Add a cache for class variables

Redo of 34a2acdac788602c14bf05fb616215187badd504 and
931138b00696419945dc03e10f033b1f53cd50f3 which were reverted.

GitHub PR #4340.

This change implements a cache for class variables. Previously there was
no cache for cvars. Cvar access is slow due to needing to travel all the
way up th ancestor tree before returning the cvar value. The deeper the
ancestor tree the slower cvar access will be.

The benefits of the cache are more visible with a higher number of
included modules due to the way Ruby looks up class variables. The
benchmark here includes 26 modules and shows with the cache, this branch
is 6.5x faster when accessing class variables.

```
compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19]
built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19]

|         |compare-ruby|built-ruby|
|:--------|-----------:|---------:|
|vm_cvar  |      5.681M|   36.980M|
|         |           -|     6.51x|
```

Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails
application. ActiveRecord::Base.logger has 71 ancestors. The more
ancestors a tree has, the more clear the speed increase. IE if Base had
only one ancestor we'd see no improvement. This benchmark is run on a
vanilla Rails application.

Benchmark code:

```ruby
require "benchmark/ips"
require_relative "config/environment"

Benchmark.ips do |x|
  x.report "logger" do
    ActiveRecord::Base.logger
  end
end
```

Ruby 3.0 master / Rails 6.1:

```
Warming up --------------------------------------
              logger   155.251k i/100ms
Calculating -------------------------------------
```

Ruby 3.0 with cvar cache /  Rails 6.1:

```
Warming up --------------------------------------
              logger     1.546M i/100ms
Calculating -------------------------------------
              logger     14.857M (± 4.8%) i/s -     74.198M in   5.006202s
```

Lastly we ran a benchmark to demonstate the difference between master
and our cache when the number of modules increases. This benchmark
measures 1 ancestor, 30 ancestors, and 100 ancestors.

Ruby 3.0 master:

```
Warming up --------------------------------------
            1 module     1.231M i/100ms
          30 modules   432.020k i/100ms
         100 modules   145.399k i/100ms
Calculating -------------------------------------
            1 module     12.210M (± 2.1%) i/s -     61.553M in   5.043400s
          30 modules      4.354M (± 2.7%) i/s -     22.033M in   5.063839s
         100 modules      1.434M (± 2.9%) i/s -      7.270M in   5.072531s

Comparison:
            1 module: 12209958.3 i/s
          30 modules:  4354217.8 i/s - 2.80x  (± 0.00) slower
         100 modules:  1434447.3 i/s - 8.51x  (± 0.00) slower
```

Ruby 3.0 with cvar cache:

```
Warming up --------------------------------------
            1 module     1.641M i/100ms
          30 modules     1.655M i/100ms
         100 modules     1.620M i/100ms
Calculating -------------------------------------
            1 module     16.279M (± 3.8%) i/s -     82.038M in   5.046923s
          30 modules     15.891M (± 3.9%) i/s -     79.459M in   5.007958s
         100 modules     16.087M (± 3.6%) i/s -     81.005M in   5.041931s

Comparison:
            1 module: 16279458.0 i/s
         100 modules: 16087484.6 i/s - same-ish: difference falls within error
          30 modules: 15891406.2 i/s - same-ish: difference falls within error
```

Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>

15 files changed, 271 insertions, 22 deletions

diff --git a/benchmark/vm_cvar.yml b/benchmark/vm_cvar.yml
new file mode 100644
index 0000000000..1d0e161829
--- /dev/null
+++ b/benchmark/vm_cvar.yml
@@ -0,0 +1,20 @@
+prelude: |
+  class A
+    @@foo = 1
+
+    def self.foo
+      @@foo
+    end
+
+    ("A".."Z").each do |module_name|
+      eval <<-EOM
+        module #{module_name}
+        end
+
+        include #{module_name}
+      EOM
+    end
+  end
+benchmark:
+  vm_cvar: A.foo
+loop_count: 600000
diff --git a/class.c b/class.c
index 9ac2b3ff47..ef3db6dab3 100644
--- a/class.c
+++ b/class.c
@@ -27,6 +27,7 @@
 #include <ctype.h>
 
