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_pte
)(struct mm_struct
*mm
, u32 pfn
);
224 void (*alloc_pmd
)(struct mm_struct
*mm
, u32 pfn
);
225 void (*alloc_pmd_clone
)(u32 pfn
, u32 clonepfn
, u32 start
, u32 count
);
226 void (*alloc_pud
)(struct mm_struct
*mm
, u32 pfn
);
227 void (*release_pte
)(u32 pfn
);
228 void (*release_pmd
)(u32 pfn
);
229 void (*release_pud
)(u32 pfn
);
231 /* Pagetable manipulation functions */
232 void (*set_pte
)(pte_t
*ptep
, pte_t pteval
);
233 void (*set_pte_at
)(struct mm_struct
*mm
, unsigned long addr
,
234 pte_t
*ptep
, pte_t pteval
);
235 void (*set_pmd
)(pmd_t
*pmdp
, pmd_t pmdval
);
236 void (*pte_update
)(struct mm_struct
*mm
, unsigned long addr
,
238 void (*pte_update_defer
)(struct mm_struct
*mm
,
239 unsigned long addr
, pte_t
*ptep
);
241 pteval_t (*pte_val
)(pte_t
);
242 pte_t (*make_pte
)(pteval_t pte
);
244 pgdval_t (*pgd_val
)(pgd_t
);
245 pgd_t (*make_pgd
)(pgdval_t pgd
);
247 #if PAGETABLE_LEVELS >= 3
248 #ifdef CONFIG_X86_PAE
249 void (*set_pte_atomic
)(pte_t
*ptep
, pte_t pteval
);
250 void (*set_pte_present
)(struct mm_struct
*mm
, unsigned long addr
,
251 pte_t
*ptep
, pte_t pte
);
252 void (*pte_clear
)(struct mm_struct
*mm
, unsigned long addr
,
254 void (*pmd_clear
)(pmd_t
*pmdp
);
256 #endif /* CONFIG_X86_PAE */
258 void (*set_pud
)(pud_t
*pudp
, pud_t pudval
);
260 pmdval_t (*pmd_val
)(pmd_t
);
261 pmd_t (*make_pmd
)(pmdval_t pmd
);
263 #if PAGETABLE_LEVELS == 4
264 pudval_t (*pud_val
)(pud_t
);
265 pud_t (*make_pud
)(pudval_t pud
);
267 void (*set_pgd
)(pgd_t
*pudp
, pgd_t pgdval
);
268 #endif /* PAGETABLE_LEVELS == 4 */
269 #endif /* PAGETABLE_LEVELS >= 3 */
271 #ifdef CONFIG_HIGHPTE
272 void *(*kmap_atomic_pte
)(struct page
*page
, enum km_type type
);
275 struct pv_lazy_ops lazy_mode
;
278 /* This contains all the paravirt structures: we get a convenient
279 * number for each function using the offset which we use to indicate
281 struct paravirt_patch_template
{
282 struct pv_init_ops pv_init_ops
;
283 struct pv_time_ops pv_time_ops
;
284 struct pv_cpu_ops pv_cpu_ops
;
285 struct pv_irq_ops pv_irq_ops
;
286 struct pv_apic_ops pv_apic_ops
;
287 struct pv_mmu_ops pv_mmu_ops
;
290 extern struct pv_info pv_info
;
291 extern struct pv_init_ops pv_init_ops
;
292 extern struct pv_time_ops pv_time_ops
;
293 extern struct pv_cpu_ops pv_cpu_ops
;
294 extern struct pv_irq_ops pv_irq_ops
;
295 extern struct pv_apic_ops pv_apic_ops
;
296 extern struct pv_mmu_ops pv_mmu_ops
;
298 #define PARAVIRT_PATCH(x) \
299 (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
301 #define paravirt_type(op) \
302 [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \
303 [paravirt_opptr] "m" (op)
304 #define paravirt_clobber(clobber) \
305 [paravirt_clobber] "i" (clobber)
308 * Generate some code, and mark it as patchable by the
309 * apply_paravirt() alternate instruction patcher.
