Thanks, Chris for the detailed reply. On Thu, 19 Nov 2020 at 22:56, Chris Johns <chr...@rtems.org> wrote: > > On 19/11/20 5:33 pm, Hesham Almatary wrote: > > Thanks for the info, Chris. > > > > On Wed, 18 Nov 2020 at 02:43, Chris Johns <chr...@rtems.org> wrote: > >> On 17/11/20 10:43 pm, Hesham Almatary wrote: > >>> Hello, > >>> > >>> I am trying to debug a dynamically loaded ELF object with GDB: My setup > >>> is: > >>> rtems-riscv64: > >>> GDB: GNU gdb (GDB) 10.0.50.20200904-git (build with RSB/rtems6) > >>> Host: Ubuntu 20.04.1 LTS > >>> > >>> GDB doesn't seem to notice when I dlopen an object (which gets loaded, > >>> linked and run fine). > >> > >> libdl has the NetBSD type support built in for debugging. A function is > >> called > >> when a module is loaded and there is a table of modules GDB reads and then > >> loads > >> the object file. > >> > > Is that expected to work on Linux with recent GDB versions? > > We need a patch for GDB to add the support needed for RTEMS. > Right, that's what I thought too.
> > GDB > > doesn't seem to detect when "_rtld_debug_state" is called from dlopen, > > and consequently isn't aware of the list of objects. Does that need > > GDB to be modified with solib-rtems* or something? > > A patch was created as part the 2013 RTL GSoC project (an excellent year for > students :)). The repo is https://github.com/MrVan/gdb/commits/rtems-gdb. The > only problem is GDB has moved to C++ since then and I am not sure what effect > that has had on the patch. > That's really helpful. It would have been great if that got merged, still, it's a good reference now for someone who might add back RTEMS support. > >>> Any instructions or idea how to troubleshoot that further? > >> > >> At this point the simplest path is to get the load address(es) and load the > >> symbols with the VMA. > >> > > I can use "add-symbol-file" filename for each object and give it the > > text_base address. This only allows me to break on functions within > > loaded objects. If that's not what you mean, how can I load the > > symbols with the VMA? > > I think using a command is always going to be a bit of a hack. The patch Peng > created allowed you to set breakpoints in code that was not loaded and it > would > break when loaded _and_ executed. > > Chris -- Hesham _______________________________________________ devel mailing list devel@rtems.org http://lists.rtems.org/mailman/listinfo/devel
