[Qemu-devel] [RFC 08/14] Add instruction-counting infrastructure to target-arm

[Christopher Covington]

This (partially) divorces counting instructions from basic block
collection so instructions can be counted without the bbv plugin being
enabled.

Written by Aaron Lindsay.

Signed-off-by: Christopher Covington <address@hidden>
---
 include/exec/cpu-defs.h    |  2 ++
 target-arm/helper.c        | 43 +++++++++++++++++++++++++++++++++++++++++++
 target-arm/helper.h        |  2 ++
 target-arm/translate-a64.c | 18 ++++++++++--------
 target-arm/translate.c     | 18 ++++++++++--------
 5 files changed, 67 insertions(+), 16 deletions(-)

diff --git a/include/exec/cpu-defs.h b/include/exec/cpu-defs.h
index 1eed930..9af080f 100644
--- a/include/exec/cpu-defs.h
+++ b/include/exec/cpu-defs.h
@@ -136,6 +136,8 @@ typedef struct CPUIOTLBEntry {
     CPU_COMMON_TLB                                                      \
     /* Instruction Count (for profiling) */                                    
\
     uint64_t prof_ic;                                                   \
+    /* How much of prof_ic's value have we processed? */                \
+    uint64_t prof_ic_last;                                              \
     /* PC (for profiling) */                                            \
     uint64_t prof_pc;                                                   \
     /* next page start (for profiling) */                               \
diff --git a/target-arm/helper.c b/target-arm/helper.c
index 1843ec5..be3ad01 100644
--- a/target-arm/helper.c
+++ b/target-arm/helper.c
@@ -673,6 +673,15 @@ static inline bool arm_ccnt_enabled(CPUARMState *env)
     return true;
 }
 
+/* Called by anything that wants to be an input for event counts to the PMU
+ * (except for SWINC, event 0x000, since its events can target specific
+ * counters)
+ */
+static void pmevcntr_increment(CPUARMState *env, uint8_t event_type,
+                               uint64_t increment_by)
+{
+}
+
 void pmccntr_sync(CPUARMState *env)
 {
     uint64_t temp_ticks;
@@ -4060,6 +4069,40 @@ void HELPER(context_check_pid)(CPUARMState *env)
 
     bb_context_check_pid(env->prof_ic, pid);
 }
+
+void HELPER(update_instruction_count)(CPUARMState *env)
+{
+    if (bbtrace_initialized()) {
+      /*
+       * If the bbv plugin is compiled in and enabled, we must account for the
+       * fact that bbv_profile needs to see prof_ic before we clear it.
+       * However, it doesn't always clear the counter every time this gets
+       * called, so we must keep track of the last value seen to ensure we
+       * update the instruction counter correctly in that case.
+       */
+        increment_instruction_counters(env->prof_ic - env->prof_ic_last);
+        if (env->prof_pc && env->prof_is_jmp) {
+            // If this is the end of a basic block, zero out last_seen counter 
too
+            env->prof_ic_last = 0;
+        } else {
+            env->prof_ic_last = env->prof_ic;
+        }
+    } else {
+        pmevcntr_increment(env, PMU_COUNTER_TYPE_INSTRUCTIONS, env->prof_ic);
+        pmevcntr_increment(env, PMU_COUNTER_TYPE_CYCLES, env->prof_ic);
+        env->prof_ic = 0;
+    }
+}
+
+#else //!CONFIG_BBVEC
+
+void HELPER(update_instruction_count)(CPUARMState *env)
+{
+    pmevcntr_increment(env, PMU_COUNTER_TYPE_INSTRUCTIONS, env->prof_ic);
+    pmevcntr_increment(env, PMU_COUNTER_TYPE_CYCLES, env->prof_ic);
+    env->prof_ic = 0;
+}
+
 #endif //CONFIG_BBVEC
 
 /* Sign/zero extend */
diff --git a/target-arm/helper.h b/target-arm/helper.h
index 41291c9..02edf3a 100644
--- a/target-arm/helper.h
+++ b/target-arm/helper.h
@@ -43,6 +43,8 @@ DEF_HELPER_1(context_check_mode, void, env)
 DEF_HELPER_1(context_check_pid, void, env)
 #endif // CONFIG_BBVEC
 
+DEF_HELPER_1(update_instruction_count, void, env)
+
 DEF_HELPER_3(ssat, i32, env, i32, i32)
 DEF_HELPER_3(usat, i32, env, i32, i32)
 DEF_HELPER_3(ssat16, i32, env, i32, i32)
diff --git a/target-arm/translate-a64.c b/target-arm/translate-a64.c
index 80b27ed..f6f8832 100644
--- a/target-arm/translate-a64.c
+++ b/target-arm/translate-a64.c
@@ -55,9 +55,9 @@ static TCGv_i64 cpu_exclusive_high;
  * keep track of the current pc, so the last pc in the block can be
  * captured.   */
 static TCGv_i64 cpu_prof_pc;
-static TCGv_i64 cpu_prof_ic;
 static TCGv_i64 cpu_prof_is_jmp;
 #endif // CONFIG_BBVEC
+static TCGv_i64 cpu_prof_ic;
 #ifdef CONFIG_USER_ONLY
 static TCGv_i64 cpu_exclusive_test;
 static TCGv_i32 cpu_exclusive_info;
@@ -124,10 +124,11 @@ void a64_translate_init(void)
 
 #ifdef CONFIG_BBVEC
     // bbvec profiling globals
-    cpu_prof_ic = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_ic), "prof_ic");
     cpu_prof_pc = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_pc), "prof_pc");
     cpu_prof_is_jmp = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_is_jmp), "prof_is_jmp");
 #endif // CONFIG_BBVEC
+    cpu_prof_ic = tcg_global_mem_new_i64(TCG_AREG0,
+        offsetof(CPUARMState, prof_ic), "prof_ic");
 
     cpu_exclusive_addr = tcg_global_mem_new_i64(TCG_AREG0,
         offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
@@ -210,11 +211,6 @@ void gen_a64_set_pc_im(uint64_t val)
 
