* [PATCH v14 0/7] ARM: kprobes: OPTPROBES and other improvements.
@ 2014-12-08 6:27 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
This is v14 of optprobe related patch series. This series fixes a bug
found by Tixy: a race between patch_text() and undef handler.
Previous discussions can be found:
https://lkml.org/lkml/2014/12/4/942
https://lkml.org/lkml/2014/12/4/4
https://lkml.org/lkml/2014/12/1/64
https://lkml.org/lkml/2014/11/22/18
https://lkml.org/lkml/2014/11/21/55
https://lkml.org/lkml/2014/11/18/26
https://lkml.org/lkml/2014/11/19/31
https://lkml.org/lkml/2014/11/18/41
https://lkml.org/lkml/2014/10/25/48
https://lkml.org/lkml/2014/10/22/254
https://lkml.org/lkml/2014/8/27/255
https://lkml.org/lkml/2014/8/12/12
https://lkml.org/lkml/2014/8/8/992
https://lkml.org/lkml/2014/8/8/5
https://lkml.org/lkml/2014/8/5/63
Jon Medhurst (1):
ARM: kprobes: Add test cases for stack consuming instructions
Masami Hiramatsu (1):
kprobes: Pass the original kprobe for preparing optimized kprobe
Wang Nan (5):
ARM: probes: move all probe code to dedicate directory
ARM: kprobes: introduces checker
ARM: kprobes: collects stack consumption for store instructions
ARM: kprobes: disallow probing stack consuming instructions
ARM: kprobes: enable OPTPROBES for ARM 32
arch/arm/Kconfig | 1 +
arch/arm/Makefile | 1 +
arch/arm/{kernel => include/asm}/insn.h | 0
arch/arm/include/asm/kprobes.h | 30 +-
arch/arm/{kernel => include/asm}/patch.h | 0
arch/arm/include/asm/probes.h | 13 +
arch/arm/kernel/Makefile | 16 +-
arch/arm/kernel/entry-armv.S | 3 +-
arch/arm/kernel/ftrace.c | 3 +-
arch/arm/kernel/jump_label.c | 5 +-
arch/arm/kernel/patch.c | 3 +-
arch/arm/probes/Makefile | 7 +
.../{kernel/probes-arm.c => probes/decode-arm.c} | 12 +-
.../{kernel/probes-arm.h => probes/decode-arm.h} | 7 +-
.../probes-thumb.c => probes/decode-thumb.c} | 16 +-
.../probes-thumb.h => probes/decode-thumb.h} | 10 +-
arch/arm/{kernel/probes.c => probes/decode.c} | 74 ++++-
arch/arm/{kernel/probes.h => probes/decode.h} | 13 +-
arch/arm/probes/kprobes/Makefile | 12 +
.../kprobes-arm.c => probes/kprobes/actions-arm.c} | 9 +-
.../kprobes/actions-common.c} | 4 +-
.../kprobes/actions-thumb.c} | 10 +-
arch/arm/probes/kprobes/checkers-arm.c | 99 +++++++
arch/arm/probes/kprobes/checkers-common.c | 101 +++++++
arch/arm/probes/kprobes/checkers-thumb.c | 110 +++++++
arch/arm/probes/kprobes/checkers.h | 54 ++++
.../{kernel/kprobes.c => probes/kprobes/core.c} | 23 +-
.../{kernel/kprobes.h => probes/kprobes/core.h} | 10 +-
arch/arm/probes/kprobes/opt-arm.c | 322 +++++++++++++++++++++
.../kprobes/test-arm.c} | 31 +-
.../kprobes-test.c => probes/kprobes/test-core.c} | 8 +-
.../kprobes-test.h => probes/kprobes/test-core.h} | 2 +-
.../kprobes/test-thumb.c} | 16 +-
arch/arm/probes/uprobes/Makefile | 2 +
.../uprobes-arm.c => probes/uprobes/actions-arm.c} | 6 +-
.../{kernel/uprobes.c => probes/uprobes/core.c} | 8 +-
.../{kernel/uprobes.h => probes/uprobes/core.h} | 0
arch/x86/kernel/kprobes/opt.c | 3 +-
include/linux/kprobes.h | 3 +-
kernel/kprobes.c | 4 +-
samples/kprobes/kprobe_example.c | 2 +-
41 files changed, 961 insertions(+), 92 deletions(-)
rename arch/arm/{kernel => include/asm}/insn.h (100%)
rename arch/arm/{kernel => include/asm}/patch.h (100%)
create mode 100644 arch/arm/probes/Makefile
rename arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} (99%)
rename arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} (93%)
rename arch/arm/{kernel/probes-thumb.c => probes/decode-thumb.c} (98%)
rename arch/arm/{kernel/probes-thumb.h => probes/decode-thumb.h} (90%)
rename arch/arm/{kernel/probes.c => probes/decode.c} (85%)
rename arch/arm/{kernel/probes.h => probes/decode.h} (97%)
create mode 100644 arch/arm/probes/kprobes/Makefile
rename arch/arm/{kernel/kprobes-arm.c => probes/kprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/kprobes-common.c => probes/kprobes/actions-common.c} (98%)
rename arch/arm/{kernel/kprobes-thumb.c => probes/kprobes/actions-thumb.c} (98%)
create mode 100644 arch/arm/probes/kprobes/checkers-arm.c
create mode 100644 arch/arm/probes/kprobes/checkers-common.c
create mode 100644 arch/arm/probes/kprobes/checkers-thumb.c
create mode 100644 arch/arm/probes/kprobes/checkers.h
rename arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} (96%)
rename arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} (83%)
create mode 100644 arch/arm/probes/kprobes/opt-arm.c
rename arch/arm/{kernel/kprobes-test-arm.c => probes/kprobes/test-arm.c} (97%)
rename arch/arm/{kernel/kprobes-test.c => probes/kprobes/test-core.c} (99%)
rename arch/arm/{kernel/kprobes-test.h => probes/kprobes/test-core.h} (99%)
rename arch/arm/{kernel/kprobes-test-thumb.c => probes/kprobes/test-thumb.c} (98%)
create mode 100644 arch/arm/probes/uprobes/Makefile
rename arch/arm/{kernel/uprobes-arm.c => probes/uprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} (97%)
rename arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} (100%)
--
1.8.4
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 0/7] ARM: kprobes: OPTPROBES and other improvements.
@ 2014-12-08 6:27 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: linux-arm-kernel
This is v14 of optprobe related patch series. This series fixes a bug
found by Tixy: a race between patch_text() and undef handler.
Previous discussions can be found:
https://lkml.org/lkml/2014/12/4/942
https://lkml.org/lkml/2014/12/4/4
https://lkml.org/lkml/2014/12/1/64
https://lkml.org/lkml/2014/11/22/18
https://lkml.org/lkml/2014/11/21/55
https://lkml.org/lkml/2014/11/18/26
https://lkml.org/lkml/2014/11/19/31
https://lkml.org/lkml/2014/11/18/41
https://lkml.org/lkml/2014/10/25/48
https://lkml.org/lkml/2014/10/22/254
https://lkml.org/lkml/2014/8/27/255
https://lkml.org/lkml/2014/8/12/12
https://lkml.org/lkml/2014/8/8/992
https://lkml.org/lkml/2014/8/8/5
https://lkml.org/lkml/2014/8/5/63
Jon Medhurst (1):
ARM: kprobes: Add test cases for stack consuming instructions
Masami Hiramatsu (1):
kprobes: Pass the original kprobe for preparing optimized kprobe
Wang Nan (5):
ARM: probes: move all probe code to dedicate directory
ARM: kprobes: introduces checker
ARM: kprobes: collects stack consumption for store instructions
ARM: kprobes: disallow probing stack consuming instructions
ARM: kprobes: enable OPTPROBES for ARM 32
arch/arm/Kconfig | 1 +
arch/arm/Makefile | 1 +
arch/arm/{kernel => include/asm}/insn.h | 0
arch/arm/include/asm/kprobes.h | 30 +-
arch/arm/{kernel => include/asm}/patch.h | 0
arch/arm/include/asm/probes.h | 13 +
arch/arm/kernel/Makefile | 16 +-
arch/arm/kernel/entry-armv.S | 3 +-
arch/arm/kernel/ftrace.c | 3 +-
arch/arm/kernel/jump_label.c | 5 +-
arch/arm/kernel/patch.c | 3 +-
arch/arm/probes/Makefile | 7 +
.../{kernel/probes-arm.c => probes/decode-arm.c} | 12 +-
.../{kernel/probes-arm.h => probes/decode-arm.h} | 7 +-
.../probes-thumb.c => probes/decode-thumb.c} | 16 +-
.../probes-thumb.h => probes/decode-thumb.h} | 10 +-
arch/arm/{kernel/probes.c => probes/decode.c} | 74 ++++-
arch/arm/{kernel/probes.h => probes/decode.h} | 13 +-
arch/arm/probes/kprobes/Makefile | 12 +
.../kprobes-arm.c => probes/kprobes/actions-arm.c} | 9 +-
.../kprobes/actions-common.c} | 4 +-
.../kprobes/actions-thumb.c} | 10 +-
arch/arm/probes/kprobes/checkers-arm.c | 99 +++++++
arch/arm/probes/kprobes/checkers-common.c | 101 +++++++
arch/arm/probes/kprobes/checkers-thumb.c | 110 +++++++
arch/arm/probes/kprobes/checkers.h | 54 ++++
.../{kernel/kprobes.c => probes/kprobes/core.c} | 23 +-
.../{kernel/kprobes.h => probes/kprobes/core.h} | 10 +-
arch/arm/probes/kprobes/opt-arm.c | 322 +++++++++++++++++++++
.../kprobes/test-arm.c} | 31 +-
.../kprobes-test.c => probes/kprobes/test-core.c} | 8 +-
.../kprobes-test.h => probes/kprobes/test-core.h} | 2 +-
.../kprobes/test-thumb.c} | 16 +-
arch/arm/probes/uprobes/Makefile | 2 +
.../uprobes-arm.c => probes/uprobes/actions-arm.c} | 6 +-
.../{kernel/uprobes.c => probes/uprobes/core.c} | 8 +-
.../{kernel/uprobes.h => probes/uprobes/core.h} | 0
arch/x86/kernel/kprobes/opt.c | 3 +-
include/linux/kprobes.h | 3 +-
kernel/kprobes.c | 4 +-
samples/kprobes/kprobe_example.c | 2 +-
41 files changed, 961 insertions(+), 92 deletions(-)
rename arch/arm/{kernel => include/asm}/insn.h (100%)
rename arch/arm/{kernel => include/asm}/patch.h (100%)
create mode 100644 arch/arm/probes/Makefile
rename arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} (99%)
rename arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} (93%)
rename arch/arm/{kernel/probes-thumb.c => probes/decode-thumb.c} (98%)
rename arch/arm/{kernel/probes-thumb.h => probes/decode-thumb.h} (90%)
rename arch/arm/{kernel/probes.c => probes/decode.c} (85%)
rename arch/arm/{kernel/probes.h => probes/decode.h} (97%)
create mode 100644 arch/arm/probes/kprobes/Makefile
rename arch/arm/{kernel/kprobes-arm.c => probes/kprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/kprobes-common.c => probes/kprobes/actions-common.c} (98%)
rename arch/arm/{kernel/kprobes-thumb.c => probes/kprobes/actions-thumb.c} (98%)
create mode 100644 arch/arm/probes/kprobes/checkers-arm.c
create mode 100644 arch/arm/probes/kprobes/checkers-common.c
create mode 100644 arch/arm/probes/kprobes/checkers-thumb.c
create mode 100644 arch/arm/probes/kprobes/checkers.h
rename arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} (96%)
rename arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} (83%)
create mode 100644 arch/arm/probes/kprobes/opt-arm.c
rename arch/arm/{kernel/kprobes-test-arm.c => probes/kprobes/test-arm.c} (97%)
rename arch/arm/{kernel/kprobes-test.c => probes/kprobes/test-core.c} (99%)
rename arch/arm/{kernel/kprobes-test.h => probes/kprobes/test-core.h} (99%)
rename arch/arm/{kernel/kprobes-test-thumb.c => probes/kprobes/test-thumb.c} (98%)
create mode 100644 arch/arm/probes/uprobes/Makefile
rename arch/arm/{kernel/uprobes-arm.c => probes/uprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} (97%)
rename arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} (100%)
--
1.8.4
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 1/7] ARM: probes: move all probe code to dedicate directory
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:27 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
In discussion on LKML (https://lkml.org/lkml/2014/11/28/158), Russell
King suggests to move all probe related code to arch/arm/probes. This
patch does the work. Due to dependency on 'arch/arm/kernel/patch.h', this
patch also moves patch.h to 'arch/arm/include/asm/patch.h', and related
'#include' directives are also midified to '#include <asm/patch.h>'.
Following is an overview of this patch:
./arch/arm/kernel/ ./arch/arm/probes/
|-- Makefile |-- Makefile
|-- probes-arm.c ==> |-- decode-arm.c
|-- probes-arm.h ==> |-- decode-arm.h
|-- probes-thumb.c ==> |-- decode-thumb.c
|-- probes-thumb.h ==> |-- decode-thumb.h
|-- probes.c ==> |-- decode.c
|-- probes.h ==> |-- decode.h
| |-- kprobes
| | |-- Makefile
|-- kprobes-arm.c ==> | |-- actions-arm.c
|-- kprobes-common.c ==> | |-- actions-common.c
|-- kprobes-thumb.c ==> | |-- actions-thumb.c
|-- kprobes.c ==> | |-- core.c
|-- kprobes.h ==> | |-- core.h
|-- kprobes-test-arm.c ==> | |-- test-arm.c
|-- kprobes-test.c ==> | |-- test-core.c
|-- kprobes-test.h ==> | |-- test-core.h
|-- kprobes-test-thumb.c ==> | `-- test-thumb.c
| `-- uprobes
| |-- Makefile
|-- uprobes-arm.c ==> |-- actions-arm.c
|-- uprobes.c ==> |-- core.c
|-- uprobes.h ==> `-- core.h
|
`-- patch.h ==> arch/arm/include/asm/patch.h
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
v1 -> v2:
- Rename source files to describe their functions.
- Add Makefiles in kprobes and uprobes directories.
---
arch/arm/Makefile | 1 +
arch/arm/{kernel => include/asm}/patch.h | 0
arch/arm/kernel/Makefile | 16 ++--------------
arch/arm/kernel/jump_label.c | 2 +-
arch/arm/kernel/patch.c | 3 +--
arch/arm/probes/Makefile | 7 +++++++
arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} | 7 ++++---
arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} | 4 +++-
.../arm/{kernel/probes-thumb.c => probes/decode-thumb.c} | 6 +++---
.../arm/{kernel/probes-thumb.h => probes/decode-thumb.h} | 4 +++-
arch/arm/{kernel/probes.c => probes/decode.c} | 4 ++--
arch/arm/{kernel/probes.h => probes/decode.h} | 2 +-
arch/arm/probes/kprobes/Makefile | 11 +++++++++++
.../kprobes-arm.c => probes/kprobes/actions-arm.c} | 6 +++---
.../kprobes-common.c => probes/kprobes/actions-common.c} | 4 ++--
.../kprobes-thumb.c => probes/kprobes/actions-thumb.c} | 6 +++---
arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} | 8 ++++----
arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} | 3 ++-
.../kprobes-test-arm.c => probes/kprobes/test-arm.c} | 2 +-
.../kprobes-test.c => probes/kprobes/test-core.c} | 8 ++++----
.../kprobes-test.h => probes/kprobes/test-core.h} | 2 +-
.../kprobes-test-thumb.c => probes/kprobes/test-thumb.c} | 4 ++--
arch/arm/probes/uprobes/Makefile | 2 ++
.../uprobes-arm.c => probes/uprobes/actions-arm.c} | 6 +++---
arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} | 6 +++---
arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} | 0
26 files changed, 69 insertions(+), 55 deletions(-)
rename arch/arm/{kernel => include/asm}/patch.h (100%)
create mode 100644 arch/arm/probes/Makefile
rename arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} (99%)
rename arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} (97%)
rename arch/arm/{kernel/probes-thumb.c => probes/decode-thumb.c} (99%)
rename arch/arm/{kernel/probes-thumb.h => probes/decode-thumb.h} (97%)
rename arch/arm/{kernel/probes.c => probes/decode.c} (99%)
rename arch/arm/{kernel/probes.h => probes/decode.h} (99%)
create mode 100644 arch/arm/probes/kprobes/Makefile
rename arch/arm/{kernel/kprobes-arm.c => probes/kprobes/actions-arm.c} (99%)
rename arch/arm/{kernel/kprobes-common.c => probes/kprobes/actions-common.c} (98%)
rename arch/arm/{kernel/kprobes-thumb.c => probes/kprobes/actions-thumb.c} (99%)
rename arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} (99%)
rename arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} (96%)
rename arch/arm/{kernel/kprobes-test-arm.c => probes/kprobes/test-arm.c} (99%)
rename arch/arm/{kernel/kprobes-test.c => probes/kprobes/test-core.c} (99%)
rename arch/arm/{kernel/kprobes-test.h => probes/kprobes/test-core.h} (99%)
rename arch/arm/{kernel/kprobes-test-thumb.c => probes/kprobes/test-thumb.c} (99%)
create mode 100644 arch/arm/probes/uprobes/Makefile
rename arch/arm/{kernel/uprobes-arm.c => probes/uprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} (98%)
rename arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} (100%)
diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 034a949..a57d9bb 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -266,6 +266,7 @@ core-$(CONFIG_KVM_ARM_HOST) += arch/arm/kvm/
# If we have a machine-specific directory, then include it in the build.
core-y += arch/arm/kernel/ arch/arm/mm/ arch/arm/common/
+core-y += arch/arm/probes/
core-y += arch/arm/net/
core-y += arch/arm/crypto/
core-y += arch/arm/firmware/
diff --git a/arch/arm/kernel/patch.h b/arch/arm/include/asm/patch.h
similarity index 100%
rename from arch/arm/kernel/patch.h
rename to arch/arm/include/asm/patch.h
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 38ddd9f..40d3e00 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -51,20 +51,8 @@ obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o insn.o
obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o insn.o
obj-$(CONFIG_JUMP_LABEL) += jump_label.o insn.o patch.o
obj-$(CONFIG_KEXEC) += machine_kexec.o relocate_kernel.o
-obj-$(CONFIG_UPROBES) += probes.o probes-arm.o uprobes.o uprobes-arm.o
-obj-$(CONFIG_KPROBES) += probes.o kprobes.o kprobes-common.o patch.o
-ifdef CONFIG_THUMB2_KERNEL
-obj-$(CONFIG_KPROBES) += kprobes-thumb.o probes-thumb.o
-else
-obj-$(CONFIG_KPROBES) += kprobes-arm.o probes-arm.o
-endif
-obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
-test-kprobes-objs := kprobes-test.o
-ifdef CONFIG_THUMB2_KERNEL
-test-kprobes-objs += kprobes-test-thumb.o
-else
-test-kprobes-objs += kprobes-test-arm.o
-endif
+# Main staffs in KPROBES are in arch/arm/probes/ .
