Skip to contentSkip to conversation

iOS and Apple Silicon host support

PATCH Superseded Joelle van Dyne 7 participants 39 messages Oct 19, 2020 – Oct 26, 2020

Subsystems (4)

v2 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v2 cover iOS and Apple Silicon host support

These set of changes brings QEMU TCG to iOS devices and future Apple Silicon devices. They were originally developed last year and have been working in the UTM app. Recently, we ported the changes to master, re-wrote a lot of the build script changes for meson, and broke up the patches into more distinct units.

A summary of the changes:

* CONFIG_IOS and CONFIG_IOS_JIT defined when building for iOS and iOS specific changes (as well as unsupported code) are gated behind it.
* A new dependency, libucontext is added since iOS does not have native ucontext and broken support for sigaltstack. libucontext is available as a new option for coroutine backend.
* On stock iOS devices, there is a workaround for running JIT code without any special entitlement. It requires the JIT region to be mirror mapped with one region RW and another one RX. To support this style of JIT, TCG is changed to support writing to a different code_ptr. These changes are gated by the CONFIG_IOS_JIT.
* For (recent) jailbroken iOS devices as well as upcoming Apple Silicon devices, there are new rules for applications supporting JIT (with the proper entitlement). These rules are implemented as well.

Since v2:

* Changed getting mirror pointer from a macro to inline functions
* Split constification of TCG code pointers to separate patch
* Removed slirp updates (will send future patch once slirp changes are in)
* Removed shared library patch (will send future patch)

-j

osy (9):
configure: option to disable host block devices
configure: cross-compiling without cross_prefix
qemu: add support for iOS host
coroutine: add libucontext as external library
tcg: add const hints for code pointers
tcg: implement mirror mapped JIT for iOS
tcg: mirror mapping RWX pages for iOS optional
tcg: support JIT on Apple Silicon
block: check availablity for preadv/pwritev on mac

 docs/devel/ios.rst           |  40 +++++++++
 configure                    | 104 ++++++++++++++++++++--
 meson.build                  |  32 ++++++-
 include/exec/exec-all.h      |  10 +++
 include/sysemu/tcg.h         |   2 +-
 include/tcg/tcg-apple-jit.h  |  85 ++++++++++++++++++
 include/tcg/tcg.h            |  28 +++++-
 tcg/aarch64/tcg-target.h     |  23 ++++-
 tcg/arm/tcg-target.h         |   9 +-
 tcg/i386/tcg-target.h        |  24 ++++-
 tcg/mips/tcg-target.h        |   8 +-
 tcg/ppc/tcg-target.h         |   8 +-
 tcg/riscv/tcg-target.h       |   9 +-
 tcg/s390/tcg-target.h        |  13 ++-
 tcg/sparc/tcg-target.h       |   8 +-
 tcg/tci/tcg-target.h         |   9 +-
 accel/tcg/cpu-exec-common.c  |   2 +
 accel/tcg/cpu-exec.c         |   9 +-
 accel/tcg/tcg-all.c          |  27 +++++-
 accel/tcg/translate-all.c    | 168 ++++++++++++++++++++++++++++++++---
 block.c                      |   2 +-
 block/file-posix.c           |  50 ++++++++---
 bsd-user/main.c              |   2 +-
 linux-user/main.c            |   2 +-
 net/slirp.c                  |  16 ++--
 qga/commands-posix.c         |   6 ++
 target/arm/arm-semi.c        |   2 +
 target/m68k/m68k-semi.c      |   2 +
 target/nios2/nios2-semi.c    |   2 +
 tcg/tcg.c                    |  64 ++++++++-----
 util/coroutine-ucontext.c    |   9 ++
 .gitmodules                  |   3 +
 libucontext                  |   1 +
 meson_options.txt            |   2 +
 qemu-options.hx              |  11 +++
 tcg/aarch64/tcg-target.c.inc |  48 ++++++----
 tcg/arm/tcg-target.c.inc     |  33 ++++---
 tcg/i386/tcg-target.c.inc    |  28 +++---
 tcg/mips/tcg-target.c.inc    |  64 +++++++------
 tcg/ppc/tcg-target.c.inc     |  55 +++++++-----
 tcg/riscv/tcg-target.c.inc   |  51 ++++++-----
 tcg/s390/tcg-target.c.inc    |  25 +++---
 tcg/sparc/tcg-target.c.inc   |  33 ++++---
 tcg/tcg-ldst.c.inc           |   2 +-
 tcg/tcg-pool.c.inc           |   9 +-
 tcg/tci/tcg-target.c.inc     |   8 +-
 tests/qtest/meson.build      |   7 +-
 47 files changed, 919 insertions(+), 236 deletions(-)
 create mode 100644 docs/devel/ios.rst
 create mode 100644 include/tcg/tcg-apple-jit.h
 create mode 160000 libucontext
Signature · 1 line
2.24.3 (Apple Git-128)
#1 lore raw

PATCH v2 1/9 configure: option to disable host block devices

From: osy <[email protected]>

Some hosts (iOS) have a sandboxed filesystem and do not provide low-level APIs for interfacing with host block devices.

---
 configure          | 4 ++++
 meson.build        | 1 +
 block/file-posix.c | 8 +++++++-
 3 files changed, 12 insertions(+), 1 deletion(-)
configure +4 -0
index b553288c5e..3c63879750 100755
@@ -446,6 +446,7 @@ meson=""
 ninja=""
 skip_meson=no
 gettext=""
+host_block_device_support="yes"
 
 bogus_os="no"
 malloc_trim="auto"
@@ -6098,6 +6099,9 @@ if test "$default_devices" = "yes" ; then
 else
   echo "CONFIG_MINIKCONF_MODE=--allnoconfig" >> $config_host_mak
 fi
+if test "$host_block_device_support" = "yes" ; then
+  echo "CONFIG_HOST_BLOCK_DEVICE=y" >> $config_host_mak
+fi
 if test "$debug_tcg" = "yes" ; then
   echo "CONFIG_DEBUG_TCG=y" >> $config_host_mak
 fi
meson.build +1 -0
index 17c89c87c6..5d3a47784b 100644
@@ -1947,6 +1947,7 @@ summary_info += {'vvfat support':     config_host.has_key('CONFIG_VVFAT')}
 summary_info += {'qed support':       config_host.has_key('CONFIG_QED')}
 summary_info += {'parallels support': config_host.has_key('CONFIG_PARALLELS')}
 summary_info += {'sheepdog support':  config_host.has_key('CONFIG_SHEEPDOG')}
+summary_info += {'host block dev support': config_host.has_key('CONFIG_HOST_BLOCK_DEVICE')}
 summary_info += {'capstone':          capstone_opt == 'disabled' ? false : capstone_opt}
 summary_info += {'libpmem support':   config_host.has_key('CONFIG_LIBPMEM')}
 summary_info += {'libdaxctl support': config_host.has_key('CONFIG_LIBDAXCTL')}
block/file-posix.c +7 -1
index c63926d592..52f7c20525 100644
@@ -41,7 +41,7 @@
 #include "scsi/pr-manager.h"
 #include "scsi/constants.h"
 
-#if defined(__APPLE__) && (__MACH__)
+#if defined(CONFIG_HOST_BLOCK_DEVICE) && defined(__APPLE__) && (__MACH__)
 #include <paths.h>
 #include <sys/param.h>
 #include <IOKit/IOKitLib.h>
@@ -3247,6 +3247,8 @@ BlockDriver bdrv_file = {
 /***********************************************/
 /* host device */
 
+#if defined(CONFIG_HOST_BLOCK_DEVICE)
+
 #if defined(__APPLE__) && defined(__MACH__)
 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
                                 CFIndex maxPathSize, int flags);
@@ -3872,6 +3874,8 @@ static BlockDriver bdrv_host_cdrom = {
 };
 #endif /* __FreeBSD__ */
 
+#endif /* CONFIG_HOST_BLOCK_DEVICE */
+
 static void bdrv_file_init(void)
 {
     /*
@@ -3879,6 +3883,7 @@ static void bdrv_file_init(void)
      * registered last will get probed first.
      */
     bdrv_register(&bdrv_file);
+#if defined(CONFIG_HOST_BLOCK_DEVICE)
     bdrv_register(&bdrv_host_device);
 #ifdef __linux__
     bdrv_register(&bdrv_host_cdrom);
@@ -3886,6 +3891,7 @@ static void bdrv_file_init(void)
 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
     bdrv_register(&bdrv_host_cdrom);
 #endif
+#endif /* CONFIG_HOST_BLOCK_DEVICE */
 }
 
 block_init(bdrv_file_init);
Signature · 1 line
2.24.3 (Apple Git-128)
#2 lore raw

PATCH v2 2/9 configure: cross-compiling without cross_prefix

In reply to Joelle van Dyne

From: osy <[email protected]>

The iOS toolchain does not use the host prefix naming convention. We add a new option --enable-cross-compile that forces cross-compile even without a cross_prefix.

---
 configure | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
configure +12 -1
index 3c63879750..46d5db63e8 100755
@@ -234,6 +234,7 @@ cpu=""
 iasl="iasl"
 interp_prefix="/usr/gnemul/qemu-%M"
 static="no"
+cross_compile="no"
 cross_prefix=""
 audio_drv_list=""
 block_drv_rw_whitelist=""
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"
   ;;
   --cc=*) CC="$optarg"
   ;;
@@ -878,6 +884,10 @@ for opt do
   ;;
   --cross-prefix=*)
   ;;
+  --enable-cross-compile)
+  ;;
+  --disable-cross-compile)
+  ;;
   --cc=*)
   ;;
   --host-cc=*) host_cc="$optarg"
@@ -1687,6 +1697,7 @@ Advanced options (experts only):
   --efi-aarch64=PATH       PATH of efi file to use for aarch64 VMs.
   --with-suffix=SUFFIX     suffix for QEMU data inside datadir/libdir/sysconfdir/docdir [$qemu_suffix]
   --with-pkgversion=VERS   use specified string as sub-version of the package
+  --enable-cross-compile   enable cross compiling (set automatically if $cross_prefix is set)
   --enable-debug           enable common debug build options
   --enable-sanitizers      enable default sanitizers
   --enable-tsan            enable thread sanitizer
@@ -7164,7 +7175,7 @@ if has $sdl2_config; then
 fi
 echo "strip = [$(meson_quote $strip)]" >> $cross
 echo "windres = [$(meson_quote $windres)]" >> $cross
-if test -n "$cross_prefix"; then
+if test "$cross_compile" = "yes"; then
     cross_arg="--cross-file config-meson.cross"
     echo "[host_machine]" >> $cross
     if test "$mingw32" = "yes" ; then
Signature · 1 line
2.24.3 (Apple Git-128)
#3 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 24 lines · Joelle van Dyne

From: osy <[email protected]>

The iOS toolchain does not use the host prefix naming convention. We add a new option --enable-cross-compile that forces cross-compile even without a cross_prefix.

---
 configure | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
configure +1 -0
index 3c63879750..46d5db63e8 100755
@@ -234,6 +234,7 @@ cpu=""
 iasl="iasl"
 interp_prefix="/usr/gnemul/qemu-%M"
 static="no"
+cross_compile="no"
 cross_prefix=""
 audio_drv_list=""
 block_drv_rw_whitelist=""

configure for opt do · line 464

Quoted 8 lines
configure +5 -0
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

Thomas

#4 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 27 lines · Thomas Huth
Quoted 26 lines · Joelle van Dyne

From: osy <[email protected]>

The iOS toolchain does not use the host prefix naming convention. We add a new option --enable-cross-compile that forces cross-compile even without a cross_prefix.

---
 configure | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
configure +1 -0
index 3c63879750..46d5db63e8 100755
@@ -234,6 +234,7 @@ cpu=""
 iasl="iasl"
 interp_prefix="/usr/gnemul/qemu-%M"
 static="no"
+cross_compile="no"
 cross_prefix=""
 audio_drv_list=""
 block_drv_rw_whitelist=""
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
Quoted 8 lines

configure for opt do · line 464

Quoted 6 lines
configure +5 -0
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

Thomas

#5 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 37 lines · Thomas Huth
Quoted 27 lines · Thomas Huth
Quoted 26 lines · Joelle van Dyne

From: osy <[email protected]>

The iOS toolchain does not use the host prefix naming convention. We add a new option --enable-cross-compile that forces cross-compile even without a cross_prefix.

---
 configure | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
configure +1 -0
index 3c63879750..46d5db63e8 100755
@@ -234,6 +234,7 @@ cpu=""
 iasl="iasl"
 interp_prefix="/usr/gnemul/qemu-%M"
 static="no"
+cross_compile="no"
 cross_prefix=""
 audio_drv_list=""
 block_drv_rw_whitelist=""
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
Quoted 8 lines

configure for opt do · line 464

Quoted 6 lines
configure +5 -0
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

Quoted 2 lines

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

That seems less intuitive for people trying to find this option. If --help lists --enable-cross-compile I can guess what that means but there's no way I could guess --cros-prefix="" unless I've been told or searched and stumbled upon it. So unless it's a big problem I like the explicit options better. Or is that a convention in other projects to use empty prefix to enable cross compile that I don't know about?

Regards,
BALATON Zoltan

#6 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Correct me if I'm wrong but wouldn't the following test still fail with --cross-prefix=""

if test -n "$cross_prefix"; then
...

That was my main reason for making this change.

-j

Quoted 51 lines · BALATON Zoltan
Quoted 37 lines · Thomas Huth
Quoted 27 lines · Thomas Huth
Quoted 26 lines · Joelle van Dyne

From: osy <[email protected]>

The iOS toolchain does not use the host prefix naming convention. We add a new option --enable-cross-compile that forces cross-compile even without a cross_prefix.

---
 configure | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
configure +1 -0
index 3c63879750..46d5db63e8 100755
@@ -234,6 +234,7 @@ cpu=""
 iasl="iasl"
 interp_prefix="/usr/gnemul/qemu-%M"
 static="no"
+cross_compile="no"
 cross_prefix=""
 audio_drv_list=""
 block_drv_rw_whitelist=""
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
Quoted 8 lines

configure for opt do · line 464

Quoted 6 lines
configure +5 -0
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

Quoted 2 lines

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

That seems less intuitive for people trying to find this option. If --help lists --enable-cross-compile I can guess what that means but there's no way I could guess --cros-prefix="" unless I've been told or searched and stumbled upon it. So unless it's a big problem I like the explicit options better. Or is that a convention in other projects to use empty prefix to enable cross compile that I don't know about?

Regards,
BALATON Zoltan

#7 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 7 lines · Joelle van Dyne

Correct me if I'm wrong but wouldn't the following test still fail with --cross-prefix=""

if test -n "$cross_prefix"; then
...

That was my main reason for making this change.

That's why I wrote "still introduce the cross_compile=yes variable" ... that change is certainly required anyway.

Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 11 lines

configure for opt do · line 464

Quoted 9 lines
configure +5 -0
@@ -456,6 +457,11 @@ for opt do
   optarg=$(expr "x$opt" : 'x[^=]*=\(.*\)')
   case "$opt" in
   --cross-prefix=*) cross_prefix="$optarg"
+                    cross_compile="yes"
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

Quoted 6 lines
Quoted 6 lines

That seems less intuitive for people trying to find this option. If --help lists --enable-cross-compile I can guess what that means but there's no way I could guess --cros-prefix="" unless I've been told or searched and stumbled upon it. So unless it's a big problem I like the explicit options better. Or is that a convention in other projects to use empty prefix to enable cross compile that I don't know about?

I don't think that --cross-prefix is a "standard" option... Most other (GNU-tools related) projects use "--build" and "--host" instead... so I guess we're free to chose here. Let's see whether other people here have an opionion on this...

Thomas

#8 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 17 lines · Thomas Huth
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
Quoted 6 lines

configure for opt do · line 464

Quoted 4 lines
configure +4 -0
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

Quoted 6 lines
Quoted 6 lines

That seems less intuitive for people trying to find this option. If --help lists --enable-cross-compile I can guess what that means but there's no way I could guess --cros-prefix="" unless I've been told or searched and stumbled upon it. So unless it's a big problem I like the explicit options better. Or is that a convention in other projects to use empty prefix to enable cross compile that I don't know about?

Quoted 4 lines

I don't think that --cross-prefix is a "standard" option... Most other (GNU-tools related) projects use "--build" and "--host" instead... so I guess we're free to chose here. Let's see whether other people here have an opionion on this...

Yeah, the way GNU tools do it is that you specify --build and --host, and --host triggers cross compilation. I'm not sure how they'd handle the situation where the cross prefix is empty.

For QEMU, I agree with Thomas that --cross-prefix="" is enough.

Paolo

#9 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

As the patch stands --cross-prefix="" works but I like having the --enable-cross-compile option because it's more clear of the intention (in logs and such). I've been struggling to come up with a concise modification to the help text for --cross-prefix that explains that empty PREFIX is allowed and why you would want to do that.

-j

Quoted 32 lines · Paolo Bonzini
Quoted 17 lines · Thomas Huth
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
Quoted 6 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+  ;;
+  --enable-cross-compile) cross_compile="yes"
+  ;;
+  --disable-cross-compile) cross_compile="no"

Can't you simply use --cros-prefix="" instead?

I mean, still introduce the "cross_compile=yes" variable, just omit the new options.

Quoted 6 lines
Quoted 6 lines

That seems less intuitive for people trying to find this option. If --help lists --enable-cross-compile I can guess what that means but there's no way I could guess --cros-prefix="" unless I've been told or searched and stumbled upon it. So unless it's a big problem I like the explicit options better. Or is that a convention in other projects to use empty prefix to enable cross compile that I don't know about?

Quoted 4 lines

I don't think that --cross-prefix is a "standard" option... Most other (GNU-tools related) projects use "--build" and "--host" instead... so I guess we're free to chose here. Let's see whether other people here have an opionion on this...

Yeah, the way GNU tools do it is that you specify --build and --host, and --host triggers cross compilation. I'm not sure how they'd handle the situation where the cross prefix is empty.

For QEMU, I agree with Thomas that --cross-prefix="" is enough.

Paolo

#10 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 5 lines · Joelle van Dyne

As the patch stands --cross-prefix="" works but I like having the --enable-cross-compile option because it's more clear of the intention (in logs and such). I've been struggling to come up with a concise modification to the help text for --cross-prefix that explains that empty PREFIX is allowed and why you would want to do that.

--cross-prefix=PREFIX enable cross compilation, using PREFIX
in front of compile tools

Paolo

#11 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Just to be clear, you're in favor of removing the --enable-cross-compile option? I can do that if others agree but I'm not sure what the downside of adding this extra option is?

-j

Quoted 13 lines · Paolo Bonzini
Quoted 5 lines · Joelle van Dyne

As the patch stands --cross-prefix="" works but I like having the --enable-cross-compile option because it's more clear of the intention (in logs and such). I've been struggling to come up with a concise modification to the help text for --cross-prefix that explains that empty PREFIX is allowed and why you would want to do that.

--cross-prefix=PREFIX enable cross compilation, using PREFIX
in front of compile tools

Paolo

#12 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 3 lines · Joelle van Dyne

Just to be clear, you're in favor of removing the --enable-cross-compile option? I can do that if others agree but I'm not sure what the downside of adding this extra option is?

For me the downside is that it's not clear that --cross-prefix=foo- implies --enable-cross-compile.

Paolo

Quoted 17 lines

-j

Quoted 13 lines · Paolo Bonzini
Quoted 5 lines · Joelle van Dyne

As the patch stands --cross-prefix="" works but I like having the --enable-cross-compile option because it's more clear of the intention (in logs and such). I've been struggling to come up with a concise modification to the help text for --cross-prefix that explains that empty PREFIX is allowed and why you would want to do that.

--cross-prefix=PREFIX enable cross compilation, using PREFIX
in front of compile tools

Paolo

#13 lore raw

Re: [PATCH v2 2/9] configure: cross-compiling without cross_prefix

Quoted 7 lines · Paolo Bonzini
Quoted 3 lines · Joelle van Dyne

Just to be clear, you're in favor of removing the --enable-cross-compile option? I can do that if others agree but I'm not sure what the downside of adding this extra option is?

For me the downside is that it's not clear that --cross-prefix=foo- implies --enable-cross-compile.

I agree with Paolo ... two options sound rather confusing to me than just having one option. And the amount of options of the configure script is too big already anyway, so I'd recommend to drop --enable-cross-compile.

Thomas

#14 lore raw

PATCH v2 3/9 qemu: add support for iOS host

In reply to Joelle van Dyne

From: osy <[email protected]>

This introduces support for building for iOS hosts. When the correct Xcode toolchain is used, iOS host will be detected automatically.

block: disable features not supported by iOS sandbox
slirp: disable SMB features for iOS
target: disable system() calls for iOS
tcg: use sys_icache_invalidate() instead of GCC builtin for iOS
tests: disable tests on iOS which uses system()

---
 configure                 | 43 ++++++++++++++++++++++++++++++++++++++-
 meson.build               |  2 +-
 tcg/aarch64/tcg-target.h  | 10 +++++++++
 block.c                   |  2 +-
 block/file-posix.c        | 30 ++++++++++++++++-----------
 net/slirp.c               | 16 +++++++--------
 qga/commands-posix.c      |  6 ++++++
 target/arm/arm-semi.c     |  2 ++
 target/m68k/m68k-semi.c   |  2 ++
 target/nios2/nios2-semi.c |  2 ++
 tests/qtest/meson.build   |  7 +++----
 11 files changed, 95 insertions(+), 27 deletions(-)
configure +42 -1
index 46d5db63e8..c474d7c221 100755
@@ -561,6 +561,19 @@ EOF
   compile_object
 }
 
+check_ios() {
+  cat > $TMPC <<EOF
+#ifdef __APPLE__
+#import "TargetConditionals.h"
+#if !TARGET_OS_IPHONE
+#error TARGET_OS_IPHONE not true
+#endif
+#endif
+int main(void) { return 0; }
+EOF
+  compile_object
+}
+
 check_include() {
 cat > $TMPC <<EOF
 #include <$1>
@@ -603,7 +616,11 @@ elif check_define __DragonFly__ ; then
 elif check_define __NetBSD__; then
   targetos='NetBSD'
 elif check_define __APPLE__; then
-  targetos='Darwin'
+  if check_ios ; then
+    targetos='iOS'
+  else
+    targetos='Darwin'
+  fi
 else
   # This is a fatal error, but don't report it yet, because we
   # might be going to just print the --help text, or it might
@@ -780,6 +797,22 @@ Darwin)
   # won't work when we're compiling with gcc as a C compiler.
   QEMU_CFLAGS="-DOS_OBJECT_USE_OBJC=0 $QEMU_CFLAGS"
 ;;
+iOS)
+  bsd="yes"
+  darwin="yes"
+  ios="yes"
+  if [ "$cpu" = "x86_64" ] ; then
+    QEMU_CFLAGS="-arch x86_64 $QEMU_CFLAGS"
+    QEMU_LDFLAGS="-arch x86_64 $QEMU_LDFLAGS"
+  fi
+  host_block_device_support="no"
+  audio_drv_list=""
+  audio_possible_drivers=""
+  QEMU_LDFLAGS="-framework CoreFoundation $QEMU_LDFLAGS"
+  # Disable attempts to use ObjectiveC features in os/object.h since they
+  # won't work when we're compiling with gcc as a C compiler.
+  QEMU_CFLAGS="-DOS_OBJECT_USE_OBJC=0 $QEMU_CFLAGS"
+;;
 SunOS)
   solaris="yes"
   make="${MAKE-gmake}"
@@ -6162,6 +6195,10 @@ if test "$darwin" = "yes" ; then
   echo "CONFIG_DARWIN=y" >> $config_host_mak
 fi
 
+if test "$ios" = "yes" ; then
+  echo "CONFIG_IOS=y" >> $config_host_mak
+fi
+
 if test "$solaris" = "yes" ; then
   echo "CONFIG_SOLARIS=y" >> $config_host_mak
 fi
@@ -7166,6 +7203,7 @@ echo "cpp_link_args = [${LDFLAGS:+$(meson_quote $LDFLAGS)}]" >> $cross
 echo "[binaries]" >> $cross
 echo "c = [$(meson_quote $cc)]" >> $cross
 test -n "$cxx" && echo "cpp = [$(meson_quote $cxx)]" >> $cross
+test -n "$objcc" && echo "objc = [$(meson_quote $objcc)]" >> $cross
 echo "ar = [$(meson_quote $ar)]" >> $cross
 echo "nm = [$(meson_quote $nm)]" >> $cross
 echo "pkgconfig = [$(meson_quote $pkg_config_exe)]" >> $cross
@@ -7184,6 +7222,9 @@ if test "$cross_compile" = "yes"; then
     if test "$linux" = "yes" ; then
         echo "system = 'linux'" >> $cross
     fi
+    if test "$darwin" = "yes" ; then
+        echo "system = 'darwin'" >> $cross
+    fi
     case "$ARCH" in
         i386|x86_64)
             echo "cpu_family = 'x86'" >> $cross
meson.build +1 -1
index 5d3a47784b..69a3c00cce 100644
@@ -140,7 +140,7 @@ if targetos == 'windows'
                                       include_directories: include_directories('.'))
 elif targetos == 'darwin'
   coref = dependency('appleframeworks', modules: 'CoreFoundation')
-  iokit = dependency('appleframeworks', modules: 'IOKit')
+  iokit = dependency('appleframeworks', modules: 'IOKit', required: 'CONFIG_IOS' not in config_host)
   cocoa = dependency('appleframeworks', modules: 'Cocoa', required: get_option('cocoa'))
 elif targetos == 'sunos'
   socket = [cc.find_library('socket'),
tcg/aarch64/tcg-target.h +10 -0
index 663dd0b95e..a2b22b4305 100644
@@ -148,9 +148,19 @@ typedef enum {
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
 
+#if defined(__APPLE__)
+void sys_icache_invalidate(void *start, size_t len);
+#endif
+
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
+#if defined(__APPLE__)
+    sys_icache_invalidate((char *)start, stop - start);
+#elif defined(__GNUC__)
     __builtin___clear_cache((char *)start, (char *)stop);
+#else
+#error "Missing builtin to flush instruction cache"
+#endif
 }
 
 void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
block.c +1 -1
index 430edf79bb..5d49869d02 100644
@@ -53,7 +53,7 @@
 #ifdef CONFIG_BSD
 #include <sys/ioctl.h>
 #include <sys/queue.h>
-#ifndef __DragonFly__
+#if !defined(__DragonFly__) && !defined(CONFIG_IOS)
 #include <sys/disk.h>
 #endif
 #endif
block/file-posix.c +18 -12
index 52f7c20525..cdc73b5f1d 100644
@@ -181,7 +181,16 @@ typedef struct BDRVRawReopenState {
     bool check_cache_dropped;
 } BDRVRawReopenState;
 
