[RFC PATCH 00/13] mm/kwatch: dynamic hardware watchpoints for hunting memory corruption

[RFC PATCH 00/13] mm/kwatch: dynamic hardware watchpoints for hunting memory corruption

[Jinchao Wang]

Motivation
==========

The hardest memory corruption bugs are the silent ones: a rogue writer
scribbles over a live object through a stale pointer or a race, and
the victim crashes in a code path far away from the culprit. Any
single developer hits such a bug rarely, but across the kernel's code
base and install base they keep arriving, and each one is
disproportionately expensive to localize. The existing tools report
the symptom, not the writer:

 - KASAN/KFENCE catch invalid accesses, but a targeted use-after-free
   or an in-bounds logical overwrite (a *valid* pointer written at the
   wrong time) never violates memory safety, so they stay silent - and
   KASAN's rebuild, overhead and redzones often perturb the bug away.
 - Hardware watchpoints via kgdb or perf can watch one fixed address,
   system-wide, for the whole run. But the interesting address is
   usually per-object and per-invocation ("field X of the object this
   function is currently operating on"), which cannot be expressed.

Design
======

KWatch implements two key mechanisms: a function-scoped watch window
that decides when the hardware breakpoint is armed and released, and
a flexible expression engine that resolves which address it guards.
A watch is configured through debugfs with a single line; hits are
reported through a tracepoint, carrying the writing instruction and
a stack trace.

The window: a kretprobe pair opens the watch at function entry and
closes it on return; inside, a per-CPU hardware breakpoint from a
preallocated pool is re-pointed at the target address. The pool is
managed locklessly, so a window can open in whatever context the
watched function runs in - real NMI excepted - and a hit can fire
and be handled in any context.

The window is also what makes the scarce hardware affordable: x86
has only four hardware breakpoint slots per CPU, and every corruption
happens within some execution context, so a breakpoint is armed only
while that context runs. The rest of the system runs at full speed and
only the watched function pays the kprobe cost, which keeps KWatch
usable on busy, highly concurrent systems. Global variables can also
be watched without a window, in a time-bounded anchor session.

The expression engine: at each entry it evaluates the configured
watch expression to resolve that invocation's target address. The
base can be a function argument, the stack pointer, a symbol or an
absolute address; offsets and pointer dereferences chain on top, so
heap fields reachable from an argument, globals and stack slots are
all expressible - no objdump session needed.

KWatch is also designed for painless deployment: it is fully
self-contained and can be built as a module, loaded only when a
corruption hunt needs it. It is just a debugfs entry until a watch
is configured, then just a kprobe on the watched function, with the
breakpoint armed only when needed.

A real case: dummy_hcd
======================

Gadget requests were completing through a clobbered req->complete.
Months of KASAN-enabled syzkaller runs produced only downstream
symptoms, with no lead on the root cause. Watching the victim field
with KWatch:

  func_name=usb_gadget_giveback_request watch_expr=arg2+56 \
  watch_len=8 access_type=0

caught the writer in the act:

  kwatch_hit: KWatch HIT: time=370.399836 ip=memcpy+0xc/0x30
              addr=0xffff888109cf5218
   => usb_ep_queue+0xf1/0x3c0
   => raw_process_ep_io+0x5e4/0xd80
   => raw_ioctl+0x251c/0x41c0
   => __se_sys_ioctl+0xfc/0x170
   => do_syscall_64+0x174/0x580
   => entry_SYSCALL_64_after_hwframe+0x77/0x7f

on the same request that crashed an instant later - the crash RIP was
the just-written garbage value. Root cause: dummy_queue()'s single
shared fifo_req is struct-copied over while dummy_timer() is
mid-giveback. Neither a use-after-free nor list corruption, so KASAN
and CONFIG_DEBUG_LIST are blind to it by design. A fix based on this
diagnosis is under review [1] - KWatch's part was pinpointing the
root cause: who clobbers the pointer, and from where.

Series layout
=============

Patches 1-4: a minimal "reinstall" operation for hw_breakpoint.
Re-pointing an already-installed breakpoint from a kprobe handler is
not possible with the current API (register/unregister may sleep and
rebalances constraints); reinstall lets the arch rewrite a slot it
already owns, and modify_wide_hw_breakpoint_local() exposes that for
the local CPU - cross-CPU propagation is the caller's job (KWatch
uses async IPIs). Patch 4 is Masami Hiramatsu's work, carried
unchanged.

Patches 5-11: KWatch itself, in mm/kwatch/ (patch 7 exports
stack_trace_save_regs() for the modular build).

Patches 12-13: KUnit tests and documentation.

Testing
=======

The dummy_hcd hunt above exercised the function-window path against
a live reproducer. Global watching, session auto-stop and the KUnit
parser suite were verified end to end under QEMU on x86_64. Both
KWATCH=y and KWATCH=m build.

arm64
=====

This RFC deliberately targets x86 only. On arm64 the watchpoint
exception fires before the access, so the arch must single-step over
hits, and today it only does that for the default overflow handler.
Rather than hardcoding a KWatch hook into arm64 core code, I plan a
follow-up that adds a generic way for in-kernel breakpoint consumers
to request stepping, and arm64 support on top of it (a prototype
exists).

Relationship to KStackWatch
===========================

KWatch grew out of KStackWatch [2], an earlier tool aimed at stack
corruption only, and has been substantially reworked since. The
hw_breakpoint prerequisites are carried over from that series.

Major changes since the KStackWatch v8 posting:

 - The watch expression engine widens the watchable range from the
   stack to any address expressible via function arguments, globals
   or stack addresses plus pointer dereference chains.
 - The task_struct and scheduler hooks are gone; KWatch is now fully
   self-contained, as described above.
 - A time-bounded anchor context was added for watching global
   variables (duration=N, auto-stop on expiry).
 - Hits are reported through a tracepoint carrying a stack trace
   instead of printk: safe in NMI-like contexts, and recoverable
   after a crash (ftrace_dump_on_oops, kdump, pstore).
 - Invocations in real NMI(-like) context are detected and rejected,
   with a visible nmi_rejected counter.
 - arm64 support and the auto-canary, profiling and test-module
   extras were dropped from this first posting to keep it reviewable.

Feedback on the design, the implementation or the usage is welcome;
if you are staring at a corruption that KASAN and friends cannot
attribute, give KWatch a try, or simply Cc me - I am glad to help.

[1] https://lore.kernel.org/all/20260714064829.172098-1-wangjinchao600@gmail.com/
[2] https://lore.kernel.org/all/20251110163634.3686676-1-wangjinchao600@gmail.com/

Jinchao Wang (12):
  arch: add HAVE_REINSTALL_HW_BREAKPOINT
  x86/hw_breakpoint: Unify breakpoint install/uninstall
  x86/hw_breakpoint: Add arch_reinstall_hw_breakpoint
  mm/kwatch: add watch expression parser and dereference engine
  mm/kwatch: add lockless per-task context pool
  stacktrace: export stack_trace_save_regs()
  mm/kwatch: add hardware breakpoint backend
  mm/kwatch: add probe lifecycle runtime
  mm/kwatch: add anchor thread for global watchpoints
  mm/kwatch: add debugfs control plane
  mm/kwatch: add KUnit tests for the watch expression parser
  Documentation/dev-tools: document KWatch

Masami Hiramatsu (Google) (1):
  HWBP: Add modify_wide_hw_breakpoint_local() API

 Documentation/dev-tools/index.rst    |   1 +
 Documentation/dev-tools/kwatch.rst   | 193 +++++++++++++++
 MAINTAINERS                          |   8 +
 arch/Kconfig                         |  10 +
 arch/x86/Kconfig                     |   1 +
 arch/x86/include/asm/hw_breakpoint.h |   8 +
 arch/x86/kernel/hw_breakpoint.c      | 151 ++++++-----
 include/linux/hw_breakpoint.h        |   6 +
 include/trace/events/kwatch.h        |  57 +++++
 kernel/events/hw_breakpoint.c        |  37 +++
 kernel/stacktrace.c                  |   2 +
 mm/Kconfig                           |   1 +
 mm/Makefile                          |   1 +
 mm/kwatch/.kunitconfig               |   9 +
 mm/kwatch/Kconfig                    |  27 ++
 mm/kwatch/Makefile                   |   4 +
 mm/kwatch/anchor.c                   |  82 ++++++
 mm/kwatch/core.c                     | 325 ++++++++++++++++++++++++
 mm/kwatch/deref.c                    | 174 +++++++++++++
 mm/kwatch/deref_test.c               | 137 ++++++++++
 mm/kwatch/hwbp.c                     | 358 +++++++++++++++++++++++++++
 mm/kwatch/kwatch.h                   | 107 ++++++++
 mm/kwatch/probe.c                    | 263 ++++++++++++++++++++
 mm/kwatch/task_ctx.c                 | 105 ++++++++
 24 files changed, 2005 insertions(+), 62 deletions(-)
 create mode 100644 Documentation/dev-tools/kwatch.rst
 create mode 100644 include/trace/events/kwatch.h
 create mode 100644 mm/kwatch/.kunitconfig
 create mode 100644 mm/kwatch/Kconfig
 create mode 100644 mm/kwatch/Makefile
 create mode 100644 mm/kwatch/anchor.c
 create mode 100644 mm/kwatch/core.c
 create mode 100644 mm/kwatch/deref.c
 create mode 100644 mm/kwatch/deref_test.c
 create mode 100644 mm/kwatch/hwbp.c
 create mode 100644 mm/kwatch/kwatch.h
 create mode 100644 mm/kwatch/probe.c
 create mode 100644 mm/kwatch/task_ctx.c

-- 
2.53.0

[RFC PATCH 01/13] arch: add HAVE_REINSTALL_HW_BREAKPOINT

[Jinchao Wang]

Some architectures can update the address, length or type of an
installed hardware breakpoint in place, without releasing and
re-reserving its slot. Add an opt-in Kconfig symbol so generic code
can offer such an operation on architectures that implement
arch_reinstall_hw_breakpoint().

This is a prerequisite for KWatch, which re-points preallocated
per-CPU breakpoints from atomic context, where the register/release
path (which may sleep and rebalances slot constraints) cannot be
used.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 arch/Kconfig | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/arch/Kconfig b/arch/Kconfig
index fa7507ac8e13..41b3784e0ddd 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -457,6 +457,16 @@ config HAVE_MIXED_BREAKPOINTS_REGS
 	  Select this option if your arch implements breakpoints under the
 	  latter fashion.
 
+config HAVE_REINSTALL_HW_BREAKPOINT
+	bool
+	depends on HAVE_HW_BREAKPOINT
+	help
+	  Depending on the arch implementation of hardware breakpoints,
+	  some of them are able to update the breakpoint configuration
+	  without release and reserve the hardware breakpoint register.
+	  What configuration is able to update depends on hardware and
+	  software implementation.
+
 config HAVE_USER_RETURN_NOTIFIER
 	bool
 
-- 
2.53.0

[RFC PATCH 02/13] x86/hw_breakpoint: Unify breakpoint install/uninstall

[Jinchao Wang]

Consolidate breakpoint management to reduce code duplication.
The diffstat was misleading, so the stripped code size is compared instead.
After refactoring, it is reduced from 11976 bytes to 11448 bytes on my
x86_64 system built with clang.

This also makes it easier to introduce arch_reinstall_hw_breakpoint().

