Re: [PATCH 1/4] tests/qtest: Remove TPM tests

[Philippe Mathieu-Daudé]

Hi Stefan,

On 1/15/21 8:56 PM, Stefan Berger wrote:
> On 1/15/21 1:40 PM, Stefan Berger wrote:
>> On 1/15/21 11:06 AM, Philippe Mathieu-Daudé wrote:
>>> On 1/15/21 4:53 PM, Stefan Berger wrote:
>>>> On 1/15/21 10:52 AM, Philippe Mathieu-Daudé wrote:
>>>>> Subject is incorrect, this is not a removal of the tests, but
>>>>> removal of their execution. The tests are still in the repository.
>>>>> This is more of a disablement.
>>>> How do you compile / run them to have the LeakSanitizer checks?
>>> I used:
>>>
>>> ../configure --cc=clang --enable-sanitizers && make check-qtest
>>>
>>> $ clang -v
>>> clang version 10.0.1 (Fedora 10.0.1-3.fc32)
>>>
>>> This was previously covered by patchew CI. I just figured
>>> patchew is running without the LeakSanitizer since commit
>>> 6f89ec7442e ("docker: test-debug: disable LeakSanitizer"):
>>>
>>>   docker: test-debug: disable LeakSanitizer
>>>
>>>   There are just too many leaks in device-introspect-test (especially
>>> for
>>>   the plethora of arm and aarch64 boards) to make LeakSanitizer useful;
>>>   disable it for now.
>>>
>> I only get short stack traces:
>>
>>
>> Indirect leak of 852840 byte(s) in 207 object(s) allocated from:
>>     #0 0x561a8c2f8b57 in calloc
>> (/home/stefanb/tmp/qemu-tip/build/tests/qtest/tpm-crb-swtpm-test+0x23fb57)
>>
>>     #1 0x14f0963069b0 in g_malloc0 (/lib64/libglib-2.0.so.0+0x589b0)
>>     #2 0x561a8c4c2508 in json_parser_parse
>> /home/stefanb/tmp/qemu-tip/build/../qobject/json-parser.c:580:14
>>     #3 0x561a8c4a99aa in json_message_process_token
>> /home/stefanb/tmp/qemu-tip/build/../qobject/json-streamer.c:92:12
>>     #4 0x561a8c4b6cfb in json_lexer_feed_char
>> /home/stefanb/tmp/qemu-tip/build/../qobject/json-lexer.c:313:13
>>
>> Indirect leak of 6624 byte(s) in 207 object(s) allocated from:
>>     #0 0x561a8c2f8b57 in calloc
>> (/home/stefanb/tmp/qemu-tip/build/tests/qtest/tpm-crb-swtpm-test+0x23fb57)
>>
>>     #1 0x14f0963069b0 in g_malloc0 (/lib64/libglib-2.0.so.0+0x589b0)
>>
>> Indirect leak of 1449 byte(s) in 207 object(s) allocated from:
>>     #0 0x561a8c2f899f in malloc
>> (/home/stefanb/tmp/qemu-tip/build/tests/qtest/tpm-crb-swtpm-test+0x23f99f)
>>
>>     #1 0x14f096306958 in g_malloc (/lib64/libglib-2.0.so.0+0x58958)
>>
>> How can I see more of those?
> 
> 
> I now added -fno-omit-frame-pointer to configure (should it not be
> there?) and it now shows some useful stacktraces.

No idea... Cc'ing Marc-André.

If the issue is only with ASan we could add the flag locally
to avoid generic problems with _FORTIFY_SOURCE:

-- >8 --
@@ -5309,6 +5309,9 @@ fi

 if test "$have_asan" = "yes"; then
   QEMU_CFLAGS="-fsanitize=address $QEMU_CFLAGS"
+  if test "$debug" = "no" ; then
+    QEMU_CFLAGS="-fno-omit-frame-pointer $QEMU_CFLAGS"
+  fi
   QEMU_LDFLAGS="-fsanitize=address $QEMU_LDFLAGS"
   if test "$have_asan_iface_h" = "no" ; then
       echo "ASAN build enabled, but ASAN header missing." \
---

