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 void __init default_banner(void) 60 { 61 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 62 pv_info.name); 63 } 64 65 /* Undefined instruction for dealing with missing ops pointers. */ 66 static const unsigned char ud2a[] = { 0x0f, 0x0b }; 67 68 struct branch { 69 unsigned char opcode; 70 u32 delta; 71 } __attribute__((packed)); 72 73 static unsigned paravirt_patch_call(void *insnbuf, const void *target, 74 unsigned long addr, unsigned len) 75 { 76 struct branch *b = insnbuf; 77 unsigned long delta = (unsigned long)target - (addr+5); 78 79 if (len < 5) { 80 #ifdef CONFIG_RETPOLINE 81 WARN_ONCE(1, "Failing to patch indirect CALL in %ps\n", (void *)addr); 82 #endif 83 return len; /* call too long for patch site */ 84 } 85 86 b->opcode = 0xe8; /* call */ 87 b->delta = delta; 88 BUILD_BUG_ON(sizeof(*b) != 5); 89 90 return 5; 91 } 92 93 #ifdef CONFIG_PARAVIRT_XXL 94 /* identity function, which can be inlined */ 95 u64 notrace _paravirt_ident_64(u64 x) 96 { 97 return x; 98 } 99 100 static unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 101 unsigned long addr, unsigned len) 102 { 103 struct branch *b = insnbuf; 104 unsigned long delta = (unsigned long)target - (addr+5); 105 106 if (len < 5) { 107 #ifdef CONFIG_RETPOLINE 108 WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr); 109 #endif 110 return len; /* call too long for patch site */ 111 } 112 113 b->opcode = 0xe9; /* jmp */ 114 b->delta = delta; 115 116 return 5; 117 } 118 #endif 119 120 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key); 121 122 void __init native_pv_lock_init(void) 123 { 124 if (!static_cpu_has(X86_FEATURE_HYPERVISOR)) 125 static_branch_disable(&virt_spin_lock_key); 126 } 127 128 unsigned paravirt_patch_default(u8 type, void *insnbuf, 129 unsigned long addr, unsigned len) 130 { 131 /* 132 * Neat trick to map patch type back to the call within the 133 * corresponding structure. 134 */ 135 void *opfunc = *((void **)&pv_ops + type); 136 unsigned ret; 137 138 if (opfunc == NULL) 139 /* If there's no function, patch it with a ud2a (BUG) */ 140 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a)); 141 else if (opfunc == _paravirt_nop) 142 ret = 0; 143 144 #ifdef CONFIG_PARAVIRT_XXL 145 /* identity functions just return their single argument */ 146 else if (opfunc == _paravirt_ident_64) 147 ret = paravirt_patch_ident_64(insnbuf, len); 148 149 else if (type == PARAVIRT_PATCH(cpu.iret) || 150 type == PARAVIRT_PATCH(cpu.usergs_sysret64)) 151 /* If operation requires a jmp, then jmp */ 152 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 153 #endif 154 else 155 /* Otherwise call the function. */ 156 ret = paravirt_patch_call(insnbuf, opfunc, addr, len); 157 158 return ret; 159 } 160 161 unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 162 const char *start, const char *end) 163 { 164 unsigned insn_len = end - start; 165 166 if (insn_len > len || start == NULL) 167 insn_len = len; 168 else 169 memcpy(insnbuf, start, insn_len); 170 171 return insn_len; 172 } 173 174 static void native_flush_tlb(void) 175 { 176 __native_flush_tlb(); 177 } 178 179 /* 180 * Global pages have to be flushed a bit differently. Not a real 181 * performance problem because this does not happen often. 182 */ 183 static void native_flush_tlb_global(void) 184 { 185 __native_flush_tlb_global(); 186 } 187 188 static void native_flush_tlb_one_user(unsigned long addr) 189 { 190 __native_flush_tlb_one_user(addr); 191 } 192 193 struct static_key paravirt_steal_enabled; 194 struct static_key paravirt_steal_rq_enabled; 195 196 static u64 native_steal_clock(int cpu) 197 { 198 return 0; 199 } 200 201 /* These are in entry.S */ 202 extern void native_iret(void); 203 extern void native_usergs_sysret64(void); 204 205 static struct resource reserve_ioports = { 206 .start = 0, 207 .end = IO_SPACE_LIMIT, 208 .name = "paravirt-ioport", 209 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 210 }; 211 212 /* 213 * Reserve the whole legacy IO space to prevent any legacy drivers 214 * from wasting time probing for their hardware. This is a fairly 215 * brute-force approach to disabling all non-virtual drivers. 216 * 217 * Note that this must be called very early to have any effect. 