 #include "constant.h"
+#include "debug_counter.h"
 #include "id_table.h"
 #include "internal.h"
 #include "internal/class.h"
@@ -43,6 +44,8 @@
 #define METACLASS_OF(k) RBASIC(k)->klass
 #define SET_METACLASS_OF(k, cls) RBASIC_SET_CLASS(k, cls)
 
+RUBY_EXTERN rb_serial_t ruby_vm_global_cvar_state;
+
 void
 rb_class_subclass_add(VALUE super, VALUE klass)
 {
@@ -957,6 +960,7 @@ rb_include_class_new(VALUE module, VALUE super)
 	RCLASS_CONST_TBL(module) = rb_id_table_create(0);
     }
     RCLASS_IV_TBL(klass) = RCLASS_IV_TBL(module);
+    RCLASS_CVC_TBL(klass) = RCLASS_CVC_TBL(module);
     RCLASS_CONST_TBL(klass) = RCLASS_CONST_TBL(module);
 
     RCLASS_SET_SUPER(klass, super);
@@ -1085,6 +1089,8 @@ do_include_modules_at(const VALUE klass, VALUE c, VALUE module, int search_super
         VALUE super_class = RCLASS_SUPER(c);
 
         // invalidate inline method cache
+        RB_DEBUG_COUNTER_INC(cvar_include_invalidate);
+        ruby_vm_global_cvar_state++;
         tbl = RCLASS_M_TBL(module);
         if (tbl && rb_id_table_size(tbl)) {
             if (search_super) { // include
diff --git a/common.mk b/common.mk
index 30380e0a4d..feb14b9a0d 100644
--- a/common.mk
+++ b/common.mk
@@ -2460,6 +2460,7 @@ class.$(OBJEXT): {$(VPATH)}backward/2/stdarg.h
 class.$(OBJEXT): {$(VPATH)}class.c
 class.$(OBJEXT): {$(VPATH)}config.h
 class.$(OBJEXT): {$(VPATH)}constant.h
+class.$(OBJEXT): {$(VPATH)}debug_counter.h
 class.$(OBJEXT): {$(VPATH)}defines.h
 class.$(OBJEXT): {$(VPATH)}encoding.h
 class.$(OBJEXT): {$(VPATH)}id.h
diff --git a/compile.c b/compile.c
index c3aa8f3600..c541711ff9 100644
--- a/compile.c
+++ b/compile.c
@@ -8066,8 +8066,9 @@ iseq_compile_each0(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const NODE *node, in
 	if (!popped) {
 	    ADD_INSN(ret, line_node, dup);
 	}
-	ADD_INSN1(ret, line_node, setclassvariable,
-		  ID2SYM(node->nd_vid));
+        ADD_INSN2(ret, line_node, setclassvariable,
+                  ID2SYM(node->nd_vid),
+                  get_ivar_ic_value(iseq,node->nd_vid));
 	break;
       }
       case NODE_OP_ASGN1: {
@@ -8690,8 +8691,9 @@ iseq_compile_each0(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const NODE *node, in
       }
       case NODE_CVAR:{
 	if (!popped) {
-	    ADD_INSN1(ret, line_node, getclassvariable,
-		      ID2SYM(node->nd_vid));
+	    ADD_INSN2(ret, line_node, getclassvariable,
+		      ID2SYM(node->nd_vid),
+		      get_ivar_ic_value(iseq,node->nd_vid));
 	}
 	break;
       }
diff --git a/debug_counter.h b/debug_counter.h
index 9452f4c737..3cf80cc188 100644
--- a/debug_counter.h
+++ b/debug_counter.h
@@ -24,6 +24,12 @@ RB_DEBUG_COUNTER(mc_inline_miss_same_cme)    // IMC miss, but same CME
 RB_DEBUG_COUNTER(mc_inline_miss_same_def)    // IMC miss, but same definition
 RB_DEBUG_COUNTER(mc_inline_miss_diff)        // IMC miss, different methods
 
