[PATCH v2 8/8] x86/unwind: add undwarf unwinder

[Josh Poimboeuf]

Add a new 'undwarf' unwinder which is enabled by
CONFIG_UNDWARF_UNWINDER.  It plugs into the existing x86 unwinder
framework.

It relies on objtool to generate the needed .undwarf section.

For more details on why undwarf is used instead of DWARF, see
tools/objtool/Documentation/undwarf.txt.

Thanks to Andy Lutomirski for the performance improvement ideas:
splitting the undwarf table into two parallel arrays and creating a fast
lookup table to search a subset of the undwarf table.

Signed-off-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
---
 Documentation/x86/undwarf.txt     | 146 ++++++++++
 arch/um/include/asm/unwind.h      |   8 +
 arch/x86/Kconfig                  |   1 +
 arch/x86/Kconfig.debug            |  25 ++
 arch/x86/include/asm/module.h     |   9 +
 arch/x86/include/asm/unwind.h     |  77 +++--
 arch/x86/kernel/Makefile          |   8 +-
 arch/x86/kernel/module.c          |  12 +-
 arch/x86/kernel/setup.c           |   3 +
 arch/x86/kernel/unwind_frame.c    |  39 ++-
 arch/x86/kernel/unwind_guess.c    |   5 +
 arch/x86/kernel/unwind_undwarf.c  | 589 ++++++++++++++++++++++++++++++++++++++
 arch/x86/kernel/vmlinux.lds.S     |   2 +
 include/asm-generic/vmlinux.lds.h |  20 +-
 lib/Kconfig.debug                 |   3 +
 scripts/Makefile.build            |  14 +-
 16 files changed, 898 insertions(+), 63 deletions(-)
 create mode 100644 Documentation/x86/undwarf.txt
 create mode 100644 arch/um/include/asm/unwind.h
 create mode 100644 arch/x86/kernel/unwind_undwarf.c

diff --git a/Documentation/x86/undwarf.txt b/Documentation/x86/undwarf.txt
new file mode 100644
index 0000000..d76c6b4
--- /dev/null
+++ b/Documentation/x86/undwarf.txt
@@ -0,0 +1,146 @@
+Undwarf unwinder debuginfo generation
+=====================================
+
+Overview
+--------
+
+The kernel CONFIG_UNDWARF_UNWINDER option enables objtool generation of
+undwarf debuginfo, which is out-of-band data which is used by the
+in-kernel undwarf unwinder.  It's similar in concept to DWARF CFI
+debuginfo which would be used by a DWARF unwinder.  The difference is
+that the format of the undwarf data is simpler than DWARF, which in turn
+allows the unwinder to be simpler and faster.
+
+Objtool generates the undwarf data by first doing compile-time stack
+metadata validation (CONFIG_STACK_VALIDATION).  After analyzing all the
+code paths of a .o file, it determines information about the stack state
+at each instruction address in the file and outputs that information to
+the .undwarf and .undwarf_ip sections.
+
+The undwarf sections are combined at link time and are sorted at boot
+time.  The unwinder uses the resulting data to correlate instruction
+addresses with their stack states at run time.
+
+
+Undwarf vs frame pointers
+-------------------------
+
+With frame pointers enabled, GCC adds instrumentation code to every
+function in the kernel.  The kernel's .text size increases by about
+3.2%, resulting in a broad kernel-wide slowdown.  Measurements by Mel
+Gorman [1] have shown a slowdown of 5-10% for some workloads.
+
+In contrast, the undwarf unwinder has no effect on text size or runtime
+performance, because the debuginfo is out of band.  So if you disable
+frame pointers and enable undwarf, you get a nice performance
+improvement across the board, and still have reliable stack traces.
+
+Another benefit of undwarf compared to frame pointers is that it can
+reliably unwind across interrupts and exceptions.  Frame pointer based
+unwinds can skip the caller of the interrupted function if it was a leaf
+function or if the interrupt hit before the frame pointer was saved.
+
+The main disadvantage of undwarf compared to frame pointers is that it
+needs more memory to store the undwarf table: roughly 2-4MB depending on
+the kernel config.
+
+
+Undwarf vs DWARF
+----------------
+
+Undwarf debuginfo's advantage over DWARF itself is that it's much
+simpler.  It gets rid of the complex DWARF CFI state machine and also
+gets rid of the tracking of unnecessary registers.  This allows the
+unwinder to be much simpler, meaning fewer bugs, which is especially
+important for mission critical oops code.
+
+The simpler debuginfo format also enables the unwinder to be much faster
+than DWARF, which is important for perf and lockdep.  In a basic
+performance test by Jiri Slaby [2], the undwarf unwinder was about 20x
+faster than an out-of-tree DWARF unwinder.  (Note: that measurement was
+taken before some performance tweaks were implemented, so the speedup
+may be even higher.)
+
+The undwarf format does have a few downsides compared to DWARF.  The
+undwarf table takes up ~2MB more memory than an DWARF .eh_frame table.
+
+Another potential downside is that, as GCC evolves, it's conceivable
+that the undwarf data may end up being *too* simple to describe the
+state of the stack for certain optimizations.  But IMO this is unlikely
+because GCC saves the frame pointer for any unusual stack adjustments it
+does, so I suspect we'll really only ever need to keep track of the
+stack pointer and the frame pointer between call frames.  But even if we
+do end up having to track all the registers DWARF tracks, at least we
+will still be able to control the format, e.g.  no complex state
+machines.
+
+
+Undwarf debuginfo generation
+----------------------------
+
+The undwarf data is generated by objtool.  With the existing
+compile-time stack metadata validation feature, objtool already follows
+all code paths, and so it already has all the information it needs to be
+able to generate undwarf data from scratch.  So it's an easy step to go
+from stack validation to undwarf generation.
+
+It should be possible to instead generate the undwarf data with a simple
+tool which converts DWARF to undwarf.  However, such a solution would be
+incomplete due to the kernel's extensive use of asm, inline asm, and
+special sections like exception tables.
+
+That could be rectified by manually annotating those special code paths
+using GNU assembler .cfi annotations in .S files, and homegrown
+annotations for inline asm in .c files.  But asm annotations were tried
+in the past and were found to be unmaintainable.  They were often
+incorrect/incomplete and made the code harder to read and keep updated.
+And based on looking at glibc code, annotating inline asm in .c files
+might be even worse.
+
+Objtool still needs a few annotations, but only in code which does
+unusual things to the stack like entry code.  And even then, far fewer
+annotations are needed than what DWARF would need, so they're much more
+maintainable than DWARF CFI annotations.
+
+So the advantages of using objtool to generate undwarf are that it gives
+more accurate debuginfo, with very few annotations.  It also insulates
+the kernel from toolchain bugs which can be very painful to deal with in
+the kernel since we often have to workaround issues in older versions of
+the toolchain for years.
+
+The downside is that the unwinder now becomes dependent on objtool's
+ability to reverse engineer GCC code paths.  If GCC optimizations become
+too complicated for objtool to follow, the undwarf generation might stop
+working or become incomplete.  (It's worth noting that livepatch already
+has such a dependency on objtool's ability to follow GCC code paths.)
+
+If newer versions of GCC come up with some optimizations which break
+objtool, we may need to revisit the current implementation.  Some
+possible solutions would be asking GCC to make the optimizations more
+palatable, or having objtool use DWARF as an additional input, or
+creating a GCC plugin to assist objtool with its analysis.  But for now,
+objtool follows GCC code quite well.
+
+
+Unwinder implementation details
+-------------------------------
+
+Objtool generates the undwarf data by integrating with the compile-time
+stack metadata validation feature, which is described in detail in
+tools/objtool/Documentation/stack-validation.txt.  After analyzing all
+the code paths of a .o file, it creates an array of undwarf structs, and
+a parallel array of instruction addresses associated with those structs,
+and writes them to the .undwarf and .undwarf_ip sections respectively.
+
+The undwarf data is split into the two arrays for performance reasons,
+to make the searchable part of the data (.undwarf_ip) more compact.  The
+arrays are sorted in parallel at boot time.
+
+Performance is further improved by the use of a fast lookup table which
+is created at runtime.  The fast lookup table associates a given address
+with a range of undwarf table indices, so that only a small subset of
+the undwarf table needs to be searched.
+
+
+[1] https://lkml.kernel.org/r/20170602104048.jkkzssljsompjdwy@xxxxxxx
+[2] https://lkml.kernel.org/r/d2ca5435-6386-29b8-db87-7f227c2b713a@xxxxxxx
diff --git a/arch/um/include/asm/unwind.h b/arch/um/include/asm/unwind.h
new file mode 100644
index 0000000..53f507c
--- /dev/null
+++ b/arch/um/include/asm/unwind.h
@@ -0,0 +1,8 @@
+#ifndef _ASM_UML_UNWIND_H
+#define _ASM_UML_UNWIND_H
+
+static inline void
+unwind_module_init(struct module *mod, void *undwarf_ip, size_t unward_ip_size,
+		   void *undwarf, size_t undwarf_size) {}
+
+#endif /* _ASM_UML_UNWIND_H */
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 72028a1..adf3222 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -153,6 +153,7 @@ config X86
 	select HAVE_MEMBLOCK
 	select HAVE_MEMBLOCK_NODE_MAP
 	select HAVE_MIXED_BREAKPOINTS_REGS
+	select HAVE_MOD_ARCH_SPECIFIC
 	select HAVE_NMI
 	select HAVE_OPROFILE
 	select HAVE_OPTPROBES
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index fcb7604..995434c 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -357,4 +357,29 @@ config PUNIT_ATOM_DEBUG
 	  The current power state can be read from
 	  /sys/kernel/debug/punit_atom/dev_power_state
 
