Re: [RFC PATCH] target/mips: Allow building without Inter-Thread Communication hardware

[Philippe Mathieu-Daudé]

On Wed, Apr 28, 2021 at 12:13 AM Richard Henderson
<richard.henderson@linaro.org> wrote:
> On 4/27/21 12:11 PM, Philippe Mathieu-Daudé wrote:
> > The Inter-Thread Communication unit (TYPE_MIPS_ITU) is an optional
> > device that is only selected by a few machines. However it goes
> > deep into the translation code, as the MTC0/MTHC0 SAAR helpers
> > call itc_reconfigure().
> >
> > When building with no machine selecting the ITU component (which
> > is implemented in hw/misc/mips_itu.c), we get the following link
> > failure:
> >
> >    /usr/bin/ld: target_mips_cp0_helper.c.o: in function `helper_mtc0_saar':
> >    target/mips/cp0_helper.c:1118: undefined reference to `itc_reconfigure'
> >    /usr/bin/ld: target_mips_cp0_helper.c.o: in function `helper_mthc0_saar':
> >    target/mips/cp0_helper.c:1135: undefined reference to `itc_reconfigure'
> >
> > Fix by adding a stub, built when the ITU isn't selected.
> >
> > Signed-off-by: Philippe Mathieu-Daudé<f4bug@amsat.org>
> > ---
> > RFC because too much Meson machinery to my taste.
> > But how to deal with such architectural devices else?
> >
> > To reproduce:
> >
> > $ echo CONFIG_JAZZ=y > default-configs/devices/mips64el-softmmu.mak
> > $ echo CONFIG_SEMIHOSTING=y >> default-configs/devices/mips64el-softmmu.mak
> > $ configure --without-default-devices
> > $ ninja qemu-system-mips64el
> > $ ./qemu-system-mips64el -M magnum -S
> > ---
> >   target/mips/cp0_itu-stub.c | 15 +++++++++++++++
> >   target/mips/meson.build    |  3 +++
> >   2 files changed, 18 insertions(+)
> >   create mode 100644 target/mips/cp0_itu-stub.c
>
> Perhaps use __attribute__((weak)) on itc_reconfigure?  Then you don't need the
> stub at all.  You're already protecting the actual call, so there should be no
> change needed there.
>
> We're not using weak so far, but as far as I can tell this is supported by gcc
> on windows as well.

Apparently we are:

$ git grep attribute.*weak
softmmu/memory.c:3286:void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
tests/qtest/libqtest-single.h:16:QTestState *global_qtest
__attribute__((common, weak));

I think we should use either stubs or attribute weak, not both.