Skip to contentSkip to conversation

Announce: kdb v4.4 is available for kernel 2.6.16

DISCUSSION Keith Owens 4 participants 6 messages Mar 21, 2006 – Apr 2, 2006

lore
Share & tools
lore
#0 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

KDB (Linux Kernel Debugger) has been updated for kernel 2.6.16.

ftp://oss.sgi.com/projects/kdb/download/v4.4/ ftp://ftp.ocs.com.au/pub/mirrors/oss.sgi.com/projects/kdb/download/v4.4/

Note: Due to a spam attack, the [email protected] mailing list is now subscriber only. If you reply to this mail, you may wish to trim [email protected] from the cc: list.

Thanks to Nathan Scott for updating KDB while I was on holiday.

Current versions are :-

kdb-v4.4-2.6.16-common-1.bz2
kdb-v4.4-2.6.16-i386-1.bz2
kdb-v4.4-2.6.16-ia64-1.bz2

Changelog extract since kdb-v4.4-2.6.15-common-1.

2006-03-21 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-common-1.

2006-03-14 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc6-common-1.

2006-02-28 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc5-common-1.

2006-02-20 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc4-common-1.

2006-02-06 Keith Owens <[email protected]>

* Change CONFIG_CRASH_DUMP to CONFIG_LKCD.
* Remove obsolete kdb_notifier_list.
* kdb v4.4-2.6.16-rc2-common-2.

2006-02-06 Keith Owens <[email protected]>

* Add xpcusers command. Dean Nelson, SGI.
* kdb v4.4-2.6.16-rc2-common-1.

2006-02-02 Keith Owens <[email protected]>

* Check if we have a console before using it for KDB.
* kdb v4.4-2.6.16-rc1-common-3.

2006-02-01 Keith Owens <[email protected]>

* Add option 'R' to the pid command to reset to the original task.
* Include 'pid R' in archkdb* commands to reset up the original failing task. Users may have switched to other cpus and/or tasks before issuing archkdb.
* Compile fix for kdbm_pg.c on i386.
* kdb v4.4-2.6.16-rc1-common-2.

2006-01-18 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-rc1-common-1.

2006-01-11 Keith Owens <[email protected]>

* Plug a timing race between KDB_ENTER_SLAVE and KDB_ENTER, and allow the cpu command to switch to a slave cpu.
* KDB_REASON_SILENT now waits for other cpus, to avoid spurious NMI events that were seen on some Xeon systems.
* kdb v4.4-2.6.15-common-3.

2006-01-08 Keith Owens <[email protected]>

* kdb mainline invokes DIE_KDEBUG_ENTER and DIE_KDEBUG_LEAVE via notify_die.
* Move xpc debug support from xpc to mainline kdb.
* kdbm_cm.c: check if file_lock_operations or lock_manager_operations are set before dereferencing them. Felix Blyakher, SGI.
* kdb v4.4-2.6.15-common-2.

Changelog extract since kdb-v4.4-2.6.15-i386-1.

2006-03-21 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-i386-1.

2006-03-14 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc6-i386-1.

2006-02-28 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc5-i386-1.

2006-02-20 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc4-i386-1.

2006-02-06 Keith Owens <[email protected]>

* Change CONFIG_CRASH_DUMP to CONFIG_LKCD.
* kdb v4.4-2.6.16-rc2-i386-2.

2006-02-06 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-rc2-i386-1.

2006-01-18 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-rc1-i386-1.

2006-01-08 Keith Owens <[email protected]>

* Add DIE_KDEBUG_ENTER and DIE_KDEBUG_LEAVE to notify_die.
* kdb v4.4-2.6.15-i386-2.

Changelog extract since kdb v4.4-2.6.15-ia64-1.

2006-03-21 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-ia64-1.

2006-03-14 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc6-ia64-1.

2006-02-28 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc5-ia64-1.

2006-02-20 Nathan Scott <[email protected]>

* kdb v4.4-2.6.16-rc4-ia64-1.

