target/arm: Set FPSCR.LTPSIZE for user-mode

[PATCH 0/3] target/arm: Set FPSCR.LTPSIZE for user-mode

Posted by Peter Maydell 2 years, 7 months ago

This patchset fixes https://gitlab.com/qemu-project/qemu/-/issues/613
which is a bug where we weren't setting FPSCR.LTPSIZE correctly
out of reset for the user-mode emulator. The effect is that
when using an M-profile CPU with the low-overhead-branch or MVE
extensions (ie the Cortex-M55) with the linux-user QEMU the 'LE'
instruction would take a UserFault and MVE instructions would
be incorrectly predicated.

This is the result of some over-exuberant ifdeffery in the
arm_cpu_reset() function. Patch 1 fixes that so that most of
the M-profile-specific reset handling is not ifdeffed, and
when we're in user mode we specifically set the FPU state up
cleanly. Patches 2 and 3 are just follow-on tidyup.

Christophe, if you are in a position to test this series with:
 M55 (has Security, MVE and LOB)
 M33 (has Security extension but not MVE/LOB)
 M7 or M4 (no Security, but does have FPU)
that would be ideal. I don't really have much in the way of
test cases for usermode to hand, so it's possible that I
forgot something in the init of the FPU state that might break
one of those combinations. (It does fix the test case attached
to the bug report.)

thanks
-- PMM

Peter Maydell (3):
  target/arm: Don't skip M-profile reset entirely in user mode
  target/arm: Always clear exclusive monitor on reset
  target/arm: Consolidate ifdef blocks in reset

 target/arm/cpu.c | 39 ++++++++++++++++++++++++++++-----------
 1 file changed, 28 insertions(+), 11 deletions(-)

-- 
2.20.1

Re: [PATCH 0/3] target/arm: Set FPSCR.LTPSIZE for user-mode

Posted by Richard Henderson 2 years, 7 months ago

On 9/14/21 5:07 AM, Peter Maydell wrote:
> Peter Maydell (3):
>    target/arm: Don't skip M-profile reset entirely in user mode
>    target/arm: Always clear exclusive monitor on reset
>    target/arm: Consolidate ifdef blocks in reset

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>


r~

Re: [PATCH 0/3] target/arm: Set FPSCR.LTPSIZE for user-mode

Posted by Christophe Lyon 2 years, 7 months ago

Hi Peter,

Thanks for the prompt patches!

On Tue, Sep 14, 2021 at 2:08 PM Peter Maydell <peter.maydell@linaro.org>
wrote:

> This patchset fixes https://gitlab.com/qemu-project/qemu/-/issues/613
> which is a bug where we weren't setting FPSCR.LTPSIZE correctly
> out of reset for the user-mode emulator. The effect is that
> when using an M-profile CPU with the low-overhead-branch or MVE
> extensions (ie the Cortex-M55) with the linux-user QEMU the 'LE'
> instruction would take a UserFault and MVE instructions would
> be incorrectly predicated.
>
> This is the result of some over-exuberant ifdeffery in the
> arm_cpu_reset() function. Patch 1 fixes that so that most of
> the M-profile-specific reset handling is not ifdeffed, and
> when we're in user mode we specifically set the FPU state up
> cleanly. Patches 2 and 3 are just follow-on tidyup.
>
> Christophe, if you are in a position to test this series with:
>  M55 (has Security, MVE and LOB)
>  M33 (has Security extension but not MVE/LOB)
>  M7 or M4 (no Security, but does have FPU)
> that would be ideal. I don't really have much in the way of
> test cases for usermode to hand, so it's possible that I
> forgot something in the init of the FPU state that might break
> one of those combinations. (It does fix the test case attached
> to the bug report.)
>

I ran the GCC testsuite with these configurations, and found no regression
compared to qemu-6.0, thanks.

In the M55 case, this also enabled many more tests, which is great!

Christophe


> thanks
> -- PMM
>
> Peter Maydell (3):
>   target/arm: Don't skip M-profile reset entirely in user mode
>   target/arm: Always clear exclusive monitor on reset
>   target/arm: Consolidate ifdef blocks in reset
>
>  target/arm/cpu.c | 39 ++++++++++++++++++++++++++++-----------
>  1 file changed, 28 insertions(+), 11 deletions(-)
>
> --
> 2.20.1
>
>
>