| Message ID | 20210813131809.28655-4-peter.maydell@linaro.org |
|---|---|
| State | Superseded |
| Headers | show |
| Series | linux-user: Clean up siginfo_t handling for arm, aarch64 | expand |
On 8/13/21 3:18 AM, Peter Maydell wrote: > In the Arm target code, when the fpa11 emulation code tells us we > need to send the guest a SIGFPE, we do this with queue_signal(), but > we are using the wrong si_type, and we aren't setting the _sifields > union members corresponding to either the si_type we are using or the > si_type we should be using. > > As the existing comment notes, the kernel code for this calls the old > send_sig() function to deliver the signal. This eventually results > in the kernel's signal handling code fabricating a siginfo_t with a > SI_KERNEL code and a zero pid and uid. For QEMU this means we need > to use QEMU_SI_KILL. We already have a function for that: > force_sig() sets up the whole target_siginfo_t the way we need it. > > Signed-off-by: Peter Maydell<peter.maydell@linaro.org> > --- > linux-user/arm/cpu_loop.c | 11 ++++------- > 1 file changed, 4 insertions(+), 7 deletions(-) Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
diff --git a/linux-user/arm/cpu_loop.c b/linux-user/arm/cpu_loop.c index 007752f5b74..44324976196 100644 --- a/linux-user/arm/cpu_loop.c +++ b/linux-user/arm/cpu_loop.c @@ -266,16 +266,13 @@ static bool emulate_arm_fpa11(CPUARMState *env, uint32_t opcode) ts->fpa.fpsr |= raise & ~enabled; if (raise & enabled) { - target_siginfo_t info = { }; - /* * The kernel's nwfpe emulator does not pass a real si_code. - * It merely uses send_sig(SIGFPE, current, 1). + * It merely uses send_sig(SIGFPE, current, 1), which results in + * __send_signal() filling out SI_KERNEL with pid and uid 0 (under + * the "SEND_SIG_PRIV" case). That's what our force_sig() does. */ - info.si_signo = TARGET_SIGFPE; - info.si_code = TARGET_SI_KERNEL; - - queue_signal(env, info.si_signo, QEMU_SI_FAULT, &info); + force_sig(TARGET_SIGFPE); } else { env->regs[15] += 4; }
In the Arm target code, when the fpa11 emulation code tells us we need to send the guest a SIGFPE, we do this with queue_signal(), but we are using the wrong si_type, and we aren't setting the _sifields union members corresponding to either the si_type we are using or the si_type we should be using. As the existing comment notes, the kernel code for this calls the old send_sig() function to deliver the signal. This eventually results in the kernel's signal handling code fabricating a siginfo_t with a SI_KERNEL code and a zero pid and uid. For QEMU this means we need to use QEMU_SI_KILL. We already have a function for that: force_sig() sets up the whole target_siginfo_t the way we need it. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- linux-user/arm/cpu_loop.c | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) -- 2.20.1