[RFC 00/16] kGraft

[RFC 00/16] kGraft

[Jiri Slaby]

Hi,

this is the first RFC on kGraft, the linux kernel online patching
developed at SUSE.

The patches are posted as a reply to this email and can be also
obtained as a whole tree at:
https://git.kernel.org/cgit/linux/kernel/git/jirislaby/kgraft.git/log/?h=kgraft

Jiri Kosina (4):
  kgr: initial code
  kgr: x86: refuse to build without fentry support
  kgr: add procfs interface for per-process 'kgr_in_progress'
  kgr: make a per-process 'in progress' flag a single bit

Jiri Slaby (12):
  ftrace: Add function to find fentry of function
  ftrace: Make ftrace_is_dead available globally
  kgr: add testing kgraft patch
  kgr: update Kconfig documentation
  kgr: add Documentation
  kgr: trigger the first check earlier
  kgr: sched.h, introduce kgr_task_safe helper
  kgr: mark task_safe in some kthreads
  kgr: kthreads support
  kgr: handle irqs
  kgr: add tools
  kgr: add MAINTAINERS entry

 Documentation/kgr.txt              |   26 +
 MAINTAINERS                        |    9 +
 arch/x86/Kconfig                   |    2 +
 arch/x86/include/asm/kgr.h         |   45 +
 arch/x86/include/asm/thread_info.h |    6 +-
 arch/x86/kernel/entry_64.S         |    9 +
 arch/x86/kernel/x8664_ksyms_64.c   |    1 +
 drivers/base/devtmpfs.c            |    1 +
 fs/jbd2/journal.c                  |    2 +
 fs/notify/mark.c                   |    5 +-
 fs/proc/base.c                     |   11 +
 include/linux/ftrace.h             |    4 +
 include/linux/kgr.h                |   86 +
 include/linux/sched.h              |    9 +
 kernel/Kconfig.kgr                 |   10 +
 kernel/Makefile                    |    1 +
 kernel/hung_task.c                 |    5 +-
 kernel/kgr.c                       |  338 +++
 kernel/kthread.c                   |    3 +
 kernel/rcu/tree.c                  |    6 +-
 kernel/rcu/tree_plugin.h           |    9 +-
 kernel/trace/ftrace.c              |   29 +
 kernel/trace/trace.h               |    2 -
 kernel/workqueue.c                 |    1 +
 samples/Kconfig                    |    8 +
 samples/Makefile                   |    3 +-
 samples/kgr/Makefile               |    1 +
 samples/kgr/kgr_patcher.c          |   97 +
 tools/Makefile                     |   13 +-
 tools/kgraft/Makefile              |   30 +
 tools/kgraft/README                |   50 +
 tools/kgraft/TODO                  |   20 +
 tools/kgraft/app.c                 |   35 +
 tools/kgraft/app.h                 |    7 +
 tools/kgraft/create-kgrmodule.sh   |   25 +
 tools/kgraft/create-stub.sh        |   53 +
 tools/kgraft/dwarf-inline-tree.c   |  544 +++++
 tools/kgraft/dwarf_names.awk       |  126 ++
 tools/kgraft/dwarf_names.c         | 4366 ++++++++++++++++++++++++++++++++++++
 tools/kgraft/dwarf_names.h         |   53 +
 tools/kgraft/extract-syms.sh       |   18 +
 tools/kgraft/it2rev.pl             |   40 +
 tools/kgraft/objcopy.diff          |  131 ++
 tools/kgraft/symlist               |    1 +
 44 files changed, 6225 insertions(+), 16 deletions(-)
 create mode 100644 Documentation/kgr.txt
 create mode 100644 arch/x86/include/asm/kgr.h
 create mode 100644 include/linux/kgr.h
 create mode 100644 kernel/Kconfig.kgr
 create mode 100644 kernel/kgr.c
 create mode 100644 samples/kgr/Makefile
 create mode 100644 samples/kgr/kgr_patcher.c
 create mode 100644 tools/kgraft/Makefile
 create mode 100644 tools/kgraft/README
 create mode 100644 tools/kgraft/TODO
 create mode 100644 tools/kgraft/app.c
 create mode 100644 tools/kgraft/app.h
 create mode 100755 tools/kgraft/create-kgrmodule.sh
 create mode 100755 tools/kgraft/create-stub.sh
 create mode 100644 tools/kgraft/dwarf-inline-tree.c
 create mode 100644 tools/kgraft/dwarf_names.awk
 create mode 100644 tools/kgraft/dwarf_names.c
 create mode 100644 tools/kgraft/dwarf_names.h
 create mode 100755 tools/kgraft/extract-syms.sh
 create mode 100644 tools/kgraft/it2rev.pl
 create mode 100644 tools/kgraft/objcopy.diff
 create mode 100644 tools/kgraft/symlist

-- 
1.9.2

[RFC 01/16] ftrace: Add function to find fentry of function

[Jiri Slaby]

This is needed for kgr to find fentry location to be "ftraced". We use
this to find a place where to jump to a new/old code location.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 include/linux/ftrace.h |  1 +
 kernel/trace/ftrace.c  | 29 +++++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)

diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
index ae9504b4b67d..8b447493b6a5 100644
--- a/include/linux/ftrace.h
+++ b/include/linux/ftrace.h
@@ -299,6 +299,7 @@ extern void
 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops);
 extern void unregister_ftrace_function_probe_all(char *glob);
 
+extern unsigned long ftrace_function_to_fentry(unsigned long addr);
 extern int ftrace_text_reserved(const void *start, const void *end);
 
 extern int ftrace_nr_registered_ops(void);
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 4a54a25afa2f..9968695cdcf9 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -1495,6 +1495,35 @@ ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
 		}				\
 	}
 
+/**
+ * ftrace_function_to_fentry -- lookup fentry location for a function
+ * @addr: function address to find a fentry in
+ *
+ * Perform a lookup in a list of fentry callsites to find one that fits a
+ * specified function @addr. It returns the corresponding fentry callsite or
+ * zero on failure.
+ */
+unsigned long ftrace_function_to_fentry(unsigned long addr)
+{
+	const struct dyn_ftrace *rec;
+	const struct ftrace_page *pg;
+	unsigned long ret = 0;
+
+	mutex_lock(&ftrace_lock);
+	do_for_each_ftrace_rec(pg, rec) {
+		unsigned long off;
+		if (!kallsyms_lookup_size_offset(rec->ip, NULL, &off))
+			continue;
+		if (addr + off == rec->ip) {
+			ret = rec->ip;
+			goto end;
+		}
+	} while_for_each_ftrace_rec()
+end:
+	mutex_unlock(&ftrace_lock);
+
+	return ret;
+}
 
 static int ftrace_cmp_recs(const void *a, const void *b)
 {
-- 
1.9.2

Re: [RFC 01/16] ftrace: Add function to find fentry of function

[Steven Rostedt]

On Wed, 30 Apr 2014 16:30:34 +0200
Jiri Slaby <jslaby@suse.cz> wrote:

> This is needed for kgr to find fentry location to be "ftraced". We use
> this to find a place where to jump to a new/old code location.
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Ingo Molnar <mingo@redhat.com>
> ---
>  include/linux/ftrace.h |  1 +
>  kernel/trace/ftrace.c  | 29 +++++++++++++++++++++++++++++
>  2 files changed, 30 insertions(+)
> 
> diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
> index ae9504b4b67d..8b447493b6a5 100644
> --- a/include/linux/ftrace.h
> +++ b/include/linux/ftrace.h
> @@ -299,6 +299,7 @@ extern void
>  unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops);
>  extern void unregister_ftrace_function_probe_all(char *glob);
>  
> +extern unsigned long ftrace_function_to_fentry(unsigned long addr);
>  extern int ftrace_text_reserved(const void *start, const void *end);
>  
>  extern int ftrace_nr_registered_ops(void);
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index 4a54a25afa2f..9968695cdcf9 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -1495,6 +1495,35 @@ ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
>  		}				\
>  	}
>  
> +/**
> + * ftrace_function_to_fentry -- lookup fentry location for a function
> + * @addr: function address to find a fentry in
> + *
> + * Perform a lookup in a list of fentry callsites to find one that fits a
> + * specified function @addr. It returns the corresponding fentry callsite or
> + * zero on failure.
> + */
> +unsigned long ftrace_function_to_fentry(unsigned long addr)
> +{
> +	const struct dyn_ftrace *rec;
> +	const struct ftrace_page *pg;
> +	unsigned long ret = 0;
> +
> +	mutex_lock(&ftrace_lock);
> +	do_for_each_ftrace_rec(pg, rec) {

The records are sorted within a pg. You can optimize this a lot if you
just test the first and last record and see if it is in the range. If
not, then skip to the next page. If it is, you can use a bsearch as
well, to save on the lookups.

-- Steve

> +		unsigned long off;
> +		if (!kallsyms_lookup_size_offset(rec->ip, NULL, &off))
> +			continue;
> +		if (addr + off == rec->ip) {
> +			ret = rec->ip;
> +			goto end;
> +		}
> +	} while_for_each_ftrace_rec()
> +end:
> +	mutex_unlock(&ftrace_lock);
> +
> +	return ret;
> +}
>  
>  static int ftrace_cmp_recs(const void *a, const void *b)
>  {

Re: [RFC 01/16] ftrace: Add function to find fentry of function

[Jiri Slaby]

On 04/30/2014 04:48 PM, Steven Rostedt wrote:
> On Wed, 30 Apr 2014 16:30:34 +0200
> Jiri Slaby <jslaby@suse.cz> wrote:
>> --- a/kernel/trace/ftrace.c
>> +++ b/kernel/trace/ftrace.c
>> @@ -1495,6 +1495,35 @@ ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
>>  		}				\
>>  	}
>>  
>> +/**
>> + * ftrace_function_to_fentry -- lookup fentry location for a function
>> + * @addr: function address to find a fentry in
>> + *
>> + * Perform a lookup in a list of fentry callsites to find one that fits a
>> + * specified function @addr. It returns the corresponding fentry callsite or
>> + * zero on failure.
>> + */
>> +unsigned long ftrace_function_to_fentry(unsigned long addr)
>> +{
>> +	const struct dyn_ftrace *rec;
>> +	const struct ftrace_page *pg;
>> +	unsigned long ret = 0;
>> +
>> +	mutex_lock(&ftrace_lock);
>> +	do_for_each_ftrace_rec(pg, rec) {
> 
> The records are sorted within a pg. You can optimize this a lot if you
> just test the first and last record and see if it is in the range. If
> not, then skip to the next page. If it is, you can use a bsearch as
> well, to save on the lookups.

Yes, this is a KISS (and suboptimal) version as there is a slight issue
in your suggestion: if we pass addres of a function of which the fentry
is the first in pg, it would be out of range for the particular pg. What
should work is: check the first entry and then search binary...

thanks,
-- 
js
suse labs

[RFC 02/16] ftrace: Make ftrace_is_dead available globally

[Jiri Slaby]

Kgr wants to test whether ftrace is OK with patching. If not, we just
bail out and will not initialize at all.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 include/linux/ftrace.h | 3 +++
 kernel/trace/trace.h   | 2 --
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
index 8b447493b6a5..720b7be77615 100644
--- a/include/linux/ftrace.h
+++ b/include/linux/ftrace.h
@@ -144,6 +144,8 @@ enum ftrace_tracing_type_t {
 /* Current tracing type, default is FTRACE_TYPE_ENTER */
 extern enum ftrace_tracing_type_t ftrace_tracing_type;
 
+extern int ftrace_is_dead(void);
+
 /**
  * ftrace_stop - stop function tracer.
  *
@@ -245,6 +247,7 @@ static inline int ftrace_nr_registered_ops(void)
 	return 0;
 }
 static inline void clear_ftrace_function(void) { }
+static inline int ftrace_is_dead(void) { return 0; }
 static inline void ftrace_kill(void) { }
 static inline void ftrace_stop(void) { }
 static inline void ftrace_start(void) { }
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 2e29d7ba5a52..e3d867571ffc 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -819,7 +819,6 @@ static inline int ftrace_trace_task(struct task_struct *task)
 
 	return test_tsk_trace_trace(task);
 }
-extern int ftrace_is_dead(void);
 int ftrace_create_function_files(struct trace_array *tr,
 				 struct dentry *parent);
 void ftrace_destroy_function_files(struct trace_array *tr);
@@ -828,7 +827,6 @@ static inline int ftrace_trace_task(struct task_struct *task)
 {
 	return 1;
 }
-static inline int ftrace_is_dead(void) { return 0; }
 static inline int
 ftrace_create_function_files(struct trace_array *tr,
 			     struct dentry *parent)
-- 
1.9.2

[RFC 03/16] kgr: initial code

[Jiri Slaby]

From: Jiri Kosina <jkosina@suse.cz>

Provide initial implementation. We are now able to do ftrace-based
runtime patching of the kernel code.

In addition to that, we will provide a kgr_patcher module in the next
patch to test the functionality.

Limitations/TODOs:

- rmmod of the module that provides the patch is not possible (it'd be nice
  if that'd cause reverse application of the patch -- would be necessary to
  keep a list of patched locations)
- x86_64 only

Additional squashes to this patch:
jk: add missing Kconfig.kgr
jk: fixup a header bug
jk: cleanup comments
js: port to new mcount infrastructure
js: order includes
js: fix for non-KGR (prototype and Kconfig fixes)
js: fix potential lock imbalance in kgr_patch_code
js: use insn helper for jmp generation
js: add \n to a printk
jk: externally_visible attribute warning fix
jk: symbol lookup failure handling
jk: fix race between patching and setting a flag (thanks to bpetkov)
js: add more sanity checking
js: handle missing kallsyms gracefully
js: use correct name, not alias
js: fix index in cleanup path
js: clear kgr_in_progress for all syscall paths
js: cleanup
js: do the checking in the process context
js: call kgr_mark_processes outside loop and locks
jk: convert from raw patching to ftrace API
jk: depend on regs-saving ftrace
js: make kgr_init an init_call
js: use correct offset for stub

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 arch/x86/Kconfig                   |   2 +
 arch/x86/include/asm/kgr.h         |  39 +++++
 arch/x86/include/asm/thread_info.h |   1 +
 arch/x86/kernel/asm-offsets.c      |   1 +
 arch/x86/kernel/entry_64.S         |   3 +
 arch/x86/kernel/x8664_ksyms_64.c   |   1 +
 include/linux/kgr.h                |  71 +++++++++
 kernel/Kconfig.kgr                 |   7 +
 kernel/Makefile                    |   1 +
 kernel/kgr.c                       | 308 +++++++++++++++++++++++++++++++++++++
 10 files changed, 434 insertions(+)
 create mode 100644 arch/x86/include/asm/kgr.h
 create mode 100644 include/linux/kgr.h
 create mode 100644 kernel/Kconfig.kgr
 create mode 100644 kernel/kgr.c

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 25d2c6f7325e..789a4c870ab3 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -130,6 +130,7 @@ config X86
 	select HAVE_CC_STACKPROTECTOR
 	select GENERIC_CPU_AUTOPROBE
 	select HAVE_ARCH_AUDITSYSCALL
+	select HAVE_KGR
 
 config INSTRUCTION_DECODER
 	def_bool y
@@ -263,6 +264,7 @@ config ARCH_SUPPORTS_UPROBES
 
 source "init/Kconfig"
 source "kernel/Kconfig.freezer"
+source "kernel/Kconfig.kgr"
 
 menu "Processor type and features"
 
diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
new file mode 100644
index 000000000000..172f7b966bb5
--- /dev/null
+++ b/arch/x86/include/asm/kgr.h
@@ -0,0 +1,39 @@
+#ifndef ASM_KGR_H
+#define ASM_KGR_H
+
+#include <linux/linkage.h>
+
+/*
+ * The stub needs to modify the RIP value stored in struct pt_regs
+ * so that ftrace redirects the execution properly.
+ */
+#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
+static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
+		struct ftrace_ops *ops, struct pt_regs *regs)		\
+{									\
+	struct kgr_loc_caches *c = ops->private;			\
+									\
+	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
+		pr_info("kgr: slow stub: calling old code at %lx\n",	\
+				c->old);				\
+		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
+	} else {							\
+		pr_info("kgr: slow stub: calling new code at %lx\n",	\
+				c->new);				\
+		regs->ip = c->new;					\
+	}								\
+}
+
+#define KGR_STUB_ARCH_FAST(_name, _new_function)			\
+static void _new_function ##_stub_fast (unsigned long ip,		\
+		unsigned long parent_ip, struct ftrace_ops *ops,	\
+		struct pt_regs *regs)					\
+{									\
+	struct kgr_loc_caches *c = ops->private;			\
+									\
+	BUG_ON(!c->new);				\
+	pr_info("kgr: fast stub: calling new code at %lx\n", c->new); \
+	regs->ip = c->new;				\
+}
+
+#endif
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index 47e5de25ba79..1fdc144dcc9c 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -35,6 +35,7 @@ struct thread_info {
 	void __user		*sysenter_return;
 	unsigned int		sig_on_uaccess_error:1;
 	unsigned int		uaccess_err:1;	/* uaccess failed */
+	unsigned short		kgr_in_progress;
 };
 
 #define INIT_THREAD_INFO(tsk)			\
diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
index 9f6b9341950f..0db0437967a2 100644
--- a/arch/x86/kernel/asm-offsets.c
+++ b/arch/x86/kernel/asm-offsets.c
@@ -32,6 +32,7 @@ void common(void) {
 	OFFSET(TI_flags, thread_info, flags);
 	OFFSET(TI_status, thread_info, status);
 	OFFSET(TI_addr_limit, thread_info, addr_limit);
+	OFFSET(TI_kgr_in_progress, thread_info, kgr_in_progress);
 
 	BLANK();
 	OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 1e96c3628bf2..a03b1e9d2de3 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
 	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
 	movq  %rcx,RIP-ARGOFFSET(%rsp)
 	CFI_REL_OFFSET rip,RIP-ARGOFFSET
+	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	jnz tracesys
 system_call_fastpath:
@@ -639,6 +640,7 @@ sysret_check:
 	LOCKDEP_SYS_EXIT
 	DISABLE_INTERRUPTS(CLBR_NONE)
 	TRACE_IRQS_OFF
+	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
 	andl %edi,%edx
 	jnz  sysret_careful
@@ -761,6 +763,7 @@ GLOBAL(int_ret_from_sys_call)
 GLOBAL(int_with_check)
 	LOCKDEP_SYS_EXIT_IRQ
 	GET_THREAD_INFO(%rcx)
+	movw $0, TI_kgr_in_progress(%rcx)
 	movl TI_flags(%rcx),%edx
 	andl %edi,%edx
 	jnz   int_careful
diff --git a/arch/x86/kernel/x8664_ksyms_64.c b/arch/x86/kernel/x8664_ksyms_64.c
index 040681928e9d..df6425d44fa0 100644
--- a/arch/x86/kernel/x8664_ksyms_64.c
+++ b/arch/x86/kernel/x8664_ksyms_64.c
@@ -3,6 +3,7 @@
 
 #include <linux/module.h>
 #include <linux/smp.h>
+#include <linux/kgr.h>
 
 #include <net/checksum.h>
 
diff --git a/include/linux/kgr.h b/include/linux/kgr.h
new file mode 100644
index 000000000000..d72add7f3d5d
--- /dev/null
+++ b/include/linux/kgr.h
@@ -0,0 +1,71 @@
+#ifndef LINUX_KGR_H
+#define LINUX_KGR_H
+
+#include <linux/init.h>
+#include <linux/ftrace.h>
+
+#include <asm/kgr.h>
+
+#ifdef CONFIG_KGR
+
+#define KGR_TIMEOUT 30
+#define KGR_DEBUG 1
+
+#ifdef KGR_DEBUG
+#define kgr_debug(args...)	\
+	pr_info(args);
+#else
+#define kgr_debug(args...) { }
+#endif
+
+struct kgr_patch {
+	char reserved;
+	const struct kgr_patch_fun {
+		const char *name;
+		const char *new_name;
+		void *new_function;
+		struct ftrace_ops *ftrace_ops_slow;
+		struct ftrace_ops *ftrace_ops_fast;
+
+	} *patches[];
+};
+
+/*
+ * data structure holding locations of the source and target function
+ * fentry sites to avoid repeated lookups
+ */
+struct kgr_loc_caches {
+	unsigned long old;
+	unsigned long new;
+};
+
+#define KGR_PATCHED_FUNCTION(patch, _name, _new_function)			\
+	KGR_STUB_ARCH_SLOW(_name, _new_function);				\
+	KGR_STUB_ARCH_FAST(_name, _new_function);				\
+	extern void _new_function ## _stub_slow (unsigned long, unsigned long,	\
+		                       struct ftrace_ops *, struct pt_regs *);	\
+	extern void _new_function ## _stub_fast (unsigned long, unsigned long,	\
+		                       struct ftrace_ops *, struct pt_regs *);	\
+	static struct ftrace_ops __kgr_patch_ftrace_ops_slow_ ## _name = {	\
+		.func = _new_function ## _stub_slow,				\
+		.flags = FTRACE_OPS_FL_SAVE_REGS,				\
+	};									\
+	static struct ftrace_ops __kgr_patch_ftrace_ops_fast_ ## _name = {	\
+		.func = _new_function ## _stub_fast,				\
+		.flags = FTRACE_OPS_FL_SAVE_REGS,				\
+	};									\
+	static const struct kgr_patch_fun __kgr_patch_ ## _name = {		\
+		.name = #_name,							\
+		.new_name = #_new_function,					\
+		.new_function = _new_function,					\
+		.ftrace_ops_slow = &__kgr_patch_ftrace_ops_slow_ ## _name,	\
+		.ftrace_ops_fast = &__kgr_patch_ftrace_ops_fast_ ## _name,	\
+	};									\
+
+#define KGR_PATCH(name)		&__kgr_patch_ ## name
+#define KGR_PATCH_END		NULL
+
+extern int kgr_start_patching(const struct kgr_patch *);
+#endif /* CONFIG_KGR */
+
+#endif /* LINUX_KGR_H */
diff --git a/kernel/Kconfig.kgr b/kernel/Kconfig.kgr
new file mode 100644
index 000000000000..af9125f27b6d
--- /dev/null
+++ b/kernel/Kconfig.kgr
@@ -0,0 +1,7 @@
+config HAVE_KGR
+	bool
+
+config KGR
+	tristate "Kgr infrastructure"
+	depends on DYNAMIC_FTRACE_WITH_REGS
+	depends on HAVE_KGR
diff --git a/kernel/Makefile b/kernel/Makefile
index f2a8b6246ce9..86ac7a2e5fe0 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -28,6 +28,7 @@ obj-y += printk/
 obj-y += irq/
 obj-y += rcu/
 
+obj-$(CONFIG_KGR) += kgr.o
 obj-$(CONFIG_CHECKPOINT_RESTORE) += kcmp.o
 obj-$(CONFIG_FREEZER) += freezer.o
 obj-$(CONFIG_PROFILING) += profile.o
diff --git a/kernel/kgr.c b/kernel/kgr.c
new file mode 100644
index 000000000000..6f55c7654618
--- /dev/null
+++ b/kernel/kgr.c
@@ -0,0 +1,308 @@
+/*
+ * kGraft Online Kernel Patching
+ *
+ *  Copyright (c) 2013-2014 SUSE
+ *   Authors: Jiri Kosina
+ *	      Vojtech Pavlik
+ *	      Jiri Slaby
+ */
+
+/*
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ */
+
+#include <linux/ftrace.h>
+#include <linux/kallsyms.h>
+#include <linux/kgr.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/sort.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+#include <linux/workqueue.h>
+
+static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final);
+static void kgr_work_fn(struct work_struct *work);
+
+static struct workqueue_struct *kgr_wq;
+static DECLARE_DELAYED_WORK(kgr_work, kgr_work_fn);
+static DEFINE_MUTEX(kgr_in_progress_lock);
+static bool kgr_in_progress;
+static bool kgr_initialized;
+static const struct kgr_patch *kgr_patch;
+
+static bool kgr_still_patching(void)
+{
+	struct task_struct *p;
+	bool failed = false;
+
+	read_lock(&tasklist_lock);
+	for_each_process(p) {
+		/*
+		 * TODO
+		 *   kernel thread codepaths not supported and silently ignored
+		 */
+		if (task_thread_info(p)->kgr_in_progress && p->mm) {
+			pr_info("pid %d (%s) still in kernel after timeout\n",
+					p->pid, p->comm);
+			failed = true;
+		}
+	}
+	read_unlock(&tasklist_lock);
+	return failed;
+}
+
+static void kgr_finalize(void)
+{
+	const struct kgr_patch_fun *const *patch_fun;
+
+	for (patch_fun = kgr_patch->patches; *patch_fun; patch_fun++) {
+		int ret = kgr_patch_code(*patch_fun, true);
+		/*
+		 * In case any of the symbol resolutions in the set
+		 * has failed, patch all the previously replaced fentry
+		 * callsites back to nops and fail with grace
+		 */
+		if (ret < 0)
+			pr_err("kgr: finalize for %s failed, trying to continue\n",
+					(*patch_fun)->name);
+	}
+}
+
+static void kgr_work_fn(struct work_struct *work)
+{
+	if (kgr_still_patching()) {
+		pr_info("kgr failed after timeout (%d), still in degraded mode\n",
+			KGR_TIMEOUT);
+		/* recheck again later */
+		queue_delayed_work(kgr_wq, &kgr_work, KGR_TIMEOUT * HZ);
+		return;
+	}
+
+	/*
+	 * victory, patching finished, put everything back in shape
+	 * with as less performance impact as possible again
+	 */
+	pr_info("kgr succeeded\n");
+	kgr_finalize();
+	mutex_lock(&kgr_in_progress_lock);
+	kgr_in_progress = false;
+	mutex_unlock(&kgr_in_progress_lock);
+}
+
+static void kgr_mark_processes(void)
+{
+	struct task_struct *p;
+
+	read_lock(&tasklist_lock);
+	for_each_process(p)
+		task_thread_info(p)->kgr_in_progress = true;
+	read_unlock(&tasklist_lock);
+}
+
+static unsigned long kgr_get_fentry_loc(const char *f_name)
+{
+	unsigned long orig_addr, fentry_loc;
+	const char *check_name;
+	char check_buf[KSYM_SYMBOL_LEN];
+
+	orig_addr = kallsyms_lookup_name(f_name);
+	if (!orig_addr) {
+		WARN(1, "kgr: function %s not resolved ... kernel in inconsistent state\n",
+				f_name);
+		return -EINVAL;
+	}
+
+	fentry_loc = ftrace_function_to_fentry(orig_addr);
+	if (!fentry_loc) {
+		pr_err("kgr: fentry_loc not properly resolved\n");
+		return -EINVAL;
+	}
+
+	check_name = kallsyms_lookup(fentry_loc, NULL, NULL, NULL, check_buf);
+	if (strcmp(check_name, f_name)) {
+		pr_err("kgr: we got out of bounds the intended function (%s -> %s)\n",
+				f_name, check_name);
+		return -EINVAL;
+	}
+
+	return fentry_loc;
+}
+
+static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
+{
+	struct kgr_loc_caches *caches;
+	unsigned long fentry_loc;
+
+	/*
+	 * Initialize the ftrace_ops->private with pointers to the fentry
+	 * sites of both old and new functions. This is used as a
+	 * redirection target in the per-arch stubs.
+	 *
+	 * Beware! -- freeing (once unloading will be implemented)
+	 * will require synchronize_sched() etc.
+	 */
+
+	caches = kmalloc(sizeof(*caches), GFP_KERNEL);
+	if (!caches) {
+		kgr_debug("kgr: unable to allocate fentry caches\n");
+		return -ENOMEM;
+	}
+
+	fentry_loc = kgr_get_fentry_loc(patch_fun->new_name);
+	if (IS_ERR_VALUE(fentry_loc)) {
+		kgr_debug("kgr: fentry location lookup failed\n");
+		return fentry_loc;
+	}
+	kgr_debug("kgr: storing %lx to caches->new for %s\n",
+			fentry_loc, patch_fun->new_name);
+	caches->new = fentry_loc;
+
+	fentry_loc = kgr_get_fentry_loc(patch_fun->name);
+	if (IS_ERR_VALUE(fentry_loc)) {
+		kgr_debug("kgr: fentry location lookup failed\n");
+		return fentry_loc;
+	}
+
+	kgr_debug("kgr: storing %lx to caches->old for %s\n",
+			fentry_loc, patch_fun->name);
+	caches->old = fentry_loc;
+
+	patch_fun->ftrace_ops_fast->private = caches;
+	patch_fun->ftrace_ops_slow->private = caches;
+
+	return 0;
+}
+
+static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final)
+{
+	struct ftrace_ops *new_ops;
+	struct kgr_loc_caches *caches;
+	unsigned long fentry_loc;
+	int err;
+
+	/* Choose between slow and fast stub */
+	if (!final) {
+		err = kgr_init_ftrace_ops(patch_fun);
+		if (err)
+			return err;
+		kgr_debug("kgr: patching %s to slow stub\n", patch_fun->name);
+		new_ops = patch_fun->ftrace_ops_slow;
+	} else {
+		kgr_debug("kgr: patching %s to fast stub\n", patch_fun->name);
+		new_ops = patch_fun->ftrace_ops_fast;
+	}
+
+	/* Flip the switch */
+	caches = new_ops->private;
+	fentry_loc = caches->old;
+	err = ftrace_set_filter_ip(new_ops, fentry_loc, 0, 0);
+	if (err) {
+		kgr_debug("kgr: setting filter for %lx (%s) failed\n",
+				caches->old, patch_fun->name);
+		return err;
+	}
+
+	err = register_ftrace_function(new_ops);
+	if (err) {
+		kgr_debug("kgr: registering ftrace function for %lx (%s) failed\n",
+				caches->old, patch_fun->name);
+		return err;
+	}
+
+	/*
+	 * Get rid of the slow stub. Having two stubs in the interim is fine,
+	 * the last one always "wins", as it'll be dragged earlier from the
+	 * ftrace hashtable
+	 */
+	if (final) {
+		err = unregister_ftrace_function(patch_fun->ftrace_ops_slow);
+		if (err) {
+			kgr_debug("kgr: unregistering ftrace function for %lx (%s) failed\n",
+					fentry_loc, patch_fun->name);
+			return err;
+		}
+	}
+	kgr_debug("kgr: redirection for %lx (%s) done\n", fentry_loc,
+			patch_fun->name);
+
+	return 0;
+}
+
+/**
+ * kgr_start_patching -- the entry for a kgraft patch
+ * @patch: patch to be applied
+ *
+ * Start patching of code that is neither running in IRQ context nor
+ * kernel thread.
+ */
+int kgr_start_patching(const struct kgr_patch *patch)
+{
+	const struct kgr_patch_fun *const *patch_fun;
+
+	if (!kgr_initialized) {
+		pr_err("kgr: can't patch, not initialized\n");
+		return -EINVAL;
+	}
+
+	mutex_lock(&kgr_in_progress_lock);
+	if (kgr_in_progress) {
+		pr_err("kgr: can't patch, another patching not yet finalized\n");
+		mutex_unlock(&kgr_in_progress_lock);
+		return -EAGAIN;
+	}
+
+	for (patch_fun = patch->patches; *patch_fun; patch_fun++) {
+		int ret;
+
+		ret = kgr_patch_code(*patch_fun, false);
+		/*
+		 * In case any of the symbol resolutions in the set
+		 * has failed, patch all the previously replaced fentry
+		 * callsites back to nops and fail with grace
+		 */
+		if (ret < 0) {
+			for (; patch_fun >= patch->patches; patch_fun--)
+				unregister_ftrace_function((*patch_fun)->ftrace_ops_slow);
+			mutex_unlock(&kgr_in_progress_lock);
+			return ret;
+		}
+	}
+	kgr_in_progress = true;
+	kgr_patch = patch;
+	mutex_unlock(&kgr_in_progress_lock);
+
+	kgr_mark_processes();
+
+	/*
+	 * give everyone time to exit kernel, and check after a while
+	 */
+	queue_delayed_work(kgr_wq, &kgr_work, KGR_TIMEOUT * HZ);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(kgr_start_patching);
+
+static int __init kgr_init(void)
+{
+	if (ftrace_is_dead()) {
+		pr_warning("kgr: enabled, but no fentry locations found ... aborting\n");
+		return -ENODEV;
+	}
+
+	kgr_wq = create_singlethread_workqueue("kgr");
+	if (!kgr_wq) {
+		pr_err("kgr: cannot allocate a work queue, aborting!\n");
+		return -ENOMEM;
+	}
+
+	kgr_initialized = true;
+	pr_info("kgr: successfully initialized\n");
+
+	return 0;
+}
+module_init(kgr_init);
-- 
1.9.2

Re: [RFC 03/16] kgr: initial code

[Steven Rostedt]

On Wed, 30 Apr 2014 16:30:36 +0200
Jiri Slaby <jslaby@suse.cz> wrote:

> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index 25d2c6f7325e..789a4c870ab3 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -130,6 +130,7 @@ config X86
>  	select HAVE_CC_STACKPROTECTOR
>  	select GENERIC_CPU_AUTOPROBE
>  	select HAVE_ARCH_AUDITSYSCALL
> +	select HAVE_KGR
>  
>  config INSTRUCTION_DECODER
>  	def_bool y
> @@ -263,6 +264,7 @@ config ARCH_SUPPORTS_UPROBES
>  
>  source "init/Kconfig"
>  source "kernel/Kconfig.freezer"
> +source "kernel/Kconfig.kgr"
>  
>  menu "Processor type and features"
>  
> diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
> new file mode 100644
> index 000000000000..172f7b966bb5
> --- /dev/null
> +++ b/arch/x86/include/asm/kgr.h
> @@ -0,0 +1,39 @@
> +#ifndef ASM_KGR_H
> +#define ASM_KGR_H
> +
> +#include <linux/linkage.h>
> +
> +/*
> + * The stub needs to modify the RIP value stored in struct pt_regs
> + * so that ftrace redirects the execution properly.
> + */
> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
> +{									\
> +	struct kgr_loc_caches *c = ops->private;			\
> +									\
> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
> +		pr_info("kgr: slow stub: calling old code at %lx\n",	\
> +				c->old);				\
> +		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
> +	} else {							\
> +		pr_info("kgr: slow stub: calling new code at %lx\n",	\
> +				c->new);				\
> +		regs->ip = c->new;					\
> +	}								\
> +}
> +
> +#define KGR_STUB_ARCH_FAST(_name, _new_function)			\
> +static void _new_function ##_stub_fast (unsigned long ip,		\
> +		unsigned long parent_ip, struct ftrace_ops *ops,	\
> +		struct pt_regs *regs)					\
> +{									\
> +	struct kgr_loc_caches *c = ops->private;			\
> +									\
> +	BUG_ON(!c->new);				\
> +	pr_info("kgr: fast stub: calling new code at %lx\n", c->new); \
> +	regs->ip = c->new;				\
> +}
> +
> +#endif
> diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
> index 47e5de25ba79..1fdc144dcc9c 100644
> --- a/arch/x86/include/asm/thread_info.h
> +++ b/arch/x86/include/asm/thread_info.h
> @@ -35,6 +35,7 @@ struct thread_info {
>  	void __user		*sysenter_return;
>  	unsigned int		sig_on_uaccess_error:1;
>  	unsigned int		uaccess_err:1;	/* uaccess failed */
> +	unsigned short		kgr_in_progress;
>  };
>  
>  #define INIT_THREAD_INFO(tsk)			\
> diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
> index 9f6b9341950f..0db0437967a2 100644
> --- a/arch/x86/kernel/asm-offsets.c
> +++ b/arch/x86/kernel/asm-offsets.c
> @@ -32,6 +32,7 @@ void common(void) {
>  	OFFSET(TI_flags, thread_info, flags);
>  	OFFSET(TI_status, thread_info, status);
>  	OFFSET(TI_addr_limit, thread_info, addr_limit);
> +	OFFSET(TI_kgr_in_progress, thread_info, kgr_in_progress);
>  
>  	BLANK();
>  	OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
> index 1e96c3628bf2..a03b1e9d2de3 100644
> --- a/arch/x86/kernel/entry_64.S
> +++ b/arch/x86/kernel/entry_64.S
> @@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
>  	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
>  	movq  %rcx,RIP-ARGOFFSET(%rsp)
>  	CFI_REL_OFFSET rip,RIP-ARGOFFSET
> +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)

Why is this not a entry flag? Because you just added a store into a
fast path of the kernel for something that will be hardly ever used.


>  	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
>  	jnz tracesys
>  system_call_fastpath:
> @@ -639,6 +640,7 @@ sysret_check:
>  	LOCKDEP_SYS_EXIT
>  	DISABLE_INTERRUPTS(CLBR_NONE)
>  	TRACE_IRQS_OFF
> +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
>  	movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
>  	andl %edi,%edx
>  	jnz  sysret_careful
> @@ -761,6 +763,7 @@ GLOBAL(int_ret_from_sys_call)
>  GLOBAL(int_with_check)
>  	LOCKDEP_SYS_EXIT_IRQ
>  	GET_THREAD_INFO(%rcx)
> +	movw $0, TI_kgr_in_progress(%rcx)
>  	movl TI_flags(%rcx),%edx
>  	andl %edi,%edx
>  	jnz   int_careful
> diff --git a/arch/x86/kernel/x8664_ksyms_64.c b/arch/x86/kernel/x8664_ksyms_64.c
> index 040681928e9d..df6425d44fa0 100644
> --- a/arch/x86/kernel/x8664_ksyms_64.c
> +++ b/arch/x86/kernel/x8664_ksyms_64.c
> @@ -3,6 +3,7 @@
>  
>  #include <linux/module.h>
>  #include <linux/smp.h>
> +#include <linux/kgr.h>
>  
>  #include <net/checksum.h>
>  
> diff --git a/include/linux/kgr.h b/include/linux/kgr.h
> new file mode 100644
> index 000000000000..d72add7f3d5d
> --- /dev/null
> +++ b/include/linux/kgr.h
> @@ -0,0 +1,71 @@
> +#ifndef LINUX_KGR_H
> +#define LINUX_KGR_H
> +
> +#include <linux/init.h>
> +#include <linux/ftrace.h>
> +
> +#include <asm/kgr.h>
> +
> +#ifdef CONFIG_KGR
> +
> +#define KGR_TIMEOUT 30
> +#define KGR_DEBUG 1
> +
> +#ifdef KGR_DEBUG
> +#define kgr_debug(args...)	\
> +	pr_info(args);
> +#else
> +#define kgr_debug(args...) { }
> +#endif

Why not just use pr_debug(), as that's not defined unless you add DEBUG
as a define anyway?

-- Steve

> +
> +struct kgr_patch {
> +	char reserved;
> +	const struct kgr_patch_fun {
> +		const char *name;
> +		const char *new_name;
> +		void *new_function;
> +		struct ftrace_ops *ftrace_ops_slow;
> +		struct ftrace_ops *ftrace_ops_fast;
> +
> +	} *patches[];
> +};
> +
> +/*

Re: [RFC 03/16] kgr: initial code

[Jiri Slaby]

On 04/30/2014 04:56 PM, Steven Rostedt wrote:
> On Wed, 30 Apr 2014 16:30:36 +0200
> Jiri Slaby <jslaby@suse.cz> wrote:
>> --- a/arch/x86/kernel/entry_64.S
>> +++ b/arch/x86/kernel/entry_64.S
>> @@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
>>  	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
>>  	movq  %rcx,RIP-ARGOFFSET(%rsp)
>>  	CFI_REL_OFFSET rip,RIP-ARGOFFSET
>> +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
> 
> Why is this not a entry flag? Because you just added a store into a
> fast path of the kernel for something that will be hardly ever used.

Actually it is converted later in the series, please see 16/16.

>> --- /dev/null
>> +++ b/include/linux/kgr.h
>> @@ -0,0 +1,71 @@
>> +#ifndef LINUX_KGR_H
>> +#define LINUX_KGR_H
>> +
>> +#include <linux/init.h>
>> +#include <linux/ftrace.h>
>> +
>> +#include <asm/kgr.h>
>> +
>> +#ifdef CONFIG_KGR
>> +
>> +#define KGR_TIMEOUT 30
>> +#define KGR_DEBUG 1
>> +
>> +#ifdef KGR_DEBUG
>> +#define kgr_debug(args...)	\
>> +	pr_info(args);
>> +#else
>> +#define kgr_debug(args...) { }
>> +#endif
> 
> Why not just use pr_debug(), as that's not defined unless you add DEBUG
> as a define anyway?

Yeah, OK.

thanks,
-- 
js
suse labs

Re: [RFC 03/16] kgr: initial code

[Andi Kleen]

Jiri Slaby <jslaby@suse.cz> writes:
>  	OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
> index 1e96c3628bf2..a03b1e9d2de3 100644
> --- a/arch/x86/kernel/entry_64.S
> +++ b/arch/x86/kernel/entry_64.S
> @@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
>  	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
>  	movq  %rcx,RIP-ARGOFFSET(%rsp)
>  	CFI_REL_OFFSET rip,RIP-ARGOFFSET
> +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)

Better use 4 bytes. This has the potential to cause an expensive
Length Changing Prefixes Stall on Intel CPUs.
> +
> +static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
> +{
> +	struct kgr_loc_caches *caches;
> +	unsigned long fentry_loc;
> +
> +	/*
> +	 * Initialize the ftrace_ops->private with pointers to the fentry
> +	 * sites of both old and new functions. This is used as a
> +	 * redirection target in the per-arch stubs.
> +	 *
> +	 * Beware! -- freeing (once unloading will be implemented)
> +	 * will require synchronize_sched() etc.
> +	 */
> +
> +	caches = kmalloc(sizeof(*caches), GFP_KERNEL);
> +	if (!caches) {
> +		kgr_debug("kgr: unable to allocate fentry caches\n");
> +		return -ENOMEM;
> +	}

All the error paths in this function leak memory.