+config UNDWARF_UNWINDER
+	bool "undwarf unwinder"
+	depends on X86_64
+	select STACK_VALIDATION
+	---help---
+	  This option enables the "undwarf" unwinder for unwinding kernel stack
+	  traces.  It uses a custom data format which is a simplified version
+	  of the DWARF Call Frame Information standard.
+
+	  This unwinder is more accurate across interrupt entry frames than the
+	  frame pointer unwinder.  This also can enable a small performance
+	  improvement across the entire kernel if CONFIG_FRAME_POINTER is
+	  disabled.
+
+	  Enabling this option will increase the kernel's runtime memory usage
+	  by roughly 2-4MB, depending on your kernel config.
+
+config FRAME_POINTER_UNWINDER
+	def_bool y
+	depends on !UNDWARF_UNWINDER && FRAME_POINTER
+
+config GUESS_UNWINDER
+	def_bool y
+	depends on !UNDWARF_UNWINDER && !FRAME_POINTER
+
 endmenu
diff --git a/arch/x86/include/asm/module.h b/arch/x86/include/asm/module.h
index e3b7819..4dc6427 100644
--- a/arch/x86/include/asm/module.h
+++ b/arch/x86/include/asm/module.h
@@ -2,6 +2,15 @@
 #define _ASM_X86_MODULE_H
 
 #include <asm-generic/module.h>
+#include <asm/undwarf.h>
+
+struct mod_arch_specific {
+#ifdef CONFIG_UNDWARF_UNWINDER
+	unsigned int num_undwarves;
+	int *undwarf_ip;
+	struct undwarf *undwarf;
+#endif
+};
 
