1582838 - geckoview_example fails on arm emulator

Status: NEW

(GeckoView :: General, defect, P3)

Description

[Michael Froman [:mjf]]

Building from moz-central b3ecb5aef45a on Ubuntu 18.04.

Steps To Reproduce:

1. ./mach bootstrap # select 4 for Android
2. Use unmodified suggested .mozconfig

# Build GeckoView/Firefox for Android:
ac_add_options --enable-application=mobile/android

# Targeting the following architecture.
# For regular phones, no --target is needed.
# For x86 emulators (and x86 devices, which are uncommon):
# ac_add_options --target=i686
# For newer phones.
# ac_add_options --target=aarch64
# For x86_64 emulators (and x86_64 devices, which are even less common):
# ac_add_options --target=x86_64

3. ./mach build
4. ./mach run

Emulator displays message: “Unfortunately, GeckoView Example has stopped."

Comment 1

[Michael Froman [:mjf]]

I just verified this also happens on macOS 10.14.6.

Comment 2

[Geoff Brown [:gbrown]]

I have the same experience.

I also checked, with :mjf's mozconfig and the 4.3 arm emulator:

• ./mach build
• adb install ./gradle/build/mobile/android/geckoview_example/outputs/apk/withGeckoBinaries/debug/geckoview_example-withGeckoBinaries-debug.apk
• run geckoview_example manually by clicking on it in the emulator
  ...crashes just the same.

Logcat shows:

W/dalvikvm( 2086): VFY: unable to resolve virtual method 16369: Landroid/view/ViewStructure;.setClassName (Ljava/lang/String;)V
D/dalvikvm( 2086): VFY: replacing opcode 0x6e at 0x0012
I/dalvikvm( 2086): Could not find method android.view.ViewStructure.setEnabled, referenced from method org.mozilla.geckoview.SessionTextInput.onProvideAutofillVirtualStructure
W/dalvikvm( 2086): VFY: unable to resolve virtual method 16371: Landroid/view/ViewStructure;.setEnabled (Z)V
D/dalvikvm( 2086): VFY: replacing opcode 0x6e at 0x0016
I/DEBUG   (  649): *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
I/DEBUG   (  649): Build fingerprint: 'generic/sdk/generic:4.3.1/JB_MR2/3462034:eng/test-keys'
I/DEBUG   (  649): Revision: '0'
I/DEBUG   (  649): pid: 2086, tid: 2102, name: Gecko  >>> org.mozilla.geckoview_example <<<
I/DEBUG   (  649): signal 4 (SIGILL), code 1 (ILL_ILLOPC), fault addr 99abaefa
I/DEBUG   (  649):     r0 00000000  r1 835e97d9  r2 00000000  r3 ffffffff
I/DEBUG   (  649):     r4 99b941c8  r5 b6feea06  r6 9abe7494  r7 99d9bc80
I/DEBUG   (  649):     r8 0000000a  r9 b6ff2650  sl 00000001  fp 99b90008
I/DEBUG   (  649):     ip 99d9bc18  sp 99d9bc78  lr 99ab7c9b  pc 99abaefa  cpsr 60000170
I/DEBUG   (  649):     d0  400000003eaaaaab  d1  3ff0000040200000
I/DEBUG   (  649):     d2  0000000000000000  d3  0000000000000000
I/DEBUG   (  649):     d4  406bc00000000000  d5  0000000000000000
I/DEBUG   (  649):     d6  0000000000000000  d7  4040000000000003
I/DEBUG   (  649):     d8  0000000000000000  d9  0000000000000000
I/DEBUG   (  649):     d10 0000000000000000  d11 0000000000000000
I/DEBUG   (  649):     d12 0000000000000000  d13 0000000000000000
I/DEBUG   (  649):     d14 0000000000000000  d15 0000000000000000
I/DEBUG   (  649):     scr 80000010
I/DEBUG   (  649): 
I/DEBUG   (  649): backtrace:
I/DEBUG   (  649):     #00  pc 00012efa  /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649):     #01  pc 0000fc97  /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649): 
I/DEBUG   (  649): stack:
I/DEBUG   (  649):          99d9bc38  00000000  
I/DEBUG   (  649):          99d9bc3c  40000004  
I/DEBUG   (  649):          99d9bc40  00000000  
I/DEBUG   (  649):          99d9bc44  b6fe2a41  /system/bin/linker
I/DEBUG   (  649):          99d9bc48  00000000  
I/DEBUG   (  649):          99d9bc4c  10000004  
I/DEBUG   (  649):          99d9bc50  00000000  
I/DEBUG   (  649):          99d9bc54  835e97d9  
I/DEBUG   (  649):          99d9bc58  0000000a  
I/DEBUG   (  649):          99d9bc5c  b6ff2650  /system/bin/linker
I/DEBUG   (  649):          99d9bc60  00000001  
I/DEBUG   (  649):          99d9bc64  99b941c8  
I/DEBUG   (  649):          99d9bc68  b6feea06  /system/bin/linker
I/DEBUG   (  649):          99d9bc6c  9abe7494  
I/DEBUG   (  649):          99d9bc70  df0027ad  
I/DEBUG   (  649):          99d9bc74  00000000  
I/DEBUG   (  649):     #00  99d9bc78  99b941c8  
I/DEBUG   (  649):          ........  ........
I/DEBUG   (  649):     #01  99d9bc78  99b941c8  
I/DEBUG   (  649):          99d9bc7c  9abe7494  
I/DEBUG   (  649):          99d9bc80  99d9bc90  [stack:2102]
I/DEBUG   (  649):          99d9bc84  99abb9a5  /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649):          99d9bc88  99abb995  /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649):          99d9bc8c  9abe7494  
I/DEBUG   (  649):          99d9bc90  b6feeac9  /system/bin/linker
I/DEBUG   (  649):          99d9bc94  b6fe4907  /system/bin/linker
I/DEBUG   (  649):          99d9bc98  9abe7494  
I/DEBUG   (  649):          99d9bc9c  b6feea06  /system/bin/linker
I/DEBUG   (  649):          99d9bca0  00000002  
I/DEBUG   (  649):          99d9bca4  00000004  
I/DEBUG   (  649):          99d9bca8  9abe7494  
I/DEBUG   (  649):          99d9bcac  b6fe49df  /system/bin/linker
I/DEBUG   (  649):          99d9bcb0  0000002a  
I/DEBUG   (  649):          99d9bcb4  b8bd5620  [heap]
I/DEBUG   (  649): 
...
I/DEBUG   (  649): memory map around fault addr 99abaefa:
I/DEBUG   (  649):     99a21000-99aa8000 r-- /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649):     99aa8000-99b5d000 r-x /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
I/DEBUG   (  649):     99b5d000-99b94000 rw- /data/app-lib/org.mozilla.geckoview_example-1/libmozglue.so
W/ActivityManager(  954):   Force finishing activity org.mozilla.geckoview_example/.GeckoViewActivity
D/Zygote  (  652): Process 2086 terminated by signal (4)

Comment 3

[Nick Alexander :nalexander [he/him]]

So, this is odd:

I/DEBUG   (  649): signal 4 (SIGILL), code 1 (ILL_ILLOPC), fault addr 99abaefa

We've definitely seen tickets like that in the past, usually due to compiler and/or NDK toolchain issues. I have no insight into this: if you really care to run on the ARM emulator, the GV folks will have to investigate further. You could try attaching the debugger (in Android Studio, or try to make it work following some WIP notes). If this reproduces with an artifact build, you might be able to use --enable-artifact-build-symbols to get the backtrace more easily. (Note that I witness an issue with full artifact build symbols locally where we don't pick builds with full build symbols properly. I haven't filed/fixed that yet.)