-Andi
-- 
ak@linux.intel.com -- Speaking for myself only

Re: [RFC 03/16] kgr: initial code

[Jiri Kosina]

On Thu, 1 May 2014, Andi Kleen wrote:

> > diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
> > index 1e96c3628bf2..a03b1e9d2de3 100644
> > --- a/arch/x86/kernel/entry_64.S
> > +++ b/arch/x86/kernel/entry_64.S
> > @@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
> >  	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
> >  	movq  %rcx,RIP-ARGOFFSET(%rsp)
> >  	CFI_REL_OFFSET rip,RIP-ARGOFFSET
> > +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
> 
> Better use 4 bytes. This has the potential to cause an expensive
> Length Changing Prefixes Stall on Intel CPUs.

Patch 16/16 converts this to a single bit within TI_flags.

> > +static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
> > +{
> > +	struct kgr_loc_caches *caches;
> > +	unsigned long fentry_loc;
> > +
> > +	/*
> > +	 * Initialize the ftrace_ops->private with pointers to the fentry
> > +	 * sites of both old and new functions. This is used as a
> > +	 * redirection target in the per-arch stubs.
> > +	 *
> > +	 * Beware! -- freeing (once unloading will be implemented)
> > +	 * will require synchronize_sched() etc.
> > +	 */
> > +
> > +	caches = kmalloc(sizeof(*caches), GFP_KERNEL);
> > +	if (!caches) {
> > +		kgr_debug("kgr: unable to allocate fentry caches\n");
> > +		return -ENOMEM;
> > +	}
> 
> All the error paths in this function leak memory.

Gah, good catch, thanks a lot.

-- 
Jiri Kosina
SUSE Labs

Re: [RFC 03/16] kgr: initial code

[Aravinda Prasad]

On Wednesday 30 April 2014 08:00 PM, Jiri Slaby wrote:
> From: Jiri Kosina <jkosina@suse.cz>
> 
> Provide initial implementation. We are now able to do ftrace-based
> runtime patching of the kernel code.
> 
> In addition to that, we will provide a kgr_patcher module in the next
> patch to test the functionality.

Hi Jiri,

Interesting! I have couple of comments:

I think with kgraft (also with kpatch, though have not looked into
it yet), the patched function cannot be dynamically ftraced.
Though dynamic ftrace can be enabled on the new code, the user is
required to know the function label of the new code. This could
potentially break existing scripts. I think this should be documented.

Rest of the comments in-line.

> +/*
> + * The stub needs to modify the RIP value stored in struct pt_regs
> + * so that ftrace redirects the execution properly.
> + */
> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
> +{									\
> +	struct kgr_loc_caches *c = ops->private;			\
> +									\
> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\

Is there a race here? The per task kgr_in_progress is set after
the slow stub is registered in register_ftrace_function(). If the
patched function is called in between it will be redirected to new code.


> +		pr_info("kgr: slow stub: calling old code at %lx\n",	\
> +				c->old);				\
> +		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
> +	} else {							\
> +		pr_info("kgr: slow stub: calling new code at %lx\n",	\
> +				c->new);				\
> +		regs->ip = c->new;					\
> +	}								\

[...]

> +static void kgr_mark_processes(void)
> +{
> +	struct task_struct *p;
> +
> +	read_lock(&tasklist_lock);
> +	for_each_process(p)
> +		task_thread_info(p)->kgr_in_progress = true;

Is there a need for memory barrier here (or in slow stub) to avoid
the race if the slow stub is about to be called from a thread executing
on another CPU?

> +	read_unlock(&tasklist_lock);
> +}
> +

[...]

> + * kgr_start_patching -- the entry for a kgraft patch
> + * @patch: patch to be applied
> + *
> + * Start patching of code that is neither running in IRQ context nor
> + * kernel thread.
> + */
> +int kgr_start_patching(const struct kgr_patch *patch)
> +{
> +	const struct kgr_patch_fun *const *patch_fun;
> +
> +	if (!kgr_initialized) {
> +		pr_err("kgr: can't patch, not initialized\n");
> +		return -EINVAL;
> +	}
> +
> +	mutex_lock(&kgr_in_progress_lock);
> +	if (kgr_in_progress) {
> +		pr_err("kgr: can't patch, another patching not yet finalized\n");
> +		mutex_unlock(&kgr_in_progress_lock);
> +		return -EAGAIN;
> +	}
> +
> +	for (patch_fun = patch->patches; *patch_fun; patch_fun++) {
> +		int ret;
> +
> +		ret = kgr_patch_code(*patch_fun, false);
> +		/*
> +		 * In case any of the symbol resolutions in the set
> +		 * has failed, patch all the previously replaced fentry
> +		 * callsites back to nops and fail with grace
> +		 */
> +		if (ret < 0) {
> +			for (; patch_fun >= patch->patches; patch_fun--)
> +				unregister_ftrace_function((*patch_fun)->ftrace_ops_slow);
> +			mutex_unlock(&kgr_in_progress_lock);
> +			return ret;
> +		}
> +	}
> +	kgr_in_progress = true;
> +	kgr_patch = patch;
> +	mutex_unlock(&kgr_in_progress_lock);
> +
> +	kgr_mark_processes();
> +
> +	/*
> +	 * give everyone time to exit kernel, and check after a while
> +	 */

I understand that the main intention of kgraft is to apply simple
security fixes. However, if the patch changes the locking order,
I think, there is a possibility of deadlock.

A thread which has not yet returned to user space calls the old
code (not redirected to new code in slow stub) which might acquire
the lock in the old order say lock1 followed by lock2. Meanwhile
another thread which re-enters the kernel space, with kgr_in_progress
unset, is redirected to the new code which acquires the lock in reverse
order, say lock2 and lock1. This can cause deadlock.

Thanks,
Aravinda

> +	queue_delayed_work(kgr_wq, &kgr_work, KGR_TIMEOUT * HZ);
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(kgr_start_patching);
> +
> 

-- 
Regards,
Aravinda

Re: [RFC 03/16] kgr: initial code

[Jiri Slaby]

On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
> On Wednesday 30 April 2014 08:00 PM, Jiri Slaby wrote:
>> From: Jiri Kosina <jkosina@suse.cz>
>>
>> Provide initial implementation. We are now able to do ftrace-based
>> runtime patching of the kernel code.
>>
>> In addition to that, we will provide a kgr_patcher module in the next
>> patch to test the functionality.
> 
> Hi Jiri,
> 
> Interesting! I have couple of comments:
> 
> I think with kgraft (also with kpatch, though have not looked into
> it yet), the patched function cannot be dynamically ftraced.
> Though dynamic ftrace can be enabled on the new code, the user is
> required to know the function label of the new code. This could
> potentially break existing scripts. I think this should be documented.

Hi,

of course that the functions can be traced. Look, I turned on tracing
for capable, then patched, then turned on tracing for new_capable (which
is the patched function). So now, trace shows:
  console-kit-dae-535   [001] ...1   181.729698: capable <-vt_ioctl
 console-kit-dae-539   [001] ...1   181.729741: capable <-vt_ioctl
 console-kit-dae-541   [000] .N.1   181.906014: capable <-vt_ioctl
         systemd-1     [001] ...1   181.937328: capable <-SyS_epoll_ctl
            sshd-662   [001] ...1   246.437561: capable <-sock_setsockopt
            sshd-662   [001] ...1   246.437564: new_capable
<-sock_setsockopt
            sshd-662   [001] ...1   246.444790: capable <-sock_setsockopt
            sshd-662   [001] ...1   246.444793: new_capable
<-sock_setsockopt
     dbus-daemon-128   [000] .N.1   246.456307: capable <-SyS_epoll_ctl
     dbus-daemon-128   [000] ...1   246.456611: new_capable <-SyS_epoll_ctl


There is no limitation thanks to the use of the ftrace subsystem. We are
just another user, i.e. another piece of code called in a loop for a
particular fentry location.

>> +/*
>> + * The stub needs to modify the RIP value stored in struct pt_regs
>> + * so that ftrace redirects the execution properly.
>> + */
>> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
>> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
>> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
>> +{									\
>> +	struct kgr_loc_caches *c = ops->private;			\
>> +									\
>> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
> 
> Is there a race here? The per task kgr_in_progress is set after
> the slow stub is registered in register_ftrace_function(). If the
> patched function is called in between it will be redirected to new code.

Hmm, that looks strange. I will look into that and the other comments
later (and comment separately). Thanks.

-- 
js
suse labs

Re: [RFC 03/16] kgr: initial code

[Aravinda Prasad]

On Wednesday 14 May 2014 03:42 PM, Jiri Slaby wrote:
> On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
>> On Wednesday 30 April 2014 08:00 PM, Jiri Slaby wrote:
>>> From: Jiri Kosina <jkosina@suse.cz>
>>>
>>> Provide initial implementation. We are now able to do ftrace-based
>>> runtime patching of the kernel code.
>>>
>>> In addition to that, we will provide a kgr_patcher module in the next
>>> patch to test the functionality.
>>
>> Hi Jiri,
>>
>> Interesting! I have couple of comments:
>>
>> I think with kgraft (also with kpatch, though have not looked into
>> it yet), the patched function cannot be dynamically ftraced.
>> Though dynamic ftrace can be enabled on the new code, the user is
>> required to know the function label of the new code. This could
>> potentially break existing scripts. I think this should be documented.
> 
> Hi,
> 
> of course that the functions can be traced. Look, I turned on tracing
> for capable, then patched, then turned on tracing for new_capable (which
> is the patched function). So now, trace shows:
>   console-kit-dae-535   [001] ...1   181.729698: capable <-vt_ioctl
>  console-kit-dae-539   [001] ...1   181.729741: capable <-vt_ioctl
>  console-kit-dae-541   [000] .N.1   181.906014: capable <-vt_ioctl
>          systemd-1     [001] ...1   181.937328: capable <-SyS_epoll_ctl
>             sshd-662   [001] ...1   246.437561: capable <-sock_setsockopt
>             sshd-662   [001] ...1   246.437564: new_capable
> <-sock_setsockopt
>             sshd-662   [001] ...1   246.444790: capable <-sock_setsockopt
>             sshd-662   [001] ...1   246.444793: new_capable
> <-sock_setsockopt
>      dbus-daemon-128   [000] .N.1   246.456307: capable <-SyS_epoll_ctl
>      dbus-daemon-128   [000] ...1   246.456611: new_capable <-SyS_epoll_ctl
> 
> 
> There is no limitation thanks to the use of the ftrace subsystem. We are
> just another user, i.e. another piece of code called in a loop for a
> particular fentry location.

Yes true. What I intended to mention is that: the trace is turned on
for "capable" then the function is patched. Eventually, once the patch
is finalized, there will be no trace log for "capable". Someone tracing
the function "capable", not aware of patching, may think that it has not
been invoked. The user, hence, is expected to start tracing
"new_capable". I think this should be documented.

What if someone turns on tracing for "capable" after it is patched?
Will it overwrite the slow/fast stub?


> 
>>> +/*
>>> + * The stub needs to modify the RIP value stored in struct pt_regs
>>> + * so that ftrace redirects the execution properly.
>>> + */
>>> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
>>> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
>>> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
>>> +{									\
>>> +	struct kgr_loc_caches *c = ops->private;			\
>>> +									\
>>> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
>>
>> Is there a race here? The per task kgr_in_progress is set after
>> the slow stub is registered in register_ftrace_function(). If the
>> patched function is called in between it will be redirected to new code.
> 
> Hmm, that looks strange. I will look into that and the other comments
> later (and comment separately). Thanks.
> 

-- 
Regards,
Aravinda

Re: [RFC 03/16] kgr: initial code

[Jiri Slaby]

On 05/14/2014 12:41 PM, Aravinda Prasad wrote:
> 
> 
> On Wednesday 14 May 2014 03:42 PM, Jiri Slaby wrote:
>> On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
>>> On Wednesday 30 April 2014 08:00 PM, Jiri Slaby wrote:
>>>> From: Jiri Kosina <jkosina@suse.cz>
>>>>
>>>> Provide initial implementation. We are now able to do ftrace-based
>>>> runtime patching of the kernel code.
>>>>
>>>> In addition to that, we will provide a kgr_patcher module in the next
>>>> patch to test the functionality.
>>>
>>> Hi Jiri,
>>>
>>> Interesting! I have couple of comments:
>>>
>>> I think with kgraft (also with kpatch, though have not looked into
>>> it yet), the patched function cannot be dynamically ftraced.
>>> Though dynamic ftrace can be enabled on the new code, the user is
>>> required to know the function label of the new code. This could
>>> potentially break existing scripts. I think this should be documented.
>>
>> Hi,
>>
>> of course that the functions can be traced. Look, I turned on tracing
>> for capable, then patched, then turned on tracing for new_capable (which
>> is the patched function). So now, trace shows:
>>   console-kit-dae-535   [001] ...1   181.729698: capable <-vt_ioctl
>>  console-kit-dae-539   [001] ...1   181.729741: capable <-vt_ioctl
>>  console-kit-dae-541   [000] .N.1   181.906014: capable <-vt_ioctl
>>          systemd-1     [001] ...1   181.937328: capable <-SyS_epoll_ctl
>>             sshd-662   [001] ...1   246.437561: capable <-sock_setsockopt
>>             sshd-662   [001] ...1   246.437564: new_capable
>> <-sock_setsockopt
>>             sshd-662   [001] ...1   246.444790: capable <-sock_setsockopt
>>             sshd-662   [001] ...1   246.444793: new_capable
>> <-sock_setsockopt
>>      dbus-daemon-128   [000] .N.1   246.456307: capable <-SyS_epoll_ctl
>>      dbus-daemon-128   [000] ...1   246.456611: new_capable <-SyS_epoll_ctl
>>
>>
>> There is no limitation thanks to the use of the ftrace subsystem. We are
>> just another user, i.e. another piece of code called in a loop for a
>> particular fentry location.
> 
> Yes true. What I intended to mention is that: the trace is turned on
> for "capable" then the function is patched. Eventually, once the patch
> is finalized, there will be no trace log for "capable". Someone tracing
> the function "capable", not aware of patching, may think that it has not
> been invoked. The user, hence, is expected to start tracing
> "new_capable". I think this should be documented.

As you can see in the trace log above, no. fentry of capable is still
traced (and new_capable is traced along)...

> What if someone turns on tracing for "capable" after it is patched?
> Will it overwrite the slow/fast stub?

Nope, it would look like in the example above.

thanks,
-- 
js
suse labs

Re: [RFC 03/16] kgr: initial code

[Aravinda Prasad]

On Wednesday 14 May 2014 04:14 PM, Jiri Slaby wrote:
> On 05/14/2014 12:41 PM, Aravinda Prasad wrote:
>>
>>
>> On Wednesday 14 May 2014 03:42 PM, Jiri Slaby wrote:
>>> On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
>>>> On Wednesday 30 April 2014 08:00 PM, Jiri Slaby wrote:
>>>>> From: Jiri Kosina <jkosina@suse.cz>
>>>>>
>>>>> Provide initial implementation. We are now able to do ftrace-based
>>>>> runtime patching of the kernel code.
>>>>>
>>>>> In addition to that, we will provide a kgr_patcher module in the next
>>>>> patch to test the functionality.
>>>>
>>>> Hi Jiri,
>>>>
>>>> Interesting! I have couple of comments:
>>>>
>>>> I think with kgraft (also with kpatch, though have not looked into
>>>> it yet), the patched function cannot be dynamically ftraced.
>>>> Though dynamic ftrace can be enabled on the new code, the user is
>>>> required to know the function label of the new code. This could
>>>> potentially break existing scripts. I think this should be documented.
>>>
>>> Hi,
>>>
>>> of course that the functions can be traced. Look, I turned on tracing
>>> for capable, then patched, then turned on tracing for new_capable (which
>>> is the patched function). So now, trace shows:
>>>   console-kit-dae-535   [001] ...1   181.729698: capable <-vt_ioctl
>>>  console-kit-dae-539   [001] ...1   181.729741: capable <-vt_ioctl
>>>  console-kit-dae-541   [000] .N.1   181.906014: capable <-vt_ioctl
>>>          systemd-1     [001] ...1   181.937328: capable <-SyS_epoll_ctl
>>>             sshd-662   [001] ...1   246.437561: capable <-sock_setsockopt
>>>             sshd-662   [001] ...1   246.437564: new_capable
>>> <-sock_setsockopt
>>>             sshd-662   [001] ...1   246.444790: capable <-sock_setsockopt
>>>             sshd-662   [001] ...1   246.444793: new_capable
>>> <-sock_setsockopt
>>>      dbus-daemon-128   [000] .N.1   246.456307: capable <-SyS_epoll_ctl
>>>      dbus-daemon-128   [000] ...1   246.456611: new_capable <-SyS_epoll_ctl
>>>
>>>
>>> There is no limitation thanks to the use of the ftrace subsystem. We are
>>> just another user, i.e. another piece of code called in a loop for a
>>> particular fentry location.
>>
>> Yes true. What I intended to mention is that: the trace is turned on
>> for "capable" then the function is patched. Eventually, once the patch
>> is finalized, there will be no trace log for "capable". Someone tracing
>> the function "capable", not aware of patching, may think that it has not
>> been invoked. The user, hence, is expected to start tracing
>> "new_capable". I think this should be documented.
> 
> As you can see in the trace log above, no. fentry of capable is still
> traced (and new_capable is traced along)...
> 
>> What if someone turns on tracing for "capable" after it is patched?
>> Will it overwrite the slow/fast stub?
> 
> Nope, it would look like in the example above.

Thanks for clarifying.

So if I understand correctly, at some point, for very small duration,
we will have both slow stub and fast stub registered. It is possible
that both of them could be invoked and as mentioned in the code
that should not cause any problem.

Regards,
Aravinda

> 
> thanks,
> 

-- 
Regards,
Aravinda

Re: [RFC 03/16] kgr: initial code

[Jiri Slaby]

On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
>> +/*
>> + * The stub needs to modify the RIP value stored in struct pt_regs
>> + * so that ftrace redirects the execution properly.
>> + */
>> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
>> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
>> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
>> +{									\
>> +	struct kgr_loc_caches *c = ops->private;			\
>> +									\
>> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
> 
> Is there a race here? The per task kgr_in_progress is set after
> the slow stub is registered in register_ftrace_function(). If the
> patched function is called in between it will be redirected to new code.

Hi Aravinda!

Yes, you are right. I have just fixed by first setting the flag, then
start patching.

>> +		pr_info("kgr: slow stub: calling old code at %lx\n",	\
>> +				c->old);				\
>> +		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
>> +	} else {							\
>> +		pr_info("kgr: slow stub: calling new code at %lx\n",	\
>> +				c->new);				\
>> +		regs->ip = c->new;					\
>> +	}								\
> 
> [...]
> 
>> +static void kgr_mark_processes(void)
>> +{
>> +	struct task_struct *p;
>> +
>> +	read_lock(&tasklist_lock);
>> +	for_each_process(p)
>> +		task_thread_info(p)->kgr_in_progress = true;
> 
> Is there a need for memory barrier here (or in slow stub) to avoid
> the race if the slow stub is about to be called from a thread executing
> on another CPU?

Yes, it should. But since we convert it to bit-ops in 16/16, this is no
issue in the final implementation. I will fix the "initial code" though.

>> + * kgr_start_patching -- the entry for a kgraft patch
>> + * @patch: patch to be applied
>> + *
>> + * Start patching of code that is neither running in IRQ context nor
>> + * kernel thread.
>> + */
>> +int kgr_start_patching(const struct kgr_patch *patch)
>> +{
>> +	const struct kgr_patch_fun *const *patch_fun;
>> +
>> +	if (!kgr_initialized) {
>> +		pr_err("kgr: can't patch, not initialized\n");
>> +		return -EINVAL;
>> +	}
>> +
>> +	mutex_lock(&kgr_in_progress_lock);
>> +	if (kgr_in_progress) {
>> +		pr_err("kgr: can't patch, another patching not yet finalized\n");
>> +		mutex_unlock(&kgr_in_progress_lock);
>> +		return -EAGAIN;
>> +	}
>> +
>> +	for (patch_fun = patch->patches; *patch_fun; patch_fun++) {
>> +		int ret;
>> +
>> +		ret = kgr_patch_code(*patch_fun, false);
>> +		/*
>> +		 * In case any of the symbol resolutions in the set
>> +		 * has failed, patch all the previously replaced fentry
>> +		 * callsites back to nops and fail with grace
>> +		 */
>> +		if (ret < 0) {
>> +			for (; patch_fun >= patch->patches; patch_fun--)
>> +				unregister_ftrace_function((*patch_fun)->ftrace_ops_slow);
>> +			mutex_unlock(&kgr_in_progress_lock);
>> +			return ret;
>> +		}
>> +	}
>> +	kgr_in_progress = true;
>> +	kgr_patch = patch;
>> +	mutex_unlock(&kgr_in_progress_lock);
>> +
>> +	kgr_mark_processes();
>> +
>> +	/*
>> +	 * give everyone time to exit kernel, and check after a while
>> +	 */
> 
> I understand that the main intention of kgraft is to apply simple
> security fixes. However, if the patch changes the locking order,
> I think, there is a possibility of deadlock.
> 
> A thread which has not yet returned to user space calls the old
> code (not redirected to new code in slow stub) which might acquire
> the lock in the old order say lock1 followed by lock2. Meanwhile
> another thread which re-enters the kernel space, with kgr_in_progress
> unset, is redirected to the new code which acquires the lock in reverse
> order, say lock2 and lock1. This can cause deadlock.

Yes, this is a problem I was thinking of in another context yesterday.
Patching ->read or any other file_openrations which hold state over
user<->kernel switches may be a potential threat like above. The same as
in other implementations of live patching IMO. I put that on a TODO
checklist for creating patches. This has to be investigated manually
when creating a patch.

thanks for review,
-- 
js
suse labs

Re: [RFC 03/16] kgr: initial code

[Aravinda Prasad]

On Tuesday 20 May 2014 05:06 PM, Jiri Slaby wrote:
> On 05/14/2014 11:28 AM, Aravinda Prasad wrote:
>>> +/*
>>> + * The stub needs to modify the RIP value stored in struct pt_regs
>>> + * so that ftrace redirects the execution properly.
>>> + */
>>> +#define KGR_STUB_ARCH_SLOW(_name, _new_function)			\
>>> +static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_ip,	\
>>> +		struct ftrace_ops *ops, struct pt_regs *regs)		\
>>> +{									\
>>> +	struct kgr_loc_caches *c = ops->private;			\
>>> +									\
>>> +	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
>>
>> Is there a race here? The per task kgr_in_progress is set after
>> the slow stub is registered in register_ftrace_function(). If the
>> patched function is called in between it will be redirected to new code.
> 
> Hi Aravinda!
> 
> Yes, you are right. I have just fixed by first setting the flag, then
> start patching.
> 
>>> +		pr_info("kgr: slow stub: calling old code at %lx\n",	\
>>> +				c->old);				\
>>> +		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
>>> +	} else {							\
>>> +		pr_info("kgr: slow stub: calling new code at %lx\n",	\
>>> +				c->new);				\
>>> +		regs->ip = c->new;					\
>>> +	}								\
>>
>> [...]
>>
>>> +static void kgr_mark_processes(void)
>>> +{
>>> +	struct task_struct *p;
>>> +
>>> +	read_lock(&tasklist_lock);
>>> +	for_each_process(p)
>>> +		task_thread_info(p)->kgr_in_progress = true;
>>
>> Is there a need for memory barrier here (or in slow stub) to avoid
>> the race if the slow stub is about to be called from a thread executing
>> on another CPU?
> 
> Yes, it should. But since we convert it to bit-ops in 16/16, this is no
> issue in the final implementation. I will fix the "initial code" though.

Yes. I see that in 16/16. Thanks.

> 
>>> + * kgr_start_patching -- the entry for a kgraft patch
>>> + * @patch: patch to be applied
>>> + *
>>> + * Start patching of code that is neither running in IRQ context nor
>>> + * kernel thread.
>>> + */
>>> +int kgr_start_patching(const struct kgr_patch *patch)
>>> +{
>>> +	const struct kgr_patch_fun *const *patch_fun;
>>> +
>>> +	if (!kgr_initialized) {
>>> +		pr_err("kgr: can't patch, not initialized\n");
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	mutex_lock(&kgr_in_progress_lock);
>>> +	if (kgr_in_progress) {
>>> +		pr_err("kgr: can't patch, another patching not yet finalized\n");
>>> +		mutex_unlock(&kgr_in_progress_lock);
>>> +		return -EAGAIN;
>>> +	}
>>> +
>>> +	for (patch_fun = patch->patches; *patch_fun; patch_fun++) {
>>> +		int ret;
>>> +
>>> +		ret = kgr_patch_code(*patch_fun, false);
>>> +		/*
>>> +		 * In case any of the symbol resolutions in the set
>>> +		 * has failed, patch all the previously replaced fentry
>>> +		 * callsites back to nops and fail with grace
>>> +		 */
>>> +		if (ret < 0) {
>>> +			for (; patch_fun >= patch->patches; patch_fun--)
>>> +				unregister_ftrace_function((*patch_fun)->ftrace_ops_slow);
>>> +			mutex_unlock(&kgr_in_progress_lock);
>>> +			return ret;
>>> +		}
>>> +	}
>>> +	kgr_in_progress = true;
>>> +	kgr_patch = patch;
>>> +	mutex_unlock(&kgr_in_progress_lock);
>>> +
>>> +	kgr_mark_processes();
>>> +
>>> +	/*
>>> +	 * give everyone time to exit kernel, and check after a while
>>> +	 */
>>
>> I understand that the main intention of kgraft is to apply simple
>> security fixes. However, if the patch changes the locking order,
>> I think, there is a possibility of deadlock.
>>
>> A thread which has not yet returned to user space calls the old
>> code (not redirected to new code in slow stub) which might acquire
>> the lock in the old order say lock1 followed by lock2. Meanwhile
>> another thread which re-enters the kernel space, with kgr_in_progress
>> unset, is redirected to the new code which acquires the lock in reverse
>> order, say lock2 and lock1. This can cause deadlock.
> 
> Yes, this is a problem I was thinking of in another context yesterday.
> Patching ->read or any other file_openrations which hold state over
> user<->kernel switches may be a potential threat like above. The same as
> in other implementations of live patching IMO. I put that on a TODO

I agree. Meanwhile let me think on how to overcome this.

Regards,
Aravinda


> checklist for creating patches. This has to be investigated manually
> when creating a patch.
> 
> thanks for review,
> 

-- 
Regards,
Aravinda

Re: [RFC 03/16] kgr: initial code

[Jiri Kosina]

On Tue, 20 May 2014, Jiri Slaby wrote:

> Yes, this is a problem I was thinking of in another context yesterday.
> Patching ->read or any other file_openrations which hold state over
> user<->kernel switches may be a potential threat like above. The same as
> in other implementations of live patching IMO. I put that on a TODO
> checklist for creating patches. This has to be investigated manually
> when creating a patch.

Another thing that has to be handled very carefully is patching functions 
which are using self-modifying code (static keys), to make sure that the 
logic is not switched in the new function.

-- 
Jiri Kosina
SUSE Labs

[RFC 04/16] kgr: add testing kgraft patch

[Jiri Slaby]

This is intended to be a presentation of the kgraft engine, so it is
placed into samples/ directory.

It patches sys_iopl() and sys_capable() to print an additional message
to the original functionality.

Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 samples/Kconfig           |  4 ++
 samples/Makefile          |  3 +-
 samples/kgr/Makefile      |  1 +
 samples/kgr/kgr_patcher.c | 97 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 104 insertions(+), 1 deletion(-)
 create mode 100644 samples/kgr/Makefile
 create mode 100644 samples/kgr/kgr_patcher.c

diff --git a/samples/Kconfig b/samples/Kconfig
index 6181c2cc9ca0..a923510443de 100644
--- a/samples/Kconfig
+++ b/samples/Kconfig
@@ -55,6 +55,10 @@ config SAMPLE_KDB
 	  Build an example of how to dynamically add the hello
 	  command to the kdb shell.
 
+config SAMPLE_KGR_PATCHER
+	tristate "Build kgr patcher example -- loadable modules only"
+	depends on KGR && m
+
 config SAMPLE_RPMSG_CLIENT
 	tristate "Build rpmsg client sample -- loadable modules only"
 	depends on RPMSG && m
diff --git a/samples/Makefile b/samples/Makefile
index 1a60c62e2045..a141b219c019 100644
--- a/samples/Makefile
+++ b/samples/Makefile
@@ -1,4 +1,5 @@
 # Makefile for Linux samples code
 
 obj-$(CONFIG_SAMPLES)	+= kobject/ kprobes/ trace_events/ \
-			   hw_breakpoint/ kfifo/ kdb/ hidraw/ rpmsg/ seccomp/
+			   hw_breakpoint/ kfifo/ kdb/ kgr/ \
+			   hidraw/ rpmsg/ seccomp/
diff --git a/samples/kgr/Makefile b/samples/kgr/Makefile
new file mode 100644
index 000000000000..202eee7d050e
--- /dev/null
+++ b/samples/kgr/Makefile
@@ -0,0 +1 @@
+obj-$(CONFIG_SAMPLE_KGR_PATCHER) += kgr_patcher.o
diff --git a/samples/kgr/kgr_patcher.c b/samples/kgr/kgr_patcher.c
new file mode 100644
index 000000000000..828543e36f3f
--- /dev/null
+++ b/samples/kgr/kgr_patcher.c
@@ -0,0 +1,97 @@
+/*
+ * kgr_patcher -- just kick kgr infrastructure for test
+ *
+ *  Copyright (c) 2013-2014 SUSE
+ *   Authors: Jiri Kosina
+ *	      Vojtech Pavlik
+ *	      Jiri Slaby
+ */
+
+/*
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/kgr.h>
+#include <linux/kallsyms.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/capability.h>
+#include <linux/ptrace.h>
+
+#include <asm/processor.h>
+
+/*
+ * This all should be autogenerated from the patched sources
+ *
+ * IMPORTANT TODO: we have to handle cases where the new code is calling out
+ * into functions which are not exported to modules.
+ *
+ * This can either be handled by calling all such functions indirectly, i.e
+ * obtaining pointer from kallsyms in the stub (and transforming all callsites
+ * to do pointer dereference), or by modifying the kernel module linker.
+ */
+
+asmlinkage long kgr_new_sys_iopl(unsigned int level)
+{
+        struct pt_regs *regs = current_pt_regs();
+        unsigned int old = (regs->flags >> 12) & 3;
+        struct thread_struct *t = &current->thread;
+
+	printk(KERN_DEBUG "kgr-patcher: this is a new sys_iopl()\n");
+
+        if (level > 3)
+                return -EINVAL;
+        /* Trying to gain more privileges? */
+        if (level > old) {
+                if (!capable(CAP_SYS_RAWIO))
+                        return -EPERM;
+        }
+        regs->flags = (regs->flags & ~X86_EFLAGS_IOPL) | (level << 12);
+        t->iopl = level << 12;
+        set_iopl_mask(t->iopl);
+
+        return 0;
+}
+KGR_PATCHED_FUNCTION(patch, SyS_iopl, kgr_new_sys_iopl);
+
+static bool new_capable(int cap)
+{
+	printk(KERN_DEBUG "kgr-patcher: this is a new capable()\n");
+
+        return ns_capable(&init_user_ns, cap);
+}
+KGR_PATCHED_FUNCTION(patch, capable, new_capable);
+
+static const struct kgr_patch patch = {
+	.patches = {
+		KGR_PATCH(SyS_iopl),
+		KGR_PATCH(capable),
+		KGR_PATCH_END
+	}
+};
+
+static int __init kgr_patcher_init(void)
+{
+	/* removing not supported (yet?) */
+	__module_get(THIS_MODULE);
+	kgr_start_patching(&patch);
+	return 0;
+}
+
+static void __exit kgr_patcher_cleanup(void)
+{
+	/* extra care needs to be taken when freeing ftrace_ops->private */
+	printk(KERN_ERR "removing now buggy!\n");
+}
+
+module_init(kgr_patcher_init);
+module_exit(kgr_patcher_cleanup);
+
+MODULE_LICENSE("GPL");
+
-- 
1.9.2

Re: [RFC 04/16] kgr: add testing kgraft patch

[Pavel Machek]

Hi!

> This is intended to be a presentation of the kgraft engine, so it is
> placed into samples/ directory.
> 
> It patches sys_iopl() and sys_capable() to print an additional message
> to the original functionality.
> 
> Jiri Kosina <jkosina@suse.cz>

??

> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Ingo Molnar <mingo@redhat.com>

> +++ b/samples/kgr/kgr_patcher.c
> @@ -0,0 +1,97 @@
> +/*
> + * kgr_patcher -- just kick kgr infrastructure for test
> + *
> + *  Copyright (c) 2013-2014 SUSE
> + *   Authors: Jiri Kosina
> + *	      Vojtech Pavlik
> + *	      Jiri Slaby
> + */
> +
> +/*
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms of the GNU General Public License as published by the Free
> + * Software Foundation; either version 2 of the License, or (at your option)
> + * any later version.
> + */
> +
> +#include <linux/module.h>
> +#include <linux/kernel.h>
> +#include <linux/init.h>
> +#include <linux/kgr.h>
> +#include <linux/kallsyms.h>
> +#include <linux/sched.h>
> +#include <linux/types.h>
> +#include <linux/capability.h>
> +#include <linux/ptrace.h>
> +
> +#include <asm/processor.h>
> +
> +/*
> + * This all should be autogenerated from the patched sources
> + *
> + * IMPORTANT TODO: we have to handle cases where the new code is calling out
> + * into functions which are not exported to modules.

Is this todo still valid? Hey, its important :-).

> + * This can either be handled by calling all such functions indirectly, i.e
> + * obtaining pointer from kallsyms in the stub (and transforming all callsites
> + * to do pointer dereference), or by modifying the kernel module linker.
> + */
> +
> +asmlinkage long kgr_new_sys_iopl(unsigned int level)
> +{
> +        struct pt_regs *regs = current_pt_regs();
> +        unsigned int old = (regs->flags >> 12) & 3;
> +        struct thread_struct *t = &current->thread;
> +
> +	printk(KERN_DEBUG "kgr-patcher: this is a new sys_iopl()\n");

Tabs vs. spaces problem at more than one place.

> +KGR_PATCHED_FUNCTION(patch, SyS_iopl, kgr_new_sys_iopl);
> +
> +static bool new_capable(int cap)
> +{
> +	printk(KERN_DEBUG "kgr-patcher: this is a new capable()\n");
> +
> +        return ns_capable(&init_user_ns, cap);
> +}
> +KGR_PATCHED_FUNCTION(patch, capable, new_capable);

So for some reason when replacing sys_iopl, capable needs to be replaced, too?

> +static int __init kgr_patcher_init(void)
> +{
> +	/* removing not supported (yet?) */

So.. is it?
> +	__module_get(THIS_MODULE);
> +	kgr_start_patching(&patch);
> +	return 0;
> +}
> +
> +static void __exit kgr_patcher_cleanup(void)
> +{
> +	/* extra care needs to be taken when freeing ftrace_ops->private */
> +	printk(KERN_ERR "removing now buggy!\n");
> +}
> +

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [RFC 04/16] kgr: add testing kgraft patch

[Jiri Slaby]

On 05/06/2014 01:03 PM, Pavel Machek wrote:
> Hi!

Hi!

>> This is intended to be a presentation of the kgraft engine, so it is
>> placed into samples/ directory.
>>
>> It patches sys_iopl() and sys_capable() to print an additional message
>> to the original functionality.
>>
>> Jiri Kosina <jkosina@suse.cz>
> 
> ??

This was a messed up s-o-b line. Fixed already.

>> +KGR_PATCHED_FUNCTION(patch, SyS_iopl, kgr_new_sys_iopl);
>> +
>> +static bool new_capable(int cap)
>> +{
>> +	printk(KERN_DEBUG "kgr-patcher: this is a new capable()\n");
>> +
>> +        return ns_capable(&init_user_ns, cap);
>> +}
>> +KGR_PATCHED_FUNCTION(patch, capable, new_capable);
> 
> So for some reason when replacing sys_iopl, capable needs to be replaced, too?

Oh, no. This patch just demonstrates how to patch two functions at once.
They were chosen arbitrarily.

The rest has been fixed now.

thanks,
-- 
js
suse labs

[RFC 05/16] kgr: update Kconfig documentation

[Jiri Slaby]

This is based on Udo's text which were augmented in this patch.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Udo Seidel <udoseidel@gmx.de>
Cc: Vojtech Pavlik <vojtech@suse.cz>
---
 kernel/Kconfig.kgr | 3 +++
 samples/Kconfig    | 4 ++++
 2 files changed, 7 insertions(+)

diff --git a/kernel/Kconfig.kgr b/kernel/Kconfig.kgr
index af9125f27b6d..f66fa2c20656 100644
--- a/kernel/Kconfig.kgr
+++ b/kernel/Kconfig.kgr
@@ -5,3 +5,6 @@ config KGR
 	tristate "Kgr infrastructure"
 	depends on DYNAMIC_FTRACE_WITH_REGS
 	depends on HAVE_KGR
+	help
+	 Select this to enable kGraft online kernel patching. The
+	 runtime price is zero, so it is safe to say Y here.
diff --git a/samples/Kconfig b/samples/Kconfig
index a923510443de..29eba4b77812 100644
--- a/samples/Kconfig
+++ b/samples/Kconfig
@@ -58,6 +58,10 @@ config SAMPLE_KDB
 config SAMPLE_KGR_PATCHER
 	tristate "Build kgr patcher example -- loadable modules only"
 	depends on KGR && m
+	help
+	 Sample code to replace sys_iopl() and sys_capable() via
+	 kGraft. This is only for presentation purposes. It is safe to
+	 say Y here.
 
 config SAMPLE_RPMSG_CLIENT
 	tristate "Build rpmsg client sample -- loadable modules only"
-- 
1.9.2

Re: [RFC 05/16] kgr: update Kconfig documentation

[Randy Dunlap]

On 04/30/2014 07:30 AM, Jiri Slaby wrote:
> This is based on Udo's text which were augmented in this patch.
>
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Udo Seidel <udoseidel@gmx.de>
> Cc: Vojtech Pavlik <vojtech@suse.cz>
> ---
>   kernel/Kconfig.kgr | 3 +++
>   samples/Kconfig    | 4 ++++
>   2 files changed, 7 insertions(+)
>
> diff --git a/kernel/Kconfig.kgr b/kernel/Kconfig.kgr
> index af9125f27b6d..f66fa2c20656 100644
> --- a/kernel/Kconfig.kgr
> +++ b/kernel/Kconfig.kgr
> @@ -5,3 +5,6 @@ config KGR
>   	tristate "Kgr infrastructure"
>   	depends on DYNAMIC_FTRACE_WITH_REGS
>   	depends on HAVE_KGR
> +	help
> +	 Select this to enable kGraft online kernel patching. The
> +	 runtime price is zero, so it is safe to say Y here.

Please indent help text 2 spaces instead of 1 to be consistent and to
follow CodingStyle.

Also, I would prefer that this feature be referred to as kgraft (with
any capital letters that you prefer) instead of kgr.
kgr is too meaningless to me.

> diff --git a/samples/Kconfig b/samples/Kconfig
> index a923510443de..29eba4b77812 100644
> --- a/samples/Kconfig
> +++ b/samples/Kconfig
> @@ -58,6 +58,10 @@ config SAMPLE_KDB
>   config SAMPLE_KGR_PATCHER
>   	tristate "Build kgr patcher example -- loadable modules only"
>   	depends on KGR && m
> +	help
> +	 Sample code to replace sys_iopl() and sys_capable() via
> +	 kGraft. This is only for presentation purposes. It is safe to
> +	 say Y here.
>
>   config SAMPLE_RPMSG_CLIENT
>   	tristate "Build rpmsg client sample -- loadable modules only"
>


-- 
~Randy

[RFC 06/16] kgr: add Documentation

[Jiri Slaby]

This is a text provided by Udo and polished.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Udo Seidel <udoseidel@gmx.de>
---
 Documentation/kgr.txt | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)
 create mode 100644 Documentation/kgr.txt

diff --git a/Documentation/kgr.txt b/Documentation/kgr.txt
new file mode 100644
index 000000000000..5b62415641cf
--- /dev/null
+++ b/Documentation/kgr.txt
@@ -0,0 +1,26 @@
+Live Kernel Patching with kGraft
+--------------------------------
+
+Written by Udo Seidel <udoseidel at gmx dot de>
+Based on the Blog entry by Vojtech Pavlik
+
+April 2014
+
+kGraft's developement was started by the SUSE Labs. kGraft builds on
+technologies and ideas that are already present in the kernel: ftrace
+and its mcount-based reserved space in function headers, the
+INT3/IPI-NMI patching also used in jumplabels, and RCU-like update of
+code that does not require stopping the kernel. For more information
+about ftrace please checkout the Documentation shipped with the kernel
+or search for howtos and explanations on the Internet.
+
+A kGraft patch is a kernel module and fully relies on the in-kernel
+module loader to link the new code with the kernel.  Thanks to all
+that, the design can be nicely minimalistic.
+
+While kGraft is, by choice, limited to replacing whole functions and
+constants they reference, this does not limit the set of code patches
+that can be applied significantly.  kGraft offers tools to assist in
+creating the live patch modules, identifying which functions need to
+be replaced based on a patch, and creating the patch module source
+code. They are located in /tools/kgraft/.
-- 
1.9.2

Re: [RFC 06/16] kgr: add Documentation

[Pavel Machek]

Hi!

