Opened 2 years ago

Closed 2 years ago

Last modified 2 years ago

#9894 closed defect (invalid)

signal 11 (SIGSEGV), code 2 (SEGV_ACCERR) crash on Android 12 arm64 devices

Reported by: Taner Sener Owned by:
Priority: important Component: avcodec
Version: git-master Keywords: libx264 android arm64 crash
Cc: Taner Sener Blocked By:
Blocking: Reproduced by developer: no
Analyzed by developer: no

Description (last modified by Taner Sener)

ffmpeg cross-compiled with x264 for Android arm64 architecture crashes when it is run on Android 12 (Api Level 31) devices.

Don't know if it happens because of an invalid compile flag, setting, etc. Here is what we learned during our tests:

  • This case doesn't occur if x264 is built without asm.
  • Command used to reproduce the crash uses 792x1568 dimensions. Not all dimensions cause a crash. 792x1568 is one of the dimension that does.
  • Older Android devices or emulators e.g. Android 11 (Api Level 30) doesn't have this issue. Android 11 have some new changes regarding ARM Memory Tagging Extension (MTE). Maybe those changes are triggering it. See Tagged Pointers for those changes.

This is the ffmpeg command used to reproduce this case.

ffmpeg -v 9 -loglevel 99 -loop 1 -i pyramid.jpg -vf scale=792x1568 -c:v libx264 video.mp4

Full console output is available in console-output.txt.

Below you can see the report generated.

    Build fingerprint: 'google/sdk_gphone64_arm64/emulator64_arm64:12/SE1A.220630.001/8789670:userdebug/dev-keys'
    Revision: '0'
    ABI: 'arm64'
    Timestamp: 2022-08-28 11:31:51.574993028+0100
    Process uptime: 1s
    Cmdline: ./ffmpeg -v 9 -loglevel 99 -loop 1 -i pyramid.jpg -vf scale=792x1568 -c:v libx264 video.mp4
    pid: 31986, tid: 31986, name: ffmpeg  >>> ./ffmpeg <<<
    uid: 0
    tagged_addr_ctrl: 0000000000000001
    signal 11 (SIGSEGV), code 2 (SEGV_ACCERR), fault addr 0xb400007a2a46b000
        x0  b400007b5d718960  x1  b400007a2a420cd0  x2  b400007a2a425904  x3  b400007a2a433ee4
        x4  b400007a2a46aff4  x5  0000007fc9bbd8ec  x6  00000000fffffffa  x7  0000000000000032
        x8  b400007a2a4232b0  x9  b400007a2a4689b0  x10 0000007cdd03e738  x11 00000000000025e4
        x12 0000000000008000  x13 0000000000000031  x14 0000000000000061  x15 0000000000007fff
        x16 b400007a2a433e74  x17 b400007a2a3ff3a8  x18 0000007cee20c000  x19 0000000000000027
        x20 0000007fc9bbeb28  x21 0000000000000027  x22 00000000000012f2  x23 b400007ced0d6dd0
        x24 b400007a2a431890  x25 b400007a2aa45c60  x26 b400007a2a433e74  x27 b400007b5d718900
        x28 b400007a2a420c60  x29 0000007fc9bbd8f0
        lr  0000007cdcffad9c  sp  0000007fc9bbd8a0  pc  0000007cdd03e74c  pst 0000000080001000