+obj-$(CONFIG_KPROBES) += patch.o
obj-$(CONFIG_OABI_COMPAT) += sys_oabi-compat.o
obj-$(CONFIG_ARM_THUMBEE) += thumbee.o
obj-$(CONFIG_KGDB) += kgdb.o
diff --git a/arch/arm/kernel/jump_label.c b/arch/arm/kernel/jump_label.c
index 4ce4f78..c6c73ed 100644
--- a/arch/arm/kernel/jump_label.c
+++ b/arch/arm/kernel/jump_label.c
@@ -1,8 +1,8 @@
#include <linux/kernel.h>
#include <linux/jump_label.h>
+#include <asm/patch.h>
#include "insn.h"
-#include "patch.h"
#ifdef HAVE_JUMP_LABEL
diff --git a/arch/arm/kernel/patch.c b/arch/arm/kernel/patch.c
index 07314af..08bcfa1 100644
--- a/arch/arm/kernel/patch.c
+++ b/arch/arm/kernel/patch.c
@@ -5,8 +5,7 @@
#include <asm/cacheflush.h>
#include <asm/smp_plat.h>
#include <asm/opcodes.h>
-
-#include "patch.h"
+#include <asm/patch.h>
struct patch {
void *addr;
diff --git a/arch/arm/probes/Makefile b/arch/arm/probes/Makefile
new file mode 100644
index 0000000..aa1f859
--- /dev/null
+++ b/arch/arm/probes/Makefile
@@ -0,0 +1,7 @@
+obj-$(CONFIG_UPROBES) += decode.o decode-arm.o uprobes/
+obj-$(CONFIG_KPROBES) += decode.o kprobes/
+ifdef CONFIG_THUMB2_KERNEL
+obj-$(CONFIG_KPROBES) += decode-thumb.o
+else
+obj-$(CONFIG_KPROBES) += decode-arm.o
+endif
diff --git a/arch/arm/kernel/probes-arm.c b/arch/arm/probes/decode-arm.c
similarity index 99%
rename from arch/arm/kernel/probes-arm.c
rename to arch/arm/probes/decode-arm.c
index 8eaef81..e39cc75 100644
--- a/arch/arm/kernel/probes-arm.c
+++ b/arch/arm/probes/decode-arm.c
@@ -1,5 +1,6 @@
/*
- * arch/arm/kernel/probes-arm.c
+ *
+ * arch/arm/probes/decode-arm.c
*
* Some code moved here from arch/arm/kernel/kprobes-arm.c
*
@@ -20,8 +21,8 @@
#include <linux/stddef.h>
#include <linux/ptrace.h>
-#include "probes.h"
-#include "probes-arm.h"
+#include "decode.h"
+#include "decode-arm.h"
#define sign_extend(x, signbit) ((x) | (0 - ((x) & (1 << (signbit)))))
diff --git a/arch/arm/kernel/probes-arm.h b/arch/arm/probes/decode-arm.h
similarity index 97%
rename from arch/arm/kernel/probes-arm.h
rename to arch/arm/probes/decode-arm.h
index ace6572..9c56b40 100644
--- a/arch/arm/kernel/probes-arm.h
+++ b/arch/arm/probes/decode-arm.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-arm.h
+ * arch/arm/probes/decode-arm.h
*
* Copyright 2013 Linaro Ltd.
* Written by: David A. Long
@@ -15,6 +15,8 @@
#ifndef _ARM_KERNEL_PROBES_ARM_H
#define _ARM_KERNEL_PROBES_ARM_H
+#include "decode.h"
+
enum probes_arm_action {
PROBES_EMULATE_NONE,
PROBES_SIMULATE_NOP,
diff --git a/arch/arm/kernel/probes-thumb.c b/arch/arm/probes/decode-thumb.c
similarity index 99%
rename from arch/arm/kernel/probes-thumb.c
rename to arch/arm/probes/decode-thumb.c
index 4131351..2f0453a 100644
--- a/arch/arm/kernel/probes-thumb.c
+++ b/arch/arm/probes/decode-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-thumb.c
+ * arch/arm/probes/decode-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -12,8 +12,8 @@
#include <linux/kernel.h>
#include <linux/module.h>
-#include "probes.h"
-#include "probes-thumb.h"
+#include "decode.h"
+#include "decode-thumb.h"
static const union decode_item t32_table_1110_100x_x0xx[] = {
diff --git a/arch/arm/kernel/probes-thumb.h b/arch/arm/probes/decode-thumb.h
similarity index 97%
rename from arch/arm/kernel/probes-thumb.h
rename to arch/arm/probes/decode-thumb.h
index 7c6f6eb..039013c 100644
--- a/arch/arm/kernel/probes-thumb.h
+++ b/arch/arm/probes/decode-thumb.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-thumb.h
+ * arch/arm/probes/decode-thumb.h
*
* Copyright 2013 Linaro Ltd.
* Written by: David A. Long
@@ -15,6 +15,8 @@
#ifndef _ARM_KERNEL_PROBES_THUMB_H
#define _ARM_KERNEL_PROBES_THUMB_H
+#include "decode.h"
+
/*
* True if current instruction is in an IT block.
*/
diff --git a/arch/arm/kernel/probes.c b/arch/arm/probes/decode.c
similarity index 99%
rename from arch/arm/kernel/probes.c
rename to arch/arm/probes/decode.c
index a8ab540..3b05d57 100644
--- a/arch/arm/kernel/probes.c
+++ b/arch/arm/probes/decode.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes.c
+ * arch/arm/probes/decode.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -17,7 +17,7 @@
#include <asm/ptrace.h>
#include <linux/bug.h>
-#include "probes.h"
+#include "decode.h"
#ifndef find_str_pc_offset
diff --git a/arch/arm/kernel/probes.h b/arch/arm/probes/decode.h
similarity index 99%
rename from arch/arm/kernel/probes.h
rename to arch/arm/probes/decode.h
index dba9f24..1d0b531 100644
--- a/arch/arm/kernel/probes.h
+++ b/arch/arm/probes/decode.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes.h
+ * arch/arm/probes/decode.h
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
new file mode 100644
index 0000000..eb38a42
--- /dev/null
+++ b/arch/arm/probes/kprobes/Makefile
@@ -0,0 +1,11 @@
+obj-$(CONFIG_KPROBES) += core.o actions-common.o
+obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
+test-kprobes-objs := test-core.o
+
+ifdef CONFIG_THUMB2_KERNEL
+obj-$(CONFIG_KPROBES) += actions-thumb.o
+test-kprobes-objs += test-thumb.o
+else
+obj-$(CONFIG_KPROBES) += actions-arm.o
+test-kprobes-objs += test-arm.o
+endif
diff --git a/arch/arm/kernel/kprobes-arm.c b/arch/arm/probes/kprobes/actions-arm.c
similarity index 99%
rename from arch/arm/kernel/kprobes-arm.c
rename to arch/arm/probes/kprobes/actions-arm.c
index ac300c6..8797879 100644
--- a/arch/arm/kernel/kprobes-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-decode.c
+ * arch/arm/probes/kprobes/actions-arm.c
*
* Copyright (C) 2006, 2007 Motorola Inc.
*
@@ -62,8 +62,8 @@
#include <linux/kprobes.h>
#include <linux/ptrace.h>
-#include "kprobes.h"
-#include "probes-arm.h"
+#include "../decode-arm.h"
+#include "core.h"
#if __LINUX_ARM_ARCH__ >= 6
#define BLX(reg) "blx "reg" \n\t"
diff --git a/arch/arm/kernel/kprobes-common.c b/arch/arm/probes/kprobes/actions-common.c
similarity index 98%
rename from arch/arm/kernel/kprobes-common.c
rename to arch/arm/probes/kprobes/actions-common.c
index 0bf5d64..bd20a71 100644
--- a/arch/arm/kernel/kprobes-common.c
+++ b/arch/arm/probes/kprobes/actions-common.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-common.c
+ * arch/arm/probes/kprobes/actions-common.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -15,7 +15,7 @@
#include <linux/kprobes.h>
#include <asm/opcodes.h>
-#include "kprobes.h"
+#include "core.h"
static void __kprobes simulate_ldm1stm1(probes_opcode_t insn,
diff --git a/arch/arm/kernel/kprobes-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
similarity index 99%
rename from arch/arm/kernel/kprobes-thumb.c
rename to arch/arm/probes/kprobes/actions-thumb.c
index 9495d7f..6c4e60b 100644
--- a/arch/arm/kernel/kprobes-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-thumb.c
+ * arch/arm/probes/kprobes/actions-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -13,8 +13,8 @@
#include <linux/ptrace.h>
#include <linux/kprobes.h>
-#include "kprobes.h"
-#include "probes-thumb.h"
+#include "../decode-thumb.h"
+#include "core.h"
/* These emulation encodings are functionally equivalent... */
#define t32_emulate_rd8rn16rm0ra12_noflags \
diff --git a/arch/arm/kernel/kprobes.c b/arch/arm/probes/kprobes/core.c
similarity index 99%
rename from arch/arm/kernel/kprobes.c
rename to arch/arm/probes/kprobes/core.c
index 6d64420..701f49d 100644
--- a/arch/arm/kernel/kprobes.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -30,11 +30,11 @@
#include <asm/cacheflush.h>
#include <linux/percpu.h>
#include <linux/bug.h>
+#include <asm/patch.h>
-#include "kprobes.h"
-#include "probes-arm.h"
-#include "probes-thumb.h"
-#include "patch.h"
+#include "../decode-arm.h"
+#include "../decode-thumb.h"
+#include "core.h"
#define MIN_STACK_SIZE(addr) \
min((unsigned long)MAX_STACK_SIZE, \
diff --git a/arch/arm/kernel/kprobes.h b/arch/arm/probes/kprobes/core.h
similarity index 96%
rename from arch/arm/kernel/kprobes.h
rename to arch/arm/probes/kprobes/core.h
index 9a2712e..2e1e5a3 100644
--- a/arch/arm/kernel/kprobes.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -19,7 +19,8 @@
#ifndef _ARM_KERNEL_KPROBES_H
#define _ARM_KERNEL_KPROBES_H
-#include "probes.h"
+#include <asm/kprobes.h>
+#include "../decode.h"
/*
* These undefined instructions must be unique and
diff --git a/arch/arm/kernel/kprobes-test-arm.c b/arch/arm/probes/kprobes/test-arm.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test-arm.c
rename to arch/arm/probes/kprobes/test-arm.c
index cb14242..d9a1255 100644
--- a/arch/arm/kernel/kprobes-test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -13,7 +13,7 @@
#include <asm/system_info.h>
#include <asm/opcodes.h>
-#include "kprobes-test.h"
+#include "test-core.h"
#define TEST_ISA "32"
diff --git a/arch/arm/kernel/kprobes-test.c b/arch/arm/probes/kprobes/test-core.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test.c
rename to arch/arm/probes/kprobes/test-core.c
index b206d77..7ab633d 100644
--- a/arch/arm/kernel/kprobes-test.c
+++ b/arch/arm/probes/kprobes/test-core.c
@@ -209,10 +209,10 @@
#include <linux/bug.h>
#include <asm/opcodes.h>
-#include "kprobes.h"
-#include "probes-arm.h"
-#include "probes-thumb.h"
-#include "kprobes-test.h"
+#include "core.h"
+#include "test-core.h"
+#include "../decode-arm.h"
+#include "../decode-thumb.h"
#define BENCHMARKING 1
diff --git a/arch/arm/kernel/kprobes-test.h b/arch/arm/probes/kprobes/test-core.h
similarity index 99%
rename from arch/arm/kernel/kprobes-test.h
rename to arch/arm/probes/kprobes/test-core.h
index 4430990..9991754 100644
--- a/arch/arm/kernel/kprobes-test.h
+++ b/arch/arm/probes/kprobes/test-core.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-test.h
+ * arch/arm/probes/kprobes/test-core.h
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
diff --git a/arch/arm/kernel/kprobes-test-thumb.c b/arch/arm/probes/kprobes/test-thumb.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test-thumb.c
rename to arch/arm/probes/kprobes/test-thumb.c
index 844dd10..6c6e9a9 100644
--- a/arch/arm/kernel/kprobes-test-thumb.c
+++ b/arch/arm/probes/kprobes/test-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-test-thumb.c
+ * arch/arm/probes/kprobes/test-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -12,7 +12,7 @@
#include <linux/module.h>
#include <asm/opcodes.h>
-#include "kprobes-test.h"
+#include "test-core.h"
#define TEST_ISA "16"
diff --git a/arch/arm/probes/uprobes/Makefile b/arch/arm/probes/uprobes/Makefile
new file mode 100644
index 0000000..6c4c735
--- /dev/null
+++ b/arch/arm/probes/uprobes/Makefile
@@ -0,0 +1,2 @@
+obj-$(CONFIG_UPROBES) += core.o actions-arm.o
+
diff --git a/arch/arm/kernel/uprobes-arm.c b/arch/arm/probes/uprobes/actions-arm.c
similarity index 98%
rename from arch/arm/kernel/uprobes-arm.c
rename to arch/arm/probes/uprobes/actions-arm.c
index d3b655f..1dd4916 100644
--- a/arch/arm/kernel/uprobes-arm.c
+++ b/arch/arm/probes/uprobes/actions-arm.c
@@ -13,9 +13,9 @@
#include <linux/uprobes.h>
#include <linux/module.h>
-#include "probes.h"
-#include "probes-arm.h"
-#include "uprobes.h"
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "core.h"
static int uprobes_substitute_pc(unsigned long *pinsn, u32 oregs)
{
diff --git a/arch/arm/kernel/uprobes.c b/arch/arm/probes/uprobes/core.c
similarity index 98%
rename from arch/arm/kernel/uprobes.c
rename to arch/arm/probes/uprobes/core.c
index 56adf9c..b2954f6 100644
--- a/arch/arm/kernel/uprobes.c
+++ b/arch/arm/probes/uprobes/core.c
@@ -17,9 +17,9 @@
#include <asm/opcodes.h>
#include <asm/traps.h>
-#include "probes.h"
-#include "probes-arm.h"
-#include "uprobes.h"
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "core.h"
#define UPROBE_TRAP_NR UINT_MAX
diff --git a/arch/arm/kernel/uprobes.h b/arch/arm/probes/uprobes/core.h
similarity index 100%
rename from arch/arm/kernel/uprobes.h
rename to arch/arm/probes/uprobes/core.h
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 1/7] ARM: probes: move all probe code to dedicate directory
@ 2014-12-08 6:27 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: linux-arm-kernel
In discussion on LKML (https://lkml.org/lkml/2014/11/28/158), Russell
King suggests to move all probe related code to arch/arm/probes. This
patch does the work. Due to dependency on 'arch/arm/kernel/patch.h', this
patch also moves patch.h to 'arch/arm/include/asm/patch.h', and related
'#include' directives are also midified to '#include <asm/patch.h>'.
Following is an overview of this patch:
./arch/arm/kernel/ ./arch/arm/probes/
|-- Makefile |-- Makefile
|-- probes-arm.c ==> |-- decode-arm.c
|-- probes-arm.h ==> |-- decode-arm.h
|-- probes-thumb.c ==> |-- decode-thumb.c
|-- probes-thumb.h ==> |-- decode-thumb.h
|-- probes.c ==> |-- decode.c
|-- probes.h ==> |-- decode.h
| |-- kprobes
| | |-- Makefile
|-- kprobes-arm.c ==> | |-- actions-arm.c
|-- kprobes-common.c ==> | |-- actions-common.c
|-- kprobes-thumb.c ==> | |-- actions-thumb.c
|-- kprobes.c ==> | |-- core.c
|-- kprobes.h ==> | |-- core.h
|-- kprobes-test-arm.c ==> | |-- test-arm.c
|-- kprobes-test.c ==> | |-- test-core.c
|-- kprobes-test.h ==> | |-- test-core.h
|-- kprobes-test-thumb.c ==> | `-- test-thumb.c
| `-- uprobes
| |-- Makefile
|-- uprobes-arm.c ==> |-- actions-arm.c
|-- uprobes.c ==> |-- core.c
|-- uprobes.h ==> `-- core.h
|
`-- patch.h ==> arch/arm/include/asm/patch.h
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
v1 -> v2:
- Rename source files to describe their functions.
- Add Makefiles in kprobes and uprobes directories.