> This is a text provided by Udo and polished.
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Udo Seidel <udoseidel@gmx.de>
> ---
>  Documentation/kgr.txt | 26 ++++++++++++++++++++++++++
>  1 file changed, 26 insertions(+)
>  create mode 100644 Documentation/kgr.txt
> 
> diff --git a/Documentation/kgr.txt b/Documentation/kgr.txt
> new file mode 100644
> index 000000000000..5b62415641cf
> --- /dev/null
> +++ b/Documentation/kgr.txt
> @@ -0,0 +1,26 @@
> +Live Kernel Patching with kGraft
> +--------------------------------
> +
> +Written by Udo Seidel <udoseidel at gmx dot de>
> +Based on the Blog entry by Vojtech Pavlik
> +
> +April 2014
> +
> +kGraft's developement was started by the SUSE Labs. kGraft builds on
> +technologies and ideas that are already present in the kernel: ftrace
> +and its mcount-based reserved space in function headers, the
> +INT3/IPI-NMI patching also used in jumplabels, and RCU-like update of
> +code that does not require stopping the kernel. For more information
> +about ftrace please checkout the Documentation shipped with the kernel
> +or search for howtos and explanations on the Internet.

This should really provide filename in Documentation/ directory it is refering to.

> +A kGraft patch is a kernel module and fully relies on the in-kernel
> +module loader to link the new code with the kernel.  Thanks to all
> +that, the design can be nicely minimalistic.

I feel some more details would be nice here.

> +While kGraft is, by choice, limited to replacing whole functions and
> +constants they reference, this does not limit the set of code patches
> +that can be applied significantly.  kGraft offers tools to assist in
> +creating the live patch modules, identifying which functions need to
> +be replaced based on a patch, and creating the patch module source
> +code. They are located in /tools/kgraft/.

For what functions it does not work? Anything used in interrupt context?
What about assembly? What happens if that function uses &label to
do some magic?

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

kgr: dealing with optimalizations? (was Re: [RFC 06/16] kgr: add Documentat)ion

[Pavel Machek]

Hi!

Ok, one big question: you are replacing single functions, and assume that's ok, right?

But ... is it ok? gcc is allowed to do optimalization on whole source file (and whole source
tree with LTO).  How do you prevent situation where changing function foo() breaks optimalization
in function bar()?

Is turning off inter-procedural and inter-module optimalizations needed for kgraft to work?

Best regards,
											Pavel

On Tue 2014-05-06 13:03:12, Pavel Machek wrote:
> Hi!
> 
> > This is a text provided by Udo and polished.
> > 
> > Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> > Cc: Udo Seidel <udoseidel@gmx.de>
> > ---
> >  Documentation/kgr.txt | 26 ++++++++++++++++++++++++++
> >  1 file changed, 26 insertions(+)
> >  create mode 100644 Documentation/kgr.txt
> > 
> > diff --git a/Documentation/kgr.txt b/Documentation/kgr.txt
> > new file mode 100644
> > index 000000000000..5b62415641cf
> > --- /dev/null
> > +++ b/Documentation/kgr.txt
> > @@ -0,0 +1,26 @@
> > +Live Kernel Patching with kGraft
> > +--------------------------------
> > +
> > +Written by Udo Seidel <udoseidel at gmx dot de>
> > +Based on the Blog entry by Vojtech Pavlik
> > +
> > +April 2014
> > +
> > +kGraft's developement was started by the SUSE Labs. kGraft builds on
> > +technologies and ideas that are already present in the kernel: ftrace
> > +and its mcount-based reserved space in function headers, the
> > +INT3/IPI-NMI patching also used in jumplabels, and RCU-like update of
> > +code that does not require stopping the kernel. For more information
> > +about ftrace please checkout the Documentation shipped with the kernel
> > +or search for howtos and explanations on the Internet.
> 
> This should really provide filename in Documentation/ directory it is refering to.
> 
> > +A kGraft patch is a kernel module and fully relies on the in-kernel
> > +module loader to link the new code with the kernel.  Thanks to all
> > +that, the design can be nicely minimalistic.
> 
> I feel some more details would be nice here.
> 
> > +While kGraft is, by choice, limited to replacing whole functions and
> > +constants they reference, this does not limit the set of code patches
> > +that can be applied significantly.  kGraft offers tools to assist in
> > +creating the live patch modules, identifying which functions need to
> > +be replaced based on a patch, and creating the patch module source
> > +code. They are located in /tools/kgraft/.
> 
> For what functions it does not work? Anything used in interrupt context?
> What about assembly? What happens if that function uses &label to
> do some magic?
> 
> -- 
> (english) http://www.livejournal.com/~pavelmachek
> (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: kgr: dealing with optimalizations? (was Re: [RFC 06/16] kgr: add Documentat)ion

[Michael Matz]

Hi,

On Fri, 9 May 2014, Pavel Machek wrote:

> Ok, one big question: you are replacing single functions, and assume 
> that's ok, right?
> 
> But ... is it ok? gcc is allowed to do optimalization on whole source 
> file (and whole source tree with LTO).  How do you prevent situation 
> where changing function foo() breaks optimalization in function bar()?

If such situation (behaviour changes with optimization) happens it's 
either compiler or a source bug, so is of no direct concern to kgraft.  
If you want to exchange an inlined function foo you of course have to take 
care of actually exchanging all functions into which foo was inlined.  
For that you need the inline tree for all kernels for which you want to 
support kgrafting.

> Is turning off inter-procedural and inter-module optimalizations needed 
> for kgraft to work?

No.


Ciao,
Michael.

[RFC 07/16] kgr: trigger the first check earlier

[Jiri Slaby]

In 5 seconds, not 30. This speeds up the whole process in most
scenarios.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 kernel/kgr.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/kgr.c b/kernel/kgr.c
index 6f55c7654618..5e9a07faddb9 100644
--- a/kernel/kgr.c
+++ b/kernel/kgr.c
@@ -281,7 +281,7 @@ int kgr_start_patching(const struct kgr_patch *patch)
 	/*
 	 * give everyone time to exit kernel, and check after a while
 	 */
-	queue_delayed_work(kgr_wq, &kgr_work, KGR_TIMEOUT * HZ);
+	queue_delayed_work(kgr_wq, &kgr_work, 5 * HZ);
 
 	return 0;
 }
-- 
1.9.2

[RFC 08/16] kgr: sched.h, introduce kgr_task_safe helper

[Jiri Slaby]

To be used from some kthreads.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 include/linux/sched.h | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 25f54c79f757..afd5747bc7ff 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2969,6 +2969,15 @@ static inline void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 }
 #endif /* CONFIG_MM_OWNER */
 
+#ifdef CONFIG_KGR
+static inline void kgr_task_safe(struct task_struct *p)
+{
+	task_thread_info(p)->kgr_in_progress = false;
+}
+#else
+static inline void kgr_task_safe(struct task_struct *p) { }
+#endif /* CONFIG_KGR */
+
 static inline unsigned long task_rlimit(const struct task_struct *tsk,
 		unsigned int limit)
 {
-- 
1.9.2

[RFC 09/16] kgr: mark task_safe in some kthreads

[Jiri Slaby]

Some threads do not use kthread_should_stop. Before we enable a
kthread support in kgr, we must make sure all those mark themselves
safe explicitly.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Tejun Heo <tj@kernel.org>
---
 drivers/base/devtmpfs.c  | 1 +
 fs/jbd2/journal.c        | 2 ++
 fs/notify/mark.c         | 5 ++++-
 kernel/hung_task.c       | 5 ++++-
 kernel/kthread.c         | 3 +++
 kernel/rcu/tree.c        | 6 ++++--
 kernel/rcu/tree_plugin.h | 9 +++++++--
 kernel/workqueue.c       | 1 +
 8 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/drivers/base/devtmpfs.c b/drivers/base/devtmpfs.c
index 25798db14553..c7d52d1b8c9c 100644
--- a/drivers/base/devtmpfs.c
+++ b/drivers/base/devtmpfs.c
@@ -387,6 +387,7 @@ static int devtmpfsd(void *p)
 	sys_chroot(".");
 	complete(&setup_done);
 	while (1) {
+		kgr_task_safe(current);
 		spin_lock(&req_lock);
 		while (requests) {
 			struct req *req = requests;
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 67b8e303946c..1b9c4c2e014a 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -43,6 +43,7 @@
 #include <linux/backing-dev.h>
 #include <linux/bitops.h>
 #include <linux/ratelimit.h>
+#include <linux/sched.h>
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/jbd2.h>
@@ -260,6 +261,7 @@ loop:
 			write_lock(&journal->j_state_lock);
 		}
 		finish_wait(&journal->j_wait_commit, &wait);
+		kgr_task_safe(current);
 	}
 
 	jbd_debug(1, "kjournald2 wakes\n");
diff --git a/fs/notify/mark.c b/fs/notify/mark.c
index 923fe4a5f503..a74b6175e645 100644
--- a/fs/notify/mark.c
+++ b/fs/notify/mark.c
@@ -82,6 +82,7 @@
 #include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
+#include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/srcu.h>
@@ -355,7 +356,9 @@ static int fsnotify_mark_destroy(void *ignored)
 			fsnotify_put_mark(mark);
 		}
 
-		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
+		wait_event_interruptible(destroy_waitq, ({
+					kgr_task_safe(current);
+					!list_empty(&destroy_list); }));
 	}
 
 	return 0;
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 06bb1417b063..b5f85bff2509 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -14,6 +14,7 @@
 #include <linux/kthread.h>
 #include <linux/lockdep.h>
 #include <linux/export.h>
+#include <linux/sched.h>
 #include <linux/sysctl.h>
 #include <linux/utsname.h>
 #include <trace/events/sched.h>
@@ -227,8 +228,10 @@ static int watchdog(void *dummy)
 	for ( ; ; ) {
 		unsigned long timeout = sysctl_hung_task_timeout_secs;
 
-		while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
+		while (schedule_timeout_interruptible(timeout_jiffies(timeout))) {
+			kgr_task_safe(current);
 			timeout = sysctl_hung_task_timeout_secs;
+		}
 
 		if (atomic_xchg(&reset_hung_task, 0))
 			continue;
diff --git a/kernel/kthread.c b/kernel/kthread.c
index 9a130ec06f7a..08b979dad619 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -78,6 +78,8 @@ static struct kthread *to_live_kthread(struct task_struct *k)
  */
 bool kthread_should_stop(void)
 {
+	kgr_task_safe(current);
+
 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
 }
 EXPORT_SYMBOL(kthread_should_stop);
@@ -497,6 +499,7 @@ int kthreadd(void *unused)
 		if (list_empty(&kthread_create_list))
 			schedule();
 		__set_current_state(TASK_RUNNING);
+		kgr_task_safe(current);
 
 		spin_lock(&kthread_create_lock);
 		while (!list_empty(&kthread_create_list)) {
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 0c47e300210a..5dddedacfc06 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1593,9 +1593,10 @@ static int __noreturn rcu_gp_kthread(void *arg)
 			trace_rcu_grace_period(rsp->name,
 					       ACCESS_ONCE(rsp->gpnum),
 					       TPS("reqwait"));
-			wait_event_interruptible(rsp->gp_wq,
+			wait_event_interruptible(rsp->gp_wq, ({
+						 kgr_task_safe(current);
 						 ACCESS_ONCE(rsp->gp_flags) &
-						 RCU_GP_FLAG_INIT);
+						 RCU_GP_FLAG_INIT; }));
 			/* Locking provides needed memory barrier. */
 			if (rcu_gp_init(rsp))
 				break;
@@ -1626,6 +1627,7 @@ static int __noreturn rcu_gp_kthread(void *arg)
 					(!ACCESS_ONCE(rnp->qsmask) &&
 					 !rcu_preempt_blocked_readers_cgp(rnp)),
 					j);
+			kgr_task_safe(current);
 			/* Locking provides needed memory barriers. */
 			/* If grace period done, leave loop. */
 			if (!ACCESS_ONCE(rnp->qsmask) &&
diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index 962d1d589929..8b383003b228 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -27,6 +27,7 @@
 #include <linux/delay.h>
 #include <linux/gfp.h>
 #include <linux/oom.h>
+#include <linux/sched.h>
 #include <linux/smpboot.h>
 #include "../time/tick-internal.h"
 
@@ -1273,7 +1274,8 @@ static int rcu_boost_kthread(void *arg)
 	for (;;) {
 		rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
 		trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
-		rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
+		rcu_wait(({ kgr_task_safe(current);
+					rnp->boost_tasks || rnp->exp_tasks; }));
 		trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
 		rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
 		more2boost = rcu_boost(rnp);
@@ -2283,11 +2285,14 @@ static int rcu_nocb_kthread(void *arg)
 
 	/* Each pass through this loop invokes one batch of callbacks */
 	for (;;) {
+		kgr_task_safe(current);
 		/* If not polling, wait for next batch of callbacks. */
 		if (!rcu_nocb_poll) {
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("Sleep"));
-			wait_event_interruptible(rdp->nocb_wq, rdp->nocb_head);
+			wait_event_interruptible(rdp->nocb_wq, ({
+						kgr_task_safe(current);
+						rdp->nocb_head; }));
 			/* Memory barrier provide by xchg() below. */
 		} else if (firsttime) {
 			firsttime = 0;
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 0ee63af30bd1..4b89f1dc0dd8 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2369,6 +2369,7 @@ sleep:
 	__set_current_state(TASK_INTERRUPTIBLE);
 	spin_unlock_irq(&pool->lock);
 	schedule();
+	kgr_task_safe(current);
 	goto woke_up;
 }
 
-- 
1.9.2

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Greg Kroah-Hartman]

On Wed, Apr 30, 2014 at 04:30:42PM +0200, Jiri Slaby wrote:
> Some threads do not use kthread_should_stop. Before we enable a
> kthread support in kgr, we must make sure all those mark themselves
> safe explicitly.
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: "Theodore Ts'o" <tytso@mit.edu>
> Cc: Dipankar Sarma <dipankar@in.ibm.com>
> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Cc: Tejun Heo <tj@kernel.org>
> ---
>  drivers/base/devtmpfs.c  | 1 +

Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Paul E. McKenney]

On Wed, Apr 30, 2014 at 04:30:42PM +0200, Jiri Slaby wrote:
> Some threads do not use kthread_should_stop. Before we enable a
> kthread support in kgr, we must make sure all those mark themselves
> safe explicitly.

Would it make sense to bury kgr_task_safe() in wait_event_interruptible()
and friends?  The kgr_task_safe() implementation looks pretty lightweight,
so it should not be a performance problem.

One reason might this might be a bad idea is that there are calls to
wait_event_interruptible() all over the place, which might therefore
constrain where grafting could be safely done.  That would be fair enough,
but does that also imply new constraints on where kthread_should_stop()
can be invoked?  Any new constraints might not be a big deal given that
a very large fraction of the kthreads (and maybe all of them) invoke
kthread_should_stop() from their top-level function, but would be good
to call out.

So, what is the story?

							Thanx, Paul

> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: "Theodore Ts'o" <tytso@mit.edu>
> Cc: Dipankar Sarma <dipankar@in.ibm.com>
> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Cc: Tejun Heo <tj@kernel.org>
> ---
>  drivers/base/devtmpfs.c  | 1 +
>  fs/jbd2/journal.c        | 2 ++
>  fs/notify/mark.c         | 5 ++++-
>  kernel/hung_task.c       | 5 ++++-
>  kernel/kthread.c         | 3 +++
>  kernel/rcu/tree.c        | 6 ++++--
>  kernel/rcu/tree_plugin.h | 9 +++++++--
>  kernel/workqueue.c       | 1 +
>  8 files changed, 26 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/base/devtmpfs.c b/drivers/base/devtmpfs.c
> index 25798db14553..c7d52d1b8c9c 100644
> --- a/drivers/base/devtmpfs.c
> +++ b/drivers/base/devtmpfs.c
> @@ -387,6 +387,7 @@ static int devtmpfsd(void *p)
>  	sys_chroot(".");
>  	complete(&setup_done);
>  	while (1) {
> +		kgr_task_safe(current);
>  		spin_lock(&req_lock);
>  		while (requests) {
>  			struct req *req = requests;
> diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
> index 67b8e303946c..1b9c4c2e014a 100644
> --- a/fs/jbd2/journal.c
> +++ b/fs/jbd2/journal.c
> @@ -43,6 +43,7 @@
>  #include <linux/backing-dev.h>
>  #include <linux/bitops.h>
>  #include <linux/ratelimit.h>
> +#include <linux/sched.h>
> 
>  #define CREATE_TRACE_POINTS
>  #include <trace/events/jbd2.h>
> @@ -260,6 +261,7 @@ loop:
>  			write_lock(&journal->j_state_lock);
>  		}
>  		finish_wait(&journal->j_wait_commit, &wait);
> +		kgr_task_safe(current);
>  	}
> 
>  	jbd_debug(1, "kjournald2 wakes\n");
> diff --git a/fs/notify/mark.c b/fs/notify/mark.c
> index 923fe4a5f503..a74b6175e645 100644
> --- a/fs/notify/mark.c
> +++ b/fs/notify/mark.c
> @@ -82,6 +82,7 @@
>  #include <linux/kthread.h>
>  #include <linux/module.h>
>  #include <linux/mutex.h>
> +#include <linux/sched.h>
>  #include <linux/slab.h>
>  #include <linux/spinlock.h>
>  #include <linux/srcu.h>
> @@ -355,7 +356,9 @@ static int fsnotify_mark_destroy(void *ignored)
>  			fsnotify_put_mark(mark);
>  		}
> 
> -		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
> +		wait_event_interruptible(destroy_waitq, ({
> +					kgr_task_safe(current);
> +					!list_empty(&destroy_list); }));
>  	}
> 
>  	return 0;
> diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> index 06bb1417b063..b5f85bff2509 100644
> --- a/kernel/hung_task.c
> +++ b/kernel/hung_task.c
> @@ -14,6 +14,7 @@
>  #include <linux/kthread.h>
>  #include <linux/lockdep.h>
>  #include <linux/export.h>
> +#include <linux/sched.h>
>  #include <linux/sysctl.h>
>  #include <linux/utsname.h>
>  #include <trace/events/sched.h>
> @@ -227,8 +228,10 @@ static int watchdog(void *dummy)
>  	for ( ; ; ) {
>  		unsigned long timeout = sysctl_hung_task_timeout_secs;
> 
> -		while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
> +		while (schedule_timeout_interruptible(timeout_jiffies(timeout))) {
> +			kgr_task_safe(current);
>  			timeout = sysctl_hung_task_timeout_secs;
> +		}
> 
>  		if (atomic_xchg(&reset_hung_task, 0))
>  			continue;
> diff --git a/kernel/kthread.c b/kernel/kthread.c
> index 9a130ec06f7a..08b979dad619 100644
> --- a/kernel/kthread.c
> +++ b/kernel/kthread.c
> @@ -78,6 +78,8 @@ static struct kthread *to_live_kthread(struct task_struct *k)
>   */
>  bool kthread_should_stop(void)
>  {
> +	kgr_task_safe(current);
> +
>  	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
>  }
>  EXPORT_SYMBOL(kthread_should_stop);
> @@ -497,6 +499,7 @@ int kthreadd(void *unused)
>  		if (list_empty(&kthread_create_list))
>  			schedule();
>  		__set_current_state(TASK_RUNNING);
> +		kgr_task_safe(current);
> 
>  		spin_lock(&kthread_create_lock);
>  		while (!list_empty(&kthread_create_list)) {
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 0c47e300210a..5dddedacfc06 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1593,9 +1593,10 @@ static int __noreturn rcu_gp_kthread(void *arg)
>  			trace_rcu_grace_period(rsp->name,
>  					       ACCESS_ONCE(rsp->gpnum),
>  					       TPS("reqwait"));
> -			wait_event_interruptible(rsp->gp_wq,
> +			wait_event_interruptible(rsp->gp_wq, ({
> +						 kgr_task_safe(current);
>  						 ACCESS_ONCE(rsp->gp_flags) &
> -						 RCU_GP_FLAG_INIT);
> +						 RCU_GP_FLAG_INIT; }));
>  			/* Locking provides needed memory barrier. */
>  			if (rcu_gp_init(rsp))
>  				break;
> @@ -1626,6 +1627,7 @@ static int __noreturn rcu_gp_kthread(void *arg)
>  					(!ACCESS_ONCE(rnp->qsmask) &&
>  					 !rcu_preempt_blocked_readers_cgp(rnp)),
>  					j);
> +			kgr_task_safe(current);
>  			/* Locking provides needed memory barriers. */
>  			/* If grace period done, leave loop. */
>  			if (!ACCESS_ONCE(rnp->qsmask) &&
> diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> index 962d1d589929..8b383003b228 100644
> --- a/kernel/rcu/tree_plugin.h
> +++ b/kernel/rcu/tree_plugin.h
> @@ -27,6 +27,7 @@
>  #include <linux/delay.h>
>  #include <linux/gfp.h>
>  #include <linux/oom.h>
> +#include <linux/sched.h>
>  #include <linux/smpboot.h>
>  #include "../time/tick-internal.h"
> 
> @@ -1273,7 +1274,8 @@ static int rcu_boost_kthread(void *arg)
>  	for (;;) {
>  		rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
>  		trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
> -		rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
> +		rcu_wait(({ kgr_task_safe(current);
> +					rnp->boost_tasks || rnp->exp_tasks; }));
>  		trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
>  		rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
>  		more2boost = rcu_boost(rnp);
> @@ -2283,11 +2285,14 @@ static int rcu_nocb_kthread(void *arg)
> 
>  	/* Each pass through this loop invokes one batch of callbacks */
>  	for (;;) {
> +		kgr_task_safe(current);
>  		/* If not polling, wait for next batch of callbacks. */
>  		if (!rcu_nocb_poll) {
>  			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
>  					    TPS("Sleep"));
> -			wait_event_interruptible(rdp->nocb_wq, rdp->nocb_head);
> +			wait_event_interruptible(rdp->nocb_wq, ({
> +						kgr_task_safe(current);
> +						rdp->nocb_head; }));
>  			/* Memory barrier provide by xchg() below. */
>  		} else if (firsttime) {
>  			firsttime = 0;
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 0ee63af30bd1..4b89f1dc0dd8 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -2369,6 +2369,7 @@ sleep:
>  	__set_current_state(TASK_INTERRUPTIBLE);
>  	spin_unlock_irq(&pool->lock);
>  	schedule();
> +	kgr_task_safe(current);
>  	goto woke_up;
>  }
> 
> -- 
> 1.9.2
> 

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Vojtech Pavlik]

On Wed, Apr 30, 2014 at 09:55:32AM -0700, Paul E. McKenney wrote:
> On Wed, Apr 30, 2014 at 04:30:42PM +0200, Jiri Slaby wrote:
> > Some threads do not use kthread_should_stop. Before we enable a
> > kthread support in kgr, we must make sure all those mark themselves
> > safe explicitly.
> 
> Would it make sense to bury kgr_task_safe() in wait_event_interruptible()
> and friends?  The kgr_task_safe() implementation looks pretty lightweight,
> so it should not be a performance problem.

For userspace tasks, the kGraft in progress flag is cleared when
entering or exiting userspace. At that point it is safe to switch the
task to a post-patch world view.

For kernel threads, it's a bit more complicated: They never exit the
kernel, they keep executing within the kernel continuously. The
kgr_task_safe() call is thus inserted at a location within the main loop
where a 'new loop' begins - where there are no dependencies on results
of calls of functions from the previous loop.

Hence, putting kgr_task_safe() into every wait_event_interruptible()
wouldn't work, only a few of them are at that strategic spot where a
'new loop' can be indicated to kGraft.

The reason kgr_task_safe() is called from within the condition
evaluation statement in wait_event_interruptible() in this patch is
because we want it to be called as soon as a new loop begins - even if
that loop is empty because the condition to stop waiting has not been
met.

This also means that kGraft currently cannot patch the main loops of
kernel threads themselves as the thread of execution never exits them.

Jiří (Slabý) has some ideas about how to do without calling
kgr_task_safe() from within the kernel thread main loops, but for now,
the goal is to keep things simple and easy to understand.

> One reason might this might be a bad idea is that there are calls to
> wait_event_interruptible() all over the place, which might therefore
> constrain where grafting could be safely done.  That would be fair enough,
> but does that also imply new constraints on where kthread_should_stop()
> can be invoked?  Any new constraints might not be a big deal given that
> a very large fraction of the kthreads (and maybe all of them) invoke
> kthread_should_stop() from their top-level function, but would be good
> to call out.

> So, what is the story?

kGraft currently assumes that kthread_should_stop() is always in a part
of the main loop which doesn't carry over effect dependencies from the
previous iteration. This is currently true for all the uses of
kthread_should_stop(), but indeed it is an additional constraint for the
future.


Vojtech

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Paul E. McKenney]

On Wed, Apr 30, 2014 at 08:33:27PM +0200, Vojtech Pavlik wrote:
> On Wed, Apr 30, 2014 at 09:55:32AM -0700, Paul E. McKenney wrote:
> > On Wed, Apr 30, 2014 at 04:30:42PM +0200, Jiri Slaby wrote:
> > > Some threads do not use kthread_should_stop. Before we enable a
> > > kthread support in kgr, we must make sure all those mark themselves
> > > safe explicitly.
> > 
> > Would it make sense to bury kgr_task_safe() in wait_event_interruptible()
> > and friends?  The kgr_task_safe() implementation looks pretty lightweight,
> > so it should not be a performance problem.
> 
> For userspace tasks, the kGraft in progress flag is cleared when
> entering or exiting userspace. At that point it is safe to switch the
> task to a post-patch world view.
> 
> For kernel threads, it's a bit more complicated: They never exit the
> kernel, they keep executing within the kernel continuously. The
> kgr_task_safe() call is thus inserted at a location within the main loop
> where a 'new loop' begins - where there are no dependencies on results
> of calls of functions from the previous loop.
> 
> Hence, putting kgr_task_safe() into every wait_event_interruptible()
> wouldn't work, only a few of them are at that strategic spot where a
> 'new loop' can be indicated to kGraft.
> 
> The reason kgr_task_safe() is called from within the condition
> evaluation statement in wait_event_interruptible() in this patch is
> because we want it to be called as soon as a new loop begins - even if
> that loop is empty because the condition to stop waiting has not been
> met.
> 
> This also means that kGraft currently cannot patch the main loops of
> kernel threads themselves as the thread of execution never exits them.
> 
> Jiří (Slabý) has some ideas about how to do without calling
> kgr_task_safe() from within the kernel thread main loops, but for now,
> the goal is to keep things simple and easy to understand.

OK, from an RCU perspective:

Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

> > One reason might this might be a bad idea is that there are calls to
> > wait_event_interruptible() all over the place, which might therefore
> > constrain where grafting could be safely done.  That would be fair enough,
> > but does that also imply new constraints on where kthread_should_stop()
> > can be invoked?  Any new constraints might not be a big deal given that
> > a very large fraction of the kthreads (and maybe all of them) invoke
> > kthread_should_stop() from their top-level function, but would be good
> > to call out.
> 
> > So, what is the story?
> 
> kGraft currently assumes that kthread_should_stop() is always in a part
> of the main loop which doesn't carry over effect dependencies from the
> previous iteration. This is currently true for all the uses of
> kthread_should_stop(), but indeed it is an additional constraint for the
> future.

Got it.  It would be good to document this.  ;-)

							Thanx, Paul

> Vojtech
> 

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

On Wed, Apr 30, 2014 at 04:30:42PM +0200, Jiri Slaby wrote:
> Some threads do not use kthread_should_stop. Before we enable a

Haven't really following kgraft development but is it safe to assume
that all kthread_should_stop() usages are clean side-effect-less
boundaries?  If so, why is that property guaranteed?  Is there any
mechanism for sanity checks?  Maybe I'm just failing to understand how
the whole thing is supposed to work but this looks like it could
devolve into something more broken than the freezer which we haven't
fully recovered from yet.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Jiri Kosina]

On Thu, 1 May 2014, Tejun Heo wrote:

> > Some threads do not use kthread_should_stop. Before we enable a
> 
> Haven't really following kgraft development but is it safe to assume
> that all kthread_should_stop() usages are clean side-effect-less
> boundaries?  If so, why is that property guaranteed?  Is there any
> mechanism for sanity checks?  Maybe I'm just failing to understand how
> the whole thing is supposed to work but this looks like it could
> devolve into something more broken than the freezer which we haven't
> fully recovered from yet.

Hi Tejun,

first, thanks a lot for review.

I agree that this expectation might really somewhat implicit and is not 
probably properly documented anywhere. The basic observation is "whenever 
kthread_should_stop() is being called, all data structures are in a 
consistent state and don't need any further updates in order to achieve 
consistency, because we can exit the loop immediately here", as 
kthread_should_stop() is the very last thing every freezable kernel thread 
is calling before starting a new iteration.

For the sake of collecting data points -- do you happen to have any 
counter-example to the assumption?

Thanks,

-- 
Jiri Kosina
SUSE Labs

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hello, Jiri.

On Thu, May 01, 2014 at 10:17:44PM +0200, Jiri Kosina wrote:
> I agree that this expectation might really somewhat implicit and is not 
> probably properly documented anywhere. The basic observation is "whenever 
> kthread_should_stop() is being called, all data structures are in a 
> consistent state and don't need any further updates in order to achieve 
> consistency, because we can exit the loop immediately here", as 
> kthread_should_stop() is the very last thing every freezable kernel thread 

But kthread_should_stop() doesn't necessarily imply that "we can exit
the loop *immediately*" at all.  It just indicates that it should
terminate in finite amount of time.  I don't think it'd be too
difficult to find cases where kthreads do some stuff before returning
after testing kthread_should_stop().  e.g. after pending changes,
workqueue rescuers do one final loop over pending work items after
kthread_should_stop() tests positive to ensure empty queue on exit.
Please note that there's no expectation of discontinuity over the
test.  The users may carry over any state across the test as they see
fit.

> is calling before starting a new iteration.
> 
> For the sake of collecting data points -- do you happen to have any 
> counter-example to the assumption?

Just grep for kthread_should_stop() and look for the ones which
doesn't immediately perform return?  I think there are more which
don't return *immediately*.  You'd have to audit each and everyone to
determine that they don't carry over states across the test.  Most
will hopefully be trivial but not all.  More importantly, sounds like
a maintenance nightmare to me without any means to guarantee, or even
reasonably increase, correctness.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

On Thu, May 01, 2014 at 05:02:42PM -0400, Tejun Heo wrote:
> Hello, Jiri.
> 
> On Thu, May 01, 2014 at 10:17:44PM +0200, Jiri Kosina wrote:
> > I agree that this expectation might really somewhat implicit and is not 
> > probably properly documented anywhere. The basic observation is "whenever 
> > kthread_should_stop() is being called, all data structures are in a 
> > consistent state and don't need any further updates in order to achieve 
> > consistency, because we can exit the loop immediately here", as 
> > kthread_should_stop() is the very last thing every freezable kernel thread 
> 
> But kthread_should_stop() doesn't necessarily imply that "we can exit
> the loop *immediately*" at all.  It just indicates that it should
> terminate in finite amount of time.  I don't think it'd be too

Just a bit of addition.  Please note that kthread_should_stop(), along
with the freezer test, is actually trickier than it seems.  It's very
easy to write code which works most of the time but misses wake up
from kill when the timing is just right (or wrong).  It should be
interlocked with set_current_state() and other related queueing data
structure accesses.  This was several years ago but when I audited
most kthread users in kernel, especially in combination with the
freezer test which also has similar requirement, surprising percentage
of users (at least several tens of pct) were getting it slightly
wrong, so kthread_should_stop() really isn't used as "we can exit
*immediately*".  It just isn't that simple.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Jiri Slaby]

Hi Tejun,

On 05/01/2014 11:09 PM, Tejun Heo wrote:
> On Thu, May 01, 2014 at 05:02:42PM -0400, Tejun Heo wrote:
>> Hello, Jiri.
>>
>> On Thu, May 01, 2014 at 10:17:44PM +0200, Jiri Kosina wrote:
>>> I agree that this expectation might really somewhat implicit and is not 
>>> probably properly documented anywhere. The basic observation is "whenever 
>>> kthread_should_stop() is being called, all data structures are in a 
>>> consistent state and don't need any further updates in order to achieve 
>>> consistency, because we can exit the loop immediately here", as 
>>> kthread_should_stop() is the very last thing every freezable kernel thread 
>>
>> But kthread_should_stop() doesn't necessarily imply that "we can exit
>> the loop *immediately*" at all.  It just indicates that it should
>> terminate in finite amount of time.  I don't think it'd be too
> 
> Just a bit of addition.  Please note that kthread_should_stop(), along
> with the freezer test, is actually trickier than it seems.  It's very
> easy to write code which works most of the time but misses wake up
> from kill when the timing is just right (or wrong).  It should be
> interlocked with set_current_state() and other related queueing data
> structure accesses.  This was several years ago but when I audited
> most kthread users in kernel, especially in combination with the
> freezer test which also has similar requirement, surprising percentage
> of users (at least several tens of pct) were getting it slightly
> wrong, so kthread_should_stop() really isn't used as "we can exit
> *immediately*".  It just isn't that simple.

I see the worst case scenario. (For curious readers, it is for example
this kthread body:
while (1) {
  some_paired_call(); /* invokes pre-patched code */
  if (kthread_should_stop()) { /* kgraft switches to the new code */
    its_paired_function(); /* invokes patched code (wrong) */
    break;
  }
  its_paired_function(); /* the same (wrong) */
})

What to do with that now? We have come up with a couple possibilities.
Would you consider try_to_freeze() a good state-defining function? As it
is called when a kthread expects weird things can happen, it should be
safe to switch to the patched version in our opinion.

The other possibility is to patch every kthread loop (~300) and insert
kgr_task_safe() semi-manually at some proper place.

Or if you have any other suggestions we would appreciate that?

thanks,
-- 
js
suse labs

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Vojtech Pavlik]

On Wed, May 14, 2014 at 04:59:05PM +0200, Jiri Slaby wrote:

> I see the worst case scenario. (For curious readers, it is for example
> this kthread body:
> while (1) {
>   some_paired_call(); /* invokes pre-patched code */
>   if (kthread_should_stop()) { /* kgraft switches to the new code */
>     its_paired_function(); /* invokes patched code (wrong) */
>     break;
>   }
>   its_paired_function(); /* the same (wrong) */
> })
> 
> What to do with that now? We have come up with a couple possibilities.
> Would you consider try_to_freeze() a good state-defining function? As it
> is called when a kthread expects weird things can happen, it should be
> safe to switch to the patched version in our opinion.
> 
> The other possibility is to patch every kthread loop (~300) and insert
> kgr_task_safe() semi-manually at some proper place.
> 
> Or if you have any other suggestions we would appreciate that?

A heretic idea would be to convert all kernel threads into functions
that do not sleep and exit after a single iteration and are called from
a central kthread main loop function. That would get all of
kthread_should_stop() and try_to_freeze() and kgr_task_safe() nicely
into one place and at the same time put enough constraint on what the
thread function can do to prevent it from breaking the assumptions of
each of these calls. 

-- 
Vojtech Pavlik
SUSE Labs

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Paul E. McKenney]

On Wed, May 14, 2014 at 05:15:01PM +0200, Vojtech Pavlik wrote:
> On Wed, May 14, 2014 at 04:59:05PM +0200, Jiri Slaby wrote:
> 
> > I see the worst case scenario. (For curious readers, it is for example
> > this kthread body:
> > while (1) {
> >   some_paired_call(); /* invokes pre-patched code */
> >   if (kthread_should_stop()) { /* kgraft switches to the new code */
> >     its_paired_function(); /* invokes patched code (wrong) */
> >     break;
> >   }
> >   its_paired_function(); /* the same (wrong) */
> > })
> > 
> > What to do with that now? We have come up with a couple possibilities.
> > Would you consider try_to_freeze() a good state-defining function? As it
> > is called when a kthread expects weird things can happen, it should be
> > safe to switch to the patched version in our opinion.
> > 
> > The other possibility is to patch every kthread loop (~300) and insert
> > kgr_task_safe() semi-manually at some proper place.
> > 
> > Or if you have any other suggestions we would appreciate that?
> 
> A heretic idea would be to convert all kernel threads into functions
> that do not sleep and exit after a single iteration and are called from
> a central kthread main loop function. That would get all of
> kthread_should_stop() and try_to_freeze() and kgr_task_safe() nicely
> into one place and at the same time put enough constraint on what the
> thread function can do to prevent it from breaking the assumptions of
> each of these calls. 

Some substantial restructuring would be required for several of
the kthreads I am aware of, which contain kthread_should_stop()
inside loop bodies as well as on their conditions.  Also, a number
of them do things like wait_event() and the like, which would mean
that the central kthread main loop function would need to know
about the wait queues and wait conditions and handle them properly.
See for example rcu_torture_barrier_cbs() and rcu_torture_barrier()
in kernel/rcu/rcutorture.c [*], which wait on each other in order to
test RCU's rcu_barrier() primitives.

							Thanx, Paul

* In older kernels, this is kernel/rcu/torture.c or kernel/rcutorture.c.

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hello, Jiri, Vojtech.

On Wed, May 14, 2014 at 05:15:01PM +0200, Vojtech Pavlik wrote:
> On Wed, May 14, 2014 at 04:59:05PM +0200, Jiri Slaby wrote:
> > I see the worst case scenario. (For curious readers, it is for example
> > this kthread body:
> > while (1) {
> >   some_paired_call(); /* invokes pre-patched code */
> >   if (kthread_should_stop()) { /* kgraft switches to the new code */
> >     its_paired_function(); /* invokes patched code (wrong) */
> >     break;
> >   }
> >   its_paired_function(); /* the same (wrong) */
> > })
> > 
> > What to do with that now? We have come up with a couple possibilities.
> > Would you consider try_to_freeze() a good state-defining function? As it
> > is called when a kthread expects weird things can happen, it should be
> > safe to switch to the patched version in our opinion.
> > 
> > The other possibility is to patch every kthread loop (~300) and insert
> > kgr_task_safe() semi-manually at some proper place.
> > 
> > Or if you have any other suggestions we would appreciate that?
> 
> A heretic idea would be to convert all kernel threads into functions
> that do not sleep and exit after a single iteration and are called from
> a central kthread main loop function. That would get all of

Or converting them to use workqueues instead.  Converting majority of
kthread users to workqueue is probably a good idea regardless of this
because workqueues are far easier to get right and give clear
delineation boundary between execution instances between which it's
safe to freeze and shutdown (and possibly to patch the work function).
Let alone overall lower overhead.  I converted some and was planning
on converting most of them but never got around ot it.

> kthread_should_stop() and try_to_freeze() and kgr_task_safe() nicely
> into one place and at the same time put enough constraint on what the
> thread function can do to prevent it from breaking the assumptions of
> each of these calls. 

Yeah, the exactly same rationales for using workqueue over kthreads.
That said, even with most kthread users converted to workqueue, we'd
probably want something which can really enforce correctness for the
leftovers as long as we continue to expose kthread interface.  Ooh,
there's also kthread_worker thing which puts workqueue-like semantics
on top of kthreads which can be used for whatever which can't be
converted to workqueue due to special worker attributes or whatnot.

So, yeah, I think there are enough tools available to put enough
semantic meanings over how kthreads are used such that things like
freezer or hot-code patching can be implemented in the generic
framework rather than in hundred scattered places but it's likely to
take a substantial amount of work.  The upside is that conversions are
likely beneficial on their own so they can be pushed separately.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Mike Galbraith]

On Wed, 2014-05-14 at 12:32 -0400, Tejun Heo wrote: 
> Hello, Jiri, Vojtech.
> 
> On Wed, May 14, 2014 at 05:15:01PM +0200, Vojtech Pavlik wrote:
> > On Wed, May 14, 2014 at 04:59:05PM +0200, Jiri Slaby wrote:
> > > I see the worst case scenario. (For curious readers, it is for example
> > > this kthread body:
> > > while (1) {
> > >   some_paired_call(); /* invokes pre-patched code */
> > >   if (kthread_should_stop()) { /* kgraft switches to the new code */
> > >     its_paired_function(); /* invokes patched code (wrong) */
> > >     break;
> > >   }
> > >   its_paired_function(); /* the same (wrong) */
> > > })
> > > 
> > > What to do with that now? We have come up with a couple possibilities.
> > > Would you consider try_to_freeze() a good state-defining function? As it
> > > is called when a kthread expects weird things can happen, it should be
> > > safe to switch to the patched version in our opinion.
> > > 
> > > The other possibility is to patch every kthread loop (~300) and insert
> > > kgr_task_safe() semi-manually at some proper place.
> > > 
> > > Or if you have any other suggestions we would appreciate that?
> > 
> > A heretic idea would be to convert all kernel threads into functions
> > that do not sleep and exit after a single iteration and are called from
> > a central kthread main loop function. That would get all of
> 
> Or converting them to use workqueues instead.  Converting majority of
> kthread users to workqueue is probably a good idea regardless of this
> because workqueues are far easier to get right and give clear
> delineation boundary between execution instances between which it's
> safe to freeze and shutdown (and possibly to patch the work function).
> Let alone overall lower overhead.  I converted some and was planning
> on converting most of them but never got around ot it.

