patrickcusack / libyuv

Automatically exported from code.google.com/p/libyuv
BSD 3-Clause "New" or "Revised" License
0 stars 0 forks source link

Unitialized read in CopyRow_AVX introduced between r1038 and r1130 #377

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
What steps will reproduce the problem?
1. Sync WebRTC r7489
2. set GYP_DEFINES=build_for_tool=drmemory
   gclient runhooks
3. Compile libjingle_media_unittest in Debug mode
4. Run the test with: tools\valgrind-webrtc\webrtc_tests.bat --build_dir 
src/out/Debug --test common_audio_unittests --tool drmemory_full ""

What is the expected output? What do you see instead?
Test should pass. Instead it fails like this:
http://build.chromium.org/p/client.webrtc/builders/Win%20DrMemory%20Full/builds/
1205/steps/libjingle_media_unittest/logs/stdio

UNINITIALIZED READ: reading 0x04318ea0-0x04318ec0 32 byte(s) within 
0x04318ea0-0x04318ec0
#0 CopyRow_AVX                 
#1 CopyPlane                   
#2 I420Copy                    
#3 I420Rotate                  
#4 ConvertToI420               
#5 cricket::WebRtcVideoFrame::Reset
#6 cricket::WebRtcVideoFrame::Init
#7 WebRtcVideoFrameTest::TestInit
#8 WebRtcVideoFrameTest_InitEvenSize_Test::TestBody
#9 testing::internal::HandleSehExceptionsInMethodIfSupported<>
Note: @0:02:50.174 in thread 1620
Note: instruction: vmovdqu (%eax) -> %ymm0
Suppression (error hash=#5E132DBC3BD73684#):
  For more info on using suppressions see http://dev.chromium.org/developers/how-tos/using-drmemory#TOC-Suppressing-error-reports-from-the-
{
UNINITIALIZED READ
name=<insert_a_suppression_name_here>
*!CopyRow_AVX
*!CopyPlane
*!I420Copy
*!I420Rotate
*!ConvertToI420
*!cricket::WebRtcVideoFrame::Reset
*!cricket::WebRtcVideoFrame::Init
*!WebRtcVideoFrameTest::TestInit
*!WebRtcVideoFrameTest_InitEvenSize_Test::TestBody
*!testing::internal::HandleSehExceptionsInMethodIfSupported<>
}

What version of the product are you using? On what operating system?
It started failing when we rolled Chromium revision (and therefore also libyuv 
1038:1130 in https://code.google.com/p/webrtc/source/detail?r=7489

Please provide any additional information below.

Original issue reported on code.google.com by kjellan...@google.com on 22 Oct 2014 at 1:44

GoogleCodeExporter commented 9 years ago
I'm wondering whether this is DrMemory being overly aggressive, I'm not sure 
that vmovdqu should trigger an uninitialized read, it's just copying stuff, not 
acting on it. I've asked zhaoqin@ to take a look, but I can't +cc people here.

Original comment by pbos@google.com on 22 Oct 2014 at 2:00

GoogleCodeExporter commented 9 years ago
Unfortunately, Dr. Memory does not yet propagate shadow values for ymm 
registers (covered by https://code.google.com/p/drmemory/issues/detail?id=1485).

Original comment by bruen...@chromium.org on 22 Oct 2014 at 3:15

GoogleCodeExporter commented 9 years ago
The propagation tracking is not completely finished for multi-media 
instructions, so there are some instructions doing the uninit-check on first 
touching it instead of first significant use. Will look into that.

Original comment by zhao...@google.com on 22 Oct 2014 at 3:17

GoogleCodeExporter commented 9 years ago
This needs to be tested on a CPU with AVX2 I assume.
For purposes of Dr Memory, you can turn off AVX2 with compile flags or 
environment variables.
set LIBYUV_DISABLE_AVX2=1
set LIBYUV_DISABLE_AVX=1

Original comment by fbarch...@google.com on 22 Oct 2014 at 6:55

GoogleCodeExporter commented 9 years ago
As the bug appears to be in DrMemory not libyuv, putting this as 'New', not 
accepted.

Original comment by fbarch...@google.com on 22 Oct 2014 at 6:56

GoogleCodeExporter commented 9 years ago
As the issue is in Dr Memory, I'd like to close the libyuv bug

Original comment by fbarch...@google.com on 9 Feb 2015 at 7:46

GoogleCodeExporter commented 9 years ago
FYI- the bug is real but not in YUV. The test case itself allocates memory but 
do not initalize it before calling libyuv.

Fixed by magjed in https://webrtc-codereview.appspot.com/41029004/.
I will remove the suppression.

Original comment by perkj@webrtc.org on 3 Mar 2015 at 9:33