views:

488

answers:

3

I'm developing an iPhone application. I have an EXC_BAD_ACCESS that occurs only in the release target; when I build the debug target the exception does not occur. However, when I set the NSZombieEnabled environment variable to YES, I still get the EXC_BAD_ACCESS with no further information. Is it even possible for NSZombieEnabled to work when executing the release target? I don't see why not, since gdb is running in both cases...

Update: here is a printout of the top of the stack:

#0  0x33369ebc in objc_msgSend ()
#1  0x3144f968 in -[EAInputStream _streamEventTrigger] ()
#2  0x3144fe78 in __streamEventTrigger ()
#3  0x338ae3a6 in CFRunLoopRunSpecific ()
#4  0x338adc1e in CFRunLoopRunInMode ()
#5  0x32ed6966 in -[NSRunLoop(NSRunLoop) runMode:beforeDate:] ()
#6  0x00005b06 in -[IOStreamDelegate removeMsg:] (self=0x142cc0, _cmd=<value temporarily unavailable, due to optimizations>, message=0x2fffe544) at /Users/robertmoretti/Documents/XXXXXXX/IOStreamDelegate.m:191

Here is a gdb session from inside the objc_msgSend call at the top:

(gdb) p/x $r0
$6 = 0x3100000
(gdb) x/s $r1
0x32d7cff8:  "release"
(gdb) disassemble $pc
Dump of assembler code for function objc_msgSend:
0x33369ea8 <objc_msgSend+0>:    teq r0, #0  ; 0x0
0x33369eac <objc_msgSend+4>:    moveq   r1, #0  ; 0x0
0x33369eb0 <objc_msgSend+8>:    bxeq    lr
0x33369eb4 <objc_msgSend+12>:   push    {r3, r4, r5, r6}
0x33369eb8 <objc_msgSend+16>:   ldr r4, [r0]
0x33369ebc <objc_msgSend+20>:   ldr r5, [r4, #8]
0x33369ec0 <objc_msgSend+24>:   ldr r6, [r5]
0x33369ec4 <objc_msgSend+28>:   add r3, r5, #8  ; 0x8
0x33369ec8 <objc_msgSend+32>:   and r5, r6, r1, lsr #2
0x33369ecc <objc_msgSend+36>:   ldr r4, [r3, r5, lsl #2]
0x33369ed0 <objc_msgSend+40>:   teq r4, #0  ; 0x0
0x33369ed4 <objc_msgSend+44>:   add r5, r5, #1  ; 0x1
0x33369ed8 <objc_msgSend+48>:   beq 0x33369efc <objc_msgSend+84>
0x33369edc <objc_msgSend+52>:   ldr r12, [r4]
0x33369ee0 <objc_msgSend+56>:   teq r1, r12
0x33369ee4 <objc_msgSend+60>:   and r5, r5, r6
0x33369ee8 <objc_msgSend+64>:   bne 0x33369ecc <objc_msgSend+36>
0x33369eec <objc_msgSend+68>:   ldr r12, [r4, #8]
0x33369ef0 <objc_msgSend+72>:   teq r4, r4
0x33369ef4 <objc_msgSend+76>:   pop {r3, r4, r5, r6}
0x33369ef8 <objc_msgSend+80>:   bx  r12
0x33369efc <objc_msgSend+84>:   pop {r3, r4, r5, r6}
0x33369f00 <objc_msgSend+88>:   b   0x33369f04 <objc_msgSend_uncached>
End of assembler dump.
(gdb) p/x *$r0
$9 = 0x0

The exception occurs on the line 0x33369ebc <objc_msgSend+20>: ldr r5, [r4, #8]. r4 has just been given the value pointed to by r0, which happens to be 0. I am wondering what is supposed to be in the memory region of 0x3100000. Here's a memory dump of that area:

(gdb) x/256w 0x3100000
0x3100000:  0x00000000  0x0000a293  0xaa650505  0x00000000
0x3100010:  0x0000a294  0xaa670505  0x00000000  0x0000a295
0x3100020:  0xaa690505  0x00000000  0x0000a296  0xaa6b0505
0x3100030:  0x00000000  0x0000a297  0xaa6d0505  0x00000000
0x3100040:  0x0000a298  0xaa6f0505  0x00000000  0x0000a299
0x3100050:  0xaa710505  0x00000000  0x0000a29a  0xaa730505
0x3100060:  0x00000000  0x0000a29b  0xaa750505  0x00000000
0x3100070:  0x0000a29c  0xaa770505  0x00000000  0x0000a29d
0x3100080:  0xaa790505  0x00000000  0x0000a29e  0xaa7b0505
0x3100090:  0x00000000  0x0000a29f  0xaa7d0505  0x00000000
0x31000a0:  0x0000a2a0  0xaa7f0505  0x00000000  0x0000a2a1
0x31000b0:  0xaa810505  0x00000000  0x0000a2a2  0xaa830505
0x31000c0:  0x00000000  0x0000a2a3  0xaa850505  0x00000000
0x31000d0:  0x0000a2a4  0xaa870505  0x00000000  0x0000a2a5
0x31000e0:  0xaa890505  0x00000000  0x0000a2a6  0xaa8b0505
0x31000f0:  0x00000000  0x0000a2a7  0xaa8d0505  0x00000000
0x3100100:  0x0000a2a8  0xaa8f0505  0x00000000  0x0000a2a9
0x3100110:  0xaa910505  0x00000000  0x0000a2aa  0xaa930505
0x3100120:  0x00000000  0x0000a2ab  0xaa950505  0x00000000
0x3100130:  0x0000a2ac  0xaa970505  0x00000000  0x0000a2ad
0x3100140:  0xaa990505  0x00000000  0x0000a2ae  0xaa9b0505
0x3100150:  0x00000000  0x0000a2af  0xaa9d0505  0x00000000
0x3100160:  0x0000a2b0  0xaa9f0505  0x00000000  0x0000a2b1
0x3100170:  0xaaa10505  0x00000000  0x0000a2b2  0xaaa30505
0x3100180:  0x00000000  0x0000a2b3  0xaaa50505  0x00000000
0x3100190:  0x0000a2b4  0xaaa70505  0x00000000  0x0000a2b5
0x31001a0:  0xaaa90505  0x00000000  0x0000a2b6  0xaaab0505
0x31001b0:  0x00000000  0x0000a2b7  0xaaad0505  0x00000000
0x31001c0:  0x0000a2b8  0xaaaf0505  0x00000000  0x0000a2b9
0x31001d0:  0xaab10505  0x00000000  0x0000a2ba  0xaab30505
0x31001e0:  0x00000000  0x0000a2bb  0xaab50505  0x00000000
0x31001f0:  0x0000a2bc  0xaab70505  0x00000000  0x0000a2bd
0x3100200:  0xaab90505  0x00000000  0x0000a2be  0xaabb0505
0x3100210:  0x00000000  0x0000a2bf  0xaabd0505  0x00000000
0x3100220:  0x0000a2c0  0xaabf0505  0x00000000  0x0000a2c1
0x3100230:  0xaac10505  0x00000000  0x0000a2c2  0xaac30505
0x3100240:  0x00000000  0x0000a2c3  0xaac50505  0x00000000
0x3100250:  0x0000a2c4  0xaac70505  0x00000000  0x0000a2c5
0x3100260:  0xaac90505  0x00000000  0x0000a2c6  0xaacb0505
0x3100270:  0x00000000  0x0000a2c7  0xaacd0505  0x00000000
0x3100280:  0x0000a2c8  0xaacf0505  0x00000000  0x0000a2c9
0x3100290:  0xaad10505  0x00000000  0x0000a2ca  0xaad30505
0x31002a0:  0x00000000  0x0000a2cb  0xaad50505  0x00000000
0x31002b0:  0x0000a2cc  0xaad70505  0x00000000  0x0000a2cd
0x31002c0:  0xaad90505  0x00000000  0x0000a2ce  0xaadb0505
0x31002d0:  0x00000000  0x0000a2cf  0xaadd0505  0x00000000
0x31002e0:  0x0000a2d0  0xaadf0505  0x00000000  0x0000a2d1
0x31002f0:  0xaae10505  0x00000000  0x0000a2d2  0xaae30505
0x3100300:  0x00000000  0x0000a2d3  0xaae50505  0x00000000
0x3100310:  0x0000a2d4  0xaae70505  0x00000000  0x0000a2d5
0x3100320:  0xaae90505  0x00000000  0x0000a2d6  0xaaeb0505
0x3100330:  0x00000000  0x0000a2d7  0xaaed0505  0x00000000
0x3100340:  0x0000a2d8  0xaaef0505  0x00000000  0x0000a2d9
0x3100350:  0xaaf10505  0x00000000  0x0000a2da  0xaaf30505
0x3100360:  0x00000000  0x0000a2db  0xaaf50505  0x00000000
0x3100370:  0x0000a2dc  0xaaf70505  0x00000000  0x0000a2dd
0x3100380:  0xaaf90505  0x00000000  0x0000a2de  0xaafb0505
0x3100390:  0x00000000  0x0000a2df  0xaafd0505  0x00000000
0x31003a0:  0x0000a2e0  0xab050505  0x00000000  0x0000a2e1
0x31003b0:  0xab070505  0x00000000  0x0000a2e2  0xab090505
0x31003c0:  0x00000000  0x0000a2e3  0xab0b0505  0x00000000
0x31003d0:  0x0000a2e4  0xab0d0505  0x00000000  0x0000a2e5
0x31003e0:  0xab0f0505  0x00000000  0x0000a2e6  0xab110505
0x31003f0:  0x00000000  0x0000a2e7  0xab130505  0x00000000

