1 #ifndef __ASM_PARAVIRT_H
2 #define __ASM_PARAVIRT_H
3 /* Various instructions on x86 need to be replaced for
4 * para-virtualization: those hooks are defined here. */
10 /* Bitmask of what can be clobbered: usually at least eax. */
12 #define CLBR_EAX (1 << 0)
13 #define CLBR_ECX (1 << 1)
14 #define CLBR_EDX (1 << 2)
17 #define CLBR_RSI (1 << 3)
18 #define CLBR_RDI (1 << 4)
19 #define CLBR_R8 (1 << 5)
20 #define CLBR_R9 (1 << 6)
21 #define CLBR_R10 (1 << 7)
22 #define CLBR_R11 (1 << 8)
23 #define CLBR_ANY ((1 << 9) - 1)
24 #include <asm/desc_defs.h>
26 /* CLBR_ANY should match all regs platform has. For i386, that's just it */
27 #define CLBR_ANY ((1 << 3) - 1)
31 #include <linux/types.h>
32 #include <linux/cpumask.h>
33 #include <asm/kmap_types.h>
34 #include <asm/desc_defs.h>
45 unsigned int kernel_rpl
;
46 int shared_kernel_pmd
;
53 * Patch may replace one of the defined code sequences with
54 * arbitrary code, subject to the same register constraints.
55 * This generally means the code is not free to clobber any
56 * registers other than EAX. The patch function should return
57 * the number of bytes of code generated, as we nop pad the
58 * rest in generic code.
60 unsigned (*patch
)(u8 type
, u16 clobber
, void *insnbuf
,
61 unsigned long addr
, unsigned len
);
63 /* Basic arch-specific setup */
64 void (*arch_setup
)(void);
65 char *(*memory_setup
)(void);
66 void (*post_allocator_init
)(void);
68 /* Print a banner to identify the environment */
74 /* Set deferred update mode, used for batching operations. */
80 void (*time_init
)(void);
82 /* Set and set time of day */
83 unsigned long (*get_wallclock
)(void);
84 int (*set_wallclock
)(unsigned long);
86 unsigned long long (*sched_clock
)(void);
87 unsigned long (*get_cpu_khz
)(void);
91 /* hooks for various privileged instructions */
92 unsigned long (*get_debugreg
)(int regno
);
93 void (*set_debugreg
)(int regno
, unsigned long value
);
97 unsigned long (*read_cr0
)(void);
98 void (*write_cr0
)(unsigned long);
100 unsigned long (*read_cr4_safe
)(void);
101 unsigned long (*read_cr4
)(void);
102 void (*write_cr4
)(unsigned long);
105 unsigned long (*read_cr8
)(void);
106 void (*write_cr8
)(unsigned long);
109 /* Segment descriptor handling */
110 void (*load_tr_desc
)(void);
111 void (*load_gdt
)(const struct desc_ptr
*);
112 void (*load_idt
)(const struct desc_ptr
*);
113 void (*store_gdt
)(struct desc_ptr
*);
114 void (*store_idt
)(struct desc_ptr
*);
115 void (*set_ldt
)(const void *desc
, unsigned entries
);
116 unsigned long (*store_tr
)(void);
117 void (*load_tls
)(struct thread_struct
*t
, unsigned int cpu
);
118 void (*write_ldt_entry
)(struct desc_struct
*ldt
, int entrynum
,
120 void (*write_gdt_entry
)(struct desc_struct
*,
121 int entrynum
, const void *desc
, int size
);
122 void (*write_idt_entry
)(gate_desc
*,
123 int entrynum
, const gate_desc
*gate
);
124 void (*load_sp0
)(struct tss_struct
*tss
, struct thread_struct
*t
);
126 void (*set_iopl_mask
)(unsigned mask
);
128 void (*wbinvd
)(void);
129 void (*io_delay
)(void);
131 /* cpuid emulation, mostly so that caps bits can be disabled */
132 void (*cpuid
)(unsigned int *eax
, unsigned int *ebx
,
133 unsigned int *ecx
, unsigned int *edx
);
135 /* MSR, PMC and TSR operations.
136 err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
137 u64 (*read_msr
)(unsigned int msr
, int *err
);
138 int (*write_msr
)(unsigned int msr
, unsigned low
, unsigned high
);
140 u64 (*read_tsc
)(void);
141 u64 (*read_pmc
)(int counter
);
142 unsigned long long (*read_tscp
)(unsigned int *aux
);
144 /* These two are jmp to, not actually called. */
145 void (*irq_enable_syscall_ret
)(void);
148 void (*swapgs
)(void);
150 struct pv_lazy_ops lazy_mode
;
154 void (*init_IRQ
)(void);
157 * Get/set interrupt state. save_fl and restore_fl are only
158 * expected to use X86_EFLAGS_IF; all other bits
159 * returned from save_fl are undefined, and may be ignored by
162 unsigned long (*save_fl
)(void);
163 void (*restore_fl
)(unsigned long);
164 void (*irq_disable
)(void);
165 void (*irq_enable
)(void);
166 void (*safe_halt
)(void);
171 #ifdef CONFIG_X86_LOCAL_APIC
173 * Direct APIC operations, principally for VMI. Ideally
174 * these shouldn't be in this interface.
