Re: [RFC PATCH 1/2 v2] Unified trace buffer

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Steven Rostedt (rostedt@goodmis.org) wrote:
> 
> On Thu, 25 Sep 2008, Mathieu Desnoyers wrote:
> > 
> > Is there a reason to use delta between events rather than simply write
> > the 27 LSBs that I would have missed ?
> 
> One answer is that your counter wrap problem is extended.
> 
> That is, you have 27 bits of time between each event to not worry about
> wraps. But if you go against the page itself, the last event on that page
> is more likely to suffer.
> 

You can do the exact same thing and manage to keep the absolute time.
You just have to adapt the reader like this : (this would be for
per-event cycle count in the 32 LSBs, slight bitmask adaptation needed
for 27 bits only).

keep a 64 bits TSC value in a per-buffer variable. The previous value is
always re-used for the next read.

let's call it :
tf->buffer.tsc  (u64)

read_event() would look like :

u32 timetamp = read_event_timetamp();

if(timestamp < (0xFFFFFFFFULL & tf->buffer.tsc)) {
    /* overflow */
    tf->buffer.tsc = ((tf->buffer.tsc & 0xFFFFFFFF00000000ULL)
                     + 0x100000000ULL) | (u64)timestamp;
} else {
    /* no overflow */
    tf->buffer.tsc = (tf->buffer.tsc & 0xFFFFFFFF00000000ULL)
                     | (u64)timestamp;
}

This will detect 32 bits overflow and keep the tf->buffer.tsc in sync
with the TSC representation on the traced machine as long as events are
less then 27 bits apart. A "full tsc" header can also be easily managed
with this by updating the tf->buffer.tsc value completely when such
event is met.

Mathieu

> -- Steve
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68