Merge branch 'x86/paravirt' into x86/cpu
authorIngo Molnar <mingo@elte.hu>
Tue, 1 Sep 2009 10:13:30 +0000 (12:13 +0200)
committerIngo Molnar <mingo@elte.hu>
Tue, 1 Sep 2009 10:13:30 +0000 (12:13 +0200)
Conflicts:
arch/x86/include/asm/paravirt.h

Manual merge:
arch/x86/include/asm/paravirt_types.h

Merge reason: x86/paravirt conflicts non-trivially with x86/cpu,
              resolve it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
1  2 
arch/x86/include/asm/paravirt.h
arch/x86/include/asm/paravirt_types.h

Simple merge
index 0000000,2b3371b..25402d0
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,720 +1,721 @@@
 -      u64 (*read_msr_amd)(unsigned int msr, int *err);
+ #ifndef _ASM_X86_PARAVIRT_TYPES_H
+ #define _ASM_X86_PARAVIRT_TYPES_H
+ /* Bitmask of what can be clobbered: usually at least eax. */
+ #define CLBR_NONE 0
+ #define CLBR_EAX  (1 << 0)
+ #define CLBR_ECX  (1 << 1)
+ #define CLBR_EDX  (1 << 2)
+ #define CLBR_EDI  (1 << 3)
+ #ifdef CONFIG_X86_32
+ /* CLBR_ANY should match all regs platform has. For i386, that's just it */
+ #define CLBR_ANY  ((1 << 4) - 1)
+ #define CLBR_ARG_REGS (CLBR_EAX | CLBR_EDX | CLBR_ECX)
+ #define CLBR_RET_REG  (CLBR_EAX | CLBR_EDX)
+ #define CLBR_SCRATCH  (0)
+ #else
+ #define CLBR_RAX  CLBR_EAX
+ #define CLBR_RCX  CLBR_ECX
+ #define CLBR_RDX  CLBR_EDX
+ #define CLBR_RDI  CLBR_EDI
+ #define CLBR_RSI  (1 << 4)
+ #define CLBR_R8   (1 << 5)
+ #define CLBR_R9   (1 << 6)
+ #define CLBR_R10  (1 << 7)
+ #define CLBR_R11  (1 << 8)
+ #define CLBR_ANY  ((1 << 9) - 1)
+ #define CLBR_ARG_REGS (CLBR_RDI | CLBR_RSI | CLBR_RDX | \
+                        CLBR_RCX | CLBR_R8 | CLBR_R9)
+ #define CLBR_RET_REG  (CLBR_RAX)
+ #define CLBR_SCRATCH  (CLBR_R10 | CLBR_R11)
+ #endif /* X86_64 */
+ #define CLBR_CALLEE_SAVE ((CLBR_ARG_REGS | CLBR_SCRATCH) & ~CLBR_RET_REG)
+ #ifndef __ASSEMBLY__
+ #include <asm/desc_defs.h>
+ #include <asm/kmap_types.h>
+ struct page;
+ struct thread_struct;
+ struct desc_ptr;
+ struct tss_struct;
+ struct mm_struct;
+ struct desc_struct;
+ struct task_struct;
+ struct cpumask;
+ /*
+  * Wrapper type for pointers to code which uses the non-standard
+  * calling convention.  See PV_CALL_SAVE_REGS_THUNK below.
+  */
+ struct paravirt_callee_save {
+       void *func;
+ };
+ /* general info */
+ struct pv_info {
+       unsigned int kernel_rpl;
+       int shared_kernel_pmd;
+       int paravirt_enabled;
+       const char *name;
+ };
+ struct pv_init_ops {
+       /*
+        * Patch may replace one of the defined code sequences with
+        * arbitrary code, subject to the same register constraints.
+        * This generally means the code is not free to clobber any
+        * registers other than EAX.  The patch function should return
+        * the number of bytes of code generated, as we nop pad the
+        * rest in generic code.