    backtrace:
        #00 pc 000000000060c74c  /data/local/tmp/org.ffmpeg.test/libavcodec.so (x264_8_mbtree_propagate_cost_neon+20)
        #01 pc 00000000005c8d98  /data/local/tmp/org.ffmpeg.test/libavcodec.so (macroblock_tree_propagate+540)
        #02 pc 00000000005bd9b4  /data/local/tmp/org.ffmpeg.test/libavcodec.so (macroblock_tree+856)
        #03 pc 00000000005bcef0  /data/local/tmp/org.ffmpeg.test/libavcodec.so (x264_8_slicetype_analyse+2844)
        #04 pc 000000000060f4c4  /data/local/tmp/org.ffmpeg.test/libavcodec.so (x264_8_lookahead_get_frames+264)
        #05 pc 00000000005962ac  /data/local/tmp/org.ffmpeg.test/libavcodec.so (x264_8_encoder_encode+896)
        #06 pc 0000000000592360  /data/local/tmp/org.ffmpeg.test/libavcodec.so (x264_encoder_encode+16)
        #07 pc 000000000058c714  /data/local/tmp/org.ffmpeg.test/libavcodec.so (X264_frame+1272)
        #08 pc 00000000004838b8  /data/local/tmp/org.ffmpeg.test/libavcodec.so (ff_encode_encode_cb+36)
        #09 pc 0000000000483db0  /data/local/tmp/org.ffmpeg.test/libavcodec.so (encode_receive_packet_internal+340)
        #10 pc 0000000000483c10  /data/local/tmp/org.ffmpeg.test/libavcodec.so (avcodec_send_frame+516)
        #11 pc 0000000000035f14  /data/local/tmp/org.ffmpeg.test/ffmpeg (encode_frame+252)
        #12 pc 0000000000035950  /data/local/tmp/org.ffmpeg.test/ffmpeg (do_video_out+1840)
        #13 pc 0000000000034eec  /data/local/tmp/org.ffmpeg.test/ffmpeg (reap_filters+280)
        #14 pc 000000000003129c  /data/local/tmp/org.ffmpeg.test/ffmpeg (main+7160)
        #15 pc 00000000000488c8  /apex/com.android.runtime/lib64/bionic/libc.so (__libc_init+96) (BuildId: ba489d4985c0cf173209da67405662f9)

Steps we used to build and run ffmpeg and x264 can be found under the android-ffmpeg-x264-arm64-api-level-31 repository.

Also created a ticket on on x264 project since I don't know who should own this issue, https://code.videolan.org/videolan/x264/-/issues/53.

Attachments (3)

console-output.txt (271.0 KB ) - added by Taner Sener 2 years ago.
console output
tombstone.txt (169.6 KB ) - added by Taner Sener 2 years ago.
Tombstone generated for this vrash
disassembly.txt (38.7 KB ) - added by Taner Sener 2 years ago.
Disassembly information

Download all attachments as: .zip

Change History (13)

by Taner Sener, 2 years ago

Attachment: console-output.txt added

console output

comment:1 by Taner Sener, 2 years ago

Description: modified (diff)

comment:2 by Taner Sener, 2 years ago

Description: modified (diff)

comment:3 by Carl Eugen Hoyos, 2 years ago

Component: undeterminedavcodec
Keywords: libx264 crash added; x264 removed
Priority: normalimportant

This is missing disassembly and register dump as explained on https://ffmpeg.org/bugreports.html

Why do you think this is not a bug in x264?

by Taner Sener, 2 years ago

Attachment: tombstone.txt added

Tombstone generated for this vrash

in reply to:  3 comment:4 by Taner Sener, 2 years ago

Replying to Carl Eugen Hoyos:

This is missing disassembly and register dump as explained on https://ffmpeg.org/bugreports.html

Why do you think this is not a bug in x264?

Debugging a native Android process via gdb or lldb is not impossible but very difficult. I tried a few times but I wasn't successful. I attached the tombstone file instead. Does it include the disassembly we need? I don't know how disassembly looks like to be honest.

I'm not sure who to blame here, ffmpeg or x264. Yes, the crash happens inside x264 methods, but ffmpeg calls those methods. Maybe parameters passed by ffmpeg are wrong. And the crash is the result of that. For me, it's hard to say who should own this.

If ffmpeg devs could have a look at the crash and check if everything is looking good from ffmpeg perspective that would be quite helpful I believe.

by Taner Sener, 2 years ago

Attachment: disassembly.txt added

Disassembly information

comment:5 by Taner Sener, 2 years ago

See disassembly.txt for the disassembly information.

comment:6 by Balling, 2 years ago

If it crashes when you target Api 31 (which you would never do, since it will only work on Android 12, only OS apps do that, and even compile themselves into armv9) then this is a bug in NDK's assembler. If you compile on older NDK and it does not work on newer android, then this Android's bug. Anyway, report this to x264 gitlab or AOSP bugtracker.

Last edited 2 years ago by Balling (previous) (diff)

comment:7 by Taner Sener, 2 years ago

The crash is fixed on x264 side. They merged a fix to their master.

How should I close this ticket, with wontfix?

comment:8 by Carl Eugen Hoyos, 2 years ago

Resolution: invalid
Status: newclosed

Thank you for reporting back!

comment:10 by Balling, 2 years ago

This BTW, was key information "we are using Android NDK r25b, but we saw that this case can be reproduced with older NDK versions e.g. r24, r23c, r22b as well".

Note: See TracTickets for help on using tickets.