2006-02-07 Keith Owens <[email protected]>

* Change kdb_running_process_save from a static array to a pointer. gcc 4.0 objects to forward declarations for arrays with an incomplete type.
* kdb v4.4-2.6.16-rc2-ia64-3.

2006-02-06 Keith Owens <[email protected]>

* Change CONFIG_CRASH_DUMP to CONFIG_LKCD.
* kdb v4.4-2.6.16-rc2-ia64-2.

2006-02-06 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-rc2-ia64-1.

2006-02-01 Keith Owens <[email protected]>

* Handlers: check that the task is in kernel space before looking at the thread_info bits.
* Expose kdb_running_process_save[] so 'pid R' can get the original process, even when the MCA/INIT handlers are being used.
* kdb v4.4-2.6.16-rc1-ia64-3.

2006-01-19 Keith Owens <[email protected]>

* Add back some kdb changes to xpc_main that were lost due to a patch conflict.
* kdb v4.4-2.6.16-rc1-ia64-2.

2006-01-18 Keith Owens <[email protected]>

* kdb v4.4-2.6.16-rc1-ia64-1.

2006-01-10 Keith Owens <[email protected]>

* Build kdba_pod for generic as well as sn2 kernels and test at run time if the platform is sn2.
* kdb v4.4-2.6.15-ia64-3.

2006-01-08 Keith Owens <[email protected]>

* Convert xpc to use DIE_KDEBUG_ENTER and DIE_KDEBUG_LEAVE.
* Add debug option for xpc.
* break.b always sets a debug trap number of 0 , so pass that to kdb as well as the normal kdb traaps.
* kdb v4.4-2.6.15-ia64-2.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Exmh version 2.1.1 10/15/1999

iD8DBQFEH5Y8i4UHNye0ZOoRAgxfAKDEFyGyiC0MqpZf3g++ZcyWnkZm4QCfdGpB vrKA3BN4aKCI/MDX6bJTme0ípm
-----END PGP SIGNATURE-----

#1 lore raw
Quoted 11 lines · Keith Owens (on Tue, 21 Mar 2006 16:59:24 +1100) wrote:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

KDB (Linux Kernel Debugger) has been updated for kernel 2.6.16.

ftp://oss.sgi.com/projects/kdb/download/v4.4/ ftp://ftp.ocs.com.au/pub/mirrors/oss.sgi.com/projects/kdb/download/v4.4/

Note: Due to a spam attack, the [email protected] mailing list is now subscriber only. If you reply to this mail, you may wish to trim [email protected] from the cc: list.

Updates to kdb v4.4-2.6.16-{common,i386,ia64}-1, moving to -2. Mainly to get more forceful with stuck cpus on IA64, where the "non-maskable" interrupt is masked by local_irq_disable().

We finally get decent backtraces on stuck IA64 cpus. It took a long time to get there, KDB could not change until the base kernel could cope with INIT and recover correctly.

2006-03-22 Keith Owens <[email protected]>

* Add some more xpc flags. Dean Nelson, SGI.
* Replace open coded counter references with atomic_read().
* Pass early_uart_console to early_uart_setup(). Francois Wellenreiter, Bull.
* Replace open code with for_each_online_cpu().
* If cpus do not come into kdb after a few seconds then let architectures send a more forceful interrupt.
* Close a timing race with KDB_ENTER_SLAVE.
* kdb v4.4-2.6.16-common-2.

2006-03-24 Keith Owens <[email protected]>

* Define a dummy kdba_wait_for_cpus().
* kdb v4.4-2.6.16-i386-2.

2006-03-24 Keith Owens <[email protected]>

* Use INIT to interrupt cpus that do not respond to a normal kdb IPI.
* Remove KDBA_MCA_TRACE from arch/ia64/kernel/mca.c.
* kdb v4.4-2.6.16-ia64-2.

