On Wed, 28 Aug 2024 at 16:51, Peter Maydell <peter.mayd...@linaro.org> wrote: > > On Wed, 28 Aug 2024 at 08:22, Michael Tokarev <m...@tls.msk.ru> wrote: > > > > 05.03.2024 16:52, Peter Maydell wrote: > > > From: Richard Henderson <richard.hender...@linaro.org> > > > > > > If translation is disabled, the default memory type is Device, which > > > requires alignment checking. This is more optimally done early via > > > the MemOp given to the TCG memory operation. > > > > > > Reviewed-by: Philippe Mathieu-Daudé <phi...@linaro.org> > > > Reported-by: Idan Horowitz <idan.horow...@gmail.com> > > > Signed-off-by: Richard Henderson <richard.hender...@linaro.org> > > > Message-id: 20240301204110.656742-6-richard.hender...@linaro.org > > > Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1204 > > > Signed-off-by: Richard Henderson <richard.hender...@linaro.org> > > > Signed-off-by: Peter Maydell <peter.mayd...@linaro.org> > > > > Hi! > > > > Apparently this change also breaks picolibc testsuite (between > > 8.2 and 9.0, bisect points to this commit). > > > > For example: > > > > ./qemu-system-arm \ > > -m 1G \ > > -chardev stdio,mux=on,id=stdio0 \ > > -semihosting-config enable=on,chardev=stdio0,arg=program-name \ > > -monitor none \ > > -serial none \ > > -machine none,accel=tcg \ > > -cpu cortex-a8 \ > > -device > > loader,file=/tmp/picolibc-1.8.6/arm-none-eabi/test/printf_scanf_thumb_v7_fp_softfp,cpu-num=0 > > \ > > -nographic > > > > (yes, this testsuite uses qemu-system as a substitute of > > qemu-user, sort of, (ab)using -device loader) > > My immediate guess is that this code won't run on real hardware > either -- i.e. that is bare-metal code that was only ever tested > and run on QEMU and was previously relying on the incorrect > behaviour that we didn't enforce the alignment checks that we're > supposed to do when the MMU is off.
I had a look at the test binary you kindly provided in https://gitlab.com/qemu-project/qemu/-/issues/2542 and that confirmed my guess. This binary would never have worked on real hardware, and it only worked on older QEMU because we weren't correctly emulating this corner of the architecture. You need to either use a new enough picolibc that you can turn on its _PICOCRT_ENABLE_MMU option, or else make sure everything is built with gcc's -mstrict-align or similar option to avoid any unaligned loads. thanks -- PMM