+        */
+       unsigned (*patch)(u8 type, u16 clobber, void *insnbuf,
+                         unsigned long addr, unsigned len);
+       /* Basic arch-specific setup */
+       void (*arch_setup)(void);
+       char *(*memory_setup)(void);
+       void (*post_allocator_init)(void);
+       /* Print a banner to identify the environment */
+       void (*banner)(void);
+ };
+ struct pv_lazy_ops {
+       /* Set deferred update mode, used for batching operations. */
+       void (*enter)(void);
+       void (*leave)(void);
+ };
+ struct pv_time_ops {
+       void (*time_init)(void);
+       /* Set and set time of day */
+       unsigned long (*get_wallclock)(void);
+       int (*set_wallclock)(unsigned long);
+       unsigned long long (*sched_clock)(void);
+       unsigned long (*get_tsc_khz)(void);
+ };
+ struct pv_cpu_ops {
+       /* hooks for various privileged instructions */
+       unsigned long (*get_debugreg)(int regno);
+       void (*set_debugreg)(int regno, unsigned long value);
+       void (*clts)(void);
+       unsigned long (*read_cr0)(void);
+       void (*write_cr0)(unsigned long);
+       unsigned long (*read_cr4_safe)(void);
+       unsigned long (*read_cr4)(void);
+       void (*write_cr4)(unsigned long);
+ #ifdef CONFIG_X86_64
+       unsigned long (*read_cr8)(void);
+       void (*write_cr8)(unsigned long);
+ #endif
+       /* Segment descriptor handling */
+       void (*load_tr_desc)(void);
+       void (*load_gdt)(const struct desc_ptr *);
+       void (*load_idt)(const struct desc_ptr *);
+       void (*store_gdt)(struct desc_ptr *);
+       void (*store_idt)(struct desc_ptr *);
+       void (*set_ldt)(const void *desc, unsigned entries);
+       unsigned long (*store_tr)(void);
+       void (*load_tls)(struct thread_struct *t, unsigned int cpu);
+ #ifdef CONFIG_X86_64
+       void (*load_gs_index)(unsigned int idx);
+ #endif
+       void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum,
+                               const void *desc);
+       void (*write_gdt_entry)(struct desc_struct *,
+                               int entrynum, const void *desc, int size);
+       void (*write_idt_entry)(gate_desc *,
+                               int entrynum, const gate_desc *gate);
+       void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries);
+       void (*free_ldt)(struct desc_struct *ldt, unsigned entries);
+       void (*load_sp0)(struct tss_struct *tss, struct thread_struct *t);
+       void (*set_iopl_mask)(unsigned mask);
+       void (*wbinvd)(void);
+       void (*io_delay)(void);
+       /* cpuid emulation, mostly so that caps bits can be disabled */
+       void (*cpuid)(unsigned int *eax, unsigned int *ebx,
+                     unsigned int *ecx, unsigned int *edx);
+       /* MSR, PMC and TSR operations.
+          err = 0/-EFAULT.  wrmsr returns 0/-EFAULT. */
+       u64 (*read_msr)(unsigned int msr, int *err);
++      int (*rdmsr_regs)(u32 *regs);
+       int (*write_msr)(unsigned int msr, unsigned low, unsigned high);
++      int (*wrmsr_regs)(u32 *regs);
+       u64 (*read_tsc)(void);
+       u64 (*read_pmc)(int counter);
+       unsigned long long (*read_tscp)(unsigned int *aux);
+       /*
+        * Atomically enable interrupts and return to userspace.  This
+        * is only ever used to return to 32-bit processes; in a
+        * 64-bit kernel, it's used for 32-on-64 compat processes, but
+        * never native 64-bit processes.  (Jump, not call.)
+        */
+       void (*irq_enable_sysexit)(void);
+       /*
+        * Switch to usermode gs and return to 64-bit usermode using
+        * sysret.  Only used in 64-bit kernels to return to 64-bit
+        * processes.  Usermode register state, including %rsp, must
+        * already be restored.
+        */
+       void (*usergs_sysret64)(void);
+       /*
+        * Switch to usermode gs and return to 32-bit usermode using
+        * sysret.  Used to return to 32-on-64 compat processes.
+        * Other usermode register state, including %esp, must already
+        * be restored.
+        */
+       void (*usergs_sysret32)(void);
+       /* Normal iret.  Jump to this with the standard iret stack
+          frame set up. */
+       void (*iret)(void);
+       void (*swapgs)(void);
+       void (*start_context_switch)(struct task_struct *prev);
+       void (*end_context_switch)(struct task_struct *next);
+ };
+ struct pv_irq_ops {
+       void (*init_IRQ)(void);
+       /*
+        * Get/set interrupt state.  save_fl and restore_fl are only
+        * expected to use X86_EFLAGS_IF; all other bits
+        * returned from save_fl are undefined, and may be ignored by
+        * restore_fl.
