[RFC PATCH] Appease Valgrind over ptrace calls.

[RFC PATCH] Appease Valgrind over ptrace calls.

[Pedro Alves]

Trying out Valgrind on gdbserver, on x86_64, I get these annoying warnings:

==3987== Conditional jump or move depends on uninitialised value(s)
==3987==    at 0x430FB2: x86_linux_update_xmltarget (linux-x86-low.c:1250)
==3987==    by 0x4310F7: x86_linux_process_qsupported (linux-x86-low.c:1304)
==3987==    by 0x42BD07: linux_process_qsupported (linux-low.c:5253)
==3987==    by 0x40CF75: handle_query (server.c:1571)
==3987==    by 0x41045E: process_serial_event (server.c:2930)
==3987==    by 0x41154C: handle_serial_event (server.c:3364)
==3987==    by 0x4170BA: handle_file_event (event-loop.c:436)
==3987==    by 0x416847: process_event (event-loop.c:191)
==3987==    by 0x41762E: start_event_loop (event-loop.c:554)
==3987==    by 0x410137: main (server.c:2781)
==3987==
==3987== Conditional jump or move depends on uninitialised value(s)
==3987==    at 0x431009: x86_linux_update_xmltarget (linux-x86-low.c:1258)
==3987==    by 0x4310F7: x86_linux_process_qsupported (linux-x86-low.c:1304)
==3987==    by 0x42BD07: linux_process_qsupported (linux-low.c:5253)
==3987==    by 0x40CF75: handle_query (server.c:1571)
==3987==    by 0x41045E: process_serial_event (server.c:2930)
==3987==    by 0x41154C: handle_serial_event (server.c:3364)
==3987==    by 0x4170BA: handle_file_event (event-loop.c:436)
==3987==    by 0x416847: process_event (event-loop.c:191)
==3987==    by 0x41762E: start_event_loop (event-loop.c:554)
==3987==    by 0x410137: main (server.c:2781)
==3987==
==3987== Conditional jump or move depends on uninitialised value(s)
==3987==    at 0x435AD9: i387_xsave_to_cache (i387-fp.c:477)
==3987==    by 0x42FD67: x86_store_xstateregset (linux-x86-low.c:353)
==3987==    by 0x42A01C: regsets_fetch_inferior_registers (linux-low.c:4063)
==3987==    by 0x42A7FC: linux_fetch_registers (linux-low.c:4355)
==3987==    by 0x40675B: get_thread_regcache (regcache.c:50)
==3987==    by 0x408D0F: prepare_resume_reply (remote-utils.c:1337)
==3987==    by 0x40F5CC: handle_status (server.c:2372)
==3987==    by 0x41077F: process_serial_event (server.c:3029)
==3987==    by 0x41154C: handle_serial_event (server.c:3364)
==3987==    by 0x4170BA: handle_file_event (event-loop.c:436)
==3987==    by 0x416847: process_event (event-loop.c:191)
==3987==    by 0x41762E: start_event_loop (event-loop.c:554)

It seems like Valgrind doesn't understand that ptrace writes to the
buffers in question, so throws out false positives.

Valgrind on GDB spews out equivalent warnings.

This patch works around it in gdbserver.  Is there a better way to do
this perhaps?

Hmm, nut, I'm now wondering why we don't see many more ptrace-related
warnings in other cases.  Maybe Valgrind does know about ptrace, but
not about PTRACE_GETREGSET?

gdb/gdbserver/
2013-02-14  Pedro Alves  <palves@redhat.com>

	* linux-low.c (regsets_fetch_inferior_registers)
---
 gdb/gdbserver/linux-low.c     |    6 ++++--
 gdb/gdbserver/linux-x86-low.c |    4 +++-
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/gdb/gdbserver/linux-low.c b/gdb/gdbserver/linux-low.c
index c52cd2e..00214da 100644
--- a/gdb/gdbserver/linux-low.c
+++ b/gdb/gdbserver/linux-low.c
@@ -4022,7 +4022,8 @@ regsets_fetch_inferior_registers (struct regcache *regcache)
 	  continue;
 	}
 
-      buf = xmalloc (regset->size);
+      /* Zero out the buffer to appease Valgrind.  */
+      buf = xcalloc (1, regset->size);
 
       nt_type = regset->nt_type;
       if (nt_type)
@@ -4092,7 +4093,8 @@ regsets_store_inferior_registers (struct regcache *regcache)
 	  continue;
 	}
 
-      buf = xmalloc (regset->size);
+      /* Zero out the buffer to appease Valgrind.  */
+      buf = xcalloc (1, regset->size);
 
       /* First fill the buffer with the current register set contents,
 	 in case there are any items in the kernel's regset that are
diff --git a/gdb/gdbserver/linux-x86-low.c b/gdb/gdbserver/linux-x86-low.c
index 31657d3..604ecd6 100644
--- a/gdb/gdbserver/linux-x86-low.c
+++ b/gdb/gdbserver/linux-x86-low.c
@@ -1224,7 +1224,9 @@ x86_linux_update_xmltarget (void)
   /* Check if XSAVE extended state is supported.  */
   if (have_ptrace_getregset == -1)
     {
-      unsigned long long xstateregs[I386_XSTATE_SSE_SIZE / sizeof (long long)];
+      /* Zeroed out to appease Valgrind.  */
+      unsigned long long
+	xstateregs[I386_XSTATE_SSE_SIZE / sizeof (long long)] = {0};
       struct iovec iov;
 
       iov.iov_base = xstateregs;

Re: [RFC PATCH] Appease Valgrind over ptrace calls.

[Pedro Alves]

On 02/14/2013 06:38 PM, Pedro Alves wrote:

> Hmm, but, I'm now wondering why we don't see many more ptrace-related
> warnings in other cases.  Maybe Valgrind does know about ptrace, but
> not about PTRACE_GETREGSET?

Found it...  https://bugs.kde.org/show_bug.cgi?id=308886

Answer is yes.  Valgrind doesn't know about PTRACE_GETREGSET.  Issue
seems to be fixed upstream.

Dropping this patch.

Thanks for listening. :-)

-- 
Pedro Alves