'

Debugging

From linuxwacom
Revision as of 04:07, 1 March 2011 by Favux (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Contents

This page provides an overview on how to debug issues with the driver.

Enabling in-driver debugging statements

The simplest way to debug the driver is to enable the debug statements in the driver itself. There are two types of debug statements that can be enabled separately with xsetwacom:

  • DebugLevel enables debug statements for code paths specific to a tool.
  • CommonDBG enables debug statements for code paths generic for all tools on a physical tablet.

Each type takes a debug levels between 0 (debugging off) and 12. The higher the level, the more debug messages you'll see. All debug messages are printed to the Xorg.log file.

Example debug output:

[241820.117] (II) Wacom Intuos4 6x9 pad (7:wcmSendEvents): [PAD] o_prox=false x=0 y=0 z=0 b=false b=0 tx=0 ty=0 wl=53 rot=-875 th=0
[241820.117] (II) Wacom Intuos4 6x9 pad (10:wcmRotateAndScaleCoordinates): rotate/scaled to 0/0
[241820.117] (II) Wacom Intuos4 6x9 pad (6:wcmSendEvents): abs prox=1	x=0	y=0	z=0	v3=0	v4=0	v5=53	id=15	serial=4294967295	button=false	buttons=0
[241820.117] (II) Wacom Intuos4 6x9 pad (6:wcmSendButtons): buttons=0

The output is in the format "device name (debug level:function): actual message".

Note: if the driver was built with --disable-debug, no debug messages can be printed.

Using gdb

In most cases, gdb is the fastest and most comprehensive way to find an issue with the driver. The learning curve of getting to know gdb quickly pays off. However, because X is a windowing system you need two machines to debug the driver. Otherwise, whenever you halt the driver, you also halt the X server and thus lose control over any windows. The second machine doesn't need to be fancy, you just need it to ssh into the machine running the driver. In all the commands below, it is assumed that you are executing them from a separate machine, ssh'd into the first machine. Also, you need root/sudo rights to debug with gdb as the X server runs as root.

This is not an exhaustive list of gdb commands but merely an overview of the basic commands needed to get information from the driver. Do spend the time to learn gdb, it pays off.

Requirements

The driver must be built with debugging symbols enabled. Make sure you have -ggdb -O0 in your CFLAGS at configure time. If you are running the distribution-provided driver, install the debuginfo packages provided by your distribution.

Attaching to a running process

In most cases, you will need to attach to the running X server. This can be done by providing the binary and the process ID (pid) of the X server to gdb. The binary is usually /usr/bin/Xorg, the process ID can be found with pidof Xorg (or pidof X on some distributions). Alternatively, you can use the ps command. The example below uses the shorthand notation to get the pid and pass it to gdb.

Warning: use ssh! Do not do try to attach to the running X server from within the same server!

$> sudo gdb /usr/bin/Xorg `pidof Xorg`
GNU gdb (GDB) Fedora (7.2-38.fc14)
Copyright (C) 2010 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-redhat-linux-gnu".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /usr/bin/Xorg...done.
Attaching to program: /usr/bin/Xorg, process 17306
Reading symbols from /lib64/libudev.so.0...(no debugging symbols found)...done.
Loaded symbols for /lib64/libudev.so.0
Reading symbols from /lib64/libgcrypt.so.11...(no debugging symbols found)...done.
Loaded symbols for /lib64/libgcrypt.so.11
Reading symbols from /lib64/libdl.so.2...(no debugging symbols found)...done.
Loaded symbols for /lib64/libdl.so.2
Reading symbols from /opt/xorg/lib/libpciaccess.so.0...done.
Loaded symbols for /opt/xorg/lib/libpciaccess.so.0
Reading symbols from /opt/xorg/lib/libpixman-1.so.0...done.
...
Reading symbols from /opt/xorg/lib/xorg/modules/input/synaptics_drv.so...done.
Loaded symbols for /opt/xorg/lib/xorg/modules/input/synaptics_drv.so
Reading symbols from /lib64/libnss_files.so.2...(no debugging symbols found)...done.
Loaded symbols for /lib64/libnss_files.so.2
0x00000032748d9073 in __select_nocancel () from /lib64/libc.so.6
Missing separate debuginfos, use: debuginfo-install expat-2.0.1-10.fc13.x86_64 freetype-2.4.2-4.fc14.x86_64 glibc-2.13-1.x86_64 libgcc-4.5.1-4.fc14.x86_64 libgcrypt-1.4.5-4.fc13.x86_64 libgpg-error-1.9-1.fc14.x86_64 libstdc++-4.5.1-4.fc14.x86_64 libtalloc-2.0.1-1.fc13.x86_64 libudev-161-8.fc14.x86_64 zlib-1.2.5-2.fc14.x86_64
(gdb)

