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 #include <linux/static_call.h>
  18
  19 #include <asm/bug.h>
  20 #include <asm/paravirt.h>
  21 #include <asm/debugreg.h>
  22 #include <asm/desc.h>
  23 #include <asm/setup.h>
  24 #include <asm/time.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/irq.h>
  27 #include <asm/delay.h>
  28 #include <asm/fixmap.h>
  29 #include <asm/apic.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/timer.h>
  32 #include <asm/special_insns.h>
  33 #include <asm/tlb.h>
  34 #include <asm/io_bitmap.h>
  35 #include <asm/gsseg.h>
  36
  37 /*
  38  * nop stub, which must not clobber anything *including the stack* to
  39  * avoid confusing the entry prologues.
  40  */
  41 DEFINE_PARAVIRT_ASM(_paravirt_nop, "", .entry.text);
  42
  43 /* stub always returning 0. */
  44 DEFINE_PARAVIRT_ASM(paravirt_ret0, "xor %eax,%eax", .entry.text);
  45
  46 void __init default_banner(void)
  47 {
  48         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  49                pv_info.name);
  50 }
  51
  52 /* Undefined instruction for dealing with missing ops pointers. */
  53 noinstr void paravirt_BUG(void)
  54 {
  55         BUG();
  56 }
  57
  58 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
  59                                     unsigned long addr, unsigned len)
  60 {
  61         __text_gen_insn(insn_buff, CALL_INSN_OPCODE,
  62                         (void *)addr, target, CALL_INSN_SIZE);
  63         return CALL_INSN_SIZE;
  64 }
  65
  66 #ifdef CONFIG_PARAVIRT_XXL
  67 DEFINE_PARAVIRT_ASM(_paravirt_ident_64, "mov %rdi, %rax", .text);
  68 DEFINE_PARAVIRT_ASM(pv_native_save_fl, "pushf; pop %rax", .noinstr.text);
  69 DEFINE_PARAVIRT_ASM(pv_native_irq_disable, "cli", .noinstr.text);
  70 DEFINE_PARAVIRT_ASM(pv_native_irq_enable, "sti", .noinstr.text);
  71 DEFINE_PARAVIRT_ASM(pv_native_read_cr2, "mov %cr2, %rax", .noinstr.text);
  72 #endif
  73
  74 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
  75
  76 void __init native_pv_lock_init(void)
  77 {
  78         if (IS_ENABLED(CONFIG_PARAVIRT_SPINLOCKS) &&
  79             !boot_cpu_has(X86_FEATURE_HYPERVISOR))
  80                 static_branch_disable(&virt_spin_lock_key);
  81 }
  82
  83 static void native_tlb_remove_table(struct mmu_gather *tlb, void *table)
  84 {
  85         tlb_remove_page(tlb, table);
  86 }
  87
  88 unsigned int paravirt_patch(u8 type, void *insn_buff, unsigned long addr,
  89                             unsigned int len)
  90 {
  91         /*
  92          * Neat trick to map patch type back to the call within the
  93          * corresponding structure.
  94          */
  95         void *opfunc = *((void **)&pv_ops + type);
  96         unsigned ret;
  97
  98         if (opfunc == NULL)
  99                 /* If there's no function, patch it with paravirt_BUG() */
 100                 ret = paravirt_patch_call(insn_buff, paravirt_BUG, addr, len);
 101         else if (opfunc == _paravirt_nop)
 102                 ret = 0;
 103         else
 104                 /* Otherwise call the function. */
 105                 ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
 106
 107         return ret;
 108 }
 109
 110 struct static_key paravirt_steal_enabled;
 111 struct static_key paravirt_steal_rq_enabled;
 112
 113 static u64 native_steal_clock(int cpu)
 114 {
 115         return 0;
 116 }
 117
 118 DEFINE_STATIC_CALL(pv_steal_clock, native_steal_clock);
 119 DEFINE_STATIC_CALL(pv_sched_clock, native_sched_clock);
 120
 121 void paravirt_set_sched_clock(u64 (*func)(void))
 122 {
 123         static_call_update(pv_sched_clock, func);
 124 }
 125
 126 /* These are in entry.S */
 127 static struct resource reserve_ioports = {
 128         .start = 0,
 129         .end = IO_SPACE_LIMIT,
 130         .name = "paravirt-ioport",
 131         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 132 };
 133
 134 /*
 135  * Reserve the whole legacy IO space to prevent any legacy drivers
 136  * from wasting time probing for their hardware.  This is a fairly
 137  * brute-force approach to disabling all non-virtual drivers.