I don't really know what else to try; hopefully someone with more iphone experience will be able to recognize this memory as something meaningful.

Update 2: I just discovered that the problem only occurs when compiling with -O2, -O3, and -Os. Not sure what that implies.

A: 

When u enable NSZombieEnabled you should check the console as it will probably tell u which line you're getting the EXC_BAD_ACCESS. From there you would be able to fix it. Don't forget to disable it for final release.

thary
The console tells me nothing, unfortunately; just that it got an exception. Then it just jumps to gdb
Bobby Moretti
+1  A: 

In objc_msgSend, r0 contains a pointer to the receiver of the message. Your app is crashing when it tries to load something pointed to by the receiver. The most likely explanation is that you are either messaging an object that hasn't been initialized, or an object that has already been freed, but there are other possibilities.

You also may want to have a look at "So you crashed in objc_msgSend"; it's written about the Intel OS X platform, but much of it also applies to the iPhone OS.

Stephen Canon
Thanks. I'd already read the post you linked; that's how I learned to look at `$r0` and `$r1`. The address in `$r0` being `0x3100000` is highly suspect to me. Given all the zeroes, I don't think it's very likely that it was ever a valid pointer to an object.
Bobby Moretti
+1  A: 

Based on your gdb session, it seems possible that you're over-releasing some object, although I agree that the 0x3100000 seems improbable. Have you tried running Build and Analyze? Static analysis can catch many common over-release bugs.

Based on that memory address, a buffer overrun seems more likely. This could also explain why it only crashes in release mode when you're compiling with -O2 or greater. The optimizations may be laying out your memory in such a way that your buffer overrun happens to cause a crash in release mode but not in debug mode.

Are you doing anything funky with old school C void * pointers? Working with any null terminated C strings? Doing any pointer arithmetic?

cduhn
A buffer overrun seems very likely. When the crash occurs, I am in the middle of doing a read from an external accessory. It always happens after a certain number of reads (I forget the exact number... something like 60). A lot of my low level packet manipulation code uses C-style buffers, so there is definitely the possibility of a buffer overrun there. Thanks for the idea!
Bobby Moretti
Did you find it?
cduhn