         = __PV_IS_CALLEE_SAVE(native_irq_enable),
 318         .irq.safe_halt          = native_safe_halt,
 319         .irq.halt               = native_halt,
 320 #endif /* CONFIG_PARAVIRT_XXL */
 321
 322         /* Mmu ops. */
 323         .mmu.flush_tlb_user     = native_flush_tlb_local,
 324         .mmu.flush_tlb_kernel   = native_flush_tlb_global,
 325         .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
 326         .mmu.flush_tlb_others   = native_flush_tlb_others,
 327         .mmu.tlb_remove_table   =
 328                         (void (*)(struct mmu_gather *, void *))tlb_remove_page,
 329
 330         .mmu.exit_mmap          = paravirt_nop,
 331
 332 #ifdef CONFIG_PARAVIRT_XXL
 333         .mmu.read_cr2           = __PV_IS_CALLEE_SAVE(native_read_cr2),
 334         .mmu.write_cr2          = native_write_cr2,
 335         .mmu.read_cr3           = __native_read_cr3,
 336         .mmu.write_cr3          = native_write_cr3,
 337
 338         .mmu.pgd_alloc          = __paravirt_pgd_alloc,
 339         .mmu.pgd_free           = paravirt_nop,
 340
 341         .mmu.alloc_pte          = paravirt_nop,
 342         .mmu.alloc_pmd          = paravirt_nop,
 343         .mmu.alloc_pud          = paravirt_nop,
 344         .mmu.alloc_p4d          = paravirt_nop,
 345         .mmu.release_pte        = paravirt_nop,
 346         .mmu.release_pmd        = paravirt_nop,
 347         .mmu.release_pud        = paravirt_nop,
 348         .mmu.release_p4d        = paravirt_nop,
 349
 350         .mmu.set_pte            = native_set_pte,
 351         .mmu.set_pmd            = native_set_pmd,
 352
 353         .mmu.ptep_modify_prot_start     = __ptep_modify_prot_start,
 354         .mmu.ptep_modify_prot_commit    = __ptep_modify_prot_commit,
 355
 356         .mmu.set_pud            = native_set_pud,
 357
 358         .mmu.pmd_val            = PTE_IDENT,
 359         .mmu.make_pmd           = PTE_IDENT,
 360
 361         .mmu.pud_val            = PTE_IDENT,
 362         .mmu.make_pud           = PTE_IDENT,
 363
 364         .mmu.set_p4d            = native_set_p4d,
 365
 366 #if CONFIG_PGTABLE_LEVELS >= 5
 367         .mmu.p4d_val            = PTE_IDENT,
 368         .mmu.make_p4d           = PTE_IDENT,
 369
 370         .mmu.set_pgd            = native_set_pgd,
 371 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
 372
 373         .mmu.pte_val            = PTE_IDENT,
 374         .mmu.pgd_val            = PTE_IDENT,
 375
 376         .mmu.make_pte           = PTE_IDENT,
 377         .mmu.make_pgd           = PTE_IDENT,
 378
 379         .mmu.dup_mmap           = paravirt_nop,
 380         .mmu.activate_mm        = paravirt_nop,
 381
 382         .mmu.lazy_mode = {
 383                 .enter          = paravirt_nop,
 384                 .leave          = paravirt_nop,
 385                 .flush          = paravirt_nop,
 386         },
 387
 388         .mmu.set_fixmap         = native_set_fixmap,
 389 #endif /* CONFIG_PARAVIRT_XXL */
 390
 391 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
 392         /* Lock ops. */
 393 #ifdef CONFIG_SMP
 394         .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
 395         .lock.queued_spin_unlock        =
 396                                 PV_CALLEE_SAVE(__native_queued_spin_unlock),
 397         .lock.wait                      = paravirt_nop,
 398         .lock.kick                      = paravirt_nop,
 399         .lock.vcpu_is_preempted         =
 400                                 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
 401 #endif /* SMP */
 402 #endif
 403 };
 404
 405 #ifdef CONFIG_PARAVIRT_XXL
 406 /* At this point, native_get/set_debugreg has real function entries */
 407 NOKPROBE_SYMBOL(native_get_debugreg);
 408 NOKPROBE_SYMBOL(native_set_debugreg);
 409 NOKPROBE_SYMBOL(native_load_idt);
 410 #endif
 411
 412 EXPORT_SYMBOL(pv_ops);
 413 EXPORT_SYMBOL_GPL(pv_info);