+RB_DEBUG_COUNTER(cvar_write_inline_hit)      // cvar cache hit on write
+RB_DEBUG_COUNTER(cvar_read_inline_hit)       // cvar cache hit on read
+RB_DEBUG_COUNTER(cvar_inline_miss)           // miss inline cache
+RB_DEBUG_COUNTER(cvar_class_invalidate)      // invalidate cvar cache when define a cvar that's defined on a subclass
+RB_DEBUG_COUNTER(cvar_include_invalidate)    // invalidate cvar cache on module include or prepend
+
 RB_DEBUG_COUNTER(mc_cme_complement)          // number of acquiring complement CME
 RB_DEBUG_COUNTER(mc_cme_complement_hit)      // number of cache hit for complemented CME
 
diff --git a/gc.c b/gc.c
index 0ad090affb..09420deac2 100644
--- a/gc.c
+++ b/gc.c
@@ -2998,6 +2998,13 @@ cc_table_free(rb_objspace_t *objspace, VALUE klass, bool alive)
     }
 }
 
+static enum rb_id_table_iterator_result
+cvar_table_free_i(VALUE value, void * ctx)
+{
+    xfree((void *) value);
+    return ID_TABLE_CONTINUE;
+}
+
 void
 rb_cc_table_free(VALUE klass)
 {
@@ -3109,6 +3116,10 @@ obj_free(rb_objspace_t *objspace, VALUE obj)
 	if (RCLASS_IV_INDEX_TBL(obj)) {
             iv_index_tbl_free(RCLASS_IV_INDEX_TBL(obj));
 	}
+	if (RCLASS_CVC_TBL(obj)) {
+            rb_id_table_foreach_values(RCLASS_CVC_TBL(obj), cvar_table_free_i, NULL);
+            rb_id_table_free(RCLASS_CVC_TBL(obj));
+	}
 	if (RCLASS_SUBCLASSES(obj)) {
 	    if (BUILTIN_TYPE(obj) == T_MODULE) {
 		rb_class_detach_module_subclasses(obj);
@@ -4552,6 +4563,9 @@ obj_memsize_of(VALUE obj, int use_all_types)
 	    if (RCLASS_IV_TBL(obj)) {
 		size += st_memsize(RCLASS_IV_TBL(obj));
 	    }
+	    if (RCLASS_CVC_TBL(obj)) {
+		size += rb_id_table_memsize(RCLASS_CVC_TBL(obj));
+	    }
 	    if (RCLASS_IV_INDEX_TBL(obj)) {
                 // TODO: more correct value
 		size += st_memsize(RCLASS_IV_INDEX_TBL(obj));
@@ -9750,6 +9764,27 @@ update_cc_tbl(rb_objspace_t *objspace, VALUE klass)
 }
 
 static enum rb_id_table_iterator_result
+update_cvc_tbl_i(ID id, VALUE cvc_entry, void *data)
+{
+    struct rb_cvar_class_tbl_entry *entry;
+
+    entry = (struct rb_cvar_class_tbl_entry *)cvc_entry;
+
+    entry->class_value = rb_gc_location(entry->class_value);
+
+    return ID_TABLE_CONTINUE;
+}
+
+static void
+update_cvc_tbl(rb_objspace_t *objspace, VALUE klass)
+{
+    struct rb_id_table *tbl = RCLASS_CVC_TBL(klass);
+    if (tbl) {
+        rb_id_table_foreach_with_replace(tbl, update_cvc_tbl_i, 0, objspace);
+    }
+}
+
+static enum rb_id_table_iterator_result
 update_const_table(VALUE value, void *data)
 {
     rb_const_entry_t *ce = (rb_const_entry_t *)value;
@@ -9820,6 +9855,7 @@ gc_update_object_references(rb_objspace_t *objspace, VALUE obj)
         if (!RCLASS_EXT(obj)) break;
         update_m_tbl(objspace, RCLASS_M_TBL(obj));
         update_cc_tbl(objspace, obj);
+        update_cvc_tbl(objspace, obj);
 