Hm.  The user would need to be able to identify and prioritize the
things, and have his settings stick.  Any dynamic pool business doing
allocations and/or munging priorities would be highly annoying.

I saw a case where dynamic workers inflicted a realtime regression on a
user (but what they were getting away with previously was.. horrid).

-Mike

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hey, Mike.

On Thu, May 15, 2014 at 05:53:57AM +0200, Mike Galbraith wrote:
> Hm.  The user would need to be able to identify and prioritize the

I suppose you mean userland by "the user"?

> things, and have his settings stick.  Any dynamic pool business doing
> allocations and/or munging priorities would be highly annoying.

There are some use cases where control over worker priority or other
attributes are necessary.  I'm not sure using kthread for that reason
is a good engineering choice tho.  Many of those cases end up being
accidental.

I think it'd be healthier to identify the use cases and then provide
proper interface for it.  Note that workqueue can now expose interface
to modify concurrency, priority and cpumask to userland which
writeback workers are already using.

In general, being restricted to using kthread internally for this
reason seems wrong to me.  It's too direct influence on the
implementation mechanism.

> I saw a case where dynamic workers inflicted a realtime regression on a
> user (but what they were getting away with previously was.. horrid).

Yeah, exactly.  It'd be far better to identify the use case properly
and provide the appropriate interface for it.  That said, even if it
really requires diddling with kthread directly from userland,
kthread_worker can still be used.  It's still one dedicated kthread
but with structured usage from kernel side so that infrastructure
features like freezer and possibly kgr can be implemented in a single
place rather than scattered around all over the place.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Mike Galbraith]

On Thu, 2014-05-15 at 00:06 -0400, Tejun Heo wrote: 
> Hey, Mike.
> 
> On Thu, May 15, 2014 at 05:53:57AM +0200, Mike Galbraith wrote:
> > Hm.  The user would need to be able to identify and prioritize the
> 
> I suppose you mean userland by "the user"?

Yeah.

> > things, and have his settings stick.  Any dynamic pool business doing
> > allocations and/or munging priorities would be highly annoying.
> 
> There are some use cases where control over worker priority or other
> attributes are necessary.  I'm not sure using kthread for that reason
> is a good engineering choice tho.  Many of those cases end up being
> accidental.

It's currently the only option.  For perfection, you'd have to have fine
grained deterministic yada yada throughout all paths, which is kinda out
for generic proxies, but it's a hell of a lot better than no control.

> I think it'd be healthier to identify the use cases and then provide
> proper interface for it.  Note that workqueue can now expose interface
> to modify concurrency, priority and cpumask to userland which
> writeback workers are already using.

You can't identify a specific thing, any/all of it can land on the
user's diner plate, so he should be able to make the decisions.  Power
to the user and all that, if he does something stupid, tuff titty.  User
getting to call the shots, and getting to keep the pieces when he fscks
it all up is wonderful stuff, lets kernel people off the hook :)

-Mike

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hello, Mike.

On Thu, May 15, 2014 at 06:46:18AM +0200, Mike Galbraith wrote:
> > I think it'd be healthier to identify the use cases and then provide
> > proper interface for it.  Note that workqueue can now expose interface
> > to modify concurrency, priority and cpumask to userland which
> > writeback workers are already using.
> 
> You can't identify a specific thing, any/all of it can land on the
> user's diner plate, so he should be able to make the decisions.  Power
> to the user and all that, if he does something stupid, tuff titty.  User
> getting to call the shots, and getting to keep the pieces when he fscks
> it all up is wonderful stuff, lets kernel people off the hook :)

Do we know specific kthreads which need to be exposed with this way?
If there are good enough reasons for specific ones, sure, but I don't
think "we can't change any of the kthreads because someone might be
diddling with it" is something we can sustain in the long term.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Mike Galbraith]

On Thu, 2014-05-15 at 00:50 -0400, Tejun Heo wrote: 
> Hello, Mike.
> 
> On Thu, May 15, 2014 at 06:46:18AM +0200, Mike Galbraith wrote:
> > > I think it'd be healthier to identify the use cases and then provide
> > > proper interface for it.  Note that workqueue can now expose interface
> > > to modify concurrency, priority and cpumask to userland which
> > > writeback workers are already using.
> > 
> > You can't identify a specific thing, any/all of it can land on the
> > user's diner plate, so he should be able to make the decisions.  Power
> > to the user and all that, if he does something stupid, tuff titty.  User
> > getting to call the shots, and getting to keep the pieces when he fscks
> > it all up is wonderful stuff, lets kernel people off the hook :)
> 
> Do we know specific kthreads which need to be exposed with this way?

Soft/hard irq threads and anything having to do with IO mostly, which
including workqueues.  I had to give the user a rather fugly global
prioritization option to let users more or less safely do the evil deeds
they want to and WILL do whether I agree with their motivation to do so
or not.  I tell all users that realtime is real dangerous, but if they
want to do that, it's their box, so by definition perfectly fine.

> If there are good enough reasons for specific ones, sure, but I don't
> think "we can't change any of the kthreads because someone might be
> diddling with it" is something we can sustain in the long term.

I think the opposite.  Taking any control the user has is pure evil.

-Mike

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hello, Mike.

On Thu, May 15, 2014 at 07:04:22AM +0200, Mike Galbraith wrote:
> On Thu, 2014-05-15 at 00:50 -0400, Tejun Heo wrote: 
> > Do we know specific kthreads which need to be exposed with this way?
> 
> Soft/hard irq threads and anything having to do with IO mostly, which
> including workqueues.  I had to give the user a rather fugly global
> prioritization option to let users more or less safely do the evil deeds
> they want to and WILL do whether I agree with their motivation to do so
> or not.  I tell all users that realtime is real dangerous, but if they
> want to do that, it's their box, so by definition perfectly fine.

Frederic is working on global settings for workqueues, so that'll
resolve some of those issues at least.

> > If there are good enough reasons for specific ones, sure, but I don't
> > think "we can't change any of the kthreads because someone might be
> > diddling with it" is something we can sustain in the long term.
> 
> I think the opposite.  Taking any control the user has is pure evil.

I'm not sure good/evil is the right frame to think about it.  Is
pooling worker threads evil in nature then?  Even when not doing so
leads to serious scalibilty issues and general poor utilization of
system resources?  User control, just like everything else, is one of
the many aspects to be evaluated and traded off, not something to
uphold religiously at all cost.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Mike Galbraith]

On Thu, 2014-05-15 at 01:09 -0400, Tejun Heo wrote: 
> Hello, Mike.
> 
> On Thu, May 15, 2014 at 07:04:22AM +0200, Mike Galbraith wrote:
> > On Thu, 2014-05-15 at 00:50 -0400, Tejun Heo wrote: 
> > > Do we know specific kthreads which need to be exposed with this way?
> > 
> > Soft/hard irq threads and anything having to do with IO mostly, which
> > including workqueues.  I had to give the user a rather fugly global
> > prioritization option to let users more or less safely do the evil deeds
> > they want to and WILL do whether I agree with their motivation to do so
> > or not.  I tell all users that realtime is real dangerous, but if they
> > want to do that, it's their box, so by definition perfectly fine.
> 
> Frederic is working on global settings for workqueues, so that'll
> resolve some of those issues at least.

Yeah, wrt what runs where for unbound workqueues, but not priority. 

> > > If there are good enough reasons for specific ones, sure, but I don't
> > > think "we can't change any of the kthreads because someone might be
> > > diddling with it" is something we can sustain in the long term.
> > 
> > I think the opposite.  Taking any control the user has is pure evil.
> 
> I'm not sure good/evil is the right frame to think about it.  Is
> pooling worker threads evil in nature then?

When there may be realtime consumers, yes to some extent, because it
inserts allocations he can't control directly into his world, but that's
the least of his worries.  The instant userspace depends upon any kernel
proxy the user has no control over, he instantly has a priority
inversion he can do nothing about.  This is exactly what happened that
prompted me to do fugly global hack.  User turned pet database piggies
loose as realtime tasks for his own reasons, misguided or not, they
depend upon worker threads and kjournald et al who he can control, but
kworker threads respawn as normal tasks which can and will end up under
high priority userspace tasks.  Worst case is box becomes dead, also
killing pet, best case is pet collapses to the floor in a quivering
heap.  Neither makes Joe User particularly happy.

-Mike

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Tejun Heo]

Hello, Mike.

On Thu, May 15, 2014 at 07:32:29AM +0200, Mike Galbraith wrote:
> On Thu, 2014-05-15 at 01:09 -0400, Tejun Heo wrote: 
> > > Soft/hard irq threads and anything having to do with IO mostly, which
> > > including workqueues.  I had to give the user a rather fugly global
> > > prioritization option to let users more or less safely do the evil deeds
> > > they want to and WILL do whether I agree with their motivation to do so
> > > or not.  I tell all users that realtime is real dangerous, but if they
> > > want to do that, it's their box, so by definition perfectly fine.
> > 
> > Frederic is working on global settings for workqueues, so that'll
> > resolve some of those issues at least.
> 
> Yeah, wrt what runs where for unbound workqueues, but not priority. 

Shouldn't be too difficult to extend it to cover priorities if
necessary once the infrastructure is in place.

> > > > If there are good enough reasons for specific ones, sure, but I don't
> > > > think "we can't change any of the kthreads because someone might be
> > > > diddling with it" is something we can sustain in the long term.
> > > 
> > > I think the opposite.  Taking any control the user has is pure evil.
> > 
> > I'm not sure good/evil is the right frame to think about it.  Is
> > pooling worker threads evil in nature then?
> 
> When there may be realtime consumers, yes to some extent, because it
> inserts allocations he can't control directly into his world, but that's
> the least of his worries.  The instant userspace depends upon any kernel
> proxy the user has no control over, he instantly has a priority
> inversion he can do nothing about.  This is exactly what happened that
> prompted me to do fugly global hack.  User turned pet database piggies
> loose as realtime tasks for his own reasons, misguided or not, they
> depend upon worker threads and kjournald et al who he can control, but
> kworker threads respawn as normal tasks which can and will end up under
> high priority userspace tasks.  Worst case is box becomes dead, also
> killing pet, best case is pet collapses to the floor in a quivering
> heap.  Neither makes Joe User particularly happy.

I'm not sure how much weight I can put on the specific use case.  Even
with the direct control that the user thought to have previously, the
use case was ripe with possibilities of breakage from any number of
reasons.  For example, there are driver paths which bounce to async
execution on IO exceptions (doesn't have to be hard errors) and setups
like the above would easily lock out exception handling and how's the
setup gonna work when the filesystems have to use dynamic pool of
workers as btrfs does?

The identified problem in the above case is allowing the kernel to
make reasonable forward progress even when RT processes don't concede
CPU cycles.  If that is a use case that needs to be supported, we
better engineer an appropriate solution for that.  Such solution
doesn't necessarily have to be advanced either.  Maybe all that's
necessary is marking the async mechanisms involved in IO path as such
(we already need to mark all workqueues involved in memory reclaim
path anyway) and provide a mechanism to make all of them RT when
directed.  It might be simple but still would be a concious
engineering decision.

I think the point I'm trying to make is that it isn't possible to
continue improving and maintaining the kernel with blanket
restrictions on internal details.  If certain things shouldn't be
done, we better find out the specific reasons; otherwise, it's
impossible to weight the pros and cons of different options and make a
reasonable choice or to find out ways to accomodate those restrictions
while still achieving the original goals.

Anyways, we're getting slightly off-topic and it seems like we'll have
to agree to disagree.

Thanks.

-- 
tejun

Re: [RFC 09/16] kgr: mark task_safe in some kthreads

[Mike Galbraith]

On Thu, 2014-05-15 at 02:05 -0400, Tejun Heo wrote:

> I'm not sure how much weight I can put on the specific use case.  Even
> with the direct control that the user thought to have previously, the
> use case was ripe with possibilities of breakage from any number of
> reasons.  For example, there are driver paths which bounce to async
> execution on IO exceptions (doesn't have to be hard errors) and setups
> like the above would easily lock out exception handling and how's the
> setup gonna work when the filesystems have to use dynamic pool of
> workers as btrfs does?

Oh yeah, this case isn't about _real_ realtime at all, it's just about
unmanageable priority inversion.  With hack, any/all kthreads spawning
will start life at the priority the user specified, so as long as he
doesn't prioritize userspace above that, he can do whatever evil deeds
he sees fit, and box will function as expected.
> The identified problem in the above case is allowing the kernel to
> make reasonable forward progress even when RT processes don't concede
> CPU cycles.

Not only, but yeah, mostly.

>   If that is a use case that needs to be supported, we
> better engineer an appropriate solution for that.  Such solution
> doesn't necessarily have to be advanced either.

My solution isn't the least bit sophisticated.  Dirt simple is usually
best anyway, and is enough for the cases I've encountered.

>   Maybe all that's
> necessary is marking the async mechanisms involved in IO path as such
> (we already need to mark all workqueues involved in memory reclaim
> path anyway) and provide a mechanism to make all of them RT when
> directed.  It might be simple but still would be a concious
> engineering decision.

You could do all singing/dancing PI boost thingy like RCU, but
personally, I hate that, disable it. 

> I think the point I'm trying to make is that it isn't possible to
> continue improving and maintaining the kernel with blanket
> restrictions on internal details.  If certain things shouldn't be
> done, we better find out the specific reasons; otherwise, it's
> impossible to weight the pros and cons of different options and make a
> reasonable choice or to find out ways to accomodate those restrictions
> while still achieving the original goals.
> 
> Anyways, we're getting slightly off-topic and it seems like we'll have
> to agree to disagree.

Hey, we agree! :)

-Mike

[RFC 10/16] kgr: kthreads support

[Jiri Slaby]

Wake up kthreads so that they cycle through kgr_task_safe either
by explicit call to it or implicitly via kthread_should_stop. This
ensures nobody should use the old version of the code and kgraft core
can push everybody to use the new version by switching to the fast
path.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 arch/x86/include/asm/kgr.h |  2 +-
 kernel/kgr.c               | 25 +++++++++++++++----------
 2 files changed, 16 insertions(+), 11 deletions(-)

diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
index 172f7b966bb5..49daa46243fc 100644
--- a/arch/x86/include/asm/kgr.h
+++ b/arch/x86/include/asm/kgr.h
@@ -13,7 +13,7 @@ static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_i
 {									\
 	struct kgr_loc_caches *c = ops->private;			\
 									\
-	if (task_thread_info(current)->kgr_in_progress && current->mm) {\
+	if (task_thread_info(current)->kgr_in_progress) {		\
 		pr_info("kgr: slow stub: calling old code at %lx\n",	\
 				c->old);				\
 		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
diff --git a/kernel/kgr.c b/kernel/kgr.c
index 5e9a07faddb9..ea63e857a78a 100644
--- a/kernel/kgr.c
+++ b/kernel/kgr.c
@@ -42,11 +42,7 @@ static bool kgr_still_patching(void)
 
 	read_lock(&tasklist_lock);
 	for_each_process(p) {
-		/*
-		 * TODO
-		 *   kernel thread codepaths not supported and silently ignored
-		 */
-		if (task_thread_info(p)->kgr_in_progress && p->mm) {
+		if (task_thread_info(p)->kgr_in_progress) {
 			pr_info("pid %d (%s) still in kernel after timeout\n",
 					p->pid, p->comm);
 			failed = true;
@@ -94,13 +90,23 @@ static void kgr_work_fn(struct work_struct *work)
 	mutex_unlock(&kgr_in_progress_lock);
 }
 
-static void kgr_mark_processes(void)
+static void kgr_handle_processes(void)
 {
 	struct task_struct *p;
 
 	read_lock(&tasklist_lock);
-	for_each_process(p)
+	for_each_process(p) {
 		task_thread_info(p)->kgr_in_progress = true;
+
+		/* wake up kthreads, they will clean the progress flag */
+		if (!p->mm) {
+			/*
+			 * this is incorrect for kthreads waiting still for
+			 * their first wake_up.
+			 */
+			wake_up_process(p);
+		}
+	}
 	read_unlock(&tasklist_lock);
 }
 
@@ -237,8 +243,7 @@ static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final)
  * kgr_start_patching -- the entry for a kgraft patch
  * @patch: patch to be applied
  *
- * Start patching of code that is neither running in IRQ context nor
- * kernel thread.
+ * Start patching of code that is not running in IRQ context.
  */
 int kgr_start_patching(const struct kgr_patch *patch)
 {
@@ -276,7 +281,7 @@ int kgr_start_patching(const struct kgr_patch *patch)
 	kgr_patch = patch;
 	mutex_unlock(&kgr_in_progress_lock);
 
-	kgr_mark_processes();
+	kgr_handle_processes();
 
 	/*
 	 * give everyone time to exit kernel, and check after a while
-- 
1.9.2

[RFC 11/16] kgr: handle irqs

[Jiri Slaby]

Introduce a per-cpu flag to check whether we should use the old or new
function in the slow stub. The new functoins is being used on a
processor after a scheduled function sets the flag via
schedule_on_each_cpu. Presumably this happens in the process context,
no irq is running. And protect the flag setting by disable interrupts
so that we 1) have a barrier and 2) no interrupt triggers while
setting the flag (but the set should be atomic anyway as it is bool).

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/include/asm/kgr.h |  4 +++-
 include/linux/kgr.h        |  5 +++--
 kernel/kgr.c               | 38 ++++++++++++++++++++++++++++++++------
 samples/kgr/kgr_patcher.c  |  2 +-
 4 files changed, 39 insertions(+), 10 deletions(-)

diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
index 49daa46243fc..f36661681b33 100644
--- a/arch/x86/include/asm/kgr.h
+++ b/arch/x86/include/asm/kgr.h
@@ -12,8 +12,10 @@ static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_i
 		struct ftrace_ops *ops, struct pt_regs *regs)		\
 {									\
 	struct kgr_loc_caches *c = ops->private;			\
+	bool irq = !!in_interrupt();					\
 									\
-	if (task_thread_info(current)->kgr_in_progress) {		\
+	if ((!irq && task_thread_info(current)->kgr_in_progress) ||	\
+			(irq && !*this_cpu_ptr(c->irq_use_new))) {	\
 		pr_info("kgr: slow stub: calling old code at %lx\n",	\
 				c->old);				\
 		regs->ip = c->old + MCOUNT_INSN_SIZE;			\
diff --git a/include/linux/kgr.h b/include/linux/kgr.h
index d72add7f3d5d..ebc6f5bc1ec1 100644
--- a/include/linux/kgr.h
+++ b/include/linux/kgr.h
@@ -19,7 +19,7 @@
 #endif
 
 struct kgr_patch {
-	char reserved;
+	bool __percpu *irq_use_new;
 	const struct kgr_patch_fun {
 		const char *name;
 		const char *new_name;
@@ -37,6 +37,7 @@ struct kgr_patch {
 struct kgr_loc_caches {
 	unsigned long old;
 	unsigned long new;
+	bool __percpu *irq_use_new;
 };
 
 #define KGR_PATCHED_FUNCTION(patch, _name, _new_function)			\
@@ -65,7 +66,7 @@ struct kgr_loc_caches {
 #define KGR_PATCH(name)		&__kgr_patch_ ## name
 #define KGR_PATCH_END		NULL
 
-extern int kgr_start_patching(const struct kgr_patch *);
+extern int kgr_start_patching(struct kgr_patch *);
 #endif /* CONFIG_KGR */
 
 #endif /* LINUX_KGR_H */
diff --git a/kernel/kgr.c b/kernel/kgr.c
index ea63e857a78a..ff5afaf6f0e7 100644
--- a/kernel/kgr.c
+++ b/kernel/kgr.c
@@ -18,6 +18,7 @@
 #include <linux/kallsyms.h>
 #include <linux/kgr.h>
 #include <linux/module.h>
+#include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/sort.h>
@@ -25,7 +26,8 @@
 #include <linux/types.h>
 #include <linux/workqueue.h>
 
-static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final);
+static int kgr_patch_code(const struct kgr_patch *patch,
+		const struct kgr_patch_fun *patch_fun, bool final);
 static void kgr_work_fn(struct work_struct *work);
 
 static struct workqueue_struct *kgr_wq;
@@ -57,7 +59,7 @@ static void kgr_finalize(void)
 	const struct kgr_patch_fun *const *patch_fun;
 
 	for (patch_fun = kgr_patch->patches; *patch_fun; patch_fun++) {
-		int ret = kgr_patch_code(*patch_fun, true);
+		int ret = kgr_patch_code(kgr_patch, *patch_fun, true);
 		/*
 		 * In case any of the symbol resolutions in the set
 		 * has failed, patch all the previously replaced fentry
@@ -67,6 +69,7 @@ static void kgr_finalize(void)
 			pr_err("kgr: finalize for %s failed, trying to continue\n",
 					(*patch_fun)->name);
 	}
+	free_percpu(kgr_patch->irq_use_new);
 }
 
 static void kgr_work_fn(struct work_struct *work)
@@ -139,6 +142,20 @@ static unsigned long kgr_get_fentry_loc(const char *f_name)
 	return fentry_loc;
 }
 
+static void kgr_handle_irq_cpu(struct work_struct *work)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	*this_cpu_ptr(kgr_patch->irq_use_new) = true;
+	local_irq_restore(flags);
+}
+
+static void kgr_handle_irqs(void)
+{
+	schedule_on_each_cpu(kgr_handle_irq_cpu);
+}
+
 static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
 {
 	struct kgr_loc_caches *caches;
@@ -184,7 +201,8 @@ static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
 	return 0;
 }
 
-static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final)
+static int kgr_patch_code(const struct kgr_patch *patch,
+		const struct kgr_patch_fun *patch_fun, bool final)
 {
 	struct ftrace_ops *new_ops;
 	struct kgr_loc_caches *caches;
@@ -205,6 +223,7 @@ static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final)
 
 	/* Flip the switch */
 	caches = new_ops->private;
+	caches->irq_use_new = patch->irq_use_new;
 	fentry_loc = caches->old;
 	err = ftrace_set_filter_ip(new_ops, fentry_loc, 0, 0);
 	if (err) {
@@ -243,9 +262,9 @@ static int kgr_patch_code(const struct kgr_patch_fun *patch_fun, bool final)
  * kgr_start_patching -- the entry for a kgraft patch
  * @patch: patch to be applied
  *
- * Start patching of code that is not running in IRQ context.
+ * Start patching of code.
  */
-int kgr_start_patching(const struct kgr_patch *patch)
+int kgr_start_patching(struct kgr_patch *patch)
 {
 	const struct kgr_patch_fun *const *patch_fun;
 
@@ -254,6 +273,12 @@ int kgr_start_patching(const struct kgr_patch *patch)
 		return -EINVAL;
 	}
 
+	patch->irq_use_new = alloc_percpu(bool);
+	if (!patch->irq_use_new) {
+		pr_err("kgr: can't patch, cannot allocate percpu data\n");
+		return -ENOMEM;
+	}
+
 	mutex_lock(&kgr_in_progress_lock);
 	if (kgr_in_progress) {
 		pr_err("kgr: can't patch, another patching not yet finalized\n");
@@ -264,7 +289,7 @@ int kgr_start_patching(const struct kgr_patch *patch)
 	for (patch_fun = patch->patches; *patch_fun; patch_fun++) {
 		int ret;
 
-		ret = kgr_patch_code(*patch_fun, false);
+		ret = kgr_patch_code(patch, *patch_fun, false);
 		/*
 		 * In case any of the symbol resolutions in the set
 		 * has failed, patch all the previously replaced fentry
@@ -281,6 +306,7 @@ int kgr_start_patching(const struct kgr_patch *patch)
 	kgr_patch = patch;
 	mutex_unlock(&kgr_in_progress_lock);
 
+	kgr_handle_irqs();
 	kgr_handle_processes();
 
 	/*
diff --git a/samples/kgr/kgr_patcher.c b/samples/kgr/kgr_patcher.c
index 828543e36f3f..b1465cff8d5b 100644
--- a/samples/kgr/kgr_patcher.c
+++ b/samples/kgr/kgr_patcher.c
@@ -68,7 +68,7 @@ static bool new_capable(int cap)
 }
 KGR_PATCHED_FUNCTION(patch, capable, new_capable);
 
-static const struct kgr_patch patch = {
+static struct kgr_patch patch = {
 	.patches = {
 		KGR_PATCH(SyS_iopl),
 		KGR_PATCH(capable),
-- 
1.9.2

[RFC 12/16] kgr: add tools

[Jiri Slaby]

These are a base which can be used for kgraft patch generation.

The code was provided by Michael

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Michael Matz <matz@suse.de>
---
 tools/Makefile                   |   13 +-
 tools/kgraft/Makefile            |   30 +
 tools/kgraft/README              |   50 +
 tools/kgraft/TODO                |   20 +
 tools/kgraft/app.c               |   35 +
 tools/kgraft/app.h               |    7 +
 tools/kgraft/create-kgrmodule.sh |   25 +
 tools/kgraft/create-stub.sh      |   53 +
 tools/kgraft/dwarf-inline-tree.c |  544 +++++
 tools/kgraft/dwarf_names.awk     |  126 ++
 tools/kgraft/dwarf_names.c       | 4366 ++++++++++++++++++++++++++++++++++++++
 tools/kgraft/dwarf_names.h       |   53 +
 tools/kgraft/extract-syms.sh     |   18 +
 tools/kgraft/it2rev.pl           |   40 +
 tools/kgraft/objcopy.diff        |  131 ++
 tools/kgraft/symlist             |    1 +
 16 files changed, 5507 insertions(+), 5 deletions(-)
 create mode 100644 tools/kgraft/Makefile
 create mode 100644 tools/kgraft/README
 create mode 100644 tools/kgraft/TODO
 create mode 100644 tools/kgraft/app.c
 create mode 100644 tools/kgraft/app.h
 create mode 100755 tools/kgraft/create-kgrmodule.sh
 create mode 100755 tools/kgraft/create-stub.sh
 create mode 100644 tools/kgraft/dwarf-inline-tree.c
 create mode 100644 tools/kgraft/dwarf_names.awk
 create mode 100644 tools/kgraft/dwarf_names.c
 create mode 100644 tools/kgraft/dwarf_names.h
 create mode 100755 tools/kgraft/extract-syms.sh
 create mode 100644 tools/kgraft/it2rev.pl
 create mode 100644 tools/kgraft/objcopy.diff
 create mode 100644 tools/kgraft/symlist

diff --git a/tools/Makefile b/tools/Makefile
index bcae806b0c39..d624e61606c4 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -8,6 +8,7 @@ help:
 	@echo '  cpupower   - a tool for all things x86 CPU power'
 	@echo '  firewire   - the userspace part of nosy, an IEEE-1394 traffic sniffer'
 	@echo '  hv         - tools used when in Hyper-V clients'
+	@echo '  kgraft     - the userspace part needed for online patching'
 	@echo '  lguest     - a minimal 32-bit x86 hypervisor'
 	@echo '  perf       - Linux performance measurement and analysis tool'
 	@echo '  selftests  - various kernel selftests'
@@ -41,7 +42,7 @@ acpi: FORCE
 cpupower: FORCE
 	$(call descend,power/$@)
 
-cgroup firewire hv guest usb virtio vm net: FORCE
+cgroup firewire hv kgraft guest usb virtio vm net: FORCE
 	$(call descend,$@)
 
 libapikfs: FORCE
@@ -65,7 +66,7 @@ acpi_install:
 cpupower_install:
 	$(call descend,power/$(@:_install=),install)
 
-cgroup_install firewire_install hv_install lguest_install perf_install usb_install virtio_install vm_install net_install:
+cgroup_install firewire_install hv_install kgraft_install lguest_install perf_install usb_install virtio_install vm_install net_install:
 	$(call descend,$(@:_install=),install)
 
 selftests_install:
@@ -77,7 +78,8 @@ turbostat_install x86_energy_perf_policy_install:
 tmon_install:
 	$(call descend,thermal/$(@:_install=),install)
 
-install: acpi_install cgroup_install cpupower_install hv_install firewire_install lguest_install \
+install: acpi_install cgroup_install cpupower_install hv_install \
+		kgraft_install firewire_install lguest_install \
 		perf_install selftests_install turbostat_install usb_install \
 		virtio_install vm_install net_install x86_energy_perf_policy_install \
 	tmon
@@ -88,7 +90,7 @@ acpi_clean:
 cpupower_clean:
 	$(call descend,power/cpupower,clean)
 
-cgroup_clean hv_clean firewire_clean lguest_clean usb_clean virtio_clean vm_clean net_clean:
+cgroup_clean hv_clean kgraft_clean firewire_clean lguest_clean usb_clean virtio_clean vm_clean net_clean:
 	$(call descend,$(@:_clean=),clean)
 
 libapikfs_clean:
@@ -106,7 +108,8 @@ turbostat_clean x86_energy_perf_policy_clean:
 tmon_clean:
 	$(call descend,thermal/tmon,clean)
 
-clean: acpi_clean cgroup_clean cpupower_clean hv_clean firewire_clean lguest_clean \
+clean: acpi_clean cgroup_clean cpupower_clean hv_clean kgraft_clean \
+		firewire_clean lguest_clean \
 		perf_clean selftests_clean turbostat_clean usb_clean virtio_clean \
 		vm_clean net_clean x86_energy_perf_policy_clean tmon_clean
 