311 #define _paravirt_alt(insn_string, type, clobber) \
312 "771:\n\t" insn_string "\n" "772:\n" \
313 ".pushsection .parainstructions,\"a\"\n" \
316 " .byte " type "\n" \
317 " .byte 772b-771b\n" \
318 " .short " clobber "\n" \
321 /* Generate patchable code, with the default asm parameters. */
322 #define paravirt_alt(insn_string) \
323 _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
325 /* Simple instruction patching code. */
326 #define DEF_NATIVE(ops, name, code) \
327 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
328 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
330 unsigned paravirt_patch_nop(void);
331 unsigned paravirt_patch_ignore(unsigned len
);
332 unsigned paravirt_patch_call(void *insnbuf
,
333 const void *target
, u16 tgt_clobbers
,
334 unsigned long addr
, u16 site_clobbers
,
336 unsigned paravirt_patch_jmp(void *insnbuf
, const void *target
,
337 unsigned long addr
, unsigned len
);
338 unsigned paravirt_patch_default(u8 type
, u16 clobbers
, void *insnbuf
,
339 unsigned long addr
, unsigned len
);
341 unsigned paravirt_patch_insns(void *insnbuf
, unsigned len
,
342 const char *start
, const char *end
);
344 unsigned native_patch(u8 type
, u16 clobbers
, void *ibuf
,
345 unsigned long addr
, unsigned len
);
347 int paravirt_disable_iospace(void);
350 * This generates an indirect call based on the operation type number.
351 * The type number, computed in PARAVIRT_PATCH, is derived from the
352 * offset into the paravirt_patch_template structure, and can therefore be
353 * freely converted back into a structure offset.
355 #define PARAVIRT_CALL "call *%[paravirt_opptr];"
358 * These macros are intended to wrap calls through one of the paravirt
359 * ops structs, so that they can be later identified and patched at
362 * Normally, a call to a pv_op function is a simple indirect call:
363 * (pv_op_struct.operations)(args...).
365 * Unfortunately, this is a relatively slow operation for modern CPUs,
366 * because it cannot necessarily determine what the destination
367 * address is. In this case, the address is a runtime constant, so at
368 * the very least we can patch the call to e a simple direct call, or
369 * ideally, patch an inline implementation into the callsite. (Direct
370 * calls are essentially free, because the call and return addresses
371 * are completely predictable.)
373 * For i386, these macros rely on the standard gcc "regparm(3)" calling
374 * convention, in which the first three arguments are placed in %eax,
375 * %edx, %ecx (in that order), and the remaining arguments are placed
376 * on the stack. All caller-save registers (eax,edx,ecx) are expected
377 * to be modified (either clobbered or used for return values).
378 * X86_64, on the other hand, already specifies a register-based calling
379 * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
380 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
381 * special handling for dealing with 4 arguments, unlike i386.
382 * However, x86_64 also have to clobber all caller saved registers, which
383 * unfortunately, are quite a bit (r8 - r11)
385 * The call instruction itself is marked by placing its start address
386 * and size into the .parainstructions section, so that
387 * apply_paravirt() in arch/i386/kernel/alternative.c can do the
388 * appropriate patching under the control of the backend pv_init_ops
391 * Unfortunately there's no way to get gcc to generate the args setup
392 * for the call, and then allow the call itself to be generated by an
393 * inline asm. Because of this, we must do the complete arg setup and
394 * return value handling from within these macros. This is fairly
397 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
398 * It could be extended to more arguments, but there would be little
399 * to be gained from that. For each number of arguments, there are
400 * the two VCALL and CALL variants for void and non-void functions.
402 * When there is a return value, the invoker of the macro must specify
403 * the return type. The macro then uses sizeof() on that type to
404 * determine whether its a 32 or 64 bit value, and places the return
405 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
406 * 64-bit). For x86_64 machines, it just returns at %rax regardless of
407 * the return value size.
409 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
410 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
413 * Small structures are passed and returned in registers. The macro
414 * calling convention can't directly deal with this, so the wrapper
415 * functions must do this.
417 * These PVOP_* macros are only defined within this header. This
418 * means that all uses must be wrapped in inline functions. This also
419 * makes sure the incoming and outgoing types are always correct.
