Re: xentrace buffering

[Rob Gardner <rob.gardner@hp.com>]

pak333@comcast.net wrote:
> Hi,
>
> I am looking at the xentrace code and it looks like the individual 
> cores/cpus write their (trace data) into individual regions of 
> memory and they are all collected by xentrace and processed.
>
> I would like to have one trace buffer for all cores so that I *can get 
> an exact sequence in time order* of the events that have occurred over 
> the sampling period across cores/cpus.
>
> Is that easily doable>? If so, any pointers can someone send me 
> pointers as to where I should start looking at modifying.
>

It's doable, but would require some work. I think the reason for having 
different buffers for each core is so that the trace code can be really 
light weight, since it doesn't have to deal with acquiring a write lock, 
etc.

Theoretically, each core's clock should be very close to all the others, 
so the timestamp in the trace records should allow a reasonably accurate 
reconstruction of the order of events.

Rob