         gc_update_tbl_refs(objspace, RCLASS_IV_TBL(obj));
 
diff --git a/id_table.c b/id_table.c
index 840ab46ee3..149f52cd88 100644
--- a/id_table.c
+++ b/id_table.c
@@ -92,7 +92,7 @@ rb_id_table_init(struct rb_id_table *tbl, int capa)
     return tbl;
 }
 
-struct rb_id_table *
+MJIT_FUNC_EXPORTED struct rb_id_table *
 rb_id_table_create(size_t capa)
 {
     struct rb_id_table *tbl = ALLOC(struct rb_id_table);
@@ -223,7 +223,7 @@ hash_table_show(struct rb_id_table *tbl)
 }
 #endif
 
-int
+MJIT_FUNC_EXPORTED int
 rb_id_table_lookup(struct rb_id_table *tbl, ID id, VALUE *valp)
 {
     id_key_t key = id2key(id);
@@ -253,7 +253,7 @@ rb_id_table_insert_key(struct rb_id_table *tbl, const id_key_t key, const VALUE
     return TRUE;
 }
 
-int
+MJIT_FUNC_EXPORTED int
 rb_id_table_insert(struct rb_id_table *tbl, ID id, VALUE val)
 {
     return rb_id_table_insert_key(tbl, id2key(id), val);
diff --git a/include/ruby/internal/intern/variable.h b/include/ruby/internal/intern/variable.h
index 8210662fa0..faa0cc004f 100644
--- a/include/ruby/internal/intern/variable.h
+++ b/include/ruby/internal/intern/variable.h
@@ -72,6 +72,7 @@ VALUE rb_mod_const_missing(VALUE,VALUE);
 VALUE rb_cvar_defined(VALUE, ID);
 void rb_cvar_set(VALUE, ID, VALUE);
 VALUE rb_cvar_get(VALUE, ID);
+VALUE rb_cvar_find(VALUE, ID, VALUE*);
 void rb_cv_set(VALUE, const char*, VALUE);
 VALUE rb_cv_get(VALUE, const char*);
 void rb_define_class_variable(VALUE, const char*, VALUE);
diff --git a/insns.def b/insns.def
index c6ebf7401b..0b741af504 100644
--- a/insns.def
+++ b/insns.def
@@ -230,26 +230,28 @@ setinstancevariable
 /* Get value of class variable id of klass as val. */
 DEFINE_INSN
 getclassvariable
-(ID id)
+(ID id, IVC ic)
 ()
 (VALUE val)
 /* "class variable access from toplevel" warning can be hooked. */
 // attr bool leaf = false; /* has rb_warning() */
 {
-    val = rb_cvar_get(vm_get_cvar_base(vm_get_cref(GET_EP()), GET_CFP(), 1), id);
+    rb_cref_t * cref = vm_get_cref(GET_EP());
+    rb_control_frame_t *cfp = GET_CFP();
+    val = vm_getclassvariable(GET_ISEQ(), cref, cfp, id, (ICVARC)ic);
 }
 
 /* Set value of class variable id of klass as val. */
 DEFINE_INSN
 setclassvariable
-(ID id)
+(ID id, IVC ic)
 (VALUE val)
 ()
 /* "class variable access from toplevel" warning can be hooked. */
 // attr bool leaf = false; /* has rb_warning() */
 {
     vm_ensure_not_refinement_module(GET_SELF());
-    rb_cvar_set(vm_get_cvar_base(vm_get_cref(GET_EP()), GET_CFP(), 1), id, val);
+    vm_setclassvariable(GET_ISEQ(), vm_get_cref(GET_EP()), GET_CFP(), id,  val, (ICVARC)ic);
 }
 