---
arch/arm/Makefile | 1 +
arch/arm/{kernel => include/asm}/patch.h | 0
arch/arm/kernel/Makefile | 16 ++--------------
arch/arm/kernel/jump_label.c | 2 +-
arch/arm/kernel/patch.c | 3 +--
arch/arm/probes/Makefile | 7 +++++++
arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} | 7 ++++---
arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} | 4 +++-
.../arm/{kernel/probes-thumb.c => probes/decode-thumb.c} | 6 +++---
.../arm/{kernel/probes-thumb.h => probes/decode-thumb.h} | 4 +++-
arch/arm/{kernel/probes.c => probes/decode.c} | 4 ++--
arch/arm/{kernel/probes.h => probes/decode.h} | 2 +-
arch/arm/probes/kprobes/Makefile | 11 +++++++++++
.../kprobes-arm.c => probes/kprobes/actions-arm.c} | 6 +++---
.../kprobes-common.c => probes/kprobes/actions-common.c} | 4 ++--
.../kprobes-thumb.c => probes/kprobes/actions-thumb.c} | 6 +++---
arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} | 8 ++++----
arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} | 3 ++-
.../kprobes-test-arm.c => probes/kprobes/test-arm.c} | 2 +-
.../kprobes-test.c => probes/kprobes/test-core.c} | 8 ++++----
.../kprobes-test.h => probes/kprobes/test-core.h} | 2 +-
.../kprobes-test-thumb.c => probes/kprobes/test-thumb.c} | 4 ++--
arch/arm/probes/uprobes/Makefile | 2 ++
.../uprobes-arm.c => probes/uprobes/actions-arm.c} | 6 +++---
arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} | 6 +++---
arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} | 0
26 files changed, 69 insertions(+), 55 deletions(-)
rename arch/arm/{kernel => include/asm}/patch.h (100%)
create mode 100644 arch/arm/probes/Makefile
rename arch/arm/{kernel/probes-arm.c => probes/decode-arm.c} (99%)
rename arch/arm/{kernel/probes-arm.h => probes/decode-arm.h} (97%)
rename arch/arm/{kernel/probes-thumb.c => probes/decode-thumb.c} (99%)
rename arch/arm/{kernel/probes-thumb.h => probes/decode-thumb.h} (97%)
rename arch/arm/{kernel/probes.c => probes/decode.c} (99%)
rename arch/arm/{kernel/probes.h => probes/decode.h} (99%)
create mode 100644 arch/arm/probes/kprobes/Makefile
rename arch/arm/{kernel/kprobes-arm.c => probes/kprobes/actions-arm.c} (99%)
rename arch/arm/{kernel/kprobes-common.c => probes/kprobes/actions-common.c} (98%)
rename arch/arm/{kernel/kprobes-thumb.c => probes/kprobes/actions-thumb.c} (99%)
rename arch/arm/{kernel/kprobes.c => probes/kprobes/core.c} (99%)
rename arch/arm/{kernel/kprobes.h => probes/kprobes/core.h} (96%)
rename arch/arm/{kernel/kprobes-test-arm.c => probes/kprobes/test-arm.c} (99%)
rename arch/arm/{kernel/kprobes-test.c => probes/kprobes/test-core.c} (99%)
rename arch/arm/{kernel/kprobes-test.h => probes/kprobes/test-core.h} (99%)
rename arch/arm/{kernel/kprobes-test-thumb.c => probes/kprobes/test-thumb.c} (99%)
create mode 100644 arch/arm/probes/uprobes/Makefile
rename arch/arm/{kernel/uprobes-arm.c => probes/uprobes/actions-arm.c} (98%)
rename arch/arm/{kernel/uprobes.c => probes/uprobes/core.c} (98%)
rename arch/arm/{kernel/uprobes.h => probes/uprobes/core.h} (100%)
diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 034a949..a57d9bb 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -266,6 +266,7 @@ core-$(CONFIG_KVM_ARM_HOST) += arch/arm/kvm/
# If we have a machine-specific directory, then include it in the build.
core-y += arch/arm/kernel/ arch/arm/mm/ arch/arm/common/
+core-y += arch/arm/probes/
core-y += arch/arm/net/
core-y += arch/arm/crypto/
core-y += arch/arm/firmware/
diff --git a/arch/arm/kernel/patch.h b/arch/arm/include/asm/patch.h
similarity index 100%
rename from arch/arm/kernel/patch.h
rename to arch/arm/include/asm/patch.h
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 38ddd9f..40d3e00 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -51,20 +51,8 @@ obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o insn.o
obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o insn.o
obj-$(CONFIG_JUMP_LABEL) += jump_label.o insn.o patch.o
obj-$(CONFIG_KEXEC) += machine_kexec.o relocate_kernel.o
-obj-$(CONFIG_UPROBES) += probes.o probes-arm.o uprobes.o uprobes-arm.o
-obj-$(CONFIG_KPROBES) += probes.o kprobes.o kprobes-common.o patch.o
-ifdef CONFIG_THUMB2_KERNEL
-obj-$(CONFIG_KPROBES) += kprobes-thumb.o probes-thumb.o
-else
-obj-$(CONFIG_KPROBES) += kprobes-arm.o probes-arm.o
-endif
-obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
-test-kprobes-objs := kprobes-test.o
-ifdef CONFIG_THUMB2_KERNEL
-test-kprobes-objs += kprobes-test-thumb.o
-else
-test-kprobes-objs += kprobes-test-arm.o
-endif
+# Main staffs in KPROBES are in arch/arm/probes/ .
+obj-$(CONFIG_KPROBES) += patch.o
obj-$(CONFIG_OABI_COMPAT) += sys_oabi-compat.o
obj-$(CONFIG_ARM_THUMBEE) += thumbee.o
obj-$(CONFIG_KGDB) += kgdb.o
diff --git a/arch/arm/kernel/jump_label.c b/arch/arm/kernel/jump_label.c
index 4ce4f78..c6c73ed 100644
--- a/arch/arm/kernel/jump_label.c
+++ b/arch/arm/kernel/jump_label.c
@@ -1,8 +1,8 @@
#include <linux/kernel.h>
#include <linux/jump_label.h>
+#include <asm/patch.h>
#include "insn.h"
-#include "patch.h"
#ifdef HAVE_JUMP_LABEL
diff --git a/arch/arm/kernel/patch.c b/arch/arm/kernel/patch.c
index 07314af..08bcfa1 100644
--- a/arch/arm/kernel/patch.c
+++ b/arch/arm/kernel/patch.c
@@ -5,8 +5,7 @@
#include <asm/cacheflush.h>
#include <asm/smp_plat.h>
#include <asm/opcodes.h>
-
-#include "patch.h"
+#include <asm/patch.h>
struct patch {
void *addr;
diff --git a/arch/arm/probes/Makefile b/arch/arm/probes/Makefile
new file mode 100644
index 0000000..aa1f859
--- /dev/null
+++ b/arch/arm/probes/Makefile
@@ -0,0 +1,7 @@
+obj-$(CONFIG_UPROBES) += decode.o decode-arm.o uprobes/
+obj-$(CONFIG_KPROBES) += decode.o kprobes/
+ifdef CONFIG_THUMB2_KERNEL
+obj-$(CONFIG_KPROBES) += decode-thumb.o
+else
+obj-$(CONFIG_KPROBES) += decode-arm.o
+endif
diff --git a/arch/arm/kernel/probes-arm.c b/arch/arm/probes/decode-arm.c
similarity index 99%
rename from arch/arm/kernel/probes-arm.c
rename to arch/arm/probes/decode-arm.c
index 8eaef81..e39cc75 100644
--- a/arch/arm/kernel/probes-arm.c
+++ b/arch/arm/probes/decode-arm.c
@@ -1,5 +1,6 @@
/*
- * arch/arm/kernel/probes-arm.c
+ *
+ * arch/arm/probes/decode-arm.c
*
* Some code moved here from arch/arm/kernel/kprobes-arm.c
*
@@ -20,8 +21,8 @@
#include <linux/stddef.h>
#include <linux/ptrace.h>
-#include "probes.h"
-#include "probes-arm.h"
+#include "decode.h"
+#include "decode-arm.h"
#define sign_extend(x, signbit) ((x) | (0 - ((x) & (1 << (signbit)))))
diff --git a/arch/arm/kernel/probes-arm.h b/arch/arm/probes/decode-arm.h
similarity index 97%
rename from arch/arm/kernel/probes-arm.h
rename to arch/arm/probes/decode-arm.h
index ace6572..9c56b40 100644
--- a/arch/arm/kernel/probes-arm.h
+++ b/arch/arm/probes/decode-arm.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-arm.h
+ * arch/arm/probes/decode-arm.h
*
* Copyright 2013 Linaro Ltd.
* Written by: David A. Long
@@ -15,6 +15,8 @@
#ifndef _ARM_KERNEL_PROBES_ARM_H
#define _ARM_KERNEL_PROBES_ARM_H
+#include "decode.h"
+
enum probes_arm_action {
PROBES_EMULATE_NONE,
PROBES_SIMULATE_NOP,
diff --git a/arch/arm/kernel/probes-thumb.c b/arch/arm/probes/decode-thumb.c
similarity index 99%
rename from arch/arm/kernel/probes-thumb.c
rename to arch/arm/probes/decode-thumb.c
index 4131351..2f0453a 100644
--- a/arch/arm/kernel/probes-thumb.c
+++ b/arch/arm/probes/decode-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-thumb.c
+ * arch/arm/probes/decode-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -12,8 +12,8 @@
#include <linux/kernel.h>
#include <linux/module.h>
-#include "probes.h"
-#include "probes-thumb.h"
+#include "decode.h"
+#include "decode-thumb.h"
static const union decode_item t32_table_1110_100x_x0xx[] = {
diff --git a/arch/arm/kernel/probes-thumb.h b/arch/arm/probes/decode-thumb.h
similarity index 97%
rename from arch/arm/kernel/probes-thumb.h
rename to arch/arm/probes/decode-thumb.h
index 7c6f6eb..039013c 100644
--- a/arch/arm/kernel/probes-thumb.h
+++ b/arch/arm/probes/decode-thumb.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes-thumb.h
+ * arch/arm/probes/decode-thumb.h
*
* Copyright 2013 Linaro Ltd.
* Written by: David A. Long
@@ -15,6 +15,8 @@
#ifndef _ARM_KERNEL_PROBES_THUMB_H
#define _ARM_KERNEL_PROBES_THUMB_H
+#include "decode.h"
+
/*
* True if current instruction is in an IT block.
*/
diff --git a/arch/arm/kernel/probes.c b/arch/arm/probes/decode.c
similarity index 99%
rename from arch/arm/kernel/probes.c
rename to arch/arm/probes/decode.c
index a8ab540..3b05d57 100644
--- a/arch/arm/kernel/probes.c
+++ b/arch/arm/probes/decode.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes.c
+ * arch/arm/probes/decode.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -17,7 +17,7 @@
#include <asm/ptrace.h>
#include <linux/bug.h>
-#include "probes.h"
+#include "decode.h"
#ifndef find_str_pc_offset
diff --git a/arch/arm/kernel/probes.h b/arch/arm/probes/decode.h
similarity index 99%
rename from arch/arm/kernel/probes.h
rename to arch/arm/probes/decode.h
index dba9f24..1d0b531 100644
--- a/arch/arm/kernel/probes.h
+++ b/arch/arm/probes/decode.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/probes.h
+ * arch/arm/probes/decode.h
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
new file mode 100644
index 0000000..eb38a42
--- /dev/null
+++ b/arch/arm/probes/kprobes/Makefile
@@ -0,0 +1,11 @@
+obj-$(CONFIG_KPROBES) += core.o actions-common.o
+obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
+test-kprobes-objs := test-core.o
+
+ifdef CONFIG_THUMB2_KERNEL
+obj-$(CONFIG_KPROBES) += actions-thumb.o
+test-kprobes-objs += test-thumb.o
+else
+obj-$(CONFIG_KPROBES) += actions-arm.o
+test-kprobes-objs += test-arm.o
+endif
diff --git a/arch/arm/kernel/kprobes-arm.c b/arch/arm/probes/kprobes/actions-arm.c
similarity index 99%
rename from arch/arm/kernel/kprobes-arm.c
rename to arch/arm/probes/kprobes/actions-arm.c
index ac300c6..8797879 100644
--- a/arch/arm/kernel/kprobes-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-decode.c
+ * arch/arm/probes/kprobes/actions-arm.c
*
* Copyright (C) 2006, 2007 Motorola Inc.
*
@@ -62,8 +62,8 @@
#include <linux/kprobes.h>
#include <linux/ptrace.h>
-#include "kprobes.h"
-#include "probes-arm.h"
+#include "../decode-arm.h"
+#include "core.h"
#if __LINUX_ARM_ARCH__ >= 6
#define BLX(reg) "blx "reg" \n\t"
diff --git a/arch/arm/kernel/kprobes-common.c b/arch/arm/probes/kprobes/actions-common.c
similarity index 98%
rename from arch/arm/kernel/kprobes-common.c
rename to arch/arm/probes/kprobes/actions-common.c
index 0bf5d64..bd20a71 100644
--- a/arch/arm/kernel/kprobes-common.c
+++ b/arch/arm/probes/kprobes/actions-common.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-common.c
+ * arch/arm/probes/kprobes/actions-common.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -15,7 +15,7 @@
#include <linux/kprobes.h>
#include <asm/opcodes.h>
-#include "kprobes.h"
+#include "core.h"
static void __kprobes simulate_ldm1stm1(probes_opcode_t insn,
diff --git a/arch/arm/kernel/kprobes-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
similarity index 99%
rename from arch/arm/kernel/kprobes-thumb.c
rename to arch/arm/probes/kprobes/actions-thumb.c
index 9495d7f..6c4e60b 100644
--- a/arch/arm/kernel/kprobes-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-thumb.c
+ * arch/arm/probes/kprobes/actions-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -13,8 +13,8 @@
#include <linux/ptrace.h>
#include <linux/kprobes.h>
-#include "kprobes.h"
-#include "probes-thumb.h"
+#include "../decode-thumb.h"
+#include "core.h"
/* These emulation encodings are functionally equivalent... */
#define t32_emulate_rd8rn16rm0ra12_noflags \
diff --git a/arch/arm/kernel/kprobes.c b/arch/arm/probes/kprobes/core.c
similarity index 99%
rename from arch/arm/kernel/kprobes.c
rename to arch/arm/probes/kprobes/core.c
index 6d64420..701f49d 100644
--- a/arch/arm/kernel/kprobes.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -30,11 +30,11 @@
#include <asm/cacheflush.h>
#include <linux/percpu.h>
#include <linux/bug.h>
+#include <asm/patch.h>
-#include "kprobes.h"
-#include "probes-arm.h"
-#include "probes-thumb.h"
-#include "patch.h"
+#include "../decode-arm.h"
+#include "../decode-thumb.h"
+#include "core.h"
#define MIN_STACK_SIZE(addr) \
min((unsigned long)MAX_STACK_SIZE, \
diff --git a/arch/arm/kernel/kprobes.h b/arch/arm/probes/kprobes/core.h
similarity index 96%
rename from arch/arm/kernel/kprobes.h
rename to arch/arm/probes/kprobes/core.h
index 9a2712e..2e1e5a3 100644
--- a/arch/arm/kernel/kprobes.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -19,7 +19,8 @@
#ifndef _ARM_KERNEL_KPROBES_H
#define _ARM_KERNEL_KPROBES_H
-#include "probes.h"
+#include <asm/kprobes.h>
+#include "../decode.h"
/*
* These undefined instructions must be unique and
diff --git a/arch/arm/kernel/kprobes-test-arm.c b/arch/arm/probes/kprobes/test-arm.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test-arm.c
rename to arch/arm/probes/kprobes/test-arm.c
index cb14242..d9a1255 100644
--- a/arch/arm/kernel/kprobes-test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -13,7 +13,7 @@
#include <asm/system_info.h>
#include <asm/opcodes.h>
-#include "kprobes-test.h"
+#include "test-core.h"
#define TEST_ISA "32"
diff --git a/arch/arm/kernel/kprobes-test.c b/arch/arm/probes/kprobes/test-core.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test.c
rename to arch/arm/probes/kprobes/test-core.c
index b206d77..7ab633d 100644
--- a/arch/arm/kernel/kprobes-test.c
+++ b/arch/arm/probes/kprobes/test-core.c
@@ -209,10 +209,10 @@
#include <linux/bug.h>
#include <asm/opcodes.h>
-#include "kprobes.h"
-#include "probes-arm.h"
-#include "probes-thumb.h"
-#include "kprobes-test.h"
+#include "core.h"
+#include "test-core.h"
+#include "../decode-arm.h"
+#include "../decode-thumb.h"
#define BENCHMARKING 1
diff --git a/arch/arm/kernel/kprobes-test.h b/arch/arm/probes/kprobes/test-core.h
similarity index 99%
rename from arch/arm/kernel/kprobes-test.h
rename to arch/arm/probes/kprobes/test-core.h
index 4430990..9991754 100644
--- a/arch/arm/kernel/kprobes-test.h
+++ b/arch/arm/probes/kprobes/test-core.h
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-test.h
+ * arch/arm/probes/kprobes/test-core.h
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
diff --git a/arch/arm/kernel/kprobes-test-thumb.c b/arch/arm/probes/kprobes/test-thumb.c
similarity index 99%
rename from arch/arm/kernel/kprobes-test-thumb.c
rename to arch/arm/probes/kprobes/test-thumb.c
index 844dd10..6c6e9a9 100644
--- a/arch/arm/kernel/kprobes-test-thumb.c
+++ b/arch/arm/probes/kprobes/test-thumb.c
@@ -1,5 +1,5 @@
/*
- * arch/arm/kernel/kprobes-test-thumb.c
+ * arch/arm/probes/kprobes/test-thumb.c
*
* Copyright (C) 2011 Jon Medhurst <tixy@yxit.co.uk>.
