Re: [PATCH 2/5] clocksource: add support for entropy-generation function

[Jarod Wilson <jarod@redhat.com>]

Thomas Gleixner wrote:
> On Mon, 13 Jun 2011, Jarod Wilson wrote:
>> Add a new function pointer to struct clocksource that can optionally be
>> filled in by clocksources deemed to be high enough resolution to feed
>> the random number generator entropy pool.
>
> Uurrg.
>
>> + * @entropy:		random entropy pool addition function (optional, and
>> + *			requires a fairly high-resolution clocksource)
>
> Why do you want to do that ? Looking at the implementations of TSC and
> HPET it's the same code. We really do not want to add that all over
> the place. We can make that a property flag and the entropy function
> common to the core code.
>
>> +/**
>> + * Do we have at least one clocksource that can generate entropy?
>> + */
>> +bool clocksource_entropy_available(void)
>> +{
>> +	struct clocksource *src;
>> +	bool entropy_possible = false;
>> +
>> +	mutex_lock(&clocksource_mutex);
>> +	list_for_each_entry(src,&clocksource_list, list) {
>> +		if (src->entropy) {
>> +			entropy_possible = true;
>> +			break;
>> +		}
>> +	}
>> +	mutex_unlock(&clocksource_mutex);
>> +
>> +	return entropy_possible;
>> +}
>> +EXPORT_SYMBOL(clocksource_entropy_available);
>
> That should be evaluated when clocksources are registered not at some
> random point in time, which might return total nonsense as it's not
> guaranteed that the clocksource which has the entropy property is
> going to end up as the current clocksource.
>
>> +/**
>> + * Call the clocksource's entropy-generation function, if set
>> + */
>> +void clocksource_add_entropy(void)
>> +{
>> +	if (!curr_clocksource->entropy)
>> +		return;
>
> Why restricting it to the current clocksource? We can use the best
> registered one for this which has the entropy property set.

Yeah, John had some similar suggestions, and locally, I've got a 
modified version that instead of adding entropy functions for each 
clocksource, adds an entropy rating, then the highest rated entropy 
clocksource gets used, but a common clocksource entropy function that 
calls the chosen clocksource's read function. The entropy clocksource 
function gets picked by way of another function similar to 
clocksource_select, called in all the same places.

However, since none of this is going to be viable until the random code 
is significantly restructured, I haven't bothered with posting any of 
the updates I've made. I can toss the delta out there if anyone really 
wants to take a look at it, but otherwise, I'll just sit on it until 
said restructuring happens.

-- 
Jarod Wilson
jarod@redhat.com