>
>My wish list: (not strictly DTrace limitations, but related)
>
>       1. Since the pid provider takes instruction addresses, I
>          wish that there was a reliable and easy method to
>          determine the line number in the source given the
>          address and vice versa. The dbx program does this, so
>          it must be possible. In fact, the right compilation
>          and linker flags, maybe DTrace could do this right in
>          the pid provider!

There's a one to many mapping, so which instruction would you want to
instrument?  (the compiler will completely rearrange the instruction
stream).

>       2. Comprehensive data symbol information in CTF for the
>          kernel. Right now, DTrace cannot find any of the symbols
>          that are defined in .s files in the kernel. Even if you
>          provide the address manually, it does not know the data
>          type. So, this wish is two-fold. One, give the user a
>          way to manually tell DTrace a symbol address and data
>          type and let it treat this symbol as a first class
>          symbol from then on. And two, add some kind of macro
>          to the assembler so it would generate whatever it needs
>          to in the resulting object files to allow the variables
>          in .s files to play in the CTF game.


For which kernel variables is this a limitation?

As there's a fairly easy work around: define all globals in a .c file,
I would not necessarily see this as an important restriction.

Casper

_______________________________________________
dtrace-discuss mailing list
[email protected]

Reply via email to