*
@@ -12,7 +12,7 @@
#include <linux/module.h>
#include <asm/opcodes.h>
-#include "kprobes-test.h"
+#include "test-core.h"
#define TEST_ISA "16"
diff --git a/arch/arm/probes/uprobes/Makefile b/arch/arm/probes/uprobes/Makefile
new file mode 100644
index 0000000..6c4c735
--- /dev/null
+++ b/arch/arm/probes/uprobes/Makefile
@@ -0,0 +1,2 @@
+obj-$(CONFIG_UPROBES) += core.o actions-arm.o
+
diff --git a/arch/arm/kernel/uprobes-arm.c b/arch/arm/probes/uprobes/actions-arm.c
similarity index 98%
rename from arch/arm/kernel/uprobes-arm.c
rename to arch/arm/probes/uprobes/actions-arm.c
index d3b655f..1dd4916 100644
--- a/arch/arm/kernel/uprobes-arm.c
+++ b/arch/arm/probes/uprobes/actions-arm.c
@@ -13,9 +13,9 @@
#include <linux/uprobes.h>
#include <linux/module.h>
-#include "probes.h"
-#include "probes-arm.h"
-#include "uprobes.h"
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "core.h"
static int uprobes_substitute_pc(unsigned long *pinsn, u32 oregs)
{
diff --git a/arch/arm/kernel/uprobes.c b/arch/arm/probes/uprobes/core.c
similarity index 98%
rename from arch/arm/kernel/uprobes.c
rename to arch/arm/probes/uprobes/core.c
index 56adf9c..b2954f6 100644
--- a/arch/arm/kernel/uprobes.c
+++ b/arch/arm/probes/uprobes/core.c
@@ -17,9 +17,9 @@
#include <asm/opcodes.h>
#include <asm/traps.h>
-#include "probes.h"
-#include "probes-arm.h"
-#include "uprobes.h"
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "core.h"
#define UPROBE_TRAP_NR UINT_MAX
diff --git a/arch/arm/kernel/uprobes.h b/arch/arm/probes/uprobes/core.h
similarity index 100%
rename from arch/arm/kernel/uprobes.h
rename to arch/arm/probes/uprobes/core.h
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 2/7] ARM: kprobes: introduces checker
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:27 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
This patch introdces 'checker' to decoding phase, and calls checkers
when instruction decoding. This allows further decoding for specific
instructions. This patch introduces a stub call of checkers in kprobe
arch_prepare_kprobe() as an example and for further expansion.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
---
v1 -> v2:
- kprobe checker stubs are introduced in this patch.
v2 -> v3:
- Code cleanups following Masami Hiramatsu and Tixy's advises.
- Commit message improvements.
v3 -> v4:
- Move to arch/arm/probes.
---
arch/arm/probes/decode-arm.c | 5 +--
arch/arm/probes/decode-arm.h | 3 +-
arch/arm/probes/decode-thumb.c | 10 +++---
arch/arm/probes/decode-thumb.h | 6 ++--
arch/arm/probes/decode.c | 60 +++++++++++++++++++++++++++++----
arch/arm/probes/decode.h | 11 +++++-
arch/arm/probes/kprobes/actions-arm.c | 2 ++
arch/arm/probes/kprobes/actions-thumb.c | 3 ++
arch/arm/probes/kprobes/core.c | 6 +++-
arch/arm/probes/kprobes/core.h | 7 ++--
arch/arm/probes/uprobes/core.c | 2 +-
11 files changed, 95 insertions(+), 20 deletions(-)
diff --git a/arch/arm/probes/decode-arm.c b/arch/arm/probes/decode-arm.c
index e39cc75..f46d8fc 100644
--- a/arch/arm/probes/decode-arm.c
+++ b/arch/arm/probes/decode-arm.c
@@ -726,10 +726,11 @@ static void __kprobes arm_singlestep(probes_opcode_t insn,
*/
enum probes_insn __kprobes
arm_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = arm_singlestep;
asi->insn_check_cc = probes_condition_checks[insn>>28];
return probes_decode_insn(insn, asi, probes_decode_arm_table, false,
- emulate, actions);
+ emulate, actions, checkers);
}
diff --git a/arch/arm/probes/decode-arm.h b/arch/arm/probes/decode-arm.h
index 9c56b40..a7b0398 100644
--- a/arch/arm/probes/decode-arm.h
+++ b/arch/arm/probes/decode-arm.h
@@ -70,6 +70,7 @@ extern const union decode_item probes_decode_arm_table[];
enum probes_insn arm_probes_decode_insn(probes_opcode_t,
struct arch_probes_insn *, bool emulate,
- const union decode_action *actions);
+ const union decode_action *actions,
+ const struct decode_checker *checkers[]);
#endif
diff --git a/arch/arm/probes/decode-thumb.c b/arch/arm/probes/decode-thumb.c
index 2f0453a..985e7dd 100644
--- a/arch/arm/probes/decode-thumb.c
+++ b/arch/arm/probes/decode-thumb.c
@@ -863,20 +863,22 @@ static void __kprobes thumb32_singlestep(probes_opcode_t opcode,
enum probes_insn __kprobes
thumb16_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = thumb16_singlestep;
asi->insn_check_cc = thumb_check_cc;
return probes_decode_insn(insn, asi, probes_decode_thumb16_table, true,
- emulate, actions);
+ emulate, actions, checkers);
}
enum probes_insn __kprobes
thumb32_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = thumb32_singlestep;
asi->insn_check_cc = thumb_check_cc;
return probes_decode_insn(insn, asi, probes_decode_thumb32_table, true,
- emulate, actions);
+ emulate, actions, checkers);
}
diff --git a/arch/arm/probes/decode-thumb.h b/arch/arm/probes/decode-thumb.h
index 039013c..8457add 100644
--- a/arch/arm/probes/decode-thumb.h
+++ b/arch/arm/probes/decode-thumb.h
@@ -91,9 +91,11 @@ extern const union decode_item probes_decode_thumb16_table[];
enum probes_insn __kprobes
thumb16_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions);
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[]);
enum probes_insn __kprobes
thumb32_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions);
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[]);
#endif
diff --git a/arch/arm/probes/decode.c b/arch/arm/probes/decode.c
index 3b05d57..c7d4420 100644
--- a/arch/arm/probes/decode.c
+++ b/arch/arm/probes/decode.c
@@ -342,6 +342,31 @@ static const int decode_struct_sizes[NUM_DECODE_TYPES] = {
[DECODE_TYPE_REJECT] = sizeof(struct decode_reject)
};
+static int run_checkers(const struct decode_checker *checkers[],
+ int action, probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ const struct decode_checker **p;
+
+ if (!checkers)
+ return INSN_GOOD;
+
+ p = checkers;
+ while (*p != NULL) {
+ int retval;
+ probes_check_t *checker_func = (*p)[action].checker;
+
+ retval = INSN_GOOD;
+ if (checker_func)
+ retval = checker_func(insn, asi, h);
+ if (retval == INSN_REJECTED)
+ return retval;
+ p++;
+ }
+ return INSN_GOOD;
+}
+
/*
* probes_decode_insn operates on data tables in order to decode an ARM
* architecture instruction onto which a kprobe has been placed.
@@ -388,11 +413,17 @@ static const int decode_struct_sizes[NUM_DECODE_TYPES] = {
int __kprobes
probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
const union decode_item *table, bool thumb,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
const struct decode_header *h = (struct decode_header *)table;
const struct decode_header *next;
bool matched = false;
+ /*
+ * @insn can be modified by decode_regs. Save its original
+ * value for checkers.
+ */
+ probes_opcode_t origin_insn = insn;
if (emulate)
insn = prepare_emulated_insn(insn, asi, thumb);
@@ -422,24 +453,41 @@ probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
}
case DECODE_TYPE_CUSTOM: {
+ int err;
struct decode_custom *d = (struct decode_custom *)h;
- return actions[d->decoder.action].decoder(insn, asi, h);
+ int action = d->decoder.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
+ return actions[action].decoder(insn, asi, h);
}
case DECODE_TYPE_SIMULATE: {
+ int err;
struct decode_simulate *d = (struct decode_simulate *)h;
- asi->insn_handler = actions[d->handler.action].handler;
+ int action = d->handler.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
+ asi->insn_handler = actions[action].handler;
return INSN_GOOD_NO_SLOT;
}
case DECODE_TYPE_EMULATE: {
+ int err;
struct decode_emulate *d = (struct decode_emulate *)h;
+ int action = d->handler.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
if (!emulate)
- return actions[d->handler.action].decoder(insn,
- asi, h);
+ return actions[action].decoder(insn, asi, h);
- asi->insn_handler = actions[d->handler.action].handler;
+ asi->insn_handler = actions[action].handler;
set_emulated_insn(insn, asi, thumb);
return INSN_GOOD;
}
diff --git a/arch/arm/probes/decode.h b/arch/arm/probes/decode.h
index 1d0b531..f9b08ba 100644
--- a/arch/arm/probes/decode.h
+++ b/arch/arm/probes/decode.h
@@ -314,6 +314,14 @@ union decode_action {
probes_custom_decode_t *decoder;
};
+typedef enum probes_insn (probes_check_t)(probes_opcode_t,
+ struct arch_probes_insn *,
+ const struct decode_header *);
+
+struct decode_checker {
+ probes_check_t *checker;
+};
+
#define DECODE_END \
{.bits = DECODE_TYPE_END}
@@ -402,6 +410,7 @@ probes_insn_handler_t probes_emulate_none;
int __kprobes
probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
const union decode_item *table, bool thumb, bool emulate,
- const union decode_action *actions);
+ const union decode_action *actions,
+ const struct decode_checker **checkers);
#endif
diff --git a/arch/arm/probes/kprobes/actions-arm.c b/arch/arm/probes/kprobes/actions-arm.c
index 8797879..3c3afa4 100644
--- a/arch/arm/probes/kprobes/actions-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -341,3 +341,5 @@ const union decode_action kprobes_arm_actions[NUM_PROBES_ARM_ACTIONS] = {
[PROBES_BRANCH] = {.handler = simulate_bbl},
[PROBES_LDMSTM] = {.decoder = kprobe_decode_ldmstm}
};
+
+const struct decode_checker *kprobes_arm_checkers[] = {NULL};
diff --git a/arch/arm/probes/kprobes/actions-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
index 6c4e60b..2796121 100644
--- a/arch/arm/probes/kprobes/actions-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -664,3 +664,6 @@ const union decode_action kprobes_t32_actions[NUM_PROBES_T32_ACTIONS] = {
[PROBES_T32_MUL_ADD_LONG] = {
.handler = t32_emulate_rdlo12rdhi8rn16rm0_noflags},
};
+
+const struct decode_checker *kprobes_t32_checkers[] = {NULL};
+const struct decode_checker *kprobes_t16_checkers[] = {NULL};
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 701f49d..74f3dc3 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -61,6 +61,7 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
kprobe_decode_insn_t *decode_insn;
const union decode_action *actions;
int is;
+ const struct decode_checker **checkers;
if (in_exception_text(addr))
return -EINVAL;
@@ -74,9 +75,11 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
insn = __opcode_thumb32_compose(insn, inst2);
decode_insn = thumb32_probes_decode_insn;
actions = kprobes_t32_actions;
+ checkers = kprobes_t32_checkers;
} else {
decode_insn = thumb16_probes_decode_insn;
actions = kprobes_t16_actions;
+ checkers = kprobes_t16_checkers;
}
#else /* !CONFIG_THUMB2_KERNEL */
thumb = false;
@@ -85,12 +88,13 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
insn = __mem_to_opcode_arm(*p->addr);
decode_insn = arm_probes_decode_insn;
actions = kprobes_arm_actions;
+ checkers = kprobes_arm_checkers;
#endif
p->opcode = insn;
p->ainsn.insn = tmp_insn;
- switch ((*decode_insn)(insn, &p->ainsn, true, actions)) {
+ switch ((*decode_insn)(insn, &p->ainsn, true, actions, checkers)) {
case INSN_REJECTED: /* not supported */
return -EINVAL;
diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
index 2e1e5a3..f88c79f 100644
--- a/arch/arm/probes/kprobes/core.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -37,16 +37,19 @@ kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
typedef enum probes_insn (kprobe_decode_insn_t)(probes_opcode_t,
struct arch_probes_insn *,
bool,
- const union decode_action *);
+ const union decode_action *,
+ const struct decode_checker *[*]);
#ifdef CONFIG_THUMB2_KERNEL
extern const union decode_action kprobes_t32_actions[];
extern const union decode_action kprobes_t16_actions[];
-
+extern const struct decode_checker *kprobes_t32_checkers[];
+extern const struct decode_checker *kprobes_t16_checkers[];
#else /* !CONFIG_THUMB2_KERNEL */
extern const union decode_action kprobes_arm_actions[];
+extern const struct decode_checker *kprobes_arm_checkers[];
#endif
diff --git a/arch/arm/probes/uprobes/core.c b/arch/arm/probes/uprobes/core.c
index b2954f6..d1329f1 100644
--- a/arch/arm/probes/uprobes/core.c
+++ b/arch/arm/probes/uprobes/core.c
@@ -88,7 +88,7 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe, struct mm_struct *mm,
auprobe->ixol[1] = __opcode_to_mem_arm(UPROBE_SS_ARM_INSN);
ret = arm_probes_decode_insn(insn, &auprobe->asi, false,
- uprobes_probes_actions);
+ uprobes_probes_actions, NULL);
switch (ret) {
case INSN_REJECTED:
return -EINVAL;
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 2/7] ARM: kprobes: introduces checker
@ 2014-12-08 6:27 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:27 UTC (permalink / raw)
To: linux-arm-kernel
This patch introdces 'checker' to decoding phase, and calls checkers
when instruction decoding. This allows further decoding for specific
instructions. This patch introduces a stub call of checkers in kprobe
arch_prepare_kprobe() as an example and for further expansion.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
---
v1 -> v2:
- kprobe checker stubs are introduced in this patch.
v2 -> v3:
- Code cleanups following Masami Hiramatsu and Tixy's advises.
- Commit message improvements.
v3 -> v4:
- Move to arch/arm/probes.
---
arch/arm/probes/decode-arm.c | 5 +--
arch/arm/probes/decode-arm.h | 3 +-
arch/arm/probes/decode-thumb.c | 10 +++---
arch/arm/probes/decode-thumb.h | 6 ++--
arch/arm/probes/decode.c | 60 +++++++++++++++++++++++++++++----
arch/arm/probes/decode.h | 11 +++++-
arch/arm/probes/kprobes/actions-arm.c | 2 ++
arch/arm/probes/kprobes/actions-thumb.c | 3 ++
arch/arm/probes/kprobes/core.c | 6 +++-
arch/arm/probes/kprobes/core.h | 7 ++--
arch/arm/probes/uprobes/core.c | 2 +-
11 files changed, 95 insertions(+), 20 deletions(-)
diff --git a/arch/arm/probes/decode-arm.c b/arch/arm/probes/decode-arm.c
index e39cc75..f46d8fc 100644
--- a/arch/arm/probes/decode-arm.c
+++ b/arch/arm/probes/decode-arm.c
@@ -726,10 +726,11 @@ static void __kprobes arm_singlestep(probes_opcode_t insn,
*/
enum probes_insn __kprobes
arm_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = arm_singlestep;
asi->insn_check_cc = probes_condition_checks[insn>>28];
return probes_decode_insn(insn, asi, probes_decode_arm_table, false,
- emulate, actions);
+ emulate, actions, checkers);
}
diff --git a/arch/arm/probes/decode-arm.h b/arch/arm/probes/decode-arm.h
index 9c56b40..a7b0398 100644
--- a/arch/arm/probes/decode-arm.h
+++ b/arch/arm/probes/decode-arm.h
@@ -70,6 +70,7 @@ extern const union decode_item probes_decode_arm_table[];
enum probes_insn arm_probes_decode_insn(probes_opcode_t,
struct arch_probes_insn *, bool emulate,
- const union decode_action *actions);
+ const union decode_action *actions,
+ const struct decode_checker *checkers[]);
#endif
diff --git a/arch/arm/probes/decode-thumb.c b/arch/arm/probes/decode-thumb.c
index 2f0453a..985e7dd 100644
--- a/arch/arm/probes/decode-thumb.c
+++ b/arch/arm/probes/decode-thumb.c
@@ -863,20 +863,22 @@ static void __kprobes thumb32_singlestep(probes_opcode_t opcode,
enum probes_insn __kprobes
thumb16_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = thumb16_singlestep;
asi->insn_check_cc = thumb_check_cc;
return probes_decode_insn(insn, asi, probes_decode_thumb16_table, true,
- emulate, actions);
+ emulate, actions, checkers);
}
enum probes_insn __kprobes
thumb32_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
asi->insn_singlestep = thumb32_singlestep;
asi->insn_check_cc = thumb_check_cc;
return probes_decode_insn(insn, asi, probes_decode_thumb32_table, true,
- emulate, actions);
+ emulate, actions, checkers);
}
diff --git a/arch/arm/probes/decode-thumb.h b/arch/arm/probes/decode-thumb.h
index 039013c..8457add 100644
--- a/arch/arm/probes/decode-thumb.h
+++ b/arch/arm/probes/decode-thumb.h
@@ -91,9 +91,11 @@ extern const union decode_item probes_decode_thumb16_table[];
enum probes_insn __kprobes
thumb16_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions);
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[]);
enum probes_insn __kprobes
thumb32_probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
- bool emulate, const union decode_action *actions);
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[]);
#endif
diff --git a/arch/arm/probes/decode.c b/arch/arm/probes/decode.c
index 3b05d57..c7d4420 100644
--- a/arch/arm/probes/decode.c
+++ b/arch/arm/probes/decode.c
@@ -342,6 +342,31 @@ static const int decode_struct_sizes[NUM_DECODE_TYPES] = {
[DECODE_TYPE_REJECT] = sizeof(struct decode_reject)
};
+static int run_checkers(const struct decode_checker *checkers[],
+ int action, probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ const struct decode_checker **p;
+
+ if (!checkers)
+ return INSN_GOOD;
+
+ p = checkers;
+ while (*p != NULL) {
+ int retval;
+ probes_check_t *checker_func = (*p)[action].checker;
+
+ retval = INSN_GOOD;
+ if (checker_func)
+ retval = checker_func(insn, asi, h);
+ if (retval == INSN_REJECTED)
+ return retval;
+ p++;
+ }
+ return INSN_GOOD;
+}
+
/*
* probes_decode_insn operates on data tables in order to decode an ARM
* architecture instruction onto which a kprobe has been placed.
@@ -388,11 +413,17 @@ static const int decode_struct_sizes[NUM_DECODE_TYPES] = {
int __kprobes
probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
const union decode_item *table, bool thumb,
- bool emulate, const union decode_action *actions)
+ bool emulate, const union decode_action *actions,
+ const struct decode_checker *checkers[])
{
const struct decode_header *h = (struct decode_header *)table;
const struct decode_header *next;
bool matched = false;
+ /*
+ * @insn can be modified by decode_regs. Save its original
+ * value for checkers.