176 void (*apic_write
)(unsigned long reg
, u32 v
);
177 void (*apic_write_atomic
)(unsigned long reg
, u32 v
);
178 u32 (*apic_read
)(unsigned long reg
);
179 void (*setup_boot_clock
)(void);
180 void (*setup_secondary_clock
)(void);
182 void (*startup_ipi_hook
)(int phys_apicid
,
183 unsigned long start_eip
,
184 unsigned long start_esp
);
190 * Called before/after init_mm pagetable setup. setup_start
191 * may reset %cr3, and may pre-install parts of the pagetable;
192 * pagetable setup is expected to preserve any existing
195 void (*pagetable_setup_start
)(pgd_t
*pgd_base
);
196 void (*pagetable_setup_done
)(pgd_t
*pgd_base
);
198 unsigned long (*read_cr2
)(void);
199 void (*write_cr2
)(unsigned long);
201 unsigned long (*read_cr3
)(void);
202 void (*write_cr3
)(unsigned long);
205 * Hooks for intercepting the creation/use/destruction of an
208 void (*activate_mm
)(struct mm_struct
*prev
,
209 struct mm_struct
*next
);
210 void (*dup_mmap
)(struct mm_struct
*oldmm
,
211 struct mm_struct
*mm
);
212 void (*exit_mmap
)(struct mm_struct
*mm
);
216 void (*flush_tlb_user
)(void);
217 void (*flush_tlb_kernel
)(void);
218 void (*flush_tlb_single
)(unsigned long addr
);
219 void (*flush_tlb_others
)(const cpumask_t
*cpus
, struct mm_struct
*mm
,
222 /* Hooks for allocating/releasing pagetable pages */
223 void (*alloc_pt
)(struct mm_struct
*mm
, u32 pfn
);
224 void (*alloc_pd
)(struct mm_struct
*mm
, u32 pfn
);
225 void (*alloc_pd_clone
)(u32 pfn
, u32 clonepfn
, u32 start
, u32 count
);
226 void (*release_pt
)(u32 pfn
);
227 void (*release_pd
)(u32 pfn
);
229 /* Pagetable manipulation functions */
230 void (*set_pte
)(pte_t
*ptep
, pte_t pteval
);
231 void (*set_pte_at
)(struct mm_struct
*mm
, unsigned long addr
,
232 pte_t
*ptep
, pte_t pteval
);
233 void (*set_pmd
)(pmd_t
*pmdp
, pmd_t pmdval
);
234 void (*pte_update
)(struct mm_struct
*mm
, unsigned long addr
,
236 void (*pte_update_defer
)(struct mm_struct
*mm
,
237 unsigned long addr
, pte_t
*ptep
);
239 pteval_t (*pte_val
)(pte_t
);
240 pte_t (*make_pte
)(pteval_t pte
);
242 pgdval_t (*pgd_val
)(pgd_t
);
243 pgd_t (*make_pgd
)(pgdval_t pgd
);
245 #if PAGETABLE_LEVELS >= 3
246 #ifdef CONFIG_X86_PAE
247 void (*set_pte_atomic
)(pte_t
*ptep
, pte_t pteval
);
248 void (*set_pte_present
)(struct mm_struct
*mm
, unsigned long addr
,
249 pte_t
*ptep
, pte_t pte
);
250 void (*pte_clear
)(struct mm_struct
*mm
, unsigned long addr
,
252 void (*pmd_clear
)(pmd_t
*pmdp
);
254 #endif /* CONFIG_X86_PAE */
256 void (*set_pud
)(pud_t
*pudp
, pud_t pudval
);
258 pmdval_t (*pmd_val
)(pmd_t
);
259 pmd_t (*make_pmd
)(pmdval_t pmd
);
261 #if PAGETABLE_LEVELS == 4
262 pudval_t (*pud_val
)(pud_t
);
263 pud_t (*make_pud
)(pudval_t pud
);
265 void (*set_pgd
)(pgd_t
*pudp
, pgd_t pgdval
);
266 #endif /* PAGETABLE_LEVELS == 4 */
267 #endif /* PAGETABLE_LEVELS >= 3 */
269 #ifdef CONFIG_HIGHPTE
270 void *(*kmap_atomic_pte
)(struct page
*page
, enum km_type type
);
273 struct pv_lazy_ops lazy_mode
;
276 /* This contains all the paravirt structures: we get a convenient
277 * number for each function using the offset which we use to indicate
279 struct paravirt_patch_template
{
280 struct pv_init_ops pv_init_ops
;
281 struct pv_time_ops pv_time_ops
;
282 struct pv_cpu_ops pv_cpu_ops
;
283 struct pv_irq_ops pv_irq_ops
;
284 struct pv_apic_ops pv_apic_ops
;
285 struct pv_mmu_ops pv_mmu_ops
;
288 extern struct pv_info pv_info
;
289 extern struct pv_init_ops pv_init_ops
;
290 extern struct pv_time_ops pv_time_ops
;
291 extern struct pv_cpu_ops pv_cpu_ops
;
292 extern struct pv_irq_ops pv_irq_ops
;
293 extern struct pv_apic_ops pv_apic_ops
;
294 extern struct pv_mmu_ops pv_mmu_ops
;
296 #define PARAVIRT_PATCH(x) \
297 (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
299 #define paravirt_type(op) \
300 [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \
301 [paravirt_opptr] "m" (op)
302 #define paravirt_clobber(clobber) \
303 [paravirt_clobber] "i" (clobber)
306 * Generate some code, and mark it as patchable by the
307 * apply_paravirt() alternate instruction patcher.