422 #define PVOP_VCALL_ARGS unsigned long __eax, __edx, __ecx
423 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS
424 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \
426 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS
427 #define EXTRA_CLOBBERS
428 #define VEXTRA_CLOBBERS
430 #define PVOP_VCALL_ARGS unsigned long __edi, __esi, __edx, __ecx
431 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS, __eax
432 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \
433 "=S" (__esi), "=d" (__edx), \
436 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax)
438 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11"
439 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11"
442 #define __PVOP_CALL(rettype, op, pre, post, ...) \
446 /* This is 32-bit specific, but is okay in 64-bit */ \
447 /* since this condition will never hold */ \
448 if (sizeof(rettype) > sizeof(unsigned long)) { \
450 paravirt_alt(PARAVIRT_CALL) \
452 : PVOP_CALL_CLOBBERS \
453 : paravirt_type(op), \
454 paravirt_clobber(CLBR_ANY), \
456 : "memory", "cc" EXTRA_CLOBBERS); \
457 __ret = (rettype)((((u64)__edx) << 32) | __eax); \
460 paravirt_alt(PARAVIRT_CALL) \
462 : PVOP_CALL_CLOBBERS \
463 : paravirt_type(op), \
464 paravirt_clobber(CLBR_ANY), \
466 : "memory", "cc" EXTRA_CLOBBERS); \
467 __ret = (rettype)__eax; \
471 #define __PVOP_VCALL(op, pre, post, ...) \
475 paravirt_alt(PARAVIRT_CALL) \
477 : PVOP_VCALL_CLOBBERS \
478 : paravirt_type(op), \
479 paravirt_clobber(CLBR_ANY), \
481 : "memory", "cc" VEXTRA_CLOBBERS); \
484 #define PVOP_CALL0(rettype, op) \
485 __PVOP_CALL(rettype, op, "", "")
486 #define PVOP_VCALL0(op) \
487 __PVOP_VCALL(op, "", "")
489 #define PVOP_CALL1(rettype, op, arg1) \
490 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)))
491 #define PVOP_VCALL1(op, arg1) \
492 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)))
494 #define PVOP_CALL2(rettype, op, arg1, arg2) \
495 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
496 "1" ((unsigned long)(arg2)))
497 #define PVOP_VCALL2(op, arg1, arg2) \
498 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
499 "1" ((unsigned long)(arg2)))
501 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
502 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
503 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
504 #define PVOP_VCALL3(op, arg1, arg2, arg3) \
505 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
506 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
508 /* This is the only difference in x86_64. We can make it much simpler */
510 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
511 __PVOP_CALL(rettype, op, \
512 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
513 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
514 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
515 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
517 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
518 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
519 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
521 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
522 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
523 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
524 "3"((unsigned long)(arg4)))
525 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
526 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
527 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
528 "3"((unsigned long)(arg4)))
531 static inline int paravirt_enabled(void)
533 return pv_info
.paravirt_enabled
;
536 static inline void load_sp0(struct tss_struct
*tss
,
537 struct thread_struct
*thread
)
539 PVOP_VCALL2(pv_cpu_ops
.load_sp0
, tss
, thread
);
542 #define ARCH_SETUP pv_init_ops.arch_setup();
543 static inline unsigned long get_wallclock(void)
545 return PVOP_CALL0(unsigned long, pv_time_ops
.get_wallclock
);
548 static inline int set_wallclock(unsigned long nowtime
)
550 return PVOP_CALL1(int, pv_time_ops
.set_wallclock
, nowtime
);
553 static inline void (*choose_time_init(void))(void)
555 return pv_time_ops
.time_init
;
558 /* The paravirtualized CPUID instruction. */
559 static inline void __cpuid(unsigned int *eax
, unsigned int *ebx
,
560 unsigned int *ecx
, unsigned int *edx
)
562 PVOP_VCALL4(pv_cpu_ops
.cpuid
, eax
, ebx
, ecx
, edx
);
566 * These special macros can be used to get or set a debugging register
568 static inline unsigned long paravirt_get_debugreg(int reg
)
570 return PVOP_CALL1(unsigned long, pv_cpu_ops
.get_debugreg
, reg
);
572 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
573 static inline void set_debugreg(unsigned long val
, int reg
)
575 PVOP_VCALL2(pv_cpu_ops
.set_debugreg
, reg
, val
);
578 static inline void clts(void)
580 PVOP_VCALL0(pv_cpu_ops
.clts
);
583 static inline unsigned long read_cr0(void)
585 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr0
);
588 static inline void write_cr0(unsigned long x
)
590 PVOP_VCALL1(pv_cpu_ops
.write_cr0
, x
);
593 static inline unsigned long read_cr2(void)
