G'Day, On Mon, Jul 11, 2011 at 11:33 PM, Bryan Cantrill <br...@joyent.com> wrote:
> All, > > With my apologies for having been somewhat on a tear through old > DTrace RFEs, another that has been annoying as of late is that > tracemem() does not take a dynamic size. Of course, EPIDs must > correspond to a static size, so the fix is not to make the size > entirely dynamic, but rather to add an optional third argument to > tracemem() that is a D expression that captures the dynamic size -- > provided that it is no greater than the static size (still) provided > as the second argument. A patch that implements this is attached for > your review and consideration; as with our other work, this will be > showing up at http://github.com/joyent/illumos-joyent in the near > future (our ticket for this is OS-481, "DTrace tracemem() action > should take a dynamic size argument"). Note the fix for another > long-standing falls out of this: if you tracemem() something that is > a string, it will show up as a proper hex dump instead of being > interpreted as a string. Please let me know if you have any questions > or comments! > > Many thanks! If anyone is wondering about an example use case, I ran into this only a couple of weeks ago when blogging about copyin(), writing: "1st caveat: The "100" in the above one-liner is the length for tracemem(); unfortunately, it must be a scalar constant and can't be a variable, otherwise I'd use "arg1" to print the entire returned length." (The blog post was: http://dtrace.org/blogs/brendan/2011/06/30/tweaking-memory-on-the-fly/) This fix will let me pick a max-expected returned length for arg1, and then use the extra arg to truncate it. I'll need to update the blog post. :-) Brendan -- Brendan Gregg, Joyent http://dtrace.org/blogs/brendan
_______________________________________________ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
