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 14struct page; 15struct thread_struct; 16struct desc_ptr; 17struct tss_struct; 18struct mm_struct; 19struct desc_struct; 20struct task_struct; 21struct cpumask; 22struct flush_tlb_info; 23struct mmu_gather; 24struct 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 */ 30struct paravirt_callee_save { 31 void *func; 32}; 33 34/* general info */ 35struct 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 44struct 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 52struct 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 void (*wbinvd)(void); 90 91 /* cpuid emulation, mostly so that caps bits can be disabled */ 92 void (*cpuid)(unsigned int *eax, unsigned int *ebx, 93 unsigned int *ecx, unsigned int *edx); 94 95 /* Unsafe MSR operations. These will warn or panic on failure. */ 96 u64 (*read_msr)(unsigned int msr); 97 void (*write_msr)(unsigned int msr, unsigned low, unsigned high); 98 99 /* 100 * Safe MSR operations. 101 * read sets err to 0 or -EIO. write returns 0 or -EIO. 102 */ 103 u64 (*read_msr_safe)(unsigned int msr, int *err); 104 int (*write_msr_safe)(unsigned int msr, unsigned low, unsigned high); 105 106 u64 (*read_pmc)(int counter); 107 108 void (*start_context_switch)(struct task_struct *prev); 109 void (*end_context_switch)(struct task_struct *next); 110#endif 111} __no_randomize_layout; 112 113struct pv_irq_ops { 114#ifdef CONFIG_PARAVIRT_XXL 115 /* 116 * Get/set interrupt state. save_fl is expected to use X86_EFLAGS_IF; 117 * all other bits returned from save_fl are undefined. 118 * 119 * NOTE: These functions callers expect the callee to preserve 120 * more registers than the standard C calling convention. 121 */ 122 struct paravirt_callee_save save_fl; 123 struct paravirt_callee_save irq_disable; 124 struct paravirt_callee_save irq_enable; 125 126 void (*safe_halt)(void); 127 void (*halt)(void); 128#endif 129} __no_randomize_layout; 130 131struct pv_mmu_ops { 132 /* TLB operations */ 133 void (*flush_tlb_user)(void); 134 void (*flush_tlb_kernel)(void); 135 void (*flush_tlb_one_user)(unsigned long addr); 136 void (*flush_tlb_multi)(const struct cpumask *cpus, 137 const struct flush_tlb_info *info); 138 139 void (*tlb_remove_table)(struct mmu_gather *tlb, void *table); 140 141 /* Hook for intercepting the destruction of an mm_struct. */ 142 void (*exit_mmap)(struct mm_struct *mm); 143 void (*notify_page_enc_status_changed)(unsigned long pfn, int npages, bool enc); 144 145#ifdef CONFIG_PARAVIRT_XXL 146 struct paravirt_callee_save read_cr2; 147 void (*write_cr2)(unsigned long); 148 149 unsigned long (*read_cr3)(void); 150 void (*write_cr3)(unsigned long); 151 152 /* Hook for intercepting the creation/use of an mm_struct. */ 153 void (*enter_mmap)(struct mm_struct *mm); 154 155 /* Hooks for allocating and freeing a pagetable top-level */ 156 int (*pgd_alloc)(struct mm_struct *mm); 157 void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd); 158 159 /* 160 * Hooks for allocating/releasing pagetable pages when they're 161 * attached to a pagetable 162 */ 163 void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn); 164 void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn); 165 void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn); 166 void (*alloc_p4d)(struct mm_struct *mm, unsigned long pfn); 167 void (*release_pte)(unsigned long pfn); 168 void (*release_pmd)(unsigned long pfn); 169 void (*release_pud)(unsigned long pfn); 170 void (*release_p4d)(unsigned long pfn); 171 172 /* Pagetable manipulation functions */ 173 void (*set_pte)(pte_t *ptep, pte_t pteval); 174 void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval); 175 176 pte_t (*ptep_modify_prot_start)(struct vm_area_struct *vma, unsigned long addr, 177 pte_t *ptep); 178 void (*ptep_modify_prot_commit)(struct vm_area_struct *vma, unsigned long addr, 179 pte_t *ptep, pte_t pte); 180 181 struct paravirt_callee_save pte_val; 182 struct paravirt_callee_save make_pte; 183 184 struct paravirt_callee_save pgd_val; 185 struct paravirt_callee_save make_pgd; 186 187 void (*set_pud)(pud_t *pudp, pud_t pudval); 188 189 struct paravirt_callee_save pmd_val; 190 struct paravirt_callee_save make_pmd; 191 192 struct paravirt_callee_save pud_val; 193 struct paravirt_callee_save make_pud; 194 195 void (*set_p4d)(p4d_t *p4dp, p4d_t p4dval); 196 197#if CONFIG_PGTABLE_LEVELS >= 5 198 struct paravirt_callee_save p4d_val; 199 struct paravirt_callee_save make_p4d; 200 201 void (*set_pgd)(pgd_t *pgdp, pgd_t pgdval); 202#endif /* CONFIG_PGTABLE_LEVELS >= 5 */ 203 204 struct pv_lazy_ops lazy_mode; 205 206 /* dom0 ops */ 207 208 /* Sometimes the physical address is a pfn, and sometimes its 209 an mfn. We can tell which is which from the index. */ 210 void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx, 211 phys_addr_t phys, pgprot_t flags); 212#endif 213} __no_randomize_layout; 214 215struct arch_spinlock; 216#ifdef CONFIG_SMP 217#include <asm/spinlock_types.h> 218#endif 219 220struct qspinlock; 221 222struct pv_lock_ops { 223 void (*queued_spin_lock_slowpath)(struct qspinlock *lock, u32 val); 224 struct paravirt_callee_save queued_spin_unlock; 225 226 void (*wait)(u8 *ptr, u8 val); 227 void (*kick)(int cpu); 228 229 struct paravirt_callee_save vcpu_is_preempted; 230} __no_randomize_layout; 231 232/* This contains all the paravirt structures: we get a convenient 233 * number for each function using the offset which we use to indicate 234 * what to patch. */ 235struct paravirt_patch_template { 236 struct pv_cpu_ops cpu; 237 struct pv_irq_ops irq; 238 struct pv_mmu_ops mmu; 239 struct pv_lock_ops lock; 240} __no_randomize_layout; 241 242extern struct pv_info pv_info; 243extern struct paravirt_patch_template pv_ops; 244 245#define paravirt_ptr(op) [paravirt_opptr] "m" (pv_ops.op) 246 247int paravirt_disable_iospace(void); 248 249/* This generates an indirect call based on the operation type number. */ 250#define PARAVIRT_CALL \ 251 ANNOTATE_RETPOLINE_SAFE \ 252 "call *%[paravirt_opptr];" 253 254/* 255 * These macros are intended to wrap calls through one of the paravirt 256 * ops structs, so that they can be later identified and patched at 257 * runtime. 258 * 259 * Normally, a call to a pv_op function is a simple indirect call: 260 * (pv_op_struct.operations)(args...). 261 * 262 * Unfortunately, this is a relatively slow operation for modern CPUs, 263 * because it cannot necessarily determine what the destination 264 * address is. In this case, the address is a runtime constant, so at 265 * the very least we can patch the call to a simple direct call, or, 266 * ideally, patch an inline implementation into the callsite. (Direct 267 * calls are essentially free, because the call and return addresses 268 * are completely predictable.) 269 * 270 * For i386, these macros rely on the standard gcc "regparm(3)" calling 271 * convention, in which the first three arguments are placed in %eax, 272 * %edx, %ecx (in that order), and the remaining arguments are placed 273 * on the stack. All caller-save registers (eax,edx,ecx) are expected 274 * to be modified (either clobbered or used for return values). 275 * X86_64, on the other hand, already specifies a register-based calling 276 * conventions, returning at %rax, with parameters going in %rdi, %rsi, 277 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any 278 * special handling for dealing with 4 arguments, unlike i386. 279 * However, x86_64 also has to clobber all caller saved registers, which 280 * unfortunately, are quite a bit (r8 - r11) 281 * 282 * Unfortunately there's no way to get gcc to generate the args setup 283 * for the call, and then allow the call itself to be generated by an 284 * inline asm. Because of this, we must do the complete arg setup and 285 * return value handling from within these macros. This is fairly 286 * cumbersome. 287 * 288 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments. 289 * It could be extended to more arguments, but there would be little 290 * to be gained from that. For each number of arguments, there are 291 * two VCALL and CALL variants for void and non-void functions. 292 * 293 * When there is a return value, the invoker of the macro must specify 294 * the return type. The macro then uses sizeof() on that type to 295 * determine whether it's a 32 or 64 bit value and places the return 296 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for 297 * 64-bit). For x86_64 machines, it just returns in %rax regardless of 298 * the return value size. 299 * 300 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments; 301 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments 302 * in low,high order 303 * 304 * Small structures are passed and returned in registers. The macro 305 * calling convention can't directly deal with this, so the wrapper 306 * functions must do it. 307 * 308 * These PVOP_* macros are only defined within this header. This 309 * means that all uses must be wrapped in inline functions. This also 310 * makes sure the incoming and outgoing types are always correct. 