[Ltrace-devel] [PATCH] Check return value of PTRACE_PEEKTEXT for ppc64

[Edjunior Barbosa Machado]

Hi,

the current upstream version of ltrace shows an ugly testsuite result on ppc64:

# of expected passes            23
# of unexpected failures        74

Taking a closer look, I noticed ltrace is unable to track mostly of the library calls on PowerPC 64-bit. This seems to be happening due to a failure in a PTRACE_PEEKTEXT call in the function sym2addr(), which returns -1 ('No such process', according to perror()) that is saved as a valid address for a breakpoint. The following patch ignores this error and sets addr = 0, which means this address will be checked again in the future. This patch fixes mostly of the failures in the testsuite on ppc64:

# of expected passes            88
# of unexpected failures        9

Please let me know if you have any comments or concerns.

Thanks!
-- 
Edjunior Barbosa Machado
IBM Linux Technology Center

diff --git a/sysdeps/linux-gnu/ppc/plt.c b/sysdeps/linux-gnu/ppc/plt.c
index 668f63d..9308678 100644
--- a/sysdeps/linux-gnu/ppc/plt.c
+++ b/sysdeps/linux-gnu/ppc/plt.c
@@ -1,5 +1,6 @@
 #include <gelf.h>
 #include <sys/ptrace.h>
+#include <errno.h>
 #include "common.h"
 
 GElf_Addr
@@ -42,7 +43,15 @@ sym2addr(Process *proc, struct library_symbol *sym) {
 	// non-secure case, the PLT is executable code and we can put the
 	// break-point right in the PLT.
 
-	pt_ret = ptrace(PTRACE_PEEKTEXT, proc->pid, addr, 0);
+	if ((pt_ret = ptrace(PTRACE_PEEKTEXT, proc->pid, addr, 0)) == -1) {
+		// ptrace might return -1 (No such process) on PowerPC-64 when
+		// the process is not initialized yet. Return 0 for now and the
+		// address will be collected again in the future.
+		if (errno == ESRCH) 
+			return 0;
+		else 
+			perror("PTRACE_PEEKTEXT");
+	}
 
 #if SIZEOF_LONG == 8
 	if (proc->mask_32bit) {