On Sun, 9 Jan 2022 at 16:36, Warner Losh <imp@bsdimp.com> wrote:
>
> setup_frame sets up a signalled stack frame. Associated routines to
> extract the pointer to the stack frame and to support alternate stacks.
>
> Signed-off-by: Stacey Son <sson@FreeBSD.org>
> Signed-off-by: Kyle Evans <kevans@freebsd.org>
> Signed-off-by: Warner Losh <imp@bsdimp.com>
> ---
>  bsd-user/signal.c | 166 ++++++++++++++++++++++++++++++++++++++++------
>  1 file changed, 144 insertions(+), 22 deletions(-)
>
> diff --git a/bsd-user/signal.c b/bsd-user/signal.c
> index 8dadc9a39a7..8e1427553da 100644
> --- a/bsd-user/signal.c
> +++ b/bsd-user/signal.c
> @@ -30,11 +30,27 @@
>   * fork.
>   */
>
> +static target_stack_t target_sigaltstack_used = {
> +    .ss_sp = 0,
> +    .ss_size = 0,
> +    .ss_flags = TARGET_SS_DISABLE,
> +};

sigaltstacks are per-thread, so this needs to be in the TaskState,
not global. (We fixed this in linux-user in commit 5bfce0b74fbd5d5
in 2019: the change is relatively small.)

> +
>  static struct target_sigaction sigact_table[TARGET_NSIG];
>  static void host_signal_handler(int host_sig, siginfo_t *info, void *puc);
>  static void target_to_host_sigset_internal(sigset_t *d,
>          const target_sigset_t *s);
>
> +static inline int on_sig_stack(unsigned long sp)
> +{
> +    return sp - target_sigaltstack_used.ss_sp < target_sigaltstack_used.ss_size;
> +}
> +
> +static inline int sas_ss_flags(unsigned long sp)
> +{
> +    return target_sigaltstack_used.ss_size == 0 ? SS_DISABLE : on_sig_stack(sp)
> +        ? SS_ONSTACK : 0;
> +}
>
>  int host_to_target_signal(int sig)
>  {
> @@ -336,28 +352,6 @@ void queue_signal(CPUArchState *env, int sig, target_siginfo_t *info)
>      return;
>  }
>
> -static int fatal_signal(int sig)
> -{
> -
> -    switch (sig) {
> -    case TARGET_SIGCHLD:
> -    case TARGET_SIGURG:
> -    case TARGET_SIGWINCH:
> -    case TARGET_SIGINFO:
> -        /* Ignored by default. */
> -        return 0;
> -    case TARGET_SIGCONT:
> -    case TARGET_SIGSTOP:
> -    case TARGET_SIGTSTP:
> -    case TARGET_SIGTTIN:
> -    case TARGET_SIGTTOU:
> -        /* Job control signals.  */
> -        return 0;
> -    default:
> -        return 1;
> -    }
> -}

There wasn't any need to move this function, I think ?

> -
>  /*
>   * Force a synchronously taken QEMU_SI_FAULT signal. For QEMU the
>   * 'force' part is handled in process_pending_signals().
> @@ -484,6 +478,134 @@ static void host_signal_handler(int host_sig, siginfo_t *info, void *puc)
>      cpu_exit(thread_cpu);
>  }
>
> +static int fatal_signal(int sig)
> +{
> +
> +    switch (sig) {
> +    case TARGET_SIGCHLD:
> +    case TARGET_SIGURG:
> +    case TARGET_SIGWINCH:
> +    case TARGET_SIGINFO:
> +        /* Ignored by default. */
> +        return 0;
> +    case TARGET_SIGCONT:
> +    case TARGET_SIGSTOP:
> +    case TARGET_SIGTSTP:
> +    case TARGET_SIGTTIN:
> +    case TARGET_SIGTTOU:
> +        /* Job control signals.  */
> +        return 0;
> +    default:
> +        return 1;
> +    }
> +}
> +
> +static inline abi_ulong get_sigframe(struct target_sigaction *ka,
> +        CPUArchState *regs, size_t frame_size)
> +{
> +    abi_ulong sp;
> +
> +    /* Use default user stack */
> +    sp = get_sp_from_cpustate(regs);
> +
> +    if ((ka->sa_flags & TARGET_SA_ONSTACK) && (sas_ss_flags(sp) == 0)) {
> +        sp = target_sigaltstack_used.ss_sp +
> +            target_sigaltstack_used.ss_size;
> +    }
> +
> +#if defined(TARGET_MIPS) || defined(TARGET_ARM)
> +    return (sp - frame_size) & ~7;
> +#elif defined(TARGET_AARCH64)
> +    return (sp - frame_size) & ~15;
> +#else
> +    return sp - frame_size;
> +#endif

We don't need to do it in this patchseries, but you should strongly
consider pulling the architecture-specifics out in a way that
avoids this kind of ifdef ladder.