218 */ 219 int paravirt_disable_iospace(void) 220 { 221 return request_resource(&ioport_resource, &reserve_ioports); 222 } 223 224 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 225 226 static inline void enter_lazy(enum paravirt_lazy_mode mode) 227 { 228 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 229 230 this_cpu_write(paravirt_lazy_mode, mode); 231 } 232 233 static void leave_lazy(enum paravirt_lazy_mode mode) 234 { 235 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode); 236 237 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE); 238 } 239 240 void paravirt_enter_lazy_mmu(void) 241 { 242 enter_lazy(PARAVIRT_LAZY_MMU); 243 } 244 245 void paravirt_leave_lazy_mmu(void) 246 { 247 leave_lazy(PARAVIRT_LAZY_MMU); 248 } 249 250 void paravirt_flush_lazy_mmu(void) 251 { 252 preempt_disable(); 253 254 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 255 arch_leave_lazy_mmu_mode(); 256 arch_enter_lazy_mmu_mode(); 257 } 258 259 preempt_enable(); 260 } 261 262 #ifdef CONFIG_PARAVIRT_XXL 263 void paravirt_start_context_switch(struct task_struct *prev) 264 { 265 BUG_ON(preemptible()); 266 267 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) { 268 arch_leave_lazy_mmu_mode(); 269 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES); 270 } 271 enter_lazy(PARAVIRT_LAZY_CPU); 272 } 273 274 void paravirt_end_context_switch(struct task_struct *next) 275 { 276 BUG_ON(preemptible()); 277 278 leave_lazy(PARAVIRT_LAZY_CPU); 279 280 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES)) 281 arch_enter_lazy_mmu_mode(); 282 } 283 #endif 284 285 enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 286 { 287 if (in_interrupt()) 288 return PARAVIRT_LAZY_NONE; 289 290 return this_cpu_read(paravirt_lazy_mode); 291 } 292 293 struct pv_info pv_info = { 294 .name = "bare hardware", 295 #ifdef CONFIG_PARAVIRT_XXL 296 .kernel_rpl = 0, 297 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 298 299 #ifdef CONFIG_X86_64 300 .extra_user_64bit_cs = __USER_CS, 301 #endif 302 #endif 303 }; 304 305 /* 64-bit pagetable entries */ 306 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 307 308 struct paravirt_patch_template pv_ops = { 309 /* Init ops. */ 310 .init.patch = native_patch, 311 312 /* Time ops. */ 313 .time.sched_clock = native_sched_clock, 314 .time.steal_clock = native_steal_clock, 315 316 /* Cpu ops. */ 317 .cpu.io_delay = native_io_delay, 318 319 #ifdef CONFIG_PARAVIRT_XXL 320 .cpu.cpuid = native_cpuid, 321 .cpu.get_debugreg = native_get_debugreg, 322 .cpu.set_debugreg = native_set_debugreg, 323 .cpu.read_cr0 = native_read_cr0, 324 .cpu.write_cr0 = native_write_cr0, 325 .cpu.write_cr4 = native_write_cr4, 326 #ifdef CONFIG_X86_64 327 .cpu.read_cr8 = native_read_cr8, 328 .cpu.write_cr8 = native_write_cr8, 329 #endif 330 .cpu.wbinvd = native_wbinvd, 331 .cpu.read_msr = native_read_msr, 332 .cpu.write_msr = native_write_msr, 333 .cpu.read_msr_safe = native_read_msr_safe, 334 .cpu.write_msr_safe = native_write_msr_safe, 335 .cpu.read_pmc = native_read_pmc, 336 .cpu.load_tr_desc = native_load_tr_desc, 337 .cpu.set_ldt = native_set_ldt, 338 .cpu.load_gdt = native_load_gdt, 339 .cpu.load_idt = native_load_idt, 340 .cpu.store_tr = native_store_tr, 341 .cpu.load_tls = native_load_tls, 342 #ifdef CONFIG_X86_64 343 .cpu.load_gs_index = native_load_gs_index, 344 #endif 345 .cpu.write_ldt_entry = native_write_ldt_entry, 346 .cpu.write_gdt_entry = native_write_gdt_entry, 347 .cpu.write_idt_entry = native_write_idt_entry, 348 349 .cpu.alloc_ldt = paravirt_nop, 350 .cpu.free_ldt = paravirt_nop, 351 352 .cpu.load_sp0 = native_load_sp0, 353 354 #ifdef CONFIG_X86_64 355 .cpu.usergs_sysret64 = native_usergs_sysret64, 356 #endif 357 .cpu.iret = native_iret, 358 .cpu.swapgs = native_swapgs, 359 360 .cpu.set_iopl_mask = native_set_iopl_mask, 361 362 .cpu.start_context_switch = paravirt_nop, 363 .cpu.end_context_switch = paravirt_nop, 364 365 /* Irq ops. */ 366 .irq.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 367 .irq.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 368 .irq.irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 369 .irq.irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 370 .irq.safe_halt = native_safe_halt, 371 .irq.halt = native_halt, 372 #endif /* CONFIG_PARAVIRT_XXL */ 373 374 /* Mmu ops. */ 375 .mmu.flush_tlb_user = native_flush_tlb, 376 .mmu.flush_tlb_kernel = native_flush_tlb_global, 377 .mmu.flush_tlb_one_user = native_flush_tlb_one_user, 378 .mmu.flush_tlb_others = native_flush_tlb_others, 379 .mmu.tlb_remove_table = 380 (void (*)(struct mmu_gather *, void *))tlb_remove_page, 381 382 .mmu.exit_mmap = paravirt_nop, 383 384 #ifdef CONFIG_PARAVIRT_XXL 385 .mmu.read_cr2 = native_read_cr2, 386 .mmu.write_cr2 = native_write_cr2, 387 .mmu.read_cr3 = __native_read_cr3, 388 .mmu.write_cr3 = native_write_cr3, 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.activate_mm = paravirt_nop, 443 444 .mmu.lazy_mode = { 445 .enter = paravirt_nop, 446 .leave = paravirt_nop, 447 .flush = paravirt_nop, 448 }, 449 450 .mmu.set_fixmap = native_set_fixmap, 451 #endif /* CONFIG_PARAVIRT_XXL */ 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 #ifdef CONFIG_PARAVIRT_XXL 468 /* At this point, native_get/set_debugreg has real function entries */ 469 NOKPROBE_SYMBOL(native_get_debugreg); 470 NOKPROBE_SYMBOL(native_set_debugreg); 471 NOKPROBE_SYMBOL(native_load_idt); 472 #endif 473 474 EXPORT_SYMBOL(pv_ops); 475 EXPORT_SYMBOL_GPL(pv_info); 476