#2 lore raw
In reply to Keith Owens
Quoted 12 lines · Keith Owens

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

KDB (Linux Kernel Debugger) has been updated for kernel 2.6.16.

ftp://oss.sgi.com/projects/kdb/download/v4.4/ ftp://ftp.ocs.com.au/pub/mirrors/oss.sgi.com/projects/kdb/download/v4.4/

Note: Due to a spam attack, the [email protected] mailing list is now subscriber only. If you reply to this mail, you may wish to trim [email protected] from the cc: list.

Quoted 7 lines

Thanks to Nathan Scott for updating KDB while I was on holiday.

Current versions are :-

kdb-v4.4-2.6.16-common-1.bz2
kdb-v4.4-2.6.16-i386-1.bz2
kdb-v4.4-2.6.16-ia64-1.bz2

Thanks for this new version, however I'm looking forward to see kdb maintained also for the x86_64 architecture. Currently I have got as far as forward-porting it to a level where it "works" except for one annoying issue where setjmp/longjmp looks to be broken:

[0]more> q
invalid opcode: 0000 [1] SMP
kdb: Debugger re-entered on cpu 0, new reason = 5
Attempting to abort command and recover
invalid opcode: 0000 [2] SMP
kdb: Debugger re-entered on cpu 0, new reason = 5
Attempting to abort command and recover
invalid opcode: 0000 [3] SMP
kdb: Debugger re-entered on cpu 0, new reason = 5
Attempting to abort command and recover
invalid opcode: 0000 [4] SMP
kdb: Debugger re-entered on cpu 0, new reason = 5
Attempting to abort command and recover
invalid opcode: 0000 [5] SMP
[...]

An inescapable loop.

Kernel configured with CONFIG_FRAME_POINTER=y,
gcc version 3.3.5 (Debian 1:3.3.5-13).

ffffffff8032fc90 <kdba_setjmp>:

ffffffff8032fc90:       55                      push   %rbp
ffffffff8032fc91:       48 89 e5                mov    %rsp,%rbp
ffffffff8032fc94:       48 89 5f 00             mov    %rbx,0x0(%rdi)
ffffffff8032fc98:       48 89 6f 08             mov    %rbp,0x8(%rdi)
ffffffff8032fc9c:       4c 89 67 10             mov    %r12,0x10(%rdi)
ffffffff8032fca0:       4c 89 6f 18             mov    %r13,0x18(%rdi)
ffffffff8032fca4:       4c 89 77 20             mov    %r14,0x20(%rdi)
ffffffff8032fca8:       4c 89 7f 28             mov    %r15,0x28(%rdi)
ffffffff8032fcac:       48 8d 54 24 10          lea    0x10(%rsp),%rdx
ffffffff8032fcb1:       48 89 57 30             mov    %rdx,0x30(%rdi)
ffffffff8032fcb5:       48 8b 04 24             mov    (%rsp),%rax
ffffffff8032fcb9:       48 89 47 38             mov    %rax,0x38(%rdi)
ffffffff8032fcbd:       65 8b 04 25 24 00 00    mov    %gs:0x24,%eax
ffffffff8032fcc4:       00 
ffffffff8032fcc5:       48 c7 c1 c0 86 57 80    mov    $0xffffffff805786c0,%rcx
ffffffff8032fccc:       48 98                   cltq   
ffffffff8032fcce:       8b 14 81                mov    (%rcx,%rax,4),%edx
ffffffff8032fcd1:       80 ce 04                or     $0x4,%dh
ffffffff8032fcd4:       89 14 81                mov    %edx,(%rcx,%rax,4)
ffffffff8032fcd7:       31 c0                   xor    %eax,%eax
ffffffff8032fcd9:       c9                      leaveq 
ffffffff8032fcda:       c3                      retq   
ffffffff8032fcdb:       66                      data16
ffffffff8032fcdc:       66                      data16
ffffffff8032fcdd:       90                      nop    
ffffffff8032fcde:       66                      data16
ffffffff8032fcdf:       90                      nop    

ffffffff8032fce0 <kdba_longjmp>:

ffffffff8032fce0:       55                      push   %rbp
ffffffff8032fce1:       48 89 e5                mov    %rsp,%rbp
ffffffff8032fce4:       48 8b 5f 00             mov    0x0(%rdi),%rbx
ffffffff8032fce8:       48 8b 6f 08             mov    0x8(%rdi),%rbp
ffffffff8032fcec:       4c 8b 67 10             mov    0x10(%rdi),%r12
ffffffff8032fcf0:       4c 8b 6f 18             mov    0x18(%rdi),%r13
ffffffff8032fcf4:       4c 8b 77 20             mov    0x20(%rdi),%r14
ffffffff8032fcf8:       4c 8b 7f 28             mov    0x28(%rdi),%r15
ffffffff8032fcfc:       85 f6                   test   %esi,%esi
ffffffff8032fcfe:       b8 01 00 00 00          mov    $0x1,%eax
ffffffff8032fd03:       0f 44 f0                cmove  %eax,%esi
ffffffff8032fd06:       89 f0                   mov    %esi,%eax
ffffffff8032fd08:       48 8b 57 38             mov    0x38(%rdi),%rdx
ffffffff8032fd0c:       48 8b 67 30             mov    0x30(%rdi),%rsp
ffffffff8032fd10:       ff e2                   jmpq   *%edx
ffffffff8032fd12:       c9                      leaveq 
ffffffff8032fd13:       c3                      retq   
ffffffff8032fd14:       66                      data16
ffffffff8032fd15:       66                      data16
ffffffff8032fd16:       66                      data16
ffffffff8032fd17:       90                      nop    
ffffffff8032fd18:       66                      data16
ffffffff8032fd19:       66                      data16
ffffffff8032fd1a:       66                      data16
ffffffff8032fd1b:       90                      nop    
ffffffff8032fd1c:       66                      data16
ffffffff8032fd1d:       66                      data16
ffffffff8032fd1e:       66                      data16
ffffffff8032fd1f:       90                      nop
Signature · 3 lines
Dan Aloni, Linux specialist
XIV LTD, http://www.xivstorage.com
[email protected], [email protected], [email protected], [email protected]
#3 lore raw
Quoted 7 lines · Dan Aloni

Thanks for this new version, however I'm looking forward to see kdb maintained also for the x86_64 architecture. Currently I have got as far as forward-porting it to a level where it "works" except for one annoying issue where setjmp/longjmp looks to be broken:

Kernel configured with CONFIG_FRAME_POINTER=y,

I've not tried latest kdb, but you should find the patch below still works (removing that unnecessary KDB_STATE_SET(LONGJMP) is probably irrelevant, just something I did on the way, and which does no harm).

Hugh

arch/x86_64/kdb/kdbasupport.c +3 -3
@@ -1035,14 +1004,15 @@ kdba_setjmp(kdb_jmp_buf *jb)
 {
 #if defined(CONFIG_FRAME_POINTER)
 	__asm__("movq %rbx, (0*8)(%rdi);"
-		"movq %rbp, (1*8)(%rdi);"
+		"movq (%rsp), %rax;"
+		"movq %rax, (1*8)(%rdi);"
 		"movq %r12, (2*8)(%rdi);"
 		"movq %r13, (3*8)(%rdi);"
 		"movq %r14, (4*8)(%rdi);"
 		"movq %r15, (5*8)(%rdi);"
 		"leaq 16(%rsp), %rdx;"
 		"movq %rdx, (6*8)(%rdi);"
-		"movq (%rsp), %rax;"
+		"movq 8(%rsp), %rax;"
 		"movq %rax, (7*8)(%rdi)");
 #else	 /* CONFIG_FRAME_POINTER */
 	__asm__("movq %rbx, (0*8)(%rdi);"
@@ -1056,7 +1026,6 @@ kdba_setjmp(kdb_jmp_buf *jb)
 		"movq (%rsp), %rax;"
 		"movq %rax, (7*8)(%rdi)");
 #endif   /* CONFIG_FRAME_POINTER */
-	KDB_STATE_SET(LONGJMP);
 	return 0;
 }
 