+        *
+        * NOTE: These functions callers expect the callee to preserve
+        * more registers than the standard C calling convention.
+        */
+       struct paravirt_callee_save save_fl;
+       struct paravirt_callee_save restore_fl;
+       struct paravirt_callee_save irq_disable;
+       struct paravirt_callee_save irq_enable;
+       void (*safe_halt)(void);
+       void (*halt)(void);
+ #ifdef CONFIG_X86_64
+       void (*adjust_exception_frame)(void);
+ #endif
+ };
+ struct pv_apic_ops {
+ #ifdef CONFIG_X86_LOCAL_APIC
+       void (*setup_boot_clock)(void);
+       void (*setup_secondary_clock)(void);
+       void (*startup_ipi_hook)(int phys_apicid,
+                                unsigned long start_eip,
+                                unsigned long start_esp);
+ #endif
+ };
+ struct pv_mmu_ops {
+       /*
+        * Called before/after init_mm pagetable setup. setup_start
+        * may reset %cr3, and may pre-install parts of the pagetable;
+        * pagetable setup is expected to preserve any existing
+        * mapping.
+        */
+       void (*pagetable_setup_start)(pgd_t *pgd_base);
+       void (*pagetable_setup_done)(pgd_t *pgd_base);
+       unsigned long (*read_cr2)(void);
+       void (*write_cr2)(unsigned long);
+       unsigned long (*read_cr3)(void);
+       void (*write_cr3)(unsigned long);
+       /*
+        * Hooks for intercepting the creation/use/destruction of an
+        * mm_struct.
+        */
+       void (*activate_mm)(struct mm_struct *prev,
+                           struct mm_struct *next);
+       void (*dup_mmap)(struct mm_struct *oldmm,
+                        struct mm_struct *mm);
+       void (*exit_mmap)(struct mm_struct *mm);
+       /* TLB operations */
+       void (*flush_tlb_user)(void);
+       void (*flush_tlb_kernel)(void);
+       void (*flush_tlb_single)(unsigned long addr);
+       void (*flush_tlb_others)(const struct cpumask *cpus,
+                                struct mm_struct *mm,
+                                unsigned long va);
+       /* Hooks for allocating and freeing a pagetable top-level */
+       int  (*pgd_alloc)(struct mm_struct *mm);
+       void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd);
+       /*
+        * Hooks for allocating/releasing pagetable pages when they're
+        * attached to a pagetable
+        */
+       void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn);
+       void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn);
+       void (*alloc_pmd_clone)(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count);
+       void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn);
+       void (*release_pte)(unsigned long pfn);
+       void (*release_pmd)(unsigned long pfn);
+       void (*release_pud)(unsigned long pfn);
+       /* Pagetable manipulation functions */
+       void (*set_pte)(pte_t *ptep, pte_t pteval);
+       void (*set_pte_at)(struct mm_struct *mm, unsigned long addr,
+                          pte_t *ptep, pte_t pteval);
+       void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
+       void (*pte_update)(struct mm_struct *mm, unsigned long addr,
+                          pte_t *ptep);
+       void (*pte_update_defer)(struct mm_struct *mm,
+                                unsigned long addr, pte_t *ptep);
+       pte_t (*ptep_modify_prot_start)(struct mm_struct *mm, unsigned long addr,
+                                       pte_t *ptep);
+       void (*ptep_modify_prot_commit)(struct mm_struct *mm, unsigned long addr,
+                                       pte_t *ptep, pte_t pte);
+       struct paravirt_callee_save pte_val;
+       struct paravirt_callee_save make_pte;
+       struct paravirt_callee_save pgd_val;
+       struct paravirt_callee_save make_pgd;
+ #if PAGETABLE_LEVELS >= 3
+ #ifdef CONFIG_X86_PAE
+       void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
+       void (*pte_clear)(struct mm_struct *mm, unsigned long addr,
+                         pte_t *ptep);
+       void (*pmd_clear)(pmd_t *pmdp);
+ #endif        /* CONFIG_X86_PAE */
+       void (*set_pud)(pud_t *pudp, pud_t pudval);
+       struct paravirt_callee_save pmd_val;
+       struct paravirt_callee_save make_pmd;
+ #if PAGETABLE_LEVELS == 4
+       struct paravirt_callee_save pud_val;