-static int fd_open(BlockDriverState *bs);
+static int fd_open(BlockDriverState *bs)
+{
+    BDRVRawState *s = bs->opaque;
+
+    /* this is just to ensure s->fd is sane (its called by io ops) */
+    if (s->fd >= 0)
+        return 0;
+    return -EIO;
+}
+
 static int64_t raw_getlength(BlockDriverState *bs);
 
 typedef struct RawPosixAIOData {
@@ -252,6 +261,12 @@ static int raw_normalize_devicepath(const char **filename, Error **errp)
 }
 #endif
 
+#if defined(CONFIG_IOS)
+static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
+{
+    return -ENOTSUP; /* not supported on iOS */
+}
+#else /* CONFIG_IOS */
 /*
  * Get logical block size via ioctl. On success store it in @sector_size_p.
  */
@@ -284,6 +299,7 @@ static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
 
     return success ? 0 : -errno;
 }
+#endif /* !CONFIG_IOS */
 
 /**
  * Get physical block size of @fd.
@@ -2306,7 +2322,7 @@ again:
         }
         if (size == 0)
 #endif
-#if defined(__APPLE__) && defined(__MACH__)
+#if !defined(CONFIG_IOS) && defined(__APPLE__) && defined(__MACH__)
         {
             uint64_t sectors = 0;
             uint32_t sector_size = 0;
@@ -3541,16 +3557,6 @@ hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 }
 #endif /* linux */
 
-static int fd_open(BlockDriverState *bs)
-{
-    BDRVRawState *s = bs->opaque;
-
-    /* this is just to ensure s->fd is sane (its called by io ops) */
-    if (s->fd >= 0)
-        return 0;
-    return -EIO;
-}
-
 static coroutine_fn int
 hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
 {
net/slirp.c +8 -8
index 77042e6df7..8413042c09 100644
@@ -27,7 +27,7 @@
 #include "net/slirp.h"
 
 
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
 #include <pwd.h>
 #include <sys/wait.h>
 #endif
@@ -90,7 +90,7 @@ typedef struct SlirpState {
     Slirp *slirp;
     Notifier poll_notifier;
     Notifier exit_notifier;
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
     gchar *smb_dir;
 #endif
     GSList *fwd;
@@ -103,7 +103,7 @@ static QTAILQ_HEAD(, SlirpState) slirp_stacks =
 static int slirp_hostfwd(SlirpState *s, const char *redir_str, Error **errp);
 static int slirp_guestfwd(SlirpState *s, const char *config_str, Error **errp);
 
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
 static int slirp_smb(SlirpState *s, const char *exported_dir,
                      struct in_addr vserver_addr, Error **errp);
 static void slirp_smb_cleanup(SlirpState *s);
@@ -368,7 +368,7 @@ static int net_slirp_init(NetClientState *peer, const char *model,
     struct in6_addr ip6_prefix;
     struct in6_addr ip6_host;
     struct in6_addr ip6_dns;
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
     struct in_addr smbsrv = { .s_addr = 0 };
 #endif
     NetClientState *nc;
@@ -478,7 +478,7 @@ static int net_slirp_init(NetClientState *peer, const char *model,
         return -1;
     }
 
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
     if (vsmbserver && !inet_aton(vsmbserver, &smbsrv)) {
         error_setg(errp, "Failed to parse SMB address");
         return -1;
@@ -593,7 +593,7 @@ static int net_slirp_init(NetClientState *peer, const char *model,
             }
         }
     }
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
     if (smb_export) {
         if (slirp_smb(s, smb_export, smbsrv, errp) < 0) {
             goto error;
@@ -785,7 +785,7 @@ void hmp_hostfwd_add(Monitor *mon, const QDict *qdict)
 
 }
 
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(CONFIG_IOS)
 
 /* automatic user mode samba server configuration */
 static void slirp_smb_cleanup(SlirpState *s)
@@ -900,7 +900,7 @@ static int slirp_smb(SlirpState* s, const char *exported_dir,
     return 0;
 }
 
-#endif /* !defined(_WIN32) */
+#endif /* !defined(_WIN32) && !defined(CONFIG_IOS) */
 
 static int guestfwd_can_read(void *opaque)
 {
qga/commands-posix.c +6 -0
index 3bffee99d4..ebb63b2188 100644
@@ -34,6 +34,12 @@
 
 #ifndef CONFIG_HAS_ENVIRON
 #ifdef __APPLE__
+#include "TargetConditionals.h"
+#if !TARGET_OS_IPHONE && !TARGET_IPHONE_SIMULATOR
+#define APPLE_USE_CRT_EXTERNS
+#endif
+#endif
+#ifdef APPLE_USE_CRT_EXTERNS
 #include <crt_externs.h>
 #define environ (*_NSGetEnviron())
 #else
target/arm/arm-semi.c +2 -0
index 8718fd0194..3704f19df6 100644
@@ -891,6 +891,7 @@ target_ulong do_arm_semihosting(CPUARMState *env)
         return clock() / (CLOCKS_PER_SEC / 100);
     case TARGET_SYS_TIME:
         return set_swi_errno(env, time(NULL));
+#if !defined(CONFIG_IOS) /* iOS does not have system() */
     case TARGET_SYS_SYSTEM:
         GET_ARG(0);
         GET_ARG(1);
@@ -907,6 +908,7 @@ target_ulong do_arm_semihosting(CPUARMState *env)
             unlock_user(s, arg0, 0);
             return ret;
         }
+#endif /* CONFIG_IOS */
     case TARGET_SYS_ERRNO:
         return get_swi_errno(env);
     case TARGET_SYS_GET_CMDLINE:
target/m68k/m68k-semi.c +2 -0
index 8e5fbfc8fa..6b8941839e 100644
@@ -402,6 +402,7 @@ void do_m68k_semihosting(CPUM68KState *env, int nr)
             result = isatty(arg0);
         }
         break;
+#if !defined(CONFIG_IOS) /* iOS does not have system() */
     case HOSTED_SYSTEM:
         GET_ARG(0);
         GET_ARG(1);
@@ -420,6 +421,7 @@ void do_m68k_semihosting(CPUM68KState *env, int nr)
             }
         }
         break;
+#endif /* CONFIG_IOS */
     case HOSTED_INIT_SIM:
 #if defined(CONFIG_USER_ONLY)
         {
target/nios2/nios2-semi.c +2 -0
index d7a80dd303..bb029070d3 100644
@@ -426,6 +426,7 @@ void do_nios2_semihosting(CPUNios2State *env)
             result = isatty(arg0);
         }
         break;
+#if !defined(CONFIG_IOS) /* iOS does not have system() */
     case HOSTED_SYSTEM:
         GET_ARG(0);
         GET_ARG(1);
@@ -444,6 +445,7 @@ void do_nios2_semihosting(CPUNios2State *env)
             }
         }
         break;
+#endif
     default:
         qemu_log_mask(LOG_GUEST_ERROR, "nios2-semihosting: unsupported "
                       "semihosting syscall %d\n", nr);
tests/qtest/meson.build +3 -4
index 0f32ca0895..7184edcafa 100644
@@ -46,12 +46,11 @@ qtests_i386 = \
   (config_all_devices.has_key('CONFIG_TPM_TIS_ISA') ? ['tpm-tis-test'] : []) +              \
   (config_all_devices.has_key('CONFIG_TPM_TIS_ISA') ? ['tpm-tis-swtpm-test'] : []) +        \
   (config_all_devices.has_key('CONFIG_RTL8139_PCI') ? ['rtl8139-test'] : []) +              \
+  (not config_host.has_key('CONFIG_IOS') ? ['bios-tables-test', 'hd-geo-test'] : []) +      \
   qtests_pci +                                                                              \
   ['fdc-test',
    'ide-test',
-   'hd-geo-test',
    'boot-order-test',
-   'bios-tables-test',
    'rtc-test',
    'i440fx-test',
    'fw_cfg-test',
@@ -141,9 +140,9 @@ qtests_arm = \
    'boot-serial-test',
    'hexloader-test']
 
-# TODO: once aarch64 TCG is fixed on ARM 32 bit host, make bios-tables-test unconditional
+# TODO: once aarch64 TCG is fixed on ARM 32 bit host, make bios-tables-test unconditional (except on iOS)
 qtests_aarch64 = \
-  (cpu != 'arm' ? ['bios-tables-test'] : []) +                                                  \
+  (cpu != 'arm' and not config_host.has_key('CONFIG_IOS') ? ['bios-tables-test'] : []) +        \
   (config_all_devices.has_key('CONFIG_TPM_TIS_SYSBUS') ? ['tpm-tis-device-test'] : []) +        \
   (config_all_devices.has_key('CONFIG_TPM_TIS_SYSBUS') ? ['tpm-tis-device-swtpm-test'] : []) +  \
   ['arm-cpu-features',
Signature · 1 line
2.24.3 (Apple Git-128)
#15 lore raw

PATCH v2 4/9 coroutine: add libucontext as external library

In reply to Joelle van Dyne

From: osy <[email protected]>

iOS does not support ucontext natively for aarch64 and the sigaltstack is also unsupported (even worse, it fails silently, see:
https://openradar.appspot.com/13002712 )

As a workaround we include a library implementation of ucontext and add it as a build option.

---
 configure                 | 23 ++++++++++++++++++++---
 meson.build               | 29 ++++++++++++++++++++++++++++-
 util/coroutine-ucontext.c |  9 +++++++++
 .gitmodules               |  3 +++
 libucontext               |  1 +
 meson_options.txt         |  2 ++
 6 files changed, 63 insertions(+), 4 deletions(-)
 create mode 160000 libucontext
configure +20 -3
index c474d7c221..0b7e25e7a5 100755
@@ -1756,7 +1756,7 @@ Advanced options (experts only):
   --oss-lib                path to OSS library
   --cpu=CPU                Build for host CPU [$cpu]
   --with-coroutine=BACKEND coroutine backend. Supported options:
-                           ucontext, sigaltstack, windows
+                           ucontext, libucontext, sigaltstack, windows
   --enable-gcov            enable test coverage analysis with gcov
   --disable-blobs          disable installing provided firmware blobs
   --with-vss-sdk=SDK-path  enable Windows VSS support in QEMU Guest Agent
@@ -5058,6 +5058,8 @@ if test "$coroutine" = ""; then
     coroutine=win32
   elif test "$ucontext_works" = "yes"; then
     coroutine=ucontext
+  elif test "$ios" = "yes"; then
+    coroutine=libucontext
   else
     coroutine=sigaltstack
   fi
@@ -5081,12 +5083,27 @@ else
       error_exit "only the 'windows' coroutine backend is valid for Windows"
     fi
     ;;
+  libucontext)
+  ;;
   *)
     error_exit "unknown coroutine backend $coroutine"
     ;;
   esac
 fi
 
+case $coroutine in
+libucontext)
+  git_submodules="${git_submodules} libucontext"
+  mkdir -p libucontext
+  coroutine_impl=ucontext
+  libucontext="enabled"
+  ;;
+*)
+  coroutine_impl=$coroutine
+  libucontext="disabled"
+  ;;
+esac
+
 if test "$coroutine_pool" = ""; then
   coroutine_pool=yes
 fi
@@ -6676,7 +6693,7 @@ if test "$rbd" = "yes" ; then
   echo "RBD_LIBS=$rbd_libs" >> $config_host_mak
 fi
 
-echo "CONFIG_COROUTINE_BACKEND=$coroutine" >> $config_host_mak
+echo "CONFIG_COROUTINE_BACKEND=$coroutine_impl" >> $config_host_mak
 if test "$coroutine_pool" = "yes" ; then
   echo "CONFIG_COROUTINE_POOL=1" >> $config_host_mak
 else
@@ -7273,7 +7290,7 @@ NINJA=${ninja:-$PWD/ninjatool} $meson setup \
 	-Dcocoa=$cocoa -Dmpath=$mpath -Dsdl=$sdl -Dsdl_image=$sdl_image \
 	-Dvnc=$vnc -Dvnc_sasl=$vnc_sasl -Dvnc_jpeg=$vnc_jpeg -Dvnc_png=$vnc_png \
 	-Dgettext=$gettext -Dxkbcommon=$xkbcommon -Du2f=$u2f \
-	-Dcapstone=$capstone -Dslirp=$slirp -Dfdt=$fdt \
+	-Dcapstone=$capstone -Dslirp=$slirp -Dfdt=$fdt -Ducontext=$libucontext \
         $cross_arg \
         "$PWD" "$source_path"
 
meson.build +28 -1
index 69a3c00cce..e3ff35f46b 100644
@@ -1099,9 +1099,35 @@ if not fdt.found() and fdt_required.length() > 0
   error('fdt not available but required by targets ' + ', '.join(fdt_required))
 endif
 
+ucontext = not_found
+slirp_opt = 'disabled'
+if get_option('ucontext').enabled()
+  if not fs.is_dir(meson.current_source_dir() / 'libucontext/arch' / cpu)
+    error('libucontext is wanted but not implemented for host ' + cpu)
+  endif
+  arch = host_machine.cpu()
+  ucontext_cargs = ['-DG_LOG_DOMAIN="ucontext"', '-DCUSTOM_IMPL']
+  ucontext_files = [
+    'libucontext/arch' / arch / 'getcontext.S',
+    'libucontext/arch' / arch / 'setcontext.S',
+    'libucontext/arch' / arch / 'makecontext.c',
+    'libucontext/arch' / arch / 'startcontext.S',
+    'libucontext/arch' / arch / 'swapcontext.S',
+  ]
+
+  ucontext_inc = include_directories('libucontext/include')
+  libucontext = static_library('ucontext',
+                               sources: ucontext_files,
+                               c_args: ucontext_cargs,
+                               include_directories: ucontext_inc)
+  ucontext = declare_dependency(link_with: libucontext,
+                                include_directories: ucontext_inc)
+endif
+
 config_host_data.set('CONFIG_CAPSTONE', capstone.found())
 config_host_data.set('CONFIG_FDT', fdt.found())
 config_host_data.set('CONFIG_SLIRP', slirp.found())
+config_host_data.set('CONFIG_LIBUCONTEXT', ucontext.found())
 
 genh += configure_file(output: 'config-host.h', configuration: config_host_data)
 
@@ -1321,7 +1347,7 @@ util_ss.add_all(trace_ss)
 util_ss = util_ss.apply(config_all, strict: false)
 libqemuutil = static_library('qemuutil',
                              sources: util_ss.sources() + stub_ss.sources() + genh,
-                             dependencies: [util_ss.dependencies(), m, glib, socket, malloc])
+                             dependencies: [util_ss.dependencies(), m, glib, socket, malloc, ucontext])
 qemuutil = declare_dependency(link_with: libqemuutil,
                               sources: genh + version_res)
 
@@ -1916,6 +1942,7 @@ if targetos == 'windows'
   summary_info += {'QGA MSI support':   config_host.has_key('CONFIG_QGA_MSI')}
 endif
 summary_info += {'seccomp support':   config_host.has_key('CONFIG_SECCOMP')}
+summary_info += {'libucontext support': ucontext.found()}
 summary_info += {'coroutine backend': config_host['CONFIG_COROUTINE_BACKEND']}
 summary_info += {'coroutine pool':    config_host['CONFIG_COROUTINE_POOL'] == '1'}
 summary_info += {'debug stack usage': config_host.has_key('CONFIG_DEBUG_STACK_USAGE')}
util/coroutine-ucontext.c +9 -0
index 904b375192..1e1dd43512 100644
@@ -23,7 +23,16 @@
 #undef _FORTIFY_SOURCE
 #endif
 #include "qemu/osdep.h"
+#if defined(CONFIG_LIBUCONTEXT)
+#include <libucontext.h>
+#define ucontext_t libucontext_ucontext_t
+#define getcontext libucontext_getcontext
+#define setcontext libucontext_setcontext
+#define swapcontext libucontext_swapcontext
+#define makecontext libucontext_makecontext
+#else
 #include <ucontext.h>
+#endif
 #include "qemu/coroutine_int.h"
 
 #ifdef CONFIG_VALGRIND_H
.gitmodules +3 -0
index 2bdeeacef8..065b52867f 100644
@@ -64,3 +64,6 @@
 [submodule "roms/vbootrom"]
 	path = roms/vbootrom
 	url = https://git.qemu.org/git/vbootrom.git
+[submodule "libucontext"]
+	path = libucontext
+	url = https://github.com/utmapp/libucontext.git
libucontext new file +1 -0
new file mode 160000
index 0000000000..7094e4c427
@@ -0,0 +1 @@
+Subproject commit 7094e4c42723b6178a4e2b60d4631d8a88f40719
meson_options.txt +2 -0
index 1d3c94840a..a04a00649c 100644
@@ -48,6 +48,8 @@ option('vnc_sasl', type : 'feature', value : 'auto',
        description: 'SASL authentication for VNC server')
 option('xkbcommon', type : 'feature', value : 'auto',
        description: 'xkbcommon support')
+option('ucontext', type : 'feature', value : 'disabled',
+       description: 'libucontext support')
 
 option('capstone', type: 'combo', value: 'auto',
        choices: ['disabled', 'enabled', 'auto', 'system', 'internal'],
Signature · 1 line
2.24.3 (Apple Git-128)
#16 lore raw

PATCH v2 5/9 tcg: add const hints for code pointers

In reply to Joelle van Dyne

From: osy <[email protected]>

We will introduce mirror mapping for JIT segment with separate RX and RW access. Adding 'const' hints will make it easier to identify read-only accesses and allow us to easier catch bugs at compile time in the future.

---
 include/tcg/tcg.h            |  8 ++++----
 tcg/tcg.c                    |  4 ++--
 tcg/aarch64/tcg-target.c.inc | 19 +++++++++++--------
 tcg/arm/tcg-target.c.inc     | 12 +++++++-----
 tcg/i386/tcg-target.c.inc    | 10 +++++-----
 tcg/mips/tcg-target.c.inc    | 33 +++++++++++++++++++--------------
 tcg/ppc/tcg-target.c.inc     | 21 +++++++++++++--------
 tcg/riscv/tcg-target.c.inc   | 11 ++++++-----
 tcg/s390/tcg-target.c.inc    |  9 +++++----
 tcg/sparc/tcg-target.c.inc   | 10 +++++-----
 tcg/tcg-ldst.c.inc           |  2 +-
 tcg/tci/tcg-target.c.inc     |  2 +-
 12 files changed, 79 insertions(+), 62 deletions(-)
include/tcg/tcg.h +4 -4
index 8804a8c4a2..79c5ff8dab 100644
@@ -261,7 +261,7 @@ struct TCGLabel {
     unsigned refs : 16;
     union {
         uintptr_t value;
-        tcg_insn_unit *value_ptr;
+        const tcg_insn_unit *value_ptr;
     } u;
     QSIMPLEQ_HEAD(, TCGRelocation) relocs;
     QSIMPLEQ_ENTRY(TCGLabel) next;
@@ -593,7 +593,7 @@ struct TCGContext {
     int nb_ops;
 
     /* goto_tb support */
-    tcg_insn_unit *code_buf;
+    const tcg_insn_unit *code_buf;
     uint16_t *tb_jmp_reset_offset; /* tb->jmp_reset_offset */
     uintptr_t *tb_jmp_insn_offset; /* tb->jmp_target_arg if direct_jump */
     uintptr_t *tb_jmp_target_addr; /* tb->jmp_target_arg if !direct_jump */
@@ -1099,7 +1099,7 @@ static inline TCGLabel *arg_label(TCGArg i)
  * correct result.
  */
 
-static inline ptrdiff_t tcg_ptr_byte_diff(void *a, void *b)
+static inline ptrdiff_t tcg_ptr_byte_diff(const void *a, const void *b)
 {
     return a - b;
 }
@@ -1113,7 +1113,7 @@ static inline ptrdiff_t tcg_ptr_byte_diff(void *a, void *b)
  * to the destination address.
  */
 
-static inline ptrdiff_t tcg_pcrel_diff(TCGContext *s, void *target)
+static inline ptrdiff_t tcg_pcrel_diff(TCGContext *s, const void *target)
 {
     return tcg_ptr_byte_diff(target, s->code_ptr);
 }
tcg/tcg.c +2 -2
index a8c28440e2..bb890c506d 100644
@@ -148,7 +148,7 @@ static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1,
                        intptr_t arg2);
 static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
                         TCGReg base, intptr_t ofs);
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *target);
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target);
 static int tcg_target_const_match(tcg_target_long val, TCGType type,
                                   const TCGArgConstraint *arg_ct);
 #ifdef TCG_TARGET_NEED_LDST_LABELS
@@ -295,7 +295,7 @@ static void tcg_out_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
     QSIMPLEQ_INSERT_TAIL(&l->relocs, r, next);
 }
 
-static void tcg_out_label(TCGContext *s, TCGLabel *l, tcg_insn_unit *ptr)
+static void tcg_out_label(TCGContext *s, TCGLabel *l, const tcg_insn_unit *ptr)
 {
     tcg_debug_assert(!l->has_value);
     l->has_value = 1;
tcg/aarch64/tcg-target.c.inc +11 -8
index 26f71cb599..1aa5f37fc6 100644
@@ -78,7 +78,8 @@ static const int tcg_target_call_oarg_regs[1] = {
 #define TCG_REG_GUEST_BASE TCG_REG_X28
 #endif
 
-static inline bool reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static inline bool reloc_pc26(tcg_insn_unit *code_ptr,
+                              const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - code_ptr;
     if (offset == sextract64(offset, 0, 26)) {
@@ -90,7 +91,8 @@ static inline bool reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
     return false;
 }
 
-static inline bool reloc_pc19(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static inline bool reloc_pc19(tcg_insn_unit *code_ptr,
+                              const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - code_ptr;
     if (offset == sextract64(offset, 0, 19)) {
@@ -1306,14 +1308,14 @@ static void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg a,
     }
 }
 
-static inline void tcg_out_goto(TCGContext *s, tcg_insn_unit *target)
+static inline void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - s->code_ptr;
     tcg_debug_assert(offset == sextract64(offset, 0, 26));
     tcg_out_insn(s, 3206, B, offset);
 }
 
-static inline void tcg_out_goto_long(TCGContext *s, tcg_insn_unit *target)
+static inline void tcg_out_goto_long(TCGContext *s, const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - s->code_ptr;
     if (offset == sextract64(offset, 0, 26)) {
@@ -1329,7 +1331,7 @@ static inline void tcg_out_callr(TCGContext *s, TCGReg reg)
     tcg_out_insn(s, 3207, BLR, reg);
 }
 
-static inline void tcg_out_call(TCGContext *s, tcg_insn_unit *target)
+static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - s->code_ptr;
     if (offset == sextract64(offset, 0, 26)) {
@@ -1568,7 +1570,7 @@ static void * const qemu_st_helpers[16] = {
     [MO_BEQ]  = helper_be_stq_mmu,
 };
 
-static inline void tcg_out_adr(TCGContext *s, TCGReg rd, void *target)
+static inline void tcg_out_adr(TCGContext *s, TCGReg rd, const void *target)
 {
     ptrdiff_t offset = tcg_pcrel_diff(s, target);
     tcg_debug_assert(offset == sextract64(offset, 0, 21));
@@ -1622,7 +1624,8 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
 
 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
                                 TCGType ext, TCGReg data_reg, TCGReg addr_reg,
-                                tcg_insn_unit *raddr, tcg_insn_unit *label_ptr)
+                                const tcg_insn_unit *raddr,
+                                tcg_insn_unit *label_ptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
 
@@ -1849,7 +1852,7 @@ static void tcg_out_qemu_st(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
 #endif /* CONFIG_SOFTMMU */
 }
 
-static tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *tb_ret_addr;
 
 static void tcg_out_op(TCGContext *s, TCGOpcode opc,
                        const TCGArg args[TCG_MAX_OP_ARGS],
tcg/arm/tcg-target.c.inc +7 -5
index 62c37a954b..5ceb06a971 100644
@@ -187,7 +187,8 @@ static const uint8_t tcg_cond_to_arm_cond[] = {
     [TCG_COND_GTU] = COND_HI,
 };
 
-static inline bool reloc_pc24(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static inline bool reloc_pc24(tcg_insn_unit *code_ptr,
+                              const tcg_insn_unit *target)
 {
     ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
     if (offset == sextract32(offset, 0, 24)) {
@@ -197,7 +198,8 @@ static inline bool reloc_pc24(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
     return false;
 }
 
-static inline bool reloc_pc13(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
+                              const tcg_insn_unit *target)
 {
     ptrdiff_t offset = tcg_ptr_byte_diff(target, code_ptr) - 8;
 
@@ -1019,7 +1021,7 @@ static inline void tcg_out_st8(TCGContext *s, int cond,
  * with the code buffer limited to 16MB we wouldn't need the long case.
  * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
  */
-static void tcg_out_goto(TCGContext *s, int cond, tcg_insn_unit *addr)
+static void tcg_out_goto(TCGContext *s, int cond, const tcg_insn_unit *addr)
 {
     intptr_t addri = (intptr_t)addr;
     ptrdiff_t disp = tcg_pcrel_diff(s, addr);
@@ -1033,7 +1035,7 @@ static void tcg_out_goto(TCGContext *s, int cond, tcg_insn_unit *addr)
 
 /* The call case is mostly used for helpers - so it's not unreasonable
  * for them to be beyond branch range */
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *addr)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr)
 {
     intptr_t addri = (intptr_t)addr;
     ptrdiff_t disp = tcg_pcrel_diff(s, addr);
@@ -1326,7 +1328,7 @@ static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
    helper code.  */
 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
                                 TCGReg datalo, TCGReg datahi, TCGReg addrlo,
-                                TCGReg addrhi, tcg_insn_unit *raddr,
+                                TCGReg addrhi, const tcg_insn_unit *raddr,
                                 tcg_insn_unit *label_ptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
tcg/i386/tcg-target.c.inc +5 -5
index d8797ed398..69acb51f5d 100644
@@ -165,7 +165,7 @@ static bool have_lzcnt;
 # define have_lzcnt 0
 #endif
 
-static tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *tb_ret_addr;
 
 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
@@ -1591,7 +1591,7 @@ static void tcg_out_clz(TCGContext *s, int rexw, TCGReg dest, TCGReg arg1,
     }
 }
 