diff --git a/tools/kgraft/Makefile b/tools/kgraft/Makefile
new file mode 100644
index 000000000000..75e0030b550d
--- /dev/null
+++ b/tools/kgraft/Makefile
@@ -0,0 +1,30 @@
+CC=gcc
+CFLAGS=-g
+
+all: objcopy-hacked dwarf-inline-tree it2rev.pl
+
+objcopy-hacked: objcopy.diff
+	echo "Build by hand!"
+	exit 1
+
+dwarf-inline-tree: dwarf-inline-tree.o dwarf_names.o
+	gcc -o $@ $^ -ldwarf -lelf
+
+dwarf-inline-tree.o: dwarf_names.h
+dwarf_names.o: dwarf_names.h
+
+check: app.o symlist all
+	@echo "inline tree"
+	./dwarf-inline-tree app.o
+	@echo "inline pairs"
+	./dwarf-inline-tree app.o | perl it2rev.pl
+	@echo "extract stuff"
+	./objcopy-hacked --strip-unneeded -j .doesntexist. --keep-symbols symlist app.o app-extract.o
+	@echo "symbols"
+	readelf -sW app.o app-extract.o
+
+app.c: app.h
+app.o: CFLAGS=-g -ffunction-sections -fdata-sections
+
+clean:
+	rm -f dwarf-inline-tree.o dwarf_names.o dwarf-inline-tree app.o app-extract.o
diff --git a/tools/kgraft/README b/tools/kgraft/README
new file mode 100644
index 000000000000..179db470a5b8
--- /dev/null
+++ b/tools/kgraft/README
@@ -0,0 +1,50 @@
+Some tools for kgraft.
+
+# make && make check
+
+will build most of them, and the check target contains example invocations.
+The only thing not built automatically is the hacked objcopy (objcopy-hacked),
+as usually the necessary binutils headers aren't installed.  You'll
+have to have (recent) binutils sources, apply the patch objcopy.diff
+and build it yourself.
+
+objcopy-hacked:
+  Given a list of symbols (e.g. in a file symlist) this will extract
+  all sections defining those symbols.  It will also recursively extract
+  sections needed by those (e.g. by section based relocations).
+
+dwarf-inline-tree:
+  Given an ELF file with debug info this will generate a parsable
+  output of the inline tree, like so:
+    U somesymbol
+    D filename.c:anothersym
+    I bla.h:helper
+  Meaning there's a reference to 'somesymbol', there's a definition
+  of function anothersym() from file filename.c, and that one contains
+  an inline expansion of function helper() from bla.h.
+
+  Filenames are directly from the debuginfo, and so can contain
+  directory prefixes dependending on how the objects were compiled.
+
+it2rev.pl [<path-prefix>]
+  This transforms the output of dwarf-inline-tree into a list of
+  whats-inlined-where lists, like
+    bla.h:helper filename.c:anothersym file2.c:bar
+  (helper is inlined into anothersym and bar).  If path-prefix
+  is given it is removed from all filenames from the input list.
+
+extract-syms.sh <list of symbol names>
+  This will use objcopy-hacked to extract the sections for the given
+  symbols from vmlinux.o into extracted.o.  All given symbols will
+  be prefixed with "new_" in the generated output.
+
+create-stub.sh <list of symbol names>
+  This will generate on stdout a C source that is the module source
+  code for a kgraft module patching the given symbols.
+
+create-kgrmodule.sh <list of symbol names>
+  This will generated a full kgraft module from a list of symbol
+  names (by using the above scripts).  Take care for compiling
+  the kernel providing the new code with
+    -ffunction-sections -fdata-sections .
+  The module will be in kgrafttmp/kgrmodule.ko .
diff --git a/tools/kgraft/TODO b/tools/kgraft/TODO
new file mode 100644
index 000000000000..498104c555d8
--- /dev/null
+++ b/tools/kgraft/TODO
@@ -0,0 +1,20 @@
+TODO list for kgraft tools
+
+extract-syms.sh shouldn't use vmlinux.o for extraction, but the
+underlying individual .o files.  The sections in vmlinux.o will be
+catted together already (for things like data.mostly_read), referencing
+too much unrelated stuff.
+
+extract-syms.sh should use an optional inliner tree to expand the set
+of symbols to those that have them inlined.  Further it should use
+(optionally) filename:symbol pairs for those cases where static functions
+need to be extracted whose names happen to occur multiple times in different
+units.
+
+Perhaps a top-level script taking a kernel patch, and pulling everything
+together should be created (applying patch, building kernel the right way,
+extracting stuff and so on).
+
+The seeding symbol list currently needs to come from a human.  It's probably
+feasible to generate that list for most cases by interpreting a kernel
+diff.  Binary comparison should _not_ be used to generate it.
diff --git a/tools/kgraft/app.c b/tools/kgraft/app.c
new file mode 100644
index 000000000000..16d8b313b438
--- /dev/null
+++ b/tools/kgraft/app.c
@@ -0,0 +1,35 @@
+#include <stdio.h>
+#include "app.h"
+
+static int local_data;
+int global_data;
+
+static void __attribute__((noinline)) in_app (void)
+{
+  printf ("in_app\n");
+  in_app_inline ();
+  local_data = 42;
+}
+
+static inline void __attribute__((always_inline)) in_app_inline_twice (void)
+{
+  global_data++;
+  in_app_inline ();
+}
+
+void in_app_global (void)
+{
+  printf ("in_app_global\n");
+  in_app();
+  in_app_inline_twice ();
+  global_data = 43;
+}
+
+int main ()
+{
+  in_app_global();
+  second_file ();
+  printf ("local_data = %d\n", local_data);
+  printf ("global_data = %d\n", global_data);
+  return 0;
+}
diff --git a/tools/kgraft/app.h b/tools/kgraft/app.h
new file mode 100644
index 000000000000..c07e10b23367
--- /dev/null
+++ b/tools/kgraft/app.h
@@ -0,0 +1,7 @@
+static inline void __attribute__((always_inline)) in_app_inline (void)
+{
+  static int local_static_data;
+  printf ("in_app_inline: %d\n", local_static_data++);
+}
+
+void second_file (void);
diff --git a/tools/kgraft/create-kgrmodule.sh b/tools/kgraft/create-kgrmodule.sh
new file mode 100755
index 000000000000..d73cccf01b22
--- /dev/null
+++ b/tools/kgraft/create-kgrmodule.sh
@@ -0,0 +1,25 @@
+#!/bin/bash
+TOOLPATH=`dirname $0`
+if ! test -f vmlinux.o; then
+    echo "vmlinux.o needs to exist in cwd"
+    exit 1
+fi
+if test -z "$1"; then
+    echo "usage: $0 [list of symbols to extract]"
+    exit 2
+fi
+mkdir -p kgrafttmp
+$TOOLPATH/extract-syms.sh $@
+mv extracted.o kgrafttmp
+cd kgrafttmp
+$TOOLPATH/create-stub.sh $@ > kgrstub.c
+cat <<EOF > Makefile
+obj-m = kgrmodule.o
+kgrmodule-y += kgrstub.o extracted.o
+
+all:
+	make -C .. M=\$(PWD) modules
+EOF
+make
+cd ..
+ls -l kgrafttmp/kgrmodule.ko
diff --git a/tools/kgraft/create-stub.sh b/tools/kgraft/create-stub.sh
new file mode 100755
index 000000000000..9551ebae5b31
--- /dev/null
+++ b/tools/kgraft/create-stub.sh
@@ -0,0 +1,53 @@
+#!/bin/bash
+
+if test -z "$1"; then
+    echo "usage: $0 [list of symbols]"
+fi
+
+cat <<EOF
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/kgr.h>
+#include <linux/kallsyms.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/capability.h>
+#include <linux/ptrace.h>
+
+EOF
+
+for i in $@; do
+    echo "extern void new_$i (void);"
+    echo "KGR_PATCHED_FUNCTION(patch, $i, new_$i);"
+done
+
+echo "static const struct kgr_patch patch = {"
+echo "	.patches = {"
+for i in $@; do
+    echo "		KGR_PATCH($i),"
+done
+echo "		KGR_PATCH_END"
+echo "	}"
+echo "};"
+
+cat <<EOF
+static int __init kgr_patcher_init(void)
+{
+        /* removing not supported (yet?) */
+        __module_get(THIS_MODULE);
+        /* +4 to skip push rbb / mov rsp,rbp prologue */
+        kgr_start_patching(&patch);
+        return 0;
+}
+
+static void __exit kgr_patcher_cleanup(void)
+{
+        printk(KERN_ERR "removing now buggy!\n");
+}
+
+module_init(kgr_patcher_init);
+module_exit(kgr_patcher_cleanup);
+
+MODULE_LICENSE("GPL");
+EOF
diff --git a/tools/kgraft/dwarf-inline-tree.c b/tools/kgraft/dwarf-inline-tree.c
new file mode 100644
index 000000000000..e8aea10f687d
--- /dev/null
+++ b/tools/kgraft/dwarf-inline-tree.c
@@ -0,0 +1,544 @@
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <getopt.h>
+
+#include <libelf.h>
+
+#include <libdwarf/dwarf.h>
+#include <libdwarf/libdwarf.h>
+
+#define HAVE_ELF64_GETEHDR
+
+#define string char*
+#include "dwarf_names.h"
+#undef string
+
+int ellipsis = 0;
+int verbose = 0;
+static char *file_name;
+static char *program_name;
+Dwarf_Error err;
+
+void
+print_error(Dwarf_Debug dbg, char *msg, int dwarf_code,
+	    Dwarf_Error err)
+{
+    if (dwarf_code == DW_DLV_ERROR) {
+	char *errmsg = dwarf_errmsg(err);
+	long long myerr = dwarf_errno(err);
+
+	fprintf(stderr, "%s ERROR:  %s:  %s (%lld)\n",
+		program_name, msg, errmsg, myerr);
+    } else if (dwarf_code == DW_DLV_NO_ENTRY) {
+	fprintf(stderr, "%s NO ENTRY:  %s: \n", program_name, msg);
+    } else if (dwarf_code == DW_DLV_OK) {
+	fprintf(stderr, "%s:  %s \n", program_name, msg);
+    } else {
+	fprintf(stderr, "%s InternalError:  %s:  code %d\n",
+		program_name, msg, dwarf_code);
+    }
+    exit(1);
+}
+
+static int indent_level;
+
+static void
+print_attribute(Dwarf_Debug dbg, Dwarf_Die die,
+		Dwarf_Half attr,
+		Dwarf_Attribute attr_in,
+		char **srcfiles, Dwarf_Signed cnt, Dwarf_Half tag)
+{
+    Dwarf_Attribute attrib = 0;
+    char *atname = 0;
+    int tres = 0;
+    Dwarf_Half form = 0;
+
+    attrib = attr_in;
+    atname = get_AT_name(dbg, attr);
+
+    tres = dwarf_whatform (attrib, &form, &err);
+    if (tres != DW_DLV_OK)
+	print_error (dbg, "dwarf_whatform", tres, err);
+    printf("\t\t%-28s%s\t", atname, get_FORM_name (dbg, form));
+    /* Don't move over the attributes for the top-level compile_unit
+     * DIEs.  */
+    if (tag == DW_TAG_compile_unit)
+      {
+	printf ("\n");
+	return;
+      }
+    switch (form) {
+	case DW_FORM_addr:
+	    {
+		Dwarf_Addr a;
+		tres = dwarf_formaddr (attrib, &a, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_formaddr", tres, err);
+		printf ("0x%llx", (unsigned long long)a);
+	    }
+	    break;
+	case DW_FORM_data4:
+	case DW_FORM_data8:
+	case DW_FORM_data1:
+	case DW_FORM_data2:
+	case DW_FORM_udata:
+	    {
+		/* Bah.  From just looking at FORM_data[1248] we don't
+		 * really know if it's signed or unsigned.  We have to
+		 * look at the context.  Luckily only two ATs can be signed. */
+		switch (attr) {
+		    case DW_AT_upper_bound:
+		    case DW_AT_lower_bound:
+			  {
+			    Dwarf_Signed s;
+			    tres = dwarf_formsdata (attrib, &s, &err);
+			    if (tres != DW_DLV_OK)
+			      print_error (dbg, "dwarf_formudata", tres, err);
+			    printf ("%lld", s);
+			  }
+			break;
+		    default:
+			  {
+			    Dwarf_Unsigned u;
+			    tres = dwarf_formudata (attrib, &u, &err);
+			    if (tres != DW_DLV_OK)
+			      print_error (dbg, "dwarf_formudata", tres, err);
+			    printf ("%llu", u);
+			  }
+			break;
+		}
+	    }
+	    break;
+	case DW_FORM_sdata:
+	    {
+		Dwarf_Signed s;
+		tres = dwarf_formsdata (attrib, &s, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_formsdata", tres, err);
+		printf ("%lld", s);
+	    }
+	    break;
+	case DW_FORM_string:
+	case DW_FORM_strp:
+	    {
+		char *s;
+		tres = dwarf_formstring (attrib, &s, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_formstring", tres, err);
+		printf ("%s\n", s);
+	    }
+	    break;
+	case DW_FORM_block:
+	case DW_FORM_block1:
+	case DW_FORM_block2:
+	case DW_FORM_block4:
+	    {
+		Dwarf_Block *b;
+		tres = dwarf_formblock (attrib, &b, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_formblock", tres, err);
+		printf ("[block data]");
+	    }
+	    break;
+	case DW_FORM_flag:
+	    {
+		Dwarf_Bool b;
+		tres = dwarf_formflag (attrib, &b, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_formflag", tres, err);
+		printf ("%s", b ? "true" : "false");
+	    }
+	    break;
+	case DW_FORM_ref_addr:
+	case DW_FORM_ref1:
+	case DW_FORM_ref2:
+	case DW_FORM_ref4:
+	case DW_FORM_ref8:
+	case DW_FORM_ref_udata:
+	    {
+		Dwarf_Off o;
+		tres = dwarf_global_formref (attrib, &o, &err);
+		if (tres != DW_DLV_OK)
+		    print_error (dbg, "dwarf_global_formref", tres, err);
+		printf ("ref <0x%x>\n", o);
+	    }
+	    break;
+	case DW_FORM_indirect:
+	default:
+	    print_error (dbg, "broken DW_FORM", 0, 0);
+	    break;
+    }
+    printf ("\n");
+}
+
+int
+get_file_and_name (Dwarf_Debug dbg, Dwarf_Die die, int *file, char **name)
+{
+  Dwarf_Attribute attr;
+  Dwarf_Half form = 0;
+  int tres = 0;
+  int ret = DW_DLV_OK;
+
+  if (dwarf_attr (die, DW_AT_abstract_origin, &attr, &err) == DW_DLV_OK
+      && dwarf_whatform (attr, &form, &err) == DW_DLV_OK)
+    {
+      Dwarf_Off o;
+      Dwarf_Die ref;
+      tres = dwarf_global_formref (attr, &o, &err);
+      if (tres != DW_DLV_OK)
+	print_error (dbg, "dwarf_global_formref", tres, err);
+      else
+	{
+	  if (dwarf_offdie (dbg, o, &ref, &err) == DW_DLV_OK)
+	    get_file_and_name (dbg, ref, file, name);
+	}
+    }
+
+  if (dwarf_attr (die, DW_AT_decl_file, &attr, &err) == DW_DLV_OK
+      && dwarf_whatform (attr, &form, &err) == DW_DLV_OK)
+    {
+      if (form == DW_FORM_sdata)
+	{
+	  Dwarf_Signed s;
+	  if ((tres = dwarf_formsdata (attr, &s, &err)) == DW_DLV_OK)
+	    *file = s;
+	  else
+	    ret = DW_DLV_ERROR, print_error (dbg, "dwarf_formsdata", tres, err);
+	}
+      else
+	{
+	  Dwarf_Unsigned u;
+	  if ((tres = dwarf_formudata (attr, &u, &err)) == DW_DLV_OK)
+	    *file = u;
+	  else
+	    ret = DW_DLV_ERROR, print_error (dbg, "dwarf_formudata", tres, err);
+	}
+    }
+
+  if ((dwarf_attr (die, DW_AT_MIPS_linkage_name, &attr, &err) == DW_DLV_OK
+       && dwarf_whatform (attr, &form, &err) == DW_DLV_OK)
+      || (dwarf_attr (die, DW_AT_name, &attr, &err) == DW_DLV_OK
+	  && dwarf_whatform (attr, &form, &err) == DW_DLV_OK))
+    {
+      char *s;
+      tres = dwarf_formstring (attr, &s, &err);
+      if (tres != DW_DLV_OK)
+	ret = DW_DLV_ERROR, print_error (dbg, "dwarf_formstring", tres, err);
+      *name = s;
+    }
+  return ret;
+}
+
+/* handle one die */
+void
+print_one_die(Dwarf_Debug dbg, Dwarf_Die die,
+	      char **srcfiles, Dwarf_Signed cnt)
+{
+    Dwarf_Signed i;
+    Dwarf_Off offset, overall_offset;
+    char *tagname;
+    Dwarf_Half tag;
+    Dwarf_Signed atcnt;
+    Dwarf_Attribute *atlist;
+    int tres;
+    int ores;
+    int atres;
+
+    tres = dwarf_tag(die, &tag, &err);
+    if (tres != DW_DLV_OK) {
+	print_error(dbg, "accessing tag of die!", tres, err);
+    }
+    tagname = get_TAG_name(dbg, tag);
+    ores = dwarf_dieoffset(die, &overall_offset, &err);
+    if (ores != DW_DLV_OK) {
+	print_error(dbg, "dwarf_dieoffset", ores, err);
+    }
+    ores = dwarf_die_CU_offset(die, &offset, &err);
+    if (ores != DW_DLV_OK) {
+	print_error(dbg, "dwarf_die_CU_offset", ores, err);
+    }
+
+    if (verbose)
+      {
+	if (indent_level == 0) {
+		printf
+		    ("\nCOMPILE_UNIT<header overall offset = %llu>:\n",
+		     overall_offset - offset);
+	}
+	printf("<%d><%5llu>\t%s\n", indent_level, offset, tagname);
+      }
+
+    if (tag == DW_TAG_subprogram || tag == DW_TAG_inlined_subroutine)
+      {
+	char *name = 0;
+	int filenum = -1;
+	char *prefix;
+	Dwarf_Attribute attr;
+	if (tag == DW_TAG_inlined_subroutine)
+	  prefix = "I";
+	else if (dwarf_attr (die, DW_AT_low_pc, &attr, &err) == DW_DLV_OK)
+	  prefix = "D";
+	else
+	  prefix = "U";
+	if (get_file_and_name (dbg, die, &filenum, &name) == DW_DLV_OK)
+	  {
+	    char *filename;
+	    if (filenum > 0 && filenum <= cnt)
+	      filename = srcfiles[filenum - 1];
+	    else
+	      filename = "";
+	    printf ("%s %s:%s\n", prefix, filename, name);
+	  }
+	else
+	  printf ("%s couldn't decode name or file\n", prefix);
+      }
+
+    if (!verbose)
+      return;
+
+    atres = dwarf_attrlist(die, &atlist, &atcnt, &err);
+    if (atres == DW_DLV_ERROR) {
+	print_error(dbg, "dwarf_attrlist", atres, err);
+    } else if (atres == DW_DLV_NO_ENTRY) {
+	/* indicates there are no attrs.  It is not an error. */
+	atcnt = 0;
+    }
+
+    for (i = 0; i < atcnt; i++) {
+	Dwarf_Half attr;
+	int ares;
+
+	ares = dwarf_whatattr(atlist[i], &attr, &err);
+	if (ares == DW_DLV_OK) {
+	    print_attribute(dbg, die, attr,
+			    atlist[i], srcfiles, cnt, tag);
+	} else {
+	    print_error(dbg, "dwarf_whatattr entry missing", ares, err);
+	}
+    }
+
+    for (i = 0; i < atcnt; i++) {
+	dwarf_dealloc(dbg, atlist[i], DW_DLA_ATTR);
+    }
+    if (atres == DW_DLV_OK) {
+	dwarf_dealloc(dbg, atlist, DW_DLA_LIST);
+    }
+}
+
+/* recursively follow the die tree */
+void
+print_die_and_children(Dwarf_Debug dbg, Dwarf_Die in_die_in,
+		       char **srcfiles, Dwarf_Signed cnt)
+{
+    Dwarf_Die child;
+    Dwarf_Die sibling;
+    Dwarf_Error err;
+    int tres;
+    int cdres;
+    Dwarf_Die in_die = in_die_in;
+
+    for (;;) {
+	/* here to pre-descent processing of the die */
+	print_one_die(dbg, in_die, srcfiles, cnt);
+
+	cdres = dwarf_child(in_die, &child, &err);
+	/* child first: we are doing depth-first walk */
+	if (cdres == DW_DLV_OK) {
+	    indent_level++;
+	    print_die_and_children(dbg, child, srcfiles, cnt);
+	    indent_level--;
+	    dwarf_dealloc(dbg, child, DW_DLA_DIE);
+	} else if (cdres == DW_DLV_ERROR) {
+	    print_error(dbg, "dwarf_child", cdres, err);
+	}
+
+	cdres = dwarf_siblingof(dbg, in_die, &sibling, &err);
+	if (cdres == DW_DLV_OK) {
+	    /* print_die_and_children(dbg, sibling, srcfiles, cnt); We 
+	       loop around to actually print this, rather than
+	       recursing. Recursing is horribly wasteful of stack
+	       space. */
+	} else if (cdres == DW_DLV_ERROR) {
+	    print_error(dbg, "dwarf_siblingof", cdres, err);
+	}
+
+	/* Here do any post-descent (ie post-dwarf_child) processing
+	   of the in_die. */
+
+	if (in_die != in_die_in) {
+	    /* Dealloc our in_die, but not the argument die, it belongs 
+	       to our caller. Whether the siblingof call worked or not. 
+	     */
+	    dwarf_dealloc(dbg, in_die, DW_DLA_DIE);
+	}
+	if (cdres == DW_DLV_OK) {
+	    /* Set to process the sibling, loop again. */
+	    in_die = sibling;
+	} else {
+	    /* We are done, no more siblings at this level. */
+
+	    break;
+	}
+    }				/* end for loop on siblings */
+}
+
+static void
+print_infos(Dwarf_Debug dbg)
+{
+    Dwarf_Unsigned cu_header_length = 0;
+    Dwarf_Unsigned abbrev_offset = 0;
+    Dwarf_Half version_stamp = 0;
+    Dwarf_Half address_size = 0;
+    Dwarf_Die cu_die = 0;
+    Dwarf_Unsigned next_cu_offset = 0;
+    int nres = DW_DLV_OK;
+
+    /* Loop until it fails.  */
+    while ((nres =
+	    dwarf_next_cu_header(dbg, &cu_header_length, &version_stamp,
+				 &abbrev_offset, &address_size,
+				 &next_cu_offset, &err))
+	   == DW_DLV_OK) {
+	int sres;
+
+	if (verbose)
+	{
+		printf("\nCU_HEADER:\n");
+		printf("\t\t%-28s%llu\n", "cu_header_length",
+		       cu_header_length);
+		printf("\t\t%-28s%d\n", "version_stamp", version_stamp);
+		printf("\t\t%-28s%llu\n", "abbrev_offset",
+		       abbrev_offset);
+		printf("\t\t%-28s%d", "address_size", address_size);
+	}
+
+	/* process a single compilation unit in .debug_info. */
+	sres = dwarf_siblingof(dbg, NULL, &cu_die, &err);
+	if (sres == DW_DLV_OK) {
+	    {
+		Dwarf_Signed cnt = 0;
+		char **srcfiles = 0;
+		int srcf = dwarf_srcfiles(cu_die,
+					  &srcfiles, &cnt, &err);
+
+		if (srcf != DW_DLV_OK) {
+		    srcfiles = 0;
+		    cnt = 0;
+		}
+
+		print_die_and_children(dbg, cu_die, srcfiles, cnt);
+		if (srcf == DW_DLV_OK) {
+		    int si;
+
+		    for (si = 0; si < cnt; ++si) {
+			dwarf_dealloc(dbg, srcfiles[si], DW_DLA_STRING);
+		    }
+		    dwarf_dealloc(dbg, srcfiles, DW_DLA_LIST);
+		}
+	    }
+	    dwarf_dealloc(dbg, cu_die, DW_DLA_DIE);
+	} else if (sres == DW_DLV_NO_ENTRY) {
+	    /* do nothing I guess. */
+	} else {
+	    print_error(dbg, "Regetting cu_die", sres, err);
+	}
+    }
+    if (nres == DW_DLV_ERROR) {
+	char *errmsg = dwarf_errmsg(err);
+	long long myerr = dwarf_errno(err);
+
+	fprintf(stderr, "%s ERROR:  %s:  %s (%lld)\n",
+		program_name, "attempting to print .debug_info",
+		errmsg, myerr);
+	fprintf(stderr, "attempting to continue.\n");
+    }
+}
+
+static void
+process_one_file (Elf *elf, char *file_name)
+{
+    Dwarf_Debug dbg;
+    int dres;
+
+    if (verbose)
+      printf ("processing %s\n", file_name);
+    dres = dwarf_elf_init(elf, DW_DLC_READ, NULL, NULL, &dbg, &err);
+    if (dres == DW_DLV_NO_ENTRY) {
+	printf("No DWARF information present in %s\n", file_name);
+	return;
+    }
+    if (dres != DW_DLV_OK) {
+	print_error(dbg, "dwarf_elf_init", dres, err);
+    }
+
+    print_infos(dbg);
+
+    dres = dwarf_finish(dbg, &err);
+    if (dres != DW_DLV_OK) {
+	print_error(dbg, "dwarf_finish", dres, err);
+    }
+    return;
+}
+	
+int
+main(int argc, char *argv[])
+{
+    int f;
+    Elf_Cmd cmd;
+    Elf *arf, *elf;
+
+    program_name = argv[0];
+    
+    (void) elf_version(EV_NONE);
+    if (elf_version(EV_CURRENT) == EV_NONE) {
+	(void) fprintf(stderr, "dwarf-inline-tree: libelf.a out of date.\n");
+	exit(1);
+    }
+
+    if (argc < 2)
+    {
+	fprintf (stderr, "dwarf-inline-tree <input-elf>\n");
+	exit (2);
+    }
+    file_name = argv[1];
+    f = open(file_name, O_RDONLY);
+    if (f == -1) {
+	fprintf(stderr, "%s ERROR:  can't open %s\n", program_name,
+		file_name);
+	return 1;
+    }
+
+    cmd = ELF_C_READ;
+    arf = elf_begin(f, cmd, (Elf *) 0);
+    while ((elf = elf_begin(f, cmd, arf)) != 0) {
+	Elf32_Ehdr *eh32;
+
+#ifdef HAVE_ELF64_GETEHDR
+	Elf64_Ehdr *eh64;
+#endif /* HAVE_ELF64_GETEHDR */
+	eh32 = elf32_getehdr(elf);
+	if (!eh32) {
+#ifdef HAVE_ELF64_GETEHDR
+	    /* not a 32-bit obj */
+	    eh64 = elf64_getehdr(elf);
+	    if (!eh64) {
+		/* not a 64-bit obj either! */
+		/* dwarfdump is quiet when not an object */
+	    } else {
+		process_one_file(elf, file_name);
+	    }
+#endif /* HAVE_ELF64_GETEHDR */
+	} else {
+	    process_one_file(elf, file_name);
+	}
+	cmd = elf_next(elf);
+	elf_end(elf);
+    }
+    elf_end(arf);
+    return 0;
+}
diff --git a/tools/kgraft/dwarf_names.awk b/tools/kgraft/dwarf_names.awk
new file mode 100644
index 000000000000..e5b39726fe1a
--- /dev/null
+++ b/tools/kgraft/dwarf_names.awk
@@ -0,0 +1,126 @@
+# Print routines to return constant name for associated value.
+# The input is dwarf.h
+# For each set of names with a common prefix, we create a routine
+# to return the name given the value.
+# Also print header file that gives prototypes of routines.
+# To handle cases where there are multiple names for a single
+# value (DW_AT_* has some due to ambiguities in the DWARF2 spec)
+# we take the first of a given value as the definitive name.
+# TAGs, Attributes, etc are given distinct checks.
+BEGIN {
+	prefix = "foo"
+	prefix_id = "foo"
+	prefix_len = length(prefix)
+	dw_prefix = "DW_"
+	dw_len = length(dw_prefix)
+	start_routine = 0
+	printf "#include <stdio.h>\n"
+	printf "#include <string.h>\n"
+	printf "#include <libdwarf/dwarf.h>\n"
+	printf "#include <libdwarf/libdwarf.h>\n"
+
+	printf "typedef char * string;\n"
+	printf "#define makename strdup\n"
+	printf "extern int ellipsis;\n"
+
+	header = "dwarf_names.h"
+	printf "/* automatically generated routines */\n" > header
+	dup_arr["0"] = ""
+}
+{
+	if (skipit && $1 == "#endif") {
+		skipit = 0
+		next
+	}
+	if ($2 == 0 || skipit) {
+		# if 0, skip to endif
+		skipit = 1
+		next
+	}
+	if ($1 == "#define") {
+		if (substr($2,1,prefix_len) != prefix) {
+			# new prefix
+			if (substr($2,1,dw_len) != dw_prefix) {
+				# skip
+				next
+			} else if (substr($2,1,dw_len+3) == "DW_CFA") {
+				# skip, cause numbers conflict
+				# (have both high-order and low-order bits)
+				next
+			} else {
+				# New prefix, empty the dup_arr
+				for (k in dup_arr)
+					dup_arr[k] = ""
+				if (start_routine) {
+					# end routine
+					printf "\tdefault:\n"
+printf "\t\t{ \n"
+printf "\t\t    char buf[100]; \n"
+printf "\t\t    char *n; \n"
+printf "\t\t    sprintf(buf,\"<Unknown %s value 0x%%x>\",(int)val);\n",prefix_id
+printf "\t\t fprintf(stderr,\"%s of %%d (0x%%x) is unknown to dwarfdump. \" \n ", prefix_id
+printf "\t\t \"Continuing. \\n\",(int)val,(int)val );  \n"
+printf "\t\t    n = makename(buf);\n"
+printf "\t\t    return n; \n"
+printf "\t\t} \n"
+					printf "\t}\n"
+					printf "/*NOTREACHED*/\n"
+					printf "}\n\n"
+				}
+				start_routine = 1
+				post_dw = substr($2,dw_len+1, length($2))
+				second_underscore = index(post_dw,"_")
+				prefix = substr($2,1,second_underscore+dw_len)
+				prefix_len = length(prefix)
+				# prefix id is unique part after DW_, e.g. LANG
+				prefix_id = substr(prefix,dw_len+1,prefix_len-dw_len-1)
+				printf "/* ARGSUSED */\n"
+				printf "extern string\n"
+				printf "get_%s_name (Dwarf_Debug dbg, Dwarf_Half val)\n", prefix_id
+				printf "{\n"
+				printf "\tswitch (val) {\n"
+				printf "extern string get_%s_name (Dwarf_Debug dbg, Dwarf_Half val);\n\n", prefix_id >> header
+			}
+		}
+		if (substr($2,1,prefix_len) == prefix) {
+			if (substr($2,1,dw_len+8) == "DW_CHILDREN" \
+			    || substr($2,1,dw_len+8) == "DW_children" \
+			    || substr($2,1,dw_len+4) == "DW_ADDR") {
+				main_part = substr($2,dw_len+1, length($2))
+			}
+			else {
+				post_dw = substr($2,dw_len+1, length($2))
+				second_underscore = index(post_dw,"_")
+				main_part = substr($2,dw_len+second_underscore+1, length($2))
+			}
+			if( dup_arr[$3] != $3 ) {
+			  # Take first of those with identical value,
+			  # ignore others.
+			  dup_arr[$3] = $3
+			  printf "\tcase %s:\n", $2
+			  printf "\t\tif (ellipsis)\n"
+			  printf "\t\t\treturn \"%s\";\n", main_part
+			  printf "\t\telse\n"
+			  printf "\t\t\treturn \"%s\";\n", $2
+		        }
+		}
+	}
+}
+END {
+	if (start_routine) {
+					printf "\tdefault:\n"
+printf "\t\t{ \n"
+printf "\t\t    char buf[100]; \n"
+printf "\t\t    char *n; \n"
+printf "\t\t    sprintf(buf,\"<Unknown %s value 0x%%x>\",(int)val);\n",prefix_id
+printf "\t\t fprintf(stderr,\"%s of %%d (0x%%x) is unknown to dwarfdump. \" \n ", prefix_id
+printf "\t\t \"Continuing. \\n\",(int)val,(int)val );  \n"
+printf "\t\t    n = makename(buf);\n"
+printf "\t\t    return n; \n"
+printf "\t\t} \n"
+					printf "\t}\n"
+					printf "/*NOTREACHED*/\n"
+					printf "}\n\n"
+	}
+}
+
diff --git a/tools/kgraft/dwarf_names.c b/tools/kgraft/dwarf_names.c
new file mode 100644
index 000000000000..e0afd80c9cd4
--- /dev/null
+++ b/tools/kgraft/dwarf_names.c
@@ -0,0 +1,4366 @@
+#include <stdio.h>
+#include <string.h>
+#include <libdwarf/dwarf.h>
+#include <libdwarf/libdwarf.h>
+typedef char * string;
+#define makename strdup
+extern int ellipsis;
+/* ARGSUSED */
+extern string
+get_TAG_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_TAG_array_type:
+		if (ellipsis)
+			return "array_type";
+		else
+			return "DW_TAG_array_type";
+	case DW_TAG_class_type:
+		if (ellipsis)
+			return "class_type";
+		else
+			return "DW_TAG_class_type";
+	case DW_TAG_entry_point:
+		if (ellipsis)
+			return "entry_point";
+		else
+			return "DW_TAG_entry_point";
+	case DW_TAG_enumeration_type:
+		if (ellipsis)
+			return "enumeration_type";
+		else
+			return "DW_TAG_enumeration_type";
+	case DW_TAG_formal_parameter:
+		if (ellipsis)
+			return "formal_parameter";
+		else
+			return "DW_TAG_formal_parameter";
+	case DW_TAG_imported_declaration:
+		if (ellipsis)
+			return "imported_declaration";
+		else
+			return "DW_TAG_imported_declaration";
+	case DW_TAG_label:
+		if (ellipsis)
+			return "label";
+		else
+			return "DW_TAG_label";
+	case DW_TAG_lexical_block:
+		if (ellipsis)
+			return "lexical_block";
+		else
+			return "DW_TAG_lexical_block";
+	case DW_TAG_member:
+		if (ellipsis)
+			return "member";
+		else
+			return "DW_TAG_member";
+	case DW_TAG_pointer_type:
+		if (ellipsis)
+			return "pointer_type";
+		else
+			return "DW_TAG_pointer_type";
+	case DW_TAG_reference_type:
+		if (ellipsis)
+			return "reference_type";
+		else
+			return "DW_TAG_reference_type";
+	case DW_TAG_compile_unit:
+		if (ellipsis)
+			return "compile_unit";
+		else
+			return "DW_TAG_compile_unit";
+	case DW_TAG_string_type:
+		if (ellipsis)
+			return "string_type";
+		else
+			return "DW_TAG_string_type";
+	case DW_TAG_structure_type:
+		if (ellipsis)
+			return "structure_type";
+		else
+			return "DW_TAG_structure_type";
+	case DW_TAG_subroutine_type:
+		if (ellipsis)
+			return "subroutine_type";
+		else
+			return "DW_TAG_subroutine_type";
+	case DW_TAG_typedef:
+		if (ellipsis)
+			return "typedef";
+		else
+			return "DW_TAG_typedef";
+	case DW_TAG_union_type:
+		if (ellipsis)
+			return "union_type";
+		else
+			return "DW_TAG_union_type";
+	case DW_TAG_unspecified_parameters:
+		if (ellipsis)
+			return "unspecified_parameters";
+		else
+			return "DW_TAG_unspecified_parameters";
+	case DW_TAG_variant:
+		if (ellipsis)
+			return "variant";
+		else
+			return "DW_TAG_variant";
+	case DW_TAG_common_block:
+		if (ellipsis)
+			return "common_block";
+		else
+			return "DW_TAG_common_block";
+	case DW_TAG_common_inclusion:
+		if (ellipsis)
+			return "common_inclusion";
+		else
+			return "DW_TAG_common_inclusion";
+	case DW_TAG_inheritance:
+		if (ellipsis)
+			return "inheritance";
+		else
+			return "DW_TAG_inheritance";
+	case DW_TAG_inlined_subroutine:
+		if (ellipsis)
+			return "inlined_subroutine";
+		else
+			return "DW_TAG_inlined_subroutine";
+	case DW_TAG_module:
+		if (ellipsis)
+			return "module";
+		else
+			return "DW_TAG_module";
+	case DW_TAG_ptr_to_member_type:
+		if (ellipsis)
+			return "ptr_to_member_type";
+		else
+			return "DW_TAG_ptr_to_member_type";
+	case DW_TAG_set_type:
+		if (ellipsis)
+			return "set_type";
+		else
+			return "DW_TAG_set_type";
+	case DW_TAG_subrange_type:
+		if (ellipsis)
+			return "subrange_type";
+		else
+			return "DW_TAG_subrange_type";
+	case DW_TAG_with_stmt:
+		if (ellipsis)
+			return "with_stmt";
+		else
+			return "DW_TAG_with_stmt";
+	case DW_TAG_access_declaration:
+		if (ellipsis)
+			return "access_declaration";
+		else
+			return "DW_TAG_access_declaration";
+	case DW_TAG_base_type:
+		if (ellipsis)
+			return "base_type";
+		else
+			return "DW_TAG_base_type";
+	case DW_TAG_catch_block:
+		if (ellipsis)
+			return "catch_block";
+		else
+			return "DW_TAG_catch_block";
+	case DW_TAG_const_type:
+		if (ellipsis)
+			return "const_type";
+		else
+			return "DW_TAG_const_type";
+	case DW_TAG_constant:
+		if (ellipsis)
+			return "constant";
+		else
+			return "DW_TAG_constant";
+	case DW_TAG_enumerator:
+		if (ellipsis)
+			return "enumerator";
+		else
+			return "DW_TAG_enumerator";
+	case DW_TAG_file_type:
+		if (ellipsis)
+			return "file_type";
+		else
+			return "DW_TAG_file_type";
+	case DW_TAG_friend:
+		if (ellipsis)
+			return "friend";
+		else
+			return "DW_TAG_friend";
+	case DW_TAG_namelist:
+		if (ellipsis)
+			return "namelist";
+		else
+			return "DW_TAG_namelist";
+	case DW_TAG_namelist_item:
+		if (ellipsis)
+			return "namelist_item";
+		else
+			return "DW_TAG_namelist_item";
+	case DW_TAG_packed_type:
+		if (ellipsis)
+			return "packed_type";
+		else
+			return "DW_TAG_packed_type";
+	case DW_TAG_subprogram:
+		if (ellipsis)
+			return "subprogram";
+		else
+			return "DW_TAG_subprogram";
+	case DW_TAG_template_type_parameter:
+		if (ellipsis)
+			return "template_type_parameter";
+		else
+			return "DW_TAG_template_type_parameter";
+	case DW_TAG_template_value_parameter:
+		if (ellipsis)
+			return "template_value_parameter";
+		else
+			return "DW_TAG_template_value_parameter";
+	case DW_TAG_thrown_type:
+		if (ellipsis)
+			return "thrown_type";
+		else
+			return "DW_TAG_thrown_type";
+	case DW_TAG_try_block:
+		if (ellipsis)
+			return "try_block";
+		else
+			return "DW_TAG_try_block";
+	case DW_TAG_variant_part:
+		if (ellipsis)
+			return "variant_part";
+		else
+			return "DW_TAG_variant_part";
+	case DW_TAG_variable:
+		if (ellipsis)
+			return "variable";
+		else
+			return "DW_TAG_variable";
+	case DW_TAG_volatile_type:
+		if (ellipsis)
+			return "volatile_type";
+		else
+			return "DW_TAG_volatile_type";
+	case DW_TAG_dwarf_procedure:
+		if (ellipsis)
+			return "dwarf_procedure";
+		else
+			return "DW_TAG_dwarf_procedure";
+	case DW_TAG_restrict_type:
+		if (ellipsis)
+			return "restrict_type";
+		else
+			return "DW_TAG_restrict_type";
+	case DW_TAG_interface_type:
+		if (ellipsis)
+			return "interface_type";
+		else
+			return "DW_TAG_interface_type";
+	case DW_TAG_namespace:
+		if (ellipsis)
+			return "namespace";
+		else
+			return "DW_TAG_namespace";
+	case DW_TAG_imported_module:
+		if (ellipsis)
+			return "imported_module";
+		else
+			return "DW_TAG_imported_module";
+	case DW_TAG_unspecified_type:
+		if (ellipsis)
+			return "unspecified_type";
+		else
+			return "DW_TAG_unspecified_type";
+	case DW_TAG_partial_unit:
+		if (ellipsis)
+			return "partial_unit";
+		else
+			return "DW_TAG_partial_unit";
+	case DW_TAG_imported_unit:
+		if (ellipsis)
+			return "imported_unit";
+		else
+			return "DW_TAG_imported_unit";
+	case DW_TAG_mutable_type:
+		if (ellipsis)
+			return "mutable_type";
+		else
+			return "DW_TAG_mutable_type";
+	case DW_TAG_condition:
+		if (ellipsis)
+			return "condition";
+		else
+			return "DW_TAG_condition";
+	case DW_TAG_shared_type:
+		if (ellipsis)
+			return "shared_type";
+		else
+			return "DW_TAG_shared_type";
+	case DW_TAG_type_unit:
+		if (ellipsis)
+			return "type_unit";
+		else
+			return "DW_TAG_type_unit";
+	case DW_TAG_rvalue_reference_type:
+		if (ellipsis)
+			return "rvalue_reference_type";
+		else
+			return "DW_TAG_rvalue_reference_type";
+	case DW_TAG_template_alias:
+		if (ellipsis)
+			return "template_alias";
+		else
+			return "DW_TAG_template_alias";
+	case DW_TAG_lo_user:
+		if (ellipsis)
+			return "lo_user";
+		else
+			return "DW_TAG_lo_user";
+	case DW_TAG_MIPS_loop:
+		if (ellipsis)
+			return "MIPS_loop";
+		else
+			return "DW_TAG_MIPS_loop";
+	case DW_TAG_HP_array_descriptor:
+		if (ellipsis)
+			return "HP_array_descriptor";
+		else
+			return "DW_TAG_HP_array_descriptor";
+	case DW_TAG_format_label:
+		if (ellipsis)
+			return "format_label";
+		else
+			return "DW_TAG_format_label";
+	case DW_TAG_function_template:
+		if (ellipsis)
+			return "function_template";
+		else
+			return "DW_TAG_function_template";
+	case DW_TAG_class_template:
+		if (ellipsis)
+			return "class_template";
+		else
+			return "DW_TAG_class_template";
+	case DW_TAG_GNU_BINCL:
+		if (ellipsis)
+			return "GNU_BINCL";
+		else
+			return "DW_TAG_GNU_BINCL";
+	case DW_TAG_GNU_EINCL:
+		if (ellipsis)
+			return "GNU_EINCL";
+		else
+			return "DW_TAG_GNU_EINCL";
+	case DW_TAG_GNU_template_template_parameter:
+		if (ellipsis)
+			return "GNU_template_template_parameter";
+		else
+			return "DW_TAG_GNU_template_template_parameter";
+	case DW_TAG_GNU_template_parameter_pack:
+		if (ellipsis)
+			return "GNU_template_parameter_pack";
+		else
+			return "DW_TAG_GNU_template_parameter_pack";
+	case DW_TAG_GNU_formal_parameter_pack:
+		if (ellipsis)
+			return "GNU_formal_parameter_pack";
+		else
+			return "DW_TAG_GNU_formal_parameter_pack";
+	case DW_TAG_GNU_call_site:
+		if (ellipsis)
+			return "GNU_call_site";
+		else
+			return "DW_TAG_GNU_call_site";
+	case DW_TAG_GNU_call_site_parameter:
+		if (ellipsis)
+			return "GNU_call_site_parameter";
+		else
+			return "DW_TAG_GNU_call_site_parameter";
+	case DW_TAG_ALTIUM_circ_type:
+		if (ellipsis)
+			return "ALTIUM_circ_type";
+		else
+			return "DW_TAG_ALTIUM_circ_type";
+	case DW_TAG_ALTIUM_mwa_circ_type:
+		if (ellipsis)
+			return "ALTIUM_mwa_circ_type";
+		else
+			return "DW_TAG_ALTIUM_mwa_circ_type";
+	case DW_TAG_ALTIUM_rev_carry_type:
+		if (ellipsis)
+			return "ALTIUM_rev_carry_type";
+		else
+			return "DW_TAG_ALTIUM_rev_carry_type";
+	case DW_TAG_ALTIUM_rom:
+		if (ellipsis)
+			return "ALTIUM_rom";
+		else
+			return "DW_TAG_ALTIUM_rom";
+	case DW_TAG_upc_shared_type:
+		if (ellipsis)
+			return "upc_shared_type";
+		else
+			return "DW_TAG_upc_shared_type";
+	case DW_TAG_upc_strict_type:
+		if (ellipsis)
+			return "upc_strict_type";
+		else
+			return "DW_TAG_upc_strict_type";
+	case DW_TAG_upc_relaxed_type:
+		if (ellipsis)
+			return "upc_relaxed_type";
+		else
+			return "DW_TAG_upc_relaxed_type";
+	case DW_TAG_PGI_kanji_type:
+		if (ellipsis)
+			return "PGI_kanji_type";
+		else
+			return "DW_TAG_PGI_kanji_type";
+	case DW_TAG_PGI_interface_block:
+		if (ellipsis)
+			return "PGI_interface_block";
+		else
+			return "DW_TAG_PGI_interface_block";
+	case DW_TAG_SUN_function_template:
+		if (ellipsis)
+			return "SUN_function_template";
+		else
+			return "DW_TAG_SUN_function_template";
+	case DW_TAG_SUN_class_template:
+		if (ellipsis)
+			return "SUN_class_template";
+		else
+			return "DW_TAG_SUN_class_template";
+	case DW_TAG_SUN_struct_template:
+		if (ellipsis)
+			return "SUN_struct_template";
+		else
+			return "DW_TAG_SUN_struct_template";
+	case DW_TAG_SUN_union_template:
+		if (ellipsis)
+			return "SUN_union_template";
+		else
+			return "DW_TAG_SUN_union_template";
+	case DW_TAG_SUN_indirect_inheritance:
+		if (ellipsis)
+			return "SUN_indirect_inheritance";
+		else
+			return "DW_TAG_SUN_indirect_inheritance";
+	case DW_TAG_SUN_codeflags:
+		if (ellipsis)
+			return "SUN_codeflags";
+		else
+			return "DW_TAG_SUN_codeflags";
+	case DW_TAG_SUN_memop_info:
+		if (ellipsis)
+			return "SUN_memop_info";
+		else
+			return "DW_TAG_SUN_memop_info";
+	case DW_TAG_SUN_omp_child_func:
+		if (ellipsis)
+			return "SUN_omp_child_func";
+		else
+			return "DW_TAG_SUN_omp_child_func";
+	case DW_TAG_SUN_rtti_descriptor:
+		if (ellipsis)
+			return "SUN_rtti_descriptor";
+		else
+			return "DW_TAG_SUN_rtti_descriptor";
+	case DW_TAG_SUN_dtor_info:
+		if (ellipsis)
+			return "SUN_dtor_info";
+		else
+			return "DW_TAG_SUN_dtor_info";
+	case DW_TAG_SUN_dtor:
+		if (ellipsis)
+			return "SUN_dtor";
+		else
+			return "DW_TAG_SUN_dtor";
+	case DW_TAG_SUN_f90_interface:
+		if (ellipsis)
+			return "SUN_f90_interface";
+		else
+			return "DW_TAG_SUN_f90_interface";
+	case DW_TAG_SUN_fortran_vax_structure:
+		if (ellipsis)
+			return "SUN_fortran_vax_structure";
+		else
+			return "DW_TAG_SUN_fortran_vax_structure";
+	case DW_TAG_SUN_hi:
+		if (ellipsis)
+			return "SUN_hi";
+		else
+			return "DW_TAG_SUN_hi";
+	case DW_TAG_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_TAG_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown TAG value 0x%x>",(int)val);
+		 fprintf(stderr,"TAG of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_children_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_children_no:
+		if (ellipsis)
+			return "children_no";
+		else
+			return "DW_children_no";
+	case DW_children_yes:
+		if (ellipsis)
+			return "children_yes";
+		else
+			return "DW_children_yes";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown children value 0x%x>",(int)val);
+		 fprintf(stderr,"children of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_FORM_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_FORM_addr:
+		if (ellipsis)
+			return "addr";
+		else
+			return "DW_FORM_addr";
+	case DW_FORM_block2:
+		if (ellipsis)
+			return "block2";
+		else
+			return "DW_FORM_block2";
+	case DW_FORM_block4:
+		if (ellipsis)
+			return "block4";
+		else
+			return "DW_FORM_block4";
+	case DW_FORM_data2:
+		if (ellipsis)
+			return "data2";
+		else
+			return "DW_FORM_data2";
+	case DW_FORM_data4:
+		if (ellipsis)
+			return "data4";
+		else
+			return "DW_FORM_data4";
+	case DW_FORM_data8:
+		if (ellipsis)
+			return "data8";
+		else
+			return "DW_FORM_data8";
+	case DW_FORM_string:
+		if (ellipsis)
+			return "string";
+		else
+			return "DW_FORM_string";
+	case DW_FORM_block:
+		if (ellipsis)
+			return "block";
+		else
+			return "DW_FORM_block";
+	case DW_FORM_block1:
+		if (ellipsis)
+			return "block1";
+		else
+			return "DW_FORM_block1";
+	case DW_FORM_data1:
+		if (ellipsis)
+			return "data1";
+		else
+			return "DW_FORM_data1";
+	case DW_FORM_flag:
+		if (ellipsis)
+			return "flag";
+		else
+			return "DW_FORM_flag";
+	case DW_FORM_sdata:
+		if (ellipsis)
+			return "sdata";
+		else
+			return "DW_FORM_sdata";
+	case DW_FORM_strp:
+		if (ellipsis)
+			return "strp";
+		else
+			return "DW_FORM_strp";
+	case DW_FORM_udata:
+		if (ellipsis)
+			return "udata";
+		else
+			return "DW_FORM_udata";
+	case DW_FORM_ref_addr:
+		if (ellipsis)
+			return "ref_addr";
+		else
+			return "DW_FORM_ref_addr";
+	case DW_FORM_ref1:
+		if (ellipsis)
+			return "ref1";
+		else
+			return "DW_FORM_ref1";
+	case DW_FORM_ref2:
+		if (ellipsis)
+			return "ref2";
+		else
+			return "DW_FORM_ref2";
+	case DW_FORM_ref4:
+		if (ellipsis)
+			return "ref4";
+		else
+			return "DW_FORM_ref4";
+	case DW_FORM_ref8:
+		if (ellipsis)
+			return "ref8";
+		else
+			return "DW_FORM_ref8";
+	case DW_FORM_ref_udata:
+		if (ellipsis)
+			return "ref_udata";
+		else
+			return "DW_FORM_ref_udata";
+	case DW_FORM_indirect:
+		if (ellipsis)
+			return "indirect";
+		else
+			return "DW_FORM_indirect";
+	case DW_FORM_sec_offset:
+		if (ellipsis)
+			return "sec_offset";
+		else
+			return "DW_FORM_sec_offset";
+	case DW_FORM_exprloc:
+		if (ellipsis)
+			return "exprloc";
+		else
+			return "DW_FORM_exprloc";
+	case DW_FORM_flag_present:
+		if (ellipsis)
+			return "flag_present";
+		else
+			return "DW_FORM_flag_present";
+	case DW_FORM_ref_sig8:
+		if (ellipsis)
+			return "ref_sig8";
+		else
+			return "DW_FORM_ref_sig8";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown FORM value 0x%x>",(int)val);
