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 /* identity function, which can be inlined */ 60 u32 notrace _paravirt_ident_32(u32 x) 61 { 62 return x; 63 } 64 65 u64 notrace _paravirt_ident_64(u64 x) 66 { 67 return x; 68 } 69 70 void __init default_banner(void) 71 { 72 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 73 pv_info.name); 74 } 75 76 /* Undefined instruction for dealing with missing ops pointers. */ 77 static const unsigned char ud2a[] = { 0x0f, 0x0b }; 78 79 struct branch { 80 unsigned char opcode; 81 u32 delta; 82 } __attribute__((packed)); 83 84 unsigned paravirt_patch_call(void *insnbuf, 85 const void *target, u16 tgt_clobbers, 86 unsigned long addr, u16 site_clobbers, 87 unsigned len) 88 { 89 struct branch *b = insnbuf; 90 unsigned long delta = (unsigned long)target - (addr+5); 91 92 if (len < 5) { 93 #ifdef CONFIG_RETPOLINE 94 WARN_ONCE("Failing to patch indirect CALL in %ps\n", (void *)addr); 95 #endif 96 return len; /* call too long for patch site */ 97 } 98 99 b->opcode = 0xe8; /* call */ 100 b->delta = delta; 101 BUILD_BUG_ON(sizeof(*b) != 5); 102 103 return 5; 104 } 105 106 unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 107 unsigned long addr, unsigned len) 108 { 109 struct branch *b = insnbuf; 110 unsigned long delta = (unsigned long)target - (addr+5); 111 112 if (len < 5) { 113 #ifdef CONFIG_RETPOLINE 114 WARN_ONCE("Failing to patch indirect JMP in %ps\n", (void *)addr); 115 #endif 116 return len; /* call too long for patch site */ 117 } 118 119 b->opcode = 0xe9; /* jmp */ 120 b->delta = delta; 121 122 return 5; 123 } 124 125 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key); 126 127 void __init native_pv_lock_init(void) 128 { 129 if (!static_cpu_has(X86_FEATURE_HYPERVISOR)) 130 static_branch_disable(&virt_spin_lock_key); 131 } 132 133 /* 134 * Neat trick to map patch type back to the call within the 135 * corresponding structure. 136 */ 137 static void *get_call_destination(u8 type) 138 { 139 struct paravirt_patch_template tmpl = { 140 .pv_init_ops = pv_init_ops, 141 .pv_time_ops = pv_time_ops, 142 .pv_cpu_ops = pv_cpu_ops, 143 .pv_irq_ops = pv_irq_ops, 144 .pv_mmu_ops = pv_mmu_ops, 145 #ifdef CONFIG_PARAVIRT_SPINLOCKS 146 .pv_lock_ops = pv_lock_ops, 147 #endif 148 }; 149 return *((void **)&tmpl + type); 150 } 151 152 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, 153 unsigned long addr, unsigned len) 154 { 155 void *opfunc = get_call_destination(type); 156 unsigned ret; 157 158 if (opfunc == NULL) 159 /* If there's no function, patch it with a ud2a (BUG) */ 160 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a)); 161 else if (opfunc == _paravirt_nop) 162 ret = 0; 163 164 /* identity functions just return their single argument */ 165 else if (opfunc == _paravirt_ident_32) 166 ret = paravirt_patch_ident_32(insnbuf, len); 167 else if (opfunc == _paravirt_ident_64) 168 ret = paravirt_patch_ident_64(insnbuf, len); 169 170 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) || 171 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64)) 172 /* If operation requires a jmp, then jmp */ 173 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 174 else 175 /* Otherwise call the function; assume target could 176 clobber any caller-save reg */ 177 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY, 178 addr, clobbers, len); 179 180 return ret; 181 } 182 183 unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 184 const char *start, const char *end) 185 { 186 unsigned insn_len = end - start; 187 188 if (insn_len > len || start == NULL) 189 insn_len = len; 190 else 191 memcpy(insnbuf, start, insn_len); 192 193 return insn_len; 194 } 195 196 static void native_flush_tlb(void) 197 { 198 __native_flush_tlb(); 199 } 200 201 /* 202 * Global pages have to be flushed a bit differently. Not a real 203 * performance problem because this does not happen often. 