dtrace discuss - Jun 2006 - tracemem() not exactly what I had in mind

I was trying to use tracemem to look at mblk contents.  First, I tried to use
mblk->b_wptr - mblk->b_rptr as the size, and was told that it had to be a 
constant.  Foo.  (RFE #1).

Then, I tried to use 8 as the size, and kept getting decimal numbers printed. 
Stumbled on #pragma D option rawbytes=true (is the =true necessary, btw?) and 
that didn''t help.

Then, in desperation, I changed the size to 16.  Voila!

I *think* this is because tracemem() doesn''t really look at rawbytes
before it
looks for size 1, 2, 4, or 8 bytes, and handles them specially.  But
I''m gonna
call this a bug in the rawbytes option implementation.  Agreed?

On Wed, Jun 28, 2006 at 08:48:11PM -0700, Dan Mick
wrote:
> I was trying to use tracemem to look at mblk contents.  First, I tried to 
> use
> mblk->b_wptr - mblk->b_rptr as the size, and was told that it had to
be a
> constant.  Foo.  (RFE #1).
It would be nice to be able to implement that RFE, but one the simplifying
assumptions that DTrace makes is that trace records from a given probe are of
a fixed size. Can you just pick a really big default?
> Then, I tried to use 8 as the size, and kept getting decimal numbers 
> printed. Stumbled on #pragma D option rawbytes=true (is the =true 
> necessary, btw?) and that didn''t help.
> 
> Then, in desperation, I changed the size to 16.  Voila!
> 
> I *think* this is because tracemem() doesn''t really look at
rawbytes before
> it looks for size 1, 2, 4, or 8 bytes, and handles them specially.  But
I''m
> gonna call this a bug in the rawbytes option implementation.  Agreed?
If the size is 1, 2, 4, or 8 bytes, DTrace assumes that you''re tracing
an
integral type and prints it as a signed decimal integer. A reasonable RFE
might be to have tracemem() always pay attention to rawbytes.

Adam

-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl

On Thu, 2006-06-29 at 03:40, Adam Leventhal wrote:
> On Wed, Jun 28, 2006 at 08:48:11PM -0700, Dan Mick wrote:
> > I was trying to use tracemem to look at mblk contents.  First, I tried
to
> > use
> > mblk->b_wptr - mblk->b_rptr as the size, and was told that it
had to be a
> > constant.  Foo.  (RFE #1).
> 
> It would be nice to be able to implement that RFE, but one the simplifying
> assumptions that DTrace makes is that trace records from a given probe are
of
> a fixed size. Can you just pick a really big default?
what i''d like to see (because mblks are so ubiquitous in certain
corners
of the kernel) is:

	tracemblk(mp, offset, maxlen)

So long as maxlen is a constant (and you have a careful mblk-chain
walker which won''t get stuck in a loop -- and I can think of a couple
of
different ways to pull this offf) I think this fits the dtrace safety
model.

						- Bill