+		 fprintf(stderr,"FORM of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_AT_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_AT_sibling:
+		if (ellipsis)
+			return "sibling";
+		else
+			return "DW_AT_sibling";
+	case DW_AT_location:
+		if (ellipsis)
+			return "location";
+		else
+			return "DW_AT_location";
+	case DW_AT_name:
+		if (ellipsis)
+			return "name";
+		else
+			return "DW_AT_name";
+	case DW_AT_ordering:
+		if (ellipsis)
+			return "ordering";
+		else
+			return "DW_AT_ordering";
+	case DW_AT_subscr_data:
+		if (ellipsis)
+			return "subscr_data";
+		else
+			return "DW_AT_subscr_data";
+	case DW_AT_byte_size:
+		if (ellipsis)
+			return "byte_size";
+		else
+			return "DW_AT_byte_size";
+	case DW_AT_bit_offset:
+		if (ellipsis)
+			return "bit_offset";
+		else
+			return "DW_AT_bit_offset";
+	case DW_AT_bit_size:
+		if (ellipsis)
+			return "bit_size";
+		else
+			return "DW_AT_bit_size";
+	case DW_AT_element_list:
+		if (ellipsis)
+			return "element_list";
+		else
+			return "DW_AT_element_list";
+	case DW_AT_stmt_list:
+		if (ellipsis)
+			return "stmt_list";
+		else
+			return "DW_AT_stmt_list";
+	case DW_AT_low_pc:
+		if (ellipsis)
+			return "low_pc";
+		else
+			return "DW_AT_low_pc";
+	case DW_AT_high_pc:
+		if (ellipsis)
+			return "high_pc";
+		else
+			return "DW_AT_high_pc";
+	case DW_AT_language:
+		if (ellipsis)
+			return "language";
+		else
+			return "DW_AT_language";
+	case DW_AT_member:
+		if (ellipsis)
+			return "member";
+		else
+			return "DW_AT_member";
+	case DW_AT_discr:
+		if (ellipsis)
+			return "discr";
+		else
+			return "DW_AT_discr";
+	case DW_AT_discr_value:
+		if (ellipsis)
+			return "discr_value";
+		else
+			return "DW_AT_discr_value";
+	case DW_AT_visibility:
+		if (ellipsis)
+			return "visibility";
+		else
+			return "DW_AT_visibility";
+	case DW_AT_import:
+		if (ellipsis)
+			return "import";
+		else
+			return "DW_AT_import";
+	case DW_AT_string_length:
+		if (ellipsis)
+			return "string_length";
+		else
+			return "DW_AT_string_length";
+	case DW_AT_common_reference:
+		if (ellipsis)
+			return "common_reference";
+		else
+			return "DW_AT_common_reference";
+	case DW_AT_comp_dir:
+		if (ellipsis)
+			return "comp_dir";
+		else
+			return "DW_AT_comp_dir";
+	case DW_AT_const_value:
+		if (ellipsis)
+			return "const_value";
+		else
+			return "DW_AT_const_value";
+	case DW_AT_containing_type:
+		if (ellipsis)
+			return "containing_type";
+		else
+			return "DW_AT_containing_type";
+	case DW_AT_default_value:
+		if (ellipsis)
+			return "default_value";
+		else
+			return "DW_AT_default_value";
+	case DW_AT_inline:
+		if (ellipsis)
+			return "inline";
+		else
+			return "DW_AT_inline";
+	case DW_AT_is_optional:
+		if (ellipsis)
+			return "is_optional";
+		else
+			return "DW_AT_is_optional";
+	case DW_AT_lower_bound:
+		if (ellipsis)
+			return "lower_bound";
+		else
+			return "DW_AT_lower_bound";
+	case DW_AT_producer:
+		if (ellipsis)
+			return "producer";
+		else
+			return "DW_AT_producer";
+	case DW_AT_prototyped:
+		if (ellipsis)
+			return "prototyped";
+		else
+			return "DW_AT_prototyped";
+	case DW_AT_return_addr:
+		if (ellipsis)
+			return "return_addr";
+		else
+			return "DW_AT_return_addr";
+	case DW_AT_start_scope:
+		if (ellipsis)
+			return "start_scope";
+		else
+			return "DW_AT_start_scope";
+	case DW_AT_bit_stride:
+		if (ellipsis)
+			return "bit_stride";
+		else
+			return "DW_AT_bit_stride";
+	case DW_AT_upper_bound:
+		if (ellipsis)
+			return "upper_bound";
+		else
+			return "DW_AT_upper_bound";
+	case DW_AT_abstract_origin:
+		if (ellipsis)
+			return "abstract_origin";
+		else
+			return "DW_AT_abstract_origin";
+	case DW_AT_accessibility:
+		if (ellipsis)
+			return "accessibility";
+		else
+			return "DW_AT_accessibility";
+	case DW_AT_address_class:
+		if (ellipsis)
+			return "address_class";
+		else
+			return "DW_AT_address_class";
+	case DW_AT_artificial:
+		if (ellipsis)
+			return "artificial";
+		else
+			return "DW_AT_artificial";
+	case DW_AT_base_types:
+		if (ellipsis)
+			return "base_types";
+		else
+			return "DW_AT_base_types";
+	case DW_AT_calling_convention:
+		if (ellipsis)
+			return "calling_convention";
+		else
+			return "DW_AT_calling_convention";
+	case DW_AT_count:
+		if (ellipsis)
+			return "count";
+		else
+			return "DW_AT_count";
+	case DW_AT_data_member_location:
+		if (ellipsis)
+			return "data_member_location";
+		else
+			return "DW_AT_data_member_location";
+	case DW_AT_decl_column:
+		if (ellipsis)
+			return "decl_column";
+		else
+			return "DW_AT_decl_column";
+	case DW_AT_decl_file:
+		if (ellipsis)
+			return "decl_file";
+		else
+			return "DW_AT_decl_file";
+	case DW_AT_decl_line:
+		if (ellipsis)
+			return "decl_line";
+		else
+			return "DW_AT_decl_line";
+	case DW_AT_declaration:
+		if (ellipsis)
+			return "declaration";
+		else
+			return "DW_AT_declaration";
+	case DW_AT_discr_list:
+		if (ellipsis)
+			return "discr_list";
+		else
+			return "DW_AT_discr_list";
+	case DW_AT_encoding:
+		if (ellipsis)
+			return "encoding";
+		else
+			return "DW_AT_encoding";
+	case DW_AT_external:
+		if (ellipsis)
+			return "external";
+		else
+			return "DW_AT_external";
+	case DW_AT_frame_base:
+		if (ellipsis)
+			return "frame_base";
+		else
+			return "DW_AT_frame_base";
+	case DW_AT_friend:
+		if (ellipsis)
+			return "friend";
+		else
+			return "DW_AT_friend";
+	case DW_AT_identifier_case:
+		if (ellipsis)
+			return "identifier_case";
+		else
+			return "DW_AT_identifier_case";
+	case DW_AT_macro_info:
+		if (ellipsis)
+			return "macro_info";
+		else
+			return "DW_AT_macro_info";
+	case DW_AT_namelist_item:
+		if (ellipsis)
+			return "namelist_item";
+		else
+			return "DW_AT_namelist_item";
+	case DW_AT_priority:
+		if (ellipsis)
+			return "priority";
+		else
+			return "DW_AT_priority";
+	case DW_AT_segment:
+		if (ellipsis)
+			return "segment";
+		else
+			return "DW_AT_segment";
+	case DW_AT_specification:
+		if (ellipsis)
+			return "specification";
+		else
+			return "DW_AT_specification";
+	case DW_AT_static_link:
+		if (ellipsis)
+			return "static_link";
+		else
+			return "DW_AT_static_link";
+	case DW_AT_type:
+		if (ellipsis)
+			return "type";
+		else
+			return "DW_AT_type";
+	case DW_AT_use_location:
+		if (ellipsis)
+			return "use_location";
+		else
+			return "DW_AT_use_location";
+	case DW_AT_variable_parameter:
+		if (ellipsis)
+			return "variable_parameter";
+		else
+			return "DW_AT_variable_parameter";
+	case DW_AT_virtuality:
+		if (ellipsis)
+			return "virtuality";
+		else
+			return "DW_AT_virtuality";
+	case DW_AT_vtable_elem_location:
+		if (ellipsis)
+			return "vtable_elem_location";
+		else
+			return "DW_AT_vtable_elem_location";
+	case DW_AT_allocated:
+		if (ellipsis)
+			return "allocated";
+		else
+			return "DW_AT_allocated";
+	case DW_AT_associated:
+		if (ellipsis)
+			return "associated";
+		else
+			return "DW_AT_associated";
+	case DW_AT_data_location:
+		if (ellipsis)
+			return "data_location";
+		else
+			return "DW_AT_data_location";
+	case DW_AT_byte_stride:
+		if (ellipsis)
+			return "byte_stride";
+		else
+			return "DW_AT_byte_stride";
+	case DW_AT_entry_pc:
+		if (ellipsis)
+			return "entry_pc";
+		else
+			return "DW_AT_entry_pc";
+	case DW_AT_use_UTF8:
+		if (ellipsis)
+			return "use_UTF8";
+		else
+			return "DW_AT_use_UTF8";
+	case DW_AT_extension:
+		if (ellipsis)
+			return "extension";
+		else
+			return "DW_AT_extension";
+	case DW_AT_ranges:
+		if (ellipsis)
+			return "ranges";
+		else
+			return "DW_AT_ranges";
+	case DW_AT_trampoline:
+		if (ellipsis)
+			return "trampoline";
+		else
+			return "DW_AT_trampoline";
+	case DW_AT_call_column:
+		if (ellipsis)
+			return "call_column";
+		else
+			return "DW_AT_call_column";
+	case DW_AT_call_file:
+		if (ellipsis)
+			return "call_file";
+		else
+			return "DW_AT_call_file";
+	case DW_AT_call_line:
+		if (ellipsis)
+			return "call_line";
+		else
+			return "DW_AT_call_line";
+	case DW_AT_description:
+		if (ellipsis)
+			return "description";
+		else
+			return "DW_AT_description";
+	case DW_AT_binary_scale:
+		if (ellipsis)
+			return "binary_scale";
+		else
+			return "DW_AT_binary_scale";
+	case DW_AT_decimal_scale:
+		if (ellipsis)
+			return "decimal_scale";
+		else
+			return "DW_AT_decimal_scale";
+	case DW_AT_small:
+		if (ellipsis)
+			return "small";
+		else
+			return "DW_AT_small";
+	case DW_AT_decimal_sign:
+		if (ellipsis)
+			return "decimal_sign";
+		else
+			return "DW_AT_decimal_sign";
+	case DW_AT_digit_count:
+		if (ellipsis)
+			return "digit_count";
+		else
+			return "DW_AT_digit_count";
+	case DW_AT_picture_string:
+		if (ellipsis)
+			return "picture_string";
+		else
+			return "DW_AT_picture_string";
+	case DW_AT_mutable:
+		if (ellipsis)
+			return "mutable";
+		else
+			return "DW_AT_mutable";
+	case DW_AT_threads_scaled:
+		if (ellipsis)
+			return "threads_scaled";
+		else
+			return "DW_AT_threads_scaled";
+	case DW_AT_explicit:
+		if (ellipsis)
+			return "explicit";
+		else
+			return "DW_AT_explicit";
+	case DW_AT_object_pointer:
+		if (ellipsis)
+			return "object_pointer";
+		else
+			return "DW_AT_object_pointer";
+	case DW_AT_endianity:
+		if (ellipsis)
+			return "endianity";
+		else
+			return "DW_AT_endianity";
+	case DW_AT_elemental:
+		if (ellipsis)
+			return "elemental";
+		else
+			return "DW_AT_elemental";
+	case DW_AT_pure:
+		if (ellipsis)
+			return "pure";
+		else
+			return "DW_AT_pure";
+	case DW_AT_recursive:
+		if (ellipsis)
+			return "recursive";
+		else
+			return "DW_AT_recursive";
+	case DW_AT_signature:
+		if (ellipsis)
+			return "signature";
+		else
+			return "DW_AT_signature";
+	case DW_AT_main_subprogram:
+		if (ellipsis)
+			return "main_subprogram";
+		else
+			return "DW_AT_main_subprogram";
+	case DW_AT_data_bit_offset:
+		if (ellipsis)
+			return "data_bit_offset";
+		else
+			return "DW_AT_data_bit_offset";
+	case DW_AT_const_expr:
+		if (ellipsis)
+			return "const_expr";
+		else
+			return "DW_AT_const_expr";
+	case DW_AT_enum_class:
+		if (ellipsis)
+			return "enum_class";
+		else
+			return "DW_AT_enum_class";
+	case DW_AT_linkage_name:
+		if (ellipsis)
+			return "linkage_name";
+		else
+			return "DW_AT_linkage_name";
+	case DW_AT_HP_block_index:
+		if (ellipsis)
+			return "HP_block_index";
+		else
+			return "DW_AT_HP_block_index";
+	case DW_AT_MIPS_fde:
+		if (ellipsis)
+			return "MIPS_fde";
+		else
+			return "DW_AT_MIPS_fde";
+	case DW_AT_MIPS_loop_begin:
+		if (ellipsis)
+			return "MIPS_loop_begin";
+		else
+			return "DW_AT_MIPS_loop_begin";
+	case DW_AT_MIPS_tail_loop_begin:
+		if (ellipsis)
+			return "MIPS_tail_loop_begin";
+		else
+			return "DW_AT_MIPS_tail_loop_begin";
+	case DW_AT_MIPS_epilog_begin:
+		if (ellipsis)
+			return "MIPS_epilog_begin";
+		else
+			return "DW_AT_MIPS_epilog_begin";
+	case DW_AT_MIPS_loop_unroll_factor:
+		if (ellipsis)
+			return "MIPS_loop_unroll_factor";
+		else
+			return "DW_AT_MIPS_loop_unroll_factor";
+	case DW_AT_MIPS_software_pipeline_depth:
+		if (ellipsis)
+			return "MIPS_software_pipeline_depth";
+		else
+			return "DW_AT_MIPS_software_pipeline_depth";
+	case DW_AT_MIPS_linkage_name:
+		if (ellipsis)
+			return "MIPS_linkage_name";
+		else
+			return "DW_AT_MIPS_linkage_name";
+	case DW_AT_MIPS_stride:
+		if (ellipsis)
+			return "MIPS_stride";
+		else
+			return "DW_AT_MIPS_stride";
+	case DW_AT_MIPS_abstract_name:
+		if (ellipsis)
+			return "MIPS_abstract_name";
+		else
+			return "DW_AT_MIPS_abstract_name";
+	case DW_AT_MIPS_clone_origin:
+		if (ellipsis)
+			return "MIPS_clone_origin";
+		else
+			return "DW_AT_MIPS_clone_origin";
+	case DW_AT_MIPS_has_inlines:
+		if (ellipsis)
+			return "MIPS_has_inlines";
+		else
+			return "DW_AT_MIPS_has_inlines";
+	case DW_AT_MIPS_stride_byte:
+		if (ellipsis)
+			return "MIPS_stride_byte";
+		else
+			return "DW_AT_MIPS_stride_byte";
+	case DW_AT_MIPS_stride_elem:
+		if (ellipsis)
+			return "MIPS_stride_elem";
+		else
+			return "DW_AT_MIPS_stride_elem";
+	case DW_AT_MIPS_ptr_dopetype:
+		if (ellipsis)
+			return "MIPS_ptr_dopetype";
+		else
+			return "DW_AT_MIPS_ptr_dopetype";
+	case DW_AT_MIPS_allocatable_dopetype:
+		if (ellipsis)
+			return "MIPS_allocatable_dopetype";
+		else
+			return "DW_AT_MIPS_allocatable_dopetype";
+	case DW_AT_MIPS_assumed_shape_dopetype:
+		if (ellipsis)
+			return "MIPS_assumed_shape_dopetype";
+		else
+			return "DW_AT_MIPS_assumed_shape_dopetype";
+	case DW_AT_MIPS_assumed_size:
+		if (ellipsis)
+			return "MIPS_assumed_size";
+		else
+			return "DW_AT_MIPS_assumed_size";
+	case DW_AT_HP_raw_data_ptr:
+		if (ellipsis)
+			return "HP_raw_data_ptr";
+		else
+			return "DW_AT_HP_raw_data_ptr";
+	case DW_AT_HP_pass_by_reference:
+		if (ellipsis)
+			return "HP_pass_by_reference";
+		else
+			return "DW_AT_HP_pass_by_reference";
+	case DW_AT_HP_opt_level:
+		if (ellipsis)
+			return "HP_opt_level";
+		else
+			return "DW_AT_HP_opt_level";
+	case DW_AT_HP_prof_version_id:
+		if (ellipsis)
+			return "HP_prof_version_id";
+		else
+			return "DW_AT_HP_prof_version_id";
+	case DW_AT_HP_opt_flags:
+		if (ellipsis)
+			return "HP_opt_flags";
+		else
+			return "DW_AT_HP_opt_flags";
+	case DW_AT_HP_cold_region_low_pc:
+		if (ellipsis)
+			return "HP_cold_region_low_pc";
+		else
+			return "DW_AT_HP_cold_region_low_pc";
+	case DW_AT_HP_cold_region_high_pc:
+		if (ellipsis)
+			return "HP_cold_region_high_pc";
+		else
+			return "DW_AT_HP_cold_region_high_pc";
+	case DW_AT_HP_all_variables_modifiable:
+		if (ellipsis)
+			return "HP_all_variables_modifiable";
+		else
+			return "DW_AT_HP_all_variables_modifiable";
+	case DW_AT_HP_linkage_name:
+		if (ellipsis)
+			return "HP_linkage_name";
+		else
+			return "DW_AT_HP_linkage_name";
+	case DW_AT_HP_prof_flags:
+		if (ellipsis)
+			return "HP_prof_flags";
+		else
+			return "DW_AT_HP_prof_flags";
+	case DW_AT_INTEL_other_endian:
+		if (ellipsis)
+			return "INTEL_other_endian";
+		else
+			return "DW_AT_INTEL_other_endian";
+	case DW_AT_sf_names:
+		if (ellipsis)
+			return "sf_names";
+		else
+			return "DW_AT_sf_names";
+	case DW_AT_src_info:
+		if (ellipsis)
+			return "src_info";
+		else
+			return "DW_AT_src_info";
+	case DW_AT_mac_info:
+		if (ellipsis)
+			return "mac_info";
+		else
+			return "DW_AT_mac_info";
+	case DW_AT_src_coords:
+		if (ellipsis)
+			return "src_coords";
+		else
+			return "DW_AT_src_coords";
+	case DW_AT_body_begin:
+		if (ellipsis)
+			return "body_begin";
+		else
+			return "DW_AT_body_begin";
+	case DW_AT_body_end:
+		if (ellipsis)
+			return "body_end";
+		else
+			return "DW_AT_body_end";
+	case DW_AT_GNU_vector:
+		if (ellipsis)
+			return "GNU_vector";
+		else
+			return "DW_AT_GNU_vector";
+	case DW_AT_GNU_guarded_by:
+		if (ellipsis)
+			return "GNU_guarded_by";
+		else
+			return "DW_AT_GNU_guarded_by";
+	case DW_AT_GNU_pt_guarded_by:
+		if (ellipsis)
+			return "GNU_pt_guarded_by";
+		else
+			return "DW_AT_GNU_pt_guarded_by";
+	case DW_AT_GNU_guarded:
+		if (ellipsis)
+			return "GNU_guarded";
+		else
+			return "DW_AT_GNU_guarded";
+	case DW_AT_GNU_pt_guarded:
+		if (ellipsis)
+			return "GNU_pt_guarded";
+		else
+			return "DW_AT_GNU_pt_guarded";
+	case DW_AT_GNU_locks_excluded:
+		if (ellipsis)
+			return "GNU_locks_excluded";
+		else
+			return "DW_AT_GNU_locks_excluded";
+	case DW_AT_GNU_exclusive_locks_required:
+		if (ellipsis)
+			return "GNU_exclusive_locks_required";
+		else
+			return "DW_AT_GNU_exclusive_locks_required";
+	case DW_AT_GNU_shared_locks_required:
+		if (ellipsis)
+			return "GNU_shared_locks_required";
+		else
+			return "DW_AT_GNU_shared_locks_required";
+	case DW_AT_GNU_odr_signature:
+		if (ellipsis)
+			return "GNU_odr_signature";
+		else
+			return "DW_AT_GNU_odr_signature";
+	case DW_AT_GNU_template_name:
+		if (ellipsis)
+			return "GNU_template_name";
+		else
+			return "DW_AT_GNU_template_name";
+	case DW_AT_GNU_call_site_value:
+		if (ellipsis)
+			return "GNU_call_site_value";
+		else
+			return "DW_AT_GNU_call_site_value";
+	case DW_AT_GNU_call_site_data_value:
+		if (ellipsis)
+			return "GNU_call_site_data_value";
+		else
+			return "DW_AT_GNU_call_site_data_value";
+	case DW_AT_GNU_call_site_target:
+		if (ellipsis)
+			return "GNU_call_site_target";
+		else
+			return "DW_AT_GNU_call_site_target";
+	case DW_AT_GNU_call_site_target_clobbered:
+		if (ellipsis)
+			return "GNU_call_site_target_clobbered";
+		else
+			return "DW_AT_GNU_call_site_target_clobbered";
+	case DW_AT_GNU_tail_call:
+		if (ellipsis)
+			return "GNU_tail_call";
+		else
+			return "DW_AT_GNU_tail_call";
+	case DW_AT_GNU_all_tail_call_sites:
+		if (ellipsis)
+			return "GNU_all_tail_call_sites";
+		else
+			return "DW_AT_GNU_all_tail_call_sites";
+	case DW_AT_GNU_all_call_sites:
+		if (ellipsis)
+			return "GNU_all_call_sites";
+		else
+			return "DW_AT_GNU_all_call_sites";
+	case DW_AT_GNU_all_source_call_sites:
+		if (ellipsis)
+			return "GNU_all_source_call_sites";
+		else
+			return "DW_AT_GNU_all_source_call_sites";
+	case DW_AT_ALTIUM_loclist:
+		if (ellipsis)
+			return "ALTIUM_loclist";
+		else
+			return "DW_AT_ALTIUM_loclist";
+	case DW_AT_SUN_template:
+		if (ellipsis)
+			return "SUN_template";
+		else
+			return "DW_AT_SUN_template";
+	case DW_AT_SUN_alignment:
+		if (ellipsis)
+			return "SUN_alignment";
+		else
+			return "DW_AT_SUN_alignment";
+	case DW_AT_SUN_vtable:
+		if (ellipsis)
+			return "SUN_vtable";
+		else
+			return "DW_AT_SUN_vtable";
+	case DW_AT_SUN_count_guarantee:
+		if (ellipsis)
+			return "SUN_count_guarantee";
+		else
+			return "DW_AT_SUN_count_guarantee";
+	case DW_AT_SUN_command_line:
+		if (ellipsis)
+			return "SUN_command_line";
+		else
+			return "DW_AT_SUN_command_line";
+	case DW_AT_SUN_vbase:
+		if (ellipsis)
+			return "SUN_vbase";
+		else
+			return "DW_AT_SUN_vbase";
+	case DW_AT_SUN_compile_options:
+		if (ellipsis)
+			return "SUN_compile_options";
+		else
+			return "DW_AT_SUN_compile_options";
+	case DW_AT_SUN_language:
+		if (ellipsis)
+			return "SUN_language";
+		else
+			return "DW_AT_SUN_language";
+	case DW_AT_SUN_browser_file:
+		if (ellipsis)
+			return "SUN_browser_file";
+		else
+			return "DW_AT_SUN_browser_file";
+	case DW_AT_SUN_vtable_abi:
+		if (ellipsis)
+			return "SUN_vtable_abi";
+		else
+			return "DW_AT_SUN_vtable_abi";
+	case DW_AT_SUN_func_offsets:
+		if (ellipsis)
+			return "SUN_func_offsets";
+		else
+			return "DW_AT_SUN_func_offsets";
+	case DW_AT_SUN_cf_kind:
+		if (ellipsis)
+			return "SUN_cf_kind";
+		else
+			return "DW_AT_SUN_cf_kind";
+	case DW_AT_SUN_vtable_index:
+		if (ellipsis)
+			return "SUN_vtable_index";
+		else
+			return "DW_AT_SUN_vtable_index";
+	case DW_AT_SUN_omp_tpriv_addr:
+		if (ellipsis)
+			return "SUN_omp_tpriv_addr";
+		else
+			return "DW_AT_SUN_omp_tpriv_addr";
+	case DW_AT_SUN_omp_child_func:
+		if (ellipsis)
+			return "SUN_omp_child_func";
+		else
+			return "DW_AT_SUN_omp_child_func";
+	case DW_AT_SUN_func_offset:
+		if (ellipsis)
+			return "SUN_func_offset";
+		else
+			return "DW_AT_SUN_func_offset";
+	case DW_AT_SUN_memop_type_ref:
+		if (ellipsis)
+			return "SUN_memop_type_ref";
+		else
+			return "DW_AT_SUN_memop_type_ref";
+	case DW_AT_SUN_profile_id:
+		if (ellipsis)
+			return "SUN_profile_id";
+		else
+			return "DW_AT_SUN_profile_id";
+	case DW_AT_SUN_memop_signature:
+		if (ellipsis)
+			return "SUN_memop_signature";
+		else
+			return "DW_AT_SUN_memop_signature";
+	case DW_AT_SUN_obj_dir:
+		if (ellipsis)
+			return "SUN_obj_dir";
+		else
+			return "DW_AT_SUN_obj_dir";
+	case DW_AT_SUN_obj_file:
+		if (ellipsis)
+			return "SUN_obj_file";
+		else
+			return "DW_AT_SUN_obj_file";
+	case DW_AT_SUN_original_name:
+		if (ellipsis)
+			return "SUN_original_name";
+		else
+			return "DW_AT_SUN_original_name";
+	case DW_AT_SUN_hwcprof_signature:
+		if (ellipsis)
+			return "SUN_hwcprof_signature";
+		else
+			return "DW_AT_SUN_hwcprof_signature";
+	case DW_AT_SUN_amd64_parmdump:
+		if (ellipsis)
+			return "SUN_amd64_parmdump";
+		else
+			return "DW_AT_SUN_amd64_parmdump";
+	case DW_AT_SUN_part_link_name:
+		if (ellipsis)
+			return "SUN_part_link_name";
+		else
+			return "DW_AT_SUN_part_link_name";
+	case DW_AT_SUN_link_name:
+		if (ellipsis)
+			return "SUN_link_name";
+		else
+			return "DW_AT_SUN_link_name";
+	case DW_AT_SUN_pass_with_const:
+		if (ellipsis)
+			return "SUN_pass_with_const";
+		else
+			return "DW_AT_SUN_pass_with_const";
+	case DW_AT_SUN_return_with_const:
+		if (ellipsis)
+			return "SUN_return_with_const";
+		else
+			return "DW_AT_SUN_return_with_const";
+	case DW_AT_SUN_import_by_name:
+		if (ellipsis)
+			return "SUN_import_by_name";
+		else
+			return "DW_AT_SUN_import_by_name";
+	case DW_AT_SUN_f90_pointer:
+		if (ellipsis)
+			return "SUN_f90_pointer";
+		else
+			return "DW_AT_SUN_f90_pointer";
+	case DW_AT_SUN_pass_by_ref:
+		if (ellipsis)
+			return "SUN_pass_by_ref";
+		else
+			return "DW_AT_SUN_pass_by_ref";
+	case DW_AT_SUN_f90_allocatable:
+		if (ellipsis)
+			return "SUN_f90_allocatable";
+		else
+			return "DW_AT_SUN_f90_allocatable";
+	case DW_AT_SUN_f90_assumed_shape_array:
+		if (ellipsis)
+			return "SUN_f90_assumed_shape_array";
+		else
+			return "DW_AT_SUN_f90_assumed_shape_array";
+	case DW_AT_SUN_c_vla:
+		if (ellipsis)
+			return "SUN_c_vla";
+		else
+			return "DW_AT_SUN_c_vla";
+	case DW_AT_SUN_return_value_ptr:
+		if (ellipsis)
+			return "SUN_return_value_ptr";
+		else
+			return "DW_AT_SUN_return_value_ptr";
+	case DW_AT_SUN_dtor_start:
+		if (ellipsis)
+			return "SUN_dtor_start";
+		else
+			return "DW_AT_SUN_dtor_start";
+	case DW_AT_SUN_dtor_length:
+		if (ellipsis)
+			return "SUN_dtor_length";
+		else
+			return "DW_AT_SUN_dtor_length";
+	case DW_AT_SUN_dtor_state_initial:
+		if (ellipsis)
+			return "SUN_dtor_state_initial";
+		else
+			return "DW_AT_SUN_dtor_state_initial";
+	case DW_AT_SUN_dtor_state_final:
+		if (ellipsis)
+			return "SUN_dtor_state_final";
+		else
+			return "DW_AT_SUN_dtor_state_final";
+	case DW_AT_SUN_dtor_state_deltas:
+		if (ellipsis)
+			return "SUN_dtor_state_deltas";
+		else
+			return "DW_AT_SUN_dtor_state_deltas";
+	case DW_AT_SUN_import_by_lname:
+		if (ellipsis)
+			return "SUN_import_by_lname";
+		else
+			return "DW_AT_SUN_import_by_lname";
+	case DW_AT_SUN_f90_use_only:
+		if (ellipsis)
+			return "SUN_f90_use_only";
+		else
+			return "DW_AT_SUN_f90_use_only";
+	case DW_AT_SUN_namelist_spec:
+		if (ellipsis)
+			return "SUN_namelist_spec";
+		else
+			return "DW_AT_SUN_namelist_spec";
+	case DW_AT_SUN_is_omp_child_func:
+		if (ellipsis)
+			return "SUN_is_omp_child_func";
+		else
+			return "DW_AT_SUN_is_omp_child_func";
+	case DW_AT_SUN_fortran_main_alias:
+		if (ellipsis)
+			return "SUN_fortran_main_alias";
+		else
+			return "DW_AT_SUN_fortran_main_alias";
+	case DW_AT_SUN_fortran_based:
+		if (ellipsis)
+			return "SUN_fortran_based";
+		else
+			return "DW_AT_SUN_fortran_based";
+	case DW_AT_use_GNAT_descriptive_type:
+		if (ellipsis)
+			return "use_GNAT_descriptive_type";
+		else
+			return "DW_AT_use_GNAT_descriptive_type";
+	case DW_AT_GNAT_descriptive_type:
+		if (ellipsis)
+			return "GNAT_descriptive_type";
+		else
+			return "DW_AT_GNAT_descriptive_type";
+	case DW_AT_upc_threads_scaled:
+		if (ellipsis)
+			return "upc_threads_scaled";
+		else
+			return "DW_AT_upc_threads_scaled";
+	case DW_AT_PGI_lbase:
+		if (ellipsis)
+			return "PGI_lbase";
+		else
+			return "DW_AT_PGI_lbase";
+	case DW_AT_PGI_soffset:
+		if (ellipsis)
+			return "PGI_soffset";
+		else
+			return "DW_AT_PGI_soffset";
+	case DW_AT_PGI_lstride:
+		if (ellipsis)
+			return "PGI_lstride";
+		else
+			return "DW_AT_PGI_lstride";
+	case DW_AT_APPLE_optimized:
+		if (ellipsis)
+			return "APPLE_optimized";
+		else
+			return "DW_AT_APPLE_optimized";
+	case DW_AT_APPLE_flags:
+		if (ellipsis)
+			return "APPLE_flags";
+		else
+			return "DW_AT_APPLE_flags";
+	case DW_AT_APPLE_isa:
+		if (ellipsis)
+			return "APPLE_isa";
+		else
+			return "DW_AT_APPLE_isa";
+	case DW_AT_APPLE_block:
+		if (ellipsis)
+			return "APPLE_block";
+		else
+			return "DW_AT_APPLE_block";
+	case DW_AT_APPLE_major_runtime_vers:
+		if (ellipsis)
+			return "APPLE_major_runtime_vers";
+		else
+			return "DW_AT_APPLE_major_runtime_vers";
+	case DW_AT_APPLE_runtime_class:
+		if (ellipsis)
+			return "APPLE_runtime_class";
+		else
+			return "DW_AT_APPLE_runtime_class";
+	case DW_AT_APPLE_omit_frame_ptr:
+		if (ellipsis)
+			return "APPLE_omit_frame_ptr";
+		else
+			return "DW_AT_APPLE_omit_frame_ptr";
+	case DW_AT_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_AT_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown AT value 0x%x>",(int)val);
+		 fprintf(stderr,"AT of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_OP_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_OP_addr:
+		if (ellipsis)
+			return "addr";
+		else
+			return "DW_OP_addr";
+	case DW_OP_deref:
+		if (ellipsis)
+			return "deref";
+		else
+			return "DW_OP_deref";
+	case DW_OP_const1u:
+		if (ellipsis)
+			return "const1u";
+		else
+			return "DW_OP_const1u";
+	case DW_OP_const1s:
+		if (ellipsis)
+			return "const1s";
+		else
+			return "DW_OP_const1s";
+	case DW_OP_const2u:
+		if (ellipsis)
+			return "const2u";
+		else
+			return "DW_OP_const2u";
+	case DW_OP_const2s:
+		if (ellipsis)
+			return "const2s";
+		else
+			return "DW_OP_const2s";
+	case DW_OP_const4u:
+		if (ellipsis)
+			return "const4u";
+		else
+			return "DW_OP_const4u";
+	case DW_OP_const4s:
+		if (ellipsis)
+			return "const4s";
+		else
+			return "DW_OP_const4s";
+	case DW_OP_const8u:
+		if (ellipsis)
+			return "const8u";
+		else
+			return "DW_OP_const8u";
+	case DW_OP_const8s:
+		if (ellipsis)
+			return "const8s";
+		else
+			return "DW_OP_const8s";
+	case DW_OP_constu:
+		if (ellipsis)
+			return "constu";
+		else
+			return "DW_OP_constu";
+	case DW_OP_consts:
+		if (ellipsis)
+			return "consts";
+		else
+			return "DW_OP_consts";
+	case DW_OP_dup:
+		if (ellipsis)
+			return "dup";
+		else
+			return "DW_OP_dup";
+	case DW_OP_drop:
+		if (ellipsis)
+			return "drop";
+		else
+			return "DW_OP_drop";
+	case DW_OP_over:
+		if (ellipsis)
+			return "over";
+		else
+			return "DW_OP_over";
+	case DW_OP_pick:
+		if (ellipsis)
+			return "pick";
+		else
+			return "DW_OP_pick";
+	case DW_OP_swap:
+		if (ellipsis)
+			return "swap";
+		else
+			return "DW_OP_swap";
+	case DW_OP_rot:
+		if (ellipsis)
+			return "rot";
+		else
+			return "DW_OP_rot";
+	case DW_OP_xderef:
+		if (ellipsis)
+			return "xderef";
+		else
+			return "DW_OP_xderef";
+	case DW_OP_abs:
+		if (ellipsis)
+			return "abs";
+		else
+			return "DW_OP_abs";
+	case DW_OP_and:
+		if (ellipsis)
+			return "and";
+		else
+			return "DW_OP_and";
+	case DW_OP_div:
+		if (ellipsis)
+			return "div";
+		else
+			return "DW_OP_div";
+	case DW_OP_minus:
+		if (ellipsis)
+			return "minus";
+		else
+			return "DW_OP_minus";
+	case DW_OP_mod:
+		if (ellipsis)
+			return "mod";
+		else
+			return "DW_OP_mod";
+	case DW_OP_mul:
+		if (ellipsis)
+			return "mul";
+		else
+			return "DW_OP_mul";
+	case DW_OP_neg:
+		if (ellipsis)
+			return "neg";
+		else
+			return "DW_OP_neg";
+	case DW_OP_not:
+		if (ellipsis)
+			return "not";
+		else
+			return "DW_OP_not";
+	case DW_OP_or:
+		if (ellipsis)
+			return "or";
+		else
+			return "DW_OP_or";
+	case DW_OP_plus:
+		if (ellipsis)
+			return "plus";
+		else
+			return "DW_OP_plus";
+	case DW_OP_plus_uconst:
+		if (ellipsis)
+			return "plus_uconst";
+		else
+			return "DW_OP_plus_uconst";
+	case DW_OP_shl:
+		if (ellipsis)
+			return "shl";
+		else
+			return "DW_OP_shl";
+	case DW_OP_shr:
+		if (ellipsis)
+			return "shr";
+		else
+			return "DW_OP_shr";
+	case DW_OP_shra:
+		if (ellipsis)
+			return "shra";
+		else
+			return "DW_OP_shra";
+	case DW_OP_xor:
+		if (ellipsis)
+			return "xor";
+		else
+			return "DW_OP_xor";
+	case DW_OP_bra:
+		if (ellipsis)
+			return "bra";
+		else
+			return "DW_OP_bra";
+	case DW_OP_eq:
+		if (ellipsis)
+			return "eq";
+		else
+			return "DW_OP_eq";
+	case DW_OP_ge:
+		if (ellipsis)
+			return "ge";
+		else
+			return "DW_OP_ge";
+	case DW_OP_gt:
+		if (ellipsis)
+			return "gt";
+		else
+			return "DW_OP_gt";
+	case DW_OP_le:
+		if (ellipsis)
+			return "le";
+		else
+			return "DW_OP_le";
+	case DW_OP_lt:
+		if (ellipsis)
+			return "lt";
+		else
+			return "DW_OP_lt";
+	case DW_OP_ne:
+		if (ellipsis)
+			return "ne";
+		else
+			return "DW_OP_ne";
+	case DW_OP_skip:
+		if (ellipsis)
+			return "skip";
+		else
+			return "DW_OP_skip";
+	case DW_OP_lit0:
+		if (ellipsis)
+			return "lit0";
+		else
+			return "DW_OP_lit0";
+	case DW_OP_lit1:
+		if (ellipsis)
+			return "lit1";
+		else
+			return "DW_OP_lit1";
+	case DW_OP_lit2:
+		if (ellipsis)
+			return "lit2";
+		else
+			return "DW_OP_lit2";
+	case DW_OP_lit3:
+		if (ellipsis)
+			return "lit3";
+		else
+			return "DW_OP_lit3";
+	case DW_OP_lit4:
+		if (ellipsis)
+			return "lit4";
+		else
+			return "DW_OP_lit4";
+	case DW_OP_lit5:
+		if (ellipsis)
+			return "lit5";
+		else
+			return "DW_OP_lit5";
+	case DW_OP_lit6:
+		if (ellipsis)
+			return "lit6";
+		else
+			return "DW_OP_lit6";
+	case DW_OP_lit7:
+		if (ellipsis)
+			return "lit7";
+		else
+			return "DW_OP_lit7";
+	case DW_OP_lit8:
+		if (ellipsis)
+			return "lit8";
+		else
+			return "DW_OP_lit8";
+	case DW_OP_lit9:
+		if (ellipsis)
+			return "lit9";
+		else
+			return "DW_OP_lit9";
+	case DW_OP_lit10:
+		if (ellipsis)
+			return "lit10";
+		else
+			return "DW_OP_lit10";
+	case DW_OP_lit11:
+		if (ellipsis)
+			return "lit11";
+		else
+			return "DW_OP_lit11";
+	case DW_OP_lit12:
+		if (ellipsis)
+			return "lit12";
+		else
+			return "DW_OP_lit12";
+	case DW_OP_lit13:
+		if (ellipsis)
+			return "lit13";
+		else
+			return "DW_OP_lit13";
+	case DW_OP_lit14:
+		if (ellipsis)
+			return "lit14";
+		else
+			return "DW_OP_lit14";
+	case DW_OP_lit15:
+		if (ellipsis)
+			return "lit15";
+		else
+			return "DW_OP_lit15";
+	case DW_OP_lit16:
+		if (ellipsis)
+			return "lit16";
+		else
+			return "DW_OP_lit16";
+	case DW_OP_lit17:
+		if (ellipsis)
+			return "lit17";
+		else
+			return "DW_OP_lit17";
+	case DW_OP_lit18:
+		if (ellipsis)
+			return "lit18";
+		else
+			return "DW_OP_lit18";
+	case DW_OP_lit19:
+		if (ellipsis)
+			return "lit19";
+		else
+			return "DW_OP_lit19";
+	case DW_OP_lit20:
+		if (ellipsis)
+			return "lit20";
+		else
+			return "DW_OP_lit20";
+	case DW_OP_lit21:
+		if (ellipsis)
+			return "lit21";
+		else
+			return "DW_OP_lit21";
+	case DW_OP_lit22:
+		if (ellipsis)
+			return "lit22";
+		else
+			return "DW_OP_lit22";
+	case DW_OP_lit23:
+		if (ellipsis)
+			return "lit23";
+		else
+			return "DW_OP_lit23";
+	case DW_OP_lit24:
+		if (ellipsis)
+			return "lit24";
+		else
+			return "DW_OP_lit24";
+	case DW_OP_lit25:
+		if (ellipsis)
+			return "lit25";
+		else
+			return "DW_OP_lit25";
+	case DW_OP_lit26:
+		if (ellipsis)
+			return "lit26";
+		else
+			return "DW_OP_lit26";
+	case DW_OP_lit27:
+		if (ellipsis)
+			return "lit27";
+		else
+			return "DW_OP_lit27";
+	case DW_OP_lit28:
+		if (ellipsis)
+			return "lit28";
+		else
+			return "DW_OP_lit28";
+	case DW_OP_lit29:
+		if (ellipsis)
+			return "lit29";
+		else
+			return "DW_OP_lit29";
+	case DW_OP_lit30:
+		if (ellipsis)
+			return "lit30";
+		else
+			return "DW_OP_lit30";
+	case DW_OP_lit31:
+		if (ellipsis)
+			return "lit31";
+		else
+			return "DW_OP_lit31";
+	case DW_OP_reg0:
+		if (ellipsis)
+			return "reg0";
+		else
+			return "DW_OP_reg0";
+	case DW_OP_reg1:
+		if (ellipsis)
+			return "reg1";
+		else
+			return "DW_OP_reg1";
+	case DW_OP_reg2:
+		if (ellipsis)
+			return "reg2";
+		else
+			return "DW_OP_reg2";
+	case DW_OP_reg3:
+		if (ellipsis)
+			return "reg3";
+		else
+			return "DW_OP_reg3";
+	case DW_OP_reg4:
+		if (ellipsis)
+			return "reg4";
+		else
+			return "DW_OP_reg4";
+	case DW_OP_reg5:
+		if (ellipsis)
+			return "reg5";
+		else
+			return "DW_OP_reg5";
+	case DW_OP_reg6:
+		if (ellipsis)
+			return "reg6";
+		else
+			return "DW_OP_reg6";
+	case DW_OP_reg7:
+		if (ellipsis)
+			return "reg7";
+		else
+			return "DW_OP_reg7";
+	case DW_OP_reg8:
+		if (ellipsis)
+			return "reg8";
+		else
+			return "DW_OP_reg8";
+	case DW_OP_reg9:
+		if (ellipsis)
+			return "reg9";
+		else
+			return "DW_OP_reg9";
+	case DW_OP_reg10:
+		if (ellipsis)
+			return "reg10";
+		else
+			return "DW_OP_reg10";
+	case DW_OP_reg11:
+		if (ellipsis)
+			return "reg11";
+		else
+			return "DW_OP_reg11";
+	case DW_OP_reg12:
+		if (ellipsis)
+			return "reg12";
+		else
+			return "DW_OP_reg12";
+	case DW_OP_reg13:
+		if (ellipsis)
+			return "reg13";
+		else
+			return "DW_OP_reg13";
+	case DW_OP_reg14:
+		if (ellipsis)
+			return "reg14";
+		else
+			return "DW_OP_reg14";
+	case DW_OP_reg15:
+		if (ellipsis)
+			return "reg15";
+		else
+			return "DW_OP_reg15";
+	case DW_OP_reg16:
+		if (ellipsis)
+			return "reg16";
+		else
+			return "DW_OP_reg16";
+	case DW_OP_reg17:
+		if (ellipsis)
+			return "reg17";
+		else
+			return "DW_OP_reg17";
+	case DW_OP_reg18:
+		if (ellipsis)
+			return "reg18";
+		else
+			return "DW_OP_reg18";
+	case DW_OP_reg19:
+		if (ellipsis)
+			return "reg19";
+		else
+			return "DW_OP_reg19";
+	case DW_OP_reg20:
+		if (ellipsis)
+			return "reg20";
+		else
+			return "DW_OP_reg20";
+	case DW_OP_reg21:
+		if (ellipsis)
+			return "reg21";
+		else
+			return "DW_OP_reg21";
+	case DW_OP_reg22:
+		if (ellipsis)
+			return "reg22";
+		else
+			return "DW_OP_reg22";
+	case DW_OP_reg23:
+		if (ellipsis)
+			return "reg23";
+		else
+			return "DW_OP_reg23";
+	case DW_OP_reg24:
+		if (ellipsis)
+			return "reg24";
+		else
+			return "DW_OP_reg24";
+	case DW_OP_reg25:
+		if (ellipsis)
+			return "reg25";
+		else
+			return "DW_OP_reg25";
+	case DW_OP_reg26:
+		if (ellipsis)
+			return "reg26";
+		else
+			return "DW_OP_reg26";
+	case DW_OP_reg27:
+		if (ellipsis)
+			return "reg27";
+		else
+			return "DW_OP_reg27";
+	case DW_OP_reg28:
+		if (ellipsis)
+			return "reg28";
+		else
+			return "DW_OP_reg28";
+	case DW_OP_reg29:
+		if (ellipsis)
+			return "reg29";
+		else
+			return "DW_OP_reg29";
+	case DW_OP_reg30:
+		if (ellipsis)
+			return "reg30";
+		else
+			return "DW_OP_reg30";
+	case DW_OP_reg31:
+		if (ellipsis)
+			return "reg31";
+		else
+			return "DW_OP_reg31";
+	case DW_OP_breg0:
+		if (ellipsis)
+			return "breg0";
+		else
+			return "DW_OP_breg0";
+	case DW_OP_breg1:
+		if (ellipsis)
+			return "breg1";
+		else
+			return "DW_OP_breg1";
+	case DW_OP_breg2:
+		if (ellipsis)
+			return "breg2";
+		else
+			return "DW_OP_breg2";
+	case DW_OP_breg3:
+		if (ellipsis)
+			return "breg3";
+		else
+			return "DW_OP_breg3";
+	case DW_OP_breg4:
+		if (ellipsis)
+			return "breg4";
+		else
+			return "DW_OP_breg4";
+	case DW_OP_breg5:
+		if (ellipsis)
+			return "breg5";
+		else
+			return "DW_OP_breg5";
+	case DW_OP_breg6:
+		if (ellipsis)
+			return "breg6";
+		else
+			return "DW_OP_breg6";
+	case DW_OP_breg7:
+		if (ellipsis)
+			return "breg7";
+		else
+			return "DW_OP_breg7";
+	case DW_OP_breg8:
+		if (ellipsis)
+			return "breg8";
+		else
+			return "DW_OP_breg8";
+	case DW_OP_breg9:
+		if (ellipsis)
+			return "breg9";
+		else
+			return "DW_OP_breg9";
+	case DW_OP_breg10:
+		if (ellipsis)
+			return "breg10";
+		else
+			return "DW_OP_breg10";
+	case DW_OP_breg11:
+		if (ellipsis)
+			return "breg11";
+		else
+			return "DW_OP_breg11";
+	case DW_OP_breg12:
+		if (ellipsis)
+			return "breg12";
+		else
+			return "DW_OP_breg12";
+	case DW_OP_breg13:
+		if (ellipsis)
+			return "breg13";
+		else
+			return "DW_OP_breg13";
+	case DW_OP_breg14:
+		if (ellipsis)
+			return "breg14";
+		else
+			return "DW_OP_breg14";
+	case DW_OP_breg15:
+		if (ellipsis)
+			return "breg15";
+		else
+			return "DW_OP_breg15";
+	case DW_OP_breg16:
+		if (ellipsis)
+			return "breg16";
+		else
+			return "DW_OP_breg16";
+	case DW_OP_breg17:
+		if (ellipsis)
+			return "breg17";
+		else
+			return "DW_OP_breg17";
+	case DW_OP_breg18:
+		if (ellipsis)
+			return "breg18";
+		else
+			return "DW_OP_breg18";
+	case DW_OP_breg19:
+		if (ellipsis)
+			return "breg19";
+		else
+			return "DW_OP_breg19";
+	case DW_OP_breg20:
+		if (ellipsis)
+			return "breg20";
+		else
+			return "DW_OP_breg20";
+	case DW_OP_breg21:
+		if (ellipsis)
+			return "breg21";
+		else
+			return "DW_OP_breg21";
+	case DW_OP_breg22:
+		if (ellipsis)
+			return "breg22";
+		else
+			return "DW_OP_breg22";
+	case DW_OP_breg23:
+		if (ellipsis)
+			return "breg23";
+		else
+			return "DW_OP_breg23";
+	case DW_OP_breg24:
+		if (ellipsis)
+			return "breg24";
+		else
+			return "DW_OP_breg24";
+	case DW_OP_breg25:
+		if (ellipsis)
+			return "breg25";
+		else
+			return "DW_OP_breg25";
+	case DW_OP_breg26:
+		if (ellipsis)
+			return "breg26";
+		else
+			return "DW_OP_breg26";
+	case DW_OP_breg27:
+		if (ellipsis)
+			return "breg27";
+		else
+			return "DW_OP_breg27";
+	case DW_OP_breg28:
+		if (ellipsis)
+			return "breg28";
+		else
+			return "DW_OP_breg28";
+	case DW_OP_breg29:
+		if (ellipsis)
+			return "breg29";
+		else
+			return "DW_OP_breg29";
+	case DW_OP_breg30:
+		if (ellipsis)
+			return "breg30";
+		else
+			return "DW_OP_breg30";
+	case DW_OP_breg31:
+		if (ellipsis)
+			return "breg31";
+		else
+			return "DW_OP_breg31";
+	case DW_OP_regx:
+		if (ellipsis)
+			return "regx";
+		else
+			return "DW_OP_regx";
+	case DW_OP_fbreg:
+		if (ellipsis)
+			return "fbreg";
+		else
+			return "DW_OP_fbreg";
+	case DW_OP_bregx:
+		if (ellipsis)
+			return "bregx";
+		else
+			return "DW_OP_bregx";
+	case DW_OP_piece:
+		if (ellipsis)
+			return "piece";
+		else
+			return "DW_OP_piece";
+	case DW_OP_deref_size:
+		if (ellipsis)
+			return "deref_size";
+		else
+			return "DW_OP_deref_size";
+	case DW_OP_xderef_size:
+		if (ellipsis)
+			return "xderef_size";
+		else
+			return "DW_OP_xderef_size";
+	case DW_OP_nop:
+		if (ellipsis)
+			return "nop";
+		else
+			return "DW_OP_nop";
+	case DW_OP_push_object_address:
+		if (ellipsis)
+			return "push_object_address";
+		else
+			return "DW_OP_push_object_address";
+	case DW_OP_call2:
+		if (ellipsis)
+			return "call2";
+		else
+			return "DW_OP_call2";
+	case DW_OP_call4:
+		if (ellipsis)
+			return "call4";
+		else
+			return "DW_OP_call4";
+	case DW_OP_call_ref:
+		if (ellipsis)
+			return "call_ref";
+		else
+			return "DW_OP_call_ref";
+	case DW_OP_form_tls_address:
+		if (ellipsis)
+			return "form_tls_address";
+		else
+			return "DW_OP_form_tls_address";
+	case DW_OP_call_frame_cfa:
+		if (ellipsis)
+			return "call_frame_cfa";
+		else
+			return "DW_OP_call_frame_cfa";
+	case DW_OP_bit_piece:
+		if (ellipsis)
+			return "bit_piece";
+		else
+			return "DW_OP_bit_piece";
+	case DW_OP_implicit_value:
+		if (ellipsis)
+			return "implicit_value";
+		else
+			return "DW_OP_implicit_value";
+	case DW_OP_stack_value:
+		if (ellipsis)
+			return "stack_value";
+		else
+			return "DW_OP_stack_value";
+	case DW_OP_GNU_push_tls_address:
+		if (ellipsis)
+			return "GNU_push_tls_address";
+		else
+			return "DW_OP_GNU_push_tls_address";
+	case DW_OP_GNU_uninit:
+		if (ellipsis)
+			return "GNU_uninit";
+		else
+			return "DW_OP_GNU_uninit";
+	case DW_OP_GNU_encoded_addr:
+		if (ellipsis)
+			return "GNU_encoded_addr";
+		else
+			return "DW_OP_GNU_encoded_addr";
+	case DW_OP_GNU_implicit_pointer:
+		if (ellipsis)
+			return "GNU_implicit_pointer";
+		else
+			return "DW_OP_GNU_implicit_pointer";
+	case DW_OP_GNU_entry_value:
+		if (ellipsis)
+			return "GNU_entry_value";
+		else
+			return "DW_OP_GNU_entry_value";
+	case DW_OP_HP_is_value:
+		if (ellipsis)
+			return "HP_is_value";
+		else
+			return "DW_OP_HP_is_value";
+	case DW_OP_HP_fltconst4:
+		if (ellipsis)
+			return "HP_fltconst4";
+		else
+			return "DW_OP_HP_fltconst4";
+	case DW_OP_HP_fltconst8:
+		if (ellipsis)
+			return "HP_fltconst8";
+		else
+			return "DW_OP_HP_fltconst8";
+	case DW_OP_HP_mod_range:
+		if (ellipsis)
+			return "HP_mod_range";
+		else
+			return "DW_OP_HP_mod_range";
+	case DW_OP_HP_unmod_range:
+		if (ellipsis)
+			return "HP_unmod_range";
+		else
+			return "DW_OP_HP_unmod_range";
+	case DW_OP_HP_tls:
+		if (ellipsis)
+			return "HP_tls";
+		else
+			return "DW_OP_HP_tls";
+	case DW_OP_INTEL_bit_piece:
+		if (ellipsis)
+			return "INTEL_bit_piece";
+		else
+			return "DW_OP_INTEL_bit_piece";
+	case DW_OP_PGI_omp_thread_num:
+		if (ellipsis)
+			return "PGI_omp_thread_num";
+		else
+			return "DW_OP_PGI_omp_thread_num";
+	case DW_OP_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_OP_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown OP value 0x%x>",(int)val);
+		 fprintf(stderr,"OP of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ATE_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ATE_address:
+		if (ellipsis)
+			return "address";
+		else
+			return "DW_ATE_address";
+	case DW_ATE_boolean:
+		if (ellipsis)
+			return "boolean";
+		else
+			return "DW_ATE_boolean";
+	case DW_ATE_complex_float:
+		if (ellipsis)
+			return "complex_float";