309 #define _paravirt_alt(insn_string, type, clobber) \
310 "771:\n\t" insn_string "\n" "772:\n" \
311 ".pushsection .parainstructions,\"a\"\n" \
314 " .byte " type "\n" \
315 " .byte 772b-771b\n" \
316 " .short " clobber "\n" \
319 /* Generate patchable code, with the default asm parameters. */
320 #define paravirt_alt(insn_string) \
321 _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
323 /* Simple instruction patching code. */
324 #define DEF_NATIVE(ops, name, code) \
325 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
326 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
328 unsigned paravirt_patch_nop(void);
329 unsigned paravirt_patch_ignore(unsigned len
);
330 unsigned paravirt_patch_call(void *insnbuf
,
331 const void *target
, u16 tgt_clobbers
,
332 unsigned long addr
, u16 site_clobbers
,
334 unsigned paravirt_patch_jmp(void *insnbuf
, const void *target
,
335 unsigned long addr
, unsigned len
);
336 unsigned paravirt_patch_default(u8 type
, u16 clobbers
, void *insnbuf
,
337 unsigned long addr
, unsigned len
);
339 unsigned paravirt_patch_insns(void *insnbuf
, unsigned len
,
340 const char *start
, const char *end
);
342 unsigned native_patch(u8 type
, u16 clobbers
, void *ibuf
,
343 unsigned long addr
, unsigned len
);
345 int paravirt_disable_iospace(void);
348 * This generates an indirect call based on the operation type number.
349 * The type number, computed in PARAVIRT_PATCH, is derived from the
350 * offset into the paravirt_patch_template structure, and can therefore be
351 * freely converted back into a structure offset.
353 #define PARAVIRT_CALL "call *%[paravirt_opptr];"
356 * These macros are intended to wrap calls through one of the paravirt
357 * ops structs, so that they can be later identified and patched at
360 * Normally, a call to a pv_op function is a simple indirect call:
361 * (pv_op_struct.operations)(args...).
363 * Unfortunately, this is a relatively slow operation for modern CPUs,
364 * because it cannot necessarily determine what the destination
365 * address is. In this case, the address is a runtime constant, so at
366 * the very least we can patch the call to e a simple direct call, or
367 * ideally, patch an inline implementation into the callsite. (Direct
368 * calls are essentially free, because the call and return addresses
369 * are completely predictable.)
371 * For i386, these macros rely on the standard gcc "regparm(3)" calling
372 * convention, in which the first three arguments are placed in %eax,
373 * %edx, %ecx (in that order), and the remaining arguments are placed
374 * on the stack. All caller-save registers (eax,edx,ecx) are expected
375 * to be modified (either clobbered or used for return values).
376 * X86_64, on the other hand, already specifies a register-based calling
377 * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
378 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
379 * special handling for dealing with 4 arguments, unlike i386.
380 * However, x86_64 also have to clobber all caller saved registers, which
381 * unfortunately, are quite a bit (r8 - r11)
383 * The call instruction itself is marked by placing its start address
384 * and size into the .parainstructions section, so that
385 * apply_paravirt() in arch/i386/kernel/alternative.c can do the
386 * appropriate patching under the control of the backend pv_init_ops
389 * Unfortunately there's no way to get gcc to generate the args setup
390 * for the call, and then allow the call itself to be generated by an
391 * inline asm. Because of this, we must do the complete arg setup and
392 * return value handling from within these macros. This is fairly
395 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
396 * It could be extended to more arguments, but there would be little
397 * to be gained from that. For each number of arguments, there are
398 * the two VCALL and CALL variants for void and non-void functions.
400 * When there is a return value, the invoker of the macro must specify
401 * the return type. The macro then uses sizeof() on that type to
402 * determine whether its a 32 or 64 bit value, and places the return
403 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
404 * 64-bit). For x86_64 machines, it just returns at %rax regardless of
405 * the return value size.
407 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
408 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
411 * Small structures are passed and returned in registers. The macro
412 * calling convention can't directly deal with this, so the wrapper
413 * functions must do this.
415 * These PVOP_* macros are only defined within this header. This
416 * means that all uses must be wrapped in inline functions. This also
417 * makes sure the incoming and outgoing types are always correct.