 #ifdef CONFIG_X86_64
 /* X86_64 does not define MODULE_PROC_FAMILY */
diff --git a/arch/x86/include/asm/unwind.h b/arch/x86/include/asm/unwind.h
index e667649..1f8cb78 100644
--- a/arch/x86/include/asm/unwind.h
+++ b/arch/x86/include/asm/unwind.h
@@ -12,11 +12,14 @@ struct unwind_state {
 	struct task_struct *task;
 	int graph_idx;
 	bool error;
-#ifdef CONFIG_FRAME_POINTER
+#if defined(CONFIG_UNDWARF_UNWINDER)
+	bool signal, full_regs;
+	unsigned long sp, bp, ip;
+	struct pt_regs *regs;
+#elif defined(CONFIG_FRAME_POINTER)
 	bool got_irq;
-	unsigned long *bp, *orig_sp;
+	unsigned long *bp, *orig_sp, ip;
 	struct pt_regs *regs;
-	unsigned long ip;
 #else
 	unsigned long *sp;
 #endif
@@ -24,41 +27,30 @@ struct unwind_state {
 
 void __unwind_start(struct unwind_state *state, struct task_struct *task,
 		    struct pt_regs *regs, unsigned long *first_frame);
-
 bool unwind_next_frame(struct unwind_state *state);
-
 unsigned long unwind_get_return_address(struct unwind_state *state);
+unsigned long *unwind_get_return_address_ptr(struct unwind_state *state);
 
 static inline bool unwind_done(struct unwind_state *state)
 {
 	return state->stack_info.type == STACK_TYPE_UNKNOWN;
 }
 
-static inline
-void unwind_start(struct unwind_state *state, struct task_struct *task,
-		  struct pt_regs *regs, unsigned long *first_frame)
-{
-	first_frame = first_frame ? : get_stack_pointer(task, regs);
-
-	__unwind_start(state, task, regs, first_frame);
-}
-
 static inline bool unwind_error(struct unwind_state *state)
 {
 	return state->error;
 }
 
-#ifdef CONFIG_FRAME_POINTER
-
 static inline
-unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
+void unwind_start(struct unwind_state *state, struct task_struct *task,
+		  struct pt_regs *regs, unsigned long *first_frame)
 {
-	if (unwind_done(state))
-		return NULL;
+	first_frame = first_frame ? : get_stack_pointer(task, regs);
 
-	return state->regs ? &state->regs->ip : state->bp + 1;
+	__unwind_start(state, task, regs, first_frame);
 }
 
+#if defined(CONFIG_UNDWARF_UNWINDER) || defined(CONFIG_FRAME_POINTER)
 static inline struct pt_regs *unwind_get_entry_regs(struct unwind_state *state)
 {
 	if (unwind_done(state))
@@ -66,20 +58,47 @@ static inline struct pt_regs *unwind_get_entry_regs(struct unwind_state *state)
 
 	return state->regs;
 }
-
-#else /* !CONFIG_FRAME_POINTER */
-
-static inline
-unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
+#else
+static inline struct pt_regs *unwind_get_entry_regs(struct unwind_state *state)
 {
 	return NULL;
 }
+#endif
 
-static inline struct pt_regs *unwind_get_entry_regs(struct unwind_state *state)
+#ifdef CONFIG_UNDWARF_UNWINDER
+void unwind_init(void);
+void unwind_module_init(struct module *mod, void *undwarf_ip,
+			size_t undwarf_ip_size, void *undwarf,
+			size_t undwarf_size);
+#else
+static inline void unwind_init(void) {}
+static inline void unwind_module_init(struct module *mod, void *undwarf_ip,
+				      size_t undwarf_ip_size, void *undwarf,
+				      size_t undwarf_size) {}
+#endif
+
+/*
+ * This disables KASAN checking when reading a value from another task's stack,
+ * since the other task could be running on another CPU and could have poisoned
+ * the stack in the meantime.
+ */
+#define READ_ONCE_TASK_STACK(task, x)			\
+({							\
+	unsigned long val;				\
+	if (task == current)				\
+		val = READ_ONCE(x);			\
+	else						\
+		val = READ_ONCE_NOCHECK(x);		\
+	val;						\
+})
+
+static inline bool task_on_another_cpu(struct task_struct *task)
 {
-	return NULL;
+#ifdef CONFIG_SMP
+	return task != current && task->on_cpu;
+#else
+	return false;
+#endif
 }
 
-#endif /* CONFIG_FRAME_POINTER */
-
 #endif /* _ASM_X86_UNWIND_H */
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 3c7c419..4865889 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -125,11 +125,9 @@ obj-$(CONFIG_PERF_EVENTS)		+= perf_regs.o
 obj-$(CONFIG_TRACING)			+= tracepoint.o
 obj-$(CONFIG_SCHED_MC_PRIO)		+= itmt.o
 
-ifdef CONFIG_FRAME_POINTER
-obj-y					+= unwind_frame.o
-else
-obj-y					+= unwind_guess.o
-endif
+obj-$(CONFIG_UNDWARF_UNWINDER)		+= unwind_undwarf.o
+obj-$(CONFIG_FRAME_POINTER_UNWINDER)	+= unwind_frame.o
+obj-$(CONFIG_GUESS_UNWINDER)		+= unwind_guess.o
 
 ###
 # 64 bit specific files
diff --git a/arch/x86/kernel/module.c b/arch/x86/kernel/module.c
index f67bd32..203b5a7 100644
--- a/arch/x86/kernel/module.c
+++ b/arch/x86/kernel/module.c
@@ -35,6 +35,7 @@
 #include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/setup.h>
+#include <asm/unwind.h>
 
 #if 0
 #define DEBUGP(fmt, ...)				\
@@ -213,7 +214,7 @@ int module_finalize(const Elf_Ehdr *hdr,
 		    struct module *me)
 {
 	const Elf_Shdr *s, *text = NULL, *alt = NULL, *locks = NULL,
-		*para = NULL;
+		*para = NULL, *undwarf = NULL, *undwarf_ip = NULL;
 	char *secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
 
 	for (s = sechdrs; s < sechdrs + hdr->e_shnum; s++) {
@@ -225,6 +226,10 @@ int module_finalize(const Elf_Ehdr *hdr,
 			locks = s;
 		if (!strcmp(".parainstructions", secstrings + s->sh_name))
 			para = s;
+		if (!strcmp(".undwarf", secstrings + s->sh_name))
+			undwarf = s;
+		if (!strcmp(".undwarf_ip", secstrings + s->sh_name))
+			undwarf_ip = s;
 	}
 
 	if (alt) {
@@ -248,6 +253,11 @@ int module_finalize(const Elf_Ehdr *hdr,
 	/* make jump label nops */
 	jump_label_apply_nops(me);
 
