kgdb(1) ... is it broken ?

[John Baldwin]

On Friday 23 February 2007 06:34, Kostik Belousov wrote:
> On Fri, Feb 23, 2007 at 03:18:23PM +0900, Wilkinson, Alex wrote:
> > Hi all,
> > 
> > I have a reasonably recent version of current that is panic'ing at least 
once
> > every 2 days. When I run kgdb(1) to do a backtrace it aint working 
correctly.
> > 
> > [FreeBSD 7.0-CURRENT #0: Wed Jan 24 14:24:54 WST 2007]
> > 
> > e.g.
> > 
> > The panic:
> > 
> > 	NVRM: Xid (0001:00): 8, Channel 00000000
> > 	panic: Bad link elm 0xc4dc8900 next->prev != elm
> > 	cpuid = 0
> > 	KDB: enter: panic
> > 	[thread pid 909 tid 100080 ]
> > 	Stopped at      kdb_enter+0x32: leave
> > 	db>tr
> > 	Tracing pid 909 tid 100080 td 0xc47231b0
> > 	kdb_enter(c09ecabf,0,c09a4b15,e6a69a20,c47231b0,...) at kdb_enter+0x32
> > 	panic(c09a4b15,c4dc8900,4c,c09e8778,64,...) at panic+0x191
> > 	destroy_devl(c4714e80,e6a69a70,c0fe6cf0,c4dc8900,40,...) at 
destroy_devl+0x330
> > 	destroy_dev(c4dc8900,40,c47231b0,0,c4dc8900,...) at destroy_dev+0x13
> > 	nvidia_dev_close(c4dc8900,3,2000,c47231b0,c4e287d8,...) at 
nvidia_dev_close+0xa4
> > 	
> > 	giant_close(c4dc8900,3,2000,c47231b0,e6a69adc,...) at giant_close+0x4f
> > 	devfs_close(e6a69b28,3,c4e28754) at devfs_close+0x2d1
> > 	VOP_CLOSE_APV(c0a8de20,e6a69b28,c47231b0,c09f7b4c,11f,...) at 
VOP_CLOSE_APV+0x69
> > 	
> > 	vn_close(c4e28754,3,c4306a80,c47231b0,203246,...) at vn_close+0x99
> > 	vn_closefile(c4bf0a20,c47231b0,c09e9165,889,c4e28754,...) at 
vn_closefile+0x88
> > 	
fdrop_locked(c4bf0a20,c47231b0,2,c09ee59f,de,c47231b0,0,203246,c0b3b920,e6a69c24
> 
> 	,c07517fb,c0af5494,0,c4b3522c,401,c09e9165,e6a69c4c,c0716a82,c4b3522c,1,c09ebc01
> > 	,ae,0) at fdrop_locked+0xb9
> > 	closef(c4bf0a20,c47231b0,c09e9165,401,c0739bd6,...) at closef+0x1f4
> > 	kern_close(c47231b0,e,4,c4b346c0,1,...) at kern_close+0x188
> > 	syscall(e6a69d38) at syscall+0x155
> > 	Xint0x80_syscall() at Xint0x80_syscall+0x20
> > 	--- syscall (0, FreeBSD ELF32, nosys), eip = 0x2, esp = 0x203292, ebp = 
0xc1d000
> > 	01 ---
> > 	MAXCPU(4000000,90ffff00,10c19ee7,58c28e8c,34c22fbb,...) at 0x2
> > 	db>panic
> > 	panic: from debugger
> > 	cpuid = 0
> > 	Uptime: 3d5h29m19s
> > 	Physical memory: 1007 MB
> > 	Dumping 219 MB: 204 188 172 156 140 124 108 92 76 60 44 28 12
> > 	Dump complete
> > 
> > Upon a reboot I see this error:
> > 
> > 	savecore: reboot after panic: Bad link elm 0xc4dc8900 next->prev != elm
> > 	Feb 23 15:02:22 obelix savecore: reboot after panic: Bad link elm 
0xc4dc8900 next->prev != elm
> > 
> > And then the backtrace:
> > 
> > 	#0  doadump () at pcpu.h:166
> > 	166     pcpu.h: No such file or directory.
> > 	        in pcpu.h
> > 	(kgdb) where
> > 	#0  doadump () at pcpu.h:166
> > 	#1  0xc0720c1b in boot (howto=260) 
at /usr/src/sys/kern/kern_shutdown.c:411
> > 	#2  0xc0720693 in panic (fmt=0xc09ab848 "from debugger") at
> > 	/usr/src/sys/kern/kern_shutdown.c:567
> > 	#3  0xc047e490 in db_panic (addr=-1066121253, have_addr=0, count=-1,
> > 	modif=0xe6a69810 "") at /usr/src/sys/ddb/db_command.c:433
> > 	#4  0xc047e870 in db_command_loop () at /usr/src/sys/ddb/db_command.c:401
> > 	#5  0xc04805fb in db_trap (type=3, code=0) 
at /usr/src/sys/ddb/db_main.c:222
> > 	#6  0xc0744c19 in kdb_trap (type=0, code=0, tf=0xe6a699a4) at
> > 	/usr/src/sys/kern/subr_kdb.c:502
> > 	#7  0xc0960ea5 in trap (frame=0xe6a699a4) 
at /usr/src/sys/i386/i386/trap.c:621
> > 	#8  0xc0948dbb in calltrap () at /usr/src/sys/i386/i386/exception.s:139
> > 	#9  0x00000000 in ?? ()
> > 	(kgdb) 
> > 
> > Things just aint working as per normal.
> > 
> > Has anyone had problems with running backtraces of kernel core dumps with 
kgdb(1) ?
> 
> Try this patch, it shall allow to see useful backtrace in kgdb (I really
> like to receive feedback on this one):
> 
> Index: gnu/usr.bin/gdb/kgdb/trgt_i386.c
> ===================================================================
> RCS file: /usr/local/arch/ncvs/src/gnu/usr.bin/gdb/kgdb/trgt_i386.c,v
> retrieving revision 1.5
> diff -u -r1.5 trgt_i386.c
> --- gnu/usr.bin/gdb/kgdb/trgt_i386.c	11 Sep 2005 05:36:30 -0000	1.5
> +++ gnu/usr.bin/gdb/kgdb/trgt_i386.c	23 Feb 2007 11:31:39 -0000
> @@ -146,7 +146,7 @@
>  	*realnump = -1;
>  
>  	ofs = (regnum >= I386_EAX_REGNUM && regnum <= I386_FS_REGNUM)
> -	    ? kgdb_trgt_frame_offset[regnum] : -1;
> +	    ? kgdb_trgt_frame_offset[regnum] + 4 : -1;
>  	if (ofs == -1)
>  		return;

You can make the patch by dependent on the kern.osreldate (__FreeBSD_version)
which is accesible as the global var 'osreldate' in the kernel and use the
old offset for kernels before Kip's change so it works for both old and new.

-- 
John Baldwin