Re: [Qemu-devel] [PATCH] tests/check-block: Skip iotests when sanitizers are enabled

[Peter Maydell]

On Fri, 23 Aug 2019 at 10:35, Thomas Huth <address@hidden> wrote:
>
> On 8/23/19 11:04 AM, Peter Maydell wrote:
> > On Fri, 23 Aug 2019 at 09:43, Thomas Huth <address@hidden> wrote:
> >>
> >> The sanitizers (especially the address sanitizer from Clang) are
> >> sometimes printing out warnings or false positives - this spoils
> >> the output of the iotests, causing some of the tests to fail.
> >> Thus let's skip the automatic iotests during "make check" when the
> >> user configured QEMU with --enable-sanitizers.
> >
> > Do you have a log of what the sanitizer is saying?
>
> https://patchew.org/logs/QEMU/testing.asan/?type=project

(I get a "Not Found" error for that URL.)

> Example:
>
> +Indirect leak of 4120 byte(s) in 1 object(s) allocated from:
> +    #0 0x562a2ffc3c4e in calloc
> (TEST_DIR/build/x86_64-softmmu/qemu-system-x86_64+0x1a16c4e)
> +    #1 0x7fca6acf3cf0 in g_malloc0 (/lib64/libglib-2.0.so.0+0x55cf0)
> +    #2 0x562a3200c3d0 in bdrv_refresh_filename TEST_DIR/src/block.c:6416:12
> +    #3 0x562a3200b8f7 in bdrv_refresh_filename TEST_DIR/src/block.c:6388:9
> +    #4 0x562a3200b8f7 in bdrv_refresh_filename TEST_DIR/src/block.c:6388:9
> +    #5 0x562a31ffa461 in bdrv_backing_attach TEST_DIR/src/block.c:1064:5
> +    #6 0x562a320212c6 in bdrv_replace_child_noperm
> TEST_DIR/src/block.c:2283:13
> +    #7 0x562a3201ed50 in bdrv_replace_node TEST_DIR/src/block.c:4210:9
> +    #8 0x562a32021649 in bdrv_append TEST_DIR/src/block.c:4250:5
> +    #9 0x562a3234e573 in commit_start TEST_DIR/src/block/commit.c:307:5
> +    #10 0x562a30cc6cce in qmp_block_commit TEST_DIR/src/blockdev.c:3480:9
> +    #11 0x562a31dceb33 in qmp_marshal_block_commit
> TEST_DIR/build/qapi/qapi-commands-block-core.c:407:5
> +    #12 0x562a3260be28 in do_qmp_dispatch
> TEST_DIR/src/qapi/qmp-dispatch.c:131:5
> +    #13 0x562a3260b105 in qmp_dispatch
> TEST_DIR/src/qapi/qmp-dispatch.c:174:11
> +    #14 0x562a31cd1b15 in monitor_qmp_dispatch
> TEST_DIR/src/monitor/qmp.c:120:11
> +    #15 0x562a31ccfd45 in monitor_qmp_bh_dispatcher
> TEST_DIR/src/monitor/qmp.c:209:9
> +    #16 0x562a327a91ea in aio_bh_call TEST_DIR/src/util/async.c:89:5
> +    #17 0x562a327a9902 in aio_bh_poll TEST_DIR/src/util/async.c:117:13
> +    #18 0x562a327cb590 in aio_dispatch TEST_DIR/src/util/aio-posix.c:459:5
> +    #19 0x562a327ae933 in aio_ctx_dispatch TEST_DIR/src/util/async.c:260:5
> +    #20 0x7fca6acededc in g_main_context_dispatch
> (/lib64/libglib-2.0.so.0+0x4fedc)
>
> Since there are also lots of these warnings in the output:
>
> +==24683==WARNING: ASan doesn't fully support makecontext/swapcontext
> functions and may produce false positives in some cases!
>
> ... I'm really not sure whether it makes sense to go bug hunting here.

I haven't ever seen anything that's really a false positive
as a result of those warnings. Someday I might go and try to
find out exactly what the makecontext/swapcontext issue is.

> OK. Anyway, since there are also these "WARNING: ASan doesn't fully
> support ..." messages in the output, it simply does not make sense to
> run the iotests automatically in this case, since the output of the
> tests gets spoiled and thus the tests are failing.

I guess you're checking both stdout and stderr, then?
I think it is possible to redirect sanitizer output to
some other file with by adding log_path=somefile to
ASAN_OPTIONS (it then writes to somefile.$PID) but
fishing out the results again would be really annoying,
so that sounds more trouble than it's worth. Just skipping
the iotests seems like a simple fix for now, and we can always
come back to this if/when we've tackled some of the leaks that
show up in the rest of the test suite.

thanks
-- PMM