595 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr2
);
598 static inline void write_cr2(unsigned long x
)
600 PVOP_VCALL1(pv_mmu_ops
.write_cr2
, x
);
603 static inline unsigned long read_cr3(void)
605 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr3
);
608 static inline void write_cr3(unsigned long x
)
610 PVOP_VCALL1(pv_mmu_ops
.write_cr3
, x
);
613 static inline unsigned long read_cr4(void)
615 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4
);
617 static inline unsigned long read_cr4_safe(void)
619 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4_safe
);
622 static inline void write_cr4(unsigned long x
)
624 PVOP_VCALL1(pv_cpu_ops
.write_cr4
, x
);
628 static inline unsigned long read_cr8(void)
630 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr8
);
633 static inline void write_cr8(unsigned long x
)
635 PVOP_VCALL1(pv_cpu_ops
.write_cr8
, x
);
639 static inline void raw_safe_halt(void)
641 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
644 static inline void halt(void)
646 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
649 static inline void wbinvd(void)
651 PVOP_VCALL0(pv_cpu_ops
.wbinvd
);
654 #define get_kernel_rpl() (pv_info.kernel_rpl)
656 static inline u64
paravirt_read_msr(unsigned msr
, int *err
)
658 return PVOP_CALL2(u64
, pv_cpu_ops
.read_msr
, msr
, err
);
660 static inline int paravirt_write_msr(unsigned msr
, unsigned low
, unsigned high
)
662 return PVOP_CALL3(int, pv_cpu_ops
.write_msr
, msr
, low
, high
);
665 /* These should all do BUG_ON(_err), but our headers are too tangled. */
666 #define rdmsr(msr, val1, val2) \
669 u64 _l = paravirt_read_msr(msr, &_err); \
674 #define wrmsr(msr, val1, val2) \
676 paravirt_write_msr(msr, val1, val2); \
679 #define rdmsrl(msr, val) \
682 val = paravirt_read_msr(msr, &_err); \
685 #define wrmsrl(msr, val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
686 #define wrmsr_safe(msr, a, b) paravirt_write_msr(msr, a, b)
688 /* rdmsr with exception handling */
689 #define rdmsr_safe(msr, a, b) \
692 u64 _l = paravirt_read_msr(msr, &_err); \
698 static inline int rdmsrl_safe(unsigned msr
, unsigned long long *p
)
702 *p
= paravirt_read_msr(msr
, &err
);
706 static inline u64
paravirt_read_tsc(void)
708 return PVOP_CALL0(u64
, pv_cpu_ops
.read_tsc
);
711 #define rdtscl(low) \
713 u64 _l = paravirt_read_tsc(); \
717 #define rdtscll(val) (val = paravirt_read_tsc())
719 static inline unsigned long long paravirt_sched_clock(void)
721 return PVOP_CALL0(unsigned long long, pv_time_ops
.sched_clock
);
723 #define calculate_cpu_khz() (pv_time_ops.get_cpu_khz())
725 static inline unsigned long long paravirt_read_pmc(int counter
)
727 return PVOP_CALL1(u64
, pv_cpu_ops
.read_pmc
, counter
);
730 #define rdpmc(counter, low, high) \
732 u64 _l = paravirt_read_pmc(counter); \
737 static inline unsigned long long paravirt_rdtscp(unsigned int *aux
)
739 return PVOP_CALL1(u64
, pv_cpu_ops
.read_tscp
, aux
);
742 #define rdtscp(low, high, aux) \
745 unsigned long __val = paravirt_rdtscp(&__aux); \
746 (low) = (u32)__val; \
747 (high) = (u32)(__val >> 32); \
751 #define rdtscpll(val, aux) \
753 unsigned long __aux; \
754 val = paravirt_rdtscp(&__aux); \
758 static inline void load_TR_desc(void)
760 PVOP_VCALL0(pv_cpu_ops
.load_tr_desc
);
762 static inline void load_gdt(const struct desc_ptr
*dtr
)
764 PVOP_VCALL1(pv_cpu_ops
.load_gdt
, dtr
);
766 static inline void load_idt(const struct desc_ptr
*dtr
)
768 PVOP_VCALL1(pv_cpu_ops
.load_idt
, dtr
);
770 static inline void set_ldt(const void *addr
, unsigned entries
)
772 PVOP_VCALL2(pv_cpu_ops
.set_ldt
, addr
, entries
);
774 static inline void store_gdt(struct desc_ptr
*dtr
)
776 PVOP_VCALL1(pv_cpu_ops
.store_gdt
, dtr
);
778 static inline void store_idt(struct desc_ptr
*dtr
)
780 PVOP_VCALL1(pv_cpu_ops
.store_idt
, dtr
);
782 static inline unsigned long paravirt_store_tr(void)
784 return PVOP_CALL0(unsigned long, pv_cpu_ops
.store_tr
);
786 #define store_tr(tr) ((tr) = paravirt_store_tr())
787 static inline void load_TLS(struct thread_struct
*t
, unsigned cpu
)
789 PVOP_VCALL2(pv_cpu_ops
.load_tls
, t
, cpu
);
792 static inline void write_ldt_entry(struct desc_struct
*dt
, int entry
,
795 PVOP_VCALL3(pv_cpu_ops
.write_ldt_entry
, dt
, entry
, desc
);
798 static inline void write_gdt_entry(struct desc_struct
*dt
, int entry
,
799 void *desc
, int type
)
801 PVOP_VCALL4(pv_cpu_ops
.write_gdt_entry
, dt
, entry
, desc
, type
);
804 static inline void write_idt_entry(gate_desc
*dt
, int entry
, const gate_desc
*g
)
806 PVOP_VCALL3(pv_cpu_ops
.write_idt_entry
, dt
, entry
, g
);
808 static inline void set_iopl_mask(unsigned mask
)
810 PVOP_VCALL1(pv_cpu_ops
.set_iopl_mask
, mask
);
813 /* The paravirtualized I/O functions */
814 static inline void slow_down_io(void)
816 pv_cpu_ops
.io_delay();
817 #ifdef REALLY_SLOW_IO
818 pv_cpu_ops
.io_delay();
819 pv_cpu_ops
.io_delay();
820 pv_cpu_ops
.io_delay();
824 #ifdef CONFIG_X86_LOCAL_APIC
826 * Basic functions accessing APICs.