+ */
+ probes_opcode_t origin_insn = insn;
if (emulate)
insn = prepare_emulated_insn(insn, asi, thumb);
@@ -422,24 +453,41 @@ probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
}
case DECODE_TYPE_CUSTOM: {
+ int err;
struct decode_custom *d = (struct decode_custom *)h;
- return actions[d->decoder.action].decoder(insn, asi, h);
+ int action = d->decoder.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
+ return actions[action].decoder(insn, asi, h);
}
case DECODE_TYPE_SIMULATE: {
+ int err;
struct decode_simulate *d = (struct decode_simulate *)h;
- asi->insn_handler = actions[d->handler.action].handler;
+ int action = d->handler.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
+ asi->insn_handler = actions[action].handler;
return INSN_GOOD_NO_SLOT;
}
case DECODE_TYPE_EMULATE: {
+ int err;
struct decode_emulate *d = (struct decode_emulate *)h;
+ int action = d->handler.action;
+
+ err = run_checkers(checkers, action, origin_insn, asi, h);
+ if (err == INSN_REJECTED)
+ return INSN_REJECTED;
if (!emulate)
- return actions[d->handler.action].decoder(insn,
- asi, h);
+ return actions[action].decoder(insn, asi, h);
- asi->insn_handler = actions[d->handler.action].handler;
+ asi->insn_handler = actions[action].handler;
set_emulated_insn(insn, asi, thumb);
return INSN_GOOD;
}
diff --git a/arch/arm/probes/decode.h b/arch/arm/probes/decode.h
index 1d0b531..f9b08ba 100644
--- a/arch/arm/probes/decode.h
+++ b/arch/arm/probes/decode.h
@@ -314,6 +314,14 @@ union decode_action {
probes_custom_decode_t *decoder;
};
+typedef enum probes_insn (probes_check_t)(probes_opcode_t,
+ struct arch_probes_insn *,
+ const struct decode_header *);
+
+struct decode_checker {
+ probes_check_t *checker;
+};
+
#define DECODE_END \
{.bits = DECODE_TYPE_END}
@@ -402,6 +410,7 @@ probes_insn_handler_t probes_emulate_none;
int __kprobes
probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
const union decode_item *table, bool thumb, bool emulate,
- const union decode_action *actions);
+ const union decode_action *actions,
+ const struct decode_checker **checkers);
#endif
diff --git a/arch/arm/probes/kprobes/actions-arm.c b/arch/arm/probes/kprobes/actions-arm.c
index 8797879..3c3afa4 100644
--- a/arch/arm/probes/kprobes/actions-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -341,3 +341,5 @@ const union decode_action kprobes_arm_actions[NUM_PROBES_ARM_ACTIONS] = {
[PROBES_BRANCH] = {.handler = simulate_bbl},
[PROBES_LDMSTM] = {.decoder = kprobe_decode_ldmstm}
};
+
+const struct decode_checker *kprobes_arm_checkers[] = {NULL};
diff --git a/arch/arm/probes/kprobes/actions-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
index 6c4e60b..2796121 100644
--- a/arch/arm/probes/kprobes/actions-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -664,3 +664,6 @@ const union decode_action kprobes_t32_actions[NUM_PROBES_T32_ACTIONS] = {
[PROBES_T32_MUL_ADD_LONG] = {
.handler = t32_emulate_rdlo12rdhi8rn16rm0_noflags},
};
+
+const struct decode_checker *kprobes_t32_checkers[] = {NULL};
+const struct decode_checker *kprobes_t16_checkers[] = {NULL};
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 701f49d..74f3dc3 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -61,6 +61,7 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
kprobe_decode_insn_t *decode_insn;
const union decode_action *actions;
int is;
+ const struct decode_checker **checkers;
if (in_exception_text(addr))
return -EINVAL;
@@ -74,9 +75,11 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
insn = __opcode_thumb32_compose(insn, inst2);
decode_insn = thumb32_probes_decode_insn;
actions = kprobes_t32_actions;
+ checkers = kprobes_t32_checkers;
} else {
decode_insn = thumb16_probes_decode_insn;
actions = kprobes_t16_actions;
+ checkers = kprobes_t16_checkers;
}
#else /* !CONFIG_THUMB2_KERNEL */
thumb = false;
@@ -85,12 +88,13 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
insn = __mem_to_opcode_arm(*p->addr);
decode_insn = arm_probes_decode_insn;
actions = kprobes_arm_actions;
+ checkers = kprobes_arm_checkers;
#endif
p->opcode = insn;
p->ainsn.insn = tmp_insn;
- switch ((*decode_insn)(insn, &p->ainsn, true, actions)) {
+ switch ((*decode_insn)(insn, &p->ainsn, true, actions, checkers)) {
case INSN_REJECTED: /* not supported */
return -EINVAL;
diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
index 2e1e5a3..f88c79f 100644
--- a/arch/arm/probes/kprobes/core.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -37,16 +37,19 @@ kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
typedef enum probes_insn (kprobe_decode_insn_t)(probes_opcode_t,
struct arch_probes_insn *,
bool,
- const union decode_action *);
+ const union decode_action *,
+ const struct decode_checker *[*]);
#ifdef CONFIG_THUMB2_KERNEL
extern const union decode_action kprobes_t32_actions[];
extern const union decode_action kprobes_t16_actions[];
-
+extern const struct decode_checker *kprobes_t32_checkers[];
+extern const struct decode_checker *kprobes_t16_checkers[];
#else /* !CONFIG_THUMB2_KERNEL */
extern const union decode_action kprobes_arm_actions[];
+extern const struct decode_checker *kprobes_arm_checkers[];
#endif
diff --git a/arch/arm/probes/uprobes/core.c b/arch/arm/probes/uprobes/core.c
index b2954f6..d1329f1 100644
--- a/arch/arm/probes/uprobes/core.c
+++ b/arch/arm/probes/uprobes/core.c
@@ -88,7 +88,7 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe, struct mm_struct *mm,
auprobe->ixol[1] = __opcode_to_mem_arm(UPROBE_SS_ARM_INSN);
ret = arm_probes_decode_insn(insn, &auprobe->asi, false,
- uprobes_probes_actions);
+ uprobes_probes_actions, NULL);
switch (ret) {
case INSN_REJECTED:
return -EINVAL;
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 3/7] ARM: kprobes: collects stack consumption for store instructions
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:28 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
This patch uses the previously introduced checker functionality on
store instructions to record their stack consumption information to
arch_probes_insn.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Jon Medhurst <tixy@linaro.org>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
---
v1 -> v2:
- Bugfix and code improvements following Tixy's suggestion. See:
http://www.spinics.net/lists/arm-kernel/msg372912.html
v2 -> v3:
- Totaly reconstructed following Tixy' instruction. See:
https://lkml.org/lkml/2014/10/27/662 .
Add his SOB.
v3 -> v4:
- Commit message improvements.
- Comments improvements and code cleanup.
- A bug is found and fixed in decode table in arm_check_stack().
v4 -> v5:
- Bugs in thumb2 decoding table is found by Tixy and fixed.
v5 -> v6:
- Move to arch/arm/probes/ .
---
arch/arm/include/asm/probes.h | 1 +
arch/arm/probes/decode.c | 10 +++
arch/arm/probes/kprobes/Makefile | 6 +-
arch/arm/probes/kprobes/actions-arm.c | 3 +-
arch/arm/probes/kprobes/actions-thumb.c | 5 +-
arch/arm/probes/kprobes/checkers-arm.c | 99 +++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers-common.c | 101 +++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers-thumb.c | 110 ++++++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers.h | 54 +++++++++++++++
9 files changed, 383 insertions(+), 6 deletions(-)
create mode 100644 arch/arm/probes/kprobes/checkers-arm.c
create mode 100644 arch/arm/probes/kprobes/checkers-common.c
create mode 100644 arch/arm/probes/kprobes/checkers-thumb.c
create mode 100644 arch/arm/probes/kprobes/checkers.h
diff --git a/arch/arm/include/asm/probes.h b/arch/arm/include/asm/probes.h
index 806cfe6..ccf9af3 100644
--- a/arch/arm/include/asm/probes.h
+++ b/arch/arm/include/asm/probes.h
@@ -38,6 +38,7 @@ struct arch_probes_insn {
probes_check_cc *insn_check_cc;
probes_insn_singlestep_t *insn_singlestep;
probes_insn_fn_t *insn_fn;
+ int stack_space;
};
#endif
diff --git a/arch/arm/probes/decode.c b/arch/arm/probes/decode.c
index c7d4420..f9d7c42 100644
--- a/arch/arm/probes/decode.c
+++ b/arch/arm/probes/decode.c
@@ -425,6 +425,16 @@ probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
*/
probes_opcode_t origin_insn = insn;
+ /*
+ * stack_space is initialized to 0 here. Checker functions
+ * should update is value if they find this is a stack store
+ * instruction: positive value means bytes of stack usage,
+ * negitive value means unable to determine stack usage
+ * statically. For instruction doesn't store to stack, checker
+ * do nothing with it.
+ */
+ asi->stack_space = 0;
+
if (emulate)
insn = prepare_emulated_insn(insn, asi, thumb);
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
index eb38a42..bc8d504 100644
--- a/arch/arm/probes/kprobes/Makefile
+++ b/arch/arm/probes/kprobes/Makefile
@@ -1,11 +1,11 @@
-obj-$(CONFIG_KPROBES) += core.o actions-common.o
+obj-$(CONFIG_KPROBES) += core.o actions-common.o checkers-common.o
obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
test-kprobes-objs := test-core.o
ifdef CONFIG_THUMB2_KERNEL
-obj-$(CONFIG_KPROBES) += actions-thumb.o
+obj-$(CONFIG_KPROBES) += actions-thumb.o checkers-thumb.o
test-kprobes-objs += test-thumb.o
else
-obj-$(CONFIG_KPROBES) += actions-arm.o
+obj-$(CONFIG_KPROBES) += actions-arm.o checkers-arm.o
test-kprobes-objs += test-arm.o
endif
diff --git a/arch/arm/probes/kprobes/actions-arm.c b/arch/arm/probes/kprobes/actions-arm.c
index 3c3afa4..1e67839 100644
--- a/arch/arm/probes/kprobes/actions-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -64,6 +64,7 @@
#include "../decode-arm.h"
#include "core.h"
+#include "checkers.h"
#if __LINUX_ARM_ARCH__ >= 6
#define BLX(reg) "blx "reg" \n\t"
@@ -342,4 +343,4 @@ const union decode_action kprobes_arm_actions[NUM_PROBES_ARM_ACTIONS] = {
[PROBES_LDMSTM] = {.decoder = kprobe_decode_ldmstm}
};
-const struct decode_checker *kprobes_arm_checkers[] = {NULL};
+const struct decode_checker *kprobes_arm_checkers[] = {arm_stack_checker, NULL};
diff --git a/arch/arm/probes/kprobes/actions-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
index 2796121..07cfd9b 100644
--- a/arch/arm/probes/kprobes/actions-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -15,6 +15,7 @@
#include "../decode-thumb.h"
#include "core.h"
+#include "checkers.h"
/* These emulation encodings are functionally equivalent... */
#define t32_emulate_rd8rn16rm0ra12_noflags \
@@ -665,5 +666,5 @@ const union decode_action kprobes_t32_actions[NUM_PROBES_T32_ACTIONS] = {
.handler = t32_emulate_rdlo12rdhi8rn16rm0_noflags},
};
-const struct decode_checker *kprobes_t32_checkers[] = {NULL};
-const struct decode_checker *kprobes_t16_checkers[] = {NULL};
+const struct decode_checker *kprobes_t32_checkers[] = {t32_stack_checker, NULL};
+const struct decode_checker *kprobes_t16_checkers[] = {t16_stack_checker, NULL};
diff --git a/arch/arm/probes/kprobes/checkers-arm.c b/arch/arm/probes/kprobes/checkers-arm.c
new file mode 100644
index 0000000..f817663
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-arm.c
@@ -0,0 +1,99 @@
+/*
+ * arch/arm/probes/kprobes/checkers-arm.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "checkers.h"
+
+static enum probes_insn __kprobes arm_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ /*
+ * PROBES_LDRSTRD, PROBES_LDMSTM, PROBES_STORE,
+ * PROBES_STORE_EXTRA may get here. Simply mark all normal
+ * insns as STACK_USE_NONE.
+ */
+ static const union decode_item table[] = {
+ /*
+ * 'STR{,D,B,H}, Rt, [Rn, Rm]' should be marked as UNKNOWN
+ * if Rn or Rm is SP.
+ * x
+ * STR (register) cccc 011x x0x0 xxxx xxxx xxxx xxxx xxxx
+ * STRB (register) cccc 011x x1x0 xxxx xxxx xxxx xxxx xxxx
+ */
+ DECODE_OR (0x0e10000f, 0x0600000d),
+ DECODE_OR (0x0e1f0000, 0x060d0000),
+
+ /*
+ * x
+ * STRD (register) cccc 000x x0x0 xxxx xxxx xxxx 1111 xxxx
+ * STRH (register) cccc 000x x0x0 xxxx xxxx xxxx 1011 xxxx
+ */
+ DECODE_OR (0x0e5000bf, 0x000000bd),
+ DECODE_CUSTOM (0x0e5f00b0, 0x000d00b0, STACK_USE_UNKNOWN),
+
+ /*
+ * For PROBES_LDMSTM, only stmdx sp, [...] need to examine
+ *
+ * Bit B/A (bit 24) encodes arithmetic operation order. 1 means
+ * before, 0 means after.
+ * Bit I/D (bit 23) encodes arithmetic operation. 1 means
+ * increment, 0 means decrement.
+ *
+ * So:
+ * B I
+ * / /
+ * A D | Rn |
+ * STMDX SP, [...] cccc 100x 00x0 xxxx xxxx xxxx xxxx xxxx
+ */
+ DECODE_CUSTOM (0x0edf0000, 0x080d0000, STACK_USE_STMDX),
+
+ /* P U W | Rn | Rt | imm12 |*/
+ /* STR (immediate) cccc 010x x0x0 1101 xxxx xxxx xxxx xxxx */
+ /* STRB (immediate) cccc 010x x1x0 1101 xxxx xxxx xxxx xxxx */
+ /* P U W | Rn | Rt |imm4| |imm4|*/
+ /* STRD (immediate) cccc 000x x1x0 1101 xxxx xxxx 1111 xxxx */
+ /* STRH (immediate) cccc 000x x1x0 1101 xxxx xxxx 1011 xxxx */
+ /*
+ * index = (P == '1'); add = (U == '1').
+ * Above insns with:
+ * index == 0 (str{,d,h} rx, [sp], #+/-imm) or
+ * add == 1 (str{,d,h} rx, [sp, #+<imm>])
+ * should be STACK_USE_NONE.
+ * Only str{,b,d,h} rx,[sp,#-n] (P == 1 and U == 0) are
+ * required to be examined.
+ */
+ /* STR{,B} Rt,[SP,#-n] cccc 0101 0xx0 1101 xxxx xxxx xxxx xxxx */
+ DECODE_CUSTOM (0x0f9f0000, 0x050d0000, STACK_USE_FIXED_XXX),
+
+ /* STR{D,H} Rt,[SP,#-n] cccc 0001 01x0 1101 xxxx xxxx 1x11 xxxx */
+ DECODE_CUSTOM (0x0fdf00b0, 0x014d00b0, STACK_USE_FIXED_X0X),
+
+ /* fall through */
+ DECODE_CUSTOM (0, 0, STACK_USE_NONE),
+ DECODE_END
+ };
+
+ return probes_decode_insn(insn, asi, table, false, false, stack_check_actions, NULL);
+}
+
+const struct decode_checker arm_stack_checker[NUM_PROBES_ARM_ACTIONS] = {
+ [PROBES_LDRSTRD] = {.checker = arm_check_stack},
+ [PROBES_STORE_EXTRA] = {.checker = arm_check_stack},
+ [PROBES_STORE] = {.checker = arm_check_stack},
+ [PROBES_LDMSTM] = {.checker = arm_check_stack},
+};
diff --git a/arch/arm/probes/kprobes/checkers-common.c b/arch/arm/probes/kprobes/checkers-common.c
new file mode 100644
index 0000000..971119c
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-common.c
@@ -0,0 +1,101 @@
+/*
+ * arch/arm/probes/kprobes/checkers-common.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "checkers.h"
+
+enum probes_insn checker_stack_use_none(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ asi->stack_space = 0;
+ return INSN_GOOD_NO_SLOT;
+}
+
+enum probes_insn checker_stack_use_unknown(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ asi->stack_space = -1;
+ return INSN_GOOD_NO_SLOT;
+}
+
+#ifdef CONFIG_THUMB2_KERNEL
+enum probes_insn checker_stack_use_imm_0xx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xff;
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+
+/*
+ * Different from other insn uses imm8, the real addressing offset of
+ * STRD in T32 encoding should be imm8 * 4. See ARMARM description.