+		else
+			return "DW_ATE_complex_float";
+	case DW_ATE_float:
+		if (ellipsis)
+			return "float";
+		else
+			return "DW_ATE_float";
+	case DW_ATE_signed:
+		if (ellipsis)
+			return "signed";
+		else
+			return "DW_ATE_signed";
+	case DW_ATE_signed_char:
+		if (ellipsis)
+			return "signed_char";
+		else
+			return "DW_ATE_signed_char";
+	case DW_ATE_unsigned:
+		if (ellipsis)
+			return "unsigned";
+		else
+			return "DW_ATE_unsigned";
+	case DW_ATE_unsigned_char:
+		if (ellipsis)
+			return "unsigned_char";
+		else
+			return "DW_ATE_unsigned_char";
+	case DW_ATE_imaginary_float:
+		if (ellipsis)
+			return "imaginary_float";
+		else
+			return "DW_ATE_imaginary_float";
+	case DW_ATE_packed_decimal:
+		if (ellipsis)
+			return "packed_decimal";
+		else
+			return "DW_ATE_packed_decimal";
+	case DW_ATE_numeric_string:
+		if (ellipsis)
+			return "numeric_string";
+		else
+			return "DW_ATE_numeric_string";
+	case DW_ATE_edited:
+		if (ellipsis)
+			return "edited";
+		else
+			return "DW_ATE_edited";
+	case DW_ATE_signed_fixed:
+		if (ellipsis)
+			return "signed_fixed";
+		else
+			return "DW_ATE_signed_fixed";
+	case DW_ATE_unsigned_fixed:
+		if (ellipsis)
+			return "unsigned_fixed";
+		else
+			return "DW_ATE_unsigned_fixed";
+	case DW_ATE_decimal_float:
+		if (ellipsis)
+			return "decimal_float";
+		else
+			return "DW_ATE_decimal_float";
+	case DW_ATE_ALTIUM_fract:
+		if (ellipsis)
+			return "ALTIUM_fract";
+		else
+			return "DW_ATE_ALTIUM_fract";
+	case DW_ATE_ALTIUM_accum:
+		if (ellipsis)
+			return "ALTIUM_accum";
+		else
+			return "DW_ATE_ALTIUM_accum";
+	case DW_ATE_HP_float128:
+		if (ellipsis)
+			return "HP_float128";
+		else
+			return "DW_ATE_HP_float128";
+	case DW_ATE_HP_complex_float128:
+		if (ellipsis)
+			return "HP_complex_float128";
+		else
+			return "DW_ATE_HP_complex_float128";
+	case DW_ATE_HP_floathpintel:
+		if (ellipsis)
+			return "HP_floathpintel";
+		else
+			return "DW_ATE_HP_floathpintel";
+	case DW_ATE_HP_imaginary_float80:
+		if (ellipsis)
+			return "HP_imaginary_float80";
+		else
+			return "DW_ATE_HP_imaginary_float80";
+	case DW_ATE_HP_imaginary_float128:
+		if (ellipsis)
+			return "HP_imaginary_float128";
+		else
+			return "DW_ATE_HP_imaginary_float128";
+	case DW_ATE_SUN_interval_float:
+		if (ellipsis)
+			return "SUN_interval_float";
+		else
+			return "DW_ATE_SUN_interval_float";
+	case DW_ATE_SUN_imaginary_float:
+		if (ellipsis)
+			return "SUN_imaginary_float";
+		else
+			return "DW_ATE_SUN_imaginary_float";
+	case DW_ATE_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_ATE_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ATE value 0x%x>",(int)val);
+		 fprintf(stderr,"ATE of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_DS_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_DS_unsigned:
+		if (ellipsis)
+			return "unsigned";
+		else
+			return "DW_DS_unsigned";
+	case DW_DS_leading_overpunch:
+		if (ellipsis)
+			return "leading_overpunch";
+		else
+			return "DW_DS_leading_overpunch";
+	case DW_DS_trailing_overpunch:
+		if (ellipsis)
+			return "trailing_overpunch";
+		else
+			return "DW_DS_trailing_overpunch";
+	case DW_DS_leading_separate:
+		if (ellipsis)
+			return "leading_separate";
+		else
+			return "DW_DS_leading_separate";
+	case DW_DS_trailing_separate:
+		if (ellipsis)
+			return "trailing_separate";
+		else
+			return "DW_DS_trailing_separate";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown DS value 0x%x>",(int)val);
+		 fprintf(stderr,"DS of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_END_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_END_default:
+		if (ellipsis)
+			return "default";
+		else
+			return "DW_END_default";
+	case DW_END_big:
+		if (ellipsis)
+			return "big";
+		else
+			return "DW_END_big";
+	case DW_END_little:
+		if (ellipsis)
+			return "little";
+		else
+			return "DW_END_little";
+	case DW_END_lo_user:
+		if (ellipsis)
+			return "lo_user";
+		else
+			return "DW_END_lo_user";
+	case DW_END_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_END_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown END value 0x%x>",(int)val);
+		 fprintf(stderr,"END of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ATCF_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ATCF_lo_user:
+		if (ellipsis)
+			return "lo_user";
+		else
+			return "DW_ATCF_lo_user";
+	case DW_ATCF_SUN_mop_bitfield:
+		if (ellipsis)
+			return "SUN_mop_bitfield";
+		else
+			return "DW_ATCF_SUN_mop_bitfield";
+	case DW_ATCF_SUN_mop_spill:
+		if (ellipsis)
+			return "SUN_mop_spill";
+		else
+			return "DW_ATCF_SUN_mop_spill";
+	case DW_ATCF_SUN_mop_scopy:
+		if (ellipsis)
+			return "SUN_mop_scopy";
+		else
+			return "DW_ATCF_SUN_mop_scopy";
+	case DW_ATCF_SUN_func_start:
+		if (ellipsis)
+			return "SUN_func_start";
+		else
+			return "DW_ATCF_SUN_func_start";
+	case DW_ATCF_SUN_end_ctors:
+		if (ellipsis)
+			return "SUN_end_ctors";
+		else
+			return "DW_ATCF_SUN_end_ctors";
+	case DW_ATCF_SUN_branch_target:
+		if (ellipsis)
+			return "SUN_branch_target";
+		else
+			return "DW_ATCF_SUN_branch_target";
+	case DW_ATCF_SUN_mop_stack_probe:
+		if (ellipsis)
+			return "SUN_mop_stack_probe";
+		else
+			return "DW_ATCF_SUN_mop_stack_probe";
+	case DW_ATCF_SUN_func_epilog:
+		if (ellipsis)
+			return "SUN_func_epilog";
+		else
+			return "DW_ATCF_SUN_func_epilog";
+	case DW_ATCF_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_ATCF_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ATCF value 0x%x>",(int)val);
+		 fprintf(stderr,"ATCF of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ACCESS_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ACCESS_public:
+		if (ellipsis)
+			return "public";
+		else
+			return "DW_ACCESS_public";
+	case DW_ACCESS_protected:
+		if (ellipsis)
+			return "protected";
+		else
+			return "DW_ACCESS_protected";
+	case DW_ACCESS_private:
+		if (ellipsis)
+			return "private";
+		else
+			return "DW_ACCESS_private";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ACCESS value 0x%x>",(int)val);
+		 fprintf(stderr,"ACCESS of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_VIS_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_VIS_local:
+		if (ellipsis)
+			return "local";
+		else
+			return "DW_VIS_local";
+	case DW_VIS_exported:
+		if (ellipsis)
+			return "exported";
+		else
+			return "DW_VIS_exported";
+	case DW_VIS_qualified:
+		if (ellipsis)
+			return "qualified";
+		else
+			return "DW_VIS_qualified";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown VIS value 0x%x>",(int)val);
+		 fprintf(stderr,"VIS of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_VIRTUALITY_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_VIRTUALITY_none:
+		if (ellipsis)
+			return "none";
+		else
+			return "DW_VIRTUALITY_none";
+	case DW_VIRTUALITY_virtual:
+		if (ellipsis)
+			return "virtual";
+		else
+			return "DW_VIRTUALITY_virtual";
+	case DW_VIRTUALITY_pure_virtual:
+		if (ellipsis)
+			return "pure_virtual";
+		else
+			return "DW_VIRTUALITY_pure_virtual";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown VIRTUALITY value 0x%x>",(int)val);
+		 fprintf(stderr,"VIRTUALITY of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_LANG_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_LANG_C89:
+		if (ellipsis)
+			return "C89";
+		else
+			return "DW_LANG_C89";
+	case DW_LANG_C:
+		if (ellipsis)
+			return "C";
+		else
+			return "DW_LANG_C";
+	case DW_LANG_Ada83:
+		if (ellipsis)
+			return "Ada83";
+		else
+			return "DW_LANG_Ada83";
+	case DW_LANG_C_plus_plus:
+		if (ellipsis)
+			return "C_plus_plus";
+		else
+			return "DW_LANG_C_plus_plus";
+	case DW_LANG_Cobol74:
+		if (ellipsis)
+			return "Cobol74";
+		else
+			return "DW_LANG_Cobol74";
+	case DW_LANG_Cobol85:
+		if (ellipsis)
+			return "Cobol85";
+		else
+			return "DW_LANG_Cobol85";
+	case DW_LANG_Fortran77:
+		if (ellipsis)
+			return "Fortran77";
+		else
+			return "DW_LANG_Fortran77";
+	case DW_LANG_Fortran90:
+		if (ellipsis)
+			return "Fortran90";
+		else
+			return "DW_LANG_Fortran90";
+	case DW_LANG_Pascal83:
+		if (ellipsis)
+			return "Pascal83";
+		else
+			return "DW_LANG_Pascal83";
+	case DW_LANG_Modula2:
+		if (ellipsis)
+			return "Modula2";
+		else
+			return "DW_LANG_Modula2";
+	case DW_LANG_Java:
+		if (ellipsis)
+			return "Java";
+		else
+			return "DW_LANG_Java";
+	case DW_LANG_C99:
+		if (ellipsis)
+			return "C99";
+		else
+			return "DW_LANG_C99";
+	case DW_LANG_Ada95:
+		if (ellipsis)
+			return "Ada95";
+		else
+			return "DW_LANG_Ada95";
+	case DW_LANG_Fortran95:
+		if (ellipsis)
+			return "Fortran95";
+		else
+			return "DW_LANG_Fortran95";
+	case DW_LANG_PLI:
+		if (ellipsis)
+			return "PLI";
+		else
+			return "DW_LANG_PLI";
+	case DW_LANG_ObjC:
+		if (ellipsis)
+			return "ObjC";
+		else
+			return "DW_LANG_ObjC";
+	case DW_LANG_ObjC_plus_plus:
+		if (ellipsis)
+			return "ObjC_plus_plus";
+		else
+			return "DW_LANG_ObjC_plus_plus";
+	case DW_LANG_UPC:
+		if (ellipsis)
+			return "UPC";
+		else
+			return "DW_LANG_UPC";
+	case DW_LANG_D:
+		if (ellipsis)
+			return "D";
+		else
+			return "DW_LANG_D";
+	case DW_LANG_Python:
+		if (ellipsis)
+			return "Python";
+		else
+			return "DW_LANG_Python";
+	case DW_LANG_OpenCL:
+		if (ellipsis)
+			return "OpenCL";
+		else
+			return "DW_LANG_OpenCL";
+	case DW_LANG_Go:
+		if (ellipsis)
+			return "Go";
+		else
+			return "DW_LANG_Go";
+	case DW_LANG_Modula3:
+		if (ellipsis)
+			return "Modula3";
+		else
+			return "DW_LANG_Modula3";
+	case DW_LANG_Haskel:
+		if (ellipsis)
+			return "Haskel";
+		else
+			return "DW_LANG_Haskel";
+	case DW_LANG_lo_user:
+		if (ellipsis)
+			return "lo_user";
+		else
+			return "DW_LANG_lo_user";
+	case DW_LANG_Mips_Assembler:
+		if (ellipsis)
+			return "Mips_Assembler";
+		else
+			return "DW_LANG_Mips_Assembler";
+	case DW_LANG_Upc:
+		if (ellipsis)
+			return "Upc";
+		else
+			return "DW_LANG_Upc";
+	case DW_LANG_ALTIUM_Assembler:
+		if (ellipsis)
+			return "ALTIUM_Assembler";
+		else
+			return "DW_LANG_ALTIUM_Assembler";
+	case DW_LANG_SUN_Assembler:
+		if (ellipsis)
+			return "SUN_Assembler";
+		else
+			return "DW_LANG_SUN_Assembler";
+	case DW_LANG_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_LANG_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown LANG value 0x%x>",(int)val);
+		 fprintf(stderr,"LANG of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ID_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ID_case_sensitive:
+		if (ellipsis)
+			return "case_sensitive";
+		else
+			return "DW_ID_case_sensitive";
+	case DW_ID_up_case:
+		if (ellipsis)
+			return "up_case";
+		else
+			return "DW_ID_up_case";
+	case DW_ID_down_case:
+		if (ellipsis)
+			return "down_case";
+		else
+			return "DW_ID_down_case";
+	case DW_ID_case_insensitive:
+		if (ellipsis)
+			return "case_insensitive";
+		else
+			return "DW_ID_case_insensitive";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ID value 0x%x>",(int)val);
+		 fprintf(stderr,"ID of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_CC_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_CC_normal:
+		if (ellipsis)
+			return "normal";
+		else
+			return "DW_CC_normal";
+	case DW_CC_program:
+		if (ellipsis)
+			return "program";
+		else
+			return "DW_CC_program";
+	case DW_CC_nocall:
+		if (ellipsis)
+			return "nocall";
+		else
+			return "DW_CC_nocall";
+	case DW_CC_lo_user:
+		if (ellipsis)
+			return "lo_user";
+		else
+			return "DW_CC_lo_user";
+	case DW_CC_GNU_borland_fastcall_i386:
+		if (ellipsis)
+			return "GNU_borland_fastcall_i386";
+		else
+			return "DW_CC_GNU_borland_fastcall_i386";
+	case DW_CC_ALTIUM_interrupt:
+		if (ellipsis)
+			return "ALTIUM_interrupt";
+		else
+			return "DW_CC_ALTIUM_interrupt";
+	case DW_CC_ALTIUM_near_system_stack:
+		if (ellipsis)
+			return "ALTIUM_near_system_stack";
+		else
+			return "DW_CC_ALTIUM_near_system_stack";
+	case DW_CC_ALTIUM_near_user_stack:
+		if (ellipsis)
+			return "ALTIUM_near_user_stack";
+		else
+			return "DW_CC_ALTIUM_near_user_stack";
+	case DW_CC_ALTIUM_huge_user_stack:
+		if (ellipsis)
+			return "ALTIUM_huge_user_stack";
+		else
+			return "DW_CC_ALTIUM_huge_user_stack";
+	case DW_CC_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_CC_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown CC value 0x%x>",(int)val);
+		 fprintf(stderr,"CC of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_INL_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_INL_not_inlined:
+		if (ellipsis)
+			return "not_inlined";
+		else
+			return "DW_INL_not_inlined";
+	case DW_INL_inlined:
+		if (ellipsis)
+			return "inlined";
+		else
+			return "DW_INL_inlined";
+	case DW_INL_declared_not_inlined:
+		if (ellipsis)
+			return "declared_not_inlined";
+		else
+			return "DW_INL_declared_not_inlined";
+	case DW_INL_declared_inlined:
+		if (ellipsis)
+			return "declared_inlined";
+		else
+			return "DW_INL_declared_inlined";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown INL value 0x%x>",(int)val);
+		 fprintf(stderr,"INL of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ORD_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ORD_row_major:
+		if (ellipsis)
+			return "row_major";
+		else
+			return "DW_ORD_row_major";
+	case DW_ORD_col_major:
+		if (ellipsis)
+			return "col_major";
+		else
+			return "DW_ORD_col_major";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ORD value 0x%x>",(int)val);
+		 fprintf(stderr,"ORD of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_DSC_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_DSC_label:
+		if (ellipsis)
+			return "label";
+		else
+			return "DW_DSC_label";
+	case DW_DSC_range:
+		if (ellipsis)
+			return "range";
+		else
+			return "DW_DSC_range";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown DSC value 0x%x>",(int)val);
+		 fprintf(stderr,"DSC of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_LNS_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_LNS_copy:
+		if (ellipsis)
+			return "copy";
+		else
+			return "DW_LNS_copy";
+	case DW_LNS_advance_pc:
+		if (ellipsis)
+			return "advance_pc";
+		else
+			return "DW_LNS_advance_pc";
+	case DW_LNS_advance_line:
+		if (ellipsis)
+			return "advance_line";
+		else
+			return "DW_LNS_advance_line";
+	case DW_LNS_set_file:
+		if (ellipsis)
+			return "set_file";
+		else
+			return "DW_LNS_set_file";
+	case DW_LNS_set_column:
+		if (ellipsis)
+			return "set_column";
+		else
+			return "DW_LNS_set_column";
+	case DW_LNS_negate_stmt:
+		if (ellipsis)
+			return "negate_stmt";
+		else
+			return "DW_LNS_negate_stmt";
+	case DW_LNS_set_basic_block:
+		if (ellipsis)
+			return "set_basic_block";
+		else
+			return "DW_LNS_set_basic_block";
+	case DW_LNS_const_add_pc:
+		if (ellipsis)
+			return "const_add_pc";
+		else
+			return "DW_LNS_const_add_pc";
+	case DW_LNS_fixed_advance_pc:
+		if (ellipsis)
+			return "fixed_advance_pc";
+		else
+			return "DW_LNS_fixed_advance_pc";
+	case DW_LNS_set_prologue_end:
+		if (ellipsis)
+			return "set_prologue_end";
+		else
+			return "DW_LNS_set_prologue_end";
+	case DW_LNS_set_epilogue_begin:
+		if (ellipsis)
+			return "set_epilogue_begin";
+		else
+			return "DW_LNS_set_epilogue_begin";
+	case DW_LNS_set_isa:
+		if (ellipsis)
+			return "set_isa";
+		else
+			return "DW_LNS_set_isa";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown LNS value 0x%x>",(int)val);
+		 fprintf(stderr,"LNS of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_LNE_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_LNE_end_sequence:
+		if (ellipsis)
+			return "end_sequence";
+		else
+			return "DW_LNE_end_sequence";
+	case DW_LNE_set_address:
+		if (ellipsis)
+			return "set_address";
+		else
+			return "DW_LNE_set_address";
+	case DW_LNE_define_file:
+		if (ellipsis)
+			return "define_file";
+		else
+			return "DW_LNE_define_file";
+	case DW_LNE_set_discriminator:
+		if (ellipsis)
+			return "set_discriminator";
+		else
+			return "DW_LNE_set_discriminator";
+	case DW_LNE_HP_negate_is_UV_update:
+		if (ellipsis)
+			return "HP_negate_is_UV_update";
+		else
+			return "DW_LNE_HP_negate_is_UV_update";
+	case DW_LNE_HP_push_context:
+		if (ellipsis)
+			return "HP_push_context";
+		else
+			return "DW_LNE_HP_push_context";
+	case DW_LNE_HP_pop_context:
+		if (ellipsis)
+			return "HP_pop_context";
+		else
+			return "DW_LNE_HP_pop_context";
+	case DW_LNE_HP_set_file_line_column:
+		if (ellipsis)
+			return "HP_set_file_line_column";
+		else
+			return "DW_LNE_HP_set_file_line_column";
+	case DW_LNE_HP_set_routine_name:
+		if (ellipsis)
+			return "HP_set_routine_name";
+		else
+			return "DW_LNE_HP_set_routine_name";
+	case DW_LNE_HP_set_sequence:
+		if (ellipsis)
+			return "HP_set_sequence";
+		else
+			return "DW_LNE_HP_set_sequence";
+	case DW_LNE_HP_negate_post_semantics:
+		if (ellipsis)
+			return "HP_negate_post_semantics";
+		else
+			return "DW_LNE_HP_negate_post_semantics";
+	case DW_LNE_HP_negate_function_exit:
+		if (ellipsis)
+			return "HP_negate_function_exit";
+		else
+			return "DW_LNE_HP_negate_function_exit";
+	case DW_LNE_HP_negate_front_end_logical:
+		if (ellipsis)
+			return "HP_negate_front_end_logical";
+		else
+			return "DW_LNE_HP_negate_front_end_logical";
+	case DW_LNE_HP_define_proc:
+		if (ellipsis)
+			return "HP_define_proc";
+		else
+			return "DW_LNE_HP_define_proc";
+	case DW_LNE_HP_source_file_correlation:
+		if (ellipsis)
+			return "HP_source_file_correlation";
+		else
+			return "DW_LNE_HP_source_file_correlation";
+	case DW_LNE_hi_user:
+		if (ellipsis)
+			return "hi_user";
+		else
+			return "DW_LNE_hi_user";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown LNE value 0x%x>",(int)val);
+		 fprintf(stderr,"LNE of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ISA_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ISA_UNKNOWN:
+		if (ellipsis)
+			return "UNKNOWN";
+		else
+			return "DW_ISA_UNKNOWN";
+	case DW_ISA_ARM_thumb:
+		if (ellipsis)
+			return "ARM_thumb";
+		else
+			return "DW_ISA_ARM_thumb";
+	case DW_ISA_ARM_arm:
+		if (ellipsis)
+			return "ARM_arm";
+		else
+			return "DW_ISA_ARM_arm";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ISA value 0x%x>",(int)val);
+		 fprintf(stderr,"ISA of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_MACINFO_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_MACINFO_define:
+		if (ellipsis)
+			return "define";
+		else
+			return "DW_MACINFO_define";
+	case DW_MACINFO_undef:
+		if (ellipsis)
+			return "undef";
+		else
+			return "DW_MACINFO_undef";
+	case DW_MACINFO_start_file:
+		if (ellipsis)
+			return "start_file";
+		else
+			return "DW_MACINFO_start_file";
+	case DW_MACINFO_end_file:
+		if (ellipsis)
+			return "end_file";
+		else
+			return "DW_MACINFO_end_file";
+	case DW_MACINFO_vendor_ext:
+		if (ellipsis)
+			return "vendor_ext";
+		else
+			return "DW_MACINFO_vendor_ext";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown MACINFO value 0x%x>",(int)val);
+		 fprintf(stderr,"MACINFO of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_EH_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_EH_PE_absptr:
+		if (ellipsis)
+			return "PE_absptr";
+		else
+			return "DW_EH_PE_absptr";
+	case DW_EH_PE_uleb128:
+		if (ellipsis)
+			return "PE_uleb128";
+		else
+			return "DW_EH_PE_uleb128";
+	case DW_EH_PE_udata2:
+		if (ellipsis)
+			return "PE_udata2";
+		else
+			return "DW_EH_PE_udata2";
+	case DW_EH_PE_udata4:
+		if (ellipsis)
+			return "PE_udata4";
+		else
+			return "DW_EH_PE_udata4";
+	case DW_EH_PE_udata8:
+		if (ellipsis)
+			return "PE_udata8";
+		else
+			return "DW_EH_PE_udata8";
+	case DW_EH_PE_sleb128:
+		if (ellipsis)
+			return "PE_sleb128";
+		else
+			return "DW_EH_PE_sleb128";
+	case DW_EH_PE_sdata2:
+		if (ellipsis)
+			return "PE_sdata2";
+		else
+			return "DW_EH_PE_sdata2";
+	case DW_EH_PE_sdata4:
+		if (ellipsis)
+			return "PE_sdata4";
+		else
+			return "DW_EH_PE_sdata4";
+	case DW_EH_PE_sdata8:
+		if (ellipsis)
+			return "PE_sdata8";
+		else
+			return "DW_EH_PE_sdata8";
+	case DW_EH_PE_pcrel:
+		if (ellipsis)
+			return "PE_pcrel";
+		else
+			return "DW_EH_PE_pcrel";
+	case DW_EH_PE_textrel:
+		if (ellipsis)
+			return "PE_textrel";
+		else
+			return "DW_EH_PE_textrel";
+	case DW_EH_PE_datarel:
+		if (ellipsis)
+			return "PE_datarel";
+		else
+			return "DW_EH_PE_datarel";
+	case DW_EH_PE_funcrel:
+		if (ellipsis)
+			return "PE_funcrel";
+		else
+			return "DW_EH_PE_funcrel";
+	case DW_EH_PE_aligned:
+		if (ellipsis)
+			return "PE_aligned";
+		else
+			return "DW_EH_PE_aligned";
+	case DW_EH_PE_omit:
+		if (ellipsis)
+			return "PE_omit";
+		else
+			return "DW_EH_PE_omit";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown EH value 0x%x>",(int)val);
+		 fprintf(stderr,"EH of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_FRAME_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_FRAME_CFA_COL:
+		if (ellipsis)
+			return "CFA_COL";
+		else
+			return "DW_FRAME_CFA_COL";
+	case DW_FRAME_REG1:
+		if (ellipsis)
+			return "REG1";
+		else
+			return "DW_FRAME_REG1";
+	case DW_FRAME_REG2:
+		if (ellipsis)
+			return "REG2";
+		else
+			return "DW_FRAME_REG2";
+	case DW_FRAME_REG3:
+		if (ellipsis)
+			return "REG3";
+		else
+			return "DW_FRAME_REG3";
+	case DW_FRAME_REG4:
+		if (ellipsis)
+			return "REG4";
+		else
+			return "DW_FRAME_REG4";
+	case DW_FRAME_REG5:
+		if (ellipsis)
+			return "REG5";
+		else
+			return "DW_FRAME_REG5";
+	case DW_FRAME_REG6:
+		if (ellipsis)
+			return "REG6";
+		else
+			return "DW_FRAME_REG6";
+	case DW_FRAME_REG7:
+		if (ellipsis)
+			return "REG7";
+		else
+			return "DW_FRAME_REG7";
+	case DW_FRAME_REG8:
+		if (ellipsis)
+			return "REG8";
+		else
+			return "DW_FRAME_REG8";
+	case DW_FRAME_REG9:
+		if (ellipsis)
+			return "REG9";
+		else
+			return "DW_FRAME_REG9";
+	case DW_FRAME_REG10:
+		if (ellipsis)
+			return "REG10";
+		else
+			return "DW_FRAME_REG10";
+	case DW_FRAME_REG11:
+		if (ellipsis)
+			return "REG11";
+		else
+			return "DW_FRAME_REG11";
+	case DW_FRAME_REG12:
+		if (ellipsis)
+			return "REG12";
+		else
+			return "DW_FRAME_REG12";
+	case DW_FRAME_REG13:
+		if (ellipsis)
+			return "REG13";
+		else
+			return "DW_FRAME_REG13";
+	case DW_FRAME_REG14:
+		if (ellipsis)
+			return "REG14";
+		else
+			return "DW_FRAME_REG14";
+	case DW_FRAME_REG15:
+		if (ellipsis)
+			return "REG15";
+		else
+			return "DW_FRAME_REG15";
+	case DW_FRAME_REG16:
+		if (ellipsis)
+			return "REG16";
+		else
+			return "DW_FRAME_REG16";
+	case DW_FRAME_REG17:
+		if (ellipsis)
+			return "REG17";
+		else
+			return "DW_FRAME_REG17";
+	case DW_FRAME_REG18:
+		if (ellipsis)
+			return "REG18";
+		else
+			return "DW_FRAME_REG18";
+	case DW_FRAME_REG19:
+		if (ellipsis)
+			return "REG19";
+		else
+			return "DW_FRAME_REG19";
+	case DW_FRAME_REG20:
+		if (ellipsis)
+			return "REG20";
+		else
+			return "DW_FRAME_REG20";
+	case DW_FRAME_REG21:
+		if (ellipsis)
+			return "REG21";
+		else
+			return "DW_FRAME_REG21";
+	case DW_FRAME_REG22:
+		if (ellipsis)
+			return "REG22";
+		else
+			return "DW_FRAME_REG22";
+	case DW_FRAME_REG23:
+		if (ellipsis)
+			return "REG23";
+		else
+			return "DW_FRAME_REG23";
+	case DW_FRAME_REG24:
+		if (ellipsis)
+			return "REG24";
+		else
+			return "DW_FRAME_REG24";
+	case DW_FRAME_REG25:
+		if (ellipsis)
+			return "REG25";
+		else
+			return "DW_FRAME_REG25";
+	case DW_FRAME_REG26:
+		if (ellipsis)
+			return "REG26";
+		else
+			return "DW_FRAME_REG26";
+	case DW_FRAME_REG27:
+		if (ellipsis)
+			return "REG27";
+		else
+			return "DW_FRAME_REG27";
+	case DW_FRAME_REG28:
+		if (ellipsis)
+			return "REG28";
+		else
+			return "DW_FRAME_REG28";
+	case DW_FRAME_REG29:
+		if (ellipsis)
+			return "REG29";
+		else
+			return "DW_FRAME_REG29";
+	case DW_FRAME_REG30:
+		if (ellipsis)
+			return "REG30";
+		else
+			return "DW_FRAME_REG30";
+	case DW_FRAME_REG31:
+		if (ellipsis)
+			return "REG31";
+		else
+			return "DW_FRAME_REG31";
+	case DW_FRAME_FREG0:
+		if (ellipsis)
+			return "FREG0";
+		else
+			return "DW_FRAME_FREG0";
+	case DW_FRAME_FREG1:
+		if (ellipsis)
+			return "FREG1";
+		else
+			return "DW_FRAME_FREG1";
+	case DW_FRAME_FREG2:
+		if (ellipsis)
+			return "FREG2";
+		else
+			return "DW_FRAME_FREG2";
+	case DW_FRAME_FREG3:
+		if (ellipsis)
+			return "FREG3";
+		else
+			return "DW_FRAME_FREG3";
+	case DW_FRAME_FREG4:
+		if (ellipsis)
+			return "FREG4";
+		else
+			return "DW_FRAME_FREG4";
+	case DW_FRAME_FREG5:
+		if (ellipsis)
+			return "FREG5";
+		else
+			return "DW_FRAME_FREG5";
+	case DW_FRAME_FREG6:
+		if (ellipsis)
+			return "FREG6";
+		else
+			return "DW_FRAME_FREG6";
+	case DW_FRAME_FREG7:
+		if (ellipsis)
+			return "FREG7";
+		else
+			return "DW_FRAME_FREG7";
+	case DW_FRAME_FREG8:
+		if (ellipsis)
+			return "FREG8";
+		else
+			return "DW_FRAME_FREG8";
+	case DW_FRAME_FREG9:
+		if (ellipsis)
+			return "FREG9";
+		else
+			return "DW_FRAME_FREG9";
+	case DW_FRAME_FREG10:
+		if (ellipsis)
+			return "FREG10";
+		else
+			return "DW_FRAME_FREG10";
+	case DW_FRAME_FREG11:
+		if (ellipsis)
+			return "FREG11";
+		else
+			return "DW_FRAME_FREG11";
+	case DW_FRAME_FREG12:
+		if (ellipsis)
+			return "FREG12";
+		else
+			return "DW_FRAME_FREG12";
+	case DW_FRAME_FREG13:
+		if (ellipsis)
+			return "FREG13";
+		else
+			return "DW_FRAME_FREG13";
+	case DW_FRAME_FREG14:
+		if (ellipsis)
+			return "FREG14";
+		else
+			return "DW_FRAME_FREG14";
+	case DW_FRAME_FREG15:
+		if (ellipsis)
+			return "FREG15";
+		else
+			return "DW_FRAME_FREG15";
+	case DW_FRAME_FREG16:
+		if (ellipsis)
+			return "FREG16";
+		else
+			return "DW_FRAME_FREG16";
+	case DW_FRAME_FREG17:
+		if (ellipsis)
+			return "FREG17";
+		else
+			return "DW_FRAME_FREG17";
+	case DW_FRAME_FREG18:
+		if (ellipsis)
+			return "FREG18";
+		else
+			return "DW_FRAME_FREG18";
+	case DW_FRAME_FREG19:
+		if (ellipsis)
+			return "FREG19";
+		else
+			return "DW_FRAME_FREG19";
+	case DW_FRAME_FREG20:
+		if (ellipsis)
+			return "FREG20";
+		else
+			return "DW_FRAME_FREG20";
+	case DW_FRAME_FREG21:
+		if (ellipsis)
+			return "FREG21";
+		else
+			return "DW_FRAME_FREG21";
+	case DW_FRAME_FREG22:
+		if (ellipsis)
+			return "FREG22";
+		else
+			return "DW_FRAME_FREG22";
+	case DW_FRAME_FREG23:
+		if (ellipsis)
+			return "FREG23";
+		else
+			return "DW_FRAME_FREG23";
+	case DW_FRAME_FREG24:
+		if (ellipsis)
+			return "FREG24";
+		else
+			return "DW_FRAME_FREG24";
+	case DW_FRAME_FREG25:
+		if (ellipsis)
+			return "FREG25";
+		else
+			return "DW_FRAME_FREG25";
+	case DW_FRAME_FREG26:
+		if (ellipsis)
+			return "FREG26";
+		else
+			return "DW_FRAME_FREG26";
+	case DW_FRAME_FREG27:
+		if (ellipsis)
+			return "FREG27";
+		else
+			return "DW_FRAME_FREG27";
+	case DW_FRAME_FREG28:
+		if (ellipsis)
+			return "FREG28";
+		else
+			return "DW_FRAME_FREG28";
+	case DW_FRAME_FREG29:
+		if (ellipsis)
+			return "FREG29";
+		else
+			return "DW_FRAME_FREG29";
+	case DW_FRAME_FREG30:
+		if (ellipsis)
+			return "FREG30";
+		else
+			return "DW_FRAME_FREG30";
+	case DW_FRAME_FREG31:
+		if (ellipsis)
+			return "FREG31";
+		else
+			return "DW_FRAME_FREG31";
+	case DW_FRAME_FREG32:
+		if (ellipsis)
+			return "FREG32";
+		else
+			return "DW_FRAME_FREG32";
+	case DW_FRAME_FREG33:
+		if (ellipsis)
+			return "FREG33";
+		else
+			return "DW_FRAME_FREG33";
+	case DW_FRAME_FREG34:
+		if (ellipsis)
+			return "FREG34";
+		else
+			return "DW_FRAME_FREG34";
+	case DW_FRAME_FREG35:
+		if (ellipsis)
+			return "FREG35";
+		else
+			return "DW_FRAME_FREG35";
+	case DW_FRAME_FREG36:
+		if (ellipsis)
+			return "FREG36";
+		else
+			return "DW_FRAME_FREG36";
+	case DW_FRAME_FREG37:
+		if (ellipsis)
+			return "FREG37";
+		else
+			return "DW_FRAME_FREG37";
+	case DW_FRAME_FREG38:
+		if (ellipsis)
+			return "FREG38";
+		else
+			return "DW_FRAME_FREG38";
+	case DW_FRAME_FREG39:
+		if (ellipsis)
+			return "FREG39";
+		else
+			return "DW_FRAME_FREG39";
+	case DW_FRAME_FREG40:
+		if (ellipsis)
+			return "FREG40";
+		else
+			return "DW_FRAME_FREG40";
+	case DW_FRAME_FREG41:
+		if (ellipsis)
+			return "FREG41";
+		else
+			return "DW_FRAME_FREG41";
+	case DW_FRAME_FREG42:
+		if (ellipsis)
+			return "FREG42";
+		else
+			return "DW_FRAME_FREG42";
+	case DW_FRAME_FREG43:
+		if (ellipsis)
+			return "FREG43";
+		else
+			return "DW_FRAME_FREG43";
+	case DW_FRAME_FREG44:
+		if (ellipsis)
+			return "FREG44";
+		else
+			return "DW_FRAME_FREG44";
+	case DW_FRAME_FREG45:
+		if (ellipsis)
+			return "FREG45";
+		else
+			return "DW_FRAME_FREG45";
+	case DW_FRAME_FREG46:
+		if (ellipsis)
+			return "FREG46";
+		else
+			return "DW_FRAME_FREG46";
+	case DW_FRAME_FREG47:
+		if (ellipsis)
+			return "FREG47";
+		else
+			return "DW_FRAME_FREG47";
+	case DW_FRAME_FREG48:
+		if (ellipsis)
+			return "FREG48";
+		else
+			return "DW_FRAME_FREG48";
+	case DW_FRAME_FREG49:
+		if (ellipsis)
+			return "FREG49";
+		else
+			return "DW_FRAME_FREG49";
+	case DW_FRAME_FREG50:
+		if (ellipsis)
+			return "FREG50";
+		else
+			return "DW_FRAME_FREG50";
+	case DW_FRAME_FREG51:
+		if (ellipsis)
+			return "FREG51";
+		else
+			return "DW_FRAME_FREG51";
+	case DW_FRAME_FREG52:
+		if (ellipsis)
+			return "FREG52";
+		else
+			return "DW_FRAME_FREG52";
+	case DW_FRAME_FREG53:
+		if (ellipsis)
+			return "FREG53";
+		else
+			return "DW_FRAME_FREG53";
+	case DW_FRAME_FREG54:
+		if (ellipsis)
+			return "FREG54";
+		else
+			return "DW_FRAME_FREG54";
+	case DW_FRAME_FREG55:
+		if (ellipsis)
+			return "FREG55";
+		else
+			return "DW_FRAME_FREG55";
+	case DW_FRAME_FREG56:
+		if (ellipsis)
+			return "FREG56";
+		else
+			return "DW_FRAME_FREG56";
+	case DW_FRAME_FREG57:
+		if (ellipsis)
+			return "FREG57";
+		else
+			return "DW_FRAME_FREG57";
+	case DW_FRAME_FREG58:
+		if (ellipsis)
+			return "FREG58";
+		else
+			return "DW_FRAME_FREG58";
+	case DW_FRAME_FREG59:
+		if (ellipsis)
+			return "FREG59";
+		else
+			return "DW_FRAME_FREG59";
+	case DW_FRAME_FREG60:
+		if (ellipsis)
+			return "FREG60";
+		else
+			return "DW_FRAME_FREG60";
+	case DW_FRAME_FREG61:
+		if (ellipsis)
+			return "FREG61";
+		else
+			return "DW_FRAME_FREG61";
+	case DW_FRAME_FREG62:
+		if (ellipsis)
+			return "FREG62";
+		else
+			return "DW_FRAME_FREG62";
+	case DW_FRAME_FREG63:
+		if (ellipsis)
+			return "FREG63";
+		else
+			return "DW_FRAME_FREG63";
+	case DW_FRAME_FREG64:
+		if (ellipsis)
+			return "FREG64";
+		else
+			return "DW_FRAME_FREG64";
+	case DW_FRAME_FREG65:
+		if (ellipsis)
+			return "FREG65";
+		else
+			return "DW_FRAME_FREG65";
+	case DW_FRAME_FREG66:
+		if (ellipsis)
+			return "FREG66";
+		else
+			return "DW_FRAME_FREG66";
+	case DW_FRAME_FREG67:
+		if (ellipsis)
+			return "FREG67";
+		else
+			return "DW_FRAME_FREG67";
+	case DW_FRAME_FREG68:
+		if (ellipsis)
+			return "FREG68";
+		else
+			return "DW_FRAME_FREG68";
+	case DW_FRAME_FREG69:
+		if (ellipsis)
+			return "FREG69";
+		else
+			return "DW_FRAME_FREG69";
+	case DW_FRAME_FREG70:
+		if (ellipsis)
+			return "FREG70";
+		else
+			return "DW_FRAME_FREG70";
+	case DW_FRAME_FREG71:
+		if (ellipsis)
+			return "FREG71";
+		else
+			return "DW_FRAME_FREG71";
+	case DW_FRAME_FREG72:
+		if (ellipsis)
+			return "FREG72";
+		else
+			return "DW_FRAME_FREG72";
+	case DW_FRAME_FREG73:
+		if (ellipsis)
+			return "FREG73";
+		else
+			return "DW_FRAME_FREG73";
+	case DW_FRAME_FREG74:
+		if (ellipsis)
+			return "FREG74";
+		else
+			return "DW_FRAME_FREG74";
+	case DW_FRAME_FREG75:
+		if (ellipsis)
+			return "FREG75";
+		else
+			return "DW_FRAME_FREG75";
+	case DW_FRAME_FREG76:
+		if (ellipsis)
+			return "FREG76";
+		else
+			return "DW_FRAME_FREG76";
+	case DW_FRAME_HIGHEST_NORMAL_REGISTER:
+		if (ellipsis)
+			return "HIGHEST_NORMAL_REGISTER";
+		else
+			return "DW_FRAME_HIGHEST_NORMAL_REGISTER";
+	case DW_FRAME_LAST_REG_NUM:
+		if (ellipsis)
+			return "LAST_REG_NUM";
+		else
+			return "DW_FRAME_LAST_REG_NUM";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown FRAME value 0x%x>",(int)val);
+		 fprintf(stderr,"FRAME of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_CHILDREN_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_CHILDREN_no:
+		if (ellipsis)
+			return "CHILDREN_no";
+		else
+			return "DW_CHILDREN_no";
+	case DW_CHILDREN_yes:
+		if (ellipsis)
+			return "CHILDREN_yes";
+		else
+			return "DW_CHILDREN_yes";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown CHILDREN value 0x%x>",(int)val);
+		 fprintf(stderr,"CHILDREN of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
+/* ARGSUSED */
+extern string
+get_ADDR_name (Dwarf_Debug dbg, Dwarf_Half val)
+{
+	switch (val) {
+	case DW_ADDR_none:
+		if (ellipsis)
+			return "ADDR_none";
+		else
+			return "DW_ADDR_none";
+	default:
+		{ 
+		    char buf[100]; 
+		    char *n; 
+		    sprintf(buf,"<Unknown ADDR value 0x%x>",(int)val);
+		 fprintf(stderr,"ADDR of %d (0x%x) is unknown to dwarfdump. " 
+ 		 "Continuing. \n",(int)val,(int)val );  
+		    n = makename(buf);
+		    return n; 
+		} 
+	}
+/*NOTREACHED*/
+}
+
diff --git a/tools/kgraft/dwarf_names.h b/tools/kgraft/dwarf_names.h
new file mode 100644
index 000000000000..5a9ffaef9059
--- /dev/null
+++ b/tools/kgraft/dwarf_names.h
@@ -0,0 +1,53 @@
+/* automatically generated routines */
+extern string get_TAG_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_children_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_FORM_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_AT_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_OP_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ATE_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_DS_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_END_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ATCF_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ACCESS_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_VIS_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_VIRTUALITY_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_LANG_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ID_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_CC_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_INL_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ORD_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_DSC_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_LNS_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_LNE_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ISA_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_MACINFO_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_EH_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_FRAME_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_CHILDREN_name (Dwarf_Debug dbg, Dwarf_Half val);
+
+extern string get_ADDR_name (Dwarf_Debug dbg, Dwarf_Half val);
+
diff --git a/tools/kgraft/extract-syms.sh b/tools/kgraft/extract-syms.sh
new file mode 100755
index 000000000000..280b1829b25b
--- /dev/null
+++ b/tools/kgraft/extract-syms.sh
@@ -0,0 +1,18 @@
+#!/bin/bash
+TOOLPATH=`dirname $0`
+if ! test -f vmlinux.o; then
+    echo "vmlinux.o needs to exist in cwd"
+    exit 1
+fi
+if test -z "$1"; then
+    echo "usage: $0 [list of symbols to extract]"
+    exit 2
+fi
+
+rm -f symlist symlist.rename extracted.o
+for i in $@; do
+    echo $i >> symlist
+    echo $i new_$i >> symlist.rename
+done
+$TOOLPATH/objcopy-hacked --strip-unneeded -j .doesntexist. --keep-symbols symlist --redefine-syms symlist.rename vmlinux.o extracted.o
+nm extracted.o
diff --git a/tools/kgraft/it2rev.pl b/tools/kgraft/it2rev.pl
new file mode 100644
index 000000000000..3e0cf8a55138
--- /dev/null
+++ b/tools/kgraft/it2rev.pl
@@ -0,0 +1,40 @@
+#!/usr/bin/perl -w
+
+my $prefix='';
+foreach (@ARGV) {
+    if (/^-/) {
+    } else {
+	$prefix = $_;
+    }
+}
+my %files=();
+my $func='';
+my $file='';
+my $ffref=0;
+while(<STDIN>) {
+    chomp;
+    if (/^U/) {
+	#print "ignore $_\n";
+    } elsif (/^D (.*):([^: ]*)$/) {
+	#print "func $2 in $1\n";
+	$func=$2;
+	($file = $1) =~ s/^$prefix//;
+    } elsif (/^I (.*):([^: ]*)$/) {
+	#print "inline $2 in $1\n";
+	my $t = $1;
+	my $u = $2;
+	$t =~ s/^$prefix//;
+	$files{$t}->{$u}->{$file}->{$func} = 1;
+    }
+}
+foreach (sort keys %files) {
+    foreach my $inlinee (sort keys %{$files{$_}}) {
+	print "$_:$inlinee";
+	foreach my $ifile (sort keys %{$files{$_}->{$inlinee}}) {
+	    foreach my $ifunc (sort keys %{$files{$_}->{$inlinee}->{$ifile}}) {
+		print " $ifile:$ifunc";
+	    }
+	}
+	print "\n";
+    }
+}
diff --git a/tools/kgraft/objcopy.diff b/tools/kgraft/objcopy.diff
new file mode 100644
index 000000000000..53697612f696
--- /dev/null
+++ b/tools/kgraft/objcopy.diff
@@ -0,0 +1,131 @@
+diff --git a/binutils/objcopy.c b/binutils/objcopy.c
+index 14f6b96..a6d59de 100644
+--- a/binutils/objcopy.c
++++ b/binutils/objcopy.c
+@@ -1301,7 +1301,14 @@ filter_symbols (bfd *abfd, bfd *obfd, asymbol **osyms,
+ 	keep = TRUE;
+ 
+       if (keep && is_strip_section (abfd, bfd_get_section (sym)))
+-	keep = FALSE;
++	{
++	  if (relocatable && used_in_reloc)
++	    {
++	      sym->section = bfd_und_section_ptr;
++	    }
++	  else
++	    keep = FALSE;
++	}
+ 
+       if (keep)
+ 	{
+@@ -1564,6 +1571,72 @@ copy_unknown_object (bfd *ibfd, bfd *obfd)
+   return TRUE;
+ }
+ 
++static bfd_boolean traverse_reloc_changed;
++
++static void
++traverse_relocs (bfd *ibfd, sec_ptr isection, void *symbolsarg)
++{
++  asymbol **symbols = (asymbol **) symbolsarg;
++  long relsize;
++  arelent **relpp;
++  long relcount, i;
++
++  /* If we don't keep this section, don't look at it.  */
++  if (!find_section_list (bfd_get_section_name (ibfd, isection),
++			  FALSE, SECTION_CONTEXT_COPY))
++    return;
++
++  relsize = bfd_get_reloc_upper_bound (ibfd, isection);
++  if (relsize < 0)
++    {
++      /* Do not complain if the target does not support relocations.  */
++      if (relsize == -1 && bfd_get_error () == bfd_error_invalid_operation)
++	return;
++      bfd_fatal (bfd_get_filename (ibfd));
++    }
++
++  if (relsize == 0)
++    return;
++
++  relpp = (arelent **) xmalloc (relsize);
++  relcount = bfd_canonicalize_reloc (ibfd, isection, relpp, symbols);
++  if (relcount < 0)
++    bfd_fatal (bfd_get_filename (ibfd));
++
++  /* Examine each symbol used in a relocation.  If it's not one of the
++     special bfd section symbols, then mark it with BSF_KEEP.  */
++  for (i = 0; i < relcount; i++)
++    {
++      asymbol *sym = *relpp[i]->sym_ptr_ptr;
++      asection *sec;
++      if (sym != bfd_com_section_ptr->symbol
++	  && sym != bfd_abs_section_ptr->symbol
++	  && sym != bfd_und_section_ptr->symbol)
++	{
++	  if (sym->flags & BSF_KEEP)
++	    continue;
++	  sym->flags |= BSF_KEEP;
++	}
++      /* We need to copy sections defining stuff we need for section-based
++         relocs.  For the others we can just emit undef symbols.  */
++      if (!(sym->flags & BSF_SECTION_SYM))
++	continue;
++      sec = bfd_get_section (sym);
++      if (find_section_list (bfd_get_section_name (ibfd, sec),
++			     FALSE, SECTION_CONTEXT_COPY))
++	continue;
++      printf ("copying section %s because of symbol %s (reloc from %s).\n",
++	      bfd_get_section_name (ibfd, sec), bfd_asymbol_name (sym),
++	      bfd_get_section_name (ibfd, isection));
++      find_section_list (bfd_get_section_name (ibfd, sec), TRUE,
++			 SECTION_CONTEXT_COPY);
++      traverse_reloc_changed = TRUE;
++    }
++
++  if (relpp != NULL)
++    free (relpp);
++}
++
+ /* Copy object file IBFD onto OBFD.
+    Returns TRUE upon success, FALSE otherwise.  */
+ 
+@@ -1746,6 +1819,37 @@ copy_object (bfd *ibfd, bfd *obfd, const bfd_arch_info_type *input_arch)
+       return FALSE;
+     }
+ 
++  if (1)
++    {
++      long i;
++      for (i = 0; i < symcount; i++)
++	{
++	  asymbol *sym = isympp[i];
++	  asection *sec;
++	  char *name = (char *) bfd_asymbol_name (sym);
++	  bfd_boolean undefined;
++
++	  sec = bfd_get_section (sym);
++	  undefined = bfd_is_und_section (sec);
++	  if (!undefined
++	      && is_specified_symbol (name, keep_specific_htab))
++	    {
++	      find_section_list (bfd_get_section_name (ibfd, sec),
++				 TRUE, SECTION_CONTEXT_COPY);
++	      sections_copied = TRUE;
++	      traverse_reloc_changed = TRUE;
++	      printf ("copying section %s because of symbol %s.\n",
++		      bfd_get_section_name (ibfd, sec), name);
++	    }
++	}
++      /* Now mark all sections copied that are referred to from
++         relocations of sections that are already copied, transitively.  */
++      while (traverse_reloc_changed)
++	{
++	  traverse_reloc_changed = FALSE;
++	  bfd_map_over_sections (ibfd, traverse_relocs, isympp);
++	}
++    }
+   /* BFD mandates that all output sections be created and sizes set before
+      any output is done.  Thus, we traverse all sections multiple times.  */
+   bfd_map_over_sections (ibfd, setup_section, obfd);
diff --git a/tools/kgraft/symlist b/tools/kgraft/symlist
new file mode 100644
index 000000000000..79a58de6e1ca
--- /dev/null
+++ b/tools/kgraft/symlist
@@ -0,0 +1 @@
+in_app
-- 
1.9.2