420 #define PVOP_VCALL_ARGS unsigned long __eax, __edx, __ecx
421 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS
422 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \
424 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS
425 #define EXTRA_CLOBBERS
426 #define VEXTRA_CLOBBERS
428 #define PVOP_VCALL_ARGS unsigned long __edi, __esi, __edx, __ecx
429 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS, __eax
430 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \
431 "=S" (__esi), "=d" (__edx), \
434 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax)
436 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11"
437 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11"
440 #define __PVOP_CALL(rettype, op, pre, post, ...) \
444 /* This is 32-bit specific, but is okay in 64-bit */ \
445 /* since this condition will never hold */ \
446 if (sizeof(rettype) > sizeof(unsigned long)) { \
448 paravirt_alt(PARAVIRT_CALL) \
450 : PVOP_CALL_CLOBBERS \
451 : paravirt_type(op), \
452 paravirt_clobber(CLBR_ANY), \
454 : "memory", "cc" EXTRA_CLOBBERS); \
455 __ret = (rettype)((((u64)__edx) << 32) | __eax); \
458 paravirt_alt(PARAVIRT_CALL) \
460 : PVOP_CALL_CLOBBERS \
461 : paravirt_type(op), \
462 paravirt_clobber(CLBR_ANY), \
464 : "memory", "cc" EXTRA_CLOBBERS); \
465 __ret = (rettype)__eax; \
469 #define __PVOP_VCALL(op, pre, post, ...) \
473 paravirt_alt(PARAVIRT_CALL) \
475 : PVOP_VCALL_CLOBBERS \
476 : paravirt_type(op), \
477 paravirt_clobber(CLBR_ANY), \
479 : "memory", "cc" VEXTRA_CLOBBERS); \
482 #define PVOP_CALL0(rettype, op) \
483 __PVOP_CALL(rettype, op, "", "")
484 #define PVOP_VCALL0(op) \
485 __PVOP_VCALL(op, "", "")
487 #define PVOP_CALL1(rettype, op, arg1) \
488 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)))
489 #define PVOP_VCALL1(op, arg1) \
490 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)))
492 #define PVOP_CALL2(rettype, op, arg1, arg2) \
493 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
494 "1" ((unsigned long)(arg2)))
495 #define PVOP_VCALL2(op, arg1, arg2) \
496 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
497 "1" ((unsigned long)(arg2)))
499 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
500 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
501 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
502 #define PVOP_VCALL3(op, arg1, arg2, arg3) \
503 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
504 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
506 /* This is the only difference in x86_64. We can make it much simpler */
508 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
509 __PVOP_CALL(rettype, op, \
510 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
511 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
512 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
513 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
515 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
516 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
517 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
519 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
520 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
521 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
522 "3"((unsigned long)(arg4)))
523 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
524 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
525 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
526 "3"((unsigned long)(arg4)))
529 static inline int paravirt_enabled(void)
531 return pv_info
.paravirt_enabled
;
534 static inline void load_sp0(struct tss_struct
*tss
,
535 struct thread_struct
*thread
)
537 PVOP_VCALL2(pv_cpu_ops
.load_sp0
, tss
, thread
);
540 #define ARCH_SETUP pv_init_ops.arch_setup();
541 static inline unsigned long get_wallclock(void)
543 return PVOP_CALL0(unsigned long, pv_time_ops
.get_wallclock
);
546 static inline int set_wallclock(unsigned long nowtime
)
548 return PVOP_CALL1(int, pv_time_ops
.set_wallclock
, nowtime
);
551 static inline void (*choose_time_init(void))(void)
553 return pv_time_ops
.time_init
;
556 /* The paravirtualized CPUID instruction. */
557 static inline void __cpuid(unsigned int *eax
, unsigned int *ebx
,
558 unsigned int *ecx
, unsigned int *edx
)
560 PVOP_VCALL4(pv_cpu_ops
.cpuid
, eax
, ebx
, ecx
, edx
);
564 * These special macros can be used to get or set a debugging register
566 static inline unsigned long paravirt_get_debugreg(int reg
)
568 return PVOP_CALL1(unsigned long, pv_cpu_ops
.get_debugreg
, reg
);
570 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
571 static inline void set_debugreg(unsigned long val
, int reg
)
573 PVOP_VCALL2(pv_cpu_ops
.set_debugreg
, reg
, val
);
576 static inline void clts(void)
578 PVOP_VCALL0(pv_cpu_ops
.clts
);
581 static inline unsigned long read_cr0(void)
583 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr0
);
586 static inline void write_cr0(unsigned long x
)
588 PVOP_VCALL1(pv_cpu_ops
.write_cr0
, x
);
591 static inline unsigned long read_cr2(void)
593 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr2
);
596 static inline void write_cr2(unsigned long x
)
598 PVOP_VCALL1(pv_mmu_ops
.write_cr2
, x
);
601 static inline unsigned long read_cr3(void)
603 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr3
);
606 static inline void write_cr3(unsigned long x
)
608 PVOP_VCALL1(pv_mmu_ops
.write_cr3
, x
);
611 static inline unsigned long read_cr4(void)
613 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4
);
615 static inline unsigned long read_cr4_safe(void)
617 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4_safe
);
620 static inline void write_cr4(unsigned long x
)
622 PVOP_VCALL1(pv_cpu_ops
.write_cr4
, x
);
626 static inline unsigned long read_cr8(void)
628 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr8
);
631 static inline void write_cr8(unsigned long x
)
633 PVOP_VCALL1(pv_cpu_ops
.write_cr8
, x
);
637 static inline void raw_safe_halt(void)
639 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
642 static inline void halt(void)
644 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
647 static inline void wbinvd(void)
649 PVOP_VCALL0(pv_cpu_ops
.wbinvd
);
652 #define get_kernel_rpl() (pv_info.kernel_rpl)
654 static inline u64
paravirt_read_msr(unsigned msr
, int *err
)
656 return PVOP_CALL2(u64
, pv_cpu_ops
.read_msr
, msr
, err
);
658 static inline int paravirt_write_msr(unsigned msr
, unsigned low
, unsigned high
)
660 return PVOP_CALL3(int, pv_cpu_ops
.write_msr
, msr
, low
, high
);
663 /* These should all do BUG_ON(_err), but our headers are too tangled. */
664 #define rdmsr(msr, val1, val2) \
667 u64 _l = paravirt_read_msr(msr, &_err); \
672 #define wrmsr(msr, val1, val2) \
674 paravirt_write_msr(msr, val1, val2); \
677 #define rdmsrl(msr, val) \
680 val = paravirt_read_msr(msr, &_err); \
683 #define wrmsrl(msr, val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
684 #define wrmsr_safe(msr, a, b) paravirt_write_msr(msr, a, b)
686 /* rdmsr with exception handling */
687 #define rdmsr_safe(msr, a, b) \
690 u64 _l = paravirt_read_msr(msr, &_err); \
696 static inline int rdmsrl_safe(unsigned msr
, unsigned long long *p
)
700 *p
= paravirt_read_msr(msr
, &err
);
704 static inline u64
paravirt_read_tsc(void)
706 return PVOP_CALL0(u64
, pv_cpu_ops
.read_tsc
);
709 #define rdtscl(low) \
711 u64 _l = paravirt_read_tsc(); \
715 #define rdtscll(val) (val = paravirt_read_tsc())
717 static inline unsigned long long paravirt_sched_clock(void)
719 return PVOP_CALL0(unsigned long long, pv_time_ops
.sched_clock
);
721 #define calculate_cpu_khz() (pv_time_ops.get_cpu_khz())
723 static inline unsigned long long paravirt_read_pmc(int counter
)
725 return PVOP_CALL1(u64
, pv_cpu_ops
.read_pmc
, counter
);
728 #define rdpmc(counter, low, high) \
730 u64 _l = paravirt_read_pmc(counter); \
735 static inline unsigned long long paravirt_rdtscp(unsigned int *aux
)
737 return PVOP_CALL1(u64
, pv_cpu_ops
.read_tscp
, aux
);
740 #define rdtscp(low, high, aux) \
743 unsigned long __val = paravirt_rdtscp(&__aux); \
744 (low) = (u32)__val; \
745 (high) = (u32)(__val >> 32); \
749 #define rdtscpll(val, aux) \
751 unsigned long __aux; \
752 val = paravirt_rdtscp(&__aux); \
756 static inline void load_TR_desc(void)
758 PVOP_VCALL0(pv_cpu_ops
.load_tr_desc
);
760 static inline void load_gdt(const struct desc_ptr
*dtr
)
762 PVOP_VCALL1(pv_cpu_ops
.load_gdt
, dtr
);
764 static inline void load_idt(const struct desc_ptr
*dtr
)
766 PVOP_VCALL1(pv_cpu_ops
.load_idt
, dtr
);
768 static inline void set_ldt(const void *addr
, unsigned entries
)
770 PVOP_VCALL2(pv_cpu_ops
.set_ldt
, addr
, entries
);
772 static inline void store_gdt(struct desc_ptr
*dtr
)
774 PVOP_VCALL1(pv_cpu_ops
.store_gdt
, dtr
);
776 static inline void store_idt(struct desc_ptr
*dtr
)
778 PVOP_VCALL1(pv_cpu_ops
.store_idt
, dtr
);
780 static inline unsigned long paravirt_store_tr(void)
782 return PVOP_CALL0(unsigned long, pv_cpu_ops
.store_tr
);
784 #define store_tr(tr) ((tr) = paravirt_store_tr())
785 static inline void load_TLS(struct thread_struct
*t
, unsigned cpu
)
787 PVOP_VCALL2(pv_cpu_ops
.load_tls
, t
, cpu
);
790 static inline void write_ldt_entry(struct desc_struct
*dt
, int entry
,
793 PVOP_VCALL3(pv_cpu_ops
.write_ldt_entry
, dt
, entry
, desc
);
796 static inline void write_gdt_entry(struct desc_struct
*dt
, int entry
,
797 void *desc
, int type
)
799 PVOP_VCALL4(pv_cpu_ops
.write_gdt_entry
, dt
, entry
, desc
, type
);
802 static inline void write_idt_entry(gate_desc
*dt
, int entry
, const gate_desc
*g
)
804 PVOP_VCALL3(pv_cpu_ops
.write_idt_entry
, dt
, entry
, g
);
806 static inline void set_iopl_mask(unsigned mask
)
808 PVOP_VCALL1(pv_cpu_ops
.set_iopl_mask
, mask
);
811 /* The paravirtualized I/O functions */
812 static inline void slow_down_io(void)
814 pv_cpu_ops
.io_delay();
815 #ifdef REALLY_SLOW_IO
816 pv_cpu_ops
.io_delay();
817 pv_cpu_ops
.io_delay();
818 pv_cpu_ops
.io_delay();
822 #ifdef CONFIG_X86_LOCAL_APIC
824 * Basic functions accessing APICs.