204 */ 205 static void native_flush_tlb_global(void) 206 { 207 __native_flush_tlb_global(); 208 } 209 210 static void native_flush_tlb_one_user(unsigned long addr) 211 { 212 __native_flush_tlb_one_user(addr); 213 } 214 215 struct static_key paravirt_steal_enabled; 216 struct static_key paravirt_steal_rq_enabled; 217 218 static u64 native_steal_clock(int cpu) 219 { 220 return 0; 221 } 222 223 /* These are in entry.S */ 224 extern void native_iret(void); 225 extern void native_usergs_sysret64(void); 226 227 static struct resource reserve_ioports = { 228 .start = 0, 229 .end = IO_SPACE_LIMIT, 230 .name = "paravirt-ioport", 231 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 232 }; 233 234 /* 235 * Reserve the whole legacy IO space to prevent any legacy drivers 236 * from wasting time probing for their hardware. This is a fairly 237 * brute-force approach to disabling all non-virtual drivers. 238 * 239 * Note that this must be called very early to have any effect. 240 */ 241 int paravirt_disable_iospace(void) 242 { 243 return request_resource(&ioport_resource, &reserve_ioports); 244 } 245 246 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 247 248 static inline void enter_lazy(enum paravirt_lazy_mode mode) 249 { 250 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 251 252 this_cpu_write(paravirt_lazy_mode, mode); 253 } 254 255 static void leave_lazy(enum paravirt_lazy_mode mode) 256 { 257 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode); 258 259 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE); 260 } 261 262 void paravirt_enter_lazy_mmu(void) 263 { 264 enter_lazy(PARAVIRT_LAZY_MMU); 265 } 266 267 void paravirt_leave_lazy_mmu(void) 268 { 269 leave_lazy(PARAVIRT_LAZY_MMU); 270 } 271 272 void paravirt_flush_lazy_mmu(void) 273 { 274 preempt_disable(); 275 276 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 277 arch_leave_lazy_mmu_mode(); 278 arch_enter_lazy_mmu_mode(); 279 } 280 281 preempt_enable(); 282 } 283 284 void paravirt_start_context_switch(struct task_struct *prev) 285 { 286 BUG_ON(preemptible()); 287 288 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) { 289 arch_leave_lazy_mmu_mode(); 290 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES); 291 } 292 enter_lazy(PARAVIRT_LAZY_CPU); 293 } 294 295 void paravirt_end_context_switch(struct task_struct *next) 296 { 297 BUG_ON(preemptible()); 298 299 leave_lazy(PARAVIRT_LAZY_CPU); 300 301 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES)) 302 arch_enter_lazy_mmu_mode(); 303 } 304 305 enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 306 { 307 if (in_interrupt()) 308 return PARAVIRT_LAZY_NONE; 309 310 return this_cpu_read(paravirt_lazy_mode); 311 } 312 313 struct pv_info pv_info = { 314 .name = "bare hardware", 315 .kernel_rpl = 0, 316 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 317 318 #ifdef CONFIG_X86_64 319 .extra_user_64bit_cs = __USER_CS, 320 #endif 321 }; 322 323 struct pv_init_ops pv_init_ops = { 324 .patch = native_patch, 325 }; 326 327 struct pv_time_ops pv_time_ops = { 328 .sched_clock = native_sched_clock, 329 .steal_clock = native_steal_clock, 330 }; 331 332 __visible struct pv_irq_ops pv_irq_ops = { 333 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 334 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 335 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 336 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 337 .safe_halt = native_safe_halt, 338 .halt = native_halt, 339 }; 340 341 __visible struct pv_cpu_ops pv_cpu_ops = { 342 .cpuid = native_cpuid, 343 .get_debugreg = native_get_debugreg, 344 .set_debugreg = native_set_debugreg, 345 .read_cr0 = native_read_cr0, 346 .write_cr0 = native_write_cr0, 347 .write_cr4 = native_write_cr4, 348 #ifdef CONFIG_X86_64 349 .read_cr8 = native_read_cr8, 350 .write_cr8 = native_write_cr8, 351 #endif 352 .wbinvd = native_wbinvd, 353 .read_msr = native_read_msr, 354 .write_msr = native_write_msr, 355 .read_msr_safe = native_read_msr_safe, 356 .write_msr_safe = native_write_msr_safe, 357 .read_pmc = native_read_pmc, 358 .load_tr_desc = native_load_tr_desc, 359 .set_ldt = native_set_ldt, 360 .load_gdt = native_load_gdt, 361 .load_idt = native_load_idt, 362 .store_tr = native_store_tr, 363 .load_tls = native_load_tls, 364 #ifdef CONFIG_X86_64 365 .load_gs_index = native_load_gs_index, 366 #endif 367 .write_ldt_entry = native_write_ldt_entry, 368 .write_gdt_entry = native_write_gdt_entry, 369 .write_idt_entry = native_write_idt_entry, 370 371 .alloc_ldt = paravirt_nop, 372 .free_ldt = paravirt_nop, 373 374 .load_sp0 = native_load_sp0, 375 376 #ifdef CONFIG_X86_64 377 .usergs_sysret64 = native_usergs_sysret64, 378 #endif 379 .iret = native_iret, 380 .swapgs = native_swapgs, 381 382 .set_iopl_mask = native_set_iopl_mask, 383 .io_delay = native_io_delay, 384 385 .start_context_switch = paravirt_nop, 386 .end_context_switch = paravirt_nop, 387 }; 388 389 /* At this point, native_get/set_debugreg has real function entries */ 390 NOKPROBE_SYMBOL(native_get_debugreg); 391 NOKPROBE_SYMBOL(native_set_debugreg); 392 NOKPROBE_SYMBOL(native_load_idt); 393 394 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE) 395 /* 32-bit pagetable entries */ 396 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32) 397 #else 398 /* 64-bit pagetable entries */ 399 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 400 #endif 401 402 struct pv_mmu_ops pv_mmu_ops __ro_after_init = { 403 404 .read_cr2 = native_read_cr2, 405 .write_cr2 = native_write_cr2, 406 .read_cr3 = __native_read_cr3, 407 .write_cr3 = native_write_cr3, 408 409 .flush_tlb_user = native_flush_tlb, 410 .flush_tlb_kernel = native_flush_tlb_global, 411 .flush_tlb_one_user = native_flush_tlb_one_user, 412 .flush_tlb_others = native_flush_tlb_others, 413 .tlb_remove_table = (void (*)(struct mmu_gather *, void *))tlb_remove_page, 414 415 .pgd_alloc = __paravirt_pgd_alloc, 416 .pgd_free = paravirt_nop, 417 418 .alloc_pte = paravirt_nop, 419 .alloc_pmd = paravirt_nop, 420 .alloc_pud = paravirt_nop, 421 .alloc_p4d = paravirt_nop, 422 .release_pte = paravirt_nop, 423 .release_pmd = paravirt_nop, 424 .release_pud = paravirt_nop, 425 .release_p4d = paravirt_nop, 426 427 .set_pte = native_set_pte, 428 .set_pte_at = native_set_pte_at, 429 .set_pmd = native_set_pmd, 430 431 .ptep_modify_prot_start = __ptep_modify_prot_start, 432 .ptep_modify_prot_commit = __ptep_modify_prot_commit, 433 434 #if CONFIG_PGTABLE_LEVELS >= 3 435 #ifdef CONFIG_X86_PAE 436 .set_pte_atomic = native_set_pte_atomic, 437 .pte_clear = native_pte_clear, 438 .pmd_clear = native_pmd_clear, 439 #endif 440 .set_pud = native_set_pud, 441 442 .pmd_val = PTE_IDENT, 443 .make_pmd = PTE_IDENT, 444 445 #if CONFIG_PGTABLE_LEVELS >= 4 446 .pud_val = PTE_IDENT, 447 .make_pud = PTE_IDENT, 448 449 .set_p4d = native_set_p4d, 450 451 #if CONFIG_PGTABLE_LEVELS >= 5 452 .p4d_val = PTE_IDENT, 453 .make_p4d = PTE_IDENT, 454 455 .set_pgd = native_set_pgd, 456 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */ 457 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */ 458 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */ 459 460 .pte_val = PTE_IDENT, 461 .pgd_val = PTE_IDENT, 462 463 .make_pte = PTE_IDENT, 464 .make_pgd = PTE_IDENT, 465 466 .dup_mmap = paravirt_nop, 467 .exit_mmap = paravirt_nop, 468 .activate_mm = paravirt_nop, 469 470 .lazy_mode = { 471 .enter = paravirt_nop, 472 .leave = paravirt_nop, 473 .flush = paravirt_nop, 474 }, 475 476 .set_fixmap = native_set_fixmap, 477 }; 478 479 EXPORT_SYMBOL_GPL(pv_time_ops); 480 EXPORT_SYMBOL (pv_cpu_ops); 481 EXPORT_SYMBOL (pv_mmu_ops); 482 EXPORT_SYMBOL_GPL(pv_info); 483 EXPORT_SYMBOL (pv_irq_ops); 484