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/module.h> 23 #include <linux/efi.h> 24 #include <linux/bcd.h> 25 #include <linux/highmem.h> 26 #include <linux/kprobes.h> 27 28 #include <asm/bug.h> 29 #include <asm/paravirt.h> 30 #include <asm/debugreg.h> 31 #include <asm/desc.h> 32 #include <asm/setup.h> 33 #include <asm/pgtable.h> 34 #include <asm/time.h> 35 #include <asm/pgalloc.h> 36 #include <asm/irq.h> 37 #include <asm/delay.h> 38 #include <asm/fixmap.h> 39 #include <asm/apic.h> 40 #include <asm/tlbflush.h> 41 #include <asm/timer.h> 42 #include <asm/special_insns.h> 43 44 /* 45 * nop stub, which must not clobber anything *including the stack* to 46 * avoid confusing the entry prologues. 47 */ 48 extern void _paravirt_nop(void); 49 asm (".pushsection .entry.text, \"ax\"\n" 50 ".global _paravirt_nop\n" 51 "_paravirt_nop:\n\t" 52 "ret\n\t" 53 ".size _paravirt_nop, . - _paravirt_nop\n\t" 54 ".type _paravirt_nop, @function\n\t" 55 ".popsection"); 56 57 /* identity function, which can be inlined */ 58 u32 _paravirt_ident_32(u32 x) 59 { 60 return x; 61 } 62 63 u64 _paravirt_ident_64(u64 x) 64 { 65 return x; 66 } 67 68 void __init default_banner(void) 69 { 70 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 71 pv_info.name); 72 } 73 74 /* Undefined instruction for dealing with missing ops pointers. */ 75 static const unsigned char ud2a[] = { 0x0f, 0x0b }; 76 77 struct branch { 78 unsigned char opcode; 79 u32 delta; 80 } __attribute__((packed)); 81 82 unsigned paravirt_patch_call(void *insnbuf, 83 const void *target, u16 tgt_clobbers, 84 unsigned long addr, u16 site_clobbers, 85 unsigned len) 86 { 87 struct branch *b = insnbuf; 88 unsigned long delta = (unsigned long)target - (addr+5); 89 90 if (tgt_clobbers & ~site_clobbers) 91 return len; /* target would clobber too much for this site */ 92 if (len < 5) 93 return len; /* call too long for patch site */ 94 95 b->opcode = 0xe8; /* call */ 96 b->delta = delta; 97 BUILD_BUG_ON(sizeof(*b) != 5); 98 99 return 5; 100 } 101 102 unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 103 unsigned long addr, unsigned len) 104 { 105 struct branch *b = insnbuf; 106 unsigned long delta = (unsigned long)target - (addr+5); 107 108 if (len < 5) 109 return len; /* call too long for patch site */ 110 111 b->opcode = 0xe9; /* jmp */ 112 b->delta = delta; 113 114 return 5; 115 } 116 117 /* Neat trick to map patch type back to the call within the 118 * corresponding structure. */ 119 static void *get_call_destination(u8 type) 120 { 121 struct paravirt_patch_template tmpl = { 122 .pv_init_ops = pv_init_ops, 123 .pv_time_ops = pv_time_ops, 124 .pv_cpu_ops = pv_cpu_ops, 125 .pv_irq_ops = pv_irq_ops, 126 .pv_mmu_ops = pv_mmu_ops, 127 #ifdef CONFIG_PARAVIRT_SPINLOCKS 128 .pv_lock_ops = pv_lock_ops, 129 #endif 130 }; 131 return *((void **)&tmpl + type); 132 } 133 134 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, 135 unsigned long addr, unsigned len) 136 { 137 void *opfunc = get_call_destination(type); 138 unsigned ret; 139 140 if (opfunc == NULL) 141 /* If there's no function, patch it with a ud2a (BUG) */ 142 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a)); 143 else if (opfunc == _paravirt_nop) 144 ret = 0; 145 146 /* identity functions just return their single argument */ 147 else if (opfunc == _paravirt_ident_32) 148 ret = paravirt_patch_ident_32(insnbuf, len); 149 else if (opfunc == _paravirt_ident_64) 150 ret = paravirt_patch_ident_64(insnbuf, len); 151 152 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) || 153 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64)) 154 /* If operation requires a jmp, then jmp */ 155 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 156 else 157 /* Otherwise call the function; assume target could 158 clobber any caller-save reg */ 159 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY, 160 addr, clobbers, len); 161 162 return ret; 163 } 164 165 unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 166 const char *start, const char *end) 167 { 168 unsigned insn_len = end - start; 169 170 if (insn_len > len || start == NULL) 171 insn_len = len; 172 else 173 memcpy(insnbuf, start, insn_len); 174 175 return insn_len; 176 } 177 178 static void native_flush_tlb(void) 179 { 180 __native_flush_tlb(); 181 } 182 183 /* 184 * Global pages have to be flushed a bit differently. Not a real 185 * performance problem because this does not happen often. 