 #ifdef CONFIG_BBVEC
 /* Basic block profiling functions */
-static void gen_insn_cnt_incr(CPUARMState * env, DisasContext *s)
-{
-    tcg_gen_addi_i64(cpu_prof_ic, cpu_prof_ic, 1);
-}
-
 static void gen_pc_incr(CPUARMState * env, DisasContext *s)
 {
     tcg_gen_movi_i64(cpu_prof_pc, s->pc);
@@ -226,6 +222,11 @@ static void gen_store_is_jmp(uint32_t jmp)
 }
 #endif // CONFIG_BBVEC
 
+static void gen_insn_cnt_incr(CPUARMState * env, DisasContext *s)
+{
+    tcg_gen_addi_i64(cpu_prof_ic, cpu_prof_ic, 1);
+}
+
 static void gen_exception_internal(int excp)
 {
     TCGv_i32 tcg_excp = tcg_const_i32(excp);
@@ -11068,6 +11069,7 @@ void gen_intermediate_code_internal_a64(ARMCPU *cpu,
 
     tcg_clear_temp_count();
 
+    gen_helper_update_instruction_count(cpu_env);
 #ifdef CONFIG_BBVEC
     /* Profile previously run block, check for PID change, and initialize
      * prof_is_jmp flag.   */
@@ -11082,10 +11084,10 @@ void gen_intermediate_code_internal_a64(ARMCPU *cpu,
 #ifdef CONFIG_BBVEC
         if (bbtrace_initialized()) {
             gen_helper_context_check_mode(cpu_env);
-            gen_insn_cnt_incr(env, dc);
             gen_pc_incr(env, dc);
         }
 #endif // CONFIG_BBVEC
+        gen_insn_cnt_incr(env, dc);
 
         if (unlikely(!QTAILQ_EMPTY(&cs->breakpoints))) {
             QTAILQ_FOREACH(bp, &cs->breakpoints, entry) {
diff --git a/target-arm/translate.c b/target-arm/translate.c
index 7a19a8b..113e3b6 100644
--- a/target-arm/translate.c
+++ b/target-arm/translate.c
@@ -71,10 +71,10 @@ static TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
 static TCGv_i64 cpu_exclusive_addr;
 static TCGv_i64 cpu_exclusive_val;
 #ifdef CONFIG_BBVEC
-static TCGv_i64 cpu_prof_ic;
 static TCGv_i64 cpu_prof_pc;
 static TCGv_i64 cpu_prof_is_jmp;
 #endif
+static TCGv_i64 cpu_prof_ic;
 #ifdef CONFIG_USER_ONLY
 static TCGv_i64 cpu_exclusive_test;
 static TCGv_i32 cpu_exclusive_info;
@@ -109,10 +109,11 @@ void arm_translate_init(void)
 
 #ifdef CONFIG_BBVEC
     // bbvec profiling globals
-    cpu_prof_ic = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_ic), "prof_ic");
     cpu_prof_pc = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_pc), "prof_pc");
     cpu_prof_is_jmp = tcg_global_mem_new_i64(TCG_AREG0, offsetof(CPUARMState, 
prof_is_jmp), "prof_is_jmp");
 #endif
+    cpu_prof_ic = tcg_global_mem_new_i64(TCG_AREG0,
+        offsetof(CPUARMState, prof_ic), "prof_ic");
 
     cpu_exclusive_addr = tcg_global_mem_new_i64(TCG_AREG0,
         offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
@@ -194,11 +195,6 @@ static inline TCGv_i32 load_reg(DisasContext *s, int reg)
 
 #ifdef CONFIG_BBVEC
 /* Basic block profiling functions */
-static void gen_insn_cnt_incr(CPUARMState * env, DisasContext *s)
-{
-    tcg_gen_addi_i64(cpu_prof_ic, cpu_prof_ic, 1);
-}
-
 static void gen_pc_incr(CPUARMState * env, DisasContext *s)
 {
     tcg_gen_movi_i64(cpu_prof_pc, s->pc);
@@ -210,6 +206,11 @@ static void gen_store_is_jmp(uint32_t jmp)
 }
 #endif
 
+static void gen_insn_cnt_incr(CPUARMState * env, DisasContext *s)
+{
+    tcg_gen_addi_i64(cpu_prof_ic, cpu_prof_ic, 1);
+}
+
 /* Set a CPU register.  The source must be a temporary and will be
    marked as dead.  */
 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
@@ -11627,6 +11628,7 @@ static inline void 
gen_intermediate_code_internal(ARMCPU *cpu,
         store_cpu_field(tmp, condexec_bits);
       }
 
+    gen_helper_update_instruction_count(cpu_env);
 #ifdef CONFIG_BBVEC
     /* Profile previously run block, check for PID change, and initialize
      * prof_is_jmp flag.   */
@@ -11641,13 +11643,13 @@ static inline void 
gen_intermediate_code_internal(ARMCPU *cpu,
 #ifdef CONFIG_BBVEC
         if (bbtrace_initialized()) {
             gen_helper_context_check_mode(cpu_env);
-            gen_insn_cnt_incr(env, dc);
             gen_pc_incr(env, dc);
             /* FIXME: this call should not be necessary if all the cases
                where the prof_is_jmp flag gets set are correct.   */
             gen_store_is_jmp(0);
         }
 #endif
+        gen_insn_cnt_incr(env, dc);
 
 #ifdef CONFIG_USER_ONLY
         /* Intercept jump to the magic kernel page.  */
-- 
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project