826 static inline void apic_write(unsigned long reg
, u32 v
)
828 PVOP_VCALL2(pv_apic_ops
.apic_write
, reg
, v
);
831 static inline void apic_write_atomic(unsigned long reg
, u32 v
)
833 PVOP_VCALL2(pv_apic_ops
.apic_write_atomic
, reg
, v
);
836 static inline u32
apic_read(unsigned long reg
)
838 return PVOP_CALL1(unsigned long, pv_apic_ops
.apic_read
, reg
);
841 static inline void setup_boot_clock(void)
843 PVOP_VCALL0(pv_apic_ops
.setup_boot_clock
);
846 static inline void setup_secondary_clock(void)
848 PVOP_VCALL0(pv_apic_ops
.setup_secondary_clock
);
852 static inline void paravirt_post_allocator_init(void)
854 if (pv_init_ops
.post_allocator_init
)
855 (*pv_init_ops
.post_allocator_init
)();
858 static inline void paravirt_pagetable_setup_start(pgd_t
*base
)
860 (*pv_mmu_ops
.pagetable_setup_start
)(base
);
863 static inline void paravirt_pagetable_setup_done(pgd_t
*base
)
865 (*pv_mmu_ops
.pagetable_setup_done
)(base
);
869 static inline void startup_ipi_hook(int phys_apicid
, unsigned long start_eip
,
870 unsigned long start_esp
)
872 PVOP_VCALL3(pv_apic_ops
.startup_ipi_hook
,
873 phys_apicid
, start_eip
, start_esp
);
877 static inline void paravirt_activate_mm(struct mm_struct
*prev
,
878 struct mm_struct
*next
)
880 PVOP_VCALL2(pv_mmu_ops
.activate_mm
, prev
, next
);
883 static inline void arch_dup_mmap(struct mm_struct
*oldmm
,
884 struct mm_struct
*mm
)
886 PVOP_VCALL2(pv_mmu_ops
.dup_mmap
, oldmm
, mm
);
889 static inline void arch_exit_mmap(struct mm_struct
*mm
)
891 PVOP_VCALL1(pv_mmu_ops
.exit_mmap
, mm
);
894 static inline void __flush_tlb(void)
896 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_user
);
898 static inline void __flush_tlb_global(void)
900 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_kernel
);
902 static inline void __flush_tlb_single(unsigned long addr
)
904 PVOP_VCALL1(pv_mmu_ops
.flush_tlb_single
, addr
);
907 static inline void flush_tlb_others(cpumask_t cpumask
, struct mm_struct
*mm
,
910 PVOP_VCALL3(pv_mmu_ops
.flush_tlb_others
, &cpumask
, mm
, va
);
913 static inline void paravirt_alloc_pt(struct mm_struct
*mm
, unsigned pfn
)
915 PVOP_VCALL2(pv_mmu_ops
.alloc_pt
, mm
, pfn
);
917 static inline void paravirt_release_pt(unsigned pfn
)
919 PVOP_VCALL1(pv_mmu_ops
.release_pt
, pfn
);
922 static inline void paravirt_alloc_pd(struct mm_struct
*mm
, unsigned pfn
)
924 PVOP_VCALL2(pv_mmu_ops
.alloc_pd
, mm
, pfn
);
927 static inline void paravirt_alloc_pd_clone(unsigned pfn
, unsigned clonepfn
,
928 unsigned start
, unsigned count
)
930 PVOP_VCALL4(pv_mmu_ops
.alloc_pd_clone
, pfn
, clonepfn
, start
, count
);
932 static inline void paravirt_release_pd(unsigned pfn
)
934 PVOP_VCALL1(pv_mmu_ops
.release_pd
, pfn
);
937 #ifdef CONFIG_HIGHPTE
938 static inline void *kmap_atomic_pte(struct page
*page
, enum km_type type
)
941 ret
= PVOP_CALL2(unsigned long, pv_mmu_ops
.kmap_atomic_pte
, page
, type
);
946 static inline void pte_update(struct mm_struct
*mm
, unsigned long addr
,
949 PVOP_VCALL3(pv_mmu_ops
.pte_update
, mm
, addr
, ptep
);
952 static inline void pte_update_defer(struct mm_struct
*mm
, unsigned long addr
,
955 PVOP_VCALL3(pv_mmu_ops
.pte_update_defer
, mm
, addr
, ptep
);
958 static inline pte_t
__pte(pteval_t val
)
962 if (sizeof(pteval_t
) > sizeof(long))
963 ret
= PVOP_CALL2(pteval_t
,
965 val
, (u64
)val
>> 32);
967 ret
= PVOP_CALL1(pteval_t
,
971 return (pte_t
) { .pte
= ret
};
974 static inline pteval_t
pte_val(pte_t pte
)
978 if (sizeof(pteval_t
) > sizeof(long))
979 ret
= PVOP_CALL2(pteval_t
, pv_mmu_ops
.pte_val
,
980 pte
.pte
, (u64
)pte
.pte
>> 32);
982 ret
= PVOP_CALL1(pteval_t
, pv_mmu_ops
.pte_val
,
988 static inline pgd_t
__pgd(pgdval_t val
)
992 if (sizeof(pgdval_t
) > sizeof(long))
993 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.make_pgd
,
994 val
, (u64
)val
>> 32);
996 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.make_pgd
,
999 return (pgd_t
) { ret
};
1002 static inline pgdval_t
pgd_val(pgd_t pgd
)
1006 if (sizeof(pgdval_t
) > sizeof(long))
1007 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.pgd_val
,
1008 pgd
.pgd
, (u64
)pgd
.pgd
>> 32);
1010 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.pgd_val
,
1016 static inline void set_pte(pte_t
*ptep
, pte_t pte
)
1018 if (sizeof(pteval_t
) > sizeof(long))
1019 PVOP_VCALL3(pv_mmu_ops
.set_pte
, ptep
,
1020 pte
.pte
, (u64
)pte
.pte
>> 32);
1022 PVOP_VCALL2(pv_mmu_ops