-static void tcg_out_branch(TCGContext *s, int call, tcg_insn_unit *dest)
+static void tcg_out_branch(TCGContext *s, int call, const tcg_insn_unit *dest)
 {
     intptr_t disp = tcg_pcrel_diff(s, dest) - 5;
 
@@ -1610,12 +1610,12 @@ static void tcg_out_branch(TCGContext *s, int call, tcg_insn_unit *dest)
     }
 }
 
-static inline void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
+static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest)
 {
     tcg_out_branch(s, 1, dest);
 }
 
-static void tcg_out_jmp(TCGContext *s, tcg_insn_unit *dest)
+static void tcg_out_jmp(TCGContext *s, const tcg_insn_unit *dest)
 {
     tcg_out_branch(s, 0, dest);
 }
@@ -1774,7 +1774,7 @@ static void add_qemu_ldst_label(TCGContext *s, bool is_ld, bool is_64,
                                 TCGMemOpIdx oi,
                                 TCGReg datalo, TCGReg datahi,
                                 TCGReg addrlo, TCGReg addrhi,
-                                tcg_insn_unit *raddr,
+                                const tcg_insn_unit *raddr,
                                 tcg_insn_unit **label_ptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
tcg/mips/tcg-target.c.inc +19 -14
index 41be574e89..ace413afde 100644
@@ -139,12 +139,13 @@ static const TCGReg tcg_target_call_oarg_regs[2] = {
     TCG_REG_V1
 };
 
-static tcg_insn_unit *tb_ret_addr;
-static tcg_insn_unit *bswap32_addr;
-static tcg_insn_unit *bswap32u_addr;
-static tcg_insn_unit *bswap64_addr;
+static const tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *bswap32_addr;
+static const tcg_insn_unit *bswap32u_addr;
+static const tcg_insn_unit *bswap64_addr;
 
-static inline uint32_t reloc_pc16_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+static inline uint32_t reloc_pc16_val(const tcg_insn_unit *pc,
+                                      const tcg_insn_unit *target)
 {
     /* Let the compiler perform the right-shift as part of the arithmetic.  */
     ptrdiff_t disp = target - (pc + 1);
@@ -152,18 +153,21 @@ static inline uint32_t reloc_pc16_val(tcg_insn_unit *pc, tcg_insn_unit *target)
     return disp & 0xffff;
 }
 
-static inline void reloc_pc16(tcg_insn_unit *pc, tcg_insn_unit *target)
+static inline void reloc_pc16(tcg_insn_unit *pc,
+                              const tcg_insn_unit *target)
 {
     *pc = deposit32(*pc, 0, 16, reloc_pc16_val(pc, target));
 }
 
-static inline uint32_t reloc_26_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
+                                    const tcg_insn_unit *target)
 {
     tcg_debug_assert((((uintptr_t)pc ^ (uintptr_t)target) & 0xf0000000) == 0);
     return ((uintptr_t)target >> 2) & 0x3ffffff;
 }
 
-static inline void reloc_26(tcg_insn_unit *pc, tcg_insn_unit *target)
+static inline void reloc_26(tcg_insn_unit *pc,
+                            const tcg_insn_unit *target)
 {
     *pc = deposit32(*pc, 0, 26, reloc_26_val(pc, target));
 }
@@ -516,7 +520,7 @@ static void tcg_out_opc_sa64(TCGContext *s, MIPSInsn opc1, MIPSInsn opc2,
  * Type jump.
  * Returns true if the branch was in range and the insn was emitted.
  */
-static bool tcg_out_opc_jmp(TCGContext *s, MIPSInsn opc, void *target)
+static bool tcg_out_opc_jmp(TCGContext *s, MIPSInsn opc, const void *target)
 {
     uintptr_t dest = (uintptr_t)target;
     uintptr_t from = (uintptr_t)s->code_ptr + 4;
@@ -631,7 +635,7 @@ static inline void tcg_out_bswap16s(TCGContext *s, TCGReg ret, TCGReg arg)
     }
 }
 
-static void tcg_out_bswap_subr(TCGContext *s, tcg_insn_unit *sub)
+static void tcg_out_bswap_subr(TCGContext *s, const tcg_insn_unit *sub)
 {
     bool ok = tcg_out_opc_jmp(s, OPC_JAL, sub);
     tcg_debug_assert(ok);
@@ -1079,7 +1083,7 @@ static void tcg_out_movcond(TCGContext *s, TCGCond cond, TCGReg ret,
     }
 }
 
-static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
+static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
 {
     /* Note that the ABI requires the called function's address to be
        loaded into T9, even if a direct branch is in range.  */
@@ -1097,7 +1101,7 @@ static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
     }
 }
 
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *arg)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg)
 {
     tcg_out_call_int(s, arg, false);
     tcg_out_nop(s);
@@ -1289,7 +1293,8 @@ static void add_qemu_ldst_label(TCGContext *s, int is_ld, TCGMemOpIdx oi,
                                 TCGType ext,
                                 TCGReg datalo, TCGReg datahi,
                                 TCGReg addrlo, TCGReg addrhi,
-                                void *raddr, tcg_insn_unit *label_ptr[2])
+                                const tcg_insn_unit *raddr,
+                                tcg_insn_unit *label_ptr[2])
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
 
@@ -2430,7 +2435,7 @@ static void tcg_target_detect_isa(void)
     sigaction(SIGILL, &sa_old, NULL);
 }
 
-static tcg_insn_unit *align_code_ptr(TCGContext *s)
+static const tcg_insn_unit *align_code_ptr(TCGContext *s)
 {
     uintptr_t p = (uintptr_t)s->code_ptr;
     if (p & 15) {
tcg/ppc/tcg-target.c.inc +13 -8
index 18ee989f95..289388b603 100644
@@ -62,7 +62,7 @@
 #define TCG_CT_CONST_MONE 0x2000
 #define TCG_CT_CONST_WSZ  0x4000
 
-static tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *tb_ret_addr;
 
 TCGPowerISA have_isa;
 static bool have_isel;
@@ -184,14 +184,16 @@ static inline bool in_range_b(tcg_target_long target)
     return target == sextract64(target, 0, 26);
 }
 
-static uint32_t reloc_pc24_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
+                               const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     tcg_debug_assert(in_range_b(disp));
     return disp & 0x3fffffc;
 }
 
-static bool reloc_pc24(tcg_insn_unit *pc, tcg_insn_unit *target)
+static bool reloc_pc24(tcg_insn_unit *pc,
+                       const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     if (in_range_b(disp)) {
@@ -201,14 +203,16 @@ static bool reloc_pc24(tcg_insn_unit *pc, tcg_insn_unit *target)
     return false;
 }
 
-static uint16_t reloc_pc14_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
+                               const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     tcg_debug_assert(disp == (int16_t) disp);
     return disp & 0xfffc;
 }
 
-static bool reloc_pc14(tcg_insn_unit *pc, tcg_insn_unit *target)
+static bool reloc_pc14(tcg_insn_unit *pc,
+                       const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     if (disp == (int16_t) disp) {
@@ -1106,7 +1110,7 @@ static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
     tcg_out_zori32(s, dst, src, c, XORI, XORIS);
 }
 
-static void tcg_out_b(TCGContext *s, int mask, tcg_insn_unit *target)
+static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_pcrel_diff(s, target);
     if (in_range_b(disp)) {
@@ -1762,7 +1766,7 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
     }
 }
 
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *target)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
 {
 #ifdef _CALL_AIX
     /* Look through the descriptor.  If the branch is in range, and we
@@ -1987,7 +1991,8 @@ static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
                                 TCGReg datalo_reg, TCGReg datahi_reg,
                                 TCGReg addrlo_reg, TCGReg addrhi_reg,
-                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
+                                const tcg_insn_unit *raddr,
+                                tcg_insn_unit *lptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
 
tcg/riscv/tcg-target.c.inc +6 -5
index d536f3ccc1..5432ee89db 100644
@@ -854,14 +854,14 @@ static void tcg_out_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
     g_assert_not_reached();
 }
 
-static inline void tcg_out_goto(TCGContext *s, tcg_insn_unit *target)
+static inline void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
 {
     ptrdiff_t offset = tcg_pcrel_diff(s, target);
     tcg_debug_assert(offset == sextreg(offset, 1, 20) << 1);
     tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, offset);
 }
 
-static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
+static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
 {
     TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
     ptrdiff_t offset = tcg_pcrel_diff(s, arg);
@@ -888,7 +888,7 @@ static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
     }
 }
 
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *arg)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg)
 {
     tcg_out_call_int(s, arg, false);
 }
@@ -1022,7 +1022,8 @@ static void add_qemu_ldst_label(TCGContext *s, int is_ld, TCGMemOpIdx oi,
                                 TCGType ext,
                                 TCGReg datalo, TCGReg datahi,
                                 TCGReg addrlo, TCGReg addrhi,
-                                void *raddr, tcg_insn_unit **label_ptr)
+                                const tcg_insn_unit *raddr,
+                                tcg_insn_unit **label_ptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
 
@@ -1274,7 +1275,7 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
 #endif
 }
 
-static tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *tb_ret_addr;
 
 static void tcg_out_op(TCGContext *s, TCGOpcode opc,
                        const TCGArg *args, const int *const_args)
tcg/s390/tcg-target.c.inc +5 -4
index c5e096449b..7c6917f3c4 100644
@@ -363,7 +363,7 @@ static void * const qemu_st_helpers[16] = {
 };
 #endif
 
-static tcg_insn_unit *tb_ret_addr;
+static const tcg_insn_unit *tb_ret_addr;
 uint64_t s390_facilities;
 
 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
@@ -1302,7 +1302,7 @@ static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
     tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
 }
 
-static void tgen_gotoi(TCGContext *s, int cc, tcg_insn_unit *dest)
+static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
 {
     ptrdiff_t off = dest - s->code_ptr;
     if (off == (int16_t)off) {
@@ -1415,7 +1415,7 @@ static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
     tgen_branch(s, cc, l);
 }
 
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest)
 {
     ptrdiff_t off = dest - s->code_ptr;
     if (off == (int32_t)off) {
@@ -1593,7 +1593,8 @@ static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, MemOp opc,
 
 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
                                 TCGReg data, TCGReg addr,
-                                tcg_insn_unit *raddr, tcg_insn_unit *label_ptr)
+                                const tcg_insn_unit *raddr,
+                                tcg_insn_unit *label_ptr)
 {
     TCGLabelQemuLdst *label = new_ldst_label(s);
 
tcg/sparc/tcg-target.c.inc +5 -5
index 6775bd30fc..5eac18b719 100644
@@ -840,7 +840,7 @@ static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
     tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
 }
 
-static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest,
+static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
                                  bool in_prologue)
 {
     ptrdiff_t disp = tcg_pcrel_diff(s, dest);
@@ -855,7 +855,7 @@ static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest,
     }
 }
 
-static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
+static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest)
 {
     tcg_out_call_nodelay(s, dest, false);
     tcg_out_nop(s);
@@ -868,8 +868,8 @@ static void tcg_out_mb(TCGContext *s, TCGArg a0)
 }
 
 #ifdef CONFIG_SOFTMMU
-static tcg_insn_unit *qemu_ld_trampoline[16];
-static tcg_insn_unit *qemu_st_trampoline[16];
+static const tcg_insn_unit *qemu_ld_trampoline[16];
+static const tcg_insn_unit *qemu_st_trampoline[16];
 
 static void emit_extend(TCGContext *s, TCGReg r, int op)
 {
@@ -1163,7 +1163,7 @@ static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
 #ifdef CONFIG_SOFTMMU
     unsigned memi = get_mmuidx(oi);
     TCGReg addrz, param;
-    tcg_insn_unit *func;
+    const tcg_insn_unit *func;
     tcg_insn_unit *label_ptr;
 
     addrz = tcg_out_tlb_load(s, addr, memi, memop,
tcg/tcg-ldst.c.inc +1 -1
index 05f9b3ccd6..eaba08700e 100644
@@ -28,7 +28,7 @@ typedef struct TCGLabelQemuLdst {
     TCGReg addrhi_reg;      /* reg index for high word of guest virtual addr */
     TCGReg datalo_reg;      /* reg index for low word to be loaded or stored */
     TCGReg datahi_reg;      /* reg index for high word to be loaded or stored */
-    tcg_insn_unit *raddr;   /* gen code addr of the next IR of qemu_ld/st IR */
+    const tcg_insn_unit *raddr; /* gen code addr of the next IR of qemu_ld/st */
     tcg_insn_unit *label_ptr[2]; /* label pointers to be updated */
     QSIMPLEQ_ENTRY(TCGLabelQemuLdst) next;
 } TCGLabelQemuLdst;
tcg/tci/tcg-target.c.inc +1 -1
index 231b9b1775..d5a4d9d37c 100644
@@ -545,7 +545,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
     old_code_ptr[1] = s->code_ptr - old_code_ptr;
 }
 
-static inline void tcg_out_call(TCGContext *s, tcg_insn_unit *arg)
+static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg)
 {
     uint8_t *old_code_ptr = s->code_ptr;
     tcg_out_op_t(s, INDEX_op_call);
Signature · 1 line
2.24.3 (Apple Git-128)
#17 lore raw

Re: [PATCH v2 5/9] tcg: add const hints for code pointers

Quoted 7 lines · Joelle van Dyne

From: osy <[email protected]>

We will introduce mirror mapping for JIT segment with separate RX and RW access. Adding 'const' hints will make it easier to identify read-only accesses and allow us to easier catch bugs at compile time in the future.

Are you "osy"? We do need S-o-b with real names from all contributors to the patch.

The patch is bigger than I would like, but it all appears to be strongly connected, and I don't see where it could be split. I'm inclined to cherry-pick this patch out now to avoid carrying it around. So if I can either fix the --author to you, or we can get a s-o-b from the author, I'll do that.

r~

#18 lore raw

Re: [PATCH v2 5/9] tcg: add const hints for code pointers

You can --author "Joelle van Dyne <[email protected]>"

-j

Quoted 20 lines · Richard Henderson
Quoted 7 lines · Joelle van Dyne

From: osy <[email protected]>

We will introduce mirror mapping for JIT segment with separate RX and RW access. Adding 'const' hints will make it easier to identify read-only accesses and allow us to easier catch bugs at compile time in the future.

Are you "osy"? We do need S-o-b with real names from all contributors to the patch.

The patch is bigger than I would like, but it all appears to be strongly connected, and I don't see where it could be split. I'm inclined to cherry-pick this patch out now to avoid carrying it around. So if I can either fix the --author to you, or we can get a s-o-b from the author, I'll do that.

r~

#19 lore raw

Re: [PATCH v2 5/9] tcg: add const hints for code pointers

In reply to Joelle van Dyne
Quoted 7 lines · Joelle van Dyne

From: osy <[email protected]>

We will introduce mirror mapping for JIT segment with separate RX and RW access. Adding 'const' hints will make it easier to identify read-only accesses and allow us to easier catch bugs at compile time in the future.

Quoted 14 lines
---
 include/tcg/tcg.h            |  8 ++++----
 tcg/tcg.c                    |  4 ++--
 tcg/aarch64/tcg-target.c.inc | 19 +++++++++++--------
 tcg/arm/tcg-target.c.inc     | 12 +++++++-----
 tcg/i386/tcg-target.c.inc    | 10 +++++-----
 tcg/mips/tcg-target.c.inc    | 33 +++++++++++++++++++--------------
 tcg/ppc/tcg-target.c.inc     | 21 +++++++++++++--------
 tcg/riscv/tcg-target.c.inc   | 11 ++++++-----
 tcg/s390/tcg-target.c.inc    |  9 +++++----
 tcg/sparc/tcg-target.c.inc   | 10 +++++-----
 tcg/tcg-ldst.c.inc           |  2 +-
 tcg/tci/tcg-target.c.inc     |  2 +-
 12 files changed, 79 insertions(+), 62 deletions(-)

tcg/ppc/tcg-target.c.inc:2349:26: error: assignment discards ‘const’ qualifier from pointer target type [-Werror]

     s->code_gen_epilogue = tb_ret_addr = s->code_ptr;

How many of the targets did you build?

r~

#20 lore raw

Re: [PATCH v2 5/9] tcg: add const hints for code pointers

Seems like I missed a few. Sorry about that. Will fix.

-j

Quoted 32 lines · Richard Henderson
Quoted 7 lines · Joelle van Dyne

From: osy <[email protected]>

We will introduce mirror mapping for JIT segment with separate RX and RW access. Adding 'const' hints will make it easier to identify read-only accesses and allow us to easier catch bugs at compile time in the future.

Quoted 14 lines
---
 include/tcg/tcg.h            |  8 ++++----
 tcg/tcg.c                    |  4 ++--
 tcg/aarch64/tcg-target.c.inc | 19 +++++++++++--------
 tcg/arm/tcg-target.c.inc     | 12 +++++++-----
 tcg/i386/tcg-target.c.inc    | 10 +++++-----
 tcg/mips/tcg-target.c.inc    | 33 +++++++++++++++++++--------------
 tcg/ppc/tcg-target.c.inc     | 21 +++++++++++++--------
 tcg/riscv/tcg-target.c.inc   | 11 ++++++-----
 tcg/s390/tcg-target.c.inc    |  9 +++++----
 tcg/sparc/tcg-target.c.inc   | 10 +++++-----
 tcg/tcg-ldst.c.inc           |  2 +-
 tcg/tci/tcg-target.c.inc     |  2 +-
 12 files changed, 79 insertions(+), 62 deletions(-)

tcg/ppc/tcg-target.c.inc:2349:26: error: assignment discards ‘const’ qualifier from pointer target type [-Werror]

     s->code_gen_epilogue = tb_ret_addr = s->code_ptr;

How many of the targets did you build?

r~

#21 lore raw

Re: [PATCH v2 5/9] tcg: add const hints for code pointers

Quoted 1 line · Joelle van Dyne

Seems like I missed a few. Sorry about that. Will fix.

I'll keep looking, Just In Case. ;-)

Quoted 1 line
Quoted 1 line
     s->code_gen_epilogue = tb_ret_addr = s->code_ptr;

In this case, just splitting the chained assignment to two statements is sufficient to fix.

r~

#22 lore raw

PATCH v2 6/9 tcg: implement mirror mapped JIT for iOS

In reply to Joelle van Dyne

From: osy <[email protected]>

On iOS, we cannot allocate RWX pages without special entitlements. As a workaround, we can allocate a RX region and then mirror map it to a separate RX region. Then we can write to one region and execute from the other one.

We also define tcg_mirror_ptr_rw and tcg_code_ptr_rw to return a pointer to RW memory. The difference between the RW and RX regions is stored in the TCG context.

To ensure cache coherency, we flush the data cache in the RW mapping and then invalidate the instruction cache in the RX mapping (where applicable). Because data cache flush is OS defined on some architectures, we do not provide implementations for non iOS platforms (ARM/x86).

---
 docs/devel/ios.rst           | 40 +++++++++++++++++++
 configure                    |  1 +
 include/exec/exec-all.h      |  8 ++++
 include/tcg/tcg.h            | 17 ++++++++
 tcg/aarch64/tcg-target.h     | 13 +++++-
 tcg/arm/tcg-target.h         |  9 ++++-
 tcg/i386/tcg-target.h        | 24 ++++++++++-
 tcg/mips/tcg-target.h        |  8 +++-
 tcg/ppc/tcg-target.h         |  8 +++-
 tcg/riscv/tcg-target.h       |  9 ++++-
 tcg/s390/tcg-target.h        | 13 +++++-
 tcg/sparc/tcg-target.h       |  8 +++-
 tcg/tci/tcg-target.h         |  9 ++++-
 accel/tcg/cpu-exec.c         |  7 +++-
 accel/tcg/translate-all.c    | 77 ++++++++++++++++++++++++++++++++++--
 tcg/tcg.c                    | 56 +++++++++++++++++---------
 tcg/aarch64/tcg-target.c.inc | 33 ++++++++++------
 tcg/arm/tcg-target.c.inc     | 25 ++++++------
 tcg/i386/tcg-target.c.inc    | 18 ++++-----
 tcg/mips/tcg-target.c.inc    | 35 +++++++++-------
 tcg/ppc/tcg-target.c.inc     | 38 +++++++++++-------
 tcg/riscv/tcg-target.c.inc   | 40 +++++++++++--------
 tcg/s390/tcg-target.c.inc    | 16 ++++----
 tcg/sparc/tcg-target.c.inc   | 23 +++++++----
 tcg/tcg-pool.c.inc           |  9 +++--
 tcg/tci/tcg-target.c.inc     |  6 +--
 26 files changed, 416 insertions(+), 134 deletions(-)
 create mode 100644 docs/devel/ios.rst
docs/devel/ios.rst new file +40 -0
new file mode 100644
index 0000000000..dba9fdd868
@@ -0,0 +1,40 @@
+===========
+iOS Support
+===========
+
+To run qemu on the iOS platform, some modifications were required. Most of the
+modifications are conditioned on the ``CONFIG_IOS`` and ``CONFIG_IOS_JIT``
+configuration variables.
+
+Build support
+-------------
+
+For the code to compile, certain changes in the block driver and the slirp
+driver had to be made. There is no ``system()`` call, so code requiring it had
+to be disabled.
+
+``ucontext`` support is broken on iOS. The implementation from ``libucontext``
+is used instead.
+
+Because ``fork()`` is not allowed on iOS apps, the option to build qemu and the
+utilities as shared libraries is added. Note that because qemu does not perform
+resource cleanup in most cases (open files, allocated memory, etc), it is
+advisable that the user implements a proxy layer for syscalls so resources can
+be kept track by the app that uses qemu as a shared library.
+
+JIT support
+-----------
+
+On iOS, allocating RWX pages require special entitlements not usually granted to
+apps. However, it is possible to use `bulletproof JIT`_ with a development
+certificate. This means that we need to allocate one chunk of memory with RX
+permissions and then mirror map the same memory with RW permissions. We generate
+code to the mirror mapping and execute the original mapping.
+
+With ``CONFIG_IOS_JIT`` defined, we store inside the TCG context the difference
+between the two mappings. Then, we make sure that any writes to JIT memory is
+done to the pointer + the difference (in order to get a pointer to the mirror
+mapped space). Additionally, we make sure to flush the data cache before we
+invalidate the instruction cache so the changes are seen in both mappings.
+
+.. _bulletproof JIT: https://www.blackhat.com/docs/us-16/materials/us-16-Krstic.pdf
configure +1 -0
index 0b7e25e7a5..93d6fd5ce2 100755
@@ -6214,6 +6214,7 @@ fi
 
 if test "$ios" = "yes" ; then
   echo "CONFIG_IOS=y" >> $config_host_mak
+  echo "CONFIG_IOS_JIT=y" >> $config_host_mak
 fi
 
 if test "$solaris" = "yes" ; then
include/exec/exec-all.h +8 -0
index 66f9b4cca6..2db155a772 100644
@@ -483,6 +483,14 @@ struct TranslationBlock {
     uintptr_t jmp_list_head;
     uintptr_t jmp_list_next[2];
     uintptr_t jmp_dest[2];
+
+#if defined(CONFIG_IOS_JIT)
+    /*
+     * Store difference to writable mirror
+     * We need this when patching the jump instructions
+     */
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 };
 
 extern bool parallel_cpus;
include/tcg/tcg.h +17 -0
index 79c5ff8dab..ade01d2e41 100644
@@ -627,6 +627,9 @@ struct TCGContext {
     size_t code_gen_buffer_size;
     void *code_gen_ptr;
     void *data_gen_ptr;
+#if defined(CONFIG_IOS_JIT)
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 
     /* Threshold to flush the translated code buffer.  */
     void *code_gen_highwater;
@@ -677,6 +680,20 @@ struct TCGContext {
     target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
 };
 
+static inline void *tcg_mirror_ptr_rw(TCGContext *s, const void *rx)
+{
+#if defined(CONFIG_IOS_JIT)
+    return (void *)rx + s->code_rw_mirror_diff;
+#else
+    return (void *)rx;
+#endif
+}
+
+static inline tcg_insn_unit *tcg_code_ptr_rw(TCGContext *s, const void *rx)
+{
+    return (tcg_insn_unit *)tcg_mirror_ptr_rw(s, rx);
+}
+
 extern TCGContext tcg_init_ctx;
 extern __thread TCGContext *tcg_ctx;
 extern TCGv_env cpu_env;
tcg/aarch64/tcg-target.h +12 -1
index a2b22b4305..78c97460a1 100644
@@ -150,6 +150,7 @@ typedef enum {
 
 #if defined(__APPLE__)
 void sys_icache_invalidate(void *start, size_t len);
+void sys_dcache_flush(void *start, size_t len);
 #endif
 
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
@@ -163,7 +164,17 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #endif
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/arm/tcg-target.h +8 -1
index 17e771374d..d8d7e7e239 100644
@@ -139,8 +139,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
     __builtin___clear_cache((char *) start, (char *) stop);
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/i386/tcg-target.h +22 -2
index abd4ac7fc0..cdc440ce36 100644
@@ -206,16 +206,36 @@ extern bool have_avx2;
 #define TCG_TARGET_extract_i64_valid(ofs, len) \
     (((ofs) == 8 && (len) == 8) || ((ofs) + (len)) == 32)
 
+#ifdef __APPLE__
+void sys_dcache_flush(void *start, size_t len);
+#endif
+
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
     /* patch the branch destination */
-    qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
+    qatomic_set((int32_t *)wr_addr, addr - (jmp_addr + 4));
     /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    /* we do need to flush mirror dcache */
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 /* This defines the natural memory order supported by this
tcg/mips/tcg-target.h +7 -1
index c6b091d849..80dcba5358 100644
@@ -212,7 +212,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
     cacheflush ((void *)start, stop-start, ICACHE);
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/ppc/tcg-target.h +7 -1
index be10363956..23d7a337c9 100644
@@ -176,7 +176,13 @@ extern bool have_vsx;
 #define TCG_TARGET_HAS_cmpsel_vec       0
 
 void flush_icache_range(uintptr_t start, uintptr_t stop);
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
tcg/riscv/tcg-target.h +8 -1
index 032439d806..d42b361991 100644
@@ -164,8 +164,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
     __builtin___clear_cache((char *)start, (char *)stop);
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #define TCG_TARGET_DEFAULT_MO (0)
 
tcg/s390/tcg-target.h +12 -1
index 63c8797bd3..d67632512d 100644
@@ -149,13 +149,24 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
     /* patch the branch destination */
     intptr_t disp = addr - (jmp_addr - 2);
     qatomic_set((int32_t *)jmp_addr, disp / 2);
     /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 #ifdef CONFIG_SOFTMMU
tcg/sparc/tcg-target.h +7 -1
index 633841ebf2..d102e13692 100644
@@ -176,7 +176,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
     }
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_NEED_POOL_LABELS
 
tcg/tci/tcg-target.h +8 -1
index 8c1c1d265d..2a258ea350 100644
@@ -195,6 +195,12 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+}
+#endif
+
 /* We could notice __i386__ or __s390x__ and reduce the barriers depending
    on the host.  But if you want performance, you use the normal backend.
    We prefer consistency across hosts on this.  */
@@ -203,7 +209,8 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
 
