ia64/pv_ops: implement binary patching optimization for native.
implement binary patching optimization for pv_cpu_ops.
With this optimization, indirect call for pv_cpu_ops methods can be
converted into inline execution or direct call.
Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Signed-off-by: Tony Luck <tony.luck@intel.com>
diff --git a/arch/ia64/include/asm/intrinsics.h b/arch/ia64/include/asm/intrinsics.h
index a3e44a5..fbe2ad9 100644
--- a/arch/ia64/include/asm/intrinsics.h
+++ b/arch/ia64/include/asm/intrinsics.h
@@ -201,7 +201,11 @@
#ifndef __ASSEMBLY__
#if defined(CONFIG_PARAVIRT) && defined(__KERNEL__)
-#define IA64_INTRINSIC_API(name) pv_cpu_ops.name
+#ifdef ASM_SUPPORTED
+# define IA64_INTRINSIC_API(name) paravirt_ ## name
+#else
+# define IA64_INTRINSIC_API(name) pv_cpu_ops.name
+#endif
#define IA64_INTRINSIC_MACRO(name) paravirt_ ## name
#else
#define IA64_INTRINSIC_API(name) ia64_native_ ## name
diff --git a/arch/ia64/include/asm/paravirt.h b/arch/ia64/include/asm/paravirt.h
index fc433f6..2eb0a98 100644
--- a/arch/ia64/include/asm/paravirt.h
+++ b/arch/ia64/include/asm/paravirt.h
@@ -118,6 +118,14 @@
int (*arch_setup_nomca)(void);
void (*post_smp_prepare_boot_cpu)(void);
+
+#ifdef ASM_SUPPORTED
+ unsigned long (*patch_bundle)(void *sbundle, void *ebundle,
+ unsigned long type);
+ unsigned long (*patch_inst)(unsigned long stag, unsigned long etag,
+ unsigned long type);
+#endif
+ void (*patch_branch)(unsigned long tag, unsigned long type);
};
extern struct pv_init_ops pv_init_ops;
diff --git a/arch/ia64/include/asm/paravirt_privop.h b/arch/ia64/include/asm/paravirt_privop.h
index 33c8e55..76d6a69 100644
--- a/arch/ia64/include/asm/paravirt_privop.h
+++ b/arch/ia64/include/asm/paravirt_privop.h
@@ -60,12 +60,18 @@
/* Instructions paravirtualized for performance */
/************************************************/
+#ifndef ASM_SUPPORTED
+#define paravirt_ssm_i() pv_cpu_ops.ssm_i()
+#define paravirt_rsm_i() pv_cpu_ops.rsm_i()
+#define __paravirt_getreg() pv_cpu_ops.getreg()
+#endif
+
/* mask for ia64_native_ssm/rsm() must be constant.("i" constraing).
* static inline function doesn't satisfy it. */
#define paravirt_ssm(mask) \
do { \
if ((mask) == IA64_PSR_I) \
- pv_cpu_ops.ssm_i(); \
+ paravirt_ssm_i(); \
else \
ia64_native_ssm(mask); \
} while (0)
@@ -73,7 +79,7 @@
#define paravirt_rsm(mask) \
do { \
if ((mask) == IA64_PSR_I) \
- pv_cpu_ops.rsm_i(); \
+ paravirt_rsm_i(); \
else \
ia64_native_rsm(mask); \
} while (0)
@@ -86,7 +92,7 @@
if ((reg) == _IA64_REG_IP) \
res = ia64_native_getreg(_IA64_REG_IP); \
else \
- res = pv_cpu_ops.getreg(reg); \
+ res = __paravirt_getreg(reg); \
res; \
})
@@ -121,4 +127,333 @@
IA64_PARAVIRT_ASM_FUNC(work_processed_syscall)
#define ia64_leave_kernel IA64_PARAVIRT_ASM_FUNC(leave_kernel)
+
+#if defined(CONFIG_PARAVIRT)
+/******************************************************************************
+ * binary patching infrastructure
+ */
+#define PARAVIRT_PATCH_TYPE_FC 1
+#define PARAVIRT_PATCH_TYPE_THASH 2