828 static inline void apic_write(unsigned long reg
, u32 v
)
830 PVOP_VCALL2(pv_apic_ops
.apic_write
, reg
, v
);
833 static inline void apic_write_atomic(unsigned long reg
, u32 v
)
835 PVOP_VCALL2(pv_apic_ops
.apic_write_atomic
, reg
, v
);
838 static inline u32
apic_read(unsigned long reg
)
840 return PVOP_CALL1(unsigned long, pv_apic_ops
.apic_read
, reg
);
843 static inline void setup_boot_clock(void)
845 PVOP_VCALL0(pv_apic_ops
.setup_boot_clock
);
848 static inline void setup_secondary_clock(void)
850 PVOP_VCALL0(pv_apic_ops
.setup_secondary_clock
);
854 static inline void paravirt_post_allocator_init(void)
856 if (pv_init_ops
.post_allocator_init
)
857 (*pv_init_ops
.post_allocator_init
)();
860 static inline void paravirt_pagetable_setup_start(pgd_t
*base
)
862 (*pv_mmu_ops
.pagetable_setup_start
)(base
);
865 static inline void paravirt_pagetable_setup_done(pgd_t
*base
)
867 (*pv_mmu_ops
.pagetable_setup_done
)(base
);
871 static inline void startup_ipi_hook(int phys_apicid
, unsigned long start_eip
,
872 unsigned long start_esp
)
874 PVOP_VCALL3(pv_apic_ops
.startup_ipi_hook
,
875 phys_apicid
, start_eip
, start_esp
);
879 static inline void paravirt_activate_mm(struct mm_struct
*prev
,
880 struct mm_struct
*next
)
882 PVOP_VCALL2(pv_mmu_ops
.activate_mm
, prev
, next
);
885 static inline void arch_dup_mmap(struct mm_struct
*oldmm
,
886 struct mm_struct
*mm
)
888 PVOP_VCALL2(pv_mmu_ops
.dup_mmap
, oldmm
, mm
);
891 static inline void arch_exit_mmap(struct mm_struct
*mm
)
893 PVOP_VCALL1(pv_mmu_ops
.exit_mmap
, mm
);
896 static inline void __flush_tlb(void)
898 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_user
);
900 static inline void __flush_tlb_global(void)
902 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_kernel
);
904 static inline void __flush_tlb_single(unsigned long addr
)
906 PVOP_VCALL1(pv_mmu_ops
.flush_tlb_single
, addr
);
909 static inline void flush_tlb_others(cpumask_t cpumask
, struct mm_struct
*mm
,
912 PVOP_VCALL3(pv_mmu_ops
.flush_tlb_others
, &cpumask
, mm
, va
);
915 static inline void paravirt_alloc_pte(struct mm_struct
*mm
, unsigned pfn
)
917 PVOP_VCALL2(pv_mmu_ops
.alloc_pte
, mm
, pfn
);
919 static inline void paravirt_release_pte(unsigned pfn
)
921 PVOP_VCALL1(pv_mmu_ops
.release_pte
, pfn
);
924 static inline void paravirt_alloc_pmd(struct mm_struct
*mm
, unsigned pfn
)
926 PVOP_VCALL2(pv_mmu_ops
.alloc_pmd
, mm
, pfn
);
929 static inline void paravirt_alloc_pmd_clone(unsigned pfn
, unsigned clonepfn
,
930 unsigned start
, unsigned count
)
932 PVOP_VCALL4(pv_mmu_ops
.alloc_pmd_clone
, pfn
, clonepfn
, start
, count
);
934 static inline void paravirt_release_pmd(unsigned pfn
)
936 PVOP_VCALL1(pv_mmu_ops
.release_pmd
, pfn
);
939 static inline void paravirt_alloc_pud(struct mm_struct
*mm
, unsigned pfn
)
941 PVOP_VCALL2(pv_mmu_ops
.alloc_pud
, mm
, pfn
);
943 static inline void paravirt_release_pud(unsigned pfn
)
945 PVOP_VCALL1(pv_mmu_ops
.release_pud
, pfn
);
948 #ifdef CONFIG_HIGHPTE
949 static inline void *kmap_atomic_pte(struct page
*page
, enum km_type type
)
952 ret
= PVOP_CALL2(unsigned long, pv_mmu_ops
.kmap_atomic_pte
, page
, type
);
957 static inline void pte_update(struct mm_struct
*mm
, unsigned long addr
,
960 PVOP_VCALL3(pv_mmu_ops
.pte_update
, mm
, addr
, ptep
);
963 static inline void pte_update_defer(struct mm_struct
*mm
, unsigned long addr