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
     /* patch the branch destination */
     qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
accel/tcg/cpu-exec.c +6 -1
index 58aea605d8..821aefdea2 100644
@@ -354,7 +354,12 @@ void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
     if (TCG_TARGET_HAS_direct_jump) {
         uintptr_t offset = tb->jmp_target_arg[n];
         uintptr_t tc_ptr = (uintptr_t)tb->tc.ptr;
-        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr);
+#if defined(CONFIG_IOS_JIT)
+        uintptr_t wr_addr = tc_ptr + offset + tb->code_rw_mirror_diff;
+#else
+        uintptr_t wr_addr = tc_ptr + offset;
+#endif
+        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr, wr_addr);
     } else {
         tb->jmp_target_arg[n] = addr;
     }
accel/tcg/translate-all.c +74 -3

122 lines not on this page · raw message

tcg/tcg.c +37 -19

146 lines not on this page · raw message

tcg/aarch64/tcg-target.c.inc +20 -13
index 1aa5f37fc6..ea3c06e917 100644
@@ -78,40 +78,44 @@ static const int tcg_target_call_oarg_regs[1] = {
 #define TCG_REG_GUEST_BASE TCG_REG_X28
 #endif
 
-static inline bool reloc_pc26(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc26(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                               const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - code_ptr;
     if (offset == sextract64(offset, 0, 26)) {
         /* read instruction, mask away previous PC_REL26 parameter contents,
            set the proper offset, then write back the instruction. */
-        *code_ptr = deposit32(*code_ptr, 0, 26, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 0, 26, offset);
         return true;
     }
     return false;
 }
 
-static inline bool reloc_pc19(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc19(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                               const tcg_insn_unit *target)
 {
     ptrdiff_t offset = target - code_ptr;
     if (offset == sextract64(offset, 0, 19)) {
-        *code_ptr = deposit32(*code_ptr, 5, 19, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 5, 19, offset);
         return true;
     }
     return false;
 }
 
-static inline bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static inline bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                                intptr_t value, intptr_t addend)
 {
     tcg_debug_assert(addend == 0);
     switch (type) {
     case R_AARCH64_JUMP26:
     case R_AARCH64_CALL26:
-        return reloc_pc26(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc26(s, code_ptr, (tcg_insn_unit *)value);
     case R_AARCH64_CONDBR19:
-        return reloc_pc19(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc19(s, code_ptr, (tcg_insn_unit *)value);
     default:
         g_assert_not_reached();
     }
@@ -1343,7 +1347,7 @@ static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
     tcg_insn_unit i1, i2;
     TCGType rt = TCG_TYPE_I64;
@@ -1364,7 +1368,10 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
         i2 = I3401_ADDI | rt << 31 | (addr & 0xfff) << 10 | rd << 5 | rd;
     }
     pair = (uint64_t)i2 << 32 | i1;
-    qatomic_set((uint64_t *)jmp_addr, pair);
+    qatomic_set((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
     flush_icache_range(jmp_addr, jmp_addr + 8);
 }
 
@@ -1583,7 +1590,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     MemOp opc = get_memop(oi);
     MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -1608,7 +1615,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     MemOp opc = get_memop(oi);
     MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -2919,11 +2926,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
     tcg_out_insn(s, 3207, RET, TCG_REG_LR);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
     int i;
     for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
     }
 }
 
tcg/arm/tcg-target.c.inc +14 -11
index 5ceb06a971..ee8ab72c2e 100644
@@ -187,18 +187,21 @@ static const uint8_t tcg_cond_to_arm_cond[] = {
     [TCG_COND_GTU] = COND_HI,
 };
 
-static inline bool reloc_pc24(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc24(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                               const tcg_insn_unit *target)
 {
     ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
     if (offset == sextract32(offset, 0, 24)) {
-        *code_ptr = (*code_ptr & ~0xffffff) | (offset & 0xffffff);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            (*tcg_code_ptr_rw(s, code_ptr) & ~0xffffff) | (offset & 0xffffff);
         return true;
     }
     return false;
 }
 
-static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc13(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                               const tcg_insn_unit *target)
 {
     ptrdiff_t offset = tcg_ptr_byte_diff(target, code_ptr) - 8;
@@ -211,21 +214,21 @@ static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
         }
         insn = deposit32(insn, 23, 1, u);
         insn = deposit32(insn, 0, 12, offset);
-        *code_ptr = insn;
+        *tcg_code_ptr_rw(s, code_ptr) = insn;
         return true;
     }
     return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     tcg_debug_assert(addend == 0);
 
     if (type == R_ARM_PC24) {
-        return reloc_pc24(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc24(s, code_ptr, (tcg_insn_unit *)value);
     } else if (type == R_ARM_PC13) {
-        return reloc_pc13(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc13(s, code_ptr, (tcg_insn_unit *)value);
     } else {
         g_assert_not_reached();
     }
@@ -1350,7 +1353,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     MemOp opc = get_memop(oi);
     void *func;
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -1413,7 +1416,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     TCGMemOpIdx oi = lb->oi;
     MemOp opc = get_memop(oi);
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -2257,11 +2260,11 @@ static inline void tcg_out_movi(TCGContext *s, TCGType type,
     tcg_out_movi32(s, COND_AL, ret, arg);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
     int i;
     for (i = 0; i < count; ++i) {
-        p[i] = INSN_NOP;
+        (tcg_code_ptr_rw(s, p))[i] = INSN_NOP;
     }
 }
 
tcg/i386/tcg-target.c.inc +9 -9
index 69acb51f5d..76bdaf748f 100644
@@ -167,7 +167,7 @@ static bool have_lzcnt;
 
 static const tcg_insn_unit *tb_ret_addr;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     value += addend;
@@ -179,14 +179,14 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
         }
         /* FALLTHRU */
     case R_386_32:
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
         break;
     case R_386_PC8:
         value -= (uintptr_t)code_ptr;
         if (value != (int8_t)value) {
             return false;
         }
-        tcg_patch8(code_ptr, value);
+        tcg_patch8(s, code_ptr, value);
         break;
     default:
         tcg_abort();
@@ -1805,9 +1805,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     int rexw = (l->type == TCG_TYPE_I64 ? P_REXW : 0);
 
     /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
     if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
     }
 
     if (TCG_TARGET_REG_BITS == 32) {
@@ -1890,9 +1890,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     TCGReg retaddr;
 
     /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
     if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
     }
 
     if (TCG_TARGET_REG_BITS == 32) {
@@ -3842,9 +3842,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
     tcg_out_opc(s, OPC_RET, 0, 0, 0);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x90, count);
+    memset(tcg_code_ptr_rw(s, p), 0x90, count);
 }
 
 static void tcg_target_init(TCGContext *s)
tcg/mips/tcg-target.c.inc +21 -14
index ace413afde..d4389f7135 100644
@@ -153,10 +153,12 @@ static inline uint32_t reloc_pc16_val(const tcg_insn_unit *pc,
     return disp & 0xffff;
 }
 
-static inline void reloc_pc16(tcg_insn_unit *pc,
+static inline void reloc_pc16(TCGContext *s,
+                              tcg_insn_unit *pc,
                               const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 16, reloc_pc16_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 16, reloc_pc16_val(pc, target));
 }
 
 static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
@@ -166,18 +168,20 @@ static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
     return ((uintptr_t)target >> 2) & 0x3ffffff;
 }
 
-static inline void reloc_26(tcg_insn_unit *pc,
+static inline void reloc_26(TCGContext *s,
+                            tcg_insn_unit *pc,
                             const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 26, reloc_26_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 26, reloc_26_val(pc, target));
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     tcg_debug_assert(type == R_MIPS_PC16);
     tcg_debug_assert(addend == 0);
-    reloc_pc16(code_ptr, (tcg_insn_unit *)value);
+    reloc_pc16(s, code_ptr, (tcg_insn_unit *)value);
     return true;
 }
 
@@ -929,7 +933,7 @@ static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
 
     tcg_out_opc_br(s, b_opc, arg1, arg2);
     if (l->has_value) {
-        reloc_pc16(s->code_ptr - 1, l->u.value_ptr);
+        reloc_pc16(s, s->code_ptr - 1, l->u.value_ptr);
     } else {
         tcg_out_reloc(s, s->code_ptr - 1, R_MIPS_PC16, l, 0);
     }
@@ -1320,9 +1324,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     int i;
 
     /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
     if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
     }
 
     i = 1;
@@ -1350,7 +1354,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     }
 
     tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO);
-    reloc_pc16(s->code_ptr - 1, l->raddr);
+    reloc_pc16(s, s->code_ptr - 1, l->raddr);
 
     /* delay slot */
     if (TCG_TARGET_REG_BITS == 64 && l->type == TCG_TYPE_I32) {
@@ -1370,9 +1374,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     int i;
 
     /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
     if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
     }
 
     i = 1;
@@ -2662,9 +2666,12 @@ static void tcg_target_init(TCGContext *s)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
-    qatomic_set((uint32_t *)jmp_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+    qatomic_set((uint32_t *)wr_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
     flush_icache_range(jmp_addr, jmp_addr + 4);
 }
 
tcg/ppc/tcg-target.c.inc +24 -14
index 289388b603..50d72529d2 100644
@@ -192,12 +192,14 @@ static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
     return disp & 0x3fffffc;
 }
 
-static bool reloc_pc24(tcg_insn_unit *pc,
+static bool reloc_pc24(TCGContext *s,
+                       tcg_insn_unit *pc,
                        const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     if (in_range_b(disp)) {
-        *pc = (*pc & ~0x3fffffc) | (disp & 0x3fffffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0x3fffffc) | (disp & 0x3fffffc);
         return true;
     }
     return false;
@@ -211,12 +213,14 @@ static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
     return disp & 0xfffc;
 }
 
-static bool reloc_pc14(tcg_insn_unit *pc,
+static bool reloc_pc14(TCGContext *s,
+                       tcg_insn_unit *pc,
                        const tcg_insn_unit *target)
 {
     ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
     if (disp == (int16_t) disp) {
-        *pc = (*pc & ~0xfffc) | (disp & 0xfffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0xfffc) | (disp & 0xfffc);
         return true;
     }
     return false;
@@ -674,7 +678,7 @@ static const uint32_t tcg_to_isel[] = {
     [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
 };
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     tcg_insn_unit *target;
@@ -686,9 +690,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
 
     switch (type) {
     case R_PPC_REL14:
-        return reloc_pc14(code_ptr, target);
+        return reloc_pc14(s, code_ptr, target);
     case R_PPC_REL24:
-        return reloc_pc24(code_ptr, target);
+        return reloc_pc24(s, code_ptr, target);
     case R_PPC_ADDR16:
         /*
          * We are (slightly) abusing this relocation type.  In particular,
@@ -1727,7 +1731,7 @@ static void tcg_out_mb(TCGContext *s, TCGArg a0)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
     if (TCG_TARGET_REG_BITS == 64) {
         tcg_insn_unit i1, i2;
@@ -1756,12 +1760,18 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
 
         /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
            within qatomic_set that would fail to build a ppc32 host.  */
-        qatomic_set__nocheck((uint64_t *)jmp_addr, pair);
+        qatomic_set__nocheck((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
         flush_icache_range(jmp_addr, jmp_addr + 8);
     } else {
         intptr_t diff = addr - jmp_addr;
         tcg_debug_assert(in_range_b(diff));
-        qatomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
+        qatomic_set((uint32_t *)wr_addr, B | (diff & 0x3fffffc));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
         flush_icache_range(jmp_addr, jmp_addr + 4);
     }
 }
@@ -2012,7 +2022,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     MemOp opc = get_memop(oi);
     TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -2060,7 +2070,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     MemOp s_bits = opc & MO_SIZE;
     TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
         return false;
     }
 
@@ -2257,11 +2267,11 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
     int i;
     for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
     }
 }
 
tcg/riscv/tcg-target.c.inc +23 -17
index 5432ee89db..19882db57f 100644
@@ -413,11 +413,11 @@ static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
     tcg_out32(s, encode_uj(opc, rd, imm));
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
     int i;
     for (i = 0; i < count; ++i) {
-        p[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
+        (tcg_code_ptr_rw(s, p))[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
     }
 }
 
@@ -425,46 +425,52 @@ static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
  * Relocations
  */
 
-static bool reloc_sbimm12(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_sbimm12(TCGContext *s,
+                          tcg_insn_unit *code_ptr,
+                          const tcg_insn_unit *target)
 {
     intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
     if (offset == sextreg(offset, 1, 12) << 1) {
-        code_ptr[0] |= encode_sbimm12(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_sbimm12(offset);
         return true;
     }
 
     return false;
 }
 
-static bool reloc_jimm20(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_jimm20(TCGContext *s,
+                         tcg_insn_unit *code_ptr,
+                         const tcg_insn_unit *target)
 {
     intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
     if (offset == sextreg(offset, 1, 20) << 1) {
-        code_ptr[0] |= encode_ujimm20(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_ujimm20(offset);
         return true;
     }
 
     return false;
 }
 
-static bool reloc_call(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_call(TCGContext *s,
+                       tcg_insn_unit *code_ptr,
+                       const tcg_insn_unit *target)
 {
     intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
     int32_t lo = sextreg(offset, 0, 12);
     int32_t hi = offset - lo;
 
     if (offset == hi + lo) {
-        code_ptr[0] |= encode_uimm20(hi);
-        code_ptr[1] |= encode_imm12(lo);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_uimm20(hi);
+        (tcg_code_ptr_rw(s, code_ptr))[1] |= encode_imm12(lo);
         return true;
     }
 
     return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     uint32_t insn = *code_ptr;
@@ -478,7 +484,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
         diff = value - (uintptr_t)code_ptr;
         short_jmp = diff == sextreg(diff, 0, 12);
         if (short_jmp) {
-            return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
+            return reloc_sbimm12(s, code_ptr, (tcg_insn_unit *)value);
         } else {
             /* Invert the condition */
             insn = insn ^ (1 << 12);
@@ -499,9 +505,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
         }
         break;
     case R_RISCV_JAL:
-        return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
+        return reloc_jimm20(s, code_ptr, (tcg_insn_unit *)value);
     case R_RISCV_CALL:
-        return reloc_call(code_ptr, (tcg_insn_unit *)value);
+        return reloc_call(s, code_ptr, (tcg_insn_unit *)value);
     default:
         tcg_abort();
     }
@@ -557,7 +563,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
     if (tmp == (int32_t)tmp) {
         tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
         tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
-        ret = reloc_call(s->code_ptr - 2, (tcg_insn_unit *)val);
+        ret = reloc_call(s, s->code_ptr - 2, (tcg_insn_unit *)val);
         tcg_debug_assert(ret == true);
         return;
     }
@@ -875,7 +881,7 @@ static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
         /* long jump: -2147483646 to 2147483648 */
         tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
         tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
-        ret = reloc_call(s->code_ptr - 2, arg);\
+        ret = reloc_call(s, s->code_ptr - 2, arg);\
         tcg_debug_assert(ret == true);
     } else if (TCG_TARGET_REG_BITS == 64) {
         /* far jump: 64-bit */
@@ -1053,7 +1059,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     }
 
     /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                      (intptr_t) s->code_ptr, 0)) {
         return false;
     }
@@ -1088,7 +1094,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
     }
 
     /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                      (intptr_t) s->code_ptr, 0)) {
         return false;
     }
tcg/s390/tcg-target.c.inc +8 -8
index 7c6917f3c4..f9498ddf63 100644
@@ -366,7 +366,7 @@ static void * const qemu_st_helpers[16] = {
 static const tcg_insn_unit *tb_ret_addr;
 uint64_t s390_facilities;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     intptr_t pcrel2;
@@ -378,13 +378,13 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
     switch (type) {
     case R_390_PC16DBL:
         if (pcrel2 == (int16_t)pcrel2) {
-            tcg_patch16(code_ptr, pcrel2);
+            tcg_patch16(s, code_ptr, pcrel2);
             return true;
         }
         break;
     case R_390_PC32DBL:
         if (pcrel2 == (int32_t)pcrel2) {
-            tcg_patch32(code_ptr, pcrel2);
+            tcg_patch32(s, code_ptr, pcrel2);
             return true;
         }
         break;
@@ -392,7 +392,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
         if (value == sextract64(value, 0, 20)) {
             old = *(uint32_t *)code_ptr & 0xf00000ff;
             old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
-            tcg_patch32(code_ptr, old);
+            tcg_patch32(s, code_ptr, old);
             return true;
         }
         break;
@@ -1613,7 +1613,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     TCGMemOpIdx oi = lb->oi;
     MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                      (intptr_t)s->code_ptr, 2)) {
         return false;
     }
@@ -1638,7 +1638,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
     TCGMemOpIdx oi = lb->oi;
     MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                      (intptr_t)s->code_ptr, 2)) {
         return false;
     }
@@ -2576,9 +2576,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
     tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x07, count * sizeof(tcg_insn_unit));
+    memset(tcg_code_ptr_rw(s, p), 0x07, count * sizeof(tcg_insn_unit));
 }
 
 typedef struct {
tcg/sparc/tcg-target.c.inc +15 -8
index 5eac18b719..41dbd21cb4 100644
@@ -291,14 +291,14 @@ static inline int check_fit_i32(int32_t val, unsigned int bits)
 # define check_fit_ptr  check_fit_i32
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     uint32_t insn = *code_ptr;
     intptr_t pcrel;
 
     value += addend;
-    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr);
+    pcrel = tcg_ptr_byte_diff((const tcg_insn_unit *)value, code_ptr);
 
     switch (type) {
     case R_SPARC_WDISP16:
@@ -1048,11 +1048,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
     int i;
     for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
     }
 }
 
@@ -1226,7 +1226,8 @@ static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
         }
     }
 
-    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
+    *tcg_code_ptr_rw(s, label_ptr) |=
+        INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
 #else
     if (SPARC64 && TARGET_LONG_BITS == 32) {
         tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL);
@@ -1822,7 +1823,7 @@ void tcg_register_jit(void *buf, size_t buf_size)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
     intptr_t tb_disp = addr - tc_ptr;
     intptr_t br_disp = addr - jmp_addr;
@@ -1834,8 +1835,11 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
     tcg_debug_assert(br_disp == (int32_t)br_disp);
 
     if (!USE_REG_TB) {
-        qatomic_set((uint32_t *)jmp_addr,
+        qatomic_set((uint32_t *)wr_addr,
 		    deposit32(CALL, 0, 30, br_disp >> 2));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
         flush_icache_range(jmp_addr, jmp_addr + 4);
         return;
     }
@@ -1859,6 +1863,9 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
               | INSN_IMM13((tb_disp & 0x3ff) | -0x400));
     }
 
-    qatomic_set((uint64_t *)jmp_addr, deposit64(i2, 32, 32, i1));
+    qatomic_set((uint64_t *)wr_addr, deposit64(i2, 32, 32, i1));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
     flush_icache_range(jmp_addr, jmp_addr + 8);
 }
tcg/tcg-pool.c.inc +5 -4
index 82cbcc89bd..8434496e72 100644
@@ -119,7 +119,7 @@ static inline void new_pool_l8(TCGContext *s, int rtype, tcg_insn_unit *label,
 }
 
 /* To be provided by cpu/tcg-target.c.inc.  */
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count);
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count);
 
 static int tcg_out_pool_finalize(TCGContext *s)
 {
@@ -135,7 +135,7 @@ static int tcg_out_pool_finalize(TCGContext *s)
        again when allocating the next TranslationBlock structure.  */
     a = (void *)ROUND_UP((uintptr_t)s->code_ptr,
                          sizeof(tcg_target_ulong) * p->nlong);
-    tcg_out_nop_fill(s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
+    tcg_out_nop_fill(s, s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
     s->data_gen_ptr = a;
 
     for (; p != NULL; p = p->next) {
@@ -144,11 +144,12 @@ static int tcg_out_pool_finalize(TCGContext *s)
             if (unlikely(a > s->code_gen_highwater)) {
                 return -1;
             }
-            memcpy(a, p->data, size);
+            memcpy(tcg_mirror_ptr_rw(s, a), p->data, size);
             a += size;
             l = p;
         }
-        if (!patch_reloc(p->label, p->rtype, (intptr_t)a - size, p->addend)) {
+        if (!patch_reloc(s, p->label, p->rtype,
+                         (intptr_t)a - size, p->addend)) {
             return -2;
         }
     }
tcg/tci/tcg-target.c.inc +3 -3
index d5a4d9d37c..133213be3a 100644
@@ -369,7 +369,7 @@ static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
 };
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
     /* tcg_out_reloc always uses the same type, addend. */
@@ -377,9 +377,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
     tcg_debug_assert(addend == 0);
     tcg_debug_assert(value != 0);
     if (TCG_TARGET_REG_BITS == 32) {
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
     } else {
-        tcg_patch64(code_ptr, value);
+        tcg_patch64(s, code_ptr, value);
     }
     return true;
 }
Signature · 1 line
2.24.3 (Apple Git-128)
#23 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

Quoted 144 lines · Joelle van Dyne

From: osy <[email protected]>

On iOS, we cannot allocate RWX pages without special entitlements. As a workaround, we can allocate a RX region and then mirror map it to a separate RX region. Then we can write to one region and execute from the other one.

We also define tcg_mirror_ptr_rw and tcg_code_ptr_rw to return a pointer to RW memory. The difference between the RW and RX regions is stored in the TCG context.

To ensure cache coherency, we flush the data cache in the RW mapping and then invalidate the instruction cache in the RX mapping (where applicable). Because data cache flush is OS defined on some architectures, we do not provide implementations for non iOS platforms (ARM/x86).

---

docs/devel/ios.rst           | 40 +++++++++++++++++++
configure                    |  1 +
include/exec/exec-all.h      |  8 ++++
include/tcg/tcg.h            | 17 ++++++++
tcg/aarch64/tcg-target.h     | 13 +++++-
tcg/arm/tcg-target.h         |  9 ++++-
tcg/i386/tcg-target.h        | 24 ++++++++++-
tcg/mips/tcg-target.h        |  8 +++-
tcg/ppc/tcg-target.h         |  8 +++-
tcg/riscv/tcg-target.h       |  9 ++++-
tcg/s390/tcg-target.h        | 13 +++++-
tcg/sparc/tcg-target.h       |  8 +++-
tcg/tci/tcg-target.h         |  9 ++++-
accel/tcg/cpu-exec.c         |  7 +++-
accel/tcg/translate-all.c    | 77 ++++++++++++++++++++++++++++++++++--
tcg/tcg.c                    | 56 +++++++++++++++++---------
tcg/aarch64/tcg-target.c.inc | 33 ++++++++++------
tcg/arm/tcg-target.c.inc     | 25 ++++++------
tcg/i386/tcg-target.c.inc    | 18 ++++-----
tcg/mips/tcg-target.c.inc    | 35 +++++++++-------
tcg/ppc/tcg-target.c.inc     | 38 +++++++++++-------
tcg/riscv/tcg-target.c.inc   | 40 +++++++++++--------
tcg/s390/tcg-target.c.inc    | 16 ++++----
tcg/sparc/tcg-target.c.inc   | 23 +++++++----
tcg/tcg-pool.c.inc           |  9 +++--
tcg/tci/tcg-target.c.inc     |  6 +--
26 files changed, 416 insertions(+), 134 deletions(-)

create mode 100644 docs/devel/ios.rst

docs/devel/ios.rst new file +40 -0
new file mode 100644
index 0000000000..dba9fdd868
@@ -0,0 +1,40 @@
+===========
+iOS Support
+===========
+
+To run qemu on the iOS platform, some modifications were required. Most of the
+modifications are conditioned on the ``CONFIG_IOS`` and ``CONFIG_IOS_JIT``
+configuration variables.
+
+Build support
+-------------
+
+For the code to compile, certain changes in the block driver and the slirp
+driver had to be made. There is no ``system()`` call, so code requiring it had
+to be disabled.
+
+``ucontext`` support is broken on iOS. The implementation from ``libucontext``
+is used instead.
+
+Because ``fork()`` is not allowed on iOS apps, the option to build qemu and the
+utilities as shared libraries is added. Note that because qemu does not perform
+resource cleanup in most cases (open files, allocated memory, etc), it is
+advisable that the user implements a proxy layer for syscalls so resources can
+be kept track by the app that uses qemu as a shared library.
+
+JIT support
+-----------
+
+On iOS, allocating RWX pages require special entitlements not usually granted to
+apps. However, it is possible to use `bulletproof JIT`_ with a development
+certificate. This means that we need to allocate one chunk of memory with RX
+permissions and then mirror map the same memory with RW permissions. We generate
+code to the mirror mapping and execute the original mapping.
+
+With ``CONFIG_IOS_JIT`` defined, we store inside the TCG context the difference
+between the two mappings. Then, we make sure that any writes to JIT memory is
+done to the pointer + the difference (in order to get a pointer to the mirror
+mapped space). Additionally, we make sure to flush the data cache before we
+invalidate the instruction cache so the changes are seen in both mappings.
+
+.. _bulletproof JIT: https://www.blackhat.com/docs/us-16/materials/us-16-Krstic.pdf
configure +0 -0
index 0b7e25e7a5..93d6fd5ce2 100755
@@ -6214,6 +6214,7 @@ fi
 

if test "$ios" = "yes" ; then
echo "CONFIG_IOS=y" >> $config_host_mak
+ echo "CONFIG_IOS_JIT=y" >> $config_host_mak
fi

if test "$solaris" = "yes" ; then

include/exec/exec-all.h +8 -0
index 66f9b4cca6..2db155a772 100644
@@ -483,6 +483,14 @@ struct TranslationBlock {
    uintptr_t jmp_list_head;
    uintptr_t jmp_list_next[2];
    uintptr_t jmp_dest[2];
+
+#if defined(CONFIG_IOS_JIT)
+    /*
+     * Store difference to writable mirror
+     * We need this when patching the jump instructions
+     */
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 };
 
 extern bool parallel_cpus;
include/tcg/tcg.h +6 -0
index 79c5ff8dab..ade01d2e41 100644
@@ -627,6 +627,9 @@ struct TCGContext {
    size_t code_gen_buffer_size;
    void *code_gen_ptr;
    void *data_gen_ptr;
+#if defined(CONFIG_IOS_JIT)
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 
    /* Threshold to flush the translated code buffer.  */
    void *code_gen_highwater;
@@ -677,6 +680,20 @@ struct TCGContext {
    target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
 };
 
+static inline void *tcg_mirror_ptr_rw(TCGContext *s, const void *rx)
+{
+#if defined(CONFIG_IOS_JIT)

include/tcg/tcg.h struct TCGContext { · line 695

Quoted 10 lines
include/tcg/tcg.h +10 -0
+    return (void *)rx + s->code_rw_mirror_diff;
+#else
+    return (void *)rx;
+#endif
+}
+
+static inline tcg_insn_unit *tcg_code_ptr_rw(TCGContext *s, const void *rx)
+{
+    return (tcg_insn_unit *)tcg_mirror_ptr_rw(s, rx);
+}

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

Quoted 40 lines
include/tcg/tcg.h +1 -0
+
 extern TCGContext tcg_init_ctx;
 extern __thread TCGContext *tcg_ctx;
 extern TCGv_env cpu_env;
tcg/aarch64/tcg-target.h +12 -1
index a2b22b4305..78c97460a1 100644
@@ -150,6 +150,7 @@ typedef enum {
 
 #if defined(__APPLE__)
 void sys_icache_invalidate(void *start, size_t len);
+void sys_dcache_flush(void *start, size_t len);
 #endif
 
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
@@ -163,7 +164,17 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #endif
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/arm/tcg-target.h +0 -0
index 17e771374d..d8d7e7e239 100644

tcg/arm/tcg-target.h static inline void flush_icache_range(uintptr_t start, uintptr_t stop) · line 142

Quoted 4 lines
tcg/arm/tcg-target.h +1 -0
@@ -139,8 +139,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    __builtin___clear_cache((char *) start, (char *) stop);
 }
 
+#if defined(CONFIG_IOS_JIT)

Is that !defined or are you missing an implementation and #else here?