+	if (undwarf && undwarf_ip)
+		unwind_module_init(me, (void *)undwarf_ip->sh_addr,
+				   undwarf_ip->sh_size,
+				   (void *)undwarf->sh_addr, undwarf->sh_size);
+
 	return 0;
 }
 
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 65622f0..d736761 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -115,6 +115,7 @@
 #include <asm/microcode.h>
 #include <asm/mmu_context.h>
 #include <asm/kaslr.h>
+#include <asm/unwind.h>
 
 /*
  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
@@ -1303,6 +1304,8 @@ void __init setup_arch(char **cmdline_p)
 	if (efi_enabled(EFI_BOOT))
 		efi_apply_memmap_quirks();
 #endif
+
+	unwind_init();
 }
 
 #ifdef CONFIG_X86_32
diff --git a/arch/x86/kernel/unwind_frame.c b/arch/x86/kernel/unwind_frame.c
index b9389d7..7574ef5 100644
--- a/arch/x86/kernel/unwind_frame.c
+++ b/arch/x86/kernel/unwind_frame.c
@@ -10,20 +10,22 @@
 
 #define FRAME_HEADER_SIZE (sizeof(long) * 2)
 
-/*
- * This disables KASAN checking when reading a value from another task's stack,
- * since the other task could be running on another CPU and could have poisoned
- * the stack in the meantime.
- */
-#define READ_ONCE_TASK_STACK(task, x)			\
-({							\
-	unsigned long val;				\
-	if (task == current)				\
-		val = READ_ONCE(x);			\
-	else						\
-		val = READ_ONCE_NOCHECK(x);		\
-	val;						\
-})
+unsigned long unwind_get_return_address(struct unwind_state *state)
+{
+	if (unwind_done(state))
+		return 0;
+
+	return __kernel_text_address(state->ip) ? state->ip : 0;
+}
+EXPORT_SYMBOL_GPL(unwind_get_return_address);
+
+unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
+{
+	if (unwind_done(state))
+		return NULL;
+
+	return state->regs ? &state->regs->ip : state->bp + 1;
+}
 
 static void unwind_dump(struct unwind_state *state)
 {
@@ -66,15 +68,6 @@ static void unwind_dump(struct unwind_state *state)
 	}
 }
 
-unsigned long unwind_get_return_address(struct unwind_state *state)
-{
-	if (unwind_done(state))
-		return 0;
-
-	return __kernel_text_address(state->ip) ? state->ip : 0;
-}
-EXPORT_SYMBOL_GPL(unwind_get_return_address);
-
 static size_t regs_size(struct pt_regs *regs)
 {
 	/* x86_32 regs from kernel mode are two words shorter: */
diff --git a/arch/x86/kernel/unwind_guess.c b/arch/x86/kernel/unwind_guess.c
index 039f367..4f0e17b 100644
--- a/arch/x86/kernel/unwind_guess.c
+++ b/arch/x86/kernel/unwind_guess.c
@@ -19,6 +19,11 @@ unsigned long unwind_get_return_address(struct unwind_state *state)
 }
 EXPORT_SYMBOL_GPL(unwind_get_return_address);
 
+unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
+{
+	return NULL;
+}
+
 bool unwind_next_frame(struct unwind_state *state)
 {
 	struct stack_info *info = &state->stack_info;
diff --git a/arch/x86/kernel/unwind_undwarf.c b/arch/x86/kernel/unwind_undwarf.c
new file mode 100644
index 0000000..44f62af
--- /dev/null
+++ b/arch/x86/kernel/unwind_undwarf.c
@@ -0,0 +1,589 @@
+#include <linux/module.h>
+#include <linux/sort.h>
+#include <asm/ptrace.h>
+#include <asm/stacktrace.h>
+#include <asm/unwind.h>
+#include <asm/undwarf.h>
+#include <asm/sections.h>
+
+#define undwarf_warn(fmt, ...) \
+	printk_deferred_once(KERN_WARNING pr_fmt("WARNING: " fmt), ##__VA_ARGS__)
+
+extern int __start_undwarf_ip[];
+extern int __stop_undwarf_ip[];
+extern struct undwarf __start_undwarf[];
+extern struct undwarf __stop_undwarf[];
+
+bool undwarf_init;
+static DEFINE_MUTEX(sort_mutex);
+
+int *cur_undwarf_ip_table = __start_undwarf_ip;
+struct undwarf *cur_undwarf_table = __start_undwarf;
+
+/*
+ * This is a lookup table for speeding up access to the undwarf table.  Given
+ * an input address offset, the corresponding lookup table entry specifies a
+ * subset of the undwarf table to search.
+ *
+ * Each block represents the end of the previous range and the start of the
+ * next range.  An extra block is added to give the last range an end.
+ *
+ * Some measured performance results for different values of LOOKUP_NUM_BLOCKS:
+ *
+ *  num blocks       array size	   lookup speedup   total speedup
+ *     2k		8k		1.5x		1.5x
+ *     4k		16k		1.6x		1.6x
+ *     8k		32k		1.8x		1.7x
+ *     16k		64k		2.0x		1.8x
+ *     32k		128k		2.5x		2.0x
+ *     64k		256k		2.9x		2.2x
+ *     128k		512k		3.3x		2.4x
+ *
+ * Go with 32k blocks because it doubles unwinder performance while only adding
+ * 3.5% to the undwarf data footprint.
+ */
+#define LOOKUP_NUM_BLOCKS		(32 * 1024)
+static unsigned int undwarf_fast_lookup[LOOKUP_NUM_BLOCKS + 1] __ro_after_init;
+
+#define LOOKUP_START_IP			(unsigned long)_stext
+#define LOOKUP_STOP_IP			(unsigned long)_etext
+#define LOOKUP_BLOCK_SIZE						\
+	(DIV_ROUND_UP(LOOKUP_STOP_IP - LOOKUP_START_IP,			\
+		      LOOKUP_NUM_BLOCKS))
+
+
+static inline unsigned long undwarf_ip(const int *ip)
+{
+	return (unsigned long)ip + *ip;
+}
+
+static struct undwarf *__undwarf_find(int *ip_table, struct undwarf *u_table,
+				      unsigned int num_entries,
+				      unsigned long ip)
+{
+	int *first = ip_table;
+	int *last = ip_table + num_entries - 1;
+	int *mid = first, *found = first;
+
+	if (!num_entries)
+		return NULL;
+
+	/*
+	 * Do a binary range search to find the rightmost duplicate of a given
+	 * starting address.  Some entries are section terminators which are
+	 * "weak" entries for ensuring there are no gaps.  They should be
+	 * ignored when they conflict with a real entry.
+	 */
+	while (first <= last) {
+		mid = first + ((last - first) / 2);
+
+		if (undwarf_ip(mid) <= ip) {
+			found = mid;
+			first = mid + 1;
+		} else
+			last = mid - 1;
+	}
+
+	return u_table + (found - ip_table);
+}
+
+static struct undwarf *undwarf_find(unsigned long ip)
+{
+	struct module *mod;
+
+	if (!undwarf_init)
+		return NULL;
+
+	/* For non-init vmlinux addresses, use the fast lookup table: */
+	if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
+		unsigned int idx, start, stop;
+
+		idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
+
+		if (WARN_ON_ONCE(idx >= LOOKUP_NUM_BLOCKS))
+			return NULL;
+
+		start = undwarf_fast_lookup[idx];
+		stop = undwarf_fast_lookup[idx + 1] + 1;
+
+		if (WARN_ON_ONCE(__start_undwarf + start >= __stop_undwarf) ||
+				 __start_undwarf + stop > __stop_undwarf)
+			return NULL;
+
+		return __undwarf_find(__start_undwarf_ip + start,
+				      __start_undwarf + start,
+				      stop - start, ip);
+	}
+
+	/* vmlinux .init slow lookup: */
+	if (ip >= (unsigned long)_sinittext && ip < (unsigned long)_einittext)
+		return __undwarf_find(__start_undwarf_ip, __start_undwarf,
+				      __stop_undwarf - __start_undwarf, ip);
+
+	/* Module lookup: */
+	mod = __module_address(ip);
+	if (!mod || !mod->arch.undwarf || !mod->arch.undwarf_ip)
+		return NULL;
+	return __undwarf_find(mod->arch.undwarf_ip, mod->arch.undwarf,
+			      mod->arch.num_undwarves, ip);
+}
+
+static void undwarf_sort_swap(void *_a, void *_b, int size)
+{
+	struct undwarf *undwarf_a, *undwarf_b;
+	struct undwarf undwarf_tmp;
+	int *a = _a, *b = _b, tmp;
+	int delta = _b - _a;
+
+	/* Swap the undwarf_ip entries: */
+	tmp = *a;
+	*a = *b + delta;
+	*b = tmp - delta;
+
+	/* Swap the corresponding undwarf entries: */
+	undwarf_a = cur_undwarf_table + (a - cur_undwarf_ip_table);
+	undwarf_b = cur_undwarf_table + (b - cur_undwarf_ip_table);
+	undwarf_tmp = *undwarf_a;
+	*undwarf_a = *undwarf_b;
+	*undwarf_b = undwarf_tmp;
+}
+
+static int undwarf_sort_cmp(const void *_a, const void *_b)
+{
+	struct undwarf *undwarf_a;
+	const int *a = _a, *b = _b;
+	unsigned long a_val = undwarf_ip(a);
+	unsigned long b_val = undwarf_ip(b);
+
+	if (a_val > b_val)
+		return 1;
+	if (a_val < b_val)
+		return -1;
+
+	/*
+	 * The "weak" section terminator entries need to always be on the left
+	 * to ensure the lookup code skips them in favor of real entries.
+	 * These terminator entries exist to handle any gaps created by
+	 * whitelisted .o files which didn't get objtool generation.
+	 */
+	undwarf_a = cur_undwarf_table + (a - cur_undwarf_ip_table);
+	return undwarf_a->cfa_reg == UNDWARF_REG_UNDEFINED ? -1 : 1;
+}
+
+void unwind_module_init(struct module *mod, void *_undwarf_ip,
+			size_t undwarf_ip_size, void *_undwarf,