.set_pte
, ptep
,
1026 static inline void set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
1027 pte_t
*ptep
, pte_t pte
)
1029 if (sizeof(pteval_t
) > sizeof(long))
1031 pv_mmu_ops
.set_pte_at(mm
, addr
, ptep
, pte
);
1033 PVOP_VCALL4(pv_mmu_ops
.set_pte_at
, mm
, addr
, ptep
, pte
.pte
);
1036 static inline void set_pmd(pmd_t
*pmdp
, pmd_t pmd
)
1038 pmdval_t val
= native_pmd_val(pmd
);
1040 if (sizeof(pmdval_t
) > sizeof(long))
1041 PVOP_VCALL3(pv_mmu_ops
.set_pmd
, pmdp
, val
, (u64
)val
>> 32);
1043 PVOP_VCALL2(pv_mmu_ops
.set_pmd
, pmdp
, val
);
1046 #if PAGETABLE_LEVELS >= 3
1047 static inline pmd_t
__pmd(pmdval_t val
)
1051 if (sizeof(pmdval_t
) > sizeof(long))
1052 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.make_pmd
,
1053 val
, (u64
)val
>> 32);
1055 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.make_pmd
,
1058 return (pmd_t
) { ret
};
1061 static inline pmdval_t
pmd_val(pmd_t pmd
)
1065 if (sizeof(pmdval_t
) > sizeof(long))
1066 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.pmd_val
,
1067 pmd
.pmd
, (u64
)pmd
.pmd
>> 32);
1069 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.pmd_val
,
1075 static inline void set_pud(pud_t
*pudp
, pud_t pud
)
1077 pudval_t val
= native_pud_val(pud
);
1079 if (sizeof(pudval_t
) > sizeof(long))
1080 PVOP_VCALL3(pv_mmu_ops
.set_pud
, pudp
,
1081 val
, (u64
)val
>> 32);
1083 PVOP_VCALL2(pv_mmu_ops
.set_pud
, pudp
,
1086 #if PAGETABLE_LEVELS == 4
1087 static inline pud_t
__pud(pudval_t val
)
1091 if (sizeof(pudval_t
) > sizeof(long))
1092 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.make_pud
,
1093 val
, (u64
)val
>> 32);
1095 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.make_pud
,
1098 return (pud_t
) { ret
};
1101 static inline pudval_t
pud_val(pud_t pud
)
1105 if (sizeof(pudval_t
) > sizeof(long))
1106 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.pud_val
,
1107 pud
.pud
, (u64
)pud
.pud
>> 32);
1109 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.pud_val
,
1115 static inline void set_pgd(pgd_t
*pgdp
, pgd_t pgd
)
1117 pgdval_t val
= native_pgd_val(pgd
);
1119 if (sizeof(pgdval_t
) > sizeof(long))
1120 PVOP_VCALL3(pv_mmu_ops
.set_pgd
, pgdp
,
1121 val
, (u64
)val
>> 32);
1123 PVOP_VCALL2(pv_mmu_ops
.set_pgd
, pgdp
,
1127 static inline void pgd_clear(pgd_t
*pgdp
)
1129 set_pgd(pgdp
, __pgd(0));
1132 static inline void pud_clear(pud_t
*pudp
)
1134 set_pud(pudp
, __pud(0));
1137 #endif /* PAGETABLE_LEVELS == 4 */
1139 #endif /* PAGETABLE_LEVELS >= 3 */
1141 #ifdef CONFIG_X86_PAE
1142 /* Special-case pte-setting operations for PAE, which can't update a
1143 64-bit pte atomically */
1144 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1146 PVOP_VCALL3(pv_mmu_ops
.set_pte_atomic
, ptep
,
1147 pte
.pte
, pte
.pte
>> 32);
1150 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1151 pte_t
*ptep
, pte_t pte
)
1154 pv_mmu_ops
.set_pte_present(mm
, addr
, ptep
, pte
);
1157 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1160 PVOP_VCALL3(pv_mmu_ops
.pte_clear
, mm
, addr
, ptep
);
1163 static inline void pmd_clear(pmd_t
*pmdp
)
1165 PVOP_VCALL1(pv_mmu_ops
.pmd_clear
, pmdp
);
1167 #else /* !CONFIG_X86_PAE */
1168 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1173 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1174 pte_t
*ptep
, pte_t pte
)
1179 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1182 set_pte_at(mm
, addr
, ptep
, __pte(0));
1185 static inline void pmd_clear(pmd_t
*pmdp
)
1187 set_pmd(pmdp
, __pmd(0));
1189 #endif /* CONFIG_X86_PAE */
1191 /* Lazy mode for batching updates / context switch */
1192 enum paravirt_lazy_mode
{
1198 enum paravirt_lazy_mode
paravirt_get_lazy_mode(void);
1199 void paravirt_enter_lazy_cpu(void);
1200 void paravirt_leave_lazy_cpu(void);
1201 void paravirt_enter_lazy_mmu(void);
1202 void paravirt_leave_lazy_mmu(void);
1203 void paravirt_leave_lazy(enum paravirt_lazy_mode mode
);
1205 #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
1206 static inline void arch_enter_lazy_cpu_mode(void)
1208 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.enter
);
1211 static inline void arch_leave_lazy_cpu_mode(void)
1213 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.leave
);
1216 static inline void arch_flush_lazy_cpu_mode(void)
1218 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)) {
1219 arch_leave_lazy_cpu_mode();
1220 arch_enter_lazy_cpu_mode();