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

Regards,
BALATON Zoltan

Quoted 1,329 lines
tcg/arm/tcg-target.h +7 -1
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/i386/tcg-target.h +22 -2
index abd4ac7fc0..cdc440ce36 100644
@@ -206,16 +206,36 @@ extern bool have_avx2;
 #define TCG_TARGET_extract_i64_valid(ofs, len) \
    (((ofs) == 8 && (len) == 8) || ((ofs) + (len)) == 32)
 
+#ifdef __APPLE__
+void sys_dcache_flush(void *start, size_t len);
+#endif
+
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
-    qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
+    qatomic_set((int32_t *)wr_addr, addr - (jmp_addr + 4));
    /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    /* we do need to flush mirror dcache */
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 /* This defines the natural memory order supported by this
tcg/mips/tcg-target.h +7 -1
index c6b091d849..80dcba5358 100644
@@ -212,7 +212,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    cacheflush ((void *)start, stop-start, ICACHE);
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/ppc/tcg-target.h +7 -1
index be10363956..23d7a337c9 100644
@@ -176,7 +176,13 @@ extern bool have_vsx;
 #define TCG_TARGET_HAS_cmpsel_vec       0
 
 void flush_icache_range(uintptr_t start, uintptr_t stop);
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
tcg/riscv/tcg-target.h +8 -1
index 032439d806..d42b361991 100644
@@ -164,8 +164,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    __builtin___clear_cache((char *)start, (char *)stop);
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #define TCG_TARGET_DEFAULT_MO (0)
 
tcg/s390/tcg-target.h +12 -1
index 63c8797bd3..d67632512d 100644
@@ -149,13 +149,24 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
    intptr_t disp = addr - (jmp_addr - 2);
    qatomic_set((int32_t *)jmp_addr, disp / 2);
    /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 #ifdef CONFIG_SOFTMMU
tcg/sparc/tcg-target.h +7 -1
index 633841ebf2..d102e13692 100644
@@ -176,7 +176,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    }
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_NEED_POOL_LABELS
 
tcg/tci/tcg-target.h +8 -1
index 8c1c1d265d..2a258ea350 100644
@@ -195,6 +195,12 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+}
+#endif
+
 /* We could notice __i386__ or __s390x__ and reduce the barriers depending
   on the host.  But if you want performance, you use the normal backend.
   We prefer consistency across hosts on this.  */
@@ -203,7 +209,8 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
 
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
    qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
accel/tcg/cpu-exec.c +6 -1
index 58aea605d8..821aefdea2 100644
@@ -354,7 +354,12 @@ void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
    if (TCG_TARGET_HAS_direct_jump) {
        uintptr_t offset = tb->jmp_target_arg[n];
        uintptr_t tc_ptr = (uintptr_t)tb->tc.ptr;
-        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr);
+#if defined(CONFIG_IOS_JIT)
+        uintptr_t wr_addr = tc_ptr + offset + tb->code_rw_mirror_diff;
+#else
+        uintptr_t wr_addr = tc_ptr + offset;
+#endif
+        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr, wr_addr);
    } else {
        tb->jmp_target_arg[n] = addr;
    }
accel/tcg/translate-all.c +74 -3
index d76097296d..eb1d8fbe2f 100644
@@ -60,6 +60,22 @@
 #include "sysemu/cpu-timers.h"
 #include "sysemu/tcg.h"
 
+#if defined(CONFIG_IOS_JIT)
+#include <mach/mach.h>
+extern kern_return_t mach_vm_remap(vm_map_t target_task,
+                                   mach_vm_address_t *target_address,
+                                   mach_vm_size_t size,
+                                   mach_vm_offset_t mask,
+                                   int flags,
+                                   vm_map_t src_task,
+                                   mach_vm_address_t src_address,
+                                   boolean_t copy,
+                                   vm_prot_t *cur_protection,
+                                   vm_prot_t *max_protection,
+                                   vm_inherit_t inheritance
+                                  );
+#endif
+
 /* #define DEBUG_TB_INVALIDATE */
 /* #define DEBUG_TB_FLUSH */
 /* make various TB consistency checks */
@@ -302,10 +318,12 @@ static target_long decode_sleb128(uint8_t **pp)
 
 static int encode_search(TranslationBlock *tb, uint8_t *block)
 {
-    uint8_t *highwater = tcg_ctx->code_gen_highwater;
-    uint8_t *p = block;
+    const uint8_t *highwater, *start;
+    uint8_t *p;
    int i, j, n;
 
+    highwater = tcg_mirror_ptr_rw(tcg_ctx, tcg_ctx->code_gen_highwater);
+    start = p = tcg_mirror_ptr_rw(tcg_ctx, block);
    for (i = 0, n = tb->icount; i < n; ++i) {
        target_ulong prev;
 
@@ -329,7 +347,7 @@ static int encode_search(TranslationBlock *tb, uint8_t *block)
        }
    }
 
-    return p - block;
+    return p - start;
 }
 
 /* The cpu state corresponding to 'searched_pc' is restored.
@@ -1067,7 +1085,11 @@ static inline void *alloc_code_gen_buffer(void)
 #else
 static inline void *alloc_code_gen_buffer(void)
 {
+#if defined(CONFIG_IOS_JIT)
+    int prot = PROT_READ | PROT_EXEC;
+#else
    int prot = PROT_WRITE | PROT_READ | PROT_EXEC;
+#endif
    int flags = MAP_PRIVATE | MAP_ANONYMOUS;
    size_t size = tcg_ctx->code_gen_buffer_size;
    void *buf;
@@ -1118,6 +1140,39 @@ static inline void *alloc_code_gen_buffer(void)
 }
 #endif /* USE_STATIC_CODE_GEN_BUFFER, WIN32, POSIX */
 
+#if defined(CONFIG_IOS_JIT)
+static inline void *alloc_jit_rw_mirror(void *base, size_t size)
+{
+    kern_return_t ret;
+    mach_vm_address_t mirror;
+    vm_prot_t cur_prot, max_prot;
+
+    mirror = 0;
+    ret = mach_vm_remap(mach_task_self(),
+                        &mirror,
+                        size,
+                        0,
+                        VM_FLAGS_ANYWHERE | VM_FLAGS_RANDOM_ADDR,
+                        mach_task_self(),
+                        (mach_vm_address_t)base,
+                        false,
+                        &cur_prot,
+                        &max_prot,
+                        VM_INHERIT_NONE
+                       );
+    if (ret != KERN_SUCCESS) {
+        return NULL;
+    }
+
+    if (mprotect((void *)mirror, size, PROT_READ | PROT_WRITE) != 0) {
+        munmap((void *)mirror, size);
+        return NULL;
+    }
+
+    return (void *)mirror;
+}
+#endif /* CONFIG_IOS_JIT */
+
 static inline void code_gen_alloc(size_t tb_size)
 {
    tcg_ctx->code_gen_buffer_size = size_code_gen_buffer(tb_size);
@@ -1126,6 +1181,19 @@ static inline void code_gen_alloc(size_t tb_size)
        fprintf(stderr, "Could not allocate dynamic translator buffer\n");
        exit(1);
    }
+#if defined(CONFIG_IOS_JIT)
+    void *mirror;
+
+    /* For iOS JIT we need a mirror mapping for code execution */
+    mirror = alloc_jit_rw_mirror(tcg_ctx->code_gen_buffer,
+                                 tcg_ctx->code_gen_buffer_size
+                                );
+    if (mirror == NULL) {
+        fprintf(stderr, "Could not remap code buffer mirror\n");
+        exit(1);
+    }
+    tcg_ctx->code_rw_mirror_diff = mirror - tcg_ctx->code_gen_buffer;
+#endif /* CONFIG_IOS_JIT */
 }
 
 static bool tb_cmp(const void *ap, const void *bp)
@@ -1721,6 +1789,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
        cpu_loop_exit(cpu);
    }
 
+#if defined(CONFIG_IOS_JIT)
+    tb->code_rw_mirror_diff = tcg_ctx->code_rw_mirror_diff;
+#endif
    gen_code_buf = tcg_ctx->code_gen_ptr;
    tb->tc.ptr = gen_code_buf;
    tb->pc = pc;
tcg/tcg.c +37 -19
index bb890c506d..6cf70c33b0 100644
@@ -70,7 +70,7 @@
 static void tcg_target_init(TCGContext *s);
 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode);
 static void tcg_target_qemu_prologue(TCGContext *s);
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend);
 
 /* The CIE and FDE header definitions will be common to all hosts.  */
@@ -203,13 +203,15 @@ static TCGRegSet tcg_target_call_clobber_regs;
 #if TCG_TARGET_INSN_UNIT_SIZE == 1
 static __attribute__((unused)) inline void tcg_out8(TCGContext *s, uint8_t v)
 {
-    *s->code_ptr++ = v;
+    *tcg_code_ptr_rw(s, s->code_ptr) = v;
+    s->code_ptr++;
 }
 
-static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch8(TCGContext *s,
+                                                      tcg_insn_unit *p,
                                                      uint8_t v)
 {
-    *p = v;
+    *tcg_code_ptr_rw(s, p) = v;
 }
 #endif
 
@@ -217,21 +219,23 @@ static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out16(TCGContext *s, uint16_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (2 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch16(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint16_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
-        *p = v;
+        *tcg_code_ptr_rw(s, p) = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -240,21 +244,23 @@ static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out32(TCGContext *s, uint32_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (4 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch32(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint32_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
        *p = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -263,21 +269,23 @@ static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out64(TCGContext *s, uint64_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (8 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch64(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch64(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint64_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
        *p = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -325,7 +333,7 @@ static bool tcg_resolve_relocs(TCGContext *s)
        uintptr_t value = l->u.value;
 
        QSIMPLEQ_FOREACH(r, &l->relocs, next) {
-            if (!patch_reloc(r->ptr, r->type, value, r->addend)) {
+            if (!patch_reloc(s, r->ptr, r->type, value, r->addend)) {
                return false;
            }
        }
@@ -1039,7 +1047,7 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s)
    }
    qatomic_set(&s->code_gen_ptr, next);
    s->data_gen_ptr = NULL;
-    return tb;
+    return (TranslationBlock *)tcg_mirror_ptr_rw(s, tb);
 }
 
 void tcg_prologue_init(TCGContext *s)
@@ -1076,6 +1084,10 @@ void tcg_prologue_init(TCGContext *s)
 #endif
 
    buf1 = s->code_ptr;
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range((uintptr_t)tcg_mirror_ptr_rw(s, buf0),
+                       (uintptr_t)tcg_mirror_ptr_rw(s, buf1));
+#endif
    flush_icache_range((uintptr_t)buf0, (uintptr_t)buf1);
 
    /* Deduct the prologue from the buffer.  */
@@ -4267,6 +4279,12 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
        return -2;
    }
 
+#if defined(CONFIG_IOS_JIT)
+    /* flush data cache on mirror */
+    flush_dcache_range((uintptr_t)tcg_mirror_ptr_rw(s, s->code_buf),
+                       (uintptr_t)tcg_mirror_ptr_rw(s, s->code_ptr));
+#endif
+
    /* flush instruction cache */
    flush_icache_range((uintptr_t)s->code_buf, (uintptr_t)s->code_ptr);
 
tcg/aarch64/tcg-target.c.inc +20 -13
index 1aa5f37fc6..ea3c06e917 100644
@@ -78,40 +78,44 @@ static const int tcg_target_call_oarg_regs[1] = {
 #define TCG_REG_GUEST_BASE TCG_REG_X28
 #endif
 
-static inline bool reloc_pc26(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc26(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = target - code_ptr;
    if (offset == sextract64(offset, 0, 26)) {
        /* read instruction, mask away previous PC_REL26 parameter contents,
           set the proper offset, then write back the instruction. */
-        *code_ptr = deposit32(*code_ptr, 0, 26, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 0, 26, offset);
        return true;
    }
    return false;
 }
 
-static inline bool reloc_pc19(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc19(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = target - code_ptr;
    if (offset == sextract64(offset, 0, 19)) {
-        *code_ptr = deposit32(*code_ptr, 5, 19, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 5, 19, offset);
        return true;
    }
    return false;
 }
 
-static inline bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static inline bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                               intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(addend == 0);
    switch (type) {
    case R_AARCH64_JUMP26:
    case R_AARCH64_CALL26:
-        return reloc_pc26(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc26(s, code_ptr, (tcg_insn_unit *)value);
    case R_AARCH64_CONDBR19:
-        return reloc_pc19(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc19(s, code_ptr, (tcg_insn_unit *)value);
    default:
        g_assert_not_reached();
    }
@@ -1343,7 +1347,7 @@ static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    tcg_insn_unit i1, i2;
    TCGType rt = TCG_TYPE_I64;
@@ -1364,7 +1368,10 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
        i2 = I3401_ADDI | rt << 31 | (addr & 0xfff) << 10 | rd << 5 | rd;
    }
    pair = (uint64_t)i2 << 32 | i1;
-    qatomic_set((uint64_t *)jmp_addr, pair);
+    qatomic_set((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 8);
 }
 
@@ -1583,7 +1590,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -1608,7 +1615,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2919,11 +2926,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_insn(s, 3207, RET, TCG_REG_LR);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
tcg/arm/tcg-target.c.inc +14 -11
index 5ceb06a971..ee8ab72c2e 100644
@@ -187,18 +187,21 @@ static const uint8_t tcg_cond_to_arm_cond[] = {
    [TCG_COND_GTU] = COND_HI,
 };
 
-static inline bool reloc_pc24(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc24(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
    if (offset == sextract32(offset, 0, 24)) {
-        *code_ptr = (*code_ptr & ~0xffffff) | (offset & 0xffffff);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            (*tcg_code_ptr_rw(s, code_ptr) & ~0xffffff) | (offset & 0xffffff);
        return true;
    }
    return false;
 }
 
-static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc13(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = tcg_ptr_byte_diff(target, code_ptr) - 8;
@@ -211,21 +214,21 @@ static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
        }
        insn = deposit32(insn, 23, 1, u);
        insn = deposit32(insn, 0, 12, offset);
-        *code_ptr = insn;
+        *tcg_code_ptr_rw(s, code_ptr) = insn;
        return true;
    }
    return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(addend == 0);
 
    if (type == R_ARM_PC24) {
-        return reloc_pc24(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc24(s, code_ptr, (tcg_insn_unit *)value);
    } else if (type == R_ARM_PC13) {
-        return reloc_pc13(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc13(s, code_ptr, (tcg_insn_unit *)value);
    } else {
        g_assert_not_reached();
    }
@@ -1350,7 +1353,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    void *func;
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -1413,7 +1416,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2257,11 +2260,11 @@ static inline void tcg_out_movi(TCGContext *s, TCGType type,
    tcg_out_movi32(s, COND_AL, ret, arg);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = INSN_NOP;
+        (tcg_code_ptr_rw(s, p))[i] = INSN_NOP;
    }
 }
 
tcg/i386/tcg-target.c.inc +9 -9
index 69acb51f5d..76bdaf748f 100644
@@ -167,7 +167,7 @@ static bool have_lzcnt;
 
 static const tcg_insn_unit *tb_ret_addr;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    value += addend;
@@ -179,14 +179,14 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        }
        /* FALLTHRU */
    case R_386_32:
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
        break;
    case R_386_PC8:
        value -= (uintptr_t)code_ptr;
        if (value != (int8_t)value) {
            return false;
        }
-        tcg_patch8(code_ptr, value);
+        tcg_patch8(s, code_ptr, value);
        break;
    default:
        tcg_abort();
@@ -1805,9 +1805,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int rexw = (l->type == TCG_TYPE_I64 ? P_REXW : 0);
 
    /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
    if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
    }
 
    if (TCG_TARGET_REG_BITS == 32) {
@@ -1890,9 +1890,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    TCGReg retaddr;
 
    /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
    if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
    }
 
    if (TCG_TARGET_REG_BITS == 32) {
@@ -3842,9 +3842,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_opc(s, OPC_RET, 0, 0, 0);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x90, count);
+    memset(tcg_code_ptr_rw(s, p), 0x90, count);
 }
 
 static void tcg_target_init(TCGContext *s)
tcg/mips/tcg-target.c.inc +21 -14
index ace413afde..d4389f7135 100644
@@ -153,10 +153,12 @@ static inline uint32_t reloc_pc16_val(const tcg_insn_unit *pc,
    return disp & 0xffff;
 }
 
-static inline void reloc_pc16(tcg_insn_unit *pc,
+static inline void reloc_pc16(TCGContext *s,
+                              tcg_insn_unit *pc,
                              const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 16, reloc_pc16_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 16, reloc_pc16_val(pc, target));
 }
 
 static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
@@ -166,18 +168,20 @@ static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
    return ((uintptr_t)target >> 2) & 0x3ffffff;
 }
 
-static inline void reloc_26(tcg_insn_unit *pc,
+static inline void reloc_26(TCGContext *s,
+                            tcg_insn_unit *pc,
                            const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 26, reloc_26_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 26, reloc_26_val(pc, target));
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(type == R_MIPS_PC16);
    tcg_debug_assert(addend == 0);
-    reloc_pc16(code_ptr, (tcg_insn_unit *)value);
+    reloc_pc16(s, code_ptr, (tcg_insn_unit *)value);
    return true;
 }
 
@@ -929,7 +933,7 @@ static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
 
    tcg_out_opc_br(s, b_opc, arg1, arg2);
    if (l->has_value) {
-        reloc_pc16(s->code_ptr - 1, l->u.value_ptr);
+        reloc_pc16(s, s->code_ptr - 1, l->u.value_ptr);
    } else {
        tcg_out_reloc(s, s->code_ptr - 1, R_MIPS_PC16, l, 0);
    }
@@ -1320,9 +1324,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int i;
 
    /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
    }
 
    i = 1;
@@ -1350,7 +1354,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO);
-    reloc_pc16(s->code_ptr - 1, l->raddr);
+    reloc_pc16(s, s->code_ptr - 1, l->raddr);
 
    /* delay slot */
    if (TCG_TARGET_REG_BITS == 64 && l->type == TCG_TYPE_I32) {
@@ -1370,9 +1374,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int i;
 
    /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
    }
 
    i = 1;
@@ -2662,9 +2666,12 @@ static void tcg_target_init(TCGContext *s)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
-    qatomic_set((uint32_t *)jmp_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+    qatomic_set((uint32_t *)wr_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 4);
 }
 
tcg/ppc/tcg-target.c.inc +24 -14
index 289388b603..50d72529d2 100644
@@ -192,12 +192,14 @@ static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
    return disp & 0x3fffffc;
 }
 
-static bool reloc_pc24(tcg_insn_unit *pc,
+static bool reloc_pc24(TCGContext *s,
+                       tcg_insn_unit *pc,
                       const tcg_insn_unit *target)
 {
    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
    if (in_range_b(disp)) {
-        *pc = (*pc & ~0x3fffffc) | (disp & 0x3fffffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0x3fffffc) | (disp & 0x3fffffc);
        return true;
    }
    return false;
@@ -211,12 +213,14 @@ static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
    return disp & 0xfffc;
 }
 
-static bool reloc_pc14(tcg_insn_unit *pc,
+static bool reloc_pc14(TCGContext *s,
+                       tcg_insn_unit *pc,
                       const tcg_insn_unit *target)
 {
    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
    if (disp == (int16_t) disp) {
-        *pc = (*pc & ~0xfffc) | (disp & 0xfffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0xfffc) | (disp & 0xfffc);
        return true;
    }
    return false;
@@ -674,7 +678,7 @@ static const uint32_t tcg_to_isel[] = {
    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
 };
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_insn_unit *target;
@@ -686,9 +690,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
 
    switch (type) {
    case R_PPC_REL14:
-        return reloc_pc14(code_ptr, target);
+        return reloc_pc14(s, code_ptr, target);
    case R_PPC_REL24:
-        return reloc_pc24(code_ptr, target);
+        return reloc_pc24(s, code_ptr, target);
    case R_PPC_ADDR16:
        /*
         * We are (slightly) abusing this relocation type.  In particular,
@@ -1727,7 +1731,7 @@ static void tcg_out_mb(TCGContext *s, TCGArg a0)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    if (TCG_TARGET_REG_BITS == 64) {
        tcg_insn_unit i1, i2;
@@ -1756,12 +1760,18 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
 
        /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
           within qatomic_set that would fail to build a ppc32 host.  */
-        qatomic_set__nocheck((uint64_t *)jmp_addr, pair);
+        qatomic_set__nocheck((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 8);
    } else {
        intptr_t diff = addr - jmp_addr;
        tcg_debug_assert(in_range_b(diff));
-        qatomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
+        qatomic_set((uint32_t *)wr_addr, B | (diff & 0x3fffffc));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 4);
    }
 }
@@ -2012,7 +2022,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2060,7 +2070,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp s_bits = opc & MO_SIZE;
    TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2257,11 +2267,11 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
tcg/riscv/tcg-target.c.inc +23 -17
index 5432ee89db..19882db57f 100644
@@ -413,11 +413,11 @@ static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
    tcg_out32(s, encode_uj(opc, rd, imm));
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
+        (tcg_code_ptr_rw(s, p))[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
    }
 }
 
@@ -425,46 +425,52 @@ static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
 * Relocations
 */
 
-static bool reloc_sbimm12(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_sbimm12(TCGContext *s,
+                          tcg_insn_unit *code_ptr,
+                          const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
    if (offset == sextreg(offset, 1, 12) << 1) {
-        code_ptr[0] |= encode_sbimm12(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_sbimm12(offset);
        return true;
    }
 
    return false;
 }
 
-static bool reloc_jimm20(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_jimm20(TCGContext *s,
+                         tcg_insn_unit *code_ptr,
+                         const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
    if (offset == sextreg(offset, 1, 20) << 1) {
-        code_ptr[0] |= encode_ujimm20(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_ujimm20(offset);
        return true;
    }
 
    return false;
 }
 
-static bool reloc_call(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_call(TCGContext *s,
+                       tcg_insn_unit *code_ptr,
+                       const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
    int32_t lo = sextreg(offset, 0, 12);
    int32_t hi = offset - lo;
 
    if (offset == hi + lo) {
-        code_ptr[0] |= encode_uimm20(hi);
-        code_ptr[1] |= encode_imm12(lo);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_uimm20(hi);
+        (tcg_code_ptr_rw(s, code_ptr))[1] |= encode_imm12(lo);
        return true;
    }
 
    return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    uint32_t insn = *code_ptr;
@@ -478,7 +484,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        diff = value - (uintptr_t)code_ptr;
        short_jmp = diff == sextreg(diff, 0, 12);
        if (short_jmp) {
-            return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
+            return reloc_sbimm12(s, code_ptr, (tcg_insn_unit *)value);
        } else {
            /* Invert the condition */
            insn = insn ^ (1 << 12);
@@ -499,9 +505,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        }
        break;
    case R_RISCV_JAL:
-        return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
+        return reloc_jimm20(s, code_ptr, (tcg_insn_unit *)value);
    case R_RISCV_CALL:
-        return reloc_call(code_ptr, (tcg_insn_unit *)value);
+        return reloc_call(s, code_ptr, (tcg_insn_unit *)value);
    default:
        tcg_abort();
    }
@@ -557,7 +563,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
    if (tmp == (int32_t)tmp) {
        tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
        tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
-        ret = reloc_call(s->code_ptr - 2, (tcg_insn_unit *)val);
+        ret = reloc_call(s, s->code_ptr - 2, (tcg_insn_unit *)val);
        tcg_debug_assert(ret == true);
        return;
    }
@@ -875,7 +881,7 @@ static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
        /* long jump: -2147483646 to 2147483648 */
        tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
-        ret = reloc_call(s->code_ptr - 2, arg);\
+        ret = reloc_call(s, s->code_ptr - 2, arg);\
        tcg_debug_assert(ret == true);
    } else if (TCG_TARGET_REG_BITS == 64) {
        /* far jump: 64-bit */
@@ -1053,7 +1059,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                     (intptr_t) s->code_ptr, 0)) {
        return false;
    }
@@ -1088,7 +1094,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                     (intptr_t) s->code_ptr, 0)) {
        return false;
    }
tcg/s390/tcg-target.c.inc +8 -8
index 7c6917f3c4..f9498ddf63 100644
@@ -366,7 +366,7 @@ static void * const qemu_st_helpers[16] = {
 static const tcg_insn_unit *tb_ret_addr;
 uint64_t s390_facilities;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    intptr_t pcrel2;
@@ -378,13 +378,13 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
    switch (type) {
    case R_390_PC16DBL:
        if (pcrel2 == (int16_t)pcrel2) {
-            tcg_patch16(code_ptr, pcrel2);
+            tcg_patch16(s, code_ptr, pcrel2);
            return true;
        }
        break;
    case R_390_PC32DBL:
        if (pcrel2 == (int32_t)pcrel2) {
-            tcg_patch32(code_ptr, pcrel2);
+            tcg_patch32(s, code_ptr, pcrel2);
            return true;
        }
        break;
@@ -392,7 +392,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        if (value == sextract64(value, 0, 20)) {
            old = *(uint32_t *)code_ptr & 0xf00000ff;
            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
-            tcg_patch32(code_ptr, old);
+            tcg_patch32(s, code_ptr, old);
            return true;
        }
        break;
@@ -1613,7 +1613,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                     (intptr_t)s->code_ptr, 2)) {
        return false;
    }
@@ -1638,7 +1638,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                     (intptr_t)s->code_ptr, 2)) {
        return false;
    }
@@ -2576,9 +2576,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x07, count * sizeof(tcg_insn_unit));
+    memset(tcg_code_ptr_rw(s, p), 0x07, count * sizeof(tcg_insn_unit));
 }
 
 typedef struct {
tcg/sparc/tcg-target.c.inc +15 -8
index 5eac18b719..41dbd21cb4 100644
@@ -291,14 +291,14 @@ static inline int check_fit_i32(int32_t val, unsigned int bits)
 # define check_fit_ptr  check_fit_i32
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    uint32_t insn = *code_ptr;
    intptr_t pcrel;
 
    value += addend;
-    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr);
+    pcrel = tcg_ptr_byte_diff((const tcg_insn_unit *)value, code_ptr);
 
    switch (type) {
    case R_SPARC_WDISP16:
@@ -1048,11 +1048,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
@@ -1226,7 +1226,8 @@ static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
        }
    }
 