+ */
+enum probes_insn checker_stack_use_t32strd(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xff;
+ asi->stack_space = imm << 2;
+ return INSN_GOOD_NO_SLOT;
+}
+#else
+enum probes_insn checker_stack_use_imm_x0x(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = ((insn & 0xf00) >> 4) + (insn & 0xf);
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+#endif
+
+enum probes_insn checker_stack_use_imm_xxx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xfff;
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+
+enum probes_insn checker_stack_use_stmdx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ unsigned int reglist = insn & 0xffff;
+ int pbit = insn & (1 << 24);
+ asi->stack_space = (hweight32(reglist) - (!pbit ? 1 : 0)) * 4;
+
+ return INSN_GOOD_NO_SLOT;
+}
+
+const union decode_action stack_check_actions[] = {
+ [STACK_USE_NONE] = {.decoder = checker_stack_use_none},
+ [STACK_USE_UNKNOWN] = {.decoder = checker_stack_use_unknown},
+#ifdef CONFIG_THUMB2_KERNEL
+ [STACK_USE_FIXED_0XX] = {.decoder = checker_stack_use_imm_0xx},
+ [STACK_USE_T32STRD] = {.decoder = checker_stack_use_t32strd},
+#else
+ [STACK_USE_FIXED_X0X] = {.decoder = checker_stack_use_imm_x0x},
+#endif
+ [STACK_USE_FIXED_XXX] = {.decoder = checker_stack_use_imm_xxx},
+ [STACK_USE_STMDX] = {.decoder = checker_stack_use_stmdx},
+};
diff --git a/arch/arm/probes/kprobes/checkers-thumb.c b/arch/arm/probes/kprobes/checkers-thumb.c
new file mode 100644
index 0000000..d608e3b
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-thumb.c
@@ -0,0 +1,110 @@
+/*
+ * arch/arm/probes/kprobes/checkers-thumb.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-thumb.h"
+#include "checkers.h"
+
+static enum probes_insn __kprobes t32_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ /*
+ * PROBES_T32_LDMSTM, PROBES_T32_LDRDSTRD and PROBES_T32_LDRSTR
+ * may get here. Simply mark all normal insns as STACK_USE_NONE.
+ */
+ static const union decode_item table[] = {
+
+ /*
+ * First, filter out all ldr insns to make our life easier.
+ * Following load insns may come here:
+ * LDM, LDRD, LDR.
+ * In T32 encoding, bit 20 is enough for distinguishing
+ * load and store. All load insns have this bit set, when
+ * all store insns have this bit clear.
+ */
+ DECODE_CUSTOM (0x00100000, 0x00100000, STACK_USE_NONE),
+
+ /*
+ * Mark all 'STR{,B,H}, Rt, [Rn, Rm]' as STACK_USE_UNKNOWN
+ * if Rn or Rm is SP. T32 doesn't encode STRD.
+ */
+ /* xx | Rn | Rt | | Rm |*/
+ /* STR (register) 1111 1000 0100 xxxx xxxx 0000 00xx xxxx */
+ /* STRB (register) 1111 1000 0000 xxxx xxxx 0000 00xx xxxx */
+ /* STRH (register) 1111 1000 0010 xxxx xxxx 0000 00xx xxxx */
+ /* INVALID INSN 1111 1000 0110 xxxx xxxx 0000 00xx xxxx */
+ /* By Introducing INVALID INSN, bit 21 and 22 can be ignored. */
+ DECODE_OR (0xff9f0fc0, 0xf80d0000),
+ DECODE_CUSTOM (0xff900fcf, 0xf800000d, STACK_USE_UNKNOWN),
+
+
+ /* xx | Rn | Rt | PUW| imm8 |*/
+ /* STR (imm 8) 1111 1000 0100 1101 xxxx 110x xxxx xxxx */
+ /* STRB (imm 8) 1111 1000 0000 1101 xxxx 110x xxxx xxxx */
+ /* STRH (imm 8) 1111 1000 0010 1101 xxxx 110x xxxx xxxx */
+ /* INVALID INSN 1111 1000 0110 1101 xxxx 110x xxxx xxxx */
+ /* Only consider U == 0 and P == 1: strx rx, [sp, #-<imm>] */
+ DECODE_CUSTOM (0xff9f0e00, 0xf80d0c00, STACK_USE_FIXED_0XX),
+
+ /* For STR{,B,H} (imm 12), offset is always positive, so ignore them. */
+
+ /* P U W | Rn | Rt | Rt2| imm8 |*/
+ /* STRD (immediate) 1110 1001 01x0 1101 xxxx xxxx xxxx xxxx */
+ /*
+ * Only consider U == 0 and P == 1.
+ * Also note that STRD in T32 encoding is special:
+ * imm = ZeroExtend(imm8:'00', 32)
+ */
+ DECODE_CUSTOM (0xffdf0000, 0xe94d0000, STACK_USE_T32STRD),
+
+ /* | Rn | */
+ /* STMDB 1110 1001 00x0 1101 xxxx xxxx xxxx xxxx */
+ DECODE_CUSTOM (0xffdf0000, 0xe90d0000, STACK_USE_STMDX),
+
+ /* fall through */
+ DECODE_CUSTOM (0, 0, STACK_USE_NONE),
+ DECODE_END
+ };
+
+ return probes_decode_insn(insn, asi, table, false, false, stack_check_actions, NULL);
+}
+
+const struct decode_checker t32_stack_checker[NUM_PROBES_T32_ACTIONS] = {
+ [PROBES_T32_LDMSTM] = {.checker = t32_check_stack},
+ [PROBES_T32_LDRDSTRD] = {.checker = t32_check_stack},
+ [PROBES_T32_LDRSTR] = {.checker = t32_check_stack},
+};
+
+/*
+ * See following comments. This insn must be 'push'.
+ */
+static enum probes_insn __kprobes t16_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ unsigned int reglist = insn & 0x1ff;
+ asi->stack_space = hweight32(reglist) * 4;
+ return INSN_GOOD;
+}
+
+/*
+ * T16 encoding is simple: only the 'push' insn can need extra stack space.
+ * Other insns, like str, can only use r0-r7 as Rn.
+ */
+const struct decode_checker t16_stack_checker[NUM_PROBES_T16_ACTIONS] = {
+ [PROBES_T16_PUSH] = {.checker = t16_check_stack},
+};
diff --git a/arch/arm/probes/kprobes/checkers.h b/arch/arm/probes/kprobes/checkers.h
new file mode 100644
index 0000000..bddfa0e
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers.h
@@ -0,0 +1,54 @@
+/*
+ * arch/arm/probes/kprobes/checkers.h
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+#ifndef _ARM_KERNEL_PROBES_CHECKERS_H
+#define _ARM_KERNEL_PROBES_CHECKERS_H
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include "../decode.h"
+
+extern probes_check_t checker_stack_use_none;
+extern probes_check_t checker_stack_use_unknown;
+#ifdef CONFIG_THUMB2_KERNEL
+extern probes_check_t checker_stack_use_imm_0xx;
+#else
+extern probes_check_t checker_stack_use_imm_x0x;
+#endif
+extern probes_check_t checker_stack_use_imm_xxx;
+extern probes_check_t checker_stack_use_stmdx;
+
+enum {
+ STACK_USE_NONE,
+ STACK_USE_UNKNOWN,
+#ifdef CONFIG_THUMB2_KERNEL
+ STACK_USE_FIXED_0XX,
+ STACK_USE_T32STRD,
+#else
+ STACK_USE_FIXED_X0X,
+#endif
+ STACK_USE_FIXED_XXX,
+ STACK_USE_STMDX,
+ NUM_STACK_USE_TYPES
+};
+
+extern const union decode_action stack_check_actions[];
+
+#ifndef CONFIG_THUMB2_KERNEL
+extern const struct decode_checker arm_stack_checker[];
+#else
+#endif
+extern const struct decode_checker t32_stack_checker[];
+extern const struct decode_checker t16_stack_checker[];
+#endif
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 3/7] ARM: kprobes: collects stack consumption for store instructions
@ 2014-12-08 6:28 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: linux-arm-kernel
This patch uses the previously introduced checker functionality on
store instructions to record their stack consumption information to
arch_probes_insn.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Jon Medhurst <tixy@linaro.org>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
---
v1 -> v2:
- Bugfix and code improvements following Tixy's suggestion. See:
http://www.spinics.net/lists/arm-kernel/msg372912.html
v2 -> v3:
- Totaly reconstructed following Tixy' instruction. See:
https://lkml.org/lkml/2014/10/27/662 .
Add his SOB.
v3 -> v4:
- Commit message improvements.
- Comments improvements and code cleanup.
- A bug is found and fixed in decode table in arm_check_stack().
v4 -> v5:
- Bugs in thumb2 decoding table is found by Tixy and fixed.
v5 -> v6:
- Move to arch/arm/probes/ .
---
arch/arm/include/asm/probes.h | 1 +
arch/arm/probes/decode.c | 10 +++
arch/arm/probes/kprobes/Makefile | 6 +-
arch/arm/probes/kprobes/actions-arm.c | 3 +-
arch/arm/probes/kprobes/actions-thumb.c | 5 +-
arch/arm/probes/kprobes/checkers-arm.c | 99 +++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers-common.c | 101 +++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers-thumb.c | 110 ++++++++++++++++++++++++++++++
arch/arm/probes/kprobes/checkers.h | 54 +++++++++++++++
9 files changed, 383 insertions(+), 6 deletions(-)
create mode 100644 arch/arm/probes/kprobes/checkers-arm.c
create mode 100644 arch/arm/probes/kprobes/checkers-common.c
create mode 100644 arch/arm/probes/kprobes/checkers-thumb.c
create mode 100644 arch/arm/probes/kprobes/checkers.h
diff --git a/arch/arm/include/asm/probes.h b/arch/arm/include/asm/probes.h
index 806cfe6..ccf9af3 100644
--- a/arch/arm/include/asm/probes.h
+++ b/arch/arm/include/asm/probes.h
@@ -38,6 +38,7 @@ struct arch_probes_insn {
probes_check_cc *insn_check_cc;
probes_insn_singlestep_t *insn_singlestep;
probes_insn_fn_t *insn_fn;
+ int stack_space;
};
#endif
diff --git a/arch/arm/probes/decode.c b/arch/arm/probes/decode.c
index c7d4420..f9d7c42 100644
--- a/arch/arm/probes/decode.c
+++ b/arch/arm/probes/decode.c
@@ -425,6 +425,16 @@ probes_decode_insn(probes_opcode_t insn, struct arch_probes_insn *asi,
*/
probes_opcode_t origin_insn = insn;
+ /*
+ * stack_space is initialized to 0 here. Checker functions
+ * should update is value if they find this is a stack store
+ * instruction: positive value means bytes of stack usage,
+ * negitive value means unable to determine stack usage
+ * statically. For instruction doesn't store to stack, checker
+ * do nothing with it.
+ */
+ asi->stack_space = 0;
+
if (emulate)
insn = prepare_emulated_insn(insn, asi, thumb);
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
index eb38a42..bc8d504 100644
--- a/arch/arm/probes/kprobes/Makefile
+++ b/arch/arm/probes/kprobes/Makefile
@@ -1,11 +1,11 @@
-obj-$(CONFIG_KPROBES) += core.o actions-common.o
+obj-$(CONFIG_KPROBES) += core.o actions-common.o checkers-common.o
obj-$(CONFIG_ARM_KPROBES_TEST) += test-kprobes.o
test-kprobes-objs := test-core.o
ifdef CONFIG_THUMB2_KERNEL
-obj-$(CONFIG_KPROBES) += actions-thumb.o
+obj-$(CONFIG_KPROBES) += actions-thumb.o checkers-thumb.o
test-kprobes-objs += test-thumb.o
else
-obj-$(CONFIG_KPROBES) += actions-arm.o
+obj-$(CONFIG_KPROBES) += actions-arm.o checkers-arm.o
test-kprobes-objs += test-arm.o
endif
diff --git a/arch/arm/probes/kprobes/actions-arm.c b/arch/arm/probes/kprobes/actions-arm.c
index 3c3afa4..1e67839 100644
--- a/arch/arm/probes/kprobes/actions-arm.c
+++ b/arch/arm/probes/kprobes/actions-arm.c
@@ -64,6 +64,7 @@
#include "../decode-arm.h"
#include "core.h"
+#include "checkers.h"
#if __LINUX_ARM_ARCH__ >= 6
#define BLX(reg) "blx "reg" \n\t"
@@ -342,4 +343,4 @@ const union decode_action kprobes_arm_actions[NUM_PROBES_ARM_ACTIONS] = {
[PROBES_LDMSTM] = {.decoder = kprobe_decode_ldmstm}
};
-const struct decode_checker *kprobes_arm_checkers[] = {NULL};
+const struct decode_checker *kprobes_arm_checkers[] = {arm_stack_checker, NULL};
diff --git a/arch/arm/probes/kprobes/actions-thumb.c b/arch/arm/probes/kprobes/actions-thumb.c
index 2796121..07cfd9b 100644
--- a/arch/arm/probes/kprobes/actions-thumb.c
+++ b/arch/arm/probes/kprobes/actions-thumb.c
@@ -15,6 +15,7 @@
#include "../decode-thumb.h"
#include "core.h"
+#include "checkers.h"
/* These emulation encodings are functionally equivalent... */
#define t32_emulate_rd8rn16rm0ra12_noflags \
@@ -665,5 +666,5 @@ const union decode_action kprobes_t32_actions[NUM_PROBES_T32_ACTIONS] = {
.handler = t32_emulate_rdlo12rdhi8rn16rm0_noflags},
};
-const struct decode_checker *kprobes_t32_checkers[] = {NULL};
-const struct decode_checker *kprobes_t16_checkers[] = {NULL};
+const struct decode_checker *kprobes_t32_checkers[] = {t32_stack_checker, NULL};
+const struct decode_checker *kprobes_t16_checkers[] = {t16_stack_checker, NULL};
diff --git a/arch/arm/probes/kprobes/checkers-arm.c b/arch/arm/probes/kprobes/checkers-arm.c
new file mode 100644
index 0000000..f817663
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-arm.c
@@ -0,0 +1,99 @@
+/*
+ * arch/arm/probes/kprobes/checkers-arm.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "checkers.h"
+
+static enum probes_insn __kprobes arm_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ /*
+ * PROBES_LDRSTRD, PROBES_LDMSTM, PROBES_STORE,
+ * PROBES_STORE_EXTRA may get here. Simply mark all normal
+ * insns as STACK_USE_NONE.
+ */
+ static const union decode_item table[] = {
+ /*
+ * 'STR{,D,B,H}, Rt, [Rn, Rm]' should be marked as UNKNOWN
+ * if Rn or Rm is SP.
+ * x
+ * STR (register) cccc 011x x0x0 xxxx xxxx xxxx xxxx xxxx
+ * STRB (register) cccc 011x x1x0 xxxx xxxx xxxx xxxx xxxx
+ */
+ DECODE_OR (0x0e10000f, 0x0600000d),
+ DECODE_OR (0x0e1f0000, 0x060d0000),
+
+ /*
+ * x
+ * STRD (register) cccc 000x x0x0 xxxx xxxx xxxx 1111 xxxx
+ * STRH (register) cccc 000x x0x0 xxxx xxxx xxxx 1011 xxxx
+ */
+ DECODE_OR (0x0e5000bf, 0x000000bd),
+ DECODE_CUSTOM (0x0e5f00b0, 0x000d00b0, STACK_USE_UNKNOWN),
+
+ /*
+ * For PROBES_LDMSTM, only stmdx sp, [...] need to examine
+ *
+ * Bit B/A (bit 24) encodes arithmetic operation order. 1 means
+ * before, 0 means after.
+ * Bit I/D (bit 23) encodes arithmetic operation. 1 means
+ * increment, 0 means decrement.
+ *
+ * So:
+ * B I
+ * / /
+ * A D | Rn |
+ * STMDX SP, [...] cccc 100x 00x0 xxxx xxxx xxxx xxxx xxxx
+ */
+ DECODE_CUSTOM (0x0edf0000, 0x080d0000, STACK_USE_STMDX),
+
+ /* P U W | Rn | Rt | imm12 |*/
+ /* STR (immediate) cccc 010x x0x0 1101 xxxx xxxx xxxx xxxx */
+ /* STRB (immediate) cccc 010x x1x0 1101 xxxx xxxx xxxx xxxx */
+ /* P U W | Rn | Rt |imm4| |imm4|*/
+ /* STRD (immediate) cccc 000x x1x0 1101 xxxx xxxx 1111 xxxx */
+ /* STRH (immediate) cccc 000x x1x0 1101 xxxx xxxx 1011 xxxx */
+ /*
+ * index = (P == '1'); add = (U == '1').
+ * Above insns with:
+ * index == 0 (str{,d,h} rx, [sp], #+/-imm) or
+ * add == 1 (str{,d,h} rx, [sp, #+<imm>])
+ * should be STACK_USE_NONE.
+ * Only str{,b,d,h} rx,[sp,#-n] (P == 1 and U == 0) are
+ * required to be examined.
+ */
+ /* STR{,B} Rt,[SP,#-n] cccc 0101 0xx0 1101 xxxx xxxx xxxx xxxx */
+ DECODE_CUSTOM (0x0f9f0000, 0x050d0000, STACK_USE_FIXED_XXX),
+
+ /* STR{D,H} Rt,[SP,#-n] cccc 0001 01x0 1101 xxxx xxxx 1x11 xxxx */
+ DECODE_CUSTOM (0x0fdf00b0, 0x014d00b0, STACK_USE_FIXED_X0X),
+
+ /* fall through */
+ DECODE_CUSTOM (0, 0, STACK_USE_NONE),
+ DECODE_END
+ };
+
+ return probes_decode_insn(insn, asi, table, false, false, stack_check_actions, NULL);
+}
+
+const struct decode_checker arm_stack_checker[NUM_PROBES_ARM_ACTIONS] = {
+ [PROBES_LDRSTRD] = {.checker = arm_check_stack},
+ [PROBES_STORE_EXTRA] = {.checker = arm_check_stack},
+ [PROBES_STORE] = {.checker = arm_check_stack},
+ [PROBES_LDMSTM] = {.checker = arm_check_stack},
+};
diff --git a/arch/arm/probes/kprobes/checkers-common.c b/arch/arm/probes/kprobes/checkers-common.c
new file mode 100644
index 0000000..971119c
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-common.c
@@ -0,0 +1,101 @@
+/*
+ * arch/arm/probes/kprobes/checkers-common.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-arm.h"
+#include "checkers.h"
+
+enum probes_insn checker_stack_use_none(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ asi->stack_space = 0;
+ return INSN_GOOD_NO_SLOT;
+}
+
+enum probes_insn checker_stack_use_unknown(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ asi->stack_space = -1;
+ return INSN_GOOD_NO_SLOT;
+}
+
+#ifdef CONFIG_THUMB2_KERNEL
+enum probes_insn checker_stack_use_imm_0xx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xff;
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+
+/*
+ * Different from other insn uses imm8, the real addressing offset of
+ * STRD in T32 encoding should be imm8 * 4. See ARMARM description.