+			size_t undwarf_size)
+{
+	int *undwarf_ip = _undwarf_ip;
+	struct undwarf *undwarf = _undwarf;
+	unsigned int num_entries = undwarf_ip_size / sizeof(int);
+
+	WARN_ON_ONCE(undwarf_ip_size % sizeof(int) != 0 ||
+		     undwarf_size % sizeof(*undwarf) != 0 ||
+		     num_entries != undwarf_size / sizeof(*undwarf));
+
+	/*
+	 * The 'cur_undwarf_*' globals allow the undwarf_sort_swap() callback
+	 * to associate an undwarf_ip table entry with its corresponding
+	 * undwarf entry so they can both be swapped.
+	 */
+	mutex_lock(&sort_mutex);
+	cur_undwarf_ip_table = undwarf_ip;
+	cur_undwarf_table = undwarf;
+	sort(undwarf_ip, num_entries, sizeof(int),undwarf_sort_cmp,
+	     undwarf_sort_swap);
+	mutex_unlock(&sort_mutex);
+
+	mod->arch.undwarf_ip = undwarf_ip;
+	mod->arch.undwarf = undwarf;
+	mod->arch.num_undwarves = num_entries;
+}
+
+void __init unwind_init(void)
+{
+	size_t undwarf_ip_size = (void *)__stop_undwarf_ip - (void *)__start_undwarf_ip;
+	size_t undwarf_size = (void *)__stop_undwarf - (void *)__start_undwarf;
+	size_t num_entries = undwarf_ip_size / sizeof(int);
+	struct undwarf *undwarf;
+	int i;
+
+	if (!num_entries || undwarf_ip_size % sizeof(int) != 0 ||
+	    undwarf_size % sizeof(struct undwarf) != 0 ||
+	    num_entries != undwarf_size / sizeof(struct undwarf)) {
+		pr_warn("WARNING: Bad or missing undwarf table.  Disabling unwinder.\n");
+		return;
+	}
+
+	/* Sort the undwarf table: */
+	sort(__start_undwarf_ip, num_entries, sizeof(int), undwarf_sort_cmp,
+	     undwarf_sort_swap);
+
+	/* Initialize the fast lookup table: */
+	for (i = 0; i < LOOKUP_NUM_BLOCKS; i++) {
+		undwarf = __undwarf_find(__start_undwarf_ip, __start_undwarf,
+					 num_entries,
+					 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
+		if (!undwarf) {
+			pr_warn("WARNING: Corrupt undwarf table.  Disabling unwinder.\n");
+			return;
+		}
+
+		undwarf_fast_lookup[i] = undwarf - __start_undwarf;
+	}
+
+	/* Initialize the last 'end' block: */
+	undwarf = __undwarf_find(__start_undwarf_ip, __start_undwarf,
+				 num_entries, LOOKUP_STOP_IP);
+	if (!undwarf) {
+		pr_warn("WARNING: Corrupt undwarf table.  Disabling unwinder.\n");
+		return;
+	}
+	undwarf_fast_lookup[LOOKUP_NUM_BLOCKS] = undwarf - __start_undwarf;
+
+	undwarf_init = true;
+}
+
+unsigned long unwind_get_return_address(struct unwind_state *state)
+{
+	if (unwind_done(state))
+		return 0;
+
+	return __kernel_text_address(state->ip) ? state->ip : 0;
+}
+EXPORT_SYMBOL_GPL(unwind_get_return_address);
+
+unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
+{
+	if (unwind_done(state))
+		return NULL;
+
+	if (state->regs)
+		return &state->regs->ip;
+
+	if (state->sp)
+		return (unsigned long *)state->sp - 1;
+
+	return NULL;
+}
+
+static bool stack_access_ok(struct unwind_state *state, unsigned long addr,
+			    size_t len)
+{
+	struct stack_info *info = &state->stack_info;
+
+	/*
+	 * If the address isn't on the current stack, switch to the next one.
+	 *
+	 * We may have to traverse multiple stacks to deal with the possibility
+	 * that info->next_sp could point to an empty stack and the address
+	 * could be on a subsequent stack.
+	 */
+	while (!on_stack(info, (void *)addr, len))
+		if (get_stack_info(info->next_sp, state->task, info,
+				   &state->stack_mask))
+			return false;
+
+	return true;
+}
+
+static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
+			    unsigned long *val)
+{
+	if (!stack_access_ok(state, addr, sizeof(long)))
+		return false;
+
+	*val = READ_ONCE_TASK_STACK(state->task, *(unsigned long *)addr);
+	return true;
+}
+
+#define REGS_SIZE (sizeof(struct pt_regs))
+#define SP_OFFSET (offsetof(struct pt_regs, sp))
+#define IRET_REGS_SIZE (REGS_SIZE - offsetof(struct pt_regs, ip))
+#define IRET_SP_OFFSET (SP_OFFSET - offsetof(struct pt_regs, ip))
+
+static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
+			     unsigned long *ip, unsigned long *sp, bool full)
+{
+	size_t regs_size = full ? REGS_SIZE : IRET_REGS_SIZE;
+	size_t sp_offset = full ? SP_OFFSET : IRET_SP_OFFSET;
+	struct pt_regs *regs = (struct pt_regs *)(addr + regs_size - REGS_SIZE);
+
+	if (IS_ENABLED(CONFIG_X86_64)) {
+		if (!stack_access_ok(state, addr, regs_size))
+			return false;
+
+		*ip = regs->ip;
+		*sp = regs->sp;
+
+		return true;
+	}
+
+	if (!stack_access_ok(state, addr, sp_offset))
+		return false;
+
+	*ip = regs->ip;
+
+	if (user_mode(regs)) {
+		if (!stack_access_ok(state, addr + sp_offset,
+				     REGS_SIZE - SP_OFFSET))
+			return false;
+
+		*sp = regs->sp;
+	} else
+		*sp = (unsigned long)&regs->sp;
+
+	return true;
+}
+
+bool unwind_next_frame(struct unwind_state *state)
+{
+	enum stack_type prev_type = state->stack_info.type;
+	unsigned long ip_p, prev_sp = state->sp;
+	unsigned long cfa, orig_ip, orig_sp;
+	struct undwarf *undwarf;
+	struct pt_regs *ptregs;
+	bool indirect = false;
+
+	if (unwind_done(state))
+		return false;
+
+	/* Don't let modules unload while we're reading their undwarf data. */
+	preempt_disable();
+
+	/* Have we reached the end? */
+	if (state->regs && user_mode(state->regs))
+		goto done;
+
+	/*
+	 * Find the undwarf table entry associated with the text address.
+	 *
+	 * Decrement call return addresses by one so they work for sibling
+	 * calls and calls to noreturn functions.
+	 */
+	undwarf = undwarf_find(state->signal ? state->ip : state->ip - 1);
+	if (!undwarf || undwarf->cfa_reg == UNDWARF_REG_UNDEFINED)
+		goto done;
+	orig_ip = state->ip;
+
+	/* Calculate the CFA (caller frame address): */
+	switch (undwarf->cfa_reg) {
+	case UNDWARF_REG_SP:
+		cfa = state->sp + undwarf->cfa_offset;