+#define PARAVIRT_PATCH_TYPE_GET_CPUID 3
+#define PARAVIRT_PATCH_TYPE_GET_PMD 4
+#define PARAVIRT_PATCH_TYPE_PTCGA 5
+#define PARAVIRT_PATCH_TYPE_GET_RR 6
+#define PARAVIRT_PATCH_TYPE_SET_RR 7
+#define PARAVIRT_PATCH_TYPE_SET_RR0_TO_RR4 8
+#define PARAVIRT_PATCH_TYPE_SSM_I 9
+#define PARAVIRT_PATCH_TYPE_RSM_I 10
+#define PARAVIRT_PATCH_TYPE_GET_PSR_I 11
+#define PARAVIRT_PATCH_TYPE_INTRIN_LOCAL_IRQ_RESTORE 12
+
+/* PARAVIRT_PATY_TYPE_[GS]ETREG + _IA64_REG_xxx */
+#define PARAVIRT_PATCH_TYPE_GETREG 0x10000000
+#define PARAVIRT_PATCH_TYPE_SETREG 0x20000000
+
+/*
+ * struct task_struct* (*ia64_switch_to)(void* next_task);
+ * void *ia64_leave_syscall;
+ * void *ia64_work_processed_syscall
+ * void *ia64_leave_kernel;
+ */
+
+#define PARAVIRT_PATCH_TYPE_BR_START 0x30000000
+#define PARAVIRT_PATCH_TYPE_BR_SWITCH_TO \
+ (PARAVIRT_PATCH_TYPE_BR_START + 0)
+#define PARAVIRT_PATCH_TYPE_BR_LEAVE_SYSCALL \
+ (PARAVIRT_PATCH_TYPE_BR_START + 1)
+#define PARAVIRT_PATCH_TYPE_BR_WORK_PROCESSED_SYSCALL \
+ (PARAVIRT_PATCH_TYPE_BR_START + 2)
+#define PARAVIRT_PATCH_TYPE_BR_LEAVE_KERNEL \
+ (PARAVIRT_PATCH_TYPE_BR_START + 3)
+
+#ifdef ASM_SUPPORTED
+#include <asm/paravirt_patch.h>
+
+/*
+ * pv_cpu_ops calling stub.
+ * normal function call convension can't be written by gcc
+ * inline assembly.
+ *
+ * from the caller's point of view,
+ * the following registers will be clobbered.
+ * r2, r3
+ * r8-r15
+ * r16, r17
+ * b6, b7
+ * p6-p15
+ * ar.ccv
+ *
+ * from the callee's point of view ,
+ * the following registers can be used.
+ * r2, r3: scratch
+ * r8: scratch, input argument0 and return value
+ * r0-r15: scratch, input argument1-5
+ * b6: return pointer
+ * b7: scratch
+ * p6-p15: scratch
+ * ar.ccv: scratch
+ *
+ * other registers must not be changed. especially
+ * b0: rp: preserved. gcc ignores b0 in clobbered register.
+ * r16: saved gp
+ */
+/* 5 bundles */
+#define __PARAVIRT_BR \
+ ";;\n" \
+ "{ .mlx\n" \
+ "nop 0\n" \
+ "movl r2 = %[op_addr]\n"/* get function pointer address */ \
+ ";;\n" \
+ "}\n" \
+ "1:\n" \
+ "{ .mii\n" \
+ "ld8 r2 = [r2]\n" /* load function descriptor address */ \
+ "mov r17 = ip\n" /* get ip to calc return address */ \
+ "mov r16 = gp\n" /* save gp */ \
+ ";;\n" \
+ "}\n" \
+ "{ .mii\n" \
+ "ld8 r3 = [r2], 8\n" /* load entry address */ \
+ "adds r17 = 1f - 1b, r17\n" /* calculate return address */ \
+ ";;\n" \
+ "mov b7 = r3\n" /* set entry address */ \
+ "}\n" \
+ "{ .mib\n" \
+ "ld8 gp = [r2]\n" /* load gp value */ \
+ "mov b6 = r17\n" /* set return address */ \
+ "br.cond.sptk.few b7\n" /* intrinsics are very short isns */ \
+ "}\n" \
+ "1:\n" \
+ "{ .mii\n" \
+ "mov gp = r16\n" /* restore gp value */ \
+ "nop 0\n" \
+ "nop 0\n" \
+ ";;\n" \
+ "}\n"
+
+#define PARAVIRT_OP(op) \
+ [op_addr] "i"(&pv_cpu_ops.op)
+
+#define PARAVIRT_TYPE(type) \
+ PARAVIRT_PATCH_TYPE_ ## type
+
+#define PARAVIRT_REG_CLOBBERS0 \
+ "r2", "r3", /*"r8",*/ "r9", "r10", "r11", "r14", \
+ "r15", "r16", "r17"
+
+#define PARAVIRT_REG_CLOBBERS1 \
+ "r2","r3", /*"r8",*/ "r9", "r10", "r11", "r14", \