1225 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1226 static inline void arch_enter_lazy_mmu_mode(void)
1228 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.enter
);
1231 static inline void arch_leave_lazy_mmu_mode(void)
1233 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.leave
);
1236 static inline void arch_flush_lazy_mmu_mode(void)
1238 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
)) {
1239 arch_leave_lazy_mmu_mode();
1240 arch_enter_lazy_mmu_mode();
1244 void _paravirt_nop(void);
1245 #define paravirt_nop ((void *)_paravirt_nop)
1247 /* These all sit in the .parainstructions section to tell us what to patch. */
1248 struct paravirt_patch_site
{
1249 u8
*instr
; /* original instructions */
1250 u8 instrtype
; /* type of this instruction */
1251 u8 len
; /* length of original instruction */
1252 u16 clobbers
; /* what registers you may clobber */
1255 extern struct paravirt_patch_site __parainstructions
[],
1256 __parainstructions_end
[];
1258 #ifdef CONFIG_X86_32
1259 #define PV_SAVE_REGS "pushl %%ecx; pushl %%edx;"
1260 #define PV_RESTORE_REGS "popl %%edx; popl %%ecx"
1261 #define PV_FLAGS_ARG "0"
1262 #define PV_EXTRA_CLOBBERS
1263 #define PV_VEXTRA_CLOBBERS
1265 /* We save some registers, but all of them, that's too much. We clobber all
1266 * caller saved registers but the argument parameter */
1267 #define PV_SAVE_REGS "pushq %%rdi;"
1268 #define PV_RESTORE_REGS "popq %%rdi;"
1269 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx"
1270 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx"
1271 #define PV_FLAGS_ARG "D"
1274 static inline unsigned long __raw_local_save_flags(void)
1278 asm volatile(paravirt_alt(PV_SAVE_REGS
1282 : paravirt_type(pv_irq_ops
.save_fl
),
1283 paravirt_clobber(CLBR_EAX
)
1284 : "memory", "cc" PV_VEXTRA_CLOBBERS
);
1288 static inline void raw_local_irq_restore(unsigned long f
)
1290 asm volatile(paravirt_alt(PV_SAVE_REGS
1295 paravirt_type(pv_irq_ops
.restore_fl
),
1296 paravirt_clobber(CLBR_EAX
)
1297 : "memory", "cc" PV_EXTRA_CLOBBERS
);
1300 static inline void raw_local_irq_disable(void)
1302 asm volatile(paravirt_alt(PV_SAVE_REGS
1306 : paravirt_type(pv_irq_ops
.irq_disable
),
1307 paravirt_clobber(CLBR_EAX
)
1308 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1311 static inline void raw_local_irq_enable(void)
1313 asm volatile(paravirt_alt(PV_SAVE_REGS
1317 : paravirt_type(pv_irq_ops
.irq_enable
),
1318 paravirt_clobber(CLBR_EAX
)
1319 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1322 static inline unsigned long __raw_local_irq_save(void)
1326 f
= __raw_local_save_flags();
1327 raw_local_irq_disable();
1331 /* Make sure as little as possible of this mess escapes. */
1332 #undef PARAVIRT_CALL
1346 #else /* __ASSEMBLY__ */
1348 #define _PVSITE(ptype, clobbers, ops, word, algn) \
1352 .pushsection .parainstructions,"a"; \
1361 #ifdef CONFIG_X86_64
1362 #define PV_SAVE_REGS pushq %rax; pushq %rdi; pushq %rcx; pushq %rdx
1363 #define PV_RESTORE_REGS popq %rdx; popq %rcx; popq %rdi; popq %rax
1364 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
1365 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1367 #define PV_SAVE_REGS pushl %eax; pushl %edi; pushl %ecx; pushl %edx
1368 #define PV_RESTORE_REGS popl %edx; popl %ecx; popl %edi; popl %eax
1369 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1370 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1373 #define INTERRUPT_RETURN \
1374 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1375 jmp *%cs:pv_cpu_ops+PV_CPU_iret)
1377 #define DISABLE_INTERRUPTS(clobbers) \
1378 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1380 call *%cs:pv_irq_ops+PV_IRQ_irq_disable; \
1383 #define ENABLE_INTERRUPTS(clobbers) \
1384 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1386 call *%cs:pv_irq_ops+PV_IRQ_irq_enable; \
1389 #define ENABLE_INTERRUPTS_SYSCALL_RET \
1390 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_syscall_ret),\
1392 jmp *%cs:pv_cpu_ops+PV_CPU_irq_enable_syscall_ret)
1395 #ifdef CONFIG_X86_32
1396 #define GET_CR0_INTO_EAX \
1397 push %ecx; push %edx; \
1398 call *pv_cpu_ops+PV_CPU_read_cr0; \
1402 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1404 call *pv_cpu_ops+PV_CPU_swapgs; \
1408 #define GET_CR2_INTO_RCX \
1409 call *pv_mmu_ops+PV_MMU_read_cr2; \
1415 #endif /* __ASSEMBLY__ */
1416 #endif /* CONFIG_PARAVIRT */
1417 #endif /* __ASM_PARAVIRT_H */