On Wed, Oct 7, 2026 at 10:29 PM Pierrick Bouvier
<[email protected]> wrote:
>
> On 10/7/2026 12:48 AM, Yonggang Luo wrote:
> > On Wed, Oct 7, 2026 at 12:03 PM Pierrick Bouvier
> > <[email protected]> wrote:
> >>
> >> For now, TARGET_ARCH only contains 32 vs 64 bit arch. The variants,
> >> based on endianness, are accessible through TARGET_NAME. However,
> >> TARGET_NAME is a string, and thus can't be concatenated to generate code
> >> and retrieve matching SYS_EMU_TARGET enum entry.
> >> Thus, export this information in TARGET_ARCH_VARIANT as an unquoted
> >> identifier.
> >>
> >> There is one more subtlety, user binaries have extra targets that are
> >> not present in SYS_EMU_TARGET, like armeb.
> >> Thus, for user binaries, simply export TARGET_ARCH.
> >
> >
> >>
> >> Signed-off-by: Pierrick Bouvier <[email protected]>
> >> ---
> >> meson.build | 11 +++++++++--
> >> 1 file changed, 9 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/meson.build b/meson.build
> >> index cae1591cd28..9bcfe84daba 100644
> >> --- a/meson.build
> >> +++ b/meson.build
> >> @@ -3288,7 +3288,14 @@ default_targets = 'CONFIG_DEFAULT_TARGETS' in
> >> config_host
> >> actual_target_dirs = []
> >> fdt_required = []
> >> foreach target : target_dirs
> >> - config_target = { 'TARGET_NAME': target.split('-')[0] }
> >> + target_name = target.split('-')[0]
> >> + config_target = { 'TARGET_NAME': target_name }
> >> + if target.endswith('linux-user')
> >> + config_target += { 'TARGET_ARCH_VARIANT': 'TARGET_ARCH' }
> >
> > If TARGET_ARCH_VARIANT or just called (TARGET_NAME_ENUM) is just a
> > direct mapping of target_name regardless linux-user/bsd-user/softmmu,
> > the logic would
> > be simpler. That would need add more enum to SYS_EMU_TARGET, the
> > current form is:
> > ##
> > # @SysEmuTarget:
> > # The comprehensive enumeration of QEMU system emulation
> > # ("softmmu") targets. Run "./configure --help" in the project
> > # root directory, and look for the \*-softmmu targets near the
> > # "--target-list" option. The individual target constants are not
> > # documented here, for the time being.
> > #
> > # @rx: since 5.0
> > #
> > # @avr: since 5.1
> > #
> > # @loongarch64: since 7.1
> > #
> > # @hexagon: since 11.0
> > #
> > # .. note:: The resulting QMP strings can be appended to the
> > # "qemu-system-" prefix to produce the corresponding QEMU
> > # executable name. This is true even for "qemu-system-x86_64".
> > #
> > # Since: 3.0
> > ##
> > { 'enum' : 'SysEmuTarget',
> > 'data' : [ 'aarch64', 'alpha', 'arm', 'avr', 'hexagon', 'hppa', 'i386',
> > 'loongarch64', 'm68k', 'microblaze', 'mips', 'mips64',
> > 'mips64el', 'mipsel', 'or1k', 'ppc',
> > 'ppc64', 'riscv32', 'riscv64', 'rx', 's390x', 'sh4',
> > 'sh4eb', 'sparc', 'sparc64', 'tricore',
> > 'x86_64', 'xtensa', 'xtensaeb' ] }
> >
> >
> > The list target names that missing from SysEmuTarget:
> >
> > Comparing the target-name prefixes of `configs/targets/*.mak` with the
> > `SysEmuTarget` enum in `qapi/machine.json`, the following seven names are
> > missing. All are Linux-user-only targets.
> >
> > | Target name | Config |
> > |---|---|
> > | `aarch64_be` | `configs/targets/aarch64_be-linux-user.mak` |
> > | `armeb` | `configs/targets/armeb-linux-user.mak` |
> > | `microblazeel` | `configs/targets/microblazeel-linux-user.mak` |
> > | `mipsn32` | `configs/targets/mipsn32-linux-user.mak` |
> > | `mipsn32el` | `configs/targets/mipsn32el-linux-user.mak` |
> > | `ppc64le` | `configs/targets/ppc64le-linux-user.mak` |
> > | `sparc32plus` | `configs/targets/sparc32plus-linux-user.mak` |
> >
> > All `*-softmmu` and `*-bsd-user` target names match enum entries.
> >
> > This comparison uses names derived from config filenames, not the
> > `TARGET_ARCH` values inside those configs.
> >
> > By doing this I think it would be more consistent.
> > If these target names are removed in future, then just remove it from
> > SysEmuTarget
> > along with. So SysEmuTarget is just a direct mapping of TARGET_NAME. Though
> > The comment of SysEmuTarget needs updated, it's not just about softmmu, but
> > also
> > for linux-user and bsd-user. But the fact is, with the current
> > changes. *-user also
> > used SysEmuTarget as an enum.