,
966 PVOP_VCALL3(pv_mmu_ops
.pte_update_defer
, mm
, addr
, ptep
);
969 static inline pte_t
__pte(pteval_t val
)
973 if (sizeof(pteval_t
) > sizeof(long))
974 ret
= PVOP_CALL2(pteval_t
,
976 val
, (u64
)val
>> 32);
978 ret
= PVOP_CALL1(pteval_t
,
982 return (pte_t
) { .pte
= ret
};
985 static inline pteval_t
pte_val(pte_t pte
)
989 if (sizeof(pteval_t
) > sizeof(long))
990 ret
= PVOP_CALL2(pteval_t
, pv_mmu_ops
.pte_val
,
991 pte
.pte
, (u64
)pte
.pte
>> 32);
993 ret
= PVOP_CALL1(pteval_t
, pv_mmu_ops
.pte_val
,
999 static inline pgd_t
__pgd(pgdval_t val
)
1003 if (sizeof(pgdval_t
) > sizeof(long))
1004 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.make_pgd
,
1005 val
, (u64
)val
>> 32);
1007 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.make_pgd
,
1010 return (pgd_t
) { ret
};
1013 static inline pgdval_t
pgd_val(pgd_t pgd
)
1017 if (sizeof(pgdval_t
) > sizeof(long))
1018 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.pgd_val
,
1019 pgd
.pgd
, (u64
)pgd
.pgd
>> 32);
1021 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.pgd_val
,
1027 static inline void set_pte(pte_t
*ptep
, pte_t pte
)
1029 if (sizeof(pteval_t
) > sizeof(long))
1030 PVOP_VCALL3(pv_mmu_ops
.set_pte
, ptep
,
1031 pte
.pte
, (u64
)pte
.pte
>> 32);
1033 PVOP_VCALL2(pv_mmu_ops
.set_pte
, ptep
,
1037 static inline void set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
1038 pte_t
*ptep
, pte_t pte
)
1040 if (sizeof(pteval_t
) > sizeof(long))
1042 pv_mmu_ops
.set_pte_at(mm
, addr
, ptep
, pte
);
1044 PVOP_VCALL4(pv_mmu_ops
.set_pte_at
, mm
, addr
, ptep
, pte
.pte
);
1047 static inline void set_pmd(pmd_t
*pmdp
, pmd_t pmd
)
1049 pmdval_t val
= native_pmd_val(pmd
);
1051 if (sizeof(pmdval_t
) > sizeof(long))
1052 PVOP_VCALL3(pv_mmu_ops
.set_pmd
, pmdp
, val
, (u64
)val
>> 32);
1054 PVOP_VCALL2(pv_mmu_ops
.set_pmd
, pmdp
, val
);
1057 #if PAGETABLE_LEVELS >= 3
1058 static inline pmd_t
__pmd(pmdval_t val
)
1062 if (sizeof(pmdval_t
) > sizeof(long))
1063 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.make_pmd
,
1064 val
, (u64
)val
>> 32);
1066 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.make_pmd
,
1069 return (pmd_t
) { ret
};
1072 static inline pmdval_t
pmd_val(pmd_t pmd
)
1076 if (sizeof(pmdval_t
) > sizeof(long))
1077 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.pmd_val
,
1078 pmd
.pmd
, (u64
)pmd
.pmd
>> 32);
1080 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.pmd_val
,
1086 static inline void set_pud(pud_t
*pudp
, pud_t pud
)
1088 pudval_t val
= native_pud_val(pud
);
1090 if (sizeof(pudval_t
) > sizeof(long))
1091 PVOP_VCALL3(pv_mmu_ops
.set_pud
, pudp
,
1092 val
, (u64
)val
>> 32);
1094 PVOP_VCALL2(pv_mmu_ops
.set_pud
, pudp
,
1097 #if PAGETABLE_LEVELS == 4
1098 static inline pud_t
__pud(pudval_t val
)
1102 if (sizeof(pudval_t
) > sizeof(long))
1103 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.make_pud
,
1104 val
, (u64
)val
>> 32);
1106 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.make_pud
,
1109 return (pud_t
) { ret
};
1112 static inline pudval_t
pud_val(pud_t pud
)
1116 if (sizeof(pudval_t
) > sizeof(long))
1117 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.pud_val
,
1118 pud
.pud
, (u64
)pud
.pud
>> 32);
1120 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.pud_val
,
1126 static inline void set_pgd(pgd_t
*pgdp
, pgd_t pgd
)
1128 pgdval_t val