 138  *
 139  * Note that this must be called very early to have any effect.
 140  */
 141 int paravirt_disable_iospace(void)
 142 {
 143         return request_resource(&ioport_resource, &reserve_ioports);
 144 }
 145
 146 #ifdef CONFIG_PARAVIRT_XXL
 147 static noinstr void pv_native_write_cr2(unsigned long val)
 148 {
 149         native_write_cr2(val);
 150 }
 151
 152 static noinstr unsigned long pv_native_get_debugreg(int regno)
 153 {
 154         return native_get_debugreg(regno);
 155 }
 156
 157 static noinstr void pv_native_set_debugreg(int regno, unsigned long val)
 158 {
 159         native_set_debugreg(regno, val);
 160 }
 161
 162 noinstr void pv_native_wbinvd(void)
 163 {
 164         native_wbinvd();
 165 }
 166
 167 static noinstr void pv_native_safe_halt(void)
 168 {
 169         native_safe_halt();
 170 }
 171 #endif
 172
 173 struct pv_info pv_info = {
 174         .name = "bare hardware",
 175 #ifdef CONFIG_PARAVIRT_XXL
 176         .extra_user_64bit_cs = __USER_CS,
 177 #endif
 178 };
 179
 180 /* 64-bit pagetable entries */
 181 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 182
 183 struct paravirt_patch_template pv_ops = {
 184         /* Cpu ops. */
 185         .cpu.io_delay           = native_io_delay,
 186
 187 #ifdef CONFIG_PARAVIRT_XXL
 188         .cpu.cpuid              = native_cpuid,
 189         .cpu.get_debugreg       = pv_native_get_debugreg,
 190         .cpu.set_debugreg       = pv_native_set_debugreg,
 191         .cpu.read_cr0           = native_read_cr0,
 192         .cpu.write_cr0          = native_write_cr0,
 193         .cpu.write_cr4          = native_write_cr4,
 194         .cpu.wbinvd             = pv_native_wbinvd,
 195         .cpu.read_msr           = native_read_msr,
 196         .cpu.write_msr          = native_write_msr,
 197         .cpu.read_msr_safe      = native_read_msr_safe,
 198         .cpu.write_msr_safe     = native_write_msr_safe,
 199         .cpu.read_pmc           = native_read_pmc,
 200         .cpu.load_tr_desc       = native_load_tr_desc,
 201         .cpu.set_ldt            = native_set_ldt,
 202         .cpu.load_gdt           = native_load_gdt,
 203         .cpu.load_idt           = native_load_idt,
 204         .cpu.store_tr           = native_store_tr,
 205         .cpu.load_tls           = native_load_tls,
 206         .cpu.load_gs_index      = native_load_gs_index,
 207         .cpu.write_ldt_entry    = native_write_ldt_entry,
 208         .cpu.write_gdt_entry    = native_write_gdt_entry,
 209         .cpu.write_idt_entry    = native_write_idt_entry,
 210
 211         .cpu.alloc_ldt          = paravirt_nop,
 212         .cpu.free_ldt           = paravirt_nop,
 213
 214         .cpu.load_sp0           = native_load_sp0,
 215
 216 #ifdef CONFIG_X86_IOPL_IOPERM
 217         .cpu.invalidate_io_bitmap       = native_tss_invalidate_io_bitmap,
 218         .cpu.update_io_bitmap           = native_tss_update_io_bitmap,
 219 #endif
 220
 221         .cpu.start_context_switch       = paravirt_nop,
 222         .cpu.end_context_switch         = paravirt_nop,
 223
 224         /* Irq ops. */
 225         .irq.save_fl            = __PV_IS_CALLEE_SAVE(pv_native_save_fl),
 226         .irq.irq_disable        = __PV_IS_CALLEE_SAVE(pv_native_irq_disable),
 227         .irq.irq_enable         = __PV_IS_CALLEE_SAVE(pv_native_irq_enable),
 228         .irq.safe_halt          = pv_native_safe_halt,
 229         .irq.halt               = native_halt,
 230 #endif /* CONFIG_PARAVIRT_XXL */
 231