Re: [RFC 12/16] kgr: add tools

[Pavel Machek]

On Wed 2014-04-30 16:30:45, Jiri Slaby wrote:
> These are a base which can be used for kgraft patch generation.
> 
> The code was provided by Michael

Should Michael Matz sign it off, then?

> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Michael Matz <matz@suse.de>

>  tools/kgraft/app.c               |   35 +
>  tools/kgraft/app.h               |    7 +

The app is just a dummy to test this on?

> +	./dwarf-inline-tree app.o
> +	@echo "inline pairs"
> +	./dwarf-inline-tree app.o | perl it2rev.pl
> +	@echo "extract stuff"
> +	./objcopy-hacked --strip-unneeded -j .doesntexist. --keep-symbols symlist app.o app-extract.o

Instead of providing local copy of objcopy, should some patch be pushed to FSF?

> +will build most of them, and the check target contains example invocations.
> +The only thing not built automatically is the hacked objcopy (objcopy-hacked),
> +as usually the necessary binutils headers aren't installed.  You'll
> +have to have (recent) binutils sources, apply the patch objcopy.diff
> +and build it yourself.

Ok, I think it should.

> +The seeding symbol list currently needs to come from a human.  It's probably
> +feasible to generate that list for most cases by interpreting a kernel
> +diff.  Binary comparison should _not_ be used to generate it.

And then we reach singularity, because computers will now be able to program
themselves? :-).


> +int global_data;
> +
> +static void __attribute__((noinline)) in_app (void)

in_app(.

> +int main ()

main(int argc, int argv[])

And add a comment that this is dummy app for objdump testing?

> +{
> +  in_app_global();
> +  second_file ();

file(.

> @@ -0,0 +1,7 @@
> +static inline void __attribute__((always_inline)) in_app_inline (void)
> +{
> +  static int local_static_data;
> +  printf ("in_app_inline: %d\n", local_static_data++);
> +}
> +
> +void second_file (void);

Some more spaces before ( need to be deleted.


> +static int __init kgr_patcher_init(void)
> +{
> +        /* removing not supported (yet?) */
> +        __module_get(THIS_MODULE);
> +        /* +4 to skip push rbb / mov rsp,rbp prologue */

What +4 ?

> --- /dev/null
> +++ b/tools/kgraft/dwarf-inline-tree.c
> @@ -0,0 +1,544 @@

GPL, authors would be cool here.


> +#define string char*
> +#include "dwarf_names.h"
> +#undef string

Ouch.

> +    attrib = attr_in;
> +    atname = get_AT_name(dbg, attr);
> +
> +    tres = dwarf_whatform (attrib, &form, &err);
> +    if (tres != DW_DLV_OK)
> +	print_error (dbg, "dwarf_whatform", tres, err);
> +    printf("\t\t%-28s%s\t", atname, get_FORM_name (dbg, form));
> +    /* Don't move over the attributes for the top-level compile_unit
> +     * DIEs.  */
> +    if (tag == DW_TAG_compile_unit)
> +      {
> +	printf ("\n");
> +	return;
> +      }

Is this inherited from GNU code?

> +	case DW_AT_allocated:
> +		if (ellipsis)
> +			return "allocated";
> +		else
> +			return "DW_AT_allocated";
> +	case DW_AT_associated:
> +		if (ellipsis)
> +			return "associated";
> +		else
> +			return "DW_AT_associated";

I have strong feeling that this code is autogenerated, or should be autogenerated. Not 
suitable for kernel git.

Also I believe patch for objdump is best reviewed at GNU mailing lists, and not suitable
for kernel git.

Best regards,
										Pavel

[RFC 13/16] kgr: add MAINTAINERS entry

[Jiri Slaby]

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Michael Matz <matz@suse.de>
Cc: Vojtech Pavlik <vojtech@suse.cz>
---
 MAINTAINERS | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index ea44a57f790e..1b2e692b398f 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5182,6 +5182,15 @@ F:	include/linux/kdb.h
 F:	include/linux/kgdb.h
 F:	kernel/debug/
 
+KGRAFT
+M:	Jiri Kosina <jkosina@suse.cz>
+M:	Jiri Slaby <jslaby@suse.cz>
+M:	Michael Matz <matz@suse.de>
+M:	Vojtech Pavlik <vojtech@suse.cz>
+F:	include/linux/kgr.h
+F:	kernel/kgr.c
+F:	tools/kgraft/
+
 KMEMCHECK
 M:	Vegard Nossum <vegardno@ifi.uio.no>
 M:	Pekka Enberg <penberg@kernel.org>
-- 
1.9.2

[RFC 14/16] kgr: x86: refuse to build without fentry support

[Jiri Slaby]

From: Jiri Kosina <jkosina@suse.cz>

The only reliable way for function redirection through ftrace_ops (when
modifying pt_regs->rip in the handler) is fentry.

The alternative -- mcount -- is problematic in several ways. Namely the
caller's function prologue (that has already been executed by the time
mcount callsite has been reached) is not known to the callee, and can be
completely incompatible to the calee, resulting in a havoc on return from
the function.

fentry doesn't suffer from this, as it's located at the very beginning of
the function, even before prologue has been executed, and therefore callee
is the owner of both function prologue and epilogue.

Fixing mcount to properly fix everything up would be non-trivial, and
Steven is not in favor of doing that.

Both kGraft and upstream kernel (patch to be submitted) should error out
when this unsupported and non-working configuration is detected.

According to Michael Matz, the -mfentry gcc option is x86 specific. Other
architectures insert the respective profile calls before the prologue by
default.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 arch/x86/include/asm/kgr.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
index f36661681b33..8a3819886e4b 100644
--- a/arch/x86/include/asm/kgr.h
+++ b/arch/x86/include/asm/kgr.h
@@ -1,6 +1,10 @@
 #ifndef ASM_KGR_H
 #define ASM_KGR_H
 
+#ifndef CC_USING_FENTRY
+#error Your compiler has to support -mfentry for kGraft to work on x86
+#endif
+
 #include <linux/linkage.h>
 
 /*
-- 
1.9.2

[RFC 15/16] kgr: add procfs interface for per-process 'kgr_in_progress'

[Jiri Slaby]

From: Jiri Kosina <jkosina@suse.cz>

Instead of flooding dmesg with data about tasks which haven't yet been
migrated to the "new universe", create a 'kgr_in_progress' in /proc/<pid>/
so that it's possible to easily script the checks/actions in userspace.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz> [simplification]
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 fs/proc/base.c | 10 ++++++++++
 kernel/kgr.c   |  3 +--
 2 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 2d696b0c93bf..70cba8b21c3f 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -2106,6 +2106,13 @@ static const struct file_operations proc_timers_operations = {
 };
 #endif /* CONFIG_CHECKPOINT_RESTORE */
 
+#ifdef CONFIG_KGR
+static int proc_pid_kgr_in_progress(struct task_struct *task, char *buffer)
+{
+	return sprintf(buffer, "%d\n", task_thread_info(task)->kgr_in_progress);
+}
+#endif /* CONFIG_KGR */
+
 static int proc_pident_instantiate(struct inode *dir,
 	struct dentry *dentry, struct task_struct *task, const void *ptr)
 {
@@ -2638,6 +2645,9 @@ static const struct pid_entry tgid_base_stuff[] = {
 #ifdef CONFIG_CHECKPOINT_RESTORE
 	REG("timers",	  S_IRUGO, proc_timers_operations),
 #endif
+#ifdef CONFIG_KGR
+	INF("kgr_in_progress",	S_IRUSR, proc_pid_kgr_in_progress),
+#endif
 };
 
 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
diff --git a/kernel/kgr.c b/kernel/kgr.c
index ff5afaf6f0e7..1fadde396021 100644
--- a/kernel/kgr.c
+++ b/kernel/kgr.c
@@ -45,9 +45,8 @@ static bool kgr_still_patching(void)
 	read_lock(&tasklist_lock);
 	for_each_process(p) {
 		if (task_thread_info(p)->kgr_in_progress) {
-			pr_info("pid %d (%s) still in kernel after timeout\n",
-					p->pid, p->comm);
 			failed = true;
+			break;
 		}
 	}
 	read_unlock(&tasklist_lock);
-- 
1.9.2

[RFC 16/16] kgr: make a per-process 'in progress' flag a single bit

[Jiri Slaby]

From: Jiri Kosina <jkosina@suse.cz>

Having the per-task 'kgr_in_progress' flag stored as int is a waste of
space, and manipulating it is likely slower than just performing a single
bit operations. Convert the flag to thread info flag.

Additionaly, makking KGR TI_flag part of _TIF_ALLWORK_MASK and
_TIF_WORK_SYSCALL_ENTRY allows for offloading the flag manipulation
to slow code paths.

js: use *_tsk_thread_flag helpers

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
---
 arch/x86/include/asm/kgr.h         |  2 +-
 arch/x86/include/asm/thread_info.h |  7 ++++---
 arch/x86/kernel/asm-offsets.c      |  1 -
 arch/x86/kernel/entry_64.S         | 12 +++++++++---
 fs/proc/base.c                     |  3 ++-
 include/linux/kgr.h                | 14 ++++++++++++++
 include/linux/sched.h              |  2 +-
 kernel/kgr.c                       |  4 ++--
 8 files changed, 33 insertions(+), 12 deletions(-)

diff --git a/arch/x86/include/asm/kgr.h b/arch/x86/include/asm/kgr.h
index 8a3819886e4b..44d32c22fbac 100644
--- a/arch/x86/include/asm/kgr.h
+++ b/arch/x86/include/asm/kgr.h
@@ -18,7 +18,7 @@ static void _new_function ##_stub_slow (unsigned long ip, unsigned long parent_i
 	struct kgr_loc_caches *c = ops->private;			\
 	bool irq = !!in_interrupt();					\
 									\
-	if ((!irq && task_thread_info(current)->kgr_in_progress) ||	\
+	if ((!irq && kgr_task_in_progress(current)) ||			\
 			(irq && !*this_cpu_ptr(c->irq_use_new))) {	\
 		pr_info("kgr: slow stub: calling old code at %lx\n",	\
 				c->old);				\
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index 1fdc144dcc9c..06ef370044cf 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -35,7 +35,6 @@ struct thread_info {
 	void __user		*sysenter_return;
 	unsigned int		sig_on_uaccess_error:1;
 	unsigned int		uaccess_err:1;	/* uaccess failed */
-	unsigned short		kgr_in_progress;
 };
 
 #define INIT_THREAD_INFO(tsk)			\
@@ -87,6 +86,7 @@ struct thread_info {
 #define TIF_IO_BITMAP		22	/* uses I/O bitmap */
 #define TIF_FORCED_TF		24	/* true if TF in eflags artificially */
 #define TIF_BLOCKSTEP		25	/* set when we want DEBUGCTLMSR_BTF */
+#define TIF_KGR_IN_PROGRESS	26	/* kgr patching running */
 #define TIF_LAZY_MMU_UPDATES	27	/* task is updating the mmu lazily */
 #define TIF_SYSCALL_TRACEPOINT	28	/* syscall tracepoint instrumentation */
 #define TIF_ADDR32		29	/* 32-bit address space on 64 bits */
@@ -110,6 +110,7 @@ struct thread_info {
 #define _TIF_IO_BITMAP		(1 << TIF_IO_BITMAP)
 #define _TIF_FORCED_TF		(1 << TIF_FORCED_TF)
 #define _TIF_BLOCKSTEP		(1 << TIF_BLOCKSTEP)
+#define _TIF_KGR_IN_PROGRESS	(1 << TIF_KGR_IN_PROGRESS)
 #define _TIF_LAZY_MMU_UPDATES	(1 << TIF_LAZY_MMU_UPDATES)
 #define _TIF_SYSCALL_TRACEPOINT	(1 << TIF_SYSCALL_TRACEPOINT)
 #define _TIF_ADDR32		(1 << TIF_ADDR32)
@@ -119,7 +120,7 @@ struct thread_info {
 #define _TIF_WORK_SYSCALL_ENTRY	\
 	(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU | _TIF_SYSCALL_AUDIT |	\
 	 _TIF_SECCOMP | _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT |	\
-	 _TIF_NOHZ)
+	 _TIF_NOHZ | _TIF_KGR_IN_PROGRESS)
 
 /* work to do in syscall_trace_leave() */
 #define _TIF_WORK_SYSCALL_EXIT	\
@@ -135,7 +136,7 @@ struct thread_info {
 /* work to do on any return to user space */
 #define _TIF_ALLWORK_MASK						\
 	((0x0000FFFF & ~_TIF_SECCOMP) | _TIF_SYSCALL_TRACEPOINT |	\
-	_TIF_NOHZ)
+	_TIF_NOHZ | _TIF_KGR_IN_PROGRESS)
 
 /* Only used for 64 bit */
 #define _TIF_DO_NOTIFY_MASK						\
diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
index 0db0437967a2..9f6b9341950f 100644
--- a/arch/x86/kernel/asm-offsets.c
+++ b/arch/x86/kernel/asm-offsets.c
@@ -32,7 +32,6 @@ void common(void) {
 	OFFSET(TI_flags, thread_info, flags);
 	OFFSET(TI_status, thread_info, status);
 	OFFSET(TI_addr_limit, thread_info, addr_limit);
-	OFFSET(TI_kgr_in_progress, thread_info, kgr_in_progress);
 
 	BLANK();
 	OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index a03b1e9d2de3..fbf391e99c46 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -615,7 +615,6 @@ GLOBAL(system_call_after_swapgs)
 	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
 	movq  %rcx,RIP-ARGOFFSET(%rsp)
 	CFI_REL_OFFSET rip,RIP-ARGOFFSET
-	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	jnz tracesys
 system_call_fastpath:
@@ -640,7 +639,6 @@ sysret_check:
 	LOCKDEP_SYS_EXIT
 	DISABLE_INTERRUPTS(CLBR_NONE)
 	TRACE_IRQS_OFF
-	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
 	andl %edi,%edx
 	jnz  sysret_careful
@@ -660,6 +658,9 @@ sysret_check:
 	/* Handle reschedules */
 	/* edx:	work, edi: workmask */
 sysret_careful:
+#ifdef CONFIG_KGR
+	andl $~_TIF_KGR_IN_PROGRESS,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+#endif
 	bt $TIF_NEED_RESCHED,%edx
 	jnc sysret_signal
 	TRACE_IRQS_ON
@@ -723,6 +724,9 @@ sysret_audit:
 
 	/* Do syscall tracing */
 tracesys:
+#ifdef CONFIG_KGR
+	andl $~_TIF_KGR_IN_PROGRESS,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+#endif
 #ifdef CONFIG_AUDITSYSCALL
 	testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
 	jz auditsys
@@ -763,7 +767,6 @@ GLOBAL(int_ret_from_sys_call)
 GLOBAL(int_with_check)
 	LOCKDEP_SYS_EXIT_IRQ
 	GET_THREAD_INFO(%rcx)
-	movw $0, TI_kgr_in_progress(%rcx)
 	movl TI_flags(%rcx),%edx
 	andl %edi,%edx
 	jnz   int_careful
@@ -774,6 +777,9 @@ GLOBAL(int_with_check)
 	/* First do a reschedule test. */
 	/* edx:	work, edi: workmask */
 int_careful:
+#ifdef CONFIG_KGR
+	andl $~_TIF_KGR_IN_PROGRESS,TI_flags(%rcx)
+#endif
 	bt $TIF_NEED_RESCHED,%edx
 	jnc  int_very_careful
 	TRACE_IRQS_ON
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 70cba8b21c3f..21d7841ec60d 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -87,6 +87,7 @@
 #include <linux/slab.h>
 #include <linux/flex_array.h>
 #include <linux/posix-timers.h>
+#include <linux/kgr.h>
 #ifdef CONFIG_HARDWALL
 #include <asm/hardwall.h>
 #endif
@@ -2109,7 +2110,7 @@ static const struct file_operations proc_timers_operations = {
 #ifdef CONFIG_KGR
 static int proc_pid_kgr_in_progress(struct task_struct *task, char *buffer)
 {
-	return sprintf(buffer, "%d\n", task_thread_info(task)->kgr_in_progress);
+	return sprintf(buffer, "%d\n", kgr_task_in_progress(task));
 }
 #endif /* CONFIG_KGR */
 
diff --git a/include/linux/kgr.h b/include/linux/kgr.h
index ebc6f5bc1ec1..7a1a4d9d97f4 100644
--- a/include/linux/kgr.h
+++ b/include/linux/kgr.h
@@ -4,6 +4,9 @@
 #include <linux/init.h>
 #include <linux/ftrace.h>
 
+static void kgr_mark_task_in_progress(struct task_struct *p);
+static bool kgr_task_in_progress(struct task_struct *p);
+
 #include <asm/kgr.h>
 
 #ifdef CONFIG_KGR
@@ -67,6 +70,17 @@ struct kgr_loc_caches {
 #define KGR_PATCH_END		NULL
 
 extern int kgr_start_patching(struct kgr_patch *);
+
+static inline void kgr_mark_task_in_progress(struct task_struct *p)
+{
+	set_tsk_thread_flag(p, TIF_KGR_IN_PROGRESS);
+}
+
+static inline bool kgr_task_in_progress(struct task_struct *p)
+{
+	return test_tsk_thread_flag(p, TIF_KGR_IN_PROGRESS);
+}
+
 #endif /* CONFIG_KGR */
 
 #endif /* LINUX_KGR_H */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index afd5747bc7ff..8efd164f1962 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2972,7 +2972,7 @@ static inline void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 #ifdef CONFIG_KGR
 static inline void kgr_task_safe(struct task_struct *p)
 {
-	task_thread_info(p)->kgr_in_progress = false;
+	clear_tsk_thread_flag(p, TIF_KGR_IN_PROGRESS);
 }
 #else
 static inline void kgr_task_safe(struct task_struct *p) { }
diff --git a/kernel/kgr.c b/kernel/kgr.c
index 1fadde396021..a55409122e77 100644
--- a/kernel/kgr.c
+++ b/kernel/kgr.c
@@ -44,7 +44,7 @@ static bool kgr_still_patching(void)
 
 	read_lock(&tasklist_lock);
 	for_each_process(p) {
-		if (task_thread_info(p)->kgr_in_progress) {
+		if (kgr_task_in_progress(p)) {
 			failed = true;
 			break;
 		}
@@ -98,7 +98,7 @@ static void kgr_handle_processes(void)
 
 	read_lock(&tasklist_lock);
 	for_each_process(p) {
-		task_thread_info(p)->kgr_in_progress = true;
+		kgr_mark_task_in_progress(p);
 
 		/* wake up kthreads, they will clean the progress flag */
 		if (!p->mm) {
-- 
1.9.2