+       struct paravirt_callee_save make_pud;
+       void (*set_pgd)(pgd_t *pudp, pgd_t pgdval);
+ #endif        /* PAGETABLE_LEVELS == 4 */
+ #endif        /* PAGETABLE_LEVELS >= 3 */
+ #ifdef CONFIG_HIGHPTE
+       void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
+ #endif
+       struct pv_lazy_ops lazy_mode;
+       /* dom0 ops */
+       /* Sometimes the physical address is a pfn, and sometimes its
+          an mfn.  We can tell which is which from the index. */
+       void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx,
+                          phys_addr_t phys, pgprot_t flags);
+ };
+ struct raw_spinlock;
+ struct pv_lock_ops {
+       int (*spin_is_locked)(struct raw_spinlock *lock);
+       int (*spin_is_contended)(struct raw_spinlock *lock);
+       void (*spin_lock)(struct raw_spinlock *lock);
+       void (*spin_lock_flags)(struct raw_spinlock *lock, unsigned long flags);
+       int (*spin_trylock)(struct raw_spinlock *lock);
+       void (*spin_unlock)(struct raw_spinlock *lock);
+ };
+ /* This contains all the paravirt structures: we get a convenient
+  * number for each function using the offset which we use to indicate
+  * what to patch. */
+ struct paravirt_patch_template {
+       struct pv_init_ops pv_init_ops;
+       struct pv_time_ops pv_time_ops;
+       struct pv_cpu_ops pv_cpu_ops;
+       struct pv_irq_ops pv_irq_ops;
+       struct pv_apic_ops pv_apic_ops;
+       struct pv_mmu_ops pv_mmu_ops;
+       struct pv_lock_ops pv_lock_ops;
+ };
+ extern struct pv_info pv_info;
+ extern struct pv_init_ops pv_init_ops;
+ extern struct pv_time_ops pv_time_ops;
+ extern struct pv_cpu_ops pv_cpu_ops;
+ extern struct pv_irq_ops pv_irq_ops;
+ extern struct pv_apic_ops pv_apic_ops;
+ extern struct pv_mmu_ops pv_mmu_ops;
+ extern struct pv_lock_ops pv_lock_ops;
+ #define PARAVIRT_PATCH(x)                                     \
+       (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
+ #define paravirt_type(op)                             \
+       [paravirt_typenum] "i" (PARAVIRT_PATCH(op)),    \
+       [paravirt_opptr] "i" (&(op))
+ #define paravirt_clobber(clobber)             \
+       [paravirt_clobber] "i" (clobber)
+ /*
+  * Generate some code, and mark it as patchable by the
+  * apply_paravirt() alternate instruction patcher.
+  */
+ #define _paravirt_alt(insn_string, type, clobber)     \
+       "771:\n\t" insn_string "\n" "772:\n"            \
+       ".pushsection .parainstructions,\"a\"\n"        \
+       _ASM_ALIGN "\n"                                 \
+       _ASM_PTR " 771b\n"                              \
+       "  .byte " type "\n"                            \
+       "  .byte 772b-771b\n"                           \
+       "  .short " clobber "\n"                        \
+       ".popsection\n"
+ /* Generate patchable code, with the default asm parameters. */
+ #define paravirt_alt(insn_string)                                     \
+       _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
+ /* Simple instruction patching code. */
+ #define DEF_NATIVE(ops, name, code)                                   \
+       extern const char start_##ops##_##name[], end_##ops##_##name[]; \
+       asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
+ unsigned paravirt_patch_nop(void);
+ unsigned paravirt_patch_ident_32(void *insnbuf, unsigned len);
+ unsigned paravirt_patch_ident_64(void *insnbuf, unsigned len);
+ unsigned paravirt_patch_ignore(unsigned len);
+ unsigned paravirt_patch_call(void *insnbuf,
+                            const void *target, u16 tgt_clobbers,
+                            unsigned long addr, u16 site_clobbers,
+                            unsigned len);
+ unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
+                           unsigned long addr, unsigned len);
+ unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
+                               unsigned long addr, unsigned len);
+ unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
+                             const char *start, const char *end);
+ unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
+                     unsigned long addr, unsigned len);
+ int paravirt_disable_iospace(void);
+ /*
+  * This generates an indirect call based on the operation type number.
