Re: [PATCH v2 2/2] linux-user/m68k: Handle EXCP_ADDRESS in cpu_loop

[Laurent Vivier]

Le 09/03/2023 à 21:16, Richard Henderson a écrit :
> This exception can be raised by illegal instructions.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   linux-user/m68k/cpu_loop.c     |  5 ++++-
>   tests/tcg/m68k/excp-address.c  | 32 ++++++++++++++++++++++++++++++++
>   tests/tcg/m68k/Makefile.target |  1 +
>   3 files changed, 37 insertions(+), 1 deletion(-)
>   create mode 100644 tests/tcg/m68k/excp-address.c
>
> diff --git a/linux-user/m68k/cpu_loop.c b/linux-user/m68k/cpu_loop.c
> index caead1cb74..84b5d290c0 100644
> --- a/linux-user/m68k/cpu_loop.c
> +++ b/linux-user/m68k/cpu_loop.c
> @@ -35,7 +35,10 @@ void cpu_loop(CPUM68KState *env)
>           cpu_exec_end(cs);
>           process_queued_cpu_work(cs);
>   
> -        switch(trapnr) {
> +        switch (trapnr) {
> +        case EXCP_ADDRESS:
> +            force_sig_fault(TARGET_SIGBUS, TARGET_BUS_ADRALN, env->pc);
> +            break;
>           case EXCP_ILLEGAL:
>           case EXCP_LINEA:
>           case EXCP_LINEF:
> diff --git a/tests/tcg/m68k/excp-address.c b/tests/tcg/m68k/excp-address.c
> new file mode 100644
> index 0000000000..1dbdddef58
> --- /dev/null
> +++ b/tests/tcg/m68k/excp-address.c
> @@ -0,0 +1,32 @@
> +/*
> + * Test m68k address exception
> + */
> +
> +#define _GNU_SOURCE 1
> +#include <signal.h>
> +#include <stdlib.h>
> +
> +static void sig_handler(int sig, siginfo_t *si, void *puc)
> +{
> +    exit(0);
> +}
> +
> +int main(int argc, char **argv)
> +{
> +    struct sigaction act = {
> +        .sa_sigaction = sig_handler,
> +        .sa_flags = SA_SIGINFO
> +    };
> +
> +    sigaction(SIGBUS, &act, NULL);
> +
> +    /*
> +     * addl %d0,#0 -- with immediate as destination is illegal.
> +     * Buggy qemu interpreted the insn as 5 words: 2 for immediate source
> +     * and another 2 for immediate destination.  Provide all that padding
> +     * so that abort gets called.
> +     */
> +    asm volatile(".word 0xd1bc,0,0,0,0");
> +
> +    abort();
> +}
> diff --git a/tests/tcg/m68k/Makefile.target b/tests/tcg/m68k/Makefile.target
> index 1163c7ef03..d3b0bc48a3 100644
> --- a/tests/tcg/m68k/Makefile.target
> +++ b/tests/tcg/m68k/Makefile.target
> @@ -4,6 +4,7 @@
>   #
>   
>   VPATH += $(SRC_PATH)/tests/tcg/m68k
> +TESTS += excp-address
>   TESTS += trap
>   
>   # On m68k Linux supports 4k and 8k pages (but 8k is currently broken)

Reviewed-by: Laurent Vivier <laurent@vivier.eu>

I've tested tests/tcg/m68k/excp-address.c on a real hardware (Q800), and the result differs from the 
one from QEMU:

On Q800 (etch m68k, kernel 5.14.0):

$ ./excp-address
Illegal instruction
$ strace ./excp-address
...
rt_sigaction(SIGBUS, {0x80000478, [], SA_SIGINFO}, NULL, 8) = 0
--- SIGILL (Illegal instruction) @ 0 (0) ---
+++ killed by SIGILL +++

With QEMU, we have:

# QEMU_STRACE= ./excp-address
...
677354 rt_sigaction(SIGBUS,0x40800454,NULL) = 0
--- SIGBUS {si_signo=SIGBUS, si_code=1, si_addr=0x800004ce} ---
677354 exit_group(0)

Thanks,
Laurent