#4 lore raw
In reply to Dan Aloni
Quoted 10 lines · Dan Aloni
Quoted 5 lines

Current versions are :-

kdb-v4.4-2.6.16-common-1.bz2
kdb-v4.4-2.6.16-i386-1.bz2
kdb-v4.4-2.6.16-ia64-1.bz2

Thanks for this new version, however I'm looking forward to see kdb maintained also for the x86_64 architecture. Currently I have got as far as forward-porting it to a level where it "works" except for one annoying issue where setjmp/longjmp looks to be broken:

Problem is due to the mixed C/asm implementation of setjmp/longjmp. Replace that with one written entirely in assemply and it will work. Here's mine:

/* setjmp / longjmp for the kernel.
* Inspired by the glibc version.
*/

	.text
	.globl	__kernel_setjmp
	.p2align 4
__kernel_setjmp:
	movq	%rbx,0x0(%rdi)
	movq	%rbp,0x8(%rdi)
	movq	%r12,0x10(%rdi)
	movq	%r13,0x18(%rdi)
	movq	%r14,0x20(%rdi)
	movq	%r15,0x28(%rdi)
	leaq	0x8(%rsp),%rdx
	movq	%rdx,0x30(%rdi)
	movq	(%rsp),%rax
	movq	%rax,0x38(%rdi)
	xorq	%rax,%rax
	ret

	.globl	__kernel_longjmp
__kernel_longjmp:
	movq	0x0(%rdi),%rbx
	movq	0x8(%rdi),%rbp
	movq	0x10(%rdi),%r12
	movq	0x18(%rdi),%r13
	movq	0x20(%rdi),%r14
	movq	0x28(%rdi),%r15
	test	%esi,%esi
	mov	$1,%eax
	cmove	%eax,%esi
	mov	%esi,%eax
	movq	0x38(%rdi),%rdx
	movq	0x30(%rdi),%rsp
	jmpq	*%rdx
#5 lore raw
Quoted 7 lines · Joe Korty
Quoted 6 lines · Dan Aloni
Quoted 5 lines

Current versions are :-

kdb-v4.4-2.6.16-common-1.bz2
kdb-v4.4-2.6.16-i386-1.bz2
kdb-v4.4-2.6.16-ia64-1.bz2

Quoted 8 lines
Quoted 4 lines

Thanks for this new version, however I'm looking forward to see kdb maintained also for the x86_64 architecture. Currently I have got as far as forward-porting it to a level where it "works" except for one annoying issue where setjmp/longjmp looks to be broken:

Problem is due to the mixed C/asm implementation of setjmp/longjmp. Replace that with one written entirely in assemply and it will work. Here's mine:

Thanks, it's obviously a better way to do it, I've tested it and made it into a patch:

arch/x86_64/kdb/kdbasupport.c +35 -65
index 57c2a72..1be4909 100644
@@ -1029,71 +1029,41 @@ kdba_clearsinglestep(struct pt_regs *reg
 		regs->eflags &= ~EF_IE;
 }
 