+ "r15", "r16", "r17"
+
+#define PARAVIRT_REG_CLOBBERS2 \
+ "r2", "r3", /*"r8", "r9",*/ "r10", "r11", "r14", \
+ "r15", "r16", "r17"
+
+#define PARAVIRT_REG_CLOBBERS5 \
+ "r2", "r3", /*"r8", "r9", "r10", "r11", "r14",*/ \
+ "r15", "r16", "r17"
+
+#define PARAVIRT_BR_CLOBBERS \
+ "b6", "b7"
+
+#define PARAVIRT_PR_CLOBBERS \
+ "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15"
+
+#define PARAVIRT_AR_CLOBBERS \
+ "ar.ccv"
+
+#define PARAVIRT_CLOBBERS0 \
+ PARAVIRT_REG_CLOBBERS0, \
+ PARAVIRT_BR_CLOBBERS, \
+ PARAVIRT_PR_CLOBBERS, \
+ PARAVIRT_AR_CLOBBERS, \
+ "memory"
+
+#define PARAVIRT_CLOBBERS1 \
+ PARAVIRT_REG_CLOBBERS1, \
+ PARAVIRT_BR_CLOBBERS, \
+ PARAVIRT_PR_CLOBBERS, \
+ PARAVIRT_AR_CLOBBERS, \
+ "memory"
+
+#define PARAVIRT_CLOBBERS2 \
+ PARAVIRT_REG_CLOBBERS2, \
+ PARAVIRT_BR_CLOBBERS, \
+ PARAVIRT_PR_CLOBBERS, \
+ PARAVIRT_AR_CLOBBERS, \
+ "memory"
+
+#define PARAVIRT_CLOBBERS5 \
+ PARAVIRT_REG_CLOBBERS5, \
+ PARAVIRT_BR_CLOBBERS, \
+ PARAVIRT_PR_CLOBBERS, \
+ PARAVIRT_AR_CLOBBERS, \
+ "memory"
+
+#define PARAVIRT_BR0(op, type) \
+ register unsigned long ia64_clobber asm ("r8"); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(type)) \
+ : "=r"(ia64_clobber) \
+ : PARAVIRT_OP(op) \
+ : PARAVIRT_CLOBBERS0)
+
+#define PARAVIRT_BR0_RET(op, type) \
+ register unsigned long ia64_intri_res asm ("r8"); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(type)) \
+ : "=r"(ia64_intri_res) \
+ : PARAVIRT_OP(op) \
+ : PARAVIRT_CLOBBERS0)
+
+#define PARAVIRT_BR1(op, type, arg1) \
+ register unsigned long __##arg1 asm ("r8") = arg1; \
+ register unsigned long ia64_clobber asm ("r8"); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(type)) \
+ : "=r"(ia64_clobber) \
+ : PARAVIRT_OP(op), "0"(__##arg1) \
+ : PARAVIRT_CLOBBERS1)
+
+#define PARAVIRT_BR1_RET(op, type, arg1) \
+ register unsigned long ia64_intri_res asm ("r8"); \
+ register unsigned long __##arg1 asm ("r8") = arg1; \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(type)) \
+ : "=r"(ia64_intri_res) \
+ : PARAVIRT_OP(op), "0"(__##arg1) \
+ : PARAVIRT_CLOBBERS1)
+
+#define PARAVIRT_BR2(op, type, arg1, arg2) \
+ register unsigned long __##arg1 asm ("r8") = arg1; \
+ register unsigned long __##arg2 asm ("r9") = arg2; \
+ register unsigned long ia64_clobber1 asm ("r8"); \
+ register unsigned long ia64_clobber2 asm ("r9"); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(type)) \
+ : "=r"(ia64_clobber1), "=r"(ia64_clobber2) \
+ : PARAVIRT_OP(op), "0"(__##arg1), "1"(__##arg2) \
+ : PARAVIRT_CLOBBERS2)
+
+
+#define PARAVIRT_DEFINE_CPU_OP0(op, type) \
+ static inline void \
+ paravirt_ ## op (void) \
+ { \
+ PARAVIRT_BR0(op, type); \
+ }
+
+#define PARAVIRT_DEFINE_CPU_OP0_RET(op, type) \
+ static inline unsigned long \
+ paravirt_ ## op (void) \
+ { \
+ PARAVIRT_BR0_RET(op, type); \
+ return ia64_intri_res; \
+ }
+
+#define PARAVIRT_DEFINE_CPU_OP1(op, type) \
+ static inline void \
+ paravirt_ ## op (unsigned long arg1) \
+ { \
+ PARAVIRT_BR1(op, type, arg1); \
+ }
+
+#define PARAVIRT_DEFINE_CPU_OP1_RET(op, type) \