 /* Get constant variable id. If klass is Qnil and allow_nil is Qtrue, constants
diff --git a/internal/class.h b/internal/class.h
index 6c03a31a4e..1d25d9e7eb 100644
--- a/internal/class.h
+++ b/internal/class.h
@@ -31,6 +31,12 @@ struct rb_iv_index_tbl_entry {
     VALUE class_value;
 };
 
+struct rb_cvar_class_tbl_entry {
+    uint32_t index;
+    rb_serial_t global_cvar_state;
+    VALUE class_value;
+};
+
 struct rb_classext_struct {
     struct st_table *iv_index_tbl; // ID -> struct rb_iv_index_tbl_entry
     struct st_table *iv_tbl;
@@ -40,6 +46,7 @@ struct rb_classext_struct {
     struct rb_id_table *const_tbl;
     struct rb_id_table *callable_m_tbl;
     struct rb_id_table *cc_tbl; /* ID -> [[ci, cc1], cc2, ...] */
+    struct rb_id_table *cvc_tbl;
     struct rb_subclass_entry *subclasses;
     struct rb_subclass_entry **parent_subclasses;
     /**
@@ -83,6 +90,7 @@ typedef struct rb_classext_struct rb_classext_t;
 #endif
 #define RCLASS_CALLABLE_M_TBL(c) (RCLASS_EXT(c)->callable_m_tbl)
 #define RCLASS_CC_TBL(c) (RCLASS_EXT(c)->cc_tbl)
+#define RCLASS_CVC_TBL(c) (RCLASS_EXT(c)->cvc_tbl)
 #define RCLASS_IV_INDEX_TBL(c) (RCLASS_EXT(c)->iv_index_tbl)
 #define RCLASS_ORIGIN(c) (RCLASS_EXT(c)->origin_)
 #define RCLASS_REFINED_CLASS(c) (RCLASS_EXT(c)->refined_class)
diff --git a/variable.c b/variable.c
index 5e5f5c4bf4..778778c866 100644
--- a/variable.c
+++ b/variable.c
@@ -39,6 +39,9 @@
 #include "ractor_core.h"
 #include "vm_sync.h"
 
+RUBY_EXTERN rb_serial_t ruby_vm_global_cvar_state;
+#define GET_GLOBAL_CVAR_STATE() (ruby_vm_global_cvar_state)
+
 typedef void rb_gvar_compact_t(void *var);
 
 static struct rb_id_table *rb_global_tbl;
@@ -3325,6 +3328,30 @@ cvar_overtaken(VALUE front, VALUE target, ID id)
     }
 }
 
+static VALUE
+find_cvar(VALUE klass, VALUE * front, VALUE * target, ID id)
+{
+    VALUE v = Qundef;
+    CVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR();
+    if (cvar_lookup_at(klass, id, (&v))) {
+        if (!*front) {
+            *front = klass;
+        }
+        *target = klass;
+    }
+
+    for (klass = cvar_front_klass(klass); klass; klass = RCLASS_SUPER(klass)) {
+	if (cvar_lookup_at(klass, id, (&v))) {
+            if (!*front) {
+                *front = klass;
+            }
+            *target = klass;
+        }
+    }
+
+    return v;
+}
+
 #define CVAR_FOREACH_ANCESTORS(klass, v, r) \
     for (klass = cvar_front_klass(klass); klass; klass = RCLASS_SUPER(klass)) { \
 	if (cvar_lookup_at(klass, id, (v))) { \
@@ -3338,6 +3365,20 @@ cvar_overtaken(VALUE front, VALUE target, ID id)
     CVAR_FOREACH_ANCESTORS(klass, v, r);\
 } while(0)
 