 232         /* Mmu ops. */
 233         .mmu.flush_tlb_user     = native_flush_tlb_local,
 234         .mmu.flush_tlb_kernel   = native_flush_tlb_global,
 235         .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
 236         .mmu.flush_tlb_multi    = native_flush_tlb_multi,
 237         .mmu.tlb_remove_table   = native_tlb_remove_table,
 238
 239         .mmu.exit_mmap          = paravirt_nop,
 240         .mmu.notify_page_enc_status_changed     = paravirt_nop,
 241
 242 #ifdef CONFIG_PARAVIRT_XXL
 243         .mmu.read_cr2           = __PV_IS_CALLEE_SAVE(pv_native_read_cr2),
 244         .mmu.write_cr2          = pv_native_write_cr2,
 245         .mmu.read_cr3           = __native_read_cr3,
 246         .mmu.write_cr3          = native_write_cr3,
 247
 248         .mmu.pgd_alloc          = __paravirt_pgd_alloc,
 249         .mmu.pgd_free           = paravirt_nop,
 250
 251         .mmu.alloc_pte          = paravirt_nop,
 252         .mmu.alloc_pmd          = paravirt_nop,
 253         .mmu.alloc_pud          = paravirt_nop,
 254         .mmu.alloc_p4d          = paravirt_nop,
 255         .mmu.release_pte        = paravirt_nop,
 256         .mmu.release_pmd        = paravirt_nop,
 257         .mmu.release_pud        = paravirt_nop,
 258         .mmu.release_p4d        = paravirt_nop,
 259
 260         .mmu.set_pte            = native_set_pte,
 261         .mmu.set_pmd            = native_set_pmd,
 262
 263         .mmu.ptep_modify_prot_start     = __ptep_modify_prot_start,
 264         .mmu.ptep_modify_prot_commit    = __ptep_modify_prot_commit,
 265
 266         .mmu.set_pud            = native_set_pud,
 267
 268         .mmu.pmd_val            = PTE_IDENT,
 269         .mmu.make_pmd           = PTE_IDENT,
 270
 271         .mmu.pud_val            = PTE_IDENT,
 272         .mmu.make_pud           = PTE_IDENT,
 273
 274         .mmu.set_p4d            = native_set_p4d,
 275
 276 #if CONFIG_PGTABLE_LEVELS >= 5
 277         .mmu.p4d_val            = PTE_IDENT,
 278         .mmu.make_p4d           = PTE_IDENT,
 279
 280         .mmu.set_pgd            = native_set_pgd,
 281 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
 282
 283         .mmu.pte_val            = PTE_IDENT,
 284         .mmu.pgd_val            = PTE_IDENT,
 285
 286         .mmu.make_pte           = PTE_IDENT,
 287         .mmu.make_pgd           = PTE_IDENT,
 288
 289         .mmu.enter_mmap         = paravirt_nop,
 290
 291         .mmu.lazy_mode = {
 292                 .enter          = paravirt_nop,
 293                 .leave          = paravirt_nop,
 294                 .flush          = paravirt_nop,
 295         },
 296
 297         .mmu.set_fixmap         = native_set_fixmap,
 298 #endif /* CONFIG_PARAVIRT_XXL */
 299
 300 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
 301         /* Lock ops. */
 302 #ifdef CONFIG_SMP
 303         .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
 304         .lock.queued_spin_unlock        =
 305                                 PV_CALLEE_SAVE(__native_queued_spin_unlock),
 306         .lock.wait                      = paravirt_nop,
 307         .lock.kick                      = paravirt_nop,
 308         .lock.vcpu_is_preempted         =
 309                                 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
 310 #endif /* SMP */
 311 #endif
 312 };
 313
 314 #ifdef CONFIG_PARAVIRT_XXL
 315 NOKPROBE_SYMBOL(native_load_idt);
 316 #endif
 317
 318 EXPORT_SYMBOL(pv_ops);
 319 EXPORT_SYMBOL_GPL(pv_info);