In addition, including linux/types.h to fix a missing build dependency.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 arch/x86/include/asm/hw_breakpoint.h |   6 ++
 arch/x86/kernel/hw_breakpoint.c      | 141 +++++++++++++++------------
 2 files changed, 84 insertions(+), 63 deletions(-)

diff --git a/arch/x86/include/asm/hw_breakpoint.h b/arch/x86/include/asm/hw_breakpoint.h
index 0bc931cd0698..aa6adac6c3a2 100644
--- a/arch/x86/include/asm/hw_breakpoint.h
+++ b/arch/x86/include/asm/hw_breakpoint.h
@@ -5,6 +5,7 @@
 #include <uapi/asm/hw_breakpoint.h>
 
 #define	__ARCH_HW_BREAKPOINT_H
+#include <linux/types.h>
 
 /*
  * The name should probably be something dealt in
@@ -18,6 +19,11 @@ struct arch_hw_breakpoint {
 	u8		type;
 };
 
+enum bp_slot_action {
+	BP_SLOT_ACTION_INSTALL,
+	BP_SLOT_ACTION_UNINSTALL,
+};
+
 #include <linux/kdebug.h>
 #include <linux/percpu.h>
 #include <linux/list.h>
diff --git a/arch/x86/kernel/hw_breakpoint.c b/arch/x86/kernel/hw_breakpoint.c
index f846c15f21ca..877509539300 100644
--- a/arch/x86/kernel/hw_breakpoint.c
+++ b/arch/x86/kernel/hw_breakpoint.c
@@ -49,7 +49,6 @@ static DEFINE_PER_CPU(unsigned long, cpu_debugreg[HBP_NUM]);
  */
 static DEFINE_PER_CPU(struct perf_event *, bp_per_reg[HBP_NUM]);
 
-
 static inline unsigned long
 __encode_dr7(int drnum, unsigned int len, unsigned int type)
 {
@@ -86,96 +85,112 @@ int decode_dr7(unsigned long dr7, int bpnum, unsigned *len, unsigned *type)
 }
 
 /*
- * Install a perf counter breakpoint.
- *
- * We seek a free debug address register and use it for this
- * breakpoint. Eventually we enable it in the debug control register.
- *
- * Atomic: we hold the counter->ctx->lock and we only handle variables
- * and registers local to this cpu.
+ * We seek a slot and change it or keep it based on the action.
+ * Returns slot number on success, negative error on failure.
+ * Must be called with IRQs disabled.
  */
-int arch_install_hw_breakpoint(struct perf_event *bp)
+static int manage_bp_slot(struct perf_event *bp, enum bp_slot_action action)
 {
-	struct arch_hw_breakpoint *info = counter_arch_bp(bp);
-	unsigned long *dr7;
-	int i;
-
-	lockdep_assert_irqs_disabled();
+	struct perf_event *old_bp;
+	struct perf_event *new_bp;
+	int slot;
+
+	switch (action) {
+	case BP_SLOT_ACTION_INSTALL:
+		old_bp = NULL;
+		new_bp = bp;
+		break;
+	case BP_SLOT_ACTION_UNINSTALL:
+		old_bp = bp;
+		new_bp = NULL;
+		break;
+	default:
+		return -EINVAL;
+	}
 
-	for (i = 0; i < HBP_NUM; i++) {
-		struct perf_event **slot = this_cpu_ptr(&bp_per_reg[i]);
+	for (slot = 0; slot < HBP_NUM; slot++) {
+		struct perf_event **curr = this_cpu_ptr(&bp_per_reg[slot]);
 
-		if (!*slot) {
-			*slot = bp;
-			break;
+		if (*curr == old_bp) {
+			*curr = new_bp;
+			return slot;
 		}
 	}
 
-	if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
-		return -EBUSY;
+	if (old_bp) {
+		WARN_ONCE(1, "Can't find matching breakpoint slot");
+		return -EINVAL;
+	}
+
+	WARN_ONCE(1, "No free breakpoint slots");
+	return -EBUSY;
+}
+
+static void setup_hwbp(struct arch_hw_breakpoint *info, int slot, bool enable)
+{
+	unsigned long dr7;
 
-	set_debugreg(info->address, i);
-	__this_cpu_write(cpu_debugreg[i], info->address);
+	set_debugreg(info->address, slot);
+	__this_cpu_write(cpu_debugreg[slot], info->address);
 
-	dr7 = this_cpu_ptr(&cpu_dr7);
-	*dr7 |= encode_dr7(i, info->len, info->type);
+	dr7 = this_cpu_read(cpu_dr7);
+	if (enable)
+		dr7 |= encode_dr7(slot, info->len, info->type);
+	else
+		dr7 &= ~__encode_dr7(slot, info->len, info->type);
 
 	/*
-	 * Ensure we first write cpu_dr7 before we set the DR7 register.
-	 * This ensures an NMI never see cpu_dr7 0 when DR7 is not.
+	 * Enabling:
+	 *   Ensure we first write cpu_dr7 before we set the DR7 register.
+	 *   This ensures an NMI never see cpu_dr7 0 when DR7 is not.
 	 */
+	if (enable)
+		this_cpu_write(cpu_dr7, dr7);
+
 	barrier();
 
-	set_debugreg(*dr7, 7);
+	set_debugreg(dr7, 7);
+
 	if (info->mask)
-		amd_set_dr_addr_mask(info->mask, i);
+		amd_set_dr_addr_mask(enable ? info->mask : 0, slot);
 
-	return 0;
+	/*
+	 * Disabling:
+	 *   Ensure the write to cpu_dr7 is after we've set the DR7 register.
+	 *   This ensures an NMI never see cpu_dr7 0 when DR7 is not.
+	 */
+	if (!enable)
+		this_cpu_write(cpu_dr7, dr7);
 }
 
 /*
- * Uninstall the breakpoint contained in the given counter.
- *
- * First we search the debug address register it uses and then we disable
- * it.
- *
- * Atomic: we hold the counter->ctx->lock and we only handle variables
- * and registers local to this cpu.
+ * find suitable breakpoint slot and set it up based on the action
  */
-void arch_uninstall_hw_breakpoint(struct perf_event *bp)
+static int arch_manage_bp(struct perf_event *bp, enum bp_slot_action action)
 {
-	struct arch_hw_breakpoint *info = counter_arch_bp(bp);
-	unsigned long dr7;
-	int i;
+	struct arch_hw_breakpoint *info;
+	int slot;
 
 	lockdep_assert_irqs_disabled();
 
-	for (i = 0; i < HBP_NUM; i++) {
-		struct perf_event **slot = this_cpu_ptr(&bp_per_reg[i]);
-
-		if (*slot == bp) {
-			*slot = NULL;
-			break;
-		}
-	}
-
-	if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
-		return;
+	slot = manage_bp_slot(bp, action);
+	if (slot < 0)
+		return slot;
 
-	dr7 = this_cpu_read(cpu_dr7);
-	dr7 &= ~__encode_dr7(i, info->len, info->type);
+	info = counter_arch_bp(bp);
+	setup_hwbp(info, slot, action != BP_SLOT_ACTION_UNINSTALL);
 
-	set_debugreg(dr7, 7);
-	if (info->mask)
-		amd_set_dr_addr_mask(0, i);
+	return 0;
+}
 
-	/*
-	 * Ensure the write to cpu_dr7 is after we've set the DR7 register.
-	 * This ensures an NMI never see cpu_dr7 0 when DR7 is not.
-	 */
-	barrier();
+int arch_install_hw_breakpoint(struct perf_event *bp)
+{
+	return arch_manage_bp(bp, BP_SLOT_ACTION_INSTALL);
+}
 
-	this_cpu_write(cpu_dr7, dr7);
+void arch_uninstall_hw_breakpoint(struct perf_event *bp)
+{
+	arch_manage_bp(bp, BP_SLOT_ACTION_UNINSTALL);
 }
 
 static int arch_bp_generic_len(int x86_len)
-- 
2.53.0

[RFC PATCH 03/13] x86/hw_breakpoint: Add arch_reinstall_hw_breakpoint

[Jinchao Wang]

The new arch_reinstall_hw_breakpoint() function can be used in an
atomic context, unlike the more expensive free and re-allocation path.
This allows callers to efficiently re-establish an existing breakpoint,
and x86 advertises the capability via HAVE_REINSTALL_HW_BREAKPOINT.

Since a REINSTALL may change bp_len, setup_hwbp() must clear the
slot's stale len/type and enable bits in DR7 before re-encoding:
OR-merging the new encoding over the old one would keep the CPU
watching with the stale width (verified in QEMU by reading DR7 after
re-arming watch_len=1 over a len8 breakpoint: 0x999906aa merged
without the clearing, 0x199906aa with it).

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 arch/x86/Kconfig                     |  1 +
 arch/x86/include/asm/hw_breakpoint.h |  2 ++
 arch/x86/kernel/hw_breakpoint.c      | 16 ++++++++++++++--
 3 files changed, 17 insertions(+), 2 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index bdad90f210e4..5be698db0241 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -246,6 +246,7 @@ config X86
 	select HAVE_FUNCTION_TRACER
 	select HAVE_GCC_PLUGINS
 	select HAVE_HW_BREAKPOINT
+	select HAVE_REINSTALL_HW_BREAKPOINT
 	select HAVE_IOREMAP_PROT
 	select HAVE_IRQ_EXIT_ON_IRQ_STACK	if X86_64
 	select HAVE_IRQ_TIME_ACCOUNTING
diff --git a/arch/x86/include/asm/hw_breakpoint.h b/arch/x86/include/asm/hw_breakpoint.h
index aa6adac6c3a2..c22cc4e87fc5 100644
--- a/arch/x86/include/asm/hw_breakpoint.h
+++ b/arch/x86/include/asm/hw_breakpoint.h
@@ -21,6 +21,7 @@ struct arch_hw_breakpoint {
 
 enum bp_slot_action {
 	BP_SLOT_ACTION_INSTALL,
+	BP_SLOT_ACTION_REINSTALL,
 	BP_SLOT_ACTION_UNINSTALL,
 };
 
@@ -65,6 +66,7 @@ extern int hw_breakpoint_exceptions_notify(struct notifier_block *unused,
 
 
 int arch_install_hw_breakpoint(struct perf_event *bp);
+int arch_reinstall_hw_breakpoint(struct perf_event *bp);
 void arch_uninstall_hw_breakpoint(struct perf_event *bp);
 void hw_breakpoint_pmu_read(struct perf_event *bp);
 void hw_breakpoint_pmu_unthrottle(struct perf_event *bp);
diff --git a/arch/x86/kernel/hw_breakpoint.c b/arch/x86/kernel/hw_breakpoint.c
index 877509539300..4221dbb899f9 100644
--- a/arch/x86/kernel/hw_breakpoint.c
+++ b/arch/x86/kernel/hw_breakpoint.c
@@ -100,6 +100,10 @@ static int manage_bp_slot(struct perf_event *bp, enum bp_slot_action action)
 		old_bp = NULL;
 		new_bp = bp;
 		break;
+	case BP_SLOT_ACTION_REINSTALL:
+		old_bp = bp;
+		new_bp = bp;
+		break;
 	case BP_SLOT_ACTION_UNINSTALL:
 		old_bp = bp;
 		new_bp = NULL;
@@ -134,10 +138,13 @@ static void setup_hwbp(struct arch_hw_breakpoint *info, int slot, bool enable)
 	__this_cpu_write(cpu_debugreg[slot], info->address);
 
 	dr7 = this_cpu_read(cpu_dr7);
+	/*
+	 * Clear the slot's stale len/type and enable bits first: a REINSTALL
+	 * with a different bp_len would otherwise OR-merge both encodings.
+	 */
+	dr7 &= ~__encode_dr7(slot, 0xf, 0);
 	if (enable)
 		dr7 |= encode_dr7(slot, info->len, info->type);
-	else
-		dr7 &= ~__encode_dr7(slot, info->len, info->type);
 
 	/*
 	 * Enabling:
@@ -188,6 +195,11 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 	return arch_manage_bp(bp, BP_SLOT_ACTION_INSTALL);
 }
 
+int arch_reinstall_hw_breakpoint(struct perf_event *bp)
+{
+	return arch_manage_bp(bp, BP_SLOT_ACTION_REINSTALL);
+}
+
 void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 {
 	arch_manage_bp(bp, BP_SLOT_ACTION_UNINSTALL);
-- 
2.53.0

[RFC PATCH 04/13] HWBP: Add modify_wide_hw_breakpoint_local() API

[Jinchao Wang]

From: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>

Add modify_wide_hw_breakpoint_local() arch-wide interface which allows
hwbp users to update watch address on-line. This is available if the
arch supports CONFIG_HAVE_REINSTALL_HW_BREAKPOINT.
Note that this allows to change the type only for compatible types,
because it does not release and reserve the hwbp slot based on type.
For instance, you can not change HW_BREAKPOINT_W to HW_BREAKPOINT_X.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 include/linux/hw_breakpoint.h |  6 ++++++
 kernel/events/hw_breakpoint.c | 37 +++++++++++++++++++++++++++++++++++
 2 files changed, 43 insertions(+)

diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h
index db199d653dd1..ea373f2587f8 100644
--- a/include/linux/hw_breakpoint.h
+++ b/include/linux/hw_breakpoint.h
@@ -81,6 +81,9 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			    perf_overflow_handler_t triggered,
 			    void *context);
 
+extern int modify_wide_hw_breakpoint_local(struct perf_event *bp,
+					   struct perf_event_attr *attr);
+
 extern int register_perf_hw_breakpoint(struct perf_event *bp);
 extern void unregister_hw_breakpoint(struct perf_event *bp);
 extern void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events);
@@ -124,6 +127,9 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			    perf_overflow_handler_t triggered,
 			    void *context)		{ return NULL; }
 static inline int
+modify_wide_hw_breakpoint_local(struct perf_event *bp,
+				struct perf_event_attr *attr) { return -ENOSYS; }
+static inline int
 register_perf_hw_breakpoint(struct perf_event *bp)	{ return -ENOSYS; }
 static inline void unregister_hw_breakpoint(struct perf_event *bp)	{ }
 static inline void
diff --git a/kernel/events/hw_breakpoint.c b/kernel/events/hw_breakpoint.c
index 789add0c185a..20ca64f30508 100644
--- a/kernel/events/hw_breakpoint.c
+++ b/kernel/events/hw_breakpoint.c
@@ -888,6 +888,43 @@ void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)
 }
 EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint);
 
+/**
+ * modify_wide_hw_breakpoint_local - update breakpoint config for local CPU
+ * @bp: the hwbp perf event for this CPU
+ * @attr: the new attribute for @bp
+ *
+ * This does not release and reserve the slot of a HWBP; it just reuses the
+ * current slot on local CPU. So the users must update the other CPUs by
+ * themselves.
+ * Also, since this does not release/reserve the slot, this can not change the
+ * type to incompatible type of the HWBP.
+ * Return err if attr is invalid or the CPU fails to update debug register
+ * for new @attr.
+ */
+#ifdef CONFIG_HAVE_REINSTALL_HW_BREAKPOINT
+int modify_wide_hw_breakpoint_local(struct perf_event *bp,
+				    struct perf_event_attr *attr)
+{
+	int ret;
+
+	if (find_slot_idx(bp->attr.bp_type) != find_slot_idx(attr->bp_type))
+		return -EINVAL;
+
+	ret = hw_breakpoint_arch_parse(bp, attr, counter_arch_bp(bp));
+	if (ret)
+		return ret;
+
+	return arch_reinstall_hw_breakpoint(bp);
+}
+#else
+int modify_wide_hw_breakpoint_local(struct perf_event *bp,
+				    struct perf_event_attr *attr)
+{
+	return -EOPNOTSUPP;
+}
+#endif
+EXPORT_SYMBOL_GPL(modify_wide_hw_breakpoint_local);
+
 /**
  * hw_breakpoint_is_used - check if breakpoints are currently used
  *
-- 
2.53.0

[RFC PATCH 05/13] mm/kwatch: add watch expression parser and dereference engine

[Jinchao Wang]

KWatch watches a memory address that is only known once the target
function runs, e.g. "argument 1, plus 8, dereferenced once". Add the
two halves of that mechanism:

- kwatch_deref_parse() turns a textual watch expression
  {base}[+-off][->[+-]off]... into a kwatch_config: a base anchor
  (arg1..arg6, stack, an absolute address or - for built-in KWatch -
  a symbol name) plus a static offset chain.