+static void
+check_for_cvar_table(VALUE subclass, VALUE key)
+{
+    st_table *tbl = RCLASS_IV_TBL(subclass);
+
+    if (tbl && st_lookup(tbl, key, NULL)) {
+        RB_DEBUG_COUNTER_INC(cvar_class_invalidate);
+        ruby_vm_global_cvar_state++;
+        return;
+    }
+
+    rb_class_foreach_subclass(subclass, check_for_cvar_table, key);
+}
+
 void
 rb_cvar_set(VALUE klass, ID id, VALUE val)
 {
@@ -3357,26 +3398,61 @@ rb_cvar_set(VALUE klass, ID id, VALUE val)
     }
     check_before_mod_set(target, id, val, "class variable");
 
-    rb_class_ivar_set(target, id, val);
+    int result = rb_class_ivar_set(target, id, val);
+
+    struct rb_id_table *rb_cvc_tbl = RCLASS_CVC_TBL(target);
+
+    if (!rb_cvc_tbl) {
+        rb_cvc_tbl = RCLASS_CVC_TBL(target) = rb_id_table_create(2);
+    }
+
+    struct rb_cvar_class_tbl_entry *ent;
+
+    if (!rb_id_table_lookup(rb_cvc_tbl, id, (VALUE*)&ent)) {
+        ent = ALLOC(struct rb_cvar_class_tbl_entry);
+        ent->class_value = target;
+        ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
+        rb_id_table_insert(rb_cvc_tbl, id, (VALUE)ent);
+        RB_DEBUG_COUNTER_INC(cvar_inline_miss);
+    } else {
+        ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
+    }
+
+    // Break the cvar cache if this is a new class variable
+    // and target is a module or a subclass with the same
+    // cvar in this lookup.
+    if (result == 0) {
+        if (RB_TYPE_P(target, T_CLASS)) {
+            if (RCLASS_SUBCLASSES(target)) {
+                rb_class_foreach_subclass(target, check_for_cvar_table, id);
+            }
+        }
+    }
 }
 
 VALUE
-rb_cvar_get(VALUE klass, ID id)
+rb_cvar_find(VALUE klass, ID id, VALUE *front)
 {
-    VALUE tmp, front = 0, target = 0;
-    st_data_t value;
+    VALUE target = 0;
+    VALUE value;
 
-    tmp = klass;
-    CVAR_LOOKUP(&value, {if (!front) front = klass; target = klass;});
+    value = find_cvar(klass, front, &target, id);
     if (!target) {
 	rb_name_err_raise("uninitialized class variable %1$s in %2$s",
-			  tmp, ID2SYM(id));
+			  klass, ID2SYM(id));
     }
-    cvar_overtaken(front, target, id);
+    cvar_overtaken(*front, target, id);
     return (VALUE)value;
 }
 
 VALUE
+rb_cvar_get(VALUE klass, ID id)
+{
+    VALUE front = 0;
+    return rb_cvar_find(klass, id, &front);
+}
+
+VALUE
 rb_cvar_defined(VALUE klass, ID id)
 {
     if (!klass) return Qfalse;
diff --git a/vm.c b/vm.c
index 6140bcf861..3963d67168 100644
--- a/vm.c
+++ b/vm.c
@@ -405,6 +405,7 @@ unsigned int    ruby_vm_event_local_num;
 
 rb_serial_t ruby_vm_global_constant_state = 1;
 rb_serial_t ruby_vm_class_serial = 1;
+rb_serial_t ruby_vm_global_cvar_state = 1;
 
 static const struct rb_callcache vm_empty_cc = {
     .flags = T_IMEMO | (imemo_callcache << FL_USHIFT) | VM_CALLCACHE_UNMARKABLE,
@@ -484,7 +485,7 @@ rb_dtrace_setup(rb_execution_context_t *ec, VALUE klass, ID id,
 static VALUE
 vm_stat(int argc, VALUE *argv, VALUE self)
 {
-    static VALUE sym_global_constant_state, sym_class_serial;
+    static VALUE sym_global_constant_state, sym_class_serial, sym_global_cvar_state;
     VALUE arg = Qnil;
     VALUE hash = Qnil, key = Qnil;
 