= native_pgd_val(pgd
);
1130 if (sizeof(pgdval_t
) > sizeof(long))
1131 PVOP_VCALL3(pv_mmu_ops
.set_pgd
, pgdp
,
1132 val
, (u64
)val
>> 32);
1134 PVOP_VCALL2(pv_mmu_ops
.set_pgd
, pgdp
,
1138 static inline void pgd_clear(pgd_t
*pgdp
)
1140 set_pgd(pgdp
, __pgd(0));
1143 static inline void pud_clear(pud_t
*pudp
)
1145 set_pud(pudp
, __pud(0));
1148 #endif /* PAGETABLE_LEVELS == 4 */
1150 #endif /* PAGETABLE_LEVELS >= 3 */
1152 #ifdef CONFIG_X86_PAE
1153 /* Special-case pte-setting operations for PAE, which can't update a
1154 64-bit pte atomically */
1155 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1157 PVOP_VCALL3(pv_mmu_ops
.set_pte_atomic
, ptep
,
1158 pte
.pte
, pte
.pte
>> 32);
1161 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1162 pte_t
*ptep
, pte_t pte
)
1165 pv_mmu_ops
.set_pte_present(mm
, addr
, ptep
, pte
);
1168 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1171 PVOP_VCALL3(pv_mmu_ops
.pte_clear
, mm
, addr
, ptep
);
1174 static inline void pmd_clear(pmd_t
*pmdp
)
1176 PVOP_VCALL1(pv_mmu_ops
.pmd_clear
, pmdp
);
1178 #else /* !CONFIG_X86_PAE */
1179 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1184 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1185 pte_t
*ptep
, pte_t pte
)
1190 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1193 set_pte_at(mm
, addr
, ptep
, __pte(0));
1196 static inline void pmd_clear(pmd_t
*pmdp
)
1198 set_pmd(pmdp
, __pmd(0));
1200 #endif /* CONFIG_X86_PAE */
1202 /* Lazy mode for batching updates / context switch */
1203 enum paravirt_lazy_mode
{
1209 enum paravirt_lazy_mode
paravirt_get_lazy_mode(void);
1210 void paravirt_enter_lazy_cpu(void);
1211 void paravirt_leave_lazy_cpu(void);
1212 void paravirt_enter_lazy_mmu(void);
1213 void paravirt_leave_lazy_mmu(void);
1214 void paravirt_leave_lazy(enum paravirt_lazy_mode mode
);
1216 #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
1217 static inline void arch_enter_lazy_cpu_mode(void)
1219 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.enter
);
1222 static inline void arch_leave_lazy_cpu_mode(void)
1224 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.leave
);
1227 static inline void arch_flush_lazy_cpu_mode(void)
1229 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)) {
1230 arch_leave_lazy_cpu_mode();
1231 arch_enter_lazy_cpu_mode();
1236 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1237 static inline void arch_enter_lazy_mmu_mode(void)
1239 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.enter
);
1242 static inline void arch_leave_lazy_mmu_mode(void)
1244 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.leave
);
1247 static inline void arch_flush_lazy_mmu_mode(void)
1249 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
)) {
1250 arch_leave_lazy_mmu_mode();
1251 arch_enter_lazy_mmu_mode();
1255 void _paravirt_nop(void);
1256 #define paravirt_nop ((void *)_paravirt_nop)
1258 /* These all sit in the .parainstructions section to tell us what to patch. */
1259 struct paravirt_patch_site
{
1260 u8
*instr
; /* original instructions */
1261 u8 instrtype
; /* type of this instruction */
1262 u8 len
; /* length of original instruction */
1263 u16 clobbers
; /* what registers you may clobber */
1266 extern struct paravirt_patch_site __parainstructions
[],
1267 __parainstructions_end
[];
1269 #ifdef CONFIG_X86_32