- kwatch_deref_resolve() replays the chain at probe time against
  pt_regs. Every pointer load goes through get_kernel_nofault() and
  the final address must be a kernel address.

Also add the internal kwatch.h header shared by the rest of the
series. Nothing is built yet; the Kconfig entry comes with the
control plane.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 mm/kwatch/Makefile |   3 +
 mm/kwatch/deref.c  | 174 +++++++++++++++++++++++++++++++++++++++++++++
 mm/kwatch/kwatch.h | 107 ++++++++++++++++++++++++++++
 3 files changed, 284 insertions(+)
 create mode 100644 mm/kwatch/Makefile
 create mode 100644 mm/kwatch/deref.c
 create mode 100644 mm/kwatch/kwatch.h

diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
new file mode 100644
index 000000000000..69c21ae62123
--- /dev/null
+++ b/mm/kwatch/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_KWATCH) += kwatch.o
+
+kwatch-y := deref.o
diff --git a/mm/kwatch/deref.c b/mm/kwatch/deref.c
new file mode 100644
index 000000000000..a93c76139e7c
--- /dev/null
+++ b/mm/kwatch/deref.c
@@ -0,0 +1,174 @@
+// SPDX-License-Identifier: GPL-2.0
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/ptrace.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+#include <linux/kallsyms.h>
+#include <linux/string.h>
+#include <linux/slab.h>
+
+#include "kwatch.h"
+
+int kwatch_deref_resolve(const struct kwatch_config *cfg, struct pt_regs *regs,
+			 unsigned long *out_addr, u16 *out_len)
+{
+	unsigned long addr = 0;
+	int i;
+
+	/* 1. Resolve the Base Anchor */
+	if (cfg->base == KWATCH_BASE_STACK) {
+		addr = kernel_stack_pointer(regs);
+		if (unlikely(!addr))
+			return -EINVAL;
+	} else if (cfg->base >= KWATCH_BASE_ARG1 &&
+		   cfg->base <= KWATCH_BASE_ARG6) {
+		int arg_idx = cfg->base - KWATCH_BASE_ARG1;
+
+		addr = regs_get_kernel_argument(regs, arg_idx);
+	} else if (cfg->base == KWATCH_BASE_ABS_ADDR ||
+		   cfg->base == KWATCH_BASE_GLOBAL_SYM) {
+		/* Zero-latency load of the static symbol location */
+		addr = cfg->sym_addr;
+	} else {
+		return -EINVAL;
+	}
+
+	/* 2. The Pointer-Chasing FSM */
+	for (i = 0; i < cfg->offset_count; i++) {
+		addr += cfg->offsets[i];
+
+		if (i < cfg->offset_count - 1) {
+			unsigned long next_addr;
+
+			/* Dynamically read the pointer contents at runtime */
+			if (get_kernel_nofault(next_addr, (unsigned long *)addr))
+				return -EFAULT;
+
+			addr = next_addr;
+		}
+	}
+
+	/* Enforce strict Kernel-Space boundary */
+	if (unlikely(addr < TASK_SIZE_MAX))
+		return -EINVAL;
+
+	*out_addr = addr;
+	*out_len = cfg->watch_len;
+	return 0;
+}
+
+int kwatch_deref_parse(struct kwatch_config *cfg, const char *watch_expr)
+{
+	char *p, *sep, *dup_expr;
+	char type = '\0';
+	bool is_deref = false;
+	int ret = 0;
+
+	dup_expr = kstrdup(watch_expr, GFP_KERNEL);
+	if (!dup_expr)
+		return -ENOMEM;
+
+	cfg->offset_count = 1;
+	cfg->offsets[0] = 0;
+
+	/* 1. Isolate and Resolve Base Anchor */
+	p = dup_expr;
+	sep = NULL;
+	while (*p) {
+		if (*p == '+') {
+			sep = p;
+			type = '+';
+			break;
+		}
+		if (*p == '-') {
+			sep = p;
+			type = '-';
+			if (p[1] == '>')
+				is_deref = true;
+			break;
+		}
+		p++;
+	}
+
+	if (type)
+		*sep = '\0';
+
+	if (!strcmp(dup_expr, "stack")) {
+		cfg->base = KWATCH_BASE_STACK;
+	} else if (!strncmp(dup_expr, "arg", 3) && strlen(dup_expr) == 4) {
+		int arg_num;
+
+		if (kstrtoint(dup_expr + 3, 10, &arg_num) || arg_num < 1 ||
+		    arg_num > 6) {
+			ret = -EINVAL;
+			goto out;
+		}
+		cfg->base = KWATCH_BASE_ARG1 + (arg_num - 1);
+	} else if (kstrtoul(dup_expr, 0, &cfg->sym_addr) == 0) {
+		cfg->base = KWATCH_BASE_ABS_ADDR;
+	} else {
+#if IS_BUILTIN(CONFIG_KWATCH)
+		cfg->sym_addr = kallsyms_lookup_name(dup_expr);
+		if (!cfg->sym_addr) {
+			pr_err("Failed to resolve symbol name: %s\n", dup_expr);
+			ret = -EINVAL;
+			goto out;
+		}
+		cfg->base = KWATCH_BASE_GLOBAL_SYM;
+#else
+		pr_err("cannot resolve symbol %s when built as a module, use a hex address\n",
+		       dup_expr);
+		ret = -EINVAL;
+		goto out;
+#endif
+	}
+
+	if (!type)
+		goto out;
+
+	/* 2. Resolve Base Offset (if + or - exists) */
+	if (!is_deref) {
+		char *next;
+
+		*sep = type; /* Restore the '+' or '-' for kstrtol */
+		next = strstr(sep, "->");
+		if (next)
+			*next = '\0';
+
+		if (kstrtol(sep, 0, &cfg->offsets[0])) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		p = next ? next + 2 : NULL;
+	} else {
+		/* Jump directly to the first dereference after '->' */
+		p = sep + 2;
+	}
+
+	/* 3. Resolve Dereference Chain */
+	while (p) {
+		char *next;
+
+		if (cfg->offset_count >= MAX_DEREF_CHAIN) {
+			ret = -E2BIG;
+			goto out;
+		}
+
+		next = strstr(p, "->");
+		if (next)
+			*next = '\0';
+
+		if (kstrtol(p, 0, &cfg->offsets[cfg->offset_count++])) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		p = next ? next + 2 : NULL;
+	}
+
+out:
+	kfree(dup_expr);
+	return ret;
+}
diff --git a/mm/kwatch/kwatch.h b/mm/kwatch/kwatch.h
new file mode 100644
index 000000000000..e1ac8ae312f6
--- /dev/null
+++ b/mm/kwatch/kwatch.h
@@ -0,0 +1,107 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _MM_KWATCH_H
+#define _MM_KWATCH_H
+
+#include <linux/fprobe.h>
+#include <linux/kprobes.h>
+#include <linux/perf_event.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/compiler.h>
+#include <linux/atomic.h>
+
+#define MAX_CONFIG_STR_LEN 512
+#define MAX_DEREF_CHAIN 4
+
+struct kwatch_watchpoint;
+
+struct kwatch_tsk_ctx {
+	struct task_struct *task;
+	struct kwatch_watchpoint *wp;
+	u16 depth;
+	u32 epoch;
+};
+
+struct kwatch_watchpoint {
+	struct perf_event *__percpu *event;
+	call_single_data_t __percpu *csd_arm;
+	call_single_data_t __percpu *csd_disarm;
+	struct perf_event_attr attr;
+	atomic_t in_use; // multi-consumer safe get/put
+	struct list_head list; // for cpu online and offline
+
+	struct task_struct *arm_tsk;
+	atomic_t pending_ipis;
+	atomic_t refcount;
+	bool teardown;
+};
+
+enum kwatch_access_type {
+	KWATCH_ACCESS_W,
+	KWATCH_ACCESS_R,
+	KWATCH_ACCESS_RW,
+	KWATCH_ACCESS_X,
+};
+
+enum kwatch_base_type {
+	KWATCH_BASE_STACK,
+	KWATCH_BASE_ABS_ADDR,
+	KWATCH_BASE_GLOBAL_SYM,
+	KWATCH_BASE_ARG1,
+	KWATCH_BASE_ARG2,
+	KWATCH_BASE_ARG3,
+	KWATCH_BASE_ARG4,
+	KWATCH_BASE_ARG5,
+	KWATCH_BASE_ARG6,
+};
+
+struct kwatch_config {
+	u16 max_watch;
+	char func_name[KSYM_NAME_LEN];
+	u16 func_offset;
+	u16 depth;
+	u16 duration;
+	enum kwatch_access_type access_type;
+	u16 watch_len;
+
+	/* Unified Deref Engine State */
+	enum kwatch_base_type base;
+	char watch_expr[MAX_CONFIG_STR_LEN];
+	unsigned long sym_addr;
+	long offsets[MAX_DEREF_CHAIN];
+	u8 offset_count;
+	u16 max_concurrency;
+};
+
+int kwatch_hwbp_prealloc(u16 max_watch, enum kwatch_access_type access_type);
+void kwatch_hwbp_free(void);
+int kwatch_hwbp_get(struct kwatch_watchpoint **out_wp);
+void kwatch_hwbp_arm(struct kwatch_watchpoint *wp, unsigned long addr, u16 len);
+int kwatch_hwbp_put(struct kwatch_watchpoint *wp);
+
+int kwatch_probe_start(struct kwatch_config *cfg);
+void kwatch_probe_stop(void);
+void kwatch_probe_mute(bool mute);
+bool kwatch_probe_validate_hit(struct pt_regs *regs, struct task_struct *arm_tsk);
+unsigned long kwatch_probe_nmi_rejected(void);
+
+int kwatch_tsk_ctx_prealloc(u16 max_concurrency);
+struct kwatch_tsk_ctx *kwatch_tsk_ctx_get(bool can_alloc);
+void kwatch_tsk_ctx_put(void);
+void kwatch_tsk_ctx_reset(struct kwatch_tsk_ctx *ctx, u32 new_epoch);
+void kwatch_tsk_ctx_release_wps(void);
+void kwatch_tsk_ctx_free(void);
+
+void kwatch_global_anchor(unsigned long duration_sec);
+int kwatch_anchor_start(u16 duration);
+void kwatch_anchor_stop(void);
+void kwatch_anchor_cancel_work(void);
+bool kwatch_anchor_has_expired(void);
+void kwatch_anchor_clear_expired(void);
+void kwatch_auto_stop(void);
+
+int kwatch_deref_resolve(const struct kwatch_config *cfg, struct pt_regs *regs,
+			 unsigned long *out_addr, u16 *out_len);
+int kwatch_deref_parse(struct kwatch_config *cfg, const char *watch_expr);
+
+#endif /* _MM_KWATCH_H */
-- 
2.53.0

[RFC PATCH 06/13] mm/kwatch: add lockless per-task context pool

[Jinchao Wang]

A task that enters the watched function needs somewhere to keep its
window state (nesting depth, owned watchpoint, config epoch). The
lookup runs in kprobe and NMI-like contexts, so it must not allocate
or take locks.