311 */ 312#ifdef CONFIG_X86_32 313#define PVOP_CALL_ARGS \ 314 unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx; 315 316#define PVOP_CALL_ARG1(x) "a" ((unsigned long)(x)) 317#define PVOP_CALL_ARG2(x) "d" ((unsigned long)(x)) 318#define PVOP_CALL_ARG3(x) "c" ((unsigned long)(x)) 319 320#define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \ 321 "=c" (__ecx) 322#define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS 323 324#define PVOP_VCALLEE_CLOBBERS "=a" (__eax), "=d" (__edx) 325#define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS 326 327#define EXTRA_CLOBBERS 328#define VEXTRA_CLOBBERS 329#else /* CONFIG_X86_64 */ 330/* [re]ax isn't an arg, but the return val */ 331#define PVOP_CALL_ARGS \ 332 unsigned long __edi = __edi, __esi = __esi, \ 333 __edx = __edx, __ecx = __ecx, __eax = __eax; 334 335#define PVOP_CALL_ARG1(x) "D" ((unsigned long)(x)) 336#define PVOP_CALL_ARG2(x) "S" ((unsigned long)(x)) 337#define PVOP_CALL_ARG3(x) "d" ((unsigned long)(x)) 338#define PVOP_CALL_ARG4(x) "c" ((unsigned long)(x)) 339 340#define PVOP_VCALL_CLOBBERS "=D" (__edi), \ 341 "=S" (__esi), "=d" (__edx), \ 342 "=c" (__ecx) 343#define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax) 344 345/* 346 * void functions are still allowed [re]ax for scratch. 347 * 348 * The ZERO_CALL_USED REGS feature may end up zeroing out callee-saved 349 * registers. Make sure we model this with the appropriate clobbers. 350 */ 351#ifdef CONFIG_ZERO_CALL_USED_REGS 352#define PVOP_VCALLEE_CLOBBERS "=a" (__eax), PVOP_VCALL_CLOBBERS 353#else 354#define PVOP_VCALLEE_CLOBBERS "=a" (__eax) 355#endif 356#define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS 357 358#define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11" 359#define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11" 360#endif /* CONFIG_X86_32 */ 361 362#ifdef CONFIG_PARAVIRT_DEBUG 363#define PVOP_TEST_NULL(op) BUG_ON(pv_ops.op == NULL) 364#else 365#define PVOP_TEST_NULL(op) ((void)pv_ops.op) 366#endif 367 368#define PVOP_RETVAL(rettype) \ 369 ({ unsigned long __mask = ~0UL; \ 370 BUILD_BUG_ON(sizeof(rettype) > sizeof(unsigned long)); \ 371 switch (sizeof(rettype)) { \ 372 case 1: __mask = 0xffUL; break; \ 373 case 2: __mask = 0xffffUL; break; \ 374 case 4: __mask = 0xffffffffUL; break; \ 375 default: break; \ 376 } \ 377 __mask & __eax; \ 378 }) 379 380/* 381 * Use alternative patching for paravirt calls: 382 * - For replacing an indirect call with a direct one, use the "normal" 383 * ALTERNATIVE() macro with the indirect call as the initial code sequence, 384 * which will be replaced with the related direct call by using the 385 * ALT_FLAG_DIRECT_CALL special case and the "always on" feature. 386 * - In case the replacement is either a direct call or a short code sequence 387 * depending on a feature bit, the ALTERNATIVE_2() macro is being used. 388 * The indirect call is the initial code sequence again, while the special 389 * code sequence is selected with the specified feature bit. In case the 390 * feature is not active, the direct call is used as above via the 391 * ALT_FLAG_DIRECT_CALL special case and the "always on" feature. 392 */ 393#define ____PVOP_CALL(ret, op, call_clbr, extra_clbr, ...) \ 394 ({ \ 395 PVOP_CALL_ARGS; \ 396 PVOP_TEST_NULL(op); \ 397 asm volatile(ALTERNATIVE(PARAVIRT_CALL, ALT_CALL_INSTR, \ 398 ALT_CALL_ALWAYS) \ 399 : call_clbr, ASM_CALL_CONSTRAINT \ 400 : paravirt_ptr(op), \ 401 ##__VA_ARGS__ \ 402 : "memory", "cc" extra_clbr); \ 403 ret; \ 404 }) 405 406#define ____PVOP_ALT_CALL(ret, op, alt, cond, call_clbr, \ 407 extra_clbr, ...) \ 408 ({ \ 409 PVOP_CALL_ARGS; \ 410 PVOP_TEST_NULL(op); \ 411 asm volatile(ALTERNATIVE_2(PARAVIRT_CALL, \ 412 ALT_CALL_INSTR, ALT_CALL_ALWAYS, \ 413 alt, cond) \ 414 : call_clbr, ASM_CALL_CONSTRAINT \ 415 : paravirt_ptr(op), \ 416 ##__VA_ARGS__ \ 417 : "memory", "cc" extra_clbr); \ 418 ret; \ 419 }) 420 421#define __PVOP_CALL(rettype, op, ...) \ 422 ____PVOP_CALL(PVOP_RETVAL(rettype), op, \ 423 PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS, ##__VA_ARGS__) 424 425#define __PVOP_ALT_CALL(rettype, op, alt, cond, ...) \ 426 ____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op, alt, cond, \ 427 PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS, \ 428 ##__VA_ARGS__) 429 430#define __PVOP_CALLEESAVE(rettype, op, ...) \ 431 ____PVOP_CALL(PVOP_RETVAL(rettype), op.func, \ 432 PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__) 433 434#define __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, ...) \ 435 ____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op.func, alt, cond, \ 436 PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__) 437 438 439#define __PVOP_VCALL(op, ...) \ 440 (void)____PVOP_CALL(, op, PVOP_VCALL_CLOBBERS, \ 441 VEXTRA_CLOBBERS, ##__VA_ARGS__) 442 443#define __PVOP_ALT_VCALL(op, alt, cond, ...) \ 444 (void)____PVOP_ALT_CALL(, op, alt, cond, \ 445 PVOP_VCALL_CLOBBERS, VEXTRA_CLOBBERS, \ 446 ##__VA_ARGS__) 447 448#define __PVOP_VCALLEESAVE(op, ...) \ 449 (void)____PVOP_CALL(, op.func, \ 450 PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__) 451 452#define __PVOP_ALT_VCALLEESAVE(op, alt, cond, ...) \ 453 (void)____PVOP_ALT_CALL(, op.func, alt, cond, \ 454 PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__) 455 456 457#define PVOP_CALL0(rettype, op) \ 458 __PVOP_CALL(rettype, op) 459#define PVOP_VCALL0(op) \ 460 __PVOP_VCALL(op) 461#define PVOP_ALT_CALL0(rettype, op, alt, cond) \ 462 __PVOP_ALT_CALL(rettype, op, alt, cond) 463#define PVOP_ALT_VCALL0(op, alt, cond) \ 464 __PVOP_ALT_VCALL(op, alt, cond) 465 466#define PVOP_CALLEE0(rettype, op) \ 467 __PVOP_CALLEESAVE(rettype, op) 468#define PVOP_VCALLEE0(op) \ 469 __PVOP_VCALLEESAVE(op) 470#define PVOP_ALT_CALLEE0(rettype, op, alt, cond) \ 471 __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond) 472#define PVOP_ALT_VCALLEE0(op, alt, cond) \ 473 __PVOP_ALT_VCALLEESAVE(op, alt, cond) 474 475 476#define PVOP_CALL1(rettype, op, arg1) \ 477 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1)) 478#define PVOP_VCALL1(op, arg1) \ 479 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1)) 480#define PVOP_ALT_VCALL1(op, arg1, alt, cond) \ 481 __PVOP_ALT_VCALL(op, alt, cond, PVOP_CALL_ARG1(arg1)) 482 483#define PVOP_CALLEE1(rettype, op, arg1) \ 484 __PVOP_CALLEESAVE(rettype, op, PVOP_CALL_ARG1(arg1)) 485#define PVOP_VCALLEE1(op, arg1) \ 486 __PVOP_VCALLEESAVE(op, PVOP_CALL_ARG1(arg1)) 487#define PVOP_ALT_CALLEE1(rettype, op, arg1, alt, cond) \ 488 __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, PVOP_CALL_ARG1(arg1)) 489#define PVOP_ALT_VCALLEE1(op, arg1, alt, cond) \ 490 __PVOP_ALT_VCALLEESAVE(op, alt, cond, PVOP_CALL_ARG1(arg1)) 491 492 493#define PVOP_CALL2(rettype, op, arg1, arg2) \ 494 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2)) 495#define PVOP_VCALL2(op, arg1, arg2) \ 496 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2)) 497 498#define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \ 499 __PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1), \ 500 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) 501#define PVOP_VCALL3(op, arg1, arg2, arg3) \ 502 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), \ 503 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) 504 505#define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \ 506 __PVOP_CALL(rettype, op, \ 507 PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ 508 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) 509#define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \ 510 __PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ 511 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) 512 513unsigned long paravirt_ret0(void); 514#ifdef CONFIG_PARAVIRT_XXL 515u64 _paravirt_ident_64(u64); 516unsigned long pv_native_save_fl(void); 517void pv_native_irq_disable(void); 518void pv_native_irq_enable(void); 519unsigned long pv_native_read_cr2(void); 520#endif 521 522#define paravirt_nop ((void *)nop_func) 523 524#endif /* __ASSEMBLY__ */ 525 526#define ALT_NOT_XEN ALT_NOT(X86_FEATURE_XENPV) 527 528#endif /* CONFIG_PARAVIRT */ 529#endif /* _ASM_X86_PARAVIRT_TYPES_H */ 530