TCTI: add TCTI TCG backend for acceleration on non-JIT AArch64

accel/tcg/translate-all.c

@@ -1129,7 +1129,7 @@ static bool alloc_code_gen_buffer_anon(size_t size, int prot,
     return true;
 }
 
-#ifndef CONFIG_TCG_INTERPRETER
+#if !defined(CONFIG_TCG_INTERPRETER) && !defined(CONFIG_TCG_THREADED_INTERPRETER)
 #ifdef CONFIG_POSIX
 #include "qemu/memfd.h"
 
@@ -1256,7 +1256,7 @@ static bool alloc_code_gen_buffer_splitwx_vmremap(size_t size, Error **errp)
 
 static bool alloc_code_gen_buffer_splitwx(size_t size, Error **errp)
 {
-#ifndef CONFIG_TCG_INTERPRETER
+#if !defined(CONFIG_TCG_INTERPRETER) && !defined(CONFIG_TCG_THREADED_INTERPRETER)
 # ifdef CONFIG_DARWIN
     return alloc_code_gen_buffer_splitwx_vmremap(size, errp);
 # endif
@@ -1289,7 +1289,7 @@ static bool alloc_code_gen_buffer(size_t size, int splitwx, Error **errp)
 
     prot = PROT_READ | PROT_WRITE | PROT_EXEC;
     flags = MAP_PRIVATE | MAP_ANONYMOUS;
-#ifdef CONFIG_TCG_INTERPRETER
+#if defined(CONFIG_TCG_INTERPRETER) || defined(CONFIG_TCG_THREADED_INTERPRETER)
     /* The tcg interpreter does not need execute permission. */
     prot = PROT_READ | PROT_WRITE;
 #elif defined(CONFIG_DARWIN)

configure

@@ -362,6 +362,7 @@ tsan="no"
 fortify_source="$default_feature"
 strip_opt="yes"
 tcg_interpreter="false"
+tcg_threaded_interpreter="false"
 bigendian="no"
 mingw32="no"
 gcov="no"
@@ -1115,6 +1116,10 @@ for opt do
   ;;
   --enable-tcg-interpreter) tcg_interpreter="true"
   ;;
+  --disable-tcg-tcti) tcg_threaded_interpreter="false"
+  ;;
+  --enable-tcg-tcti) tcg_threaded_interpreter="true"
+  ;;
   --disable-cap-ng)  cap_ng="disabled"
   ;;
   --enable-cap-ng) cap_ng="enabled"
@@ -6469,9 +6474,8 @@ NINJA=$ninja $meson setup \
         -Dvhost_user_blk_server=$vhost_user_blk_server -Dmultiprocess=$multiprocess \
         -Dfuse=$fuse -Dfuse_lseek=$fuse_lseek -Dguest_agent_msi=$guest_agent_msi \
         $(if test "$default_features" = no; then echo "-Dauto_features=disabled"; fi) \
-	-Dtcg_interpreter=$tcg_interpreter -Dshared_lib=$shared_lib \
-        $cross_arg \
-        "$PWD" "$source_path"
+	-Dtcg_interpreter=$tcg_interpreter -Dtcg_threaded_interpreter=$tcg_threaded_interpreter\
+	-Dshared_lib=$shared_lib $cross_arg "$PWD" "$source_path"
 
 if test "$?" -ne 0 ; then
     error_exit "meson setup failed"

disas.c

@@ -152,6 +152,8 @@ static void initialize_debug_host(CPUDebug *s)
 #endif
 #if defined(CONFIG_TCG_INTERPRETER)
     s->info.print_insn = print_insn_tci;
+#elif defined(CONFIG_TCG_THREADED_INTERPRETER)
+    s->info.print_insn = print_insn_tcti;
 #elif defined(__i386__)
     s->info.mach = bfd_mach_i386_i386;
     s->info.print_insn = print_insn_i386;

include/disas/dis-asm.h

@@ -411,6 +411,7 @@ typedef struct disassemble_info {
 typedef int (*disassembler_ftype) (bfd_vma, disassemble_info *);
 
 int print_insn_tci(bfd_vma, disassemble_info*);
+int print_insn_tcti(bfd_vma, disassemble_info*);
 int print_insn_big_mips         (bfd_vma, disassemble_info*);
 int print_insn_little_mips      (bfd_vma, disassemble_info*);
 int print_insn_nanomips         (bfd_vma, disassemble_info*);

include/exec/exec-all.h

@@ -543,7 +543,11 @@ void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr);
 #if defined(CONFIG_TCG_INTERPRETER)
 extern __thread uintptr_t tci_tb_ptr;
 # define GETPC() tci_tb_ptr
+#elif defined(CONFIG_TCG_THREADED_INTERPRETER)
+extern __thread uintptr_t tcti_call_return_address;
+# define GETPC() tcti_call_return_address
 #else
+/* Note that this is correct for TCTI also; whose gadget behaves like native code. */
 # define GETPC() \
     ((uintptr_t)__builtin_extract_return_addr(__builtin_return_address(0)))
 #endif

include/tcg/tcg.h

@@ -1296,7 +1296,7 @@ static inline unsigned get_mmuidx(TCGMemOpIdx oi)
 #define TB_EXIT_IDXMAX    1
 #define TB_EXIT_REQUESTED 3
 