You are now attached to the running X server, the process is halted and gdb awaits further instructions (you'll notice that the X server has stopped responding). When you're done with debugging, you simply detach from the process and quit gdb.

(gdb) det
Detaching from program: /usr/bin/Xorg, process 17306
(gdb) quit
$>

Managing the control flow

To continue the process, type continue. If you want to interrupt while the server is running, Ctrl+C brings you back to the prompt.

(gdb) continue
Continuing.
^C
Program received signal SIGINT, Interrupt.
0x00000032748d9073 in __select_nocancel () from /lib64/libc.so.6
(gdb)

Note that if the server is started through gdb (with the run command) instead of attached at runtime, Ctrl+C may not work. In this case, sending a SIGINT to the process interrupts it for you.

$> sudo kill -SIGNIT `pidof Xorg`


Breakpoints

Breakpoints can be set to any function or line in a file. Once set, gdb will automatically halt the process and provide you with the prompt whenever the control flow reaches a breakpoint.

(gdb) break wcmEvent
Breakpoint 1 at 0x7fcb8fe8c8d8: file ../src/wcmCommon.c, line 828.
(gdb) break wcmUSB.c:567
Breakpoint 2 at 0x7fcb8fe93382: file ../src/wcmUSB.c, line 567.
(gdb) info b
Num     Type           Disp Enb Address            What
1       breakpoint     keep y   0x00007fcb8fe8c8d8 in wcmEvent at ../src/wcmCommon.c:828
2       breakpoint     keep y   0x00007fcb8fe93382 in usbParse at ../src/wcmUSB.c:567
(gdb) continue
Continuing.

In the example above, we set two breakpoints: one at function wcmEvents() and one at a specific line in wcmUsb.c. Any call to wcmEvent() will not halt the server and provide you with a prompt.

Breakpoint 1, wcmEvent (common=0x2946b10, channel=0, pState=0x2946c80) at ../src/wcmCommon.c:828
828             int suppress = 0;
(gdb) continue 

Conditional breakpoints can be set on any value that is valid at the location of the breakpoint. For example, we may want to only stop at wcmEvent() if the channel is greater than 1.

(gdb) break wcmEvent if channel > 1
Note: breakpoint 1 also set at pc 0x7fcb8fe8c8d8.
Breakpoint 3 at 0x7fcb8fe8c8d8: file ../src/wcmCommon.c, line 828.
(gdb) disable 1
(gdb) continue

Note that the first breakpoint needs to be disabled, otherwise we'd still stop at every call to wcmEvent(). Disabling a breakpoint simply means that gdb will not stop there but continue on until the breakpoint is enabled. You can enable it at any time again.

Stepping through the code

Aside from the continue command that you already know, the four important commands to step through code are:

  • next: execute the current line and stop again at the next one.
  • step: step into the current function (if applicable)
  • until: step until the current block is finished (useful for quickly stepping over loops).
  • finish: continue until the function exists and then stop at the caller
Breakpoint 1, wcmEvent (common=0x2946b10, channel=0, pState=0x2946c80) at ../src/wcmCommon.c:828
828             int suppress = 0;
(gdb) n
829             WacomDevicePtr priv = common->wcmDevices;
(gdb) 
830             pChannel = common->wcmChannel + channel;
(gdb) 
831             pLast = &pChannel->valid.state;
(gdb) 
833             DBG(10, common, "channel = %d\n", channel);
(gdb) 
836             if (channel >= MAX_CHANNELS)

Note that by just hitting enter, gdb will re-execute the previous command. Below is an example for stepping into a function.