Comment 4

[Michael Froman [:mjf]]

For the record, I tried it on an artifact build from macOS and the problem is reproducible with the artifact build.

Comment 5

[Jan de Mooij [:jandem]]

This is blocking me too. I tried to investigate bug 1583907 (crashes on ARM when upgrading to Clang 9) but I can't get anything to run in the emulator.

Comment 6

[Geoff Brown [:gbrown]]

Running in emulator 29.2.1 on ubuntu 16.04 I tried installing geckoview_example.apk and geckoview-androidTest.apk from recent mozilla-central "Android 4.0 API16+" builds; they both consistently crashed on startup with SIGILL. I also tried installing a fennec build (target.apk) from esr68; it crashed on startup with SIGILL.

This seems particularly odd in that the esr68 fennec successfully runs tests in the arm emulator running android 4.3 in continuous integration. However, the arm emulator used in esr68 continuous integration is quite old (2017 or earlier).

Also, on mozilla-central, arm builds of geckoview_example are successfully run on bitbar Pixel 2 devices, for raptor tests. I even used the download link from a raptor test log to download a geckoview_example apk and installed that on my 4.3 emulator -- it crashed with SIGILL.

The 29.2.1 arm emulator appears to work in every other way: I can start Android apps like "Settings".

I think all of that points to some sort of incompatibility between our binaries and the 29.2.1 arm emulator.

Comment 7

[Geoff Brown [:gbrown]]

btw, I verified that mozilla-central x86_64 builds of geckoview_example install and run well on the x86 emulator ('mach android-emulator --version x86-7.0') -- this problem appears to be confined to arm builds on the arm emulator.

Comment 8

[Geoff Brown [:gbrown]]

Attachment: full logcat from local run of geckoview_example on emulator

Comment 9

[Geoff Brown [:gbrown]]

Attachment: tombstone from same run

Comment 10

[Geoff Brown [:gbrown]]

:glandium - Can you offer any insight?

Comment 11

[Chris Peterson [:cpeterson]]

Summary: this appears to be a regression in Android's ARM emulator version 29.2.1. Fennec and GeckoView ARM builds crash with SIGILL in ARM emulator 29.2.1, but they ran fine in old ARM emulator versions and the current x86 emulator.

What is the illegal instruction? Is this NEON code? Why would we have NEON code in mozglue?

Comment 12

[Chris Peterson [:cpeterson]]

lower priority because this is an ARM emulator regression, but most devs using ARM devices or x86 emulator. Is this bug Mac-specific?

Should we report this bug upstream?

Comment 13

[Geoff Brown [:gbrown]]

It is not Mac-specific: it fails just the same on Linux.

Comment 14

[Geoff Brown [:gbrown]]

I downgraded my emulator to 28.0.25.0, from https://dl.google.com/android/repository/emulator-linux-5395263.zip and then was able to run geckoview_example without any problem.

Comment 15

[Michael Froman [:mjf]]

(In reply to Geoff Brown [:gbrown] from comment #14)

I downgraded my emulator to 28.0.25.0, from https://dl.google.com/android/repository/emulator-linux-5395263.zip and then was able to run geckoview_example without any problem.

Should 'mach bootstrap' change to download 28.0.25.0 instead of 29.2.1? Seems a shame to have to go get a different version of the emulator after bootstrap.

Comment 16

[Geoff Brown [:gbrown]]

(In reply to Michael Froman [:mjf] from comment #15)

Should 'mach bootstrap' change to download 28.0.25.0 instead of 29.2.1?

Maybe? But on the whole, I think no. The 29.2.1 x86_64 emulator gives us no trouble, has important/relevant bug fixes, and is the version used by our tests in continuous integration. I think it may be more important to provide the same/best/latest version of the x86_64 emulator (which we encourage everyone to use whenever possible) than to provide a working arm emulator.