+		break;
+
+	case UNDWARF_REG_BP:
+		cfa = state->bp + undwarf->cfa_offset;
+		break;
+
+	case UNDWARF_REG_SP_INDIRECT:
+		cfa = state->sp + undwarf->cfa_offset;
+		indirect = true;
+		break;
+
+	case UNDWARF_REG_BP_INDIRECT:
+		cfa = state->bp + undwarf->cfa_offset;
+		indirect = true;
+		break;
+
+	case UNDWARF_REG_R10:
+		if (!state->regs || !state->full_regs) {
+			undwarf_warn("missing regs for base reg R10 at ip %p\n",
+				     (void *)state->ip);
+			goto done;
+		}
+		cfa = state->regs->r10;
+		break;
+
+	case UNDWARF_REG_R13:
+		if (!state->regs || !state->full_regs) {
+			undwarf_warn("missing regs for base reg R13 at ip %p\n",
+				     (void *)state->ip);
+			goto done;
+		}
+		cfa = state->regs->r13;
+		break;
+
+	case UNDWARF_REG_DI:
+		if (!state->regs || !state->full_regs) {
+			undwarf_warn("missing regs for base reg DI at ip %p\n",
+				     (void *)state->ip);
+			goto done;
+		}
+		cfa = state->regs->di;
+		break;
+
+	case UNDWARF_REG_DX:
+		if (!state->regs || !state->full_regs) {
+			undwarf_warn("missing regs for base reg DX at ip %p\n",
+				     (void *)state->ip);
+			goto done;
+		}
+		cfa = state->regs->dx;
+		break;
+
+	default:
+		undwarf_warn("unknown CFA base reg %d for ip %p\n",
+			     undwarf->cfa_reg, (void *)state->ip);
+		goto done;
+	}
+
+	if (indirect) {
+		if (!deref_stack_reg(state, cfa, &cfa))
+			goto done;
+	}
+
+	/* Find IP, SP and possibly regs: */
+	switch (undwarf->type) {
+	case UNDWARF_TYPE_CFA:
+		ip_p = cfa - sizeof(long);
+
+		if (!deref_stack_reg(state, ip_p, &state->ip))
+			goto done;
+
+		state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
+						  state->ip, (void *)ip_p);
+
+		state->sp = cfa;
+		state->regs = NULL;
+		state->signal = false;
+		break;
+
+	case UNDWARF_TYPE_REGS:
+		if (!deref_stack_regs(state, cfa, &state->ip, &state->sp, true)) {
+			undwarf_warn("can't dereference registers at %p for ip %p\n",
+				     (void *)cfa, (void *)orig_ip);
+			goto done;
+		}
+
+		state->regs = (struct pt_regs *)cfa;
+		state->full_regs = true;
+		state->signal = true;
+		break;
+
+	case UNDWARF_TYPE_REGS_IRET:
+		orig_sp = state->sp;
+		if (!deref_stack_regs(state, cfa, &state->ip, &state->sp, false)) {
+			undwarf_warn("can't dereference iret registers at %p for ip %p\n",
+				     (void *)cfa, (void *)orig_ip);
+			goto done;
+		}
+
+		ptregs = container_of((void *)cfa, struct pt_regs, ip);
+		if ((unsigned long)ptregs >= orig_sp &&
+		    on_stack(&state->stack_info, ptregs, REGS_SIZE)) {
+			state->regs = ptregs;
+			state->full_regs = false;
+		} else
+			state->regs = NULL;
+
+		state->signal = true;
+		break;
+
+	default:
+		undwarf_warn("unknown undwarf type %d\n", undwarf->type);
+		break;
+	}
+
+	/* Find BP: */
+	switch (undwarf->bp_reg) {
+	case UNDWARF_REG_UNDEFINED:
+		if (state->regs && state->full_regs)
+			state->bp = state->regs->bp;
+		break;
+
+	case UNDWARF_REG_CFA:
+		if (!deref_stack_reg(state, cfa + undwarf->bp_offset,&state->bp))
+			goto done;
+		break;
+
+	case UNDWARF_REG_BP:
+		if (!deref_stack_reg(state, state->bp + undwarf->bp_offset, &state->bp))
+			goto done;
+		break;
+
+	default:
+		undwarf_warn("unknown BP base reg %d for ip %p\n",
+			     undwarf->bp_reg, (void *)orig_ip);
+		goto done;
+	}
+
+	/* Prevent a recursive loop due to bad undwarf data: */
+	if (state->stack_info.type == prev_type &&
+	    on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
+	    state->sp <= prev_sp) {
+		undwarf_warn("stack going in the wrong direction? ip=%p\n",
+			     (void *)orig_ip);
+		goto done;
+	}
+
+	preempt_enable();
+	return true;
+
+done:
+	preempt_enable();
+	state->stack_info.type = STACK_TYPE_UNKNOWN;
+	return false;
+}
+EXPORT_SYMBOL_GPL(unwind_next_frame);
+
+void __unwind_start(struct unwind_state *state, struct task_struct *task,
+		    struct pt_regs *regs, unsigned long *first_frame)
+{
+	memset(state, 0, sizeof(*state));
+	state->task = task;
+
+	/*
+	 * Refuse to unwind the stack of a task while it's executing on another
+	 * CPU.  This check is racy, but that's ok: the unwinder has other
+	 * checks to prevent it from going off the rails.
+	 */
+	if (task_on_another_cpu(task))
+		goto done;
+
+	if (regs) {
+		if (user_mode(regs))
+			goto done;
+
+		state->ip = regs->ip;
+		state->sp = kernel_stack_pointer(regs);
+		state->bp = regs->bp;
+		state->regs = regs;
+		state->full_regs = true;
+		state->signal = true;
+
+	} else if (task == current) {
+		asm volatile("lea (%%rip), %0\n\t"
+			     "mov %%rsp, %1\n\t"
+			     "mov %%rbp, %2\n\t"
+			     : "=r" (state->ip), "=r" (state->sp),
+			       "=r" (state->bp));
+
+	} else {
+		struct inactive_task_frame *frame = (void *)task->thread.sp;
+
+		state->ip = frame->ret_addr;
+		state->sp = task->thread.sp;
+		state->bp = frame->bp;
+	}
+
+	if (get_stack_info((unsigned long *)state->sp, state->task,
+			   &state->stack_info, &state->stack_mask))
+		return;
+
+	/*
+	 * The caller can provide the address of the first frame directly
+	 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
+	 * to start unwinding at.  Skip ahead until we reach it.
+	 */
+	while (!unwind_done(state) &&
+	       (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
+			state->sp <= (unsigned long)first_frame))
+		unwind_next_frame(state);
+
+	return;
+
+done:
+	state->stack_info.type = STACK_TYPE_UNKNOWN;
+	return;
+}
+EXPORT_SYMBOL_GPL(__unwind_start);
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
index c8a3b61..e3b7cfc 100644
--- a/arch/x86/kernel/vmlinux.lds.S
+++ b/arch/x86/kernel/vmlinux.lds.S
@@ -148,6 +148,8 @@ SECTIONS
 
