1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PARAVIRT_TYPES_H 3 #define _ASM_X86_PARAVIRT_TYPES_H 4 5 #ifdef CONFIG_PARAVIRT 6 7 #ifndef __ASSEMBLY__ 8 #include <linux/types.h> 9 10 #include <asm/desc_defs.h> 11 #include <asm/pgtable_types.h> 12 #include <asm/nospec-branch.h> 13 14 struct page; 15 struct thread_struct; 16 struct desc_ptr; 17 struct tss_struct; 18 struct mm_struct; 19 struct desc_struct; 20 struct task_struct; 21 struct cpumask; 22 struct flush_tlb_info; 23 struct mmu_gather; 24 struct vm_area_struct; 25 26 /* 27 * Wrapper type for pointers to code which uses the non-standard 28 * calling convention. See PV_CALL_SAVE_REGS_THUNK below. 29 */ 30 struct paravirt_callee_save { 31 void *func; 32 }; 33 34 /* general info */ 35 struct pv_info { 36 #ifdef CONFIG_PARAVIRT_XXL 37 u16 extra_user_64bit_cs; /* __USER_CS if none */ 38 #endif 39 40 const char *name; 41 }; 42 43 #ifdef CONFIG_PARAVIRT_XXL 44 struct pv_lazy_ops { 45 /* Set deferred update mode, used for batching operations. */ 46 void (*enter)(void); 47 void (*leave)(void); 48 void (*flush)(void); 49 } __no_randomize_layout; 50 #endif 51 52 struct pv_cpu_ops { 53 /* hooks for various privileged instructions */ 54 void (*io_delay)(void); 55 56 #ifdef CONFIG_PARAVIRT_XXL 57 unsigned long (*get_debugreg)(int regno); 58 void (*set_debugreg)(int regno, unsigned long value); 59 60 unsigned long (*read_cr0)(void); 61 void (*write_cr0)(unsigned long); 62 63 void (*write_cr4)(unsigned long); 64 65 /* Segment descriptor handling */ 66 void (*load_tr_desc)(void); 67 void (*load_gdt)(const struct desc_ptr *); 68 void (*load_idt)(const struct desc_ptr *); 69 void (*set_ldt)(const void *desc, unsigned entries); 70 unsigned long (*store_tr)(void); 71 void (*load_tls)(struct thread_struct *t, unsigned int cpu); 72 void (*load_gs_index)(unsigned int idx); 73 void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum, 74 const void *desc); 75 void (*write_gdt_entry)(struct desc_struct *, 76 int entrynum, const void *desc, int size); 77 void (*write_idt_entry)(gate_desc *, 78 int entrynum, const gate_desc *gate); 79 void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries); 80 void (*free_ldt)(struct desc_struct *ldt, unsigned entries); 81 82 void (*load_sp0)(unsigned long sp0); 83 84 #ifdef CONFIG_X86_IOPL_IOPERM 85 void (*invalidate_io_bitmap)(void); 86 void (*update_io_bitmap)(void); 87 #endif 88 89 /* cpuid emulation, mostly so that caps bits can be disabled */ 90 void (*cpuid)(unsigned int *eax, unsigned int *ebx, 91 unsigned int *ecx, unsigned int *edx); 92 93 /* Unsafe MSR operations. These will warn or panic on failure. */ 94 u64 (*read_msr)(unsigned int msr); 95 void (*write_msr)(unsigned int msr, unsigned low, unsigned high); 96 97 /* 98 * Safe MSR operations. 99 * read sets err to 0 or -EIO. write returns 0 or -EIO. 100 */ 101 u64 (*read_msr_safe)(unsigned int msr, int *err); 102 int (*write_msr_safe)(unsigned int msr, unsigned low, unsigned high); 103 104 u64 (*read_pmc)(int counter); 105 106 void (*start_context_switch)(struct task_struct *prev); 107 void (*end_context_switch)(struct task_struct *next); 108 #endif 109 } __no_randomize_layout; 110 111 struct pv_irq_ops { 112 #ifdef CONFIG_PARAVIRT_XXL 113 /* 114 * Get/set interrupt state. save_fl is expected to use X86_EFLAGS_IF; 115 * all other bits returned from save_fl are undefined. 116 * 117 * NOTE: These functions callers expect the callee to preserve 118 * more registers than the standard C calling convention. 