@@ -505,6 +506,7 @@ vm_stat(int argc, VALUE *argv, VALUE self)
 #define S(s) sym_##s = ID2SYM(rb_intern_const(#s))
 	S(global_constant_state);
 	S(class_serial);
+	S(global_cvar_state);
 #undef S
     }
 
@@ -516,6 +518,7 @@ vm_stat(int argc, VALUE *argv, VALUE self)
 
     SET(global_constant_state, ruby_vm_global_constant_state);
     SET(class_serial, ruby_vm_class_serial);
+    SET(global_cvar_state, ruby_vm_global_cvar_state);
 #undef SET
 
     if (!NIL_P(key)) { /* matched key should return above */
diff --git a/vm_core.h b/vm_core.h
index 7eda31f11e..59b788e178 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -241,6 +241,10 @@ struct iseq_inline_iv_cache_entry {
     struct rb_iv_index_tbl_entry *entry;
 };
 
+struct iseq_inline_cvar_cache_entry {
+    struct rb_cvar_class_tbl_entry *entry;
+};
+
 union iseq_inline_storage_entry {
     struct {
 	struct rb_thread_struct *running_thread;
@@ -1157,6 +1161,7 @@ enum vm_svar_index {
 /* inline cache */
 typedef struct iseq_inline_constant_cache *IC;
 typedef struct iseq_inline_iv_cache_entry *IVC;
+typedef struct iseq_inline_cvar_cache_entry *ICVARC;
 typedef union iseq_inline_storage_entry *ISE;
 typedef const struct rb_callinfo *CALL_INFO;
 typedef const struct rb_callcache *CALL_CACHE;
diff --git a/vm_insnhelper.c b/vm_insnhelper.c
index 171b0f4273..72c90a8a5c 100644
--- a/vm_insnhelper.c
+++ b/vm_insnhelper.c
@@ -951,7 +951,7 @@ vm_ensure_not_refinement_module(VALUE self)
 }
 
 static inline VALUE
-vm_get_iclass(rb_control_frame_t *cfp, VALUE klass)
+vm_get_iclass(const rb_control_frame_t *cfp, VALUE klass)
 {
     return klass;
 }
@@ -1041,7 +1041,7 @@ vm_get_ev_const(rb_execution_context_t *ec, VALUE orig_klass, ID id, bool allow_
 }
 
 static inline VALUE
-vm_get_cvar_base(const rb_cref_t *cref, rb_control_frame_t *cfp, int top_level_raise)
+vm_get_cvar_base(const rb_cref_t *cref, const rb_control_frame_t *cfp, int top_level_raise)
 {
     VALUE klass;
 
@@ -1282,6 +1282,86 @@ vm_setivar(VALUE obj, ID id, VALUE val, const rb_iseq_t *iseq, IVC ic, const str
 }
 