 	BUG_TABLE
 
+	UNDWARF_TABLE
+
 	. = ALIGN(PAGE_SIZE);
 	__vvar_page = .;
 
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 0d64658..a8ed616 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -668,6 +668,24 @@
 #define BUG_TABLE
 #endif
 
+#ifdef CONFIG_UNDWARF_UNWINDER
+#define UNDWARF_TABLE							\
+	. = ALIGN(4);							\
+	.undwarf_ip : AT(ADDR(.undwarf_ip) - LOAD_OFFSET) {		\
+		VMLINUX_SYMBOL(__start_undwarf_ip) = .;			\
+		KEEP(*(.undwarf_ip))					\
+		VMLINUX_SYMBOL(__stop_undwarf_ip) = .;			\
+	}								\
+	. = ALIGN(8);							\
+	.undwarf : AT(ADDR(.undwarf) - LOAD_OFFSET) {			\
+		VMLINUX_SYMBOL(__start_undwarf) = .;			\
+		KEEP(*(.undwarf))					\
+		VMLINUX_SYMBOL(__stop_undwarf) = .;			\
+	}
+#else
+#define UNDWARF_TABLE
+#endif
+
 #ifdef CONFIG_PM_TRACE
 #define TRACEDATA							\
 	. = ALIGN(4);							\
@@ -854,7 +872,7 @@
 		DATA_DATA						\
 		CONSTRUCTORS						\
 	}								\
-	BUG_TABLE
+	BUG_TABLE							\
 
 #define INIT_TEXT_SECTION(inittext_align)				\
 	. = ALIGN(inittext_align);					\
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 9c5d40a..ec79366 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -374,6 +374,9 @@ config STACK_VALIDATION
 	  pointers (if CONFIG_FRAME_POINTER is enabled).  This helps ensure
 	  that runtime stack traces are more reliable.
 
+	  This is also a prerequisite for creation of the undwarf format which
+	  is needed for CONFIG_UNDWARF_UNWINDER.
+
 	  For more information, see
 	  tools/objtool/Documentation/stack-validation.txt.
 
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 733e044..7859c79 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -258,7 +258,8 @@ ifneq ($(SKIP_STACK_VALIDATION),1)
 
 __objtool_obj := $(objtree)/tools/objtool/objtool
 
-objtool_args = check
+objtool_args = $(if $(CONFIG_UNDWARF_UNWINDER),undwarf generate,check)
+
 ifndef CONFIG_FRAME_POINTER
 objtool_args += --no-fp
 endif
@@ -276,6 +277,11 @@ objtool_obj = $(if $(patsubst y%,, \
 endif # SKIP_STACK_VALIDATION
 endif # CONFIG_STACK_VALIDATION
 
+# Rebuild all objects when objtool changes, or is enabled/disabled.
+objtool_dep = $(objtool_obj)					\
+	      $(wildcard include/config/undwarf/unwinder.h	\
+			 include/config/stack/validation.h)
+
 define rule_cc_o_c
 	$(call echo-cmd,checksrc) $(cmd_checksrc)			  \
 	$(call cmd_and_fixdep,cc_o_c)					  \
@@ -298,13 +304,13 @@ cmd_undef_syms = echo
 endif
 
 # Built-in and composite module parts
-$(obj)/%.o: $(src)/%.c $(recordmcount_source) $(objtool_obj) FORCE
+$(obj)/%.o: $(src)/%.c $(recordmcount_source) $(objtool_dep) FORCE
 	$(call cmd,force_checksrc)
 	$(call if_changed_rule,cc_o_c)
 
 # Single-part modules are special since we need to mark them in $(MODVERDIR)
 
-$(single-used-m): $(obj)/%.o: $(src)/%.c $(recordmcount_source) $(objtool_obj) FORCE
+$(single-used-m): $(obj)/%.o: $(src)/%.c $(recordmcount_source) $(objtool_dep) FORCE
 	$(call cmd,force_checksrc)
 	$(call if_changed_rule,cc_o_c)
 	@{ echo $(@:.o=.ko); echo $@; \
@@ -399,7 +405,7 @@ cmd_modversions_S =								\
 endif
 endif
 
-$(obj)/%.o: $(src)/%.S $(objtool_obj) FORCE
+$(obj)/%.o: $(src)/%.S $(objtool_dep) FORCE
 	$(call if_changed_rule,as_o_S)
 
 targets += $(real-objs-y) $(real-objs-m) $(lib-y)
-- 
2.7.5