119 */ 120 struct paravirt_callee_save save_fl; 121 struct paravirt_callee_save irq_disable; 122 struct paravirt_callee_save irq_enable; 123 #endif 124 void (*safe_halt)(void); 125 void (*halt)(void); 126 } __no_randomize_layout; 127 128 struct pv_mmu_ops { 129 /* TLB operations */ 130 void (*flush_tlb_user)(void); 131 void (*flush_tlb_kernel)(void); 132 void (*flush_tlb_one_user)(unsigned long addr); 133 void (*flush_tlb_multi)(const struct cpumask *cpus, 134 const struct flush_tlb_info *info); 135 136 void (*tlb_remove_table)(struct mmu_gather *tlb, void *table); 137 138 /* Hook for intercepting the destruction of an mm_struct. */ 139 void (*exit_mmap)(struct mm_struct *mm); 140 void (*notify_page_enc_status_changed)(unsigned long pfn, int npages, bool enc); 141 142 #ifdef CONFIG_PARAVIRT_XXL 143 struct paravirt_callee_save read_cr2; 144 void (*write_cr2)(unsigned long); 145 146 unsigned long (*read_cr3)(void); 147 void (*write_cr3)(unsigned long); 148 149 /* Hook for intercepting the creation/use of an mm_struct. */ 150 void (*enter_mmap)(struct mm_struct *mm); 151 152 /* Hooks for allocating and freeing a pagetable top-level */ 153 int (*pgd_alloc)(struct mm_struct *mm); 154 void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd); 155 156 /* 157 * Hooks for allocating/releasing pagetable pages when they're 158 * attached to a pagetable 159 */ 160 void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn); 161 void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn); 162 void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn); 163 void (*alloc_p4d)(struct mm_struct *mm, unsigned long pfn); 164 void (*release_pte)(unsigned long pfn); 165 void (*release_pmd)(unsigned long pfn); 166 void (*release_pud)(unsigned long pfn); 167 void (*release_p4d)(unsigned long pfn); 168 169 /* Pagetable manipulation functions */ 170 void (*set_pte)(pte_t *ptep, pte_t pteval); 171 void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval); 172 173 pte_t (*ptep_modify_prot_start)(struct vm_area_struct *vma, unsigned long addr, 174 pte_t *ptep); 175 void (*ptep_modify_prot_commit)(struct vm_area_struct *vma, unsigned long addr, 176 pte_t *ptep, pte_t pte); 177 178 struct paravirt_callee_save pte_val; 179 struct paravirt_callee_save make_pte; 180 181 struct paravirt_callee_save pgd_val; 182 struct paravirt_callee_save make_pgd; 183 184 void (*set_pud)(pud_t *pudp, pud_t pudval); 185 186 struct paravirt_callee_save pmd_val; 187 struct paravirt_callee_save make_pmd; 188 189 struct paravirt_callee_save pud_val; 190 struct paravirt_callee_save make_pud; 191 192 void (*set_p4d)(p4d_t *p4dp, p4d_t p4dval); 193 194 #if CONFIG_PGTABLE_LEVELS >= 5 195 struct paravirt_callee_save p4d_val; 196 struct paravirt_callee_save make_p4d; 197 198 void (*set_pgd)(pgd_t *pgdp, pgd_t pgdval); 199 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */ 200 201 struct pv_lazy_ops lazy_mode; 202 203 /* dom0 ops */ 204 205 /* Sometimes the physical address is a pfn, and sometimes its 206 an mfn. We can tell which is which from the index. */ 207 void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx, 208 phys_addr_t phys, pgprot_t flags); 209 #endif 210 } __no_randomize_layout; 211 212 struct arch_spinlock; 213 #ifdef CONFIG_SMP 214 #include <asm/spinlock_types.h> 215 #endif 216 217 struct qspinlock; 218 219 struct pv_lock_ops { 220 void (*queued_spin_lock_slowpath)(struct qspinlock *lock, u32 val); 221 struct paravirt_callee_save queued_spin_unlock; 222 223 void (*wait)(u8 *ptr, u8 val); 224 void (*kick)(int cpu); 225 226 struct paravirt_callee_save vcpu_is_preempted; 227 } __no_randomize_layout; 228 229 /* This contains all the paravirt structures: we get a convenient 230 * number for each function using the offset which we use to indicate 231 * what to patch. */ 232 struct paravirt_patch_template { 233 struct pv_cpu_ops cpu; 234 struct pv_irq_ops irq; 235 struct pv_mmu_ops mmu; 236 struct pv_lock_ops lock; 237 } __no_randomize_layout; 238 239 extern struct pv_info pv_info; 240 extern struct paravirt_patch_template pv_ops; 241 242 #define paravirt_ptr(op) [paravirt_opptr] "m" (pv_ops.op) 243 244 int paravirt_disable_iospace(void); 245 246 /* This generates an indirect call based on the operation type number. */ 247 #define PARAVIRT_CALL \ 248 ANNOTATE_RETPOLINE_SAFE \ 249 "call *%[paravirt_opptr];" 250 251 /* 252 * These macros are intended to wrap calls through one of the paravirt 253 * ops structs, so that they can be later identified and patched at 254 * runtime. 