+ static inline unsigned long \
+ paravirt_ ## op (unsigned long arg1) \
+ { \
+ PARAVIRT_BR1_RET(op, type, arg1); \
+ return ia64_intri_res; \
+ }
+
+#define PARAVIRT_DEFINE_CPU_OP2(op, type) \
+ static inline void \
+ paravirt_ ## op (unsigned long arg1, \
+ unsigned long arg2) \
+ { \
+ PARAVIRT_BR2(op, type, arg1, arg2); \
+ }
+
+
+PARAVIRT_DEFINE_CPU_OP1(fc, FC);
+PARAVIRT_DEFINE_CPU_OP1_RET(thash, THASH)
+PARAVIRT_DEFINE_CPU_OP1_RET(get_cpuid, GET_CPUID)
+PARAVIRT_DEFINE_CPU_OP1_RET(get_pmd, GET_PMD)
+PARAVIRT_DEFINE_CPU_OP2(ptcga, PTCGA)
+PARAVIRT_DEFINE_CPU_OP1_RET(get_rr, GET_RR)
+PARAVIRT_DEFINE_CPU_OP2(set_rr, SET_RR)
+PARAVIRT_DEFINE_CPU_OP0(ssm_i, SSM_I)
+PARAVIRT_DEFINE_CPU_OP0(rsm_i, RSM_I)
+PARAVIRT_DEFINE_CPU_OP0_RET(get_psr_i, GET_PSR_I)
+PARAVIRT_DEFINE_CPU_OP1(intrin_local_irq_restore, INTRIN_LOCAL_IRQ_RESTORE)
+
+static inline void
+paravirt_set_rr0_to_rr4(unsigned long val0, unsigned long val1,
+ unsigned long val2, unsigned long val3,
+ unsigned long val4)
+{
+ register unsigned long __val0 asm ("r8") = val0;
+ register unsigned long __val1 asm ("r9") = val1;
+ register unsigned long __val2 asm ("r10") = val2;
+ register unsigned long __val3 asm ("r11") = val3;
+ register unsigned long __val4 asm ("r14") = val4;
+
+ register unsigned long ia64_clobber0 asm ("r8");
+ register unsigned long ia64_clobber1 asm ("r9");
+ register unsigned long ia64_clobber2 asm ("r10");
+ register unsigned long ia64_clobber3 asm ("r11");
+ register unsigned long ia64_clobber4 asm ("r14");
+
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR,
+ PARAVIRT_TYPE(SET_RR0_TO_RR4))
+ : "=r"(ia64_clobber0),
+ "=r"(ia64_clobber1),
+ "=r"(ia64_clobber2),
+ "=r"(ia64_clobber3),
+ "=r"(ia64_clobber4)
+ : PARAVIRT_OP(set_rr0_to_rr4),
+ "0"(__val0), "1"(__val1), "2"(__val2),
+ "3"(__val3), "4"(__val4)
+ : PARAVIRT_CLOBBERS5);
+}
+
+/* unsigned long paravirt_getreg(int reg) */
+#define __paravirt_getreg(reg) \
+ ({ \
+ register unsigned long ia64_intri_res asm ("r8"); \
+ register unsigned long __reg asm ("r8") = (reg); \
+ \
+ BUILD_BUG_ON(!__builtin_constant_p(reg)); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(GETREG) \
+ + (reg)) \
+ : "=r"(ia64_intri_res) \
+ : PARAVIRT_OP(getreg), "0"(__reg) \
+ : PARAVIRT_CLOBBERS1); \
+ \
+ ia64_intri_res; \
+ })
+
+/* void paravirt_setreg(int reg, unsigned long val) */
+#define paravirt_setreg(reg, val) \
+ do { \
+ register unsigned long __val asm ("r8") = val; \
+ register unsigned long __reg asm ("r9") = reg; \
+ register unsigned long ia64_clobber1 asm ("r8"); \
+ register unsigned long ia64_clobber2 asm ("r9"); \
+ \
+ BUILD_BUG_ON(!__builtin_constant_p(reg)); \
+ asm volatile (paravirt_alt_bundle(__PARAVIRT_BR, \
+ PARAVIRT_TYPE(SETREG) \
+ + (reg)) \
+ : "=r"(ia64_clobber1), \
+ "=r"(ia64_clobber2) \
+ : PARAVIRT_OP(setreg), \
+ "1"(__reg), "0"(__val) \
+ : PARAVIRT_CLOBBERS2); \
+ } while (0)
+
+#endif /* ASM_SUPPORTED */
+#endif /* CONFIG_PARAVIRT && ASM_SUPPOTED */
+
#endif /* _ASM_IA64_PARAVIRT_PRIVOP_H */