Breakpoint 4, usbParseEvent (pInfo=0x2938910, event=0x29487c8) at ../src/wcmUSB.c:715
715             WacomDevicePtr priv = (WacomDevicePtr)pInfo->private;
(gdb) n
716             WacomCommonPtr common = priv->common;
(gdb) 
717             wcmUSBData* private = common->private;
(gdb) 
719             DBG(10, common, "\n");
(gdb) 
726             if (private->wcmEventCnt >= ARRAY_SIZE(private->wcmEvents))
(gdb) 
735             private->wcmEvents[private->wcmEventCnt++] = *event;
(gdb) 
737             if (event->type == EV_MSC || event->type == EV_SYN)
(gdb) 
738                     usbParseSynEvent(pInfo, event);
(gdb) s
usbParseSynEvent (pInfo=0x2938910, event=0x29487c8) at ../src/wcmUSB.c:750
750             WacomDevicePtr priv = (WacomDevicePtr)pInfo->private;
(gdb) n
751             WacomCommonPtr common = priv->common;
(gdb) 
752             wcmUSBData* private = common->private;

Printing values and callstack information

The two important commands here are backtrace and print/display. A backtrace will give you a full list of the callstack, so you can check how we got to this particular line of the code. In the example below, you can see that we got to usbParseSynEvent() in response to a SIGIO signal handled by the server - which then called wacom's read function wcmDevReadInput(). You can also see the values of various parameters, great for debugging segfaults.

(gdb) backtrace
#0  usbParseSynEvent (pInfo=0x2938910, event=0x29487c8) at ../src/wcmUSB.c:801
#1  0x00007fcb8fe937c6 in usbParseEvent (pInfo=0x2938910, event=0x29487c8) at ../src/wcmUSB.c:738
#2  0x00007fcb8fe93378 in usbParse (pInfo=0x2938910, data=0x29487c8 "`\354YM", len=48) at ../src/wcmUSB.c:565
#3  0x00007fcb8fe8a271 in wcmReadPacket (pInfo=0x2938910) at ../src/xf86Wacom.c:812
#4  0x00007fcb8fe89f64 in wcmDevReadInput (pInfo=0x2938910) at ../src/xf86Wacom.c:762
#5  0x000000000048ab82 in xf86SigioReadInput (fd=15, closure=0x2938910) at xf86Events.c:298
#6  0x00000000005936b1 in xf86SIGIO (sig=29) at ./../shared/sigio.c:109
#7  <signal handler called>
#8  0x00000032748d9073 in __select_nocancel () from /lib64/libc.so.6
#9  0x0000000000475138 in WaitForSomething (pClientsReady=0x29c2470) at WaitFor.c:232
#10 0x00000000004291f1 in Dispatch () at dispatch.c:367
#11 0x00000000004216ab in main (argc=8, argv=0x7fffa07de6d8, envp=0x7fffa07de720) at main.c:287

Now, let's assume we want to know what's actually in pInfo and event. We use print for that:

(gdb) p pInfo->name
$1 = 0x2939800 "Wacom Intuos4 6x9 stylus"
(gdb) p event->value
$2 = -1719661677
(gdb) p event->code
$3 = 0

The display command does the same but it keeps displaying the values every time you get a prompt (you can then remove this with undisplay).

How to use all this information?

The usual way to narrow down a bug is to attach the server, trigger the bug and then inspect what values led to the bug to trigger. Then work backwards and inspect the values before the bug is triggered. As you work backwards you will eventually hit a point where a value assignment is incorrect. Congratulations, you have now found the bug! Now it's just a matter of fixing it.

Note: while gdb can speed up triaging tremendously, it is not uncommon for bugs to take several days to identify.

Personal tools
Namespaces

Variants
Actions
Navigation
Toolbox