186 */ 187 static void native_flush_tlb_global(void) 188 { 189 __native_flush_tlb_global(); 190 } 191 192 static void native_flush_tlb_single(unsigned long addr) 193 { 194 __native_flush_tlb_single(addr); 195 } 196 197 struct static_key paravirt_steal_enabled; 198 struct static_key paravirt_steal_rq_enabled; 199 200 static u64 native_steal_clock(int cpu) 201 { 202 return 0; 203 } 204 205 /* These are in entry.S */ 206 extern void native_iret(void); 207 extern void native_usergs_sysret64(void); 208 209 static struct resource reserve_ioports = { 210 .start = 0, 211 .end = IO_SPACE_LIMIT, 212 .name = "paravirt-ioport", 213 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 214 }; 215 216 /* 217 * Reserve the whole legacy IO space to prevent any legacy drivers 218 * from wasting time probing for their hardware. This is a fairly 219 * brute-force approach to disabling all non-virtual drivers. 220 * 221 * Note that this must be called very early to have any effect. 222 */ 223 int paravirt_disable_iospace(void) 224 { 225 return request_resource(&ioport_resource, &reserve_ioports); 226 } 227 228 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 229 230 static inline void enter_lazy(enum paravirt_lazy_mode mode) 231 { 232 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 233 234 this_cpu_write(paravirt_lazy_mode, mode); 235 } 236 237 static void leave_lazy(enum paravirt_lazy_mode mode) 238 { 239 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode); 240 241 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE); 242 } 243 244 void paravirt_enter_lazy_mmu(void) 245 { 246 enter_lazy(PARAVIRT_LAZY_MMU); 247 } 248 249 void paravirt_leave_lazy_mmu(void) 250 { 251 leave_lazy(PARAVIRT_LAZY_MMU); 252 } 253 254 void paravirt_flush_lazy_mmu(void) 255 { 256 preempt_disable(); 257 258 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 259 arch_leave_lazy_mmu_mode(); 260 arch_enter_lazy_mmu_mode(); 261 } 262 263 preempt_enable(); 264 } 265 266 void paravirt_start_context_switch(struct task_struct *prev) 267 { 268 BUG_ON(preemptible()); 269 270 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) { 271 arch_leave_lazy_mmu_mode(); 272 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES); 273 } 274 enter_lazy(PARAVIRT_LAZY_CPU); 275 } 276 277 void paravirt_end_context_switch(struct task_struct *next) 278 { 279 BUG_ON(preemptible()); 280 281 leave_lazy(PARAVIRT_LAZY_CPU); 282 283 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES)) 284 arch_enter_lazy_mmu_mode(); 285 } 286 287 enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 288 { 289 if (in_interrupt()) 290 return PARAVIRT_LAZY_NONE; 291 292 return this_cpu_read(paravirt_lazy_mode); 293 } 294 295 struct pv_info pv_info = { 296 .name = "bare hardware", 297 .kernel_rpl = 0, 298 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 299 300 #ifdef CONFIG_X86_64 301 .extra_user_64bit_cs = __USER_CS, 302 #endif 303 }; 304 305 struct pv_init_ops pv_init_ops = { 306 .patch = native_patch, 307 }; 308 309 struct pv_time_ops pv_time_ops = { 310 .sched_clock = native_sched_clock, 311 .steal_clock = native_steal_clock, 312 }; 313 314 __visible struct pv_irq_ops pv_irq_ops = { 315 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 316 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 317 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 318 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 319 .safe_halt = native_safe_halt, 320 .halt = native_halt, 321 #ifdef CONFIG_X86_64 322 .adjust_exception_frame = paravirt_nop, 323 #endif 