> >
> >
> > And TARGET_ARCH should also be exported by TargetInfo for other usage.
> > And it's enums is the subset
> > of enum SysEmuTarget
> >
>
> I tend to agree, but didn't find a real reason to do this change (yet).
> Also, I would be in favor to rename SysEmuTarget since in EmuTarget,
> since it would now be decoupled from system emulation only.
I think the current patch is enough at the current stage.
A better bsd-user/linux-user support needs much more work.
for example:
enum {
QEMU_ARCH_ALPHA = (1UL << SYS_EMU_TARGET_ALPHA),
QEMU_ARCH_ARM = (1UL << SYS_EMU_TARGET_ARM) |
(1UL << SYS_EMU_TARGET_AARCH64),
QEMU_ARCH_I386 = (1UL << SYS_EMU_TARGET_I386) |
(1UL << SYS_EMU_TARGET_X86_64),
QEMU_ARCH_M68K = (1UL << SYS_EMU_TARGET_M68K),
QEMU_ARCH_MICROBLAZE = (1UL << SYS_EMU_TARGET_MICROBLAZE),
QEMU_ARCH_MIPS = (1UL << SYS_EMU_TARGET_MIPS) |
(1UL << SYS_EMU_TARGET_MIPSEL) |
(1UL << SYS_EMU_TARGET_MIPS64) |
(1UL << SYS_EMU_TARGET_MIPS64EL),
QEMU_ARCH_PPC = (1UL << SYS_EMU_TARGET_PPC) |
(1UL << SYS_EMU_TARGET_PPC64),
QEMU_ARCH_S390X = (1UL << SYS_EMU_TARGET_S390X),
QEMU_ARCH_SH4 = (1UL << SYS_EMU_TARGET_SH4) |
(1UL << SYS_EMU_TARGET_SH4EB),
QEMU_ARCH_SPARC = (1UL << SYS_EMU_TARGET_SPARC) |
(1UL << SYS_EMU_TARGET_SPARC64),
QEMU_ARCH_XTENSA = (1UL << SYS_EMU_TARGET_XTENSA) |
(1UL << SYS_EMU_TARGET_XTENSAEB),
QEMU_ARCH_OR1K = (1UL << SYS_EMU_TARGET_OR1K),
QEMU_ARCH_TRICORE = (1UL << SYS_EMU_TARGET_TRICORE),
QEMU_ARCH_HPPA = (1UL << SYS_EMU_TARGET_HPPA),
QEMU_ARCH_RISCV = (1UL << SYS_EMU_TARGET_RISCV32) |
(1UL << SYS_EMU_TARGET_RISCV64),
QEMU_ARCH_RX = (1UL << SYS_EMU_TARGET_RX),
QEMU_ARCH_AVR = (1UL << SYS_EMU_TARGET_AVR),
QEMU_ARCH_HEXAGON = (1UL << SYS_EMU_TARGET_HEXAGON),
QEMU_ARCH_LOONGARCH = (1UL << SYS_EMU_TARGET_LOONGARCH64),
QEMU_ARCH_ALL = UINT32_MAX,
};
QEMU_BUILD_BUG_ON(SYS_EMU_TARGET__MAX > 32);
/**
* qemu_arch_available:
* @arch_bitmask: bitmask of QEMU_ARCH_* constants
*
* Return whether the current target architecture is contained in @arch_bitmask
*/
bool qemu_arch_available(uint32_t arch_bitmask);
arch_bitmask also needs to be increased to 64bit, and currently
SYS_EMU_TARGET__MAX is 29, very close to 32.
>
> @Daniel: Would you be favor to see such a change? I know that
> introspection/QAPI is related to system emulation, and I'm not
> completely sure adding user targets is the right move at this moment.
>
> >
> >
> >> + else
> >> + config_target += { 'TARGET_ARCH_VARIANT': target_name.to_upper() }
> >> + endif
> >> +
> >> if target.endswith('linux-user')
> >> if host_os != 'linux'
> >> if default_targets
> >> @@ -3372,7 +3379,7 @@ foreach target : target_dirs
> >> # do nothing
> >> elif ignored.contains(k)
> >> # do nothing
> >> - elif k == 'TARGET_ARCH'
> >> + elif k == 'TARGET_ARCH' or k == 'TARGET_ARCH_VARIANT'
> >> config_target_data.set(k, v.to_upper())
> >> elif k == 'TARGET_BASE_ARCH'
> >> # Note that TARGET_BASE_ARCH ends up in config-target.h but it is
> >> --
> >> 2.43.0
> >>
> >>
> >
> >
> > --
> > 此致
> > 礼
> > 罗勇刚
> > Yours
> > sincerely,
> > Yonggang Luo
>
--
此致
礼
罗勇刚
Yours
sincerely,
Yonggang Luo