Use a preallocated open-addressing array hashed by task_struct
pointer. Slots are claimed with cmpxchg() and released with
smp_store_release(); lookup is a read-only probe sequence. The pool
size (max_concurrency) bounds how many tasks can be inside watch
windows concurrently; excess tasks are simply not tracked.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 mm/kwatch/Makefile   |   2 +-
 mm/kwatch/task_ctx.c | 105 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 106 insertions(+), 1 deletion(-)
 create mode 100644 mm/kwatch/task_ctx.c

diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index 69c21ae62123..cc6574df0d68 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
-kwatch-y := deref.o
+kwatch-y := deref.o task_ctx.o
diff --git a/mm/kwatch/task_ctx.c b/mm/kwatch/task_ctx.c
new file mode 100644
index 000000000000..f8e582f0dcfe
--- /dev/null
+++ b/mm/kwatch/task_ctx.c
@@ -0,0 +1,105 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/slab.h>
+#include <linux/hash.h>
+#include <linux/sched.h>
+#include <linux/log2.h>
+#include "kwatch.h"
+
+static u16 kwatch_ctx_pool_size;
+static u16 kwatch_ctx_pool_mask;
+
+static struct kwatch_tsk_ctx *kwatch_ctx_pool;
+
+int kwatch_tsk_ctx_prealloc(u16 max_concurrency)
+{
+	if (!max_concurrency)
+		max_concurrency = 256;
+
+	kwatch_ctx_pool_size = roundup_pow_of_two(max_concurrency);
+	kwatch_ctx_pool_mask = kwatch_ctx_pool_size - 1;
+
+	if (unlikely(!kwatch_ctx_pool)) {
+		kwatch_ctx_pool = kcalloc(kwatch_ctx_pool_size,
+					  sizeof(struct kwatch_tsk_ctx),
+					  GFP_KERNEL);
+		if (!kwatch_ctx_pool)
+			return -ENOMEM;
+	}
+	return 0;
+}
+
+struct kwatch_tsk_ctx *kwatch_tsk_ctx_get(bool can_alloc)
+{
+	int start_idx, i, idx;
+	struct task_struct *t;
+
+	if (unlikely(!kwatch_ctx_pool))
+		return NULL;
+
+	start_idx = hash_ptr(current, ilog2(kwatch_ctx_pool_size));
+
+	for (i = 0; i < kwatch_ctx_pool_size; i++) {
+		idx = (start_idx + i) & kwatch_ctx_pool_mask;
+		t = READ_ONCE(kwatch_ctx_pool[idx].task);
+		if (t == current)
+			return &kwatch_ctx_pool[idx];
+	}
+
+	if (!can_alloc)
+		return NULL;
+
+	for (i = 0; i < kwatch_ctx_pool_size; i++) {
+		idx = (start_idx + i) & kwatch_ctx_pool_mask;
+		t = READ_ONCE(kwatch_ctx_pool[idx].task);
+		if (!t) {
+			if (!cmpxchg(&kwatch_ctx_pool[idx].task, NULL, current))
+				return &kwatch_ctx_pool[idx];
+		}
+	}
+
+	return NULL;
+}
+
+void kwatch_tsk_ctx_reset(struct kwatch_tsk_ctx *ctx, u32 new_epoch)
+{
+	struct kwatch_watchpoint *wp = xchg(&ctx->wp, NULL);
+
+	if (wp)
+		kwatch_hwbp_put(wp);
+	ctx->depth = 0;
+	ctx->epoch = new_epoch;
+}
+
+void kwatch_tsk_ctx_put(void)
+{
+	struct kwatch_tsk_ctx *ctx = kwatch_tsk_ctx_get(false);
+
+	if (unlikely(!ctx))
+		return;
+
+	kwatch_tsk_ctx_reset(ctx, 0);
+
+	/* Pairs with READ_ONCE() in kwatch_tsk_ctx_get() */
+	smp_store_release(&ctx->task, NULL);
+}
+
+void kwatch_tsk_ctx_release_wps(void)
+{
+	int i;
+
+	if (!kwatch_ctx_pool)
+		return;
+
+	for (i = 0; i < kwatch_ctx_pool_size; i++) {
+		struct kwatch_watchpoint *wp = xchg(&kwatch_ctx_pool[i].wp,
+						    NULL);
+		if (wp)
+			kwatch_hwbp_put(wp);
+	}
+}
+
+void kwatch_tsk_ctx_free(void)
+{
+	kfree(kwatch_ctx_pool);
+	kwatch_ctx_pool = NULL;
+}
-- 
2.53.0

[RFC PATCH 07/13] stacktrace: export stack_trace_save_regs()

[Jinchao Wang]

The other stack_trace_save_*() flavours are exported, but the regs
variant is not, so no module can capture a stack trace for a given
pt_regs. KWatch, which may be built as a module, uses it to record
who wrote to a watched address from the hardware breakpoint handler.
Export it like its siblings.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 kernel/stacktrace.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/stacktrace.c b/kernel/stacktrace.c
index afb3c116da91..d853c40f916b 100644
--- a/kernel/stacktrace.c
+++ b/kernel/stacktrace.c
@@ -175,6 +175,7 @@ unsigned int stack_trace_save_regs(struct pt_regs *regs, unsigned long *store,
 	arch_stack_walk(consume_entry, &c, current, regs);
 	return c.len;
 }
+EXPORT_SYMBOL_GPL(stack_trace_save_regs);
 
 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
 /**
@@ -325,6 +326,7 @@ unsigned int stack_trace_save_regs(struct pt_regs *regs, unsigned long *store,
 	save_stack_trace_regs(regs, &trace);
 	return trace.nr_entries;
 }
+EXPORT_SYMBOL_GPL(stack_trace_save_regs);
 
 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
 /**
-- 
2.53.0

[RFC PATCH 08/13] mm/kwatch: add hardware breakpoint backend

[Jinchao Wang]

Manage a preallocated pool of wide (per-CPU) perf hardware
breakpoints. All breakpoints are registered up front against a dummy
address; arming a watchpoint only re-points an already-registered
event, so the arm path can run from a kprobe handler.

- kwatch_hwbp_get()/put() claim and release pool entries with
  per-slot cmpxchg, safe for concurrent consumers on any CPU.
- kwatch_hwbp_arm() updates the local CPU synchronously via
  modify_wide_hw_breakpoint_local() and broadcasts asynchronous IPIs
  to the other CPUs. Arm-side IPIs are rate-limited per CPU; disarm
  IPIs are refcounted so an entry is only recycled once every CPU
  has dropped it.
- Hits are reported through the kwatch:kwatch_hit tracepoint with a
  short stack trace: the ftrace ring buffer is usable from NMI-like
  context and survives a subsequent crash, unlike printk.
- A CPU hotplug callback creates/destroys the per-CPU events as CPUs
  come and go.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 include/trace/events/kwatch.h |  57 ++++++
 mm/kwatch/Makefile            |   2 +-
 mm/kwatch/hwbp.c              | 358 ++++++++++++++++++++++++++++++++++
 3 files changed, 416 insertions(+), 1 deletion(-)
 create mode 100644 include/trace/events/kwatch.h
 create mode 100644 mm/kwatch/hwbp.c