-    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
+    *tcg_code_ptr_rw(s, label_ptr) |=
+        INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
 #else
    if (SPARC64 && TARGET_LONG_BITS == 32) {
        tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL);
@@ -1822,7 +1823,7 @@ void tcg_register_jit(void *buf, size_t buf_size)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    intptr_t tb_disp = addr - tc_ptr;
    intptr_t br_disp = addr - jmp_addr;
@@ -1834,8 +1835,11 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
    tcg_debug_assert(br_disp == (int32_t)br_disp);
 
    if (!USE_REG_TB) {
-        qatomic_set((uint32_t *)jmp_addr,
+        qatomic_set((uint32_t *)wr_addr,
 		    deposit32(CALL, 0, 30, br_disp >> 2));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 4);
        return;
    }
@@ -1859,6 +1863,9 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
              | INSN_IMM13((tb_disp & 0x3ff) | -0x400));
    }
 
-    qatomic_set((uint64_t *)jmp_addr, deposit64(i2, 32, 32, i1));
+    qatomic_set((uint64_t *)wr_addr, deposit64(i2, 32, 32, i1));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 8);
 }
tcg/tcg-pool.c.inc +5 -4
index 82cbcc89bd..8434496e72 100644
@@ -119,7 +119,7 @@ static inline void new_pool_l8(TCGContext *s, int rtype, tcg_insn_unit *label,
 }
 
 /* To be provided by cpu/tcg-target.c.inc.  */
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count);
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count);
 
 static int tcg_out_pool_finalize(TCGContext *s)
 {
@@ -135,7 +135,7 @@ static int tcg_out_pool_finalize(TCGContext *s)
       again when allocating the next TranslationBlock structure.  */
    a = (void *)ROUND_UP((uintptr_t)s->code_ptr,
                         sizeof(tcg_target_ulong) * p->nlong);
-    tcg_out_nop_fill(s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
+    tcg_out_nop_fill(s, s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
    s->data_gen_ptr = a;
 
    for (; p != NULL; p = p->next) {
@@ -144,11 +144,12 @@ static int tcg_out_pool_finalize(TCGContext *s)
            if (unlikely(a > s->code_gen_highwater)) {
                return -1;
            }
-            memcpy(a, p->data, size);
+            memcpy(tcg_mirror_ptr_rw(s, a), p->data, size);
            a += size;
            l = p;
        }
-        if (!patch_reloc(p->label, p->rtype, (intptr_t)a - size, p->addend)) {
+        if (!patch_reloc(s, p->label, p->rtype,
+                         (intptr_t)a - size, p->addend)) {
            return -2;
        }
    }
tcg/tci/tcg-target.c.inc +3 -3
index d5a4d9d37c..133213be3a 100644
@@ -369,7 +369,7 @@ static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
 };
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    /* tcg_out_reloc always uses the same type, addend. */
@@ -377,9 +377,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
    tcg_debug_assert(addend == 0);
    tcg_debug_assert(value != 0);
    if (TCG_TARGET_REG_BITS == 32) {
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
    } else {
-        tcg_patch64(code_ptr, value);
+        tcg_patch64(s, code_ptr, value);
    }
    return true;
 }
#24 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

Quoted 3 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

-j

Quoted 1,544 lines · BALATON Zoltan
Quoted 144 lines · Joelle van Dyne

From: osy <[email protected]>

On iOS, we cannot allocate RWX pages without special entitlements. As a workaround, we can allocate a RX region and then mirror map it to a separate RX region. Then we can write to one region and execute from the other one.

We also define tcg_mirror_ptr_rw and tcg_code_ptr_rw to return a pointer to RW memory. The difference between the RW and RX regions is stored in the TCG context.

To ensure cache coherency, we flush the data cache in the RW mapping and then invalidate the instruction cache in the RX mapping (where applicable). Because data cache flush is OS defined on some architectures, we do not provide implementations for non iOS platforms (ARM/x86).

---

docs/devel/ios.rst           | 40 +++++++++++++++++++
configure                    |  1 +
include/exec/exec-all.h      |  8 ++++
include/tcg/tcg.h            | 17 ++++++++
tcg/aarch64/tcg-target.h     | 13 +++++-
tcg/arm/tcg-target.h         |  9 ++++-
tcg/i386/tcg-target.h        | 24 ++++++++++-
tcg/mips/tcg-target.h        |  8 +++-
tcg/ppc/tcg-target.h         |  8 +++-
tcg/riscv/tcg-target.h       |  9 ++++-
tcg/s390/tcg-target.h        | 13 +++++-
tcg/sparc/tcg-target.h       |  8 +++-
tcg/tci/tcg-target.h         |  9 ++++-
accel/tcg/cpu-exec.c         |  7 +++-
accel/tcg/translate-all.c    | 77 ++++++++++++++++++++++++++++++++++--
tcg/tcg.c                    | 56 +++++++++++++++++---------
tcg/aarch64/tcg-target.c.inc | 33 ++++++++++------
tcg/arm/tcg-target.c.inc     | 25 ++++++------
tcg/i386/tcg-target.c.inc    | 18 ++++-----
tcg/mips/tcg-target.c.inc    | 35 +++++++++-------
tcg/ppc/tcg-target.c.inc     | 38 +++++++++++-------
tcg/riscv/tcg-target.c.inc   | 40 +++++++++++--------
tcg/s390/tcg-target.c.inc    | 16 ++++----
tcg/sparc/tcg-target.c.inc   | 23 +++++++----
tcg/tcg-pool.c.inc           |  9 +++--
tcg/tci/tcg-target.c.inc     |  6 +--
26 files changed, 416 insertions(+), 134 deletions(-)

create mode 100644 docs/devel/ios.rst

docs/devel/ios.rst new file +40 -0
new file mode 100644
index 0000000000..dba9fdd868
@@ -0,0 +1,40 @@
+===========
+iOS Support
+===========
+
+To run qemu on the iOS platform, some modifications were required. Most of the
+modifications are conditioned on the ``CONFIG_IOS`` and ``CONFIG_IOS_JIT``
+configuration variables.
+
+Build support
+-------------
+
+For the code to compile, certain changes in the block driver and the slirp
+driver had to be made. There is no ``system()`` call, so code requiring it had
+to be disabled.
+
+``ucontext`` support is broken on iOS. The implementation from ``libucontext``
+is used instead.
+
+Because ``fork()`` is not allowed on iOS apps, the option to build qemu and the
+utilities as shared libraries is added. Note that because qemu does not perform
+resource cleanup in most cases (open files, allocated memory, etc), it is
+advisable that the user implements a proxy layer for syscalls so resources can
+be kept track by the app that uses qemu as a shared library.
+
+JIT support
+-----------
+
+On iOS, allocating RWX pages require special entitlements not usually granted to
+apps. However, it is possible to use `bulletproof JIT`_ with a development
+certificate. This means that we need to allocate one chunk of memory with RX
+permissions and then mirror map the same memory with RW permissions. We generate
+code to the mirror mapping and execute the original mapping.
+
+With ``CONFIG_IOS_JIT`` defined, we store inside the TCG context the difference
+between the two mappings. Then, we make sure that any writes to JIT memory is
+done to the pointer + the difference (in order to get a pointer to the mirror
+mapped space). Additionally, we make sure to flush the data cache before we
+invalidate the instruction cache so the changes are seen in both mappings.
+
+.. _bulletproof JIT: https://www.blackhat.com/docs/us-16/materials/us-16-Krstic.pdf
configure +0 -0
index 0b7e25e7a5..93d6fd5ce2 100755
@@ -6214,6 +6214,7 @@ fi
 

if test "$ios" = "yes" ; then
echo "CONFIG_IOS=y" >> $config_host_mak
+ echo "CONFIG_IOS_JIT=y" >> $config_host_mak
fi

if test "$solaris" = "yes" ; then

include/exec/exec-all.h +8 -0
index 66f9b4cca6..2db155a772 100644
@@ -483,6 +483,14 @@ struct TranslationBlock {
    uintptr_t jmp_list_head;
    uintptr_t jmp_list_next[2];
    uintptr_t jmp_dest[2];
+
+#if defined(CONFIG_IOS_JIT)
+    /*
+     * Store difference to writable mirror
+     * We need this when patching the jump instructions
+     */
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 };
 
 extern bool parallel_cpus;
include/tcg/tcg.h +6 -0
index 79c5ff8dab..ade01d2e41 100644
@@ -627,6 +627,9 @@ struct TCGContext {
    size_t code_gen_buffer_size;
    void *code_gen_ptr;
    void *data_gen_ptr;
+#if defined(CONFIG_IOS_JIT)
+    ptrdiff_t code_rw_mirror_diff;
+#endif
 
    /* Threshold to flush the translated code buffer.  */
    void *code_gen_highwater;
@@ -677,6 +680,20 @@ struct TCGContext {
    target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
 };
 
+static inline void *tcg_mirror_ptr_rw(TCGContext *s, const void *rx)
+{
+#if defined(CONFIG_IOS_JIT)

include/tcg/tcg.h struct TCGContext { · line 695

Quoted 10 lines
include/tcg/tcg.h +10 -0
+    return (void *)rx + s->code_rw_mirror_diff;
+#else
+    return (void *)rx;
+#endif
+}
+
+static inline tcg_insn_unit *tcg_code_ptr_rw(TCGContext *s, const void *rx)
+{
+    return (tcg_insn_unit *)tcg_mirror_ptr_rw(s, rx);
+}

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

Quoted 40 lines
include/tcg/tcg.h +1 -0
+
 extern TCGContext tcg_init_ctx;
 extern __thread TCGContext *tcg_ctx;
 extern TCGv_env cpu_env;
tcg/aarch64/tcg-target.h +12 -1
index a2b22b4305..78c97460a1 100644
@@ -150,6 +150,7 @@ typedef enum {
 
 #if defined(__APPLE__)
 void sys_icache_invalidate(void *start, size_t len);
+void sys_dcache_flush(void *start, size_t len);
 #endif
 
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
@@ -163,7 +164,17 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #endif
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/arm/tcg-target.h +0 -0
index 17e771374d..d8d7e7e239 100644

tcg/arm/tcg-target.h static inline void flush_icache_range(uintptr_t start, uintptr_t stop) · line 142

Quoted 4 lines
tcg/arm/tcg-target.h +1 -0
@@ -139,8 +139,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    __builtin___clear_cache((char *) start, (char *) stop);
 }
 
+#if defined(CONFIG_IOS_JIT)

Is that !defined or are you missing an implementation and #else here?

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

Regards,
BALATON Zoltan

Quoted 1,329 lines
tcg/arm/tcg-target.h +7 -1
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/i386/tcg-target.h +22 -2
index abd4ac7fc0..cdc440ce36 100644
@@ -206,16 +206,36 @@ extern bool have_avx2;
 #define TCG_TARGET_extract_i64_valid(ofs, len) \
    (((ofs) == 8 && (len) == 8) || ((ofs) + (len)) == 32)
 
+#ifdef __APPLE__
+void sys_dcache_flush(void *start, size_t len);
+#endif
+
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#if defined(__APPLE__)
+    sys_dcache_flush((char *)start, stop - start);
+#else
+#error "Missing function to flush data cache"
+#endif
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
-    qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
+    qatomic_set((int32_t *)wr_addr, addr - (jmp_addr + 4));
    /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    /* we do need to flush mirror dcache */
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 /* This defines the natural memory order supported by this
tcg/mips/tcg-target.h +7 -1
index c6b091d849..80dcba5358 100644
@@ -212,7 +212,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    cacheflush ((void *)start, stop-start, ICACHE);
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #ifdef CONFIG_SOFTMMU
 #define TCG_TARGET_NEED_LDST_LABELS
tcg/ppc/tcg-target.h +7 -1
index be10363956..23d7a337c9 100644
@@ -176,7 +176,13 @@ extern bool have_vsx;
 #define TCG_TARGET_HAS_cmpsel_vec       0
 
 void flush_icache_range(uintptr_t start, uintptr_t stop);
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
tcg/riscv/tcg-target.h +8 -1
index 032439d806..d42b361991 100644
@@ -164,8 +164,15 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    __builtin___clear_cache((char *)start, (char *)stop);
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 /* not defined -- call should be eliminated at compile time */
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 #define TCG_TARGET_DEFAULT_MO (0)
 
tcg/s390/tcg-target.h +12 -1
index 63c8797bd3..d67632512d 100644
@@ -149,13 +149,24 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
+
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
    intptr_t disp = addr - (jmp_addr - 2);
    qatomic_set((int32_t *)jmp_addr, disp / 2);
    /* no need to flush icache explicitly */
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
 }
 
 #ifdef CONFIG_SOFTMMU
tcg/sparc/tcg-target.h +7 -1
index 633841ebf2..d102e13692 100644
@@ -176,7 +176,13 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
    }
 }
 
-void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t);
+void tb_target_set_jmp_target(uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+#error "Unimplemented dcache flush function"
+}
+#endif
 
 #define TCG_TARGET_NEED_POOL_LABELS
 
tcg/tci/tcg-target.h +8 -1
index 8c1c1d265d..2a258ea350 100644
@@ -195,6 +195,12 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
 }
 
+#if defined(CONFIG_IOS_JIT)
+static inline void flush_dcache_range(uintptr_t start, uintptr_t stop)
+{
+}
+#endif
+
 /* We could notice __i386__ or __s390x__ and reduce the barriers depending
   on the host.  But if you want performance, you use the normal backend.
   We prefer consistency across hosts on this.  */
@@ -203,7 +209,8 @@ static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 #define TCG_TARGET_HAS_MEMORY_BSWAP     1
 
 static inline void tb_target_set_jmp_target(uintptr_t tc_ptr,
-                                            uintptr_t jmp_addr, uintptr_t addr)
+                                            uintptr_t jmp_addr, uintptr_t addr,
+                                            uintptr_t wr_addr)
 {
    /* patch the branch destination */
    qatomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
accel/tcg/cpu-exec.c +6 -1
index 58aea605d8..821aefdea2 100644
@@ -354,7 +354,12 @@ void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
    if (TCG_TARGET_HAS_direct_jump) {
        uintptr_t offset = tb->jmp_target_arg[n];
        uintptr_t tc_ptr = (uintptr_t)tb->tc.ptr;
-        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr);
+#if defined(CONFIG_IOS_JIT)
+        uintptr_t wr_addr = tc_ptr + offset + tb->code_rw_mirror_diff;
+#else
+        uintptr_t wr_addr = tc_ptr + offset;
+#endif
+        tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr, wr_addr);
    } else {
        tb->jmp_target_arg[n] = addr;
    }
accel/tcg/translate-all.c +74 -3
index d76097296d..eb1d8fbe2f 100644
@@ -60,6 +60,22 @@
 #include "sysemu/cpu-timers.h"
 #include "sysemu/tcg.h"
 
+#if defined(CONFIG_IOS_JIT)
+#include <mach/mach.h>
+extern kern_return_t mach_vm_remap(vm_map_t target_task,
+                                   mach_vm_address_t *target_address,
+                                   mach_vm_size_t size,
+                                   mach_vm_offset_t mask,
+                                   int flags,
+                                   vm_map_t src_task,
+                                   mach_vm_address_t src_address,
+                                   boolean_t copy,
+                                   vm_prot_t *cur_protection,
+                                   vm_prot_t *max_protection,
+                                   vm_inherit_t inheritance
+                                  );
+#endif
+
 /* #define DEBUG_TB_INVALIDATE */
 /* #define DEBUG_TB_FLUSH */
 /* make various TB consistency checks */
@@ -302,10 +318,12 @@ static target_long decode_sleb128(uint8_t **pp)
 
 static int encode_search(TranslationBlock *tb, uint8_t *block)
 {
-    uint8_t *highwater = tcg_ctx->code_gen_highwater;
-    uint8_t *p = block;
+    const uint8_t *highwater, *start;
+    uint8_t *p;
    int i, j, n;
 
+    highwater = tcg_mirror_ptr_rw(tcg_ctx, tcg_ctx->code_gen_highwater);
+    start = p = tcg_mirror_ptr_rw(tcg_ctx, block);
    for (i = 0, n = tb->icount; i < n; ++i) {
        target_ulong prev;
 
@@ -329,7 +347,7 @@ static int encode_search(TranslationBlock *tb, uint8_t *block)
        }
    }
 
-    return p - block;
+    return p - start;
 }
 
 /* The cpu state corresponding to 'searched_pc' is restored.
@@ -1067,7 +1085,11 @@ static inline void *alloc_code_gen_buffer(void)
 #else
 static inline void *alloc_code_gen_buffer(void)
 {
+#if defined(CONFIG_IOS_JIT)
+    int prot = PROT_READ | PROT_EXEC;
+#else
    int prot = PROT_WRITE | PROT_READ | PROT_EXEC;
+#endif
    int flags = MAP_PRIVATE | MAP_ANONYMOUS;
    size_t size = tcg_ctx->code_gen_buffer_size;
    void *buf;
@@ -1118,6 +1140,39 @@ static inline void *alloc_code_gen_buffer(void)
 }
 #endif /* USE_STATIC_CODE_GEN_BUFFER, WIN32, POSIX */
 
+#if defined(CONFIG_IOS_JIT)
+static inline void *alloc_jit_rw_mirror(void *base, size_t size)
+{
+    kern_return_t ret;
+    mach_vm_address_t mirror;
+    vm_prot_t cur_prot, max_prot;
+
+    mirror = 0;
+    ret = mach_vm_remap(mach_task_self(),
+                        &mirror,
+                        size,
+                        0,
+                        VM_FLAGS_ANYWHERE | VM_FLAGS_RANDOM_ADDR,
+                        mach_task_self(),
+                        (mach_vm_address_t)base,
+                        false,
+                        &cur_prot,
+                        &max_prot,
+                        VM_INHERIT_NONE
+                       );
+    if (ret != KERN_SUCCESS) {
+        return NULL;
+    }
+
+    if (mprotect((void *)mirror, size, PROT_READ | PROT_WRITE) != 0) {
+        munmap((void *)mirror, size);
+        return NULL;
+    }
+
+    return (void *)mirror;
+}
+#endif /* CONFIG_IOS_JIT */
+
 static inline void code_gen_alloc(size_t tb_size)
 {
    tcg_ctx->code_gen_buffer_size = size_code_gen_buffer(tb_size);
@@ -1126,6 +1181,19 @@ static inline void code_gen_alloc(size_t tb_size)
        fprintf(stderr, "Could not allocate dynamic translator buffer\n");
        exit(1);
    }
+#if defined(CONFIG_IOS_JIT)
+    void *mirror;
+
+    /* For iOS JIT we need a mirror mapping for code execution */
+    mirror = alloc_jit_rw_mirror(tcg_ctx->code_gen_buffer,
+                                 tcg_ctx->code_gen_buffer_size
+                                );
+    if (mirror == NULL) {
+        fprintf(stderr, "Could not remap code buffer mirror\n");
+        exit(1);
+    }
+    tcg_ctx->code_rw_mirror_diff = mirror - tcg_ctx->code_gen_buffer;
+#endif /* CONFIG_IOS_JIT */
 }
 
 static bool tb_cmp(const void *ap, const void *bp)
@@ -1721,6 +1789,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
        cpu_loop_exit(cpu);
    }
 
+#if defined(CONFIG_IOS_JIT)
+    tb->code_rw_mirror_diff = tcg_ctx->code_rw_mirror_diff;
+#endif
    gen_code_buf = tcg_ctx->code_gen_ptr;
    tb->tc.ptr = gen_code_buf;
    tb->pc = pc;
tcg/tcg.c +37 -19
index bb890c506d..6cf70c33b0 100644
@@ -70,7 +70,7 @@
 static void tcg_target_init(TCGContext *s);
 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode);
 static void tcg_target_qemu_prologue(TCGContext *s);
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend);
 
 /* The CIE and FDE header definitions will be common to all hosts.  */
@@ -203,13 +203,15 @@ static TCGRegSet tcg_target_call_clobber_regs;
 #if TCG_TARGET_INSN_UNIT_SIZE == 1
 static __attribute__((unused)) inline void tcg_out8(TCGContext *s, uint8_t v)
 {
-    *s->code_ptr++ = v;
+    *tcg_code_ptr_rw(s, s->code_ptr) = v;
+    s->code_ptr++;
 }
 
-static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch8(TCGContext *s,
+                                                      tcg_insn_unit *p,
                                                      uint8_t v)
 {
-    *p = v;
+    *tcg_code_ptr_rw(s, p) = v;
 }
 #endif
 
@@ -217,21 +219,23 @@ static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out16(TCGContext *s, uint16_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (2 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch16(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint16_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
-        *p = v;
+        *tcg_code_ptr_rw(s, p) = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -240,21 +244,23 @@ static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out32(TCGContext *s, uint32_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (4 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch32(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint32_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
        *p = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -263,21 +269,23 @@ static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p,
 static __attribute__((unused)) inline void tcg_out64(TCGContext *s, uint64_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
-        *s->code_ptr++ = v;
+        *tcg_code_ptr_rw(s, s->code_ptr) = v;
+        s->code_ptr++;
    } else {
        tcg_insn_unit *p = s->code_ptr;
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
        s->code_ptr = p + (8 / TCG_TARGET_INSN_UNIT_SIZE);
    }
 }
 
-static __attribute__((unused)) inline void tcg_patch64(tcg_insn_unit *p,
+static __attribute__((unused)) inline void tcg_patch64(TCGContext *s,
+                                                       tcg_insn_unit *p,
                                                       uint64_t v)
 {
    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
        *p = v;
    } else {
-        memcpy(p, &v, sizeof(v));
+        memcpy(tcg_code_ptr_rw(s, p), &v, sizeof(v));
    }
 }
 #endif
@@ -325,7 +333,7 @@ static bool tcg_resolve_relocs(TCGContext *s)
        uintptr_t value = l->u.value;
 
        QSIMPLEQ_FOREACH(r, &l->relocs, next) {
-            if (!patch_reloc(r->ptr, r->type, value, r->addend)) {
+            if (!patch_reloc(s, r->ptr, r->type, value, r->addend)) {
                return false;
            }
        }
@@ -1039,7 +1047,7 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s)
    }
    qatomic_set(&s->code_gen_ptr, next);
    s->data_gen_ptr = NULL;
-    return tb;
+    return (TranslationBlock *)tcg_mirror_ptr_rw(s, tb);
 }
 
 void tcg_prologue_init(TCGContext *s)
@@ -1076,6 +1084,10 @@ void tcg_prologue_init(TCGContext *s)
 #endif
 
    buf1 = s->code_ptr;
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range((uintptr_t)tcg_mirror_ptr_rw(s, buf0),
+                       (uintptr_t)tcg_mirror_ptr_rw(s, buf1));
+#endif
    flush_icache_range((uintptr_t)buf0, (uintptr_t)buf1);
 
    /* Deduct the prologue from the buffer.  */
@@ -4267,6 +4279,12 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
        return -2;
    }
 
+#if defined(CONFIG_IOS_JIT)
+    /* flush data cache on mirror */
+    flush_dcache_range((uintptr_t)tcg_mirror_ptr_rw(s, s->code_buf),
+                       (uintptr_t)tcg_mirror_ptr_rw(s, s->code_ptr));
+#endif
+
    /* flush instruction cache */
    flush_icache_range((uintptr_t)s->code_buf, (uintptr_t)s->code_ptr);
 
tcg/aarch64/tcg-target.c.inc +20 -13
index 1aa5f37fc6..ea3c06e917 100644
@@ -78,40 +78,44 @@ static const int tcg_target_call_oarg_regs[1] = {
 #define TCG_REG_GUEST_BASE TCG_REG_X28
 #endif
 
-static inline bool reloc_pc26(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc26(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = target - code_ptr;
    if (offset == sextract64(offset, 0, 26)) {
        /* read instruction, mask away previous PC_REL26 parameter contents,
           set the proper offset, then write back the instruction. */
-        *code_ptr = deposit32(*code_ptr, 0, 26, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 0, 26, offset);
        return true;
    }
    return false;
 }
 
-static inline bool reloc_pc19(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc19(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = target - code_ptr;
    if (offset == sextract64(offset, 0, 19)) {
-        *code_ptr = deposit32(*code_ptr, 5, 19, offset);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            deposit32(*tcg_code_ptr_rw(s, code_ptr), 5, 19, offset);
        return true;
    }
    return false;
 }
 