1270 #define PV_SAVE_REGS "pushl %%ecx; pushl %%edx;"
1271 #define PV_RESTORE_REGS "popl %%edx; popl %%ecx"
1272 #define PV_FLAGS_ARG "0"
1273 #define PV_EXTRA_CLOBBERS
1274 #define PV_VEXTRA_CLOBBERS
1276 /* We save some registers, but all of them, that's too much. We clobber all
1277 * caller saved registers but the argument parameter */
1278 #define PV_SAVE_REGS "pushq %%rdi;"
1279 #define PV_RESTORE_REGS "popq %%rdi;"
1280 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx"
1281 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx"
1282 #define PV_FLAGS_ARG "D"
1285 static inline unsigned long __raw_local_save_flags(void)
1289 asm volatile(paravirt_alt(PV_SAVE_REGS
1293 : paravirt_type(pv_irq_ops
.save_fl
),
1294 paravirt_clobber(CLBR_EAX
)
1295 : "memory", "cc" PV_VEXTRA_CLOBBERS
);
1299 static inline void raw_local_irq_restore(unsigned long f
)
1301 asm volatile(paravirt_alt(PV_SAVE_REGS
1306 paravirt_type(pv_irq_ops
.restore_fl
),
1307 paravirt_clobber(CLBR_EAX
)
1308 : "memory", "cc" PV_EXTRA_CLOBBERS
);
1311 static inline void raw_local_irq_disable(void)
1313 asm volatile(paravirt_alt(PV_SAVE_REGS
1317 : paravirt_type(pv_irq_ops
.irq_disable
),
1318 paravirt_clobber(CLBR_EAX
)
1319 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1322 static inline void raw_local_irq_enable(void)
1324 asm volatile(paravirt_alt(PV_SAVE_REGS
1328 : paravirt_type(pv_irq_ops
.irq_enable
),
1329 paravirt_clobber(CLBR_EAX
)
1330 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1333 static inline unsigned long __raw_local_irq_save(void)
1337 f
= __raw_local_save_flags();
1338 raw_local_irq_disable();
1342 /* Make sure as little as possible of this mess escapes. */
1343 #undef PARAVIRT_CALL
1357 #else /* __ASSEMBLY__ */
1359 #define _PVSITE(ptype, clobbers, ops, word, algn) \
1363 .pushsection .parainstructions,"a"; \
1372 #ifdef CONFIG_X86_64
1373 #define PV_SAVE_REGS pushq %rax; pushq %rdi; pushq %rcx; pushq %rdx
1374 #define PV_RESTORE_REGS popq %rdx; popq %rcx; popq %rdi; popq %rax
1375 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
1376 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1378 #define PV_SAVE_REGS pushl %eax; pushl %edi; pushl %ecx; pushl %edx
1379 #define PV_RESTORE_REGS popl %edx; popl %ecx; popl %edi; popl %eax
1380 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1381 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1384 #define INTERRUPT_RETURN \
1385 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1386 jmp *%cs:pv_cpu_ops+PV_CPU_iret)
1388 #define DISABLE_INTERRUPTS(clobbers) \
1389 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1391 call *%cs:pv_irq_ops+PV_IRQ_irq_disable; \
1394 #define ENABLE_INTERRUPTS(clobbers) \
1395 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1397 call *%cs:pv_irq_ops+PV_IRQ_irq_enable; \
1400 #define ENABLE_INTERRUPTS_SYSCALL_RET \
1401 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_syscall_ret),\
1403 jmp *%cs:pv_cpu_ops+PV_CPU_irq_enable_syscall_ret)
1406 #ifdef CONFIG_X86_32
1407 #define GET_CR0_INTO_EAX \
1408 push %ecx; push %edx; \
1409 call *pv_cpu_ops+PV_CPU_read_cr0; \
1413 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1415 call *pv_cpu_ops+PV_CPU_swapgs; \
1419 #define GET_CR2_INTO_RCX \
1420 call *pv_mmu_ops+PV_MMU_read_cr2; \
1426 #endif /* __ASSEMBLY__ */
1427 #endif /* CONFIG_PARAVIRT */
1428 #endif /* __ASM_PARAVIRT_H */