255 * 256 * Normally, a call to a pv_op function is a simple indirect call: 257 * (pv_op_struct.operations)(args...). 258 * 259 * Unfortunately, this is a relatively slow operation for modern CPUs, 260 * because it cannot necessarily determine what the destination 261 * address is. In this case, the address is a runtime constant, so at 262 * the very least we can patch the call to a simple direct call, or, 263 * ideally, patch an inline implementation into the callsite. (Direct 264 * calls are essentially free, because the call and return addresses 265 * are completely predictable.) 266 * 267 * For i386, these macros rely on the standard gcc "regparm(3)" calling 268 * convention, in which the first three arguments are placed in %eax, 269 * %edx, %ecx (in that order), and the remaining arguments are placed 270 * on the stack. All caller-save registers (eax,edx,ecx) are expected 271 * to be modified (either clobbered or used for return values). 272 * X86_64, on the other hand, already specifies a register-based calling 273 * conventions, returning at %rax, with parameters going in %rdi, %rsi, 274 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any 275 * special handling for dealing with 4 arguments, unlike i386. 276 * However, x86_64 also has to clobber all caller saved registers, which 277 * unfortunately, are quite a bit (r8 - r11) 278 * 279 * Unfortunately there's no way to get gcc to generate the args setup 280 * for the call, and then allow the call itself to be generated by an 281 * inline asm. Because of this, we must do the complete arg setup and 282 * return value handling from within these macros. This is fairly 283 * cumbersome. 284 * 285 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments. 286 * It could be extended to more arguments, but there would be little 287 * to be gained from that. For each number of arguments, there are 288 * two VCALL and CALL variants for void and non-void functions. 289 * 290 * When there is a return value, the invoker of the macro must specify 291 * the return type. The macro then uses sizeof() on that type to 292 * determine whether it's a 32 or 64 bit value and places the return 293 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for 294 * 64-bit). For x86_64 machines, it just returns in %rax regardless of 295 * the return value size. 296 * 297 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments; 298 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments 299 * in low,high order 300 * 301 * Small structures are passed and returned in registers. The macro 302 * calling convention can't directly deal with this, so the wrapper 303 * functions must do it. 304 * 305 * These PVOP_* macros are only defined within this header. This 306 * means that all uses must be wrapped in inline functions. This also 307 * makes sure the incoming and outgoing types are always correct. 