+  * The type number, computed in PARAVIRT_PATCH, is derived from the
+  * offset into the paravirt_patch_template structure, and can therefore be
+  * freely converted back into a structure offset.
+  */
+ #define PARAVIRT_CALL "call *%c[paravirt_opptr];"
+ /*
+  * These macros are intended to wrap calls through one of the paravirt
+  * ops structs, so that they can be later identified and patched at
+  * runtime.
+  *
+  * Normally, a call to a pv_op function is a simple indirect call:
+  * (pv_op_struct.operations)(args...).
+  *
+  * Unfortunately, this is a relatively slow operation for modern CPUs,
+  * because it cannot necessarily determine what the destination
+  * address is.  In this case, the address is a runtime constant, so at
+  * the very least we can patch the call to e a simple direct call, or
+  * ideally, patch an inline implementation into the callsite.  (Direct
+  * calls are essentially free, because the call and return addresses
+  * are completely predictable.)
+  *
+  * For i386, these macros rely on the standard gcc "regparm(3)" calling
+  * convention, in which the first three arguments are placed in %eax,
+  * %edx, %ecx (in that order), and the remaining arguments are placed
+  * on the stack.  All caller-save registers (eax,edx,ecx) are expected
+  * to be modified (either clobbered or used for return values).
+  * X86_64, on the other hand, already specifies a register-based calling
+  * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
+  * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
+  * special handling for dealing with 4 arguments, unlike i386.
+  * However, x86_64 also have to clobber all caller saved registers, which
+  * unfortunately, are quite a bit (r8 - r11)
+  *
+  * The call instruction itself is marked by placing its start address
+  * and size into the .parainstructions section, so that
+  * apply_paravirt() in arch/i386/kernel/alternative.c can do the
+  * appropriate patching under the control of the backend pv_init_ops
+  * implementation.
+  *
+  * Unfortunately there's no way to get gcc to generate the args setup
+  * for the call, and then allow the call itself to be generated by an
+  * inline asm.  Because of this, we must do the complete arg setup and
+  * return value handling from within these macros.  This is fairly
+  * cumbersome.
+  *
+  * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
+  * It could be extended to more arguments, but there would be little
+  * to be gained from that.  For each number of arguments, there are
+  * the two VCALL and CALL variants for void and non-void functions.
+  *
+  * When there is a return value, the invoker of the macro must specify
+  * the return type.  The macro then uses sizeof() on that type to
+  * determine whether its a 32 or 64 bit value, and places the return
+  * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
+  * 64-bit). For x86_64 machines, it just returns at %rax regardless of
+  * the return value size.
+  *
+  * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
+  * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
+  * in low,high order
+  *
+  * Small structures are passed and returned in registers.  The macro
+  * calling convention can't directly deal with this, so the wrapper
+  * functions must do this.
+  *
+  * These PVOP_* macros are only defined within this header.  This
+  * means that all uses must be wrapped in inline functions.  This also
+  * makes sure the incoming and outgoing types are always correct.
+  */
+ #ifdef CONFIG_X86_32
+ #define PVOP_VCALL_ARGS                               \
+       unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx
+ #define PVOP_CALL_ARGS                        PVOP_VCALL_ARGS
+ #define PVOP_CALL_ARG1(x)             "a" ((unsigned long)(x))
+ #define PVOP_CALL_ARG2(x)             "d" ((unsigned long)(x))
+ #define PVOP_CALL_ARG3(x)             "c" ((unsigned long)(x))
+ #define PVOP_VCALL_CLOBBERS           "=a" (__eax), "=d" (__edx),     \
+                                       "=c" (__ecx)
+ #define PVOP_CALL_CLOBBERS            PVOP_VCALL_CLOBBERS
+ #define PVOP_VCALLEE_CLOBBERS         "=a" (__eax), "=d" (__edx)
+ #define PVOP_CALLEE_CLOBBERS          PVOP_VCALLEE_CLOBBERS
+ #define EXTRA_CLOBBERS
+ #define VEXTRA_CLOBBERS
+ #else  /* CONFIG_X86_64 */
+ #define PVOP_VCALL_ARGS                                       \
+       unsigned long __edi = __edi, __esi = __esi,     \
+               __edx = __edx, __ecx = __ecx
+ #define PVOP_CALL_ARGS                PVOP_VCALL_ARGS, __eax
+ #define PVOP_CALL_ARG1(x)             "D" ((unsigned long)(x))
+ #define PVOP_CALL_ARG2(x)             "S" ((unsigned long)(x))
+ #define PVOP_CALL_ARG3(x)             "d" ((unsigned long)(x))
+ #define PVOP_CALL_ARG4(x)             "c" ((unsigned long)(x))
+ #define PVOP_VCALL_CLOBBERS   "=D" (__edi),                           \
