From: Abhay Kandpal <abhay@linux.ibm.com>
To: gdb-patches@sourceware.org
Cc: Ulrich.Weigand@de.ibm.com, cel@linux.ibm.com, abhay.k@ibm.com,
Abhay Kandpal <abhay@linux.ibm.com>
Subject: [PATCH v2] PowerPC: fix external calls from modules built by the compile command
Date: Wed, 19 Aug 2026 13:34:29 -0500 [thread overview]
Message-ID: <20260819183429.3039030-1-abhay@linux.ibm.com> (raw)
A module built for the compile command is loaded into inferior memory
without a linker, so it has no PLT. On ELFv2 the default code model
still emits a direct bl for a call to another module, leaving r12 unset
and the caller's TOC pointer unrestored, and the callee then derives a
wrong r2 from a stale r12.
gdb.compile/compile-setjmp.exp shows this: the injected code calls
setjmp, whose global entry point computes r2 from a stale r12, and the
first TOC-relative load segfaults.
-mlongcall makes the call indirect, but on its own it reaches the target
through a PLT entry, giving R_PPC64_PLT16_HA/LO_DS relocations that BFD's
generic linker rejects. Adding -mno-pltseq resolves the call against the
module's own TOC instead, via R_PPC64_TOC16_HA/LO_DS, which GDB already
handles. GCC then emits the r2 save and restore around the call itself.
Tested on powerpc64le-linux; the two compile-setjmp.exp failures now pass
with no unexpected failures in the full testsuite.
Suggested-by: Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
---
This patch is reg tested.
gdb/ppc-linux-tdep.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/gdb/ppc-linux-tdep.c b/gdb/ppc-linux-tdep.c
index 8a55ff5d4af..b468dbcb62c 100644
--- a/gdb/ppc-linux-tdep.c
+++ b/gdb/ppc-linux-tdep.c
@@ -2196,6 +2196,18 @@ rs6000_linux_adjust_frame_regnum (struct gdbarch *gdbarch, int num,
return num;
}
+/* Implement gdbarch_gcc_target_options. A module built for the "compile"
+ command is loaded into inferior memory without a linker, so it has no
+ PLT. -mlongcall makes calls to other modules indirect, and -mno-pltseq
+ stops GCC reaching the target through a PLT entry; together they let the
+ call be resolved against the module's own TOC. */
+
+static std::string
+ppc64_gcc_target_options (struct gdbarch *gdbarch)
+{
+ return default_gcc_target_options (gdbarch) + " -mlongcall -mno-pltseq";
+}
+
static void
ppc_linux_init_abi (struct gdbarch_info info,
struct gdbarch *gdbarch)
@@ -2304,6 +2316,7 @@ ppc_linux_init_abi (struct gdbarch_info info,
(gdbarch, ppc_elfv2_elf_make_msymbol_special);
set_gdbarch_skip_entrypoint (gdbarch, ppc_elfv2_skip_entrypoint);
+ set_gdbarch_gcc_target_options (gdbarch, ppc64_gcc_target_options);
}
/* Shared library handling. */
--
2.52.0
next reply other threads:[~2026-08-19 18:35 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-08-19 18:34 Abhay Kandpal [this message]
2026-08-20 11:03 ` Ulrich Weigand
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260819183429.3039030-1-abhay@linux.ibm.com \
--to=abhay@linux.ibm.com \
--cc=Ulrich.Weigand@de.ibm.com \
--cc=abhay.k@ibm.com \
--cc=cel@linux.ibm.com \
--cc=gdb-patches@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox