Re: [Qemu-arm] [Qemu-devel] [PATCH 12/13] target/arm/arm-semi: Implement SH_EXT_STDOUT_STDERR extension

[Alex Bennée]

Peter Maydell <address@hidden> writes:

> SH_EXT_STDOUT_STDERR is a v2.0 semihosting extension: the guest
> can open ":tt" with a file mode requesting append access in
> order to open stderr, in addition to the existing "open for
> read for stdin or write for stdout". Implement this and
> report it via the :semihosting-features data.
>
> Signed-off-by: Peter Maydell <address@hidden>
> ---
>  target/arm/arm-semi.c | 22 ++++++++++++++++++++--
>  1 file changed, 20 insertions(+), 2 deletions(-)
>
> diff --git a/target/arm/arm-semi.c b/target/arm/arm-semi.c
> index 531084b7799..0df8d4d69d6 100644
> --- a/target/arm/arm-semi.c
> +++ b/target/arm/arm-semi.c
> @@ -476,12 +476,16 @@ static uint32_t gdb_flenfn(TaskState *ts, ARMCPU *cpu, 
> GuestFD *gf)
>  #define SHFB_MAGIC_2 0x46
>  #define SHFB_MAGIC_3 0x42
>
> +/* Feature bits reportable in feature byte 0 */
> +#define SH_EXT_EXIT_EXTENDED (1 << 0)

If you swap 12/13 this could be kept with the related feature. I don't
think one implies the other right?

> +#define SH_EXT_STDOUT_STDERR (1 << 1)
> +
>  static const uint8_t featurefile_data[] = {
>      SHFB_MAGIC_0,
>      SHFB_MAGIC_1,
>      SHFB_MAGIC_2,
>      SHFB_MAGIC_3,
> -    0, /* Feature byte 0 */
> +    SH_EXT_STDOUT_STDERR, /* Feature byte 0 */
>  };
>
>  static void init_featurefile_guestfd(int guestfd)
> @@ -674,7 +678,21 @@ target_ulong do_arm_semihosting(CPUARMState *env)
>          }
>
>          if (strcmp(s, ":tt") == 0) {
> -            int result_fileno = arg1 < 4 ? STDIN_FILENO : STDOUT_FILENO;
> +            int result_fileno;
> +
> +            /*
> +             * We implement SH_EXT_STDOUT_STDERR, so:
> +             *  open for read == stdin
> +             *  open for write == stdout
> +             *  open for append == stderr
> +             */

I love the way the spec documents field2 as an ISO C fopen() mode and
then an extension literally subverts the meaning to be something else.
Where the designers worried about adding a SYS_OPEN_TTY function to the
interface?

Anyway it meets the spec however weird it might be:

Reviewed-by: Alex Bennée <address@hidden>

> +            if (arg1 < 4) {
> +                result_fileno = STDIN_FILENO;
> +            } else if (arg1 < 8) {
> +                result_fileno = STDOUT_FILENO;
> +            } else {
> +                result_fileno = STDERR_FILENO;
> +            }
>              associate_guestfd(guestfd, result_fileno);
>              unlock_user(s, arg0, 0);
>              return guestfd;


--
Alex Bennée