+                               "=S" (__esi), "=d" (__edx),             \
+                               "=c" (__ecx)
+ #define PVOP_CALL_CLOBBERS    PVOP_VCALL_CLOBBERS, "=a" (__eax)
+ #define PVOP_VCALLEE_CLOBBERS "=a" (__eax)
+ #define PVOP_CALLEE_CLOBBERS  PVOP_VCALLEE_CLOBBERS
+ #define EXTRA_CLOBBERS         , "r8", "r9", "r10", "r11"
+ #define VEXTRA_CLOBBERS        , "rax", "r8", "r9", "r10", "r11"
+ #endif        /* CONFIG_X86_32 */
+ #ifdef CONFIG_PARAVIRT_DEBUG
+ #define PVOP_TEST_NULL(op)    BUG_ON(op == NULL)
+ #else
+ #define PVOP_TEST_NULL(op)    ((void)op)
+ #endif
+ #define ____PVOP_CALL(rettype, op, clbr, call_clbr, extra_clbr,               \
+                     pre, post, ...)                                   \
+       ({                                                              \
+               rettype __ret;                                          \
+               PVOP_CALL_ARGS;                                         \
+               PVOP_TEST_NULL(op);                                     \
+               /* This is 32-bit specific, but is okay in 64-bit */    \
+               /* since this condition will never hold */              \
+               if (sizeof(rettype) > sizeof(unsigned long)) {          \
+                       asm volatile(pre                                \
+                                    paravirt_alt(PARAVIRT_CALL)        \
+                                    post                               \
+                                    : call_clbr                        \
+                                    : paravirt_type(op),               \
+                                      paravirt_clobber(clbr),          \
+                                      ##__VA_ARGS__                    \
+                                    : "memory", "cc" extra_clbr);      \
+                       __ret = (rettype)((((u64)__edx) << 32) | __eax); \
+               } else {                                                \
+                       asm volatile(pre                                \
+                                    paravirt_alt(PARAVIRT_CALL)        \
+                                    post                               \
+                                    : call_clbr                        \
+                                    : paravirt_type(op),               \
+                                      paravirt_clobber(clbr),          \
+                                      ##__VA_ARGS__                    \
+                                    : "memory", "cc" extra_clbr);      \
+                       __ret = (rettype)__eax;                         \
+               }                                                       \
+               __ret;                                                  \
+       })
+ #define __PVOP_CALL(rettype, op, pre, post, ...)                      \
+       ____PVOP_CALL(rettype, op, CLBR_ANY, PVOP_CALL_CLOBBERS,        \
+                     EXTRA_CLOBBERS, pre, post, ##__VA_ARGS__)
+ #define __PVOP_CALLEESAVE(rettype, op, pre, post, ...)                        \
+       ____PVOP_CALL(rettype, op.func, CLBR_RET_REG,                   \
+                     PVOP_CALLEE_CLOBBERS, ,                           \
+                     pre, post, ##__VA_ARGS__)
+ #define ____PVOP_VCALL(op, clbr, call_clbr, extra_clbr, pre, post, ...)       \
+       ({                                                              \
+               PVOP_VCALL_ARGS;                                        \
+               PVOP_TEST_NULL(op);                                     \
+               asm volatile(pre                                        \
+                            paravirt_alt(PARAVIRT_CALL)                \
+                            post                                       \
+                            : call_clbr                                \
+                            : paravirt_type(op),                       \
+                              paravirt_clobber(clbr),                  \
+                              ##__VA_ARGS__                            \
+                            : "memory", "cc" extra_clbr);              \
+       })
+ #define __PVOP_VCALL(op, pre, post, ...)                              \
+       ____PVOP_VCALL(op, CLBR_ANY, PVOP_VCALL_CLOBBERS,               \
+                      VEXTRA_CLOBBERS,                                 \
+                      pre, post, ##__VA_ARGS__)
+ #define __PVOP_VCALLEESAVE(rettype, op, pre, post, ...)                       \
+       ____PVOP_CALL(rettype, op.func, CLBR_RET_REG,                   \
+                     PVOP_VCALLEE_CLOBBERS, ,                          \
+                     pre, post, ##__VA_ARGS__)
+ #define PVOP_CALL0(rettype, op)                                               \
