Re: [Qemu-ppc] [Qemu-devel] [PATCH 2/6] target-ppc: Implement darn instruction

[Thomas Huth]

On 12.08.2016 09:39, Nikunj A Dadhania wrote:
> Thomas Huth <address@hidden> writes:
> 
>> On 12.08.2016 08:43, Nikunj A Dadhania wrote:
>>> David Gibson <address@hidden> writes:
>>>
>>>> [ Unknown signature status ]
>>>> On Tue, Aug 09, 2016 at 02:47:46PM +0530, Nikunj A Dadhania wrote:
>>>>> Nikunj A Dadhania <address@hidden> writes:
>>>>>
>>>>>> David Gibson <address@hidden> writes:
>>>>>>
>>>>>>> [ Unknown signature status ]
>>>>>>> On Mon, Aug 08, 2016 at 07:33:37AM +1000, Benjamin Herrenschmidt wrote:
>>>>>>>> On Sun, 2016-08-07 at 23:06 +0530, Nikunj A Dadhania wrote:
>>>>>>>>> +target_ulong helper_darn(uint32_t l)
>>>>>>>>> +{
>>>>>>>>> +    target_ulong r = UINT64_MAX;
>>>>>>>>> +
>>>>>>>>> +    if (l <= 2) {
>>>>>>>>> +        do {
>>>>>>>>> +            r = random() * random();
>>>>>>>>> +            r &= l ? UINT64_MAX : UINT32_MAX;
>>>>>>>>> +        } while (r == UINT64_MAX);
>>>>>>>>> +    }
>>>>>>>>> +
>>>>>>>>> +    return r;
>>>>>>>>> +}
>>>>>>>>>  #endif
>>>>>>>>
>>>>>>>> Isn't this a bit week ? Look at the implementation of H_RANDOM...
>>>>>>>
>>>>>>> Indeed, you should be using the rng backend that H_RANDOM, virtio-rng
>>>>>>> and the Intel random number instruction all use.
>>>>>
>>>>> Can you point me to the intel instruction, I couldn't get rdrand
>>>>> implementation.
>>>>
>>>> Ah.. turns out no.  I'd assumed it was there and used the same backend
>>>> as virtio-rng and H_RANDOM, but I hadn't actually looked at the code,
>>>> and now that I'm trying I can't see it either.
>>>>
>>>>>
>>>>>> I was looking at implementing this, AFAIU, I have to get a new RNG
>>>>>> object in the initialization routine. We would need an instance of this
>>>>>> per machine. So for pseries I can add in ppc_spapr_init(). I am not sure
>>>>>> in case of linux-user where should this be initialized.
>>>>>>
>>>>>> One other place was init_proc_POWER9(), but that will be per cpu and
>>>>>> member of CPUPPCState structure. Advantage is it will work for system
>>>>>> emulation and linux-user both and we would not need a lock.
>>>>>
>>>>> More issues here. Random backend is not compiled for linux-user, adding
>>>>> that wasn't difficult, but then rng_backend_request_entropy() uses
>>>>> qemu_set_fd_handler() which is a stub for linux-user
>>>>> (stubs/set-fd-handler.c)7
>>>>
>>>>
>>>> Ah.. yeah, not sure how we'll need to handle that.
>>>
>>> I have sent updated patch, reading from /dev/random. Not sure if that is
>>> allowed in tcg. Works fine though.
>>
>> You can not rely on /dev/random for this job, since it might block. So
>> your guest would stop executing when there is not enough random data
>> available in the host, and I think that's quite a bad behavior...
> 
> Hmm.. rng-random does use this, but it might have way to time out probably:
> 
> backends/rng-random.c:    s->filename = g_strdup("/dev/random");

This is only the default value, which is likely suitable for virtio-rng,
but not for something like this instruction (or the H_RANDOM hypercall).
You can set the filename property to another file when instantiating it,
like:

 qemu-system-xxx ... -object rng-random,filename=/dev/hwrng,id=rng0 ...

That's the whole point these backends - you give the users the
possibility to chose the right source of entropy.

For example, on the POWER8 machine that I have here, /dev/random blocks
after a couple of bytes, you can see that easily by typing something
like "hexdump /dev/random". It only delivers more random data when I run
"rngd -r /dev/hwrng" in the background.

 Thomas