-#ifdef CONFIG_TCG_INTERPRETER
+#if defined(CONFIG_TCG_INTERPRETER) || defined(CONFIG_TCG_THREADED_INTERPRETER)
 uintptr_t tcg_qemu_tb_exec(CPUArchState *env, const void *tb_ptr);
 #else
 typedef uintptr_t tcg_prologue_fn(CPUArchState *env, const void *tb_ptr);

meson.build

@@ -58,6 +58,7 @@ python = import('python').find_installation()
 supported_oses = ['windows', 'freebsd', 'netbsd', 'openbsd', 'darwin', 'sunos', 'linux']
 supported_cpus = ['ppc', 'ppc64', 's390x', 'riscv32', 'riscv64', 'x86', 'x86_64',
   'arm', 'aarch64', 'mips', 'mips64', 'sparc', 'sparc64']
+tcti_supported_cpus = ['aarch64']
 
 cpu = host_machine.cpu_family()
 targetos = host_machine.system()
@@ -248,6 +249,25 @@ if not get_option('tcg').disabled()
   endif
   if get_option('tcg_interpreter')
     tcg_arch = 'tci'
+  elif get_option('tcg_threaded_interpreter')
+    if cpu not in tcti_supported_cpus
+      error('Unsupported CPU @0@ for TCTI, try --enable-tcg-interpreter'.format(cpu))
+    else
+      warning('TCTI is extremely experimental and incomplete! Things might break!')
+      tcg_arch = '@0@-tcti'.format(cpu)
+    endif
+
+    # Tell our compiler how to generate our TCTI gadgets.
+    gadget_generator = 'tcg/@0@/tcti-gadget-gen.py'.format(tcg_arch)
+    tcti_gadgets = custom_target('tcti-gadgets.c.inc',
+                                output: 'tcti-gadgets.c.inc',
+                                input: gadget_generator,
+                                command: [find_program(gadget_generator), '@OUTPUT@'],
+                                build_by_default: true,
+                                build_always_stale: false)
+
+    genh += tcti_gadgets
+
   elif config_host['ARCH'] == 'sparc64'
     tcg_arch = 'sparc'
   elif config_host['ARCH'] == 's390x'
@@ -1284,6 +1304,8 @@ foreach target : target_dirs
       config_all += { sym: 'y' }
       if sym == 'CONFIG_TCG' and tcg_arch == 'tci'
         config_target += { 'CONFIG_TCG_INTERPRETER': 'y' }
+      elif sym == 'CONFIG_TCG' and tcg_arch.endswith('tcti')
+        config_target += { 'CONFIG_TCG_THREADED_INTERPRETER': 'y' }
       elif sym == 'CONFIG_XEN' and have_xen_pci_passthrough
         config_target += { 'CONFIG_XEN_PCI_PASSTHROUGH': 'y' }
       endif
@@ -2575,6 +2597,8 @@ summary_info += {'TCG support':       config_all.has_key('CONFIG_TCG')}
 if config_all.has_key('CONFIG_TCG')
   if get_option('tcg_interpreter')
     summary_info += {'TCG backend':   'TCI (TCG with bytecode interpreter, experimental and slow)'}
+  elif get_option('tcg_threaded_interpreter')
+    summary_info += {'TCG backend':   'TCTI (TCG with threaded-dispatch bytecode interpreter, experimental and slow; but faster than TCI)'}
   else
     summary_info += {'TCG backend':   'native (@0@)'.format(cpu)}
   endif

meson_options.txt

@@ -43,6 +43,8 @@ option('tcg', type: 'feature', value: 'auto',
        description: 'TCG support')
 option('tcg_interpreter', type: 'boolean', value: false,
        description: 'TCG with bytecode interpreter (experimental and slow)')
+option('tcg_threaded_interpreter', type: 'boolean', value: false,
+       description: 'TCG with threaded-dispatch bytecode interpreter (experimental and slow, but less slow than TCI)')
 option('cfi', type: 'boolean', value: 'false',
        description: 'Control-Flow Integrity (CFI)')
 option('cfi_debug', type: 'boolean', value: 'false',

scripts/mtest2make.py

@@ -75,18 +75,18 @@ def process_tests(test, targets, suites):
     print('run-test-%d: $(.test.deps.%d)' % (i,i))
     print('\t@$(call .test.run,%d,$(.test.output-format))' % (i,))
 
-    test_suites = test['suite'] or ['default']
-    is_slow = any(s.endswith('-slow') for s in test_suites)
-    for s in test_suites:
-        # The suite name in the introspection info is "PROJECT:SUITE"
-        s = s.split(':')[1]
-        if s.endswith('-slow'):
-            s = s[:-5]
-        if is_slow:
-            suites[s].slow_tests.append(i)
-        else:
-            suites[s].tests.append(i)
-        suites[s].executables.add(executable)
+    #test_suites = test['suite'] or ['default']
+    #is_slow = any(s.endswith('-slow') for s in test_suites)
+    #for s in test_suites:
+    #    # The suite name in the introspection info is "PROJECT:SUITE"
+    #    s = s.split(':')[1]
+    #    if s.endswith('-slow'):
+    #        s = s[:-5]
+    #    if is_slow:
+    #        suites[s].slow_tests.append(i)
+    #    else:
+    #        suites[s].tests.append(i)
+    #    suites[s].executables.add(executable)
 
 def emit_prolog(suites, prefix):
     all_tap = ' '.join(('%s-report-%s.tap' % (prefix, k) for k in suites.keys()))