308 */ 309 #ifdef CONFIG_X86_32 310 #define PVOP_CALL_ARGS \ 311 unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx; 312 313 #define PVOP_CALL_ARG1(x) "a" ((unsigned long)(x)) 314 #define PVOP_CALL_ARG2(x) "d" ((unsigned long)(x)) 315 #define PVOP_CALL_ARG3(x) "c" ((unsigned long)(x)) 316 317 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \ 318 "=c" (__ecx) 319 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS 320 321 #define PVOP_VCALLEE_CLOBBERS "=a" (__eax), "=d" (__edx) 322 #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS 323 324 #define EXTRA_CLOBBERS 325 #define VEXTRA_CLOBBERS 326 #else /* CONFIG_X86_64 */ 327 /* [re]ax isn't an arg, but the return val */ 328 #define PVOP_CALL_ARGS \ 329 unsigned long __edi = __edi, __esi = __esi, \ 330 __edx = __edx, __ecx = __ecx, __eax = __eax; 331 332 #define PVOP_CALL_ARG1(x) "D" ((unsigned long)(x)) 333 #define PVOP_CALL_ARG2(x) "S" ((unsigned long)(x)) 334 #define PVOP_CALL_ARG3(x) "d" ((unsigned long)(x)) 335 #define PVOP_CALL_ARG4(x) "c" ((unsigned long)(x)) 336 337 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \ 338 "=S" (__esi), "=d" (__edx), \ 339 "=c" (__ecx) 340 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax) 341 342 /* 343 * void functions are still allowed [re]ax for scratch. 344 * 345 * The ZERO_CALL_USED REGS feature may end up zeroing out callee-saved 346 * registers. Make sure we model this with the appropriate clobbers. 347 */ 348 #ifdef CONFIG_ZERO_CALL_USED_REGS 349 #define PVOP_VCALLEE_CLOBBERS "=a" (__eax), PVOP_VCALL_CLOBBERS 350 #else 351 #define PVOP_VCALLEE_CLOBBERS "=a" (__eax) 352 #endif 353 #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS 354 355 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11" 356 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11" 357 #endif /* CONFIG_X86_32 */ 358 359 #ifdef CONFIG_PARAVIRT_DEBUG 360 #define PVOP_TEST_NULL(op) BUG_ON(pv_ops.op == NULL) 361 #else 362 #define PVOP_TEST_NULL(op) ((void)pv_ops.op) 363 #endif 364 365 #define PVOP_RETVAL(rettype) \ 366 ({ unsigned long __mask = ~0UL; \ 367 BUILD_BUG_ON(sizeof(rettype) > sizeof(unsigned long)); \ 368 switch (sizeof(rettype)) { \ 369 case 1: __mask = 0xffUL; break; \ 370 case 2: __mask = 0xffffUL; break; \ 371 case 4: __mask = 0xffffffffUL; break; \ 372 default: break; \ 373 } \ 374 __mask & __eax; \ 375 }) 376 377 /* 378 * Use alternative patching for paravirt calls: 379 * - For replacing an indirect call with a direct one, use the "normal" 380 * ALTERNATIVE() macro with the indirect call as the initial code sequence, 381 * which will be replaced with the related direct call by using the 382 * ALT_FLAG_DIRECT_CALL special case and the "always on" feature. 383 * - In case the replacement is either a direct call or a short code sequence 384 * depending on a feature bit, the ALTERNATIVE_2() macro is being used. 385 * The indirect call is the initial code sequence again, while the special 386 * code sequence is selected with the specified feature bit. In case the 387 * feature is not active, the direct call is used as above via the 388 * ALT_FLAG_DIRECT_CALL special case and the "always on" feature. 389 */ 390 #define ____PVOP_CALL(ret, op, call_clbr, extra_clbr, ...) \ 391 ({ \ 392 PVOP_CALL_ARGS; \ 393 PVOP_TEST_NULL(op); \ 394 asm volatile(ALTERNATIVE(PARAVIRT_CALL, ALT_CALL_INSTR, \ 395 ALT_CALL_ALWAYS) \ 396 : call_clbr, ASM_CALL_CONSTRAINT \ 397 : paravirt_ptr(op), \ 398 ##__VA_ARGS__ \ 399 : "memory", "cc" extra_clbr); \ 400 ret; \ 401 }) 402 403 #define ____PVOP_ALT_CALL(ret, op, alt, cond, call_clbr, \ 404 extra_clbr, ...) \ 405 ({ \ 406 PVOP_CALL_ARGS; \ 407 PVOP_TEST_NULL(op); \ 408 asm volatile(ALTERNATIVE_2(PARAVIRT_CALL, \ 409 ALT_CALL_INSTR, ALT_CALL_ALWAYS, \ 410 alt, cond) \ 411 : call_clbr, ASM_CALL_CONSTRAINT \ 412 : paravirt_ptr(op), \ 413 ##__VA_ARGS__ \ 414 : "memory", "cc" extra_clbr); \ 415 ret; \ 416 }) 417 418 #define __PVOP_CALL(rettype, op, ...) \ 419 ____PVOP_CALL(PVOP_RETVAL(rettype), op, \ 420 PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS, ##__VA_ARGS__) 421 422 #define __PVOP_ALT_CALL(rettype, op, alt, cond, ...) \ 423 ____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op, alt, cond, \ 424 PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS, \ 425 ##__VA_ARGS__) 426 427 #define __PVOP_CALLEESAVE(rettype, op, ...) \ 428 ____PVOP_CALL(PVOP_RETVAL(rettype), op.func, \ 429 PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__) 430 431 #define __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, ...) \ 432 ____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op.func, alt, cond, \ 433 PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__) 434 435 436 #define __PVOP_VCALL(op, ...) \ 437 (void)____PVOP_CALL(, op, PVOP_VCALL_CLOBBERS, \ 438 VEXTRA_CLOBBERS, ##__VA_ARGS__) 439 440 #define __PVOP_ALT_VCALL(op, alt, cond, ...) \ 441 (void)____PVOP_ALT_CALL(, op, alt, cond, \ 442 PVOP_VCALL_CLOBBERS, VEXTRA_CLOBBERS, \ 443 ##__VA_ARGS__) 444 445 #define __PVOP_VCALLEESAVE(op, ...) \ 446 (void)____PVOP_CALL(, op.func, \ 447 PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__) 448 449 #define __PVOP_ALT_VCALLEESAVE(op, alt, cond, ...) \ 450 (void)____PVOP_ALT_CALL(, op.func, alt, cond, \ 451 PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__) 452 453 454 #define PVOP_CALL0(rettype, op) \ 455 __PVOP_CALL(rettype, op) 456 #define PVOP_VCALL0(op) \ 457 __PVOP_VCALL(op) 458 #define PVOP_ALT_CALL0(rettype, op, alt, cond) \ 459 __PVOP_ALT_CALL(rettype, op, alt, cond) 460 #define PVOP_ALT_VCALL0(op, alt, cond) \ 461 __PVOP_ALT_VCALL(op, alt, cond) 462 463 #define PVOP_CALLEE0(rettype, op) \ 464 __PVOP_CALLEESAVE(rettype, op) 465 #define PVOP_VCALLEE0(op) \ 466 __PVOP_VCALLEESAVE(op) 467 #define PVOP_ALT_CALLEE0(rettype, op, alt, cond) \ 468 __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond) 469 #define PVOP_ALT_VCALLEE0(op, alt, cond) \ 470 __PVOP_ALT_VCALLEESAVE(op, alt, cond) 471 472 473 #define PVOP_CALL1(rettype, op, arg1) \ 474 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1)) 475 #define PVOP_VCALL1(op, arg1) \ 476 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1)) 477 #define PVOP_ALT_VCALL1(op, arg1, alt, cond) \ 478 __PVOP_ALT_VCALL(op, alt, cond, PVOP_CALL_ARG1(arg1)) 479 480 #define PVOP_CALLEE1(rettype, op, arg1) \ 481 __PVOP_CALLEESAVE(rettype, op, PVOP_CALL_ARG1(arg1)) 482 #define PVOP_VCALLEE1(op, arg1) \ 483 __PVOP_VCALLEESAVE(op, PVOP_CALL_ARG1(arg1)) 484 #define PVOP_ALT_CALLEE1(rettype, op, arg1, alt, cond) \ 485 __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, PVOP_CALL_ARG1(arg1)) 486 #define PVOP_ALT_VCALLEE1(op, arg1, alt, cond) \ 487 __PVOP_ALT_VCALLEESAVE(op, alt, cond, PVOP_CALL_ARG1(arg1)) 488 489 490 #define PVOP_CALL2(rettype, op, arg1, arg2) \ 491 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2)) 492 #define PVOP_VCALL2(op, arg1, arg2) \ 493 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2)) 494 495 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \ 496 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1), \ 497 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) 498 #define PVOP_VCALL3(op, arg1, arg2, arg3) \ 499 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), \ 500 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) 501 502 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \ 503 __PVOP_CALL(rettype, op, \ 504 PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ 505 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) 506 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \ 507 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ 508 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) 509 510 unsigned long paravirt_ret0(void); 511 #ifdef CONFIG_PARAVIRT_XXL 512 u64 _paravirt_ident_64(u64); 513 unsigned long pv_native_save_fl(void); 514 void pv_native_irq_disable(void); 515 void pv_native_irq_enable(void); 516 unsigned long pv_native_read_cr2(void); 517 #endif 518 519 #define paravirt_nop ((void *)nop_func) 520 521 #endif /* __ASSEMBLY__ */ 522 523 #define ALT_NOT_XEN ALT_NOT(X86_FEATURE_XENPV) 524 525 #endif /* CONFIG_PARAVIRT */ 526 #endif /* _ASM_X86_PARAVIRT_TYPES_H */ 527