324 }; 325 326 __visible struct pv_cpu_ops pv_cpu_ops = { 327 .cpuid = native_cpuid, 328 .get_debugreg = native_get_debugreg, 329 .set_debugreg = native_set_debugreg, 330 .clts = native_clts, 331 .read_cr0 = native_read_cr0, 332 .write_cr0 = native_write_cr0, 333 .read_cr4 = native_read_cr4, 334 .read_cr4_safe = native_read_cr4_safe, 335 .write_cr4 = native_write_cr4, 336 #ifdef CONFIG_X86_64 337 .read_cr8 = native_read_cr8, 338 .write_cr8 = native_write_cr8, 339 #endif 340 .wbinvd = native_wbinvd, 341 .read_msr = native_read_msr, 342 .write_msr = native_write_msr, 343 .read_msr_safe = native_read_msr_safe, 344 .write_msr_safe = native_write_msr_safe, 345 .read_pmc = native_read_pmc, 346 .load_tr_desc = native_load_tr_desc, 347 .set_ldt = native_set_ldt, 348 .load_gdt = native_load_gdt, 349 .load_idt = native_load_idt, 350 .store_idt = native_store_idt, 351 .store_tr = native_store_tr, 352 .load_tls = native_load_tls, 353 #ifdef CONFIG_X86_64 354 .load_gs_index = native_load_gs_index, 355 #endif 356 .write_ldt_entry = native_write_ldt_entry, 357 .write_gdt_entry = native_write_gdt_entry, 358 .write_idt_entry = native_write_idt_entry, 359 360 .alloc_ldt = paravirt_nop, 361 .free_ldt = paravirt_nop, 362 363 .load_sp0 = native_load_sp0, 364 365 #ifdef CONFIG_X86_64 366 .usergs_sysret64 = native_usergs_sysret64, 367 #endif 368 .iret = native_iret, 369 .swapgs = native_swapgs, 370 371 .set_iopl_mask = native_set_iopl_mask, 372 .io_delay = native_io_delay, 373 374 .start_context_switch = paravirt_nop, 375 .end_context_switch = paravirt_nop, 376 }; 377 378 /* At this point, native_get/set_debugreg has real function entries */ 379 NOKPROBE_SYMBOL(native_get_debugreg); 380 NOKPROBE_SYMBOL(native_set_debugreg); 381 NOKPROBE_SYMBOL(native_load_idt); 382 383 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE) 384 /* 32-bit pagetable entries */ 385 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32) 386 #else 387 /* 64-bit pagetable entries */ 388 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 389 #endif 390 391 struct pv_mmu_ops pv_mmu_ops = { 392 393 .read_cr2 = native_read_cr2, 394 .write_cr2 = native_write_cr2, 395 .read_cr3 = native_read_cr3, 396 .write_cr3 = native_write_cr3, 397 398 .flush_tlb_user = native_flush_tlb, 399 .flush_tlb_kernel = native_flush_tlb_global, 400 .flush_tlb_single = native_flush_tlb_single, 401 .flush_tlb_others = native_flush_tlb_others, 402 403 .pgd_alloc = __paravirt_pgd_alloc, 404 .pgd_free = paravirt_nop, 405 406 .alloc_pte = paravirt_nop, 407 .alloc_pmd = paravirt_nop, 408 .alloc_pud = paravirt_nop, 409 .release_pte = paravirt_nop, 410 .release_pmd = paravirt_nop, 411 .release_pud = paravirt_nop, 412 413 .set_pte = native_set_pte, 414 .set_pte_at = native_set_pte_at, 415 .set_pmd = native_set_pmd, 416 .set_pmd_at = native_set_pmd_at, 417 .pte_update = paravirt_nop, 418 419 .ptep_modify_prot_start = __ptep_modify_prot_start, 420 .ptep_modify_prot_commit = __ptep_modify_prot_commit, 421 422 #if CONFIG_PGTABLE_LEVELS >= 3 423 #ifdef CONFIG_X86_PAE 424 .set_pte_atomic = native_set_pte_atomic, 425 .pte_clear = native_pte_clear, 426 .pmd_clear = native_pmd_clear, 427 #endif 428 .set_pud = native_set_pud, 429 430 .pmd_val = PTE_IDENT, 431 .make_pmd = PTE_IDENT, 432 433 #if CONFIG_PGTABLE_LEVELS == 4 434 .pud_val = PTE_IDENT, 435 .make_pud = PTE_IDENT, 436 437 .set_pgd = native_set_pgd, 438 #endif 439 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */ 440 441 .pte_val = PTE_IDENT, 442 .pgd_val = PTE_IDENT, 443 444 .make_pte = PTE_IDENT, 445 .make_pgd = PTE_IDENT, 446 447 .dup_mmap = paravirt_nop, 448 .exit_mmap = paravirt_nop, 449 .activate_mm = paravirt_nop, 450 451 .lazy_mode = { 452 .enter = paravirt_nop, 453 .leave = paravirt_nop, 454 .flush = paravirt_nop, 455 }, 456 457 .set_fixmap = native_set_fixmap, 458 }; 459 460 EXPORT_SYMBOL_GPL(pv_time_ops); 461 EXPORT_SYMBOL (pv_cpu_ops); 462 EXPORT_SYMBOL (pv_mmu_ops); 463 EXPORT_SYMBOL_GPL(pv_info); 464 EXPORT_SYMBOL (pv_irq_ops); 465