diff --git a/include/trace/events/kwatch.h b/include/trace/events/kwatch.h
new file mode 100644
index 000000000000..edb95405c386
--- /dev/null
+++ b/include/trace/events/kwatch.h
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM kwatch
+
+#if !defined(_TRACE_KWATCH_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_KWATCH_H
+
+#include <linux/tracepoint.h>
+#include <linux/ptrace.h>
+
+#define KWATCH_STACK_DEPTH 8
+
+struct trace_seq;
+const char *kwatch_trace_print_stack(struct trace_seq *p,
+				     const unsigned long *stack,
+				     unsigned int nr);
+
+TRACE_EVENT(kwatch_hit,
+	TP_PROTO(unsigned long ip, unsigned long sp, unsigned long addr,
+		 u64 time_ns,
+		 unsigned long *stack_entries, unsigned int stack_nr),
+	TP_ARGS(ip, sp, addr, time_ns, stack_entries, stack_nr),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, ip)
+		__field(unsigned long, sp)
+		__field(unsigned long, addr)
+		__field(u64, time_ns)
+		__field(unsigned int, stack_nr)
+		__array(unsigned long, stack, KWATCH_STACK_DEPTH)
+	),
+
+	TP_fast_assign(
+		unsigned int i;
+
+		__entry->ip = ip;
+		__entry->sp = sp;
+		__entry->addr = addr;
+		__entry->time_ns = time_ns;
+		__entry->stack_nr = min_t(unsigned int, stack_nr,
+					  KWATCH_STACK_DEPTH);
+		for (i = 0; i < __entry->stack_nr; i++)
+			__entry->stack[i] = stack_entries[i];
+	),
+
+	TP_printk("KWatch HIT: time=%llu.%06lu ip=%pS addr=0x%lx%s",
+		  __entry->time_ns / 1000000000ULL,
+		  (unsigned long)((__entry->time_ns / 1000ULL) % 1000000ULL),
+		  (void *)__entry->ip, __entry->addr,
+		  kwatch_trace_print_stack(p, __entry->stack,
+					   __entry->stack_nr))
+);
+
+#endif /* _TRACE_KWATCH_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index cc6574df0d68..b2bc3003c89b 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
-kwatch-y := deref.o task_ctx.o
+kwatch-y := deref.o task_ctx.o hwbp.o
diff --git a/mm/kwatch/hwbp.c b/mm/kwatch/hwbp.c
new file mode 100644
index 000000000000..19498ba03826
--- /dev/null
+++ b/mm/kwatch/hwbp.c
@@ -0,0 +1,358 @@
+// SPDX-License-Identifier: GPL-2.0
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/cpuhotplug.h>
+#include <linux/ftrace.h>
+#include <linux/hw_breakpoint.h>
+#include <linux/sched/clock.h>
+#include <linux/irqflags.h>
+#include <linux/kallsyms.h>
+#include <linux/mutex.h>
+#include <linux/printk.h>
+#include <linux/slab.h>
+#include <linux/stacktrace.h>
+#include <linux/trace_seq.h>
+#include <linux/workqueue.h>
+
+#include "kwatch.h"
+
+static LIST_HEAD(kwatch_all_wp_list);
+static struct kwatch_watchpoint **kwatch_wp_slots;
+static u16 kwatch_wp_nr;
+static DEFINE_MUTEX(kwatch_all_wp_mutex);
+static unsigned long kwatch_dummy_holder __aligned(8);
+static int kwatch_hwbp_cpuhp_state = CPUHP_INVALID;
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/kwatch.h>
+
+/*
+ * Render the saved stack like the ftrace built-in stacktrace / dump_stack()
+ * style. Symbol resolution runs at trace read time, not in the hit path.
+ */
+const char *kwatch_trace_print_stack(struct trace_seq *p,
+				     const unsigned long *stack,
+				     unsigned int nr)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	unsigned int i;
+
+	for (i = 0; i < nr; i++)
+		trace_seq_printf(p, "\n => %pS", (void *)stack[i]);
+	trace_seq_putc(p, 0);
+	return ret;
+}
+
+static void kwatch_hwbp_handler(struct perf_event *bp,
+				struct perf_sample_data *data,
+				struct pt_regs *regs)
+{
+	struct kwatch_watchpoint *wp = bp->overflow_handler_context;
+	unsigned long stack_entries[KWATCH_STACK_DEPTH];
+	unsigned int stack_nr;
+
+	if (!kwatch_probe_validate_hit(regs, wp->arm_tsk))
+		return;
+
+	stack_nr = stack_trace_save_regs(regs, stack_entries, KWATCH_STACK_DEPTH, 2);
+	trace_kwatch_hit(instruction_pointer(regs), kernel_stack_pointer(regs),
+			 wp->attr.bp_addr, local_clock(),
+			 stack_entries, stack_nr);
+}
+
+static void kwatch_hwbp_arm_local(void *info)
+{
+	struct kwatch_watchpoint *wp = info;
+	struct perf_event *bp;
+	unsigned long flags;
+	int cpu, err;
+
+	local_irq_save(flags);
+
+	cpu = smp_processor_id();
+	bp = per_cpu(*wp->event, cpu);
+
+	if (unlikely(!bp))
+		goto out;
+
+	kwatch_probe_mute(true);
+	barrier();
+
+	err = modify_wide_hw_breakpoint_local(bp, &wp->attr);
+	if (unlikely(err)) {
+		WARN_ONCE(1,
+			  "KWatch: HWBP reinstall failed on CPU%d (err=%d, addr=0x%llx, len=%llu)\n",
+			  cpu, err, wp->attr.bp_addr, wp->attr.bp_len);
+	}
+
+	barrier();
+	kwatch_probe_mute(false);
+
+out:
+	local_irq_restore(flags);
+}
+
+static inline void kwatch_hwbp_try_recycle(struct kwatch_watchpoint *wp)
+{
+	if (atomic_dec_and_test(&wp->pending_ipis)) {
+		if (!READ_ONCE(wp->teardown))
+			atomic_set_release(&wp->in_use, 0);
+
+		atomic_dec(&wp->refcount);
+	}
+}
+
+static void kwatch_hwbp_disarm_local(void *info)
+{
+	struct kwatch_watchpoint *wp = info;
+
+	kwatch_hwbp_arm_local(info);
+	kwatch_hwbp_try_recycle(wp);
+}
+
+static int kwatch_hwbp_cpu_online(unsigned int cpu)
+{
+	struct perf_event_attr attr;
+	struct kwatch_watchpoint *wp;
+	struct perf_event *bp;
+
+	mutex_lock(&kwatch_all_wp_mutex);
+	list_for_each_entry(wp, &kwatch_all_wp_list, list) {
+		attr = wp->attr;
+		attr.bp_addr = (unsigned long)&kwatch_dummy_holder;
+		bp = perf_event_create_kernel_counter(&attr, cpu, NULL,
+						      kwatch_hwbp_handler, wp);
+		if (IS_ERR(bp)) {
+			pr_warn("%s failed to create watch on CPU %d: %ld\n",
+				__func__, cpu, PTR_ERR(bp));
+			continue;
+		}
+		per_cpu(*wp->event, cpu) = bp;
+	}
+	mutex_unlock(&kwatch_all_wp_mutex);
+	return 0;
+}
+
+static int kwatch_hwbp_cpu_offline(unsigned int cpu)
+{
+	struct kwatch_watchpoint *wp;
+	struct perf_event *bp;
+
+	mutex_lock(&kwatch_all_wp_mutex);
+	list_for_each_entry(wp, &kwatch_all_wp_list, list) {
+		bp = per_cpu(*wp->event, cpu);
+		if (bp) {
+			unregister_hw_breakpoint(bp);
+			per_cpu(*wp->event, cpu) = NULL;
+		}
+	}
+	mutex_unlock(&kwatch_all_wp_mutex);
+	return 0;
+}
+
+int kwatch_hwbp_get(struct kwatch_watchpoint **out_wp)
+{
+	struct kwatch_watchpoint *wp;
+	int i;
+
+	/*
+	 * Per-slot cmpxchg claim: safe for concurrent consumers on any CPU,
+	 * unlike llist_del_first() which requires a single consumer.
+	 */
+	for (i = 0; i < kwatch_wp_nr; i++) {
+		wp = kwatch_wp_slots[i];
+		if (atomic_read(&wp->in_use))
+			continue;
+		if (atomic_cmpxchg(&wp->in_use, 0, 1) == 0) {
+			atomic_inc(&wp->refcount);
+			*out_wp = wp;
+			return 0;
+		}
+	}
+	return -EBUSY;
+}
+
+void kwatch_hwbp_arm(struct kwatch_watchpoint *wp, unsigned long addr, u16 len)
+{
+	static DEFINE_PER_CPU(u64, last_ipi_time);
+	int cur_cpu;
+	call_single_data_t *csd;
+	int cpu;
+	bool is_disarm = (addr == (unsigned long)&kwatch_dummy_holder);
+
+	wp->attr.bp_addr = addr;
+	wp->attr.bp_len = len;
+
+	if (!is_disarm)
+		wp->arm_tsk = current;
+
+	/* ensure attr update visible to other cpu before sending IPI */
+	smp_wmb();
+
+	atomic_set(&wp->pending_ipis, 1);
+	cur_cpu = get_cpu();
+
+	if (!is_disarm) {
+		u64 now = local_clock();
+		u64 last = this_cpu_read(last_ipi_time);
+
+		if (now - last < 1000000ULL) {
+			put_cpu();
+			return;
+		}
+		this_cpu_write(last_ipi_time, now);
+	}
+	for_each_online_cpu(cpu) {
+		if (cpu == cur_cpu)
+			continue;
+
+		if (is_disarm)
+			atomic_inc(&wp->pending_ipis);
+
+		csd = per_cpu_ptr(is_disarm ? wp->csd_disarm : wp->csd_arm,
+				  cpu);
+		if (smp_call_function_single_async(cpu, csd) && is_disarm)
+			kwatch_hwbp_try_recycle(wp);
+	}
+	put_cpu();
+
+	if (is_disarm)
+		kwatch_hwbp_disarm_local(wp);
+	else
+		kwatch_hwbp_arm_local(wp);
+}
+
+int kwatch_hwbp_put(struct kwatch_watchpoint *wp)
+{
+	kwatch_hwbp_arm(wp, (unsigned long)&kwatch_dummy_holder,
+			sizeof(unsigned long));
+
+	return 0;
+}
+
+void kwatch_hwbp_free(void)
+{
+	struct kwatch_watchpoint *wp, *tmp;
+
+	kwatch_wp_nr = 0;
+	kfree(kwatch_wp_slots);
+	kwatch_wp_slots = NULL;
+
+	if (kwatch_hwbp_cpuhp_state != CPUHP_INVALID) {
+		cpuhp_remove_state_nocalls(kwatch_hwbp_cpuhp_state);
+		kwatch_hwbp_cpuhp_state = CPUHP_INVALID;
+	}
+
+	mutex_lock(&kwatch_all_wp_mutex);
+	list_for_each_entry_safe(wp, tmp, &kwatch_all_wp_list, list) {
+		list_del(&wp->list);
+
+		WRITE_ONCE(wp->teardown, true);
+		atomic_dec(&wp->refcount);
+
+		/* Wait for all async IPIs to finish */
+		while (atomic_read(&wp->refcount) > 0)
+			cpu_relax();
+
+		unregister_wide_hw_breakpoint(wp->event);
+		free_percpu(wp->csd_arm);
+		free_percpu(wp->csd_disarm);
+		kfree(wp);
+	}
+	mutex_unlock(&kwatch_all_wp_mutex);
+}
+
+int kwatch_hwbp_prealloc(u16 max_watch, enum kwatch_access_type access_type)
+{
+	struct kwatch_watchpoint *wp;
+	int success = 0, cpu;
+	u32 bp_type;
+	int ret;
+
+	switch (access_type) {
+	case KWATCH_ACCESS_X:
+		bp_type = HW_BREAKPOINT_X;
+		break;
+	case KWATCH_ACCESS_R:
+		bp_type = HW_BREAKPOINT_R;
+		break;
+	case KWATCH_ACCESS_RW:
+		bp_type = HW_BREAKPOINT_RW;
+		break;
+	case KWATCH_ACCESS_W:
+	default:
+		bp_type = HW_BREAKPOINT_W;
+		break;
+	}
+
+	while (!max_watch || success < max_watch) {
+		wp = kzalloc_obj(*wp);
+		if (!wp)
+			break;
+
+		wp->csd_arm = alloc_percpu(call_single_data_t);
+		wp->csd_disarm = alloc_percpu(call_single_data_t);
+		if (!wp->csd_arm || !wp->csd_disarm) {
+			free_percpu(wp->csd_arm);
+			free_percpu(wp->csd_disarm);
+			kfree(wp);
+			break;
+		}
+
+		for_each_possible_cpu(cpu) {
+			INIT_CSD(per_cpu_ptr(wp->csd_arm, cpu),
+				 kwatch_hwbp_arm_local, wp);
+			INIT_CSD(per_cpu_ptr(wp->csd_disarm, cpu),
+				 kwatch_hwbp_disarm_local, wp);
+		}
+
+		wp->teardown = false;
+
+		hw_breakpoint_init(&wp->attr);
+		wp->attr.bp_addr = (unsigned long)&kwatch_dummy_holder;
+		wp->attr.bp_len = sizeof(unsigned long);
+		wp->attr.bp_type = bp_type;
+
+		wp->event = register_wide_hw_breakpoint(&wp->attr,
+							kwatch_hwbp_handler,
+							wp);
+		if (IS_ERR((void *)wp->event)) {
+			free_percpu(wp->csd_arm);
+			free_percpu(wp->csd_disarm);
+			kfree(wp);
+			break;
+		}
+
+		atomic_set(&wp->refcount, 1);
+
+		mutex_lock(&kwatch_all_wp_mutex);
+		list_add(&wp->list, &kwatch_all_wp_list);
+		mutex_unlock(&kwatch_all_wp_mutex);
+		success++;
+	}
+
+	if (!success)
+		return -EBUSY;
+
+	kwatch_wp_slots = kcalloc(success, sizeof(*kwatch_wp_slots),
+				  GFP_KERNEL);
+	if (!kwatch_wp_slots) {
+		kwatch_hwbp_free();
+		return -ENOMEM;
+	}
+	mutex_lock(&kwatch_all_wp_mutex);
+	list_for_each_entry(wp, &kwatch_all_wp_list, list)
+		kwatch_wp_slots[kwatch_wp_nr++] = wp;
+	mutex_unlock(&kwatch_all_wp_mutex);
+
+	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "kwatch:online",
+					kwatch_hwbp_cpu_online,
+					kwatch_hwbp_cpu_offline);
+	if (ret < 0) {
+		kwatch_hwbp_free();
+		return ret;
+	}
+
+	kwatch_hwbp_cpuhp_state = ret;
+	return 0;
+}
-- 
2.53.0

[RFC PATCH 09/13] mm/kwatch: add probe lifecycle runtime

[Jinchao Wang]

Open and close the watch window with a kretprobe on the target
function: the entry handler tracks per-task nesting depth and, when
the configured depth is reached, resolves the watch expression and
arms a watchpoint; the exit handler disarms it. An optional kprobe
at func_offset arms mid-function instead of at entry.

Functions running in a real NMI(-like) context are rejected once, at
function entry, by comparing the NMI nesting count against the one
NMI-like layer that int3-based kprobe delivery itself adds; a
companion kprobe with a post_handler pins the probe point so jump
optimization cannot change the delivery mechanism after it is
sampled. Rejections are counted and exposed to the control plane.

A global epoch versioning scheme invalidates stale per-task state
across reconfigurations, and a per-CPU mute flag keeps window
management quiet while a CPU rewrites its own debug registers.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 mm/kwatch/Makefile |   2 +-
 mm/kwatch/probe.c  | 263 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 264 insertions(+), 1 deletion(-)
 create mode 100644 mm/kwatch/probe.c

diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index b2bc3003c89b..f04673cc5b1c 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
-kwatch-y := deref.o task_ctx.o hwbp.o
+kwatch-y := deref.o task_ctx.o hwbp.o probe.o
diff --git a/mm/kwatch/probe.c b/mm/kwatch/probe.c
new file mode 100644
index 000000000000..af6e0af45c10
--- /dev/null
+++ b/mm/kwatch/probe.c
@@ -0,0 +1,263 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/atomic.h>
+#include <linux/kprobes.h>
+#include <linux/kallsyms.h>
+#include <linux/percpu.h>
+#include <linux/preempt.h>
+#include <linux/sched.h>
+
+#include "kwatch.h"
+#define TRAMPOLINE_CHECK_DEPTH 16
+static DEFINE_PER_CPU(bool, kwatch_probe_cpu_muted);
+
+struct kwatch_probe_ctx {
+	struct kprobe kp;
+	struct kretprobe rp;
+	struct kprobe pin_kp;
+	const struct kwatch_config *cfg;
+	bool rp_via_int3;
+
+	u32 epoch;
+};
+
+static struct kwatch_probe_ctx kwatch_probe_ctx;
+static atomic_long_t kwatch_nmi_rejected;
+
+unsigned long kwatch_probe_nmi_rejected(void)
+{
+	return atomic_long_read(&kwatch_nmi_rejected);
+}
+
+/*
+ * True if the probed function itself runs in an NMI-like context.
+ * int3-based kprobe delivery adds one NMI-like layer of its own;
+ * delivery is pinned at registration so the subtraction stays exact.
+ */
+static bool kwatch_probed_ctx_in_nmi(bool via_int3)
+{
+	return (preempt_count() & NMI_MASK) > (via_int3 ? NMI_OFFSET : 0);
+}
+
+static void kwatch_pin_post_handler(struct kprobe *p, struct pt_regs *regs,
+				    unsigned long flags)
+{
+	/* a post_handler pins the probepoint: no jump optimization */
+}
+
+bool kwatch_probe_validate_hit(struct pt_regs *regs,
+			       struct task_struct *arm_tsk)
+{
+	struct kwatch_tsk_ctx *ctx = kwatch_tsk_ctx_get(false);
+
+	if (unlikely(!ctx))
+		return true;
+
+	if (arm_tsk != current ||
+	    ctx->depth != kwatch_probe_ctx.cfg->depth + 1)
+		return true;
+
+	return false;
+}
+
+void kwatch_probe_mute(bool mute)
+{
+	__this_cpu_write(kwatch_probe_cpu_muted, mute);
+}
+
+static inline bool kwatch_probe_is_muted(void)
+{
+	return __this_cpu_read(kwatch_probe_cpu_muted);
+}
+
+enum kwatch_probe_position {
+	KWATCH_PROBE_POSITION_ENTRY,
+	KWATCH_PROBE_POSITION_ACTIVE,
+	KWATCH_PROBE_POSITION_EXIT
+};
+
+static bool kwatch_tsk_ctx_check(enum kwatch_probe_position pos)
+{
+	struct kwatch_tsk_ctx *ctx = kwatch_tsk_ctx_get(true);
+	u32 epoch;
+
+	if (unlikely(!ctx))
+		return false;
+
+	/* Pairs with smp_store_release() in kwatch_probe_start/stop() */
+	epoch = smp_load_acquire(&kwatch_probe_ctx.epoch);
+
+	if (unlikely(ctx->epoch != epoch))
+		kwatch_tsk_ctx_reset(ctx, epoch);
+
+	if (unlikely(!epoch))
+		return false;
+
+	switch (pos) {
+	case KWATCH_PROBE_POSITION_ENTRY:
+		ctx->depth++;
+		return true;
+	case KWATCH_PROBE_POSITION_ACTIVE:
+		return true;
+	case KWATCH_PROBE_POSITION_EXIT:
+		if (unlikely(ctx->depth == 0)) {
+			kwatch_tsk_ctx_put();
+			return false;
+		}
+
+		ctx->depth--;
+		if (ctx->depth == 0) {
+			kwatch_tsk_ctx_put();
+			return false;
+		}
+		return true;
+	}
+	return false;
+}
+
+static int kwatch_activate_handler(struct kprobe *p, struct pt_regs *regs)
+{
+	struct kwatch_tsk_ctx *ctx = kwatch_tsk_ctx_get(false);
+	unsigned long watch_addr;
+	u16 watch_len;
+
+	if (unlikely(!ctx))
+		return 0;
+
+	if (unlikely(kwatch_probe_is_muted()))
+		return 0;
+
+	if (unlikely(!kwatch_tsk_ctx_check(KWATCH_PROBE_POSITION_ACTIVE)))
+		return 0;
+
+	if (ctx->depth != kwatch_probe_ctx.cfg->depth + 1 || ctx->wp)
+		return 0;
+
+	if (kwatch_deref_resolve(kwatch_probe_ctx.cfg, regs, &watch_addr,
+				 &watch_len))
+		return 0;
+
+	if (kwatch_hwbp_get(&ctx->wp))
+		return 0;
+
+	kwatch_hwbp_arm(ctx->wp, watch_addr, watch_len);
+	return 0;
+}
+
+static int kwatch_lifecycle_entry(struct kretprobe_instance *ri,
+				  struct pt_regs *regs)
+{
+	/*
+	 * Single policy point: the target function's context is judged once
+	 * here. A rejected invocation never increments depth, so the offset
+	 * kprobe path inherits the verdict through the depth check.
+	 */
+	if (unlikely(kwatch_probed_ctx_in_nmi(kwatch_probe_ctx.rp_via_int3))) {
+		atomic_long_inc(&kwatch_nmi_rejected);
+		return 1; /* NMI context is unsupported: no window, no return hook */
+	}
+
+	if (!kwatch_tsk_ctx_check(KWATCH_PROBE_POSITION_ENTRY))
+		return 0;
+
+	if (kwatch_probe_ctx.cfg->func_offset == 0)
+		kwatch_activate_handler(NULL, regs);
+
+	return 0;
+}
+
+static int kwatch_lifecycle_exit(struct kretprobe_instance *ri,
+				 struct pt_regs *regs)
+{
+	struct kwatch_tsk_ctx *ctx = kwatch_tsk_ctx_get(false);
+
+	if (unlikely(!ctx))
+		return 0;
+
+	if (!kwatch_tsk_ctx_check(KWATCH_PROBE_POSITION_EXIT))
+		return 0;
+
+	if (ctx->depth == kwatch_probe_ctx.cfg->depth) {
+		struct kwatch_watchpoint *wp = xchg(&ctx->wp, NULL);
+
+		if (wp)
+			kwatch_hwbp_put(wp);
+	}
+
+	return 0;
+}
+
+int kwatch_probe_start(struct kwatch_config *cfg)
+{
+	static u32 next_epoch;
+	u32 current_epoch;
+	int ret;
+
+	/*
+	 * Lockless check to prevent concurrent starts. Strictly serialized
+	 * by the control plane mutex, but serves as a sanity check.
+	 */
+	if (smp_load_acquire(&kwatch_probe_ctx.epoch) != 0)
+		return -EBUSY;
+
+	memset(&kwatch_probe_ctx, 0, sizeof(kwatch_probe_ctx));
+	kwatch_probe_ctx.cfg = cfg;
+
+	/*
+	 * Pin the entry probepoint before the kretprobe registers, so its
+	 * delivery (int3 vs ftrace) can never change under jump optimization.
+	 * register_kretprobe() clears kp.post_handler, hence the companion.
+	 */
+	kwatch_probe_ctx.pin_kp.symbol_name = cfg->func_name;
+	kwatch_probe_ctx.pin_kp.post_handler = kwatch_pin_post_handler;
+	ret = register_kprobe(&kwatch_probe_ctx.pin_kp);
+	if (ret < 0)
+		return ret;
+
+	kwatch_probe_ctx.rp.entry_handler = kwatch_lifecycle_entry;
+	kwatch_probe_ctx.rp.handler = kwatch_lifecycle_exit;
+	kwatch_probe_ctx.rp.kp.symbol_name = cfg->func_name;
+
+	ret = register_kretprobe(&kwatch_probe_ctx.rp);
+	if (ret < 0) {
+		unregister_kprobe(&kwatch_probe_ctx.pin_kp);
+		return ret;
+	}
+	kwatch_probe_ctx.rp_via_int3 = !kprobe_ftrace(&kwatch_probe_ctx.rp.kp);
+
+	if (cfg->func_offset) {
+		kwatch_probe_ctx.kp.symbol_name = cfg->func_name;
+		kwatch_probe_ctx.kp.offset = cfg->func_offset;
+		kwatch_probe_ctx.kp.pre_handler = kwatch_activate_handler;
+
+		ret = register_kprobe(&kwatch_probe_ctx.kp);
+		if (ret) {
+			unregister_kretprobe(&kwatch_probe_ctx.rp);
+			unregister_kprobe(&kwatch_probe_ctx.pin_kp);
+			return ret;
+		}
+	}
+
+	current_epoch = ++next_epoch;
+	if (unlikely(!current_epoch))
+		current_epoch = ++next_epoch;
+
+	/* Pairs with smp_load_acquire() in kwatch_tsk_ctx_check() */
+	smp_store_release(&kwatch_probe_ctx.epoch, current_epoch);
+
+	return 0;
+}
+
+void kwatch_probe_stop(void)
+{
+	if (!kwatch_probe_ctx.epoch)
+		return;
+
+	/* Pairs with smp_load_acquire() in kwatch_tsk_ctx_check() */
+	smp_store_release(&kwatch_probe_ctx.epoch, 0);
+
+	if (kwatch_probe_ctx.cfg->func_offset > 0)
+		unregister_kprobe(&kwatch_probe_ctx.kp);
+
+	unregister_kretprobe(&kwatch_probe_ctx.rp);
+	unregister_kprobe(&kwatch_probe_ctx.pin_kp);
+}
-- 
2.53.0

[RFC PATCH 10/13] mm/kwatch: add anchor thread for global watchpoints

[Jinchao Wang]

Global variables have no function whose execution can bound the
watch window. Provide one: a kernel thread sleeps for the configured
duration inside a dedicated noinline function,
kwatch_global_anchor(), and the probe runtime hooks that function
like any other target.

When the duration expires the thread schedules a work item that
tears the session down; the expired flag is cleared under the
control-plane mutex so a stale work item from a previous session
cannot stop a new one.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 mm/kwatch/Makefile |  2 +-
 mm/kwatch/anchor.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 83 insertions(+), 1 deletion(-)
 create mode 100644 mm/kwatch/anchor.c

diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index f04673cc5b1c..b196c794619a 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
-kwatch-y := deref.o task_ctx.o hwbp.o probe.o
+kwatch-y := deref.o task_ctx.o hwbp.o probe.o anchor.o
diff --git a/mm/kwatch/anchor.c b/mm/kwatch/anchor.c
new file mode 100644
index 000000000000..11da6aff9413
--- /dev/null
+++ b/mm/kwatch/anchor.c
@@ -0,0 +1,82 @@
+// SPDX-License-Identifier: GPL-2.0
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kthread.h>
+#include <linux/wait.h>
+#include <linux/jiffies.h>
+#include <linux/err.h>
+#include <linux/workqueue.h>
+
+#include "kwatch.h"
+
+static DECLARE_WAIT_QUEUE_HEAD(kwatch_anchor_wq);
+static struct task_struct *kwatch_anchor_tsk;
+static bool kwatch_anchor_expired;
+
+bool kwatch_anchor_has_expired(void)
+{
+	return READ_ONCE(kwatch_anchor_expired);
+}
+
+void kwatch_anchor_clear_expired(void)
+{
+	WRITE_ONCE(kwatch_anchor_expired, false);
+}
+
+static void kwatch_auto_stop_handler(struct work_struct *work)
+{
+	kwatch_auto_stop();
+}
+
+static DECLARE_WORK(kwatch_auto_stop_work, kwatch_auto_stop_handler);
+
+noinline void kwatch_global_anchor(unsigned long duration_sec)
+{
+	wait_event_timeout(kwatch_anchor_wq, kthread_should_stop(),
+			   duration_sec * HZ);
+}
+
+static int kwatch_anchor_thread_fn(void *data)
+{
+	unsigned long duration = (unsigned long)data;
+
+	kwatch_global_anchor(duration);
+
+	if (!kthread_should_stop()) {
+		/* mark before scheduling; cleared under the control mutex */
+		WRITE_ONCE(kwatch_anchor_expired, true);
+		schedule_work(&kwatch_auto_stop_work);
+	}
+
+	while (!kthread_should_stop())
+		schedule_timeout_uninterruptible(HZ);
+
+	return 0;
+}
+
+int kwatch_anchor_start(u16 duration)
+{
+	kwatch_anchor_tsk = kthread_run(kwatch_anchor_thread_fn,
+					(void *)(unsigned long)duration,
+					"kwatch_anchor");
+	if (IS_ERR(kwatch_anchor_tsk)) {
+		int ret = PTR_ERR(kwatch_anchor_tsk);
+
+		kwatch_anchor_tsk = NULL;
+		return ret;
+	}
+	return 0;
+}
+
+void kwatch_anchor_stop(void)
+{
+	if (kwatch_anchor_tsk) {
+		kthread_stop(kwatch_anchor_tsk);
+		kwatch_anchor_tsk = NULL;
+	}
+}
+
+void kwatch_anchor_cancel_work(void)
+{
+	cancel_work_sync(&kwatch_auto_stop_work);
+}
-- 
2.53.0

[RFC PATCH 11/13] mm/kwatch: add debugfs control plane

[Jinchao Wang]

Wire the pieces together behind a single debugfs file,
/sys/kernel/debug/kwatch/config. Writing a key=value configuration
string stops any active session and starts a new one; reading shows
the active configuration and the nmi_rejected counter. An open-count
guard keeps the file single-open and a mutex serializes
start/stop/auto-stop against each other.

Add the Kconfig entry and hook mm/kwatch into the mm build. KWatch
can be built in or as a module; symbol-name watch expressions need
the built-in flavour (kallsyms_lookup_name is not exported).

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 MAINTAINERS        |   8 ++
 mm/Kconfig         |   1 +
 mm/Makefile        |   1 +
 mm/kwatch/Kconfig  |  17 +++
 mm/kwatch/Makefile |   2 +-
 mm/kwatch/core.c   | 325 +++++++++++++++++++++++++++++++++++++++++++++
 6 files changed, 353 insertions(+), 1 deletion(-)
 create mode 100644 mm/kwatch/Kconfig
 create mode 100644 mm/kwatch/core.c

diff --git a/MAINTAINERS b/MAINTAINERS
index 7cc4bca5a2c5..b6371f92fe5c 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -14578,6 +14578,14 @@ S:	Supported
 T:	git git://git.kernel.org/pub/scm/virt/kvm/kvm.git
 F:	arch/x86/kvm/xen.*
 
+KWATCH
+M:	Jinchao Wang <wangjinchao600@gmail.com>
+L:	linux-mm@kvack.org
+S:	Maintained
+F:	Documentation/dev-tools/kwatch.rst
+F:	include/trace/events/kwatch.h
+F:	mm/kwatch/
+
 L3MDEV
 M:	David Ahern <dsahern@kernel.org>
 L:	netdev@vger.kernel.org
diff --git a/mm/Kconfig b/mm/Kconfig
index 9e0ca4824905..cac75a46e21a 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -1510,5 +1510,6 @@ config LAZY_MMU_MODE_KUNIT_TEST
 	  If unsure, say N.
 