+       __PVOP_CALL(rettype, op, "", "")
+ #define PVOP_VCALL0(op)                                                       \
+       __PVOP_VCALL(op, "", "")
+ #define PVOP_CALLEE0(rettype, op)                                     \
+       __PVOP_CALLEESAVE(rettype, op, "", "")
+ #define PVOP_VCALLEE0(op)                                             \
+       __PVOP_VCALLEESAVE(op, "", "")
+ #define PVOP_CALL1(rettype, op, arg1)                                 \
+       __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
+ #define PVOP_VCALL1(op, arg1)                                         \
+       __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1))
+ #define PVOP_CALLEE1(rettype, op, arg1)                                       \
+       __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
+ #define PVOP_VCALLEE1(op, arg1)                                               \
+       __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1))
+ #define PVOP_CALL2(rettype, op, arg1, arg2)                           \
+       __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1),          \
+                   PVOP_CALL_ARG2(arg2))
+ #define PVOP_VCALL2(op, arg1, arg2)                                   \
+       __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1),                  \
+                    PVOP_CALL_ARG2(arg2))
+ #define PVOP_CALLEE2(rettype, op, arg1, arg2)                         \
+       __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1),    \
+                         PVOP_CALL_ARG2(arg2))
+ #define PVOP_VCALLEE2(op, arg1, arg2)                                 \
+       __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1),            \
+                          PVOP_CALL_ARG2(arg2))
+ #define PVOP_CALL3(rettype, op, arg1, arg2, arg3)                     \
+       __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1),          \
+                   PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
+ #define PVOP_VCALL3(op, arg1, arg2, arg3)                             \
+       __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1),                  \
+                    PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
+ /* This is the only difference in x86_64. We can make it much simpler */
+ #ifdef CONFIG_X86_32
+ #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4)                       \
+       __PVOP_CALL(rettype, op,                                        \
+                   "push %[_arg4];", "lea 4(%%esp),%%esp;",            \
+                   PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),         \
+                   PVOP_CALL_ARG3(arg3), [_arg4] "mr" ((u32)(arg4)))
+ #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4)                               \
+       __PVOP_VCALL(op,                                                \
+                   "push %[_arg4];", "lea 4(%%esp),%%esp;",            \
+                   "0" ((u32)(arg1)), "1" ((u32)(arg2)),               \
+                   "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
+ #else
+ #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4)                       \
+       __PVOP_CALL(rettype, op, "", "",                                \
+                   PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),         \
+                   PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
+ #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4)                               \
+       __PVOP_VCALL(op, "", "",                                        \
+                    PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),        \
+                    PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
+ #endif
+ /* Lazy mode for batching updates / context switch */
+ enum paravirt_lazy_mode {
+       PARAVIRT_LAZY_NONE,
+       PARAVIRT_LAZY_MMU,
+       PARAVIRT_LAZY_CPU,
+ };
+ enum paravirt_lazy_mode paravirt_get_lazy_mode(void);
+ void paravirt_start_context_switch(struct task_struct *prev);
+ void paravirt_end_context_switch(struct task_struct *next);
+ void paravirt_enter_lazy_mmu(void);
+ void paravirt_leave_lazy_mmu(void);
+ void _paravirt_nop(void);
+ u32 _paravirt_ident_32(u32);
+ u64 _paravirt_ident_64(u64);
+ #define paravirt_nop  ((void *)_paravirt_nop)
+ /* These all sit in the .parainstructions section to tell us what to patch. */
+ struct paravirt_patch_site {
+       u8 *instr;              /* original instructions */
+       u8 instrtype;           /* type of this instruction */
+       u8 len;                 /* length of original instruction */
+       u16 clobbers;           /* what registers you may clobber */
+ };
+ extern struct paravirt_patch_site __parainstructions[],
+       __parainstructions_end[];
+ #endif        /* __ASSEMBLY__ */
+ #endif        /* _ASM_X86_PARAVIRT_TYPES_H */