-static inline bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static inline bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                               intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(addend == 0);
    switch (type) {
    case R_AARCH64_JUMP26:
    case R_AARCH64_CALL26:
-        return reloc_pc26(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc26(s, code_ptr, (tcg_insn_unit *)value);
    case R_AARCH64_CONDBR19:
-        return reloc_pc19(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc19(s, code_ptr, (tcg_insn_unit *)value);
    default:
        g_assert_not_reached();
    }
@@ -1343,7 +1347,7 @@ static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    tcg_insn_unit i1, i2;
    TCGType rt = TCG_TYPE_I64;
@@ -1364,7 +1368,10 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
        i2 = I3401_ADDI | rt << 31 | (addr & 0xfff) << 10 | rd << 5 | rd;
    }
    pair = (uint64_t)i2 << 32 | i1;
-    qatomic_set((uint64_t *)jmp_addr, pair);
+    qatomic_set((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 8);
 }
 
@@ -1583,7 +1590,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -1608,7 +1615,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;
 
-    if (!reloc_pc19(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc19(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2919,11 +2926,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_insn(s, 3207, RET, TCG_REG_LR);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
tcg/arm/tcg-target.c.inc +14 -11
index 5ceb06a971..ee8ab72c2e 100644
@@ -187,18 +187,21 @@ static const uint8_t tcg_cond_to_arm_cond[] = {
    [TCG_COND_GTU] = COND_HI,
 };
 
-static inline bool reloc_pc24(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc24(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
    if (offset == sextract32(offset, 0, 24)) {
-        *code_ptr = (*code_ptr & ~0xffffff) | (offset & 0xffffff);
+        *tcg_code_ptr_rw(s, code_ptr) =
+            (*tcg_code_ptr_rw(s, code_ptr) & ~0xffffff) | (offset & 0xffffff);
        return true;
    }
    return false;
 }
 
-static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
+static inline bool reloc_pc13(TCGContext *s,
+                              tcg_insn_unit *code_ptr,
                              const tcg_insn_unit *target)
 {
    ptrdiff_t offset = tcg_ptr_byte_diff(target, code_ptr) - 8;
@@ -211,21 +214,21 @@ static inline bool reloc_pc13(tcg_insn_unit *code_ptr,
        }
        insn = deposit32(insn, 23, 1, u);
        insn = deposit32(insn, 0, 12, offset);
-        *code_ptr = insn;
+        *tcg_code_ptr_rw(s, code_ptr) = insn;
        return true;
    }
    return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(addend == 0);
 
    if (type == R_ARM_PC24) {
-        return reloc_pc24(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc24(s, code_ptr, (tcg_insn_unit *)value);
    } else if (type == R_ARM_PC13) {
-        return reloc_pc13(code_ptr, (tcg_insn_unit *)value);
+        return reloc_pc13(s, code_ptr, (tcg_insn_unit *)value);
    } else {
        g_assert_not_reached();
    }
@@ -1350,7 +1353,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    void *func;
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -1413,7 +1416,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!reloc_pc24(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc24(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2257,11 +2260,11 @@ static inline void tcg_out_movi(TCGContext *s, TCGType type,
    tcg_out_movi32(s, COND_AL, ret, arg);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = INSN_NOP;
+        (tcg_code_ptr_rw(s, p))[i] = INSN_NOP;
    }
 }
 
tcg/i386/tcg-target.c.inc +9 -9
index 69acb51f5d..76bdaf748f 100644
@@ -167,7 +167,7 @@ static bool have_lzcnt;
 
 static const tcg_insn_unit *tb_ret_addr;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    value += addend;
@@ -179,14 +179,14 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        }
        /* FALLTHRU */
    case R_386_32:
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
        break;
    case R_386_PC8:
        value -= (uintptr_t)code_ptr;
        if (value != (int8_t)value) {
            return false;
        }
-        tcg_patch8(code_ptr, value);
+        tcg_patch8(s, code_ptr, value);
        break;
    default:
        tcg_abort();
@@ -1805,9 +1805,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int rexw = (l->type == TCG_TYPE_I64 ? P_REXW : 0);
 
    /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
    if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
    }
 
    if (TCG_TARGET_REG_BITS == 32) {
@@ -1890,9 +1890,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    TCGReg retaddr;
 
    /* resolve label address */
-    tcg_patch32(label_ptr[0], s->code_ptr - label_ptr[0] - 4);
+    tcg_patch32(s, label_ptr[0], s->code_ptr - label_ptr[0] - 4);
    if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
-        tcg_patch32(label_ptr[1], s->code_ptr - label_ptr[1] - 4);
+        tcg_patch32(s, label_ptr[1], s->code_ptr - label_ptr[1] - 4);
    }
 
    if (TCG_TARGET_REG_BITS == 32) {
@@ -3842,9 +3842,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_opc(s, OPC_RET, 0, 0, 0);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x90, count);
+    memset(tcg_code_ptr_rw(s, p), 0x90, count);
 }
 
 static void tcg_target_init(TCGContext *s)
tcg/mips/tcg-target.c.inc +21 -14
index ace413afde..d4389f7135 100644
@@ -153,10 +153,12 @@ static inline uint32_t reloc_pc16_val(const tcg_insn_unit *pc,
    return disp & 0xffff;
 }
 
-static inline void reloc_pc16(tcg_insn_unit *pc,
+static inline void reloc_pc16(TCGContext *s,
+                              tcg_insn_unit *pc,
                              const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 16, reloc_pc16_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 16, reloc_pc16_val(pc, target));
 }
 
 static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
@@ -166,18 +168,20 @@ static inline uint32_t reloc_26_val(const tcg_insn_unit *pc,
    return ((uintptr_t)target >> 2) & 0x3ffffff;
 }
 
-static inline void reloc_26(tcg_insn_unit *pc,
+static inline void reloc_26(TCGContext *s,
+                            tcg_insn_unit *pc,
                            const tcg_insn_unit *target)
 {
-    *pc = deposit32(*pc, 0, 26, reloc_26_val(pc, target));
+    *tcg_code_ptr_rw(s, pc) =
+        deposit32(*tcg_code_ptr_rw(s, pc), 0, 26, reloc_26_val(pc, target));
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_debug_assert(type == R_MIPS_PC16);
    tcg_debug_assert(addend == 0);
-    reloc_pc16(code_ptr, (tcg_insn_unit *)value);
+    reloc_pc16(s, code_ptr, (tcg_insn_unit *)value);
    return true;
 }
 
@@ -929,7 +933,7 @@ static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
 
    tcg_out_opc_br(s, b_opc, arg1, arg2);
    if (l->has_value) {
-        reloc_pc16(s->code_ptr - 1, l->u.value_ptr);
+        reloc_pc16(s, s->code_ptr - 1, l->u.value_ptr);
    } else {
        tcg_out_reloc(s, s->code_ptr - 1, R_MIPS_PC16, l, 0);
    }
@@ -1320,9 +1324,9 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int i;
 
    /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
    }
 
    i = 1;
@@ -1350,7 +1354,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO);
-    reloc_pc16(s->code_ptr - 1, l->raddr);
+    reloc_pc16(s, s->code_ptr - 1, l->raddr);
 
    /* delay slot */
    if (TCG_TARGET_REG_BITS == 64 && l->type == TCG_TYPE_I32) {
@@ -1370,9 +1374,9 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    int i;
 
    /* resolve label address */
-    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    reloc_pc16(s, l->label_ptr[0], s->code_ptr);
    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
-        reloc_pc16(l->label_ptr[1], s->code_ptr);
+        reloc_pc16(s, l->label_ptr[1], s->code_ptr);
    }
 
    i = 1;
@@ -2662,9 +2666,12 @@ static void tcg_target_init(TCGContext *s)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
-    qatomic_set((uint32_t *)jmp_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+    qatomic_set((uint32_t *)wr_addr, deposit32(OPC_J, 0, 26, addr >> 2));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 4);
 }
 
tcg/ppc/tcg-target.c.inc +24 -14
index 289388b603..50d72529d2 100644
@@ -192,12 +192,14 @@ static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
    return disp & 0x3fffffc;
 }
 
-static bool reloc_pc24(tcg_insn_unit *pc,
+static bool reloc_pc24(TCGContext *s,
+                       tcg_insn_unit *pc,
                       const tcg_insn_unit *target)
 {
    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
    if (in_range_b(disp)) {
-        *pc = (*pc & ~0x3fffffc) | (disp & 0x3fffffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0x3fffffc) | (disp & 0x3fffffc);
        return true;
    }
    return false;
@@ -211,12 +213,14 @@ static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
    return disp & 0xfffc;
 }
 
-static bool reloc_pc14(tcg_insn_unit *pc,
+static bool reloc_pc14(TCGContext *s,
+                       tcg_insn_unit *pc,
                       const tcg_insn_unit *target)
 {
    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
    if (disp == (int16_t) disp) {
-        *pc = (*pc & ~0xfffc) | (disp & 0xfffc);
+        *tcg_code_ptr_rw(s, pc) =
+            (*tcg_code_ptr_rw(s, pc) & ~0xfffc) | (disp & 0xfffc);
        return true;
    }
    return false;
@@ -674,7 +678,7 @@ static const uint32_t tcg_to_isel[] = {
    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
 };
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    tcg_insn_unit *target;
@@ -686,9 +690,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
 
    switch (type) {
    case R_PPC_REL14:
-        return reloc_pc14(code_ptr, target);
+        return reloc_pc14(s, code_ptr, target);
    case R_PPC_REL24:
-        return reloc_pc24(code_ptr, target);
+        return reloc_pc24(s, code_ptr, target);
    case R_PPC_ADDR16:
        /*
         * We are (slightly) abusing this relocation type.  In particular,
@@ -1727,7 +1731,7 @@ static void tcg_out_mb(TCGContext *s, TCGArg a0)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    if (TCG_TARGET_REG_BITS == 64) {
        tcg_insn_unit i1, i2;
@@ -1756,12 +1760,18 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
 
        /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
           within qatomic_set that would fail to build a ppc32 host.  */
-        qatomic_set__nocheck((uint64_t *)jmp_addr, pair);
+        qatomic_set__nocheck((uint64_t *)wr_addr, pair);
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 8);
    } else {
        intptr_t diff = addr - jmp_addr;
        tcg_debug_assert(in_range_b(diff));
-        qatomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
+        qatomic_set((uint32_t *)wr_addr, B | (diff & 0x3fffffc));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 4);
    }
 }
@@ -2012,7 +2022,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp opc = get_memop(oi);
    TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2060,7 +2070,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    MemOp s_bits = opc & MO_SIZE;
    TCGReg hi, lo, arg = TCG_REG_R3;
 
-    if (!reloc_pc14(lb->label_ptr[0], s->code_ptr)) {
+    if (!reloc_pc14(s, lb->label_ptr[0], s->code_ptr)) {
        return false;
    }
 
@@ -2257,11 +2267,11 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
tcg/riscv/tcg-target.c.inc +23 -17
index 5432ee89db..19882db57f 100644
@@ -413,11 +413,11 @@ static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
    tcg_out32(s, encode_uj(opc, rd, imm));
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
+        (tcg_code_ptr_rw(s, p))[i] = encode_i(OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
    }
 }
 
@@ -425,46 +425,52 @@ static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
 * Relocations
 */
 
-static bool reloc_sbimm12(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_sbimm12(TCGContext *s,
+                          tcg_insn_unit *code_ptr,
+                          const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
    if (offset == sextreg(offset, 1, 12) << 1) {
-        code_ptr[0] |= encode_sbimm12(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_sbimm12(offset);
        return true;
    }
 
    return false;
 }
 
-static bool reloc_jimm20(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_jimm20(TCGContext *s,
+                         tcg_insn_unit *code_ptr,
+                         const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
 
    if (offset == sextreg(offset, 1, 20) << 1) {
-        code_ptr[0] |= encode_ujimm20(offset);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_ujimm20(offset);
        return true;
    }
 
    return false;
 }
 
-static bool reloc_call(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
+static bool reloc_call(TCGContext *s,
+                       tcg_insn_unit *code_ptr,
+                       const tcg_insn_unit *target)
 {
    intptr_t offset = (intptr_t)target - (intptr_t)code_ptr;
    int32_t lo = sextreg(offset, 0, 12);
    int32_t hi = offset - lo;
 
    if (offset == hi + lo) {
-        code_ptr[0] |= encode_uimm20(hi);
-        code_ptr[1] |= encode_imm12(lo);
+        (tcg_code_ptr_rw(s, code_ptr))[0] |= encode_uimm20(hi);
+        (tcg_code_ptr_rw(s, code_ptr))[1] |= encode_imm12(lo);
        return true;
    }
 
    return false;
 }
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    uint32_t insn = *code_ptr;
@@ -478,7 +484,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        diff = value - (uintptr_t)code_ptr;
        short_jmp = diff == sextreg(diff, 0, 12);
        if (short_jmp) {
-            return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
+            return reloc_sbimm12(s, code_ptr, (tcg_insn_unit *)value);
        } else {
            /* Invert the condition */
            insn = insn ^ (1 << 12);
@@ -499,9 +505,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        }
        break;
    case R_RISCV_JAL:
-        return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
+        return reloc_jimm20(s, code_ptr, (tcg_insn_unit *)value);
    case R_RISCV_CALL:
-        return reloc_call(code_ptr, (tcg_insn_unit *)value);
+        return reloc_call(s, code_ptr, (tcg_insn_unit *)value);
    default:
        tcg_abort();
    }
@@ -557,7 +563,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
    if (tmp == (int32_t)tmp) {
        tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
        tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
-        ret = reloc_call(s->code_ptr - 2, (tcg_insn_unit *)val);
+        ret = reloc_call(s, s->code_ptr - 2, (tcg_insn_unit *)val);
        tcg_debug_assert(ret == true);
        return;
    }
@@ -875,7 +881,7 @@ static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
        /* long jump: -2147483646 to 2147483648 */
        tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
-        ret = reloc_call(s->code_ptr - 2, arg);\
+        ret = reloc_call(s, s->code_ptr - 2, arg);\
        tcg_debug_assert(ret == true);
    } else if (TCG_TARGET_REG_BITS == 64) {
        /* far jump: 64-bit */
@@ -1053,7 +1059,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                     (intptr_t) s->code_ptr, 0)) {
        return false;
    }
@@ -1088,7 +1094,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
    }
 
    /* resolve label address */
-    if (!patch_reloc(l->label_ptr[0], R_RISCV_BRANCH,
+    if (!patch_reloc(s, l->label_ptr[0], R_RISCV_BRANCH,
                     (intptr_t) s->code_ptr, 0)) {
        return false;
    }
tcg/s390/tcg-target.c.inc +8 -8
index 7c6917f3c4..f9498ddf63 100644
@@ -366,7 +366,7 @@ static void * const qemu_st_helpers[16] = {
 static const tcg_insn_unit *tb_ret_addr;
 uint64_t s390_facilities;
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    intptr_t pcrel2;
@@ -378,13 +378,13 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
    switch (type) {
    case R_390_PC16DBL:
        if (pcrel2 == (int16_t)pcrel2) {
-            tcg_patch16(code_ptr, pcrel2);
+            tcg_patch16(s, code_ptr, pcrel2);
            return true;
        }
        break;
    case R_390_PC32DBL:
        if (pcrel2 == (int32_t)pcrel2) {
-            tcg_patch32(code_ptr, pcrel2);
+            tcg_patch32(s, code_ptr, pcrel2);
            return true;
        }
        break;
@@ -392,7 +392,7 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
        if (value == sextract64(value, 0, 20)) {
            old = *(uint32_t *)code_ptr & 0xf00000ff;
            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
-            tcg_patch32(code_ptr, old);
+            tcg_patch32(s, code_ptr, old);
            return true;
        }
        break;
@@ -1613,7 +1613,7 @@ static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                     (intptr_t)s->code_ptr, 2)) {
        return false;
    }
@@ -1638,7 +1638,7 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
    TCGMemOpIdx oi = lb->oi;
    MemOp opc = get_memop(oi);
 
-    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
+    if (!patch_reloc(s, lb->label_ptr[0], R_390_PC16DBL,
                     (intptr_t)s->code_ptr, 2)) {
        return false;
    }
@@ -2576,9 +2576,9 @@ static void tcg_target_qemu_prologue(TCGContext *s)
    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
-    memset(p, 0x07, count * sizeof(tcg_insn_unit));
+    memset(tcg_code_ptr_rw(s, p), 0x07, count * sizeof(tcg_insn_unit));
 }
 
 typedef struct {
tcg/sparc/tcg-target.c.inc +15 -8
index 5eac18b719..41dbd21cb4 100644
@@ -291,14 +291,14 @@ static inline int check_fit_i32(int32_t val, unsigned int bits)
 # define check_fit_ptr  check_fit_i32
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    uint32_t insn = *code_ptr;
    intptr_t pcrel;
 
    value += addend;
-    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr);
+    pcrel = tcg_ptr_byte_diff((const tcg_insn_unit *)value, code_ptr);
 
    switch (type) {
    case R_SPARC_WDISP16:
@@ -1048,11 +1048,11 @@ static void tcg_target_qemu_prologue(TCGContext *s)
 #endif
 }
 
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count)
 {
    int i;
    for (i = 0; i < count; ++i) {
-        p[i] = NOP;
+        (tcg_code_ptr_rw(s, p))[i] = NOP;
    }
 }
 
@@ -1226,7 +1226,8 @@ static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
        }
    }
 
-    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
+    *tcg_code_ptr_rw(s, label_ptr) |=
+        INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
 #else
    if (SPARC64 && TARGET_LONG_BITS == 32) {
        tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL);
@@ -1822,7 +1823,7 @@ void tcg_register_jit(void *buf, size_t buf_size)
 }
 
 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
-                              uintptr_t addr)
+                              uintptr_t addr, uintptr_t wr_addr)
 {
    intptr_t tb_disp = addr - tc_ptr;
    intptr_t br_disp = addr - jmp_addr;
@@ -1834,8 +1835,11 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
    tcg_debug_assert(br_disp == (int32_t)br_disp);
 
    if (!USE_REG_TB) {
-        qatomic_set((uint32_t *)jmp_addr,
+        qatomic_set((uint32_t *)wr_addr,
                  deposit32(CALL, 0, 30, br_disp >> 2));
+#if defined(CONFIG_IOS_JIT)
+        flush_dcache_range(wr_addr, wr_addr + 4);
+#endif
        flush_icache_range(jmp_addr, jmp_addr + 4);
        return;
    }
@@ -1859,6 +1863,9 @@ void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
              | INSN_IMM13((tb_disp & 0x3ff) | -0x400));
    }
 
-    qatomic_set((uint64_t *)jmp_addr, deposit64(i2, 32, 32, i1));
+    qatomic_set((uint64_t *)wr_addr, deposit64(i2, 32, 32, i1));
+#if defined(CONFIG_IOS_JIT)
+    flush_dcache_range(wr_addr, wr_addr + 8);
+#endif
    flush_icache_range(jmp_addr, jmp_addr + 8);
 }
tcg/tcg-pool.c.inc +5 -4
index 82cbcc89bd..8434496e72 100644
@@ -119,7 +119,7 @@ static inline void new_pool_l8(TCGContext *s, int rtype, tcg_insn_unit *label,
 }
 
 /* To be provided by cpu/tcg-target.c.inc.  */
-static void tcg_out_nop_fill(tcg_insn_unit *p, int count);
+static void tcg_out_nop_fill(TCGContext *s, tcg_insn_unit *p, int count);
 
 static int tcg_out_pool_finalize(TCGContext *s)
 {
@@ -135,7 +135,7 @@ static int tcg_out_pool_finalize(TCGContext *s)
       again when allocating the next TranslationBlock structure.  */
    a = (void *)ROUND_UP((uintptr_t)s->code_ptr,
                         sizeof(tcg_target_ulong) * p->nlong);
-    tcg_out_nop_fill(s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
+    tcg_out_nop_fill(s, s->code_ptr, (tcg_insn_unit *)a - s->code_ptr);
    s->data_gen_ptr = a;
 
    for (; p != NULL; p = p->next) {
@@ -144,11 +144,12 @@ static int tcg_out_pool_finalize(TCGContext *s)
            if (unlikely(a > s->code_gen_highwater)) {
                return -1;
            }
-            memcpy(a, p->data, size);
+            memcpy(tcg_mirror_ptr_rw(s, a), p->data, size);
            a += size;
            l = p;
        }
-        if (!patch_reloc(p->label, p->rtype, (intptr_t)a - size, p->addend)) {
+        if (!patch_reloc(s, p->label, p->rtype,
+                         (intptr_t)a - size, p->addend)) {
            return -2;
        }
    }
tcg/tci/tcg-target.c.inc +3 -3
index d5a4d9d37c..133213be3a 100644
@@ -369,7 +369,7 @@ static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
 };
 #endif
 
-static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
+static bool patch_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
 {
    /* tcg_out_reloc always uses the same type, addend. */
@@ -377,9 +377,9 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
    tcg_debug_assert(addend == 0);
    tcg_debug_assert(value != 0);
    if (TCG_TARGET_REG_BITS == 32) {
-        tcg_patch32(code_ptr, value);
+        tcg_patch32(s, code_ptr, value);
    } else {
-        tcg_patch64(code_ptr, value);
+        tcg_patch64(s, code_ptr, value);
    }
    return true;
 }
#25 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

Quoted 11 lines · Joelle van Dyne
Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

OK that explains it, haven't looked at it at that detail.

Quoted 3 lines

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Definitely cleaner to have the cast either in a macro or inline func than manually having it everywhere. The static inline in v2 looks better than the macro in v1 so unless others disagree it's fine this way, I'm not the one who decides, I was just asking if we can avoid having two static inlines relying on casting void * but if you also dereference as above then returning the right type is needed. Let's see what Richard says who suggested the function instead of a macro but it does look more readable than the previous macro.

Quoted 4 lines
Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that

But this was in tcg/arm/tcg-target.h which is ARM but maybe you mean only x86 and 64bit ARM which is aarch64 but not 32bit ARM. I've noticed this only after sending the question.

Quoted 10 lines

Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

Quoted 3 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

Sure but it's easier to review if a single patch has only similar changes even if it touches a lot of files than if it does independent stuff intermixed unless it's a small patch (but even then QEMU tends to prefer a lot of smaller patches instead of combining changes in a single patch). That's also good for bisectability so that's also something to consider when splitting patches. Not sure if in this case this can be split up into two working changes because RW mapping may not work without cache flushes and cache flushes may not be added before having the RW split but having two patches for the review that can be squashed in the final series may still help. But lets see if this gets reviewed without further splitting or what others say.

Not sure you're aware of this: https://wiki.qemu.org/Planning/5.2 but if this series does not get merged this week don't be surprised if your next opportunity to pick it up will be in December (when most people who can review it will be on holyday so it can be easily take longer). So maybe you could try pushing it and do everything to make reviewers' job easier if you want it in the next release. Otherwise you'll have time to polish it until next year.

Also it may be good to fix checkpatch errors (warnings may be OK but errors are not) that are reported even if it's not in your code (it could be a separate clean up patch before your changes or for small things in the same patch) otherwise automated tests may not run which can also delay reviews and merging:

https://patchew.org/QEMU/[email protected]/

and I assume you already know this:

https://wiki.qemu.org/Contribute/SubmitAPatch

It might be overwhelming and off putting sometimes to try getting series into QEMU but please don't give up.

Regards,
BALATON Zoltan

#26 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

In reply to Joelle van Dyne
Quoted 8 lines · Joelle van Dyne
Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

Quoted 7 lines

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Yep.

Quoted 6 lines
Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

*shrug* As opposed to the nice compiler error you get for a missing function declaration?

That said, I think __builtin___clear_cache() may be the target-independent runtime function that you need. Both GCC and LLVM support this, and I'd be surprised if that doesn't carry through to iOS.

Quoted 7 lines
Quoted 3 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

It occurs to me that the majority of the code changes in patches 5 and 6 are due to your choice that code_gen_buffer points to the RX copy and not the RW copy.

Swap the two, and instead have an inline function that produces the executable pointer from the rw pointer, and suddenly there are very much fewer changes required.

For the most part, tcg/$cpu/ generates pc-relative code, so it need not consider the absolute address. There are a few exceptions including, obviously, 32-bit x86. But the number of places that occurs is small.

There's the assignment to tb->tc.ptr of course, and tcg_ctx.code_gen_prologue/epilogue.

In any case, each of these changes (generic, per tcg backend) can occur before you finally add a non-zero displacement that actually separates the RX and RW mappings.

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

r~

#27 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

Quoted 27 lines · Richard Henderson
Quoted 8 lines · Joelle van Dyne
Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

Quoted 7 lines

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Yep.

Quoted 6 lines
Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

Quoted 6 lines

*shrug* As opposed to the nice compiler error you get for a missing function declaration?

That said, I think __builtin___clear_cache() may be the target-independent runtime function that you need. Both GCC and LLVM support this, and I'd be surprised if that doesn't carry through to iOS.

You would think so but unfortunately iOS likes to be "special"
See: https://gitlab.haskell.org/ghc/ghc/-/issues/8561

Quoted 9 lines
Quoted 8 lines
Quoted 4 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

Quoted 6 lines

It occurs to me that the majority of the code changes in patches 5 and 6 are due to your choice that code_gen_buffer points to the RX copy and not the RW copy.

Swap the two, and instead have an inline function that produces the executable pointer from the rw pointer, and suddenly there are very much fewer changes required.

In the original incarnation of this code from a year ago, this was the chosen path, but we ran into other issues which unfortunately were not recorded anywhere. I think one issue is that the mirror offset has to be stored somewhere that can easily be retrieved or otherwise code and structs have to be modified to take a copy of the value. If we default to the RW copy, there may be less LOC changes but the changes will be more significant. This version has also been running by iOS users for the past year or so without issue and a major change in the logic would probably require a lot more testing.

Quoted 12 lines

For the most part, tcg/$cpu/ generates pc-relative code, so it need not consider the absolute address. There are a few exceptions including, obviously, 32-bit x86. But the number of places that occurs is small.

There's the assignment to tb->tc.ptr of course, and tcg_ctx.code_gen_prologue/epilogue.

In any case, each of these changes (generic, per tcg backend) can occur before you finally add a non-zero displacement that actually separates the RX and RW mappings.

Quoted 5 lines

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

Okay, I will do that.

-j

Quoted 3 lines

r~

#28 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

In reply to Richard Henderson
Quoted 5 lines

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

Just spent an hour trying to implement this for Linux and running into issues.