 source "mm/damon/Kconfig"
+source "mm/kwatch/Kconfig"
 
 endmenu
diff --git a/mm/Makefile b/mm/Makefile
index eff9f9e7e061..80c688330358 100644
--- a/mm/Makefile
+++ b/mm/Makefile
@@ -92,6 +92,7 @@ obj-$(CONFIG_PAGE_POISONING) += page_poison.o
 obj-$(CONFIG_KASAN)	+= kasan/
 obj-$(CONFIG_KFENCE) += kfence/
 obj-$(CONFIG_KMSAN)	+= kmsan/
+obj-$(CONFIG_KWATCH) += kwatch/
 obj-$(CONFIG_FAILSLAB) += failslab.o
 obj-$(CONFIG_FAIL_PAGE_ALLOC) += fail_page_alloc.o
 obj-$(CONFIG_MEMTEST)		+= memtest.o
diff --git a/mm/kwatch/Kconfig b/mm/kwatch/Kconfig
new file mode 100644
index 000000000000..b1c37a829dd5
--- /dev/null
+++ b/mm/kwatch/Kconfig
@@ -0,0 +1,17 @@
+config KWATCH
+	tristate "Kernel Watch Framework"
+	depends on PERF_EVENTS && HAVE_HW_BREAKPOINT && DEBUG_FS
+	depends on HAVE_REINSTALL_HW_BREAKPOINT
+	select KPROBES
+	select KRETPROBES
+	select STACKTRACE
+	help
+	  A generalized hardware-assisted memory monitor utility.
+	  It provides a low-overhead, real-time trigger mechanism to monitor
+	  kernel memory safely in atomic contexts using hardware breakpoints.
+
+	  KWatch is designed to catch silent memory corruptions, stack
+	  overwrites, and complex Heisenbugs by synchronously trapping the
+	  exact instruction causing the illegal access.
+
+	  If unsure, say N.
diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index b196c794619a..02d7917602f1 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
-kwatch-y := deref.o task_ctx.o hwbp.o probe.o anchor.o
+kwatch-y := core.o deref.o task_ctx.o hwbp.o probe.o anchor.o
diff --git a/mm/kwatch/core.c b/mm/kwatch/core.c
new file mode 100644
index 000000000000..548d0cdd0812
--- /dev/null
+++ b/mm/kwatch/core.c
@@ -0,0 +1,325 @@
+// SPDX-License-Identifier: GPL-2.0
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kstrtox.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/atomic.h>
+#include <linux/debugfs.h>
+#include <linux/mutex.h>
+#include "kwatch.h"
+
+static struct kwatch_config kwatch_config;
+static bool watching_active;
+
+static struct dentry *dbgfs_dir;
+static struct dentry *dbgfs_config;
+static DEFINE_MUTEX(kwatch_dbgfs_mutex);
+static atomic_t dbgfs_config_busy = ATOMIC_INIT(0);
+
+static int kwatch_start_watching(void)
+{
+	int ret;
+
+	if (!strlen(kwatch_config.func_name)) {
+		if (kwatch_config.duration > 0) {
+			strscpy(kwatch_config.func_name, "kwatch_global_anchor",
+				sizeof(kwatch_config.func_name));
+		} else {
+			pr_err("func_name or duration is required\n");
+			return -EINVAL;
+		}
+	} else if (kwatch_config.duration > 0 &&
+		   strcmp(kwatch_config.func_name, "kwatch_global_anchor")) {
+		pr_warn("duration is ignored when watching a specific function\n");
+	}
+
+	if (kwatch_config.access_type > 3) {
+		pr_err("Invalid access_type (must be 0-3)\n");
+		return -EINVAL;
+	}
+
+	ret = kwatch_hwbp_prealloc(kwatch_config.max_watch,
+				   kwatch_config.access_type);
+	if (ret) {
+		pr_err("kwatch_hwbp_prealloc ret: %d\n", ret);
+		return ret;
+	}
+
+	ret = kwatch_tsk_ctx_prealloc(kwatch_config.max_concurrency);
+	if (ret) {
+		kwatch_hwbp_free();
+		return ret;
+	}
+
+	ret = kwatch_probe_start(&kwatch_config);
+	if (ret) {
+		pr_err("kwatch_probe_start ret: %d\n", ret);
+		kwatch_tsk_ctx_free();
+		kwatch_hwbp_free();
+		return ret;
+	}
+
+	if (!strcmp(kwatch_config.func_name, "kwatch_global_anchor")) {
+		ret = kwatch_anchor_start(kwatch_config.duration);
+		if (ret) {
+			kwatch_probe_stop();
+			synchronize_rcu();
+			kwatch_tsk_ctx_release_wps();
+			kwatch_hwbp_free();
+			kwatch_tsk_ctx_free();
+			return ret;
+		}
+	}
+
+	watching_active = true;
+	return 0;
+}
+
+static void kwatch_stop_watching(void)
+{
+	watching_active = false;
+
+	kwatch_anchor_stop();
+	/* after kthread_stop: the dead thread cannot re-mark expiry */
+	kwatch_anchor_clear_expired();
+
+	kwatch_probe_stop();
+	synchronize_rcu();
+	kwatch_tsk_ctx_release_wps();
+	/*
+	 * Waits for disarm IPIs and unregisters breakpoints: no #DB can
+	 * reach the ctx pool once this returns.
+	 */
+	kwatch_hwbp_free();
+	kwatch_tsk_ctx_free();
+}
+
+void kwatch_auto_stop(void)
+{
+	mutex_lock(&kwatch_dbgfs_mutex);
+	/* the expired check neutralizes work items from torn-down sessions */
+	if (watching_active && kwatch_anchor_has_expired()) {
+		kwatch_stop_watching();
+		pr_info("watch duration expired, stopped watching\n");
+	}
+	mutex_unlock(&kwatch_dbgfs_mutex);
+}
+
+static int kwatch_config_parse(char *buf, struct kwatch_config *cfg)
+{
+	char *token, *key, *val;
+	int ret = 0;
+
+	memset(cfg, 0, sizeof(*cfg));
+	cfg->max_concurrency = 256;
+	cfg->max_watch = 4;
+	cfg->watch_len = 8;
+	cfg->access_type = 0;
+
+	while ((token = strsep(&buf, " \t\n")) != NULL) {
+		if (!*token)
+			continue;
+		key = strsep(&token, "=");
+		val = token;
+		if (!key || !val)
+			return -EINVAL;
+
+		if (!strcmp(key, "func_name")) {
+			strscpy(cfg->func_name, val, sizeof(cfg->func_name));
+		} else if (!strcmp(key, "func_offset")) {
+			ret = kstrtou16(val, 0, &cfg->func_offset);
+		} else if (!strcmp(key, "depth")) {
+			ret = kstrtou16(val, 0, &cfg->depth);
+		} else if (!strcmp(key, "max_concurrency")) {
+			ret = kstrtou16(val, 0, &cfg->max_concurrency);
+		} else if (!strcmp(key, "max_watch")) {
+			ret = kstrtou16(val, 0, &cfg->max_watch);
+		} else if (!strcmp(key, "access_type")) {
+			ret = kstrtouint(val, 0, &cfg->access_type);
+		} else if (!strcmp(key, "watch_len")) {
+			ret = kstrtou16(val, 0, &cfg->watch_len);
+			if (!ret && cfg->watch_len != 1 &&
+			    cfg->watch_len != 2 && cfg->watch_len != 4 &&
+			    cfg->watch_len != 8)
+				ret = -EINVAL;
+		} else if (!strcmp(key, "duration")) {
+			ret = kstrtou16(val, 0, &cfg->duration);
+		} else if (!strcmp(key, "watch_expr")) {
+			strscpy(cfg->watch_expr, val, sizeof(cfg->watch_expr));
+			ret = kwatch_deref_parse(cfg, val);
+		}
+
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+static int kwatch_dbgfs_open(struct inode *inode, struct file *file)
+{
+	if (atomic_cmpxchg(&dbgfs_config_busy, 0, 1))
+		return -EBUSY;
+	return 0;
+}
+
+static int kwatch_dbgfs_release(struct inode *inode, struct file *file)
+{
+	atomic_set(&dbgfs_config_busy, 0);
+	return 0;
+}
+
+static ssize_t kwatch_dbgfs_read(struct file *file, char __user *user_buf,
+				 size_t count, loff_t *ppos)
+{
+	char *out_buf;
+	size_t len = 0;
+	ssize_t ret;
+
+	out_buf = kzalloc(MAX_CONFIG_STR_LEN, GFP_KERNEL);
+	if (!out_buf)
+		return -ENOMEM;
+
+	if (watching_active) {
+		len += scnprintf(out_buf + len, MAX_CONFIG_STR_LEN - len,
+				 "func_name=%s\n"
+				 "func_offset=%u\n"
+				 "depth=%u\n"
+				 "duration=%u\n"
+				 "max_concurrency=%u\n"
+				 "max_watch=%u\n"
+				 "access_type=%u\n"
+				 "watch_len=%u\n",
+				 kwatch_config.func_name,
+				 kwatch_config.func_offset, kwatch_config.depth,
+				 kwatch_config.duration,
+				 kwatch_config.max_concurrency,
+				 kwatch_config.max_watch,
+				 kwatch_config.access_type,
+				 kwatch_config.watch_len);
+
+		if (kwatch_config.base == KWATCH_BASE_GLOBAL_SYM) {
+			len += scnprintf(out_buf + len, MAX_CONFIG_STR_LEN - len,
+					 "sym_addr=0x%lx\n", kwatch_config.sym_addr);
+		}
+
+		len += scnprintf(out_buf + len, MAX_CONFIG_STR_LEN - len,
+				 "watch_expr=%s\n"
+				 "nmi_rejected=%lu\n",
+				 kwatch_config.watch_expr,
+				 kwatch_probe_nmi_rejected());
+	} else {
+		len = scnprintf(out_buf, MAX_CONFIG_STR_LEN, "not watching\n");
+	}
+
+	ret = simple_read_from_buffer(user_buf, count, ppos, out_buf, len);
+	kfree(out_buf);
+	return ret;
+}
+
+static ssize_t kwatch_dbgfs_write(struct file *file, const char __user *buffer,
+				  size_t count, loff_t *ppos)
+{
+	char *input_alloc;
+	char *parse_str;
+	int ret;
+
+	if (count == 0 || count >= MAX_CONFIG_STR_LEN)
+		return -EINVAL;
+
+	input_alloc = memdup_user_nul(buffer, count);
+	if (IS_ERR(input_alloc))
+		return PTR_ERR(input_alloc);
+
+	mutex_lock(&kwatch_dbgfs_mutex);
+
+	if (watching_active)
+		kwatch_stop_watching();
+
+	parse_str = strim(input_alloc);
+
+	if (!strlen(parse_str)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = kwatch_config_parse(parse_str, &kwatch_config);
+	if (ret) {
+		pr_err("Failed to parse config %d\n", ret);
+		goto out;
+	}
+
+	ret = kwatch_start_watching();
+	if (ret) {
+		pr_err("Failed to start watching with %d\n", ret);
+		goto out;
+	}
+
+	ret = count;
+
+out:
+	mutex_unlock(&kwatch_dbgfs_mutex);
+	kfree(input_alloc);
+	return ret;
+}
+
+static const struct file_operations kwatch_fops = {
+	.owner = THIS_MODULE,
+	.open = kwatch_dbgfs_open,
+	.release = kwatch_dbgfs_release,
+	.read = kwatch_dbgfs_read,
+	.write = kwatch_dbgfs_write,
+};
+
+static int __init kwatch_init(void)
+{
+	int ret = 0;
+
+	memset(&kwatch_config, 0, sizeof(kwatch_config));
+
+	dbgfs_dir = debugfs_create_dir("kwatch", NULL);
+	if (IS_ERR(dbgfs_dir)) {
+		ret = PTR_ERR(dbgfs_dir);
+		goto err_dir;
+	}
+
+	dbgfs_config = debugfs_create_file("config", 0600, dbgfs_dir, NULL,
+					   &kwatch_fops);
+	if (IS_ERR(dbgfs_config)) {
+		ret = PTR_ERR(dbgfs_config);
+		goto err_file;
+	}
+
+	pr_info("module loaded\n");
+	return 0;
+
+err_file:
+	debugfs_remove_recursive(dbgfs_dir);
+	dbgfs_dir = NULL;
+err_dir:
+	return ret;
+}
+module_init(kwatch_init);
+
+static void __exit kwatch_exit(void)
+{
+	mutex_lock(&kwatch_dbgfs_mutex);
+	if (watching_active)
+		kwatch_stop_watching();
+	mutex_unlock(&kwatch_dbgfs_mutex);
+
+	/* the anchor thread is dead: nothing can schedule new work now */
+	kwatch_anchor_cancel_work();
+
+	debugfs_remove_recursive(dbgfs_dir);
+	dbgfs_dir = NULL;
+
+	pr_info("kwatch unloaded\n");
+}
+module_exit(kwatch_exit);
+
+MODULE_AUTHOR("Jinchao Wang <wangjinchao600@gmail.com>");
+MODULE_DESCRIPTION("Kernel watchpoint");
+MODULE_LICENSE("GPL");
-- 
2.53.0

[RFC PATCH 12/13] mm/kwatch: add KUnit tests for the watch expression parser

[Jinchao Wang]

Cover base anchors (stack, argN, absolute address), positive and
negative offsets, dereference chains, and rejection of malformed
expressions (missing offsets, bad argument index, junk offsets).

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 mm/kwatch/.kunitconfig |   9 +++
 mm/kwatch/Kconfig      |  10 +++
 mm/kwatch/Makefile     |   1 +
 mm/kwatch/deref_test.c | 137 +++++++++++++++++++++++++++++++++++++++++
 4 files changed, 157 insertions(+)
 create mode 100644 mm/kwatch/.kunitconfig
 create mode 100644 mm/kwatch/deref_test.c

diff --git a/mm/kwatch/.kunitconfig b/mm/kwatch/.kunitconfig
new file mode 100644
index 000000000000..7e977ddf0da1
--- /dev/null
+++ b/mm/kwatch/.kunitconfig
@@ -0,0 +1,9 @@
+CONFIG_KUNIT=y
+CONFIG_KWATCH=y
+CONFIG_KWATCH_KUNIT_TEST=y
+CONFIG_PERF_EVENTS=y
+CONFIG_HAVE_HW_BREAKPOINT=y
+CONFIG_HAVE_REINSTALL_HW_BREAKPOINT=y
+CONFIG_KPROBES=y
+CONFIG_KRETPROBES=y
+CONFIG_PRINTK=y
diff --git a/mm/kwatch/Kconfig b/mm/kwatch/Kconfig
index b1c37a829dd5..74083040a1a3 100644
--- a/mm/kwatch/Kconfig
+++ b/mm/kwatch/Kconfig
@@ -15,3 +15,13 @@ config KWATCH
 	  exact instruction causing the illegal access.
 