-#ifdef KDB_HAVE_LONGJMP
-int asmlinkage
-kdba_setjmp(kdb_jmp_buf *jb)
-{
-#if defined(CONFIG_FRAME_POINTER)
-	__asm__("movq %rbx, (0*8)(%rdi);"
-		"movq %rbp, (1*8)(%rdi);"
-		"movq %r12, (2*8)(%rdi);"
-		"movq %r13, (3*8)(%rdi);"
-		"movq %r14, (4*8)(%rdi);"
-		"movq %r15, (5*8)(%rdi);"
-		"leaq 16(%rsp), %rdx;"
-		"movq %rdx, (6*8)(%rdi);"
-		"movq (%rsp), %rax;"
-		"movq %rax, (7*8)(%rdi)");
-#else	 /* CONFIG_FRAME_POINTER */
-	__asm__("movq %rbx, (0*8)(%rdi);"
-		"movq %rbp, (1*8)(%rdi);"
-		"movq %r12, (2*8)(%rdi);"
-		"movq %r13, (3*8)(%rdi);"
-		"movq %r14, (4*8)(%rdi);"
-		"movq %r15, (5*8)(%rdi);"
-		"leaq 8(%rsp), %rdx;"
-		"movq %rdx, (6*8)(%rdi);"
-		"movq (%rsp), %rax;"
-		"movq %rax, (7*8)(%rdi)");
-#endif   /* CONFIG_FRAME_POINTER */
-	KDB_STATE_SET(LONGJMP);
-	return 0;
-}
-
-void asmlinkage
-kdba_longjmp(kdb_jmp_buf *jb, int reason)
-{
-#if defined(CONFIG_FRAME_POINTER)
-	__asm__("movq (0*8)(%rdi),%rbx;"
-		"movq (1*8)(%rdi),%rbp;"
-		"movq (2*8)(%rdi),%r12;"
-		"movq (3*8)(%rdi),%r13;"
-		"movq (4*8)(%rdi),%r14;"
-		"movq (5*8)(%rdi),%r15;"
-		"test %esi,%esi;"
-		"mov $01,%eax;"
-		"cmove %eax,%esi;"
-		"mov %esi, %eax;"
-		"movq (7*8)(%rdi),%rdx;"
-		"movq (6*8)(%rdi),%rsp;"
-		"jmpq *%rdx");
-#else    /* CONFIG_FRAME_POINTER */
-	__asm__("movq (0*8)(%rdi),%rbx;"
-		"movq (1*8)(%rdi),%rbp;"
-		"movq (2*8)(%rdi),%r12;"
-		"movq (3*8)(%rdi),%r13;"
-		"movq (4*8)(%rdi),%r14;"
-		"movq (5*8)(%rdi),%r15;"
-		"test %esi,%esi;"
-		"mov $01,%eax;"
-		"cmove %eax,%esi;"
-		"mov %esi, %eax;"
-		"movq (7*8)(%rdi),%rdx;"
-		"movq (6*8)(%rdi),%rsp;"
-		"jmpq *%rdx");
-#endif	 /* CONFIG_FRAME_POINTER */
-}
-#endif	/* KDB_HAVE_LONGJMP */
+asm(
+"        .section .text\n"
+"        .globl  kdba_setjmp\n"
+"        .p2align 4\n"
+"kdba_setjmp:\n"
+"        movq    %rbx,0x0(%rdi)\n"
+"        movq    %rbp,0x8(%rdi)\n"
+"        movq    %r12,0x10(%rdi)\n"
+"        movq    %r13,0x18(%rdi)\n"
+"        movq    %r14,0x20(%rdi)\n"
+"        movq    %r15,0x28(%rdi)\n"
+"        leaq    0x8(%rsp),%rdx\n"
+"        movq    %rdx,0x30(%rdi)\n"
+"        movq    (%rsp),%rax\n"
+"        movq    %rax,0x38(%rdi)\n"
+"        xorq    %rax,%rax\n"
+"        ret\n"
+"           \n"
+"        .globl  kdba_longjmp\n"
+"kdba_longjmp:\n"
+"        movq    0x0(%rdi),%rbx\n"
+"        movq    0x8(%rdi),%rbp\n"
+"        movq    0x10(%rdi),%r12\n"
+"        movq    0x18(%rdi),%r13\n"
+"        movq    0x20(%rdi),%r14\n"
+"        movq    0x28(%rdi),%r15\n"
+"        test    %esi,%esi\n"
+"        mov     $1,%eax\n"
+"        cmove   %eax,%esi\n"
+"        mov     %esi,%eax\n"
+"        movq    0x38(%rdi),%rdx\n"
+"        movq    0x30(%rdi),%rsp\n"
+"        jmpq    *%rdx\n"
+"        .previous\n"
+);
 
 /*
  * kdba_enable_lbr
Signature · 3 lines
Dan Aloni, Linux specialist
XIV LTD, http://www.xivstorage.com
[email protected], [email protected], [email protected], [email protected]