[DTrace-devel] [PATCH] Implementation of the printa() action

[Eugene Loh]

On 12/08/2020 10:58 PM, Kris Van Hees wrote:

> On Tue, Dec 08, 2020 at 05:09:02PM -0800, Eugene Loh wrote:
>> 3)  The aggregations have a per-CPU latch sequence number.  I thought
>> the consumer should use this number in two ways:
>>       > to check which aggregation copy to use
>>       > to confirm that the copy that was used was not written during the
>> read
>> I do not see where the consumer employs the sequence number.  I imagine
>> user-visible behavior could be quite tolerant of this omission, but it
>> is "advertised" behavior.  So if its implementation is being deferred,
>> we should at least note that.
> The consumer currently does not need to use the latch mechanism because of
> how BPF map lookups work.  The goal is to be able to use mmap'd BPF map access
> in the future which will require concurrency control.  So, the producer is
> ready for that.  So, the producer is ahead of the game but there is no need
> to mention this in the consumer patch because it doesn't affect functionality
> at all.

The omission is rather startling from a reader's point of view.  The 
producer makes such a big issue of the seq#, it is surprising that the 
consumer does nothing with it.  I think some acknowledgment of the 
situation is in order.