This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[DOC/RFA PATCH] Move comment on the 'stepping over resolver' mechanism to the internals manual.
- From: Pedro Alves <palves at redhat dot com>
- To: gdb-patches at sourceware dot org
- Date: Wed, 26 Jun 2013 22:42:37 +0100
- Subject: [DOC/RFA PATCH] Move comment on the 'stepping over resolver' mechanism to the internals manual.
This whole comment is now a bit out of place. I looked into moving it
to handle_inferior_event, close to where in_solib_dynsym_resolve_code
is used, but then there are 3 such places. I then looked at
fragmenting it, pushing bits closer to the definitions of
in_solib_dynsym_resolve_code and gdbarch_skip_solib_resolver, but then
we'd lose the main advantage which is the overview. In the end, I
realized this can fit nicely as internals manual material.
This could possibly be a subsection of a new "run control", or "source
stepping" or "stepping" or some such a bit more general section, but
we can do that when we have more related content... Even the "single
stepping" section is presently empty...
gdb/doc/
2013-06-26 Pedro Alves <palves@redhat.com>
* gdbint.texinfo (Algorithms) <Stepping over runtime loader
dynamic symbol resolution code>: New section, based on infrun.c
comment.
gdb/
2013-06-26 Pedro Alves <palves@redhat.com>
* infrun.c: Remove comment describing the 'stepping over runtime
loader dynamic symbol resolution code' mechanism; moved to
gdbint.texinfo.
---
gdb/doc/gdbint.texinfo | 42 ++++++++++++++++++++++++++++++++++++++++++
gdb/infrun.c | 40 ----------------------------------------
2 files changed, 42 insertions(+), 40 deletions(-)
diff --git a/gdb/doc/gdbint.texinfo b/gdb/doc/gdbint.texinfo
index 8f82611..55bba57 100644
--- a/gdb/doc/gdbint.texinfo
+++ b/gdb/doc/gdbint.texinfo
@@ -592,6 +592,48 @@ but @code{placed_size} may be.
@section Single Stepping
+@section Stepping over runtime loader dynamic symbol resolution code
+
+If the program uses ELF-style shared libraries, then calls to
+functions in shared libraries go through stubs, which live in a table
+called the PLT (Procedure Linkage Table). The first time the function
+is called, the stub sends control to the dynamic linker, which looks
+up the function's real address, patches the stub so that future calls
+will go directly to the function, and then passes control to the
+function.
+
+If we are stepping at the source level, we don't want to see any of
+this --- we just want to skip over the stub and the dynamic linker.
+The simple approach is to single-step until control leaves the dynamic
+linker.
+
+However, on some systems (e.g., Red Hat's 5.2 distribution) the
+dynamic linker calls functions in the shared C library, so you can't
+tell from the PC alone whether the dynamic linker is still running.
+In this case, we use a step-resume breakpoint to get us past the
+dynamic linker, as if we were using "next" to step over a function
+call.
+
+The @code{in_solib_dynsym_resolve_code} says whether we're in the
+dynamic linker code or not. Normally, this means we single-step.
+However, if @code{gdbarch_skip_solib_resolver} then returns non-zero,
+then its value is an address where we can place a step-resume
+breakpoint to get past the linker's symbol resolution function.
+
+The @code{in_dynsym_resolve_code} hook of the @code{target_so_ops}
+vector can generally be implemented in a pretty portable way, by
+comparing the PC against the address ranges of the dynamic linker's
+sections.
+
+The @code{gdbarch_skip_solib_resolver} implementation is generally
+going to be system-specific, since it depends on internal details of
+the dynamic linker. It's usually not too hard to figure out where to
+put a breakpoint, but it certainly isn't portable.
+@code{gdbarch_skip_solib_resolver} should do plenty of sanity
+checking. If it can't figure things out, returning zero and getting
+the (possibly confusing) stepping behavior is better than signaling an
+error, which will obscure the change in the inferior's state. */
+
@section Signal Handling
@section Thread Handling
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 84e4053..8fb219a 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -181,46 +181,6 @@ set_disable_randomization (char *args, int from_tty,
"this platform."));
}
-
-/* If the program uses ELF-style shared libraries, then calls to
- functions in shared libraries go through stubs, which live in a
- table called the PLT (Procedure Linkage Table). The first time the
- function is called, the stub sends control to the dynamic linker,
- which looks up the function's real address, patches the stub so
- that future calls will go directly to the function, and then passes
- control to the function.
-
- If we are stepping at the source level, we don't want to see any of
- this --- we just want to skip over the stub and the dynamic linker.
- The simple approach is to single-step until control leaves the
- dynamic linker.
-
- However, on some systems (e.g., Red Hat's 5.2 distribution) the
- dynamic linker calls functions in the shared C library, so you
- can't tell from the PC alone whether the dynamic linker is still
- running. In this case, we use a step-resume breakpoint to get us
- past the dynamic linker, as if we were using "next" to step over a
- function call.
-
- in_solib_dynsym_resolve_code() says whether we're in the dynamic
- linker code or not. Normally, this means we single-step. However,
- if gdbarch_skip_solib_resolver then returns non-zero, then its
- value is an address where we can place a step-resume breakpoint to
- get past the linker's symbol resolution function.
-
- The in_dynsym_resolve_code hook of the target_so_ops vector can
- generally be implemented in a pretty portable way, by comparing the
- PC against the address ranges of the dynamic linker's sections.
-
- The gdbarch_skip_solib_resolver implementation is generally going
- to be system-specific, since it depends on internal details of the
- dynamic linker. It's usually not too hard to figure out where to
- put a breakpoint, but it certainly isn't portable.
- gdbarch_skip_solib_resolver should do plenty of sanity checking.
- If it can't figure things out, returning zero and getting the
- (possibly confusing) stepping behavior is better than signaling an
- error, which will obscure the change in the inferior's state. */
-
/* "Observer mode" is somewhat like a more extreme version of
non-stop, in which all GDB operations that might affect the
target's execution have been disabled. */