 	  If unsure, say N.
+
+config KWATCH_KUNIT_TEST
+	bool "KUnit tests for KWatch" if !KUNIT_ALL_TESTS
+	depends on KWATCH && KUNIT
+	default KUNIT_ALL_TESTS
+	help
+	  Enable KUnit tests for the KWatch kernel module.
+	  This suite tests the core parsing logic, the pointer-chasing
+	  finite state machine, and edge cases involving complex watchpoint
+	  expressions. If unsure, say N.
diff --git a/mm/kwatch/Makefile b/mm/kwatch/Makefile
index 02d7917602f1..1d223d73b461 100644
--- a/mm/kwatch/Makefile
+++ b/mm/kwatch/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_KWATCH) += kwatch.o
 
 kwatch-y := core.o deref.o task_ctx.o hwbp.o probe.o anchor.o
+kwatch-$(CONFIG_KWATCH_KUNIT_TEST) += deref_test.o
diff --git a/mm/kwatch/deref_test.c b/mm/kwatch/deref_test.c
new file mode 100644
index 000000000000..094b7afeb235
--- /dev/null
+++ b/mm/kwatch/deref_test.c
@@ -0,0 +1,137 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <kunit/test.h>
+#include "kwatch.h"
+#include <linux/string.h>
+
+static void kwatch_test_parse_deref_chain(struct kunit *test)
+{
+	struct kwatch_config cfg;
+	int ret;
+
+	// Test 1: stack
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "stack");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_STACK);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 1);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 0);
+
+	// Test 2: arg1
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg1");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG1);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 1);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 0);
+
+	// Test 3: arg6+8
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg6+8");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG6);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 1);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 8);
+
+	// Test 4: arg2-16
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg2-16");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG2);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 1);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], -16);
+
+	// Test 5: arg3->8
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg3->8");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG3);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 2);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 0);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[1], 8);
+
+	// Test 6: arg4+8->16
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg4+8->16");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG4);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 2);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 8);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[1], 16);
+
+	// Test 7: arg5-8->-16
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg5-8->-16");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG5);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 2);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], -8);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[1], -16);
+
+	// Test 8: stack->0->8
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "stack->0->8");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_STACK);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 3);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 0);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[1], 0);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[2], 8);
+
+	// Test 9: arg1->+8
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg1->+8");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ARG1);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 2);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 0);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[1], 8);
+
+	// Test 9.1: arg1-> (implicit 0 should fail)
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg1->");
+	KUNIT_EXPECT_EQ(test, ret, -EINVAL);
+
+	// Test 9.2: stack->->8 (implicit 0 should fail)
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "stack->->8");
+	KUNIT_EXPECT_EQ(test, ret, -EINVAL);
+
+	// Test 10: Invalid base
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "invalid_base");
+	KUNIT_EXPECT_EQ(test, ret, -EINVAL);
+
+	// Test 11: Invalid offset
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg1+abc");
+	KUNIT_EXPECT_EQ(test, ret, -EINVAL);
+
+	// Test 12: Invalid arg
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "arg7");
+	KUNIT_EXPECT_EQ(test, ret, -EINVAL);
+
+	// Test 13: Absolute address
+	memset(&cfg, 0, sizeof(cfg));
+	ret = kwatch_deref_parse(&cfg, "0xffffffff81000000+8");
+	KUNIT_EXPECT_EQ(test, ret, 0);
+	KUNIT_EXPECT_EQ(test, cfg.base, KWATCH_BASE_ABS_ADDR);
+	KUNIT_EXPECT_EQ(test, cfg.sym_addr, 0xffffffff81000000UL);
+	KUNIT_EXPECT_EQ(test, cfg.offset_count, 1);
+	KUNIT_EXPECT_EQ(test, cfg.offsets[0], 8);
+}
+
+static struct kunit_case kwatch_deref_test_cases[] = {
+	KUNIT_CASE(kwatch_test_parse_deref_chain),
+	{}
+};
+
+static struct kunit_suite kwatch_deref_test_suite = {
+	.name = "kwatch_deref",
+	.test_cases = kwatch_deref_test_cases,
+};
+
+kunit_test_suite(kwatch_deref_test_suite);
+
+MODULE_DESCRIPTION("KUnit tests for the KWatch watch expression parser");
+MODULE_LICENSE("GPL");
-- 
2.53.0

[RFC PATCH 13/13] Documentation/dev-tools: document KWatch

[Jinchao Wang]

Describe what KWatch is for, how it compares with KASAN and KFENCE,
the debugfs configuration interface, the watch expression syntax,
how to read hits from the trace buffer (including after a crash),
and the current limitations.

Signed-off-by: Jinchao Wang <wangjinchao600@gmail.com>
---
 Documentation/dev-tools/index.rst  |   1 +
 Documentation/dev-tools/kwatch.rst | 193 +++++++++++++++++++++++++++++
 2 files changed, 194 insertions(+)
 create mode 100644 Documentation/dev-tools/kwatch.rst

diff --git a/Documentation/dev-tools/index.rst b/Documentation/dev-tools/index.rst
index 59cbb77b33ff..f4c748da63db 100644
--- a/Documentation/dev-tools/index.rst
+++ b/Documentation/dev-tools/index.rst
@@ -30,6 +30,7 @@ Documentation/process/debugging/index.rst
    ubsan
    kmemleak
    kcsan
+   kwatch
    lkmm/index
    kfence
    kselftest
diff --git a/Documentation/dev-tools/kwatch.rst b/Documentation/dev-tools/kwatch.rst
new file mode 100644
index 000000000000..8ead0beb06b6
--- /dev/null
+++ b/Documentation/dev-tools/kwatch.rst
@@ -0,0 +1,193 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+======================================
+KWatch - Kernel Memory Watchpoint Tool
+======================================
+
+Overview
+========
+
+KWatch is a runtime-configurable debugging tool for locating kernel memory
+corruption. It arms hardware breakpoints (watchpoints) on a target address
+while a chosen function is executing, and reports the exact instruction that
+touches the watched memory, together with a stack trace, through a
+tracepoint.
+
+Unlike shadow-memory sanitizers, KWatch does not detect invalid accesses in
+general; it answers a narrower but common question during corruption hunts:
+"who writes to this address?". This includes in-bounds logical overwrites
+that KASAN cannot see, because the rogue writer modifies valid memory
+through a valid pointer, just at the wrong time or with the wrong data.
+
+Comparison with other tools:
+
+* KASAN detects out-of-bounds and use-after-free accesses, but reports the
+  symptom (the invalid access), not the writer that corrupted the data
+  earlier. It requires a rebuild and has significant CPU and memory
+  overhead, and its redzones perturb memory layout, which can hide
+  timing-sensitive bugs.
+* KFENCE is a low-overhead sampling detector for slab objects; it cannot be
+  pointed at one specific address.
+* Hardware breakpoints via kgdb or perf can watch an address, but only a
+  fixed one, system-wide, for the whole run. KWatch resolves the address
+  dynamically at function entry (for example "argument 2 of this function,
+  plus offset 8, dereferenced once") and disarms it again at function exit,
+  so short-lived and per-invocation objects can be watched too.
+
+KWatch has near-zero overhead while armed: the watched function pays for
+one kprobe/kretprobe pair plus programming of the debug registers; the rest
+of the system runs at full speed.
+
+Requirements
+============
+
+* ``CONFIG_KWATCH=y`` or ``m``. The Kconfig symbol depends on
+  ``CONFIG_PERF_EVENTS``, ``CONFIG_DEBUG_FS`` and an architecture that
+  provides ``HAVE_REINSTALL_HW_BREAKPOINT`` (currently x86 only).
+* Resolving symbol names in watch expressions requires ``CONFIG_KWATCH=y``
+  (built-in); a module can only watch absolute hexadecimal addresses.
+
+Usage
+=====
+
+KWatch is configured through a single debugfs file::
+
+    /sys/kernel/debug/kwatch/config
+
+Writing a configuration string starts a watch session (stopping any previous
+one); reading the file shows the active configuration and hit-rejection
+counters. The configuration is a whitespace-separated list of ``key=value``
+tokens:
+
+=================== ==========================================================
+Key                 Meaning
+=================== ==========================================================
+``func_name``       Function whose execution opens the watch window.
+``func_offset``     Instruction offset inside ``func_name`` at which the
+                    watchpoint is armed (default 0 = function entry).
+``watch_expr``      Expression describing the address to watch (see below).
+``watch_len``       Watched length in bytes: 1, 2, 4 or 8 (default 8).
+``access_type``     0 = write (default), 1 = read, 2 = read/write,
+                    3 = execute.
+``depth``           Recursion depth at which the window opens (default 0).
+``max_watch``       Number of hardware watchpoints to preallocate
+                    (default 4).
+``max_concurrency`` Maximum number of tasks concurrently inside the watch
+                    window (default 256).
+``duration``        For global watches: seconds until automatic stop.
+=================== ==========================================================
+
+Watch expressions
+-----------------
+
+The address to watch is computed at function entry from::
+
+    watch_expr={base}[+-offset][->[+-]offset]...
+
+* ``base`` is one of:
+
+  - ``arg1`` ... ``arg6``: a function argument (register calling
+    convention),
+  - ``stack``: the kernel stack pointer at the probe point,
+  - an absolute hexadecimal address, e.g. ``0xffffffff81234567``,
+  - a global symbol name (built-in KWatch only).
+
+* ``+offset`` / ``-offset`` adjusts the current address.
+* ``->offset`` loads the pointer stored at the current address (via
+  ``get_kernel_nofault()``) and then applies the offset. Up to four chain
+  elements are supported; offsets must be explicit (``->`` alone is
+  rejected).
+
+Given::
+
+    struct some_struct {
+        struct some_struct *ptr;    /* offset 0 */
+        int num;                    /* offset 8 */
+    };
+
+    void target_function(struct some_struct *arg1);
+
+typical expressions are:
+
+=========================== ==============================================
+Expression                  Watches
+=========================== ==============================================
+``watch_expr=arg1``         ``&arg1->ptr`` (the pointer field itself)
+``watch_expr=arg1+8``       ``&arg1->num``
+``watch_expr=arg1->0``      ``&arg1->ptr->ptr`` (one dereference)
+``watch_expr=arg1->8``      ``&arg1->ptr->num``
+``watch_expr=0xffff...+8``  absolute address plus 8
+=========================== ==============================================
+
+Example: catch whoever overwrites ``arg1->num`` of a function while that
+function runs::
+
+    echo "func_name=target_function watch_expr=arg1+8 watch_len=4" \
+        > /sys/kernel/debug/kwatch/config
+
+Watching global variables
+-------------------------
+
+A global variable has no natural function window. When ``duration`` is
+given without ``func_name``, KWatch starts an internal anchor kernel thread
+that sleeps inside a dummy function, and uses that function as the window::
+
+    echo "watch_expr=jiffies_wobble duration=60 watch_len=8" \
+        > /sys/kernel/debug/kwatch/config
+
+The session tears itself down when the duration expires.
+
+Reading hits
+------------
+
+Hits are emitted as the ``kwatch:kwatch_hit`` tracepoint, which is safe in
+NMI-like contexts where printk is not. Each event carries the timestamp,
+the instruction pointer, the watched address and a short stack trace::
+
+    echo 1 > /sys/kernel/debug/tracing/events/kwatch/kwatch_hit/enable
+    cat /sys/kernel/debug/tracing/trace_pipe
+
+If the corruption crashes the machine, the ring buffer can still be
+recovered:
+
+* ``echo 1 > /proc/sys/kernel/ftrace_dump_on_oops`` (or the
+  ``ftrace_dump_on_oops`` boot parameter) dumps the buffer to the console
+  on an oops.
+* With kdump, the buffer is present in the vmcore and can be read with
+  ``crash> trace``.
+* ``CONFIG_PSTORE_FTRACE`` persists it across reboots on supported
+  platforms.
+
+Limitations
+===========
+
+* Functions that run in a genuine NMI(-like) context are rejected at
+  function entry; rejected invocations never open a watch window and are
+  counted in the ``nmi_rejected`` field of the config file. Watching
+  functions reachable from NMI handlers is out of scope.
+* The number of concurrent watchpoints is bounded by the CPU's debug
+  registers (typically 4).
+* If the target address cannot be resolved at arming time (for example a
+  ``get_kernel_nofault()`` failure on a swapped or unmapped page), the
+  watchpoint is not armed for that invocation.
+* Offsets in watch expressions are static; dynamic indexing such as
+  ``arg1->ptr[arg2]`` is not supported.
+* arm64 is not yet supported: stepping over a hit that has a custom
+  overflow handler needs a generic mechanism in the arch code, which is
+  planned as a follow-up series.
+
+Implementation notes
+====================
+
+The implementation lives in ``mm/kwatch/`` and is split into a control
+plane (``core.c``, the debugfs interface), an execution plane (``probe.c``
+and ``deref.c``: kprobe/kretprobe window management and address
+resolution), and a resource plane (``hwbp.c`` and ``task_ctx.c``).
+
+Hardware watchpoints are preallocated as perf events on every CPU and
+re-pointed at hit time with ``modify_wide_hw_breakpoint_local()``, a new
+hw_breakpoint API that updates the breakpoint on the local CPU without
+releasing its slot; other CPUs are updated by asynchronous IPIs. Per-task
+window state is kept in a fixed-size, lockless open-addressing array
+claimed with ``cmpxchg()``, so the hit path performs no allocation and
+takes no locks, which keeps it safe in atomic and NMI-like contexts.
-- 
2.53.0