+ */
+enum probes_insn checker_stack_use_t32strd(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xff;
+ asi->stack_space = imm << 2;
+ return INSN_GOOD_NO_SLOT;
+}
+#else
+enum probes_insn checker_stack_use_imm_x0x(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = ((insn & 0xf00) >> 4) + (insn & 0xf);
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+#endif
+
+enum probes_insn checker_stack_use_imm_xxx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ int imm = insn & 0xfff;
+ asi->stack_space = imm;
+ return INSN_GOOD_NO_SLOT;
+}
+
+enum probes_insn checker_stack_use_stmdx(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ unsigned int reglist = insn & 0xffff;
+ int pbit = insn & (1 << 24);
+ asi->stack_space = (hweight32(reglist) - (!pbit ? 1 : 0)) * 4;
+
+ return INSN_GOOD_NO_SLOT;
+}
+
+const union decode_action stack_check_actions[] = {
+ [STACK_USE_NONE] = {.decoder = checker_stack_use_none},
+ [STACK_USE_UNKNOWN] = {.decoder = checker_stack_use_unknown},
+#ifdef CONFIG_THUMB2_KERNEL
+ [STACK_USE_FIXED_0XX] = {.decoder = checker_stack_use_imm_0xx},
+ [STACK_USE_T32STRD] = {.decoder = checker_stack_use_t32strd},
+#else
+ [STACK_USE_FIXED_X0X] = {.decoder = checker_stack_use_imm_x0x},
+#endif
+ [STACK_USE_FIXED_XXX] = {.decoder = checker_stack_use_imm_xxx},
+ [STACK_USE_STMDX] = {.decoder = checker_stack_use_stmdx},
+};
diff --git a/arch/arm/probes/kprobes/checkers-thumb.c b/arch/arm/probes/kprobes/checkers-thumb.c
new file mode 100644
index 0000000..d608e3b
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers-thumb.c
@@ -0,0 +1,110 @@
+/*
+ * arch/arm/probes/kprobes/checkers-thumb.c
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include "../decode.h"
+#include "../decode-thumb.h"
+#include "checkers.h"
+
+static enum probes_insn __kprobes t32_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ /*
+ * PROBES_T32_LDMSTM, PROBES_T32_LDRDSTRD and PROBES_T32_LDRSTR
+ * may get here. Simply mark all normal insns as STACK_USE_NONE.
+ */
+ static const union decode_item table[] = {
+
+ /*
+ * First, filter out all ldr insns to make our life easier.
+ * Following load insns may come here:
+ * LDM, LDRD, LDR.
+ * In T32 encoding, bit 20 is enough for distinguishing
+ * load and store. All load insns have this bit set, when
+ * all store insns have this bit clear.
+ */
+ DECODE_CUSTOM (0x00100000, 0x00100000, STACK_USE_NONE),
+
+ /*
+ * Mark all 'STR{,B,H}, Rt, [Rn, Rm]' as STACK_USE_UNKNOWN
+ * if Rn or Rm is SP. T32 doesn't encode STRD.
+ */
+ /* xx | Rn | Rt | | Rm |*/
+ /* STR (register) 1111 1000 0100 xxxx xxxx 0000 00xx xxxx */
+ /* STRB (register) 1111 1000 0000 xxxx xxxx 0000 00xx xxxx */
+ /* STRH (register) 1111 1000 0010 xxxx xxxx 0000 00xx xxxx */
+ /* INVALID INSN 1111 1000 0110 xxxx xxxx 0000 00xx xxxx */
+ /* By Introducing INVALID INSN, bit 21 and 22 can be ignored. */
+ DECODE_OR (0xff9f0fc0, 0xf80d0000),
+ DECODE_CUSTOM (0xff900fcf, 0xf800000d, STACK_USE_UNKNOWN),
+
+
+ /* xx | Rn | Rt | PUW| imm8 |*/
+ /* STR (imm 8) 1111 1000 0100 1101 xxxx 110x xxxx xxxx */
+ /* STRB (imm 8) 1111 1000 0000 1101 xxxx 110x xxxx xxxx */
+ /* STRH (imm 8) 1111 1000 0010 1101 xxxx 110x xxxx xxxx */
+ /* INVALID INSN 1111 1000 0110 1101 xxxx 110x xxxx xxxx */
+ /* Only consider U == 0 and P == 1: strx rx, [sp, #-<imm>] */
+ DECODE_CUSTOM (0xff9f0e00, 0xf80d0c00, STACK_USE_FIXED_0XX),
+
+ /* For STR{,B,H} (imm 12), offset is always positive, so ignore them. */
+
+ /* P U W | Rn | Rt | Rt2| imm8 |*/
+ /* STRD (immediate) 1110 1001 01x0 1101 xxxx xxxx xxxx xxxx */
+ /*
+ * Only consider U == 0 and P == 1.
+ * Also note that STRD in T32 encoding is special:
+ * imm = ZeroExtend(imm8:'00', 32)
+ */
+ DECODE_CUSTOM (0xffdf0000, 0xe94d0000, STACK_USE_T32STRD),
+
+ /* | Rn | */
+ /* STMDB 1110 1001 00x0 1101 xxxx xxxx xxxx xxxx */
+ DECODE_CUSTOM (0xffdf0000, 0xe90d0000, STACK_USE_STMDX),
+
+ /* fall through */
+ DECODE_CUSTOM (0, 0, STACK_USE_NONE),
+ DECODE_END
+ };
+
+ return probes_decode_insn(insn, asi, table, false, false, stack_check_actions, NULL);
+}
+
+const struct decode_checker t32_stack_checker[NUM_PROBES_T32_ACTIONS] = {
+ [PROBES_T32_LDMSTM] = {.checker = t32_check_stack},
+ [PROBES_T32_LDRDSTRD] = {.checker = t32_check_stack},
+ [PROBES_T32_LDRSTR] = {.checker = t32_check_stack},
+};
+
+/*
+ * See following comments. This insn must be 'push'.
+ */
+static enum probes_insn __kprobes t16_check_stack(probes_opcode_t insn,
+ struct arch_probes_insn *asi,
+ const struct decode_header *h)
+{
+ unsigned int reglist = insn & 0x1ff;
+ asi->stack_space = hweight32(reglist) * 4;
+ return INSN_GOOD;
+}
+
+/*
+ * T16 encoding is simple: only the 'push' insn can need extra stack space.
+ * Other insns, like str, can only use r0-r7 as Rn.
+ */
+const struct decode_checker t16_stack_checker[NUM_PROBES_T16_ACTIONS] = {
+ [PROBES_T16_PUSH] = {.checker = t16_check_stack},
+};
diff --git a/arch/arm/probes/kprobes/checkers.h b/arch/arm/probes/kprobes/checkers.h
new file mode 100644
index 0000000..bddfa0e
--- /dev/null
+++ b/arch/arm/probes/kprobes/checkers.h
@@ -0,0 +1,54 @@
+/*
+ * arch/arm/probes/kprobes/checkers.h
+ *
+ * Copyright (C) 2014 Huawei Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+#ifndef _ARM_KERNEL_PROBES_CHECKERS_H
+#define _ARM_KERNEL_PROBES_CHECKERS_H
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include "../decode.h"
+
+extern probes_check_t checker_stack_use_none;
+extern probes_check_t checker_stack_use_unknown;
+#ifdef CONFIG_THUMB2_KERNEL
+extern probes_check_t checker_stack_use_imm_0xx;
+#else
+extern probes_check_t checker_stack_use_imm_x0x;
+#endif
+extern probes_check_t checker_stack_use_imm_xxx;
+extern probes_check_t checker_stack_use_stmdx;
+
+enum {
+ STACK_USE_NONE,
+ STACK_USE_UNKNOWN,
+#ifdef CONFIG_THUMB2_KERNEL
+ STACK_USE_FIXED_0XX,
+ STACK_USE_T32STRD,
+#else
+ STACK_USE_FIXED_X0X,
+#endif
+ STACK_USE_FIXED_XXX,
+ STACK_USE_STMDX,
+ NUM_STACK_USE_TYPES
+};
+
+extern const union decode_action stack_check_actions[];
+
+#ifndef CONFIG_THUMB2_KERNEL
+extern const struct decode_checker arm_stack_checker[];
+#else
+#endif
+extern const struct decode_checker t32_stack_checker[];
+extern const struct decode_checker t16_stack_checker[];
+#endif
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 4/7] ARM: kprobes: disallow probing stack consuming instructions
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:28 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
This patch prohibits probing instructions for which the stack
requirements are unable to be determined statically. Some test cases
are found not work again after the modification, this patch also
removes them.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
---
v1 -> v2:
- Use MAX_STACK_SIZE macro instead of hard coded stack size.
v2 -> v3:
- Commit message improvements.
v3 -> v4:
- Commit message improvements.
v4 -> v5:
- Move to arch/arm/probes.
---
arch/arm/include/asm/kprobes.h | 1 -
arch/arm/include/asm/probes.h | 12 ++++++++++++
arch/arm/kernel/entry-armv.S | 3 ++-
arch/arm/probes/kprobes/core.c | 9 +++++++++
arch/arm/probes/kprobes/test-arm.c | 16 ++++++++++------
5 files changed, 33 insertions(+), 8 deletions(-)
diff --git a/arch/arm/include/asm/kprobes.h b/arch/arm/include/asm/kprobes.h
index 49fa0df..56f9ac6 100644
--- a/arch/arm/include/asm/kprobes.h
+++ b/arch/arm/include/asm/kprobes.h
@@ -22,7 +22,6 @@
#define __ARCH_WANT_KPROBES_INSN_SLOT
#define MAX_INSN_SIZE 2
-#define MAX_STACK_SIZE 64 /* 32 would probably be OK */
#define flush_insn_slot(p) do { } while (0)
#define kretprobe_blacklist_size 0
diff --git a/arch/arm/include/asm/probes.h b/arch/arm/include/asm/probes.h
index ccf9af3..f0a1ee8 100644
--- a/arch/arm/include/asm/probes.h
+++ b/arch/arm/include/asm/probes.h
@@ -19,6 +19,8 @@
#ifndef _ASM_PROBES_H
#define _ASM_PROBES_H
+#ifndef __ASSEMBLY__
+
typedef u32 probes_opcode_t;
struct arch_probes_insn;
@@ -41,4 +43,14 @@ struct arch_probes_insn {
int stack_space;
};
+#endif /* __ASSEMBLY__ */
+
+/*
+ * We assume one instruction can consume at most 64 bytes stack, which is
+ * 'push {r0-r15}'. Instructions consume more or unknown stack space like
+ * 'str r0, [sp, #-80]' and 'str r0, [sp, r1]' should be prohibit to probe.
+ * Both kprobe and jprobe use this macro.
+ */
+#define MAX_STACK_SIZE 64
+
#endif
diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
index 2f5555d..672b219 100644
--- a/arch/arm/kernel/entry-armv.S
+++ b/arch/arm/kernel/entry-armv.S
@@ -31,6 +31,7 @@
#include "entry-header.S"
#include <asm/entry-macro-multi.S>
+#include <asm/probes.h>
/*
* Interrupt handling.
@@ -249,7 +250,7 @@ __und_svc:
@ If a kprobe is about to simulate a "stmdb sp..." instruction,
@ it obviously needs free stack space which then will belong to
@ the saved context.
- svc_entry 64
+ svc_entry MAX_STACK_SIZE
#else
svc_entry
#endif
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 74f3dc3..3a58db4 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -115,6 +115,15 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
break;
}
+ /*
+ * Never instrument insn like 'str r0, [sp, +/-r1]'. Also, insn likes
+ * 'str r0, [sp, #-68]' should also be prohibited.
+ * See __und_svc.
+ */
+ if ((p->ainsn.stack_space < 0) ||
+ (p->ainsn.stack_space > MAX_STACK_SIZE))
+ return -EINVAL;
+
return 0;
}
diff --git a/arch/arm/probes/kprobes/test-arm.c b/arch/arm/probes/kprobes/test-arm.c
index d9a1255..fdeb300 100644
--- a/arch/arm/probes/kprobes/test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -476,7 +476,8 @@ void kprobe_arm_test_cases(void)
TEST_GROUP("Extra load/store instructions")
TEST_RPR( "strh r",0, VAL1,", [r",1, 48,", -r",2, 24,"]")
- TEST_RPR( "streqh r",14,VAL2,", [r",13,0, ", r",12, 48,"]")
+ TEST_RPR( "streqh r",14,VAL2,", [r",11,0, ", r",12, 48,"]")
+ TEST_UNSUPPORTED( "streqh r14, [r13, r12]")
TEST_RPR( "strh r",1, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strneh r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strh r",2, VAL1,", [r",3, 24,"], r",4, 48,"")
@@ -565,7 +566,8 @@ void kprobe_arm_test_cases(void)
#if __LINUX_ARM_ARCH__ >= 5
TEST_RPR( "strd r",0, VAL1,", [r",1, 48,", -r",2,24,"]")
- TEST_RPR( "strccd r",8, VAL2,", [r",13,0, ", r",12,48,"]")
+ TEST_RPR( "strccd r",8, VAL2,", [r",11,0, ", r",12,48,"]")
+ TEST_UNSUPPORTED( "strccd r8, [r13, r12]")
TEST_RPR( "strd r",4, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strcsd r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strd r",2, VAL1,", [r",5, 24,"], r",4,48,"")
@@ -638,13 +640,15 @@ void kprobe_arm_test_cases(void)
TEST_RP( "str"byte" r",2, VAL1,", [r",3, 24,"], #48") \
TEST_RP( "str"byte" r",10,VAL2,", [r",9, 64,"], #-48") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 48,", -r",2, 24,"]") \
- TEST_RPR("str"byte" r",14,VAL2,", [r",13,0, ", r",12, 48,"]") \
+ TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 48,"]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 48,"]!") \
TEST_RPR("str"byte" r",12,VAL2,", [r",11,48,", -r",10,24,"]!") \
TEST_RPR("str"byte" r",2, VAL1,", [r",3, 24,"], r",4, 48,"") \
TEST_RPR("str"byte" r",10,VAL2,", [r",9, 48,"], -r",11,24,"") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 24,", r",2, 32,", asl #1]")\
- TEST_RPR("str"byte" r",14,VAL2,", [r",13,0, ", r",12, 32,", lsr #2]")\
+ TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 32,", lsr #2]")\
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]")\
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 32,", asr #3]!")\
TEST_RPR("str"byte" r",12,VAL2,", [r",11,24,", r",10, 4,", ror #31]!")\
TEST_P( "ldr"byte" r0, [r",0, 24,", #-2]") \
@@ -668,12 +672,12 @@ void kprobe_arm_test_cases(void)
LOAD_STORE("")
TEST_P( "str pc, [r",0,0,", #15*4]")
- TEST_R( "str pc, [sp, r",2,15*4,"]")
+ TEST_UNSUPPORTED( "str pc, [sp, r2]")
TEST_BF( "ldr pc, [sp, #15*4]")
TEST_BF_R("ldr pc, [sp, r",2,15*4,"]")
TEST_P( "str sp, [r",0,0,", #13*4]")
- TEST_R( "str sp, [sp, r",2,13*4,"]")
+ TEST_UNSUPPORTED( "str sp, [sp, r2]")
TEST_BF( "ldr sp, [sp, #13*4]")
TEST_BF_R("ldr sp, [sp, r",2,13*4,"]")
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 4/7] ARM: kprobes: disallow probing stack consuming instructions
@ 2014-12-08 6:28 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: linux-arm-kernel
This patch prohibits probing instructions for which the stack
requirements are unable to be determined statically. Some test cases
are found not work again after the modification, this patch also
removes them.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Reviewed-by: Jon Medhurst <tixy@linaro.org>
---
v1 -> v2:
- Use MAX_STACK_SIZE macro instead of hard coded stack size.
v2 -> v3:
- Commit message improvements.
v3 -> v4:
- Commit message improvements.
v4 -> v5:
- Move to arch/arm/probes.
---
arch/arm/include/asm/kprobes.h | 1 -
arch/arm/include/asm/probes.h | 12 ++++++++++++
arch/arm/kernel/entry-armv.S | 3 ++-
arch/arm/probes/kprobes/core.c | 9 +++++++++
arch/arm/probes/kprobes/test-arm.c | 16 ++++++++++------
5 files changed, 33 insertions(+), 8 deletions(-)
diff --git a/arch/arm/include/asm/kprobes.h b/arch/arm/include/asm/kprobes.h
index 49fa0df..56f9ac6 100644
--- a/arch/arm/include/asm/kprobes.h
+++ b/arch/arm/include/asm/kprobes.h
@@ -22,7 +22,6 @@
#define __ARCH_WANT_KPROBES_INSN_SLOT
#define MAX_INSN_SIZE 2
-#define MAX_STACK_SIZE 64 /* 32 would probably be OK */
#define flush_insn_slot(p) do { } while (0)
#define kretprobe_blacklist_size 0
diff --git a/arch/arm/include/asm/probes.h b/arch/arm/include/asm/probes.h
index ccf9af3..f0a1ee8 100644
--- a/arch/arm/include/asm/probes.h
+++ b/arch/arm/include/asm/probes.h
@@ -19,6 +19,8 @@
#ifndef _ASM_PROBES_H
#define _ASM_PROBES_H
+#ifndef __ASSEMBLY__
+
typedef u32 probes_opcode_t;
struct arch_probes_insn;
@@ -41,4 +43,14 @@ struct arch_probes_insn {
int stack_space;
};
+#endif /* __ASSEMBLY__ */
+
+/*
+ * We assume one instruction can consume at most 64 bytes stack, which is
+ * 'push {r0-r15}'. Instructions consume more or unknown stack space like
+ * 'str r0, [sp, #-80]' and 'str r0, [sp, r1]' should be prohibit to probe.
+ * Both kprobe and jprobe use this macro.
+ */
+#define MAX_STACK_SIZE 64
+
#endif
diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
index 2f5555d..672b219 100644
--- a/arch/arm/kernel/entry-armv.S
+++ b/arch/arm/kernel/entry-armv.S
@@ -31,6 +31,7 @@
#include "entry-header.S"
#include <asm/entry-macro-multi.S>
+#include <asm/probes.h>
/*
* Interrupt handling.