> +}
> +
> +/* compare to mips/mips/pm_machdep.c and sparc64/sparc64/machdep.c sendsig() */

> +static void setup_frame(int sig, int code, struct target_sigaction *ka,
> +    target_sigset_t *set, target_siginfo_t *tinfo, CPUArchState *regs)
> +{
> +    struct target_sigframe *frame;
> +    abi_ulong frame_addr;
> +    int i;
> +
> +    frame_addr = get_sigframe(ka, regs, sizeof(*frame));
> +    trace_user_setup_frame(regs, frame_addr);
> +    if (!lock_user_struct(VERIFY_WRITE, frame, frame_addr, 0)) {
> +        goto give_sigsegv;

FreeBSD for Arm (haven't checked other BSDs or other archs)
gives a SIGILL for the "can't write signal frame to stack"
case, I think:
https://github.com/freebsd/freebsd-src/blob/main/sys/arm/arm/exec_machdep.c#L316
I don't understand why they picked SIGILL, SIGSEGV seems much more
logical to me, but we should follow the kernel behaviour.

> +    }
> +
> +    memset(frame, 0, sizeof(*frame));
> +#if defined(TARGET_MIPS)
> +    int mflags = on_sig_stack(frame_addr) ? TARGET_MC_ADD_MAGIC :
> +        TARGET_MC_SET_ONSTACK | TARGET_MC_ADD_MAGIC;
> +#else
> +    int mflags = 0;
> +#endif
> +    if (get_mcontext(regs, &frame->sf_uc.uc_mcontext, mflags)) {
> +        goto give_sigsegv;

The FreeBSD kernel get_mcontext() can't fail -- why can ours ?
(This matters because SIGSEGV may not be the right response to
whatever the failure case is.)

> +    }
> +
> +    for (i = 0; i < TARGET_NSIG_WORDS; i++) {
> +        if (__put_user(set->__bits[i], &frame->sf_uc.uc_sigmask.__bits[i])) {
> +            goto give_sigsegv;

__get_user() and __put_user() in QEMU can't fail, so you don't
need to check for errors here, unlike the non-double-underscore
versions. At some point you might want to take the current linux-user
versions of these user-access functions/macros: it looks like bsd-user
has an older version which doesn't handle the case where the guest
has looser alignment restrictions than the host. The new ones
actually expand to do { ... } while(0) statements which won't
be valid inside an if() condition.

(Historical note: the reason QEMU's __put_user/__get_user ever had
return values at all is that the linux-user code was copy-and-pasted
from the Linux kernel. In the Linux kernel handling of writing
data to userspace is/was error-checked on every write, whereas
QEMU does the "is this writable" test once with the lock_user
function and then can assume all the writes to that area succeed.
But we still started with a lot of copy-pasted code that was doing
unnecessary checks on __put_user and __get_user return values.
FreeBSD seems to handle write-checking in yet a third way, by
assembling the struct in kernel-space and checking for writability
once at the end when it copies the whole block out to userspace.)

> +        }
> +    }
> +
> +    if (tinfo) {
> +        frame->sf_si.si_signo = tinfo->si_signo;
> +        frame->sf_si.si_errno = tinfo->si_errno;
> +        frame->sf_si.si_code = tinfo->si_code;
> +        frame->sf_si.si_pid = tinfo->si_pid;
> +        frame->sf_si.si_uid = tinfo->si_uid;
> +        frame->sf_si.si_status = tinfo->si_status;
> +        frame->sf_si.si_addr = tinfo->si_addr;
> +
> +        if (TARGET_SIGILL == sig || TARGET_SIGFPE == sig ||
> +                TARGET_SIGSEGV == sig || TARGET_SIGBUS == sig ||
> +                TARGET_SIGTRAP == sig) {
> +            frame->sf_si._reason._fault._trapno = tinfo->_reason._fault._trapno;
> +        }
> +
> +        /*
> +         * If si_code is one of SI_QUEUE, SI_TIMER, SI_ASYNCIO, or
> +         * SI_MESGQ, then si_value contains the application-specified
> +         * signal value. Otherwise, the contents of si_value are
> +         * undefined.
> +         */
> +        if (SI_QUEUE == code || SI_TIMER == code || SI_ASYNCIO == code ||
> +                SI_MESGQ == code) {
> +            frame->sf_si.si_value.sival_int = tinfo->si_value.sival_int;
> +        }
> +
> +        if (SI_TIMER == code) {
> +            frame->sf_si._reason._timer._timerid =
> +                tinfo->_reason._timer._timerid;
> +            frame->sf_si._reason._timer._overrun =
> +                tinfo->_reason._timer._overrun;
> +        }
> +
> +#ifdef SIGPOLL
> +        if (SIGPOLL == sig) {
> +            frame->sf_si._reason._band = tinfo->_reason._band;
> +        }
> +#endif

This seems to be yet a third set of the logic for handling
target_siginfo_t's union, to go along with tswap_siginfo() and
host_to_target_siginfo_noswap(), except that the logic here is
different. linux-user calls tswap_siginfo() in its signal-frame
setup code.

> +
> +    }
> +
> +    if (set_sigtramp_args(regs, sig, frame, frame_addr, ka)) {
> +        goto give_sigsegv;
> +    }

set_sigtramp_args() can't fail. (Not sure why it has a non-void
return type.)

> +
> +    unlock_user_struct(frame, frame_addr, 1);
> +    return;
> +
> +give_sigsegv:
> +    unlock_user_struct(frame, frame_addr, 1);
> +    force_sig(TARGET_SIGSEGV);
> +}
> +
>  void signal_init(void)
>  {
>      TaskState *ts = (TaskState *)thread_cpu->opaque;
> --
> 2.33.1

thanks