 static inline VALUE
+vm_getclassvariable(const rb_iseq_t *iseq, const rb_cref_t *cref, const rb_control_frame_t *cfp, ID id, ICVARC ic)
+{
+    if (ic->entry && ic->entry->global_cvar_state == GET_GLOBAL_CVAR_STATE()) {
+        VALUE v = Qundef;
+        RB_DEBUG_COUNTER_INC(cvar_read_inline_hit);
+
+        if (st_lookup(RCLASS_IV_TBL(ic->entry->class_value), (st_data_t)id, &v)) {
+            return v;
+        }
+    }
+
+    VALUE klass = vm_get_cvar_base(cref, cfp, 1);
+    VALUE defined_class = 0;
+
+    VALUE cvar_value = rb_cvar_find(klass, id, &defined_class);
+
+    if (RB_TYPE_P(defined_class, T_ICLASS)) {
+        defined_class = RBASIC(defined_class)->klass;
+    }
+
+    struct rb_id_table *rb_cvc_tbl = RCLASS_CVC_TBL(defined_class);
+
+    if (!rb_cvc_tbl) {
+        rb_cvc_tbl = RCLASS_CVC_TBL(defined_class) = rb_id_table_create(2);
+    }
+
+    struct rb_cvar_class_tbl_entry *ent;
+
+    if (!rb_id_table_lookup(rb_cvc_tbl, id, (VALUE*)&ent)) {
+        rb_bug("should have cvar cache entry");
+    } else {
+        ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
+    }
+
+    ic->entry = ent;
+    RB_OBJ_WRITTEN(iseq, Qundef, ent->class_value);
+
+    return cvar_value;
+}
+
+static inline void
+vm_setclassvariable(const rb_iseq_t *iseq, const rb_cref_t *cref, const rb_control_frame_t *cfp, ID id, VALUE val, ICVARC ic)
+{
+    if (ic->entry && ic->entry->global_cvar_state == GET_GLOBAL_CVAR_STATE()) {
+        RB_DEBUG_COUNTER_INC(cvar_write_inline_hit);
+
+        rb_class_ivar_set(ic->entry->class_value, id, val);
+        return;
+    }
+
+    VALUE klass = vm_get_cvar_base(cref, cfp, 1);
+
+    rb_cvar_set(klass, id, val);
+
+    VALUE defined_class = 0;
+    rb_cvar_find(klass, id, &defined_class);
+
+    if (RB_TYPE_P(defined_class, T_ICLASS)) {
+        defined_class = RBASIC(defined_class)->klass;
+    }
+
+    struct rb_id_table *rb_cvc_tbl = RCLASS_CVC_TBL(defined_class);
+
+    if (!rb_cvc_tbl) {
+        rb_bug("the cvc table should be set");
+    }
+
+    struct rb_cvar_class_tbl_entry *ent;
+
+    if (!rb_id_table_lookup(rb_cvc_tbl, id, (VALUE*)&ent)) {
+        rb_bug("should have cvar cache entry");
+    } else {
+        ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
+    }
+
+    ic->entry = ent;
+    RB_OBJ_WRITTEN(iseq, Qundef, ent->class_value);
+}
+
+static inline VALUE
 vm_getinstancevariable(const rb_iseq_t *iseq, VALUE obj, ID id, IVC ic)
 {
     return vm_getivar(obj, id, iseq, ic, NULL, FALSE);
diff --git a/vm_insnhelper.h b/vm_insnhelper.h
index e254e709e1..0d90eb9434 100644
--- a/vm_insnhelper.h
+++ b/vm_insnhelper.h
@@ -16,6 +16,7 @@ MJIT_SYMBOL_EXPORT_BEGIN
 RUBY_EXTERN VALUE ruby_vm_const_missing_count;
 RUBY_EXTERN rb_serial_t ruby_vm_global_constant_state;
 RUBY_EXTERN rb_serial_t ruby_vm_class_serial;
+RUBY_EXTERN rb_serial_t ruby_vm_global_cvar_state;
 
 MJIT_SYMBOL_EXPORT_END
 
@@ -179,6 +180,8 @@ CC_SET_FASTPATH(const struct rb_callcache *cc, vm_call_handler func, bool enable
 #define NEXT_CLASS_SERIAL() (++ruby_vm_class_serial)
 #define GET_GLOBAL_CONSTANT_STATE() (ruby_vm_global_constant_state)
 #define INC_GLOBAL_CONSTANT_STATE() (++ruby_vm_global_constant_state)
+#define GET_GLOBAL_CVAR_STATE() (ruby_vm_global_cvar_state)
+#define INC_GLOBAL_CVAR_STATE() (++ruby_vm_global_cvar_state)
 
 static inline struct vm_throw_data *
 THROW_DATA_NEW(VALUE val, const rb_control_frame_t *cf, int st)