Re: [Qemu-devel] [PATCH pic32 2/7] Stop simulation when processor is suspended forever by WAIT instruction with interrupts disabled.

[Serge Vakulenko]

Hi Peter and Leon,

With a bit of thinking, I agree, that the question of session
termination on WAIT instruction is quite complicated in case of
multi-core system, background i/o activity, mipsR6 core etc. So I'm
going to find another solution for the task. What I essentially want
here is to stop the simulator when the target Unix system is halted,
like:

$ /usr/local/qemu-mips/bin/qemu-system-mipsel -M pic32mx7-max32
-nographic -monitor none -serial stdio -bios boot-max32.hex -kernel
unix.hex -sd sdcard.img
Board: chipKIT Max32
Processor: M4K
RAM size: 128 kbytes
Load file: 'boot-max32.hex', 6720 bytes
Load file: 'unix.hex', 144992 bytes
Card0 image 'sdcard.img', 102888 kbytes
[...]
2.11 BSD UNIX (pic32) (console)

login: root
Password:
Welcome to RetroBSD!
erase, kill ^U, intr ^C
# halt
killing processes... done
syncing disks... done
halted
$ _                           <-- QEMU terminated

On BSD, the halt command uses reboot(RB_HALT) system call to terminate
the operating system. It essentially results in an endless loop on
wait instruction with interrupts disabled., like "for(;;) {
asm("wait"); }". For pic32 it makes little sense to continue
simulation in this case.

Fortunately, I've found a solution which does not require modification
of generic code. Everything can be done in the platform-specific part.

Thanks,
--Serge


2015-06-30 11:08 GMT-07:00 Peter Crosthwaite <address@hidden>:
> On Tue, Jun 30, 2015 at 7:10 AM, Leon Alrae <address@hidden> wrote:
>> Hi Serge,
>>
>> On 30/06/2015 06:02, Serge Vakulenko wrote:
>>> Signed-off-by: Serge Vakulenko <address@hidden>
>>> ---
>>>  target-mips/op_helper.c | 7 +++++++
>>>  1 file changed, 7 insertions(+)
>>>
>>> diff --git a/target-mips/op_helper.c b/target-mips/op_helper.c
>>> index 2a9ddff..1b7caeb 100644
>>> --- a/target-mips/op_helper.c
>>> +++ b/target-mips/op_helper.c
>>> @@ -22,6 +22,7 @@
>>>  #include "exec/helper-proto.h"
>>>  #include "exec/cpu_ldst.h"
>>>  #include "sysemu/kvm.h"
>>> +#include "sysemu/sysemu.h"
>>>
>>>  #ifndef CONFIG_USER_ONLY
>>>  static inline void cpu_mips_tlb_flush (CPUMIPSState *env, int 
>>> flush_global);
>>> @@ -2235,6 +2236,12 @@ void helper_wait(CPUMIPSState *env)
>>>  {
>>>      CPUState *cs = CPU(mips_env_get_cpu(env));
>>>
>>> +#ifndef CONFIG_USER_ONLY
>>> +    if (!(env->CP0_Status & (1 << CP0St_IE))) {
>>> +        /* WAIT instruction with interrupts disabled - halt the 
>>> simulation. */
>>> +        qemu_system_shutdown_request();
>>> +    }
>>> +#endif
>>
>> Why do you want to stop the simulation, wouldn't it be beneficial to leave it
>> running? For debugging for example, the user would be still able to inspect
>> the state, that could help to find a clue why CPU got suspended forever.
>>
>
> In theory, there may also be other actors in the system. Even if the
> CPU is in a never-going-to-come-back-to-life state, other-things may
> still be happening. This would be best as some sort of core feature to
> detect total-inaction of the system. That is, there are no delayed
> events pending (timed events and AIOs) and the CPUs are halted. Then
> this would be a user option to stop the system on reaching the
> machine-inactive state and it would be applicable beyond mips.
>
> Regards,
> Peter
>
>> Also, if we take into account implementations (currently not supported in
>> QEMU) where CPU can be woken up by a disabled interrupt (the Config7.WII bit
>> in P5600 for example), then this won't be correct.
>>
>> Thanks,
>> Leon
>>
>>