1) Doesn't work because MREMAP_DONTUNMAP does in fact remove the entry from the page table. According to the man pages "After completion, any access to the range specified by old_address and old_size will result in a page fault." Seems like the feature is designed around memory locking, not mirror mapping.

2) I tried doing shm_open() and mmap() but you can't PROT_EXEC on shm (see https://stackoverflow.com/questions/25275777/shared-executable-memory
)

I think it may be possible to map a file on an executable partition, but I can foresee countless issues there including some security issues. Anyone have any other ideas?

-j

Quoted 68 lines · Richard Henderson
Quoted 8 lines · Joelle van Dyne
Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

Quoted 7 lines

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Yep.

Quoted 6 lines
Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

*shrug* As opposed to the nice compiler error you get for a missing function declaration?

That said, I think __builtin___clear_cache() may be the target-independent runtime function that you need. Both GCC and LLVM support this, and I'd be surprised if that doesn't carry through to iOS.

Quoted 7 lines
Quoted 3 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

It occurs to me that the majority of the code changes in patches 5 and 6 are due to your choice that code_gen_buffer points to the RX copy and not the RW copy.

Swap the two, and instead have an inline function that produces the executable pointer from the rw pointer, and suddenly there are very much fewer changes required.

For the most part, tcg/$cpu/ generates pc-relative code, so it need not consider the absolute address. There are a few exceptions including, obviously, 32-bit x86. But the number of places that occurs is small.

There's the assignment to tb->tc.ptr of course, and tcg_ctx.code_gen_prologue/epilogue.

In any case, each of these changes (generic, per tcg backend) can occur before you finally add a non-zero displacement that actually separates the RX and RW mappings.

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

r~

#29 lore raw

Re: [PATCH v2 6/9] tcg: implement mirror mapped JIT for iOS

Quoted 13 lines · Joelle van Dyne
Quoted 5 lines

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

Just spent an hour trying to implement this for Linux and running into issues.

1) Doesn't work because MREMAP_DONTUNMAP does in fact remove the entry from the page table. According to the man pages "After completion, any access to the range specified by old_address and old_size will result in a page fault." Seems like the feature is designed around memory locking, not mirror mapping.

Quoted 7 lines

2) I tried doing shm_open() and mmap() but you can't PROT_EXEC on shm (see https://stackoverflow.com/questions/25275777/shared-executable-memory
)

I think it may be possible to map a file on an executable partition, but I can foresee countless issues there including some security issues. Anyone have any other ideas?

Maybe memfd_create(2) + mmap? It doesn't require a tmpfs mount, so it should be affected by noexec.
-Alex

Quoted 74 lines

-j

Quoted 68 lines · Richard Henderson
Quoted 8 lines · Joelle van Dyne
Quoted 3 lines

Explicit cast may not be needed here so this could be a macro if caling it differently helps or why don't you just use tcg_mirror_prr_rw directly everywhere?

There are quite a bit of code that depends on tcg_insn_unit * type such as

*tcg_code_ptr_rw(s, code_ptr) = insn;

Quoted 7 lines

and

(tcg_code_ptr_rw(s, p))[i] = NOP;

I think it's cleaner to not have to manually cast in every one of 30+ instances of this. In v1, I used a macro but was told to use an inline function instead.

Yep.

Quoted 6 lines
Quoted 1 line

Is that !defined or are you missing an implementation and #else here?

No, flush_dcache_range is only needed when mirror mapped (after writing to the RW mirror). Now there is no iOS compatible compiler for any other arch than x86 and ARM. However, in the slim chance that Apple decides to change arch again in the future and moves to RISC-V or something, then we get a nice compiler error.

*shrug* As opposed to the nice compiler error you get for a missing function declaration?

That said, I think __builtin___clear_cache() may be the target-independent runtime function that you need. Both GCC and LLVM support this, and I'd be surprised if that doesn't carry through to iOS.

Quoted 7 lines
Quoted 3 lines

Maybe this patch could be split up some more, making the RW offset handling and cache management separate patches even if they don't work separately just to make it easier to review.

I can probably do that for v3 but imo most of the LOC here is because the same change has to be done to every TCG target. No matter how you split up the patches, it will look like a lot of changes.

It occurs to me that the majority of the code changes in patches 5 and 6 are due to your choice that code_gen_buffer points to the RX copy and not the RW copy.

Swap the two, and instead have an inline function that produces the executable pointer from the rw pointer, and suddenly there are very much fewer changes required.

For the most part, tcg/$cpu/ generates pc-relative code, so it need not consider the absolute address. There are a few exceptions including, obviously, 32-bit x86. But the number of places that occurs is small.

There's the assignment to tb->tc.ptr of course, and tcg_ctx.code_gen_prologue/epilogue.

In any case, each of these changes (generic, per tcg backend) can occur before you finally add a non-zero displacement that actually separates the RX and RW mappings.

Finally, I'd like to have this implemented on Linux as well, or I'm afraid the feature will bit-rot. This can be trivially done by either (1) MREMAP_DONTUNMAP or (2) mapping from posix shared memory instead of MAP_ANON so that you can map the same memory twice. Thus virtually all of the ifdefs should go away.

r~

#30 lore raw

PATCH v2 7/9 tcg: mirror mapping RWX pages for iOS optional

In reply to Joelle van Dyne

From: osy <[email protected]>

This allows jailbroken devices with entitlements to switch the option off.

---
 include/sysemu/tcg.h      |  2 +-
 accel/tcg/tcg-all.c       | 27 +++++++++++++++++-
 accel/tcg/translate-all.c | 60 +++++++++++++++++++++++++--------------
 bsd-user/main.c           |  2 +-
 linux-user/main.c         |  2 +-
 qemu-options.hx           | 11 +++++++
 6 files changed, 79 insertions(+), 25 deletions(-)
include/sysemu/tcg.h +1 -1
index d9d3ca8559..569f90b11d 100644
@@ -8,7 +8,7 @@
 #ifndef SYSEMU_TCG_H
 #define SYSEMU_TCG_H
 
-void tcg_exec_init(unsigned long tb_size);
+void tcg_exec_init(unsigned long tb_size, bool mirror_rwx);
 #ifdef CONFIG_TCG
 extern bool tcg_allowed;
 #define tcg_enabled() (tcg_allowed)
accel/tcg/tcg-all.c +26 -1
index fa1208158f..5845744396 100644
@@ -39,6 +39,7 @@ struct TCGState {
 
     bool mttcg_enabled;
     unsigned long tb_size;
+    bool mirror_rwx;
 };
 typedef struct TCGState TCGState;
 
@@ -94,6 +95,7 @@ static void tcg_accel_instance_init(Object *obj)
     TCGState *s = TCG_STATE(obj);
 
     s->mttcg_enabled = default_mttcg_enabled();
+    s->mirror_rwx = false;
 }
 
 bool mttcg_enabled;
@@ -102,7 +104,7 @@ static int tcg_init(MachineState *ms)
 {
     TCGState *s = TCG_STATE(current_accel());
 
-    tcg_exec_init(s->tb_size * 1024 * 1024);
+    tcg_exec_init(s->tb_size * 1024 * 1024, s->mirror_rwx);
     mttcg_enabled = s->mttcg_enabled;
     cpus_register_accel(&tcg_cpus);
 
@@ -168,6 +170,22 @@ static void tcg_set_tb_size(Object *obj, Visitor *v,
     s->tb_size = value;
 }
 
+#ifdef CONFIG_IOS_JIT
+static bool tcg_get_mirror_rwx(Object *obj, Error **errp)
+{
+    TCGState *s = TCG_STATE(obj);
+
+    return s->mirror_rwx;
+}
+
+static void tcg_set_mirror_rwx(Object *obj, bool value, Error **errp)
+{
+    TCGState *s = TCG_STATE(obj);
+
+    s->mirror_rwx = value;
+}
+#endif
+
 static void tcg_accel_class_init(ObjectClass *oc, void *data)
 {
     AccelClass *ac = ACCEL_CLASS(oc);
@@ -185,6 +203,13 @@ static void tcg_accel_class_init(ObjectClass *oc, void *data)
     object_class_property_set_description(oc, "tb-size",
         "TCG translation block cache size");
 
+#ifdef CONFIG_IOS_JIT
+    object_class_property_add_bool(oc, "mirror-rwx",
+        tcg_get_mirror_rwx, tcg_set_mirror_rwx);
+    object_class_property_set_description(oc, "mirror-rwx",
+        "mirror map executable pages for TCG on iOS");
+#endif
+
 }
 
 static const TypeInfo tcg_accel_type = {
accel/tcg/translate-all.c +39 -21

114 lines not on this page · raw message

bsd-user/main.c +1 -1
index ac40d79bfa..ffd4888a26 100644
@@ -910,7 +910,7 @@ int main(int argc, char **argv)
     }
 
     /* init tcg before creating CPUs and to get qemu_host_page_size */
-    tcg_exec_init(0);
+    tcg_exec_init(0, false);
 
     cpu_type = parse_cpu_option(cpu_model);
     cpu = cpu_create(cpu_type);
linux-user/main.c +1 -1
index 75c9785157..3856b2611d 100644
@@ -705,7 +705,7 @@ int main(int argc, char **argv, char **envp)
     cpu_type = parse_cpu_option(cpu_model);
 
     /* init tcg before creating CPUs and to get qemu_host_page_size */
-    tcg_exec_init(0);
+    tcg_exec_init(0, false);
 
     cpu = cpu_create(cpu_type);
     env = cpu->env_ptr;
qemu-options.hx +11 -0
index 1da52a269c..deb39d56f5 100644
@@ -123,6 +123,9 @@ DEF("accel", HAS_ARG, QEMU_OPTION_accel,
     "                igd-passthru=on|off (enable Xen integrated Intel graphics passthrough, default=off)\n"
     "                kernel-irqchip=on|off|split controls accelerated irqchip support (default=on)\n"
     "                kvm-shadow-mem=size of KVM shadow MMU in bytes\n"
+#ifdef CONFIG_IOS_JIT
+    "                mirror-rwx=on|off (mirror map executable pages for TCG on iOS)\n"
+#endif
     "                tb-size=n (TCG translation block cache size)\n"
     "                thread=single|multi (enable multi-threaded TCG)\n", QEMU_ARCH_ALL)
 SRST
@@ -148,6 +151,14 @@ SRST
     ``kvm-shadow-mem=size``
         Defines the size of the KVM shadow MMU.
 
+#ifdef CONFIG_IOS_JIT
+
+    ``mirror-rwx=on|off``
+        Only applicable to TCG running on iOS hosts. When enabled, TB code
+        is written to a mirror mapped address separate from the address that is
+        executed. By default, this is disabled.
+#endif
+
     ``tb-size=n``
         Controls the size (in MiB) of the TCG translation block cache.
 
Signature · 1 line
2.24.3 (Apple Git-128)
#31 lore raw

Re: [PATCH v2 7/9] tcg: mirror mapping RWX pages for iOS optional

Quoted 6 lines · Joelle van Dyne

From: osy <[email protected]>

This allows jailbroken devices with entitlements to switch the option off.

---

I can guess why this performs better: half the page table entries and thus half the tlb entries required. Which for any non-trivially sized jit arena is going to add up. [*]

In line with my comments re patch 6, and making this feature available everywhere (or at least non-windows), the ifdefs would go away. I might also suggest default on for CONFIG_DEBUG_TCG and otherwise default off (when the host os allows).

r~

* Which makes me wonder how much we should use the "const TranslationBlock *" version of that structure in the rx mapping, so that we're using a tlb entry that is more likely to be present, since we've just branched from the code (or just about to branch to the code) on the same page.

#32 lore raw

PATCH v2 8/9 tcg: support JIT on Apple Silicon

In reply to Joelle van Dyne

From: osy <[email protected]>

https://developer.apple.com/documentation/apple_silicon/porting_just-in-time_compilers_to_apple_silicon

For < iOS 14, reverse engineered functions from libsystem_pthread.dylib is implemented to handle APRR supported SoCs.

The following rules apply for JIT write protect:
* JIT write-protect is enabled before tcg_qemu_tb_exec()
* JIT write-protect is disabled after tcg_qemu_tb_exec() returns
* JIT write-protect is disabled inside do_tb_phys_invalidate() but if it is called inside of tcg_qemu_tb_exec() then write-protect will be enabled again before returning.
* JIT write-protect is disabled by cpu_loop_exit() for interrupt handling.
* JIT write-protect is disabled everywhere else.

---
 configure                   | 20 +++++++++
 include/exec/exec-all.h     |  2 +
 include/tcg/tcg-apple-jit.h | 85 +++++++++++++++++++++++++++++++++++++
 include/tcg/tcg.h           |  3 ++
 accel/tcg/cpu-exec-common.c |  2 +
 accel/tcg/cpu-exec.c        |  2 +
 accel/tcg/translate-all.c   | 51 ++++++++++++++++++++++
 tcg/tcg.c                   |  4 ++
 8 files changed, 169 insertions(+)
 create mode 100644 include/tcg/tcg-apple-jit.h
configure +20 -0
index 93d6fd5ce2..2221c276f4 100755
@@ -5868,6 +5868,22 @@ but not implemented on your system"
     fi
 fi
 
+##########################################
+# check for Apple Silicon JIT function
+
+if [ "$darwin" = "yes" ] ; then
+  cat > $TMPC << EOF
+#include <pthread.h>
+int main() { pthread_jit_write_protect_np(0); return 0; }
+EOF
+  if ! compile_prog ""; then
+    have_pthread_jit_protect='no'
+  else
+    have_pthread_jit_protect='yes'
+  fi
+fi
+
+
 ##########################################
 # End of CC checks
 # After here, no more $cc or $ld runs
@@ -6988,6 +7004,10 @@ if test "$secret_keyring" = "yes" ; then
   echo "CONFIG_SECRET_KEYRING=y" >> $config_host_mak
 fi
 
+if test "$have_pthread_jit_protect" = "yes" ; then
+  echo "HAVE_PTHREAD_JIT_PROTECT=y" >> $config_host_mak
+fi
+
 if test "$tcg_interpreter" = "yes"; then
   QEMU_INCLUDES="-iquote ${source_path}/tcg/tci $QEMU_INCLUDES"
 elif test "$ARCH" = "sparc64" ; then
include/exec/exec-all.h +2 -0
index 2db155a772..253af30a2e 100644
@@ -521,6 +521,8 @@ TranslationBlock *tb_htable_lookup(CPUState *cpu, target_ulong pc,
                                    target_ulong cs_base, uint32_t flags,
                                    uint32_t cf_mask);
 void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr);
+void tb_exec_lock(void);
+void tb_exec_unlock(void);
 
 /* GETPC is the true target of the return instruction that we'll execute.  */
 #if defined(CONFIG_TCG_INTERPRETER)
include/tcg/tcg-apple-jit.h new file +85 -0

85 lines not on this page · raw message

include/tcg/tcg.h +3 -0
index ade01d2e41..0a50adc013 100644
@@ -630,6 +630,9 @@ struct TCGContext {
 #if defined(CONFIG_IOS_JIT)
     ptrdiff_t code_rw_mirror_diff;
 #endif
+#if defined(CONFIG_DARWIN) && !defined(CONFIG_TCG_INTERPRETER)
+    bool code_gen_locked; /* on Darwin each thread tracks W^X flags */
+#endif
 
     /* Threshold to flush the translated code buffer.  */
     void *code_gen_highwater;
accel/tcg/cpu-exec-common.c +2 -0
index 12c1e3e974..f1eb767b02 100644
@@ -64,6 +64,8 @@ void cpu_reloading_memory_map(void)
 
 void cpu_loop_exit(CPUState *cpu)
 {
+    /* Unlock JIT write protect if applicable. */
+    tb_exec_unlock();
     /* Undo the setting in cpu_tb_exec.  */
     cpu->can_do_io = 1;
     siglongjmp(cpu->jmp_env, 1);
accel/tcg/cpu-exec.c +2 -0
index 821aefdea2..d00d17f3c6 100644
@@ -175,7 +175,9 @@ static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
     }
 #endif /* DEBUG_DISAS */
 
+    tb_exec_lock();
     ret = tcg_qemu_tb_exec(env, tb_ptr);
+    tb_exec_unlock();
     cpu->can_do_io = 1;
     last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
     tb_exit = ret & TB_EXIT_MASK;
accel/tcg/translate-all.c +51 -0

92 lines not on this page · raw message

tcg/tcg.c +4 -0
index 6cf70c33b0..b105111858 100644
@@ -781,6 +781,8 @@ static void alloc_tcg_plugin_context(TCGContext *s)
 void tcg_register_thread(void)
 {
     tcg_ctx = &tcg_init_ctx;
+
+    tb_exec_unlock();
 }
 #else
 void tcg_register_thread(void)
@@ -815,6 +817,8 @@ void tcg_register_thread(void)
     err = tcg_region_initial_alloc__locked(tcg_ctx);
     g_assert(!err);
     qemu_mutex_unlock(&region.lock);
+
+    tb_exec_unlock();
 }
 #endif /* !CONFIG_USER_ONLY */
 
Signature · 1 line
2.24.3 (Apple Git-128)
#33 lore raw

PATCH v2 9/9 block: check availablity for preadv/pwritev on mac

In reply to Joelle van Dyne

From: osy <[email protected]>

macOS 11/iOS 14 added preadv/pwritev APIs. Due to weak linking, configure will succeed with CONFIG_PREADV even when targeting a lower OS version. We therefore need to check at run time if we can actually use these APIs.

---
 block/file-posix.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
block/file-posix.c +12 -0
index cdc73b5f1d..d7482036a3 100644
@@ -1393,12 +1393,24 @@ static bool preadv_present = true;
 static ssize_t
 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 {
+#ifdef CONFIG_DARWIN /* preadv introduced in macOS 11 */
+    if (!__builtin_available(macOS 11, iOS 14, watchOS 7, tvOS 14, *)) {
+        preadv_present = false;
+        return -ENOSYS;
+    } else
+#endif
     return preadv(fd, iov, nr_iov, offset);
 }
 
 static ssize_t
 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 {
+#ifdef CONFIG_DARWIN /* pwritev introduced in macOS 11 */
+    if (!__builtin_available(macOS 11, iOS 14, watchOS 7, tvOS 14, *)) {
+        preadv_present = false;
+        return -ENOSYS;
+    } else
+#endif
     return pwritev(fd, iov, nr_iov, offset);
 }
 
Signature · 1 line
2.24.3 (Apple Git-128)
#34 lore raw

Re: [PATCH v2 9/9] block: check availablity for preadv/pwritev on mac

Quoted 1 line · Joelle van Dyne

From: osy <[email protected]>

That "From:" line looks wrong ... could you please fix the "Author" of your patches / your git config?

Quoted 3 lines

macOS 11/iOS 14 added preadv/pwritev APIs. Due to weak linking, configure will succeed with CONFIG_PREADV even when targeting a lower OS version. We therefore need to check at run time if we can actually use these APIs.

That sounds like the wrong approach to me ... could you please try to fix the check in "configure" instead? E.g. by running compile_prog with "-Werror", so that the test fails if there is no valid prototype available?

Thomas

#35 lore raw

Re: [PATCH v2 9/9] block: check availablity for preadv/pwritev on mac

Quoted 6 lines · Thomas Huth
Quoted 1 line · Joelle van Dyne

From: osy <[email protected]>

That "From:" line looks wrong ... could you please fix the "Author" of your patches / your git config?

osy wrote the original changes. I joined the UTM project to help bring the changes upstream with permission. However, they have agreed that if required that we can use my name as the author.

Quoted 8 lines
Quoted 4 lines

macOS 11/iOS 14 added preadv/pwritev APIs. Due to weak linking, configure will succeed with CONFIG_PREADV even when targeting a lower OS version. We therefore need to check at run time if we can actually use these APIs.

That sounds like the wrong approach to me ... could you please try to fix the check in "configure" instead? E.g. by running compile_prog with "-Werror", so that the test fails if there is no valid prototype available?

It's not that simple. Xcode 11 and below (supporting macOS 10.15 and below, iOS 13 and below, etc) does not have preadv/pwritev symbols defined and would fail to compile. Xcode 12 (supporting macOS 11 and below, iOS 14 and below, etc) have preadv/pwritev weakly defined so if it runs on, for example, 10.15, it would abort. There is no way to determine at compile time if you can use preadv/pwritev or not when building with Xcode 12. The availability checks are Apple's preferred way to handle this kind of situation (they discourage directly checking if an API exists on a system).

-j

Quoted 3 lines

Thomas

#36 lore raw

Re: [PATCH v2 9/9] block: check availablity for preadv/pwritev on mac

Quoted 10 lines · Joelle van Dyne
Quoted 6 lines · Thomas Huth
Quoted 1 line · Joelle van Dyne

From: osy <[email protected]>

That "From:" line looks wrong ... could you please fix the "Author" of your patches / your git config?

osy wrote the original changes. I joined the UTM project to help bring the changes upstream with permission. However, they have agreed that if required that we can use my name as the author.

In any way, that "users.noreply.github.com" does not look like a valid e-mail address and should be replaced.

Quoted 8 lines
Quoted 8 lines
Quoted 4 lines

macOS 11/iOS 14 added preadv/pwritev APIs. Due to weak linking, configure will succeed with CONFIG_PREADV even when targeting a lower OS version. We therefore need to check at run time if we can actually use these APIs.

That sounds like the wrong approach to me ... could you please try to fix the check in "configure" instead? E.g. by running compile_prog with "-Werror", so that the test fails if there is no valid prototype available?

Quoted 9 lines

It's not that simple. Xcode 11 and below (supporting macOS 10.15 and below, iOS 13 and below, etc) does not have preadv/pwritev symbols defined and would fail to compile. Xcode 12 (supporting macOS 11 and below, iOS 14 and below, etc) have preadv/pwritev weakly defined so if it runs on, for example, 10.15, it would abort. There is no way to determine at compile time if you can use preadv/pwritev or not when building with Xcode 12. The availability checks are Apple's preferred way to handle this kind of situation (they discourage directly checking if an API exists on a system).

Ok, got it now, thanks for the detailed explanation!

Thomas

#37 lore raw
In reply to Joelle van Dyne
Quoted 30 lines · Joelle van Dyne

These set of changes brings QEMU TCG to iOS devices and future Apple Silicon devices. They were originally developed last year and have been working in the UTM app. Recently, we ported the changes to master, re-wrote a lot of the build script changes for meson, and broke up the patches into more distinct units.

A summary of the changes:

* CONFIG_IOS and CONFIG_IOS_JIT defined when building for iOS and iOS specific changes (as well as unsupported code) are gated behind it.
* A new dependency, libucontext is added since iOS does not have native ucontext and broken support for sigaltstack. libucontext is available as a new option for coroutine backend.
* On stock iOS devices, there is a workaround for running JIT code without any special entitlement. It requires the JIT region to be mirror mapped with one region RW and another one RX. To support this style of JIT, TCG is changed to support writing to a different code_ptr. These changes are gated by the CONFIG_IOS_JIT.
* For (recent) jailbroken iOS devices as well as upcoming Apple Silicon devices, there are new rules for applications supporting JIT (with the proper entitlement). These rules are implemented as well.

Since v2:

* Changed getting mirror pointer from a macro to inline functions
* Split constification of TCG code pointers to separate patch
* Removed slirp updates (will send future patch once slirp changes are in)
* Removed shared library patch (will send future patch)

-j

Quoted 10 lines

osy (9):
configure: option to disable host block devices
configure: cross-compiling without cross_prefix
qemu: add support for iOS host
coroutine: add libucontext as external library
tcg: add const hints for code pointers
tcg: implement mirror mapped JIT for iOS
tcg: mirror mapping RWX pages for iOS optional
tcg: support JIT on Apple Silicon
block: check availablity for preadv/pwritev on mac

Is there maybe also a way to compile-test this automatically in a macOS container on Cirrus-CI or Travis, too? ... otherwise I'm a little bit afraid that this support might bit-rot quite fast again, since most developers don't have the corresponding setup...

Thomas

#38 lore raw
Quoted 43 lines · Thomas Huth
Quoted 30 lines · Joelle van Dyne

These set of changes brings QEMU TCG to iOS devices and future Apple Silicon devices. They were originally developed last year and have been working in the UTM app. Recently, we ported the changes to master, re-wrote a lot of the build script changes for meson, and broke up the patches into more distinct units.

A summary of the changes:

* CONFIG_IOS and CONFIG_IOS_JIT defined when building for iOS and iOS specific changes (as well as unsupported code) are gated behind it.
* A new dependency, libucontext is added since iOS does not have native ucontext and broken support for sigaltstack. libucontext is available as a new option for coroutine backend.
* On stock iOS devices, there is a workaround for running JIT code without any special entitlement. It requires the JIT region to be mirror mapped with one region RW and another one RX. To support this style of JIT, TCG is changed to support writing to a different code_ptr. These changes are gated by the CONFIG_IOS_JIT.
* For (recent) jailbroken iOS devices as well as upcoming Apple Silicon devices, there are new rules for applications supporting JIT (with the proper entitlement). These rules are implemented as well.

Since v2:

* Changed getting mirror pointer from a macro to inline functions
* Split constification of TCG code pointers to separate patch
* Removed slirp updates (will send future patch once slirp changes are in)
* Removed shared library patch (will send future patch)

-j

Quoted 10 lines

osy (9):
configure: option to disable host block devices
configure: cross-compiling without cross_prefix
qemu: add support for iOS host
coroutine: add libucontext as external library
tcg: add const hints for code pointers
tcg: implement mirror mapped JIT for iOS
tcg: mirror mapping RWX pages for iOS optional
tcg: support JIT on Apple Silicon
block: check availablity for preadv/pwritev on mac

Quoted 4 lines

Is there maybe also a way to compile-test this automatically in a macOS container on Cirrus-CI or Travis, too? ... otherwise I'm a little bit afraid that this support might bit-rot quite fast again, since most developers don't have the corresponding setup...

That's a tough one. In theory we can use the cirrus-ci environment to build for iOS but there's no easy way to get the dependencies. For UTM, we have this complicated script
https://github.com/utmapp/UTM/blob/master/scripts/build_dependencies.sh
which downloads and builds all the dependencies from source. Adding something like that to QEMU would tie it to specific versions of those libraries as well as the download link. Not sure if that's desirable.

I am currently refactoring the code so the bigger changes (mirror JIT) are separate features that doesn't depend on iOS. Other features work on Darwin so they can be tested in the macOS builds. But there will be iOS specific code that I don't think would be easy to CI and that's a general limitation of open source projects on iOS.

-j

Quoted 3 lines

Thomas