> 
> 
> diff --git a/configure b/configure
> index 155dda124c..ed86b5ca32 100755
> --- a/configure
> +++ b/configure
> @@ -5308,7 +5308,7 @@ if test "$gprof" = "yes" ; then
>  fi
> 
>  if test "$have_asan" = "yes"; then
> -  QEMU_CFLAGS="-fsanitize=address $QEMU_CFLAGS"
> +  QEMU_CFLAGS="-fsanitize=address -fno-omit-frame-pointer $QEMU_CFLAGS"
>    QEMU_LDFLAGS="-fsanitize=address $QEMU_LDFLAGS"
>    if test "$have_asan_iface_h" = "no" ; then
>        echo "ASAN build enabled, but ASAN header missing." \
> diff --git a/tests/qtest/tpm-util.c b/tests/qtest/tpm-util.c
> 
> 
> This is my TPM related fix. Maybe it resolve the issue for you also?

Great, a trivial diff :) I'll try it next week.

> 
> 
> index 5a33a6ef0f..b70cc32d60 100644
> --- a/tests/qtest/tpm-util.c
> +++ b/tests/qtest/tpm-util.c
> @@ -250,7 +250,7 @@ void tpm_util_wait_for_migration_complete(QTestState
> *who)
>          status = qdict_get_str(rsp_return, "status");
>          completed = strcmp(status, "completed") == 0;
>          g_assert_cmpstr(status, !=,  "failed");
> -        qobject_unref(rsp_return);
> +        qobject_unref(rsp);
>          if (completed) {
>              return;
>          }
> 
> Now I see ppc64 related leaks:
> 
> Direct leak of 200 byte(s) in 1 object(s) allocated from:
>     #0 0x14c9b743c837 in __interceptor_calloc (/lib64/libasan.so.6+0xb0837)
>     #1 0x14c9b6e8b9b0 in g_malloc0 (/lib64/libglib-2.0.so.0+0x589b0)
>     #2 0x55c5e7130a1a in qemu_init_vcpu ../softmmu/cpus.c:618
>     #3 0x55c5e68b30c0 in ppc_cpu_realize
> ../target/ppc/translate_init.c.inc:10146
>     #4 0x55c5e7539c08 in device_set_realized ../hw/core/qdev.c:761
>     #5 0x55c5e714aa38 in property_set_bool ../qom/object.c:2255
>     #6 0x55c5e7145d52 in object_property_set ../qom/object.c:1400
>     #7 0x55c5e714f99f in object_property_set_qobject
> ../qom/qom-qobject.c:28
>     #8 0x55c5e71465f4 in object_property_set_bool ../qom/object.c:1470
>     #9 0x55c5e666ae21 in spapr_realize_vcpu ../hw/ppc/spapr_cpu_core.c:254
>     #10 0x55c5e666ae21 in spapr_cpu_core_realize
> ../hw/ppc/spapr_cpu_core.c:337
>     #11 0x55c5e7539c08 in device_set_realized ../hw/core/qdev.c:761
>     #12 0x55c5e714aa38 in property_set_bool ../qom/object.c:2255
>     #13 0x55c5e7145d52 in object_property_set ../qom/object.c:1400
>     #14 0x55c5e714f99f in object_property_set_qobject
> ../qom/qom-qobject.c:28
>     #15 0x55c5e71465f4 in object_property_set_bool ../qom/object.c:1470
>     #16 0x55c5e5c7553c in qdev_device_add ../softmmu/qdev-monitor.c:665
>     #17 0x55c5e6fd4cc4 in device_init_func ../softmmu/vl.c:1201
>     #18 0x55c5e78fc7bb in qemu_opts_foreach ../util/qemu-option.c:1147
>     #19 0x55c5e6fc8912 in qemu_create_cli_devices ../softmmu/vl.c:2488
>     #20 0x55c5e6fc8912 in qmp_x_exit_preconfig ../softmmu/vl.c:2527
>     #21 0x55c5e6fcfb4b in qemu_init ../softmmu/vl.c:3533
>     #22 0x55c5e5b18e78 in main ../softmmu/main.c:49
>     #23 0x14c9b50fa041 in __libc_start_main (/lib64/libc.so.6+0x27041)
> 
> [..]

Currently the fuzzed qtests are only reported for X86, so I didn't
bother testing the other targets. Cc'ing qemu-ppc@ however.

Thanks for the quick feedbacks,

Phil.