@@ -249,7 +250,7 @@ __und_svc:
@ If a kprobe is about to simulate a "stmdb sp..." instruction,
@ it obviously needs free stack space which then will belong to
@ the saved context.
- svc_entry 64
+ svc_entry MAX_STACK_SIZE
#else
svc_entry
#endif
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 74f3dc3..3a58db4 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -115,6 +115,15 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
break;
}
+ /*
+ * Never instrument insn like 'str r0, [sp, +/-r1]'. Also, insn likes
+ * 'str r0, [sp, #-68]' should also be prohibited.
+ * See __und_svc.
+ */
+ if ((p->ainsn.stack_space < 0) ||
+ (p->ainsn.stack_space > MAX_STACK_SIZE))
+ return -EINVAL;
+
return 0;
}
diff --git a/arch/arm/probes/kprobes/test-arm.c b/arch/arm/probes/kprobes/test-arm.c
index d9a1255..fdeb300 100644
--- a/arch/arm/probes/kprobes/test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -476,7 +476,8 @@ void kprobe_arm_test_cases(void)
TEST_GROUP("Extra load/store instructions")
TEST_RPR( "strh r",0, VAL1,", [r",1, 48,", -r",2, 24,"]")
- TEST_RPR( "streqh r",14,VAL2,", [r",13,0, ", r",12, 48,"]")
+ TEST_RPR( "streqh r",14,VAL2,", [r",11,0, ", r",12, 48,"]")
+ TEST_UNSUPPORTED( "streqh r14, [r13, r12]")
TEST_RPR( "strh r",1, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strneh r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strh r",2, VAL1,", [r",3, 24,"], r",4, 48,"")
@@ -565,7 +566,8 @@ void kprobe_arm_test_cases(void)
#if __LINUX_ARM_ARCH__ >= 5
TEST_RPR( "strd r",0, VAL1,", [r",1, 48,", -r",2,24,"]")
- TEST_RPR( "strccd r",8, VAL2,", [r",13,0, ", r",12,48,"]")
+ TEST_RPR( "strccd r",8, VAL2,", [r",11,0, ", r",12,48,"]")
+ TEST_UNSUPPORTED( "strccd r8, [r13, r12]")
TEST_RPR( "strd r",4, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strcsd r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strd r",2, VAL1,", [r",5, 24,"], r",4,48,"")
@@ -638,13 +640,15 @@ void kprobe_arm_test_cases(void)
TEST_RP( "str"byte" r",2, VAL1,", [r",3, 24,"], #48") \
TEST_RP( "str"byte" r",10,VAL2,", [r",9, 64,"], #-48") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 48,", -r",2, 24,"]") \
- TEST_RPR("str"byte" r",14,VAL2,", [r",13,0, ", r",12, 48,"]") \
+ TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 48,"]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 48,"]!") \
TEST_RPR("str"byte" r",12,VAL2,", [r",11,48,", -r",10,24,"]!") \
TEST_RPR("str"byte" r",2, VAL1,", [r",3, 24,"], r",4, 48,"") \
TEST_RPR("str"byte" r",10,VAL2,", [r",9, 48,"], -r",11,24,"") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 24,", r",2, 32,", asl #1]")\
- TEST_RPR("str"byte" r",14,VAL2,", [r",13,0, ", r",12, 32,", lsr #2]")\
+ TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 32,", lsr #2]")\
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]")\
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 32,", asr #3]!")\
TEST_RPR("str"byte" r",12,VAL2,", [r",11,24,", r",10, 4,", ror #31]!")\
TEST_P( "ldr"byte" r0, [r",0, 24,", #-2]") \
@@ -668,12 +672,12 @@ void kprobe_arm_test_cases(void)
LOAD_STORE("")
TEST_P( "str pc, [r",0,0,", #15*4]")
- TEST_R( "str pc, [sp, r",2,15*4,"]")
+ TEST_UNSUPPORTED( "str pc, [sp, r2]")
TEST_BF( "ldr pc, [sp, #15*4]")
TEST_BF_R("ldr pc, [sp, r",2,15*4,"]")
TEST_P( "str sp, [r",0,0,", #13*4]")
- TEST_R( "str sp, [sp, r",2,13*4,"]")
+ TEST_UNSUPPORTED( "str sp, [sp, r2]")
TEST_BF( "ldr sp, [sp, #13*4]")
TEST_BF_R("ldr sp, [sp, r",2,13*4,"]")
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 5/7] ARM: kprobes: Add test cases for stack consuming instructions
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:28 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
From: Jon Medhurst <tixy@linaro.org>
These have extra 'checker' functions associated with them so
lets make sure those get covered by testing.
Signed-off-by: Jon Medhurst <tixy@linaro.org>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
v1 -> v2:
- Move to arch/arm/probes/ .
---
arch/arm/probes/kprobes/test-arm.c | 17 +++++++++++++++--
arch/arm/probes/kprobes/test-thumb.c | 12 ++++++++++++
2 files changed, 27 insertions(+), 2 deletions(-)
diff --git a/arch/arm/probes/kprobes/test-arm.c b/arch/arm/probes/kprobes/test-arm.c
index fdeb300..9b3b1b4 100644
--- a/arch/arm/probes/kprobes/test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -12,6 +12,7 @@
#include <linux/module.h>
#include <asm/system_info.h>
#include <asm/opcodes.h>
+#include <asm/probes.h>
#include "test-core.h"
@@ -478,6 +479,7 @@ void kprobe_arm_test_cases(void)
TEST_RPR( "strh r",0, VAL1,", [r",1, 48,", -r",2, 24,"]")
TEST_RPR( "streqh r",14,VAL2,", [r",11,0, ", r",12, 48,"]")
TEST_UNSUPPORTED( "streqh r14, [r13, r12]")
+ TEST_UNSUPPORTED( "streqh r14, [r12, r13]")
TEST_RPR( "strh r",1, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strneh r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strh r",2, VAL1,", [r",3, 24,"], r",4, 48,"")
@@ -502,6 +504,9 @@ void kprobe_arm_test_cases(void)
TEST_RP( "strplh r",12,VAL2,", [r",11,24,", #-4]!")
TEST_RP( "strh r",2, VAL1,", [r",3, 24,"], #48")
TEST_RP( "strh r",10,VAL2,", [r",9, 64,"], #-48")
+ TEST_RP( "strh r",3, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strh r3, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RP( "strh r",4, VAL1,", [r",14,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_arm(0xe1efc3b0) " @ strh r12, [pc, #48]!")
TEST_UNSUPPORTED(__inst_arm(0xe0c9f3b0) " @ strh pc, [r9], #48")
@@ -568,6 +573,7 @@ void kprobe_arm_test_cases(void)
TEST_RPR( "strd r",0, VAL1,", [r",1, 48,", -r",2,24,"]")
TEST_RPR( "strccd r",8, VAL2,", [r",11,0, ", r",12,48,"]")
TEST_UNSUPPORTED( "strccd r8, [r13, r12]")
+ TEST_UNSUPPORTED( "strccd r8, [r12, r13]")
TEST_RPR( "strd r",4, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strcsd r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strd r",2, VAL1,", [r",5, 24,"], r",4,48,"")
@@ -591,6 +597,9 @@ void kprobe_arm_test_cases(void)
TEST_RP( "strvcd r",12,VAL2,", [r",11,24,", #-16]!")
TEST_RP( "strd r",2, VAL1,", [r",4, 24,"], #48")
TEST_RP( "strd r",10,VAL2,", [r",9, 64,"], #-48")
+ TEST_RP( "strd r",6, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strd r6, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RP( "strd r",4, VAL1,", [r",12,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_arm(0xe1efc3f0) " @ strd r12, [pc, #48]!")
TEST_P( "ldrd r0, [r",0, 24,", #-8]")
@@ -639,16 +648,20 @@ void kprobe_arm_test_cases(void)
TEST_RP( "str"byte" r",12,VAL2,", [r",11,24,", #-4]!") \
TEST_RP( "str"byte" r",2, VAL1,", [r",3, 24,"], #48") \
TEST_RP( "str"byte" r",10,VAL2,", [r",9, 64,"], #-48") \
+ TEST_RP( "str"byte" r",3, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!") \
+ TEST_UNSUPPORTED("str"byte" r3, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!") \
+ TEST_RP( "str"byte" r",4, VAL1,", [r",10,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 48,", -r",2, 24,"]") \
TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 48,"]") \
- TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r12, r13]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 48,"]!") \
TEST_RPR("str"byte" r",12,VAL2,", [r",11,48,", -r",10,24,"]!") \
TEST_RPR("str"byte" r",2, VAL1,", [r",3, 24,"], r",4, 48,"") \
TEST_RPR("str"byte" r",10,VAL2,", [r",9, 48,"], -r",11,24,"") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 24,", r",2, 32,", asl #1]")\
TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 32,", lsr #2]")\
- TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]")\
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 32,", asr #3]!")\
TEST_RPR("str"byte" r",12,VAL2,", [r",11,24,", r",10, 4,", ror #31]!")\
TEST_P( "ldr"byte" r0, [r",0, 24,", #-2]") \
diff --git a/arch/arm/probes/kprobes/test-thumb.c b/arch/arm/probes/kprobes/test-thumb.c
index 6c6e9a9..e8cf193 100644
--- a/arch/arm/probes/kprobes/test-thumb.c
+++ b/arch/arm/probes/kprobes/test-thumb.c
@@ -11,6 +11,7 @@
#include <linux/kernel.h>
#include <linux/module.h>
#include <asm/opcodes.h>
+#include <asm/probes.h>
#include "test-core.h"
@@ -416,6 +417,9 @@ void kprobe_thumb32_test_cases(void)
TEST_RR( "strd r",14,VAL2,", r",12,VAL1,", [sp, #16]!")
TEST_RRP("strd r",1, VAL1,", r",0, VAL2,", [r",7, 24,"], #16")
TEST_RR( "strd r",7, VAL2,", r",8, VAL1,", [sp], #-16")
+ TEST_RRP("strd r",6, VAL1,", r",7, VAL2,", [r",13, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strd r6, r7, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RRP("strd r",4, VAL1,", r",5, VAL2,", [r",14, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_thumb32(0xe9efec04) " @ strd r14, r12, [pc, #16]!")
TEST_UNSUPPORTED(__inst_thumb32(0xe8efec04) " @ strd r14, r12, [pc], #16")
@@ -821,14 +825,22 @@ CONDITION_INSTRUCTIONS(22,
TEST_RP( "str"size" r",14,VAL2,", [r",1, 256, ", #-128]!") \
TEST_RPR("str"size".w r",0, VAL1,", [r",1, 0,", r",2, 4,"]") \
TEST_RPR("str"size" r",14,VAL2,", [r",10,0,", r",11,4,", lsl #1]") \
+ TEST_UNSUPPORTED("str"size" r0, [r13, r1]") \
TEST_R( "str"size".w r",7, VAL1,", [sp, #24]") \
TEST_RP( "str"size".w r",0, VAL2,", [r",0,0, "]") \
+ TEST_RP( "str"size" r",6, VAL1,", [r",13, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!") \
+ TEST_UNSUPPORTED("str"size" r6, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!") \
+ TEST_RP( "str"size" r",4, VAL2,", [r",12, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!") \
TEST_UNSUPPORTED("str"size"t r0, [r1, #4]")
SINGLE_STORE("b")
SINGLE_STORE("h")
SINGLE_STORE("")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf801000d) " @ strb r0, [r1, r13]")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf821000d) " @ strh r0, [r1, r13]")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf841000d) " @ str r0, [r1, r13]")
+
TEST("str sp, [sp]")
TEST_UNSUPPORTED(__inst_thumb32(0xf8cfe000) " @ str r14, [pc]")
TEST_UNSUPPORTED(__inst_thumb32(0xf8cef000) " @ str pc, [r14]")
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 5/7] ARM: kprobes: Add test cases for stack consuming instructions
@ 2014-12-08 6:28 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: linux-arm-kernel
From: Jon Medhurst <tixy@linaro.org>
These have extra 'checker' functions associated with them so
lets make sure those get covered by testing.
Signed-off-by: Jon Medhurst <tixy@linaro.org>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
v1 -> v2:
- Move to arch/arm/probes/ .
---
arch/arm/probes/kprobes/test-arm.c | 17 +++++++++++++++--
arch/arm/probes/kprobes/test-thumb.c | 12 ++++++++++++
2 files changed, 27 insertions(+), 2 deletions(-)
diff --git a/arch/arm/probes/kprobes/test-arm.c b/arch/arm/probes/kprobes/test-arm.c
index fdeb300..9b3b1b4 100644
--- a/arch/arm/probes/kprobes/test-arm.c
+++ b/arch/arm/probes/kprobes/test-arm.c
@@ -12,6 +12,7 @@
#include <linux/module.h>
#include <asm/system_info.h>
#include <asm/opcodes.h>
+#include <asm/probes.h>
#include "test-core.h"
@@ -478,6 +479,7 @@ void kprobe_arm_test_cases(void)
TEST_RPR( "strh r",0, VAL1,", [r",1, 48,", -r",2, 24,"]")
TEST_RPR( "streqh r",14,VAL2,", [r",11,0, ", r",12, 48,"]")
TEST_UNSUPPORTED( "streqh r14, [r13, r12]")
+ TEST_UNSUPPORTED( "streqh r14, [r12, r13]")
TEST_RPR( "strh r",1, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strneh r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strh r",2, VAL1,", [r",3, 24,"], r",4, 48,"")
@@ -502,6 +504,9 @@ void kprobe_arm_test_cases(void)
TEST_RP( "strplh r",12,VAL2,", [r",11,24,", #-4]!")
TEST_RP( "strh r",2, VAL1,", [r",3, 24,"], #48")
TEST_RP( "strh r",10,VAL2,", [r",9, 64,"], #-48")
+ TEST_RP( "strh r",3, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strh r3, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RP( "strh r",4, VAL1,", [r",14,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_arm(0xe1efc3b0) " @ strh r12, [pc, #48]!")
TEST_UNSUPPORTED(__inst_arm(0xe0c9f3b0) " @ strh pc, [r9], #48")
@@ -568,6 +573,7 @@ void kprobe_arm_test_cases(void)
TEST_RPR( "strd r",0, VAL1,", [r",1, 48,", -r",2,24,"]")
TEST_RPR( "strccd r",8, VAL2,", [r",11,0, ", r",12,48,"]")
TEST_UNSUPPORTED( "strccd r8, [r13, r12]")
+ TEST_UNSUPPORTED( "strccd r8, [r12, r13]")
TEST_RPR( "strd r",4, VAL1,", [r",2, 24,", r",3, 48,"]!")
TEST_RPR( "strcsd r",12,VAL2,", [r",11,48,", -r",10,24,"]!")
TEST_RPR( "strd r",2, VAL1,", [r",5, 24,"], r",4,48,"")
@@ -591,6 +597,9 @@ void kprobe_arm_test_cases(void)
TEST_RP( "strvcd r",12,VAL2,", [r",11,24,", #-16]!")
TEST_RP( "strd r",2, VAL1,", [r",4, 24,"], #48")
TEST_RP( "strd r",10,VAL2,", [r",9, 64,"], #-48")
+ TEST_RP( "strd r",6, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strd r6, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RP( "strd r",4, VAL1,", [r",12,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_arm(0xe1efc3f0) " @ strd r12, [pc, #48]!")
TEST_P( "ldrd r0, [r",0, 24,", #-8]")
@@ -639,16 +648,20 @@ void kprobe_arm_test_cases(void)
TEST_RP( "str"byte" r",12,VAL2,", [r",11,24,", #-4]!") \
TEST_RP( "str"byte" r",2, VAL1,", [r",3, 24,"], #48") \
TEST_RP( "str"byte" r",10,VAL2,", [r",9, 64,"], #-48") \
+ TEST_RP( "str"byte" r",3, VAL1,", [r",13,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!") \
+ TEST_UNSUPPORTED("str"byte" r3, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!") \
+ TEST_RP( "str"byte" r",4, VAL1,", [r",10,TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 48,", -r",2, 24,"]") \
TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 48,"]") \
- TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12]") \
+ TEST_UNSUPPORTED("str"byte" r14, [r12, r13]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 48,"]!") \
TEST_RPR("str"byte" r",12,VAL2,", [r",11,48,", -r",10,24,"]!") \
TEST_RPR("str"byte" r",2, VAL1,", [r",3, 24,"], r",4, 48,"") \
TEST_RPR("str"byte" r",10,VAL2,", [r",9, 48,"], -r",11,24,"") \
TEST_RPR("str"byte" r",0, VAL1,", [r",1, 24,", r",2, 32,", asl #1]")\
TEST_RPR("str"byte" r",14,VAL2,", [r",11,0, ", r",12, 32,", lsr #2]")\
- TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]")\
+ TEST_UNSUPPORTED("str"byte" r14, [r13, r12, lsr #2]") \
TEST_RPR("str"byte" r",1, VAL1,", [r",2, 24,", r",3, 32,", asr #3]!")\
TEST_RPR("str"byte" r",12,VAL2,", [r",11,24,", r",10, 4,", ror #31]!")\
TEST_P( "ldr"byte" r0, [r",0, 24,", #-2]") \
diff --git a/arch/arm/probes/kprobes/test-thumb.c b/arch/arm/probes/kprobes/test-thumb.c
index 6c6e9a9..e8cf193 100644
--- a/arch/arm/probes/kprobes/test-thumb.c
+++ b/arch/arm/probes/kprobes/test-thumb.c
@@ -11,6 +11,7 @@
#include <linux/kernel.h>
#include <linux/module.h>
#include <asm/opcodes.h>
+#include <asm/probes.h>
#include "test-core.h"
@@ -416,6 +417,9 @@ void kprobe_thumb32_test_cases(void)
TEST_RR( "strd r",14,VAL2,", r",12,VAL1,", [sp, #16]!")
TEST_RRP("strd r",1, VAL1,", r",0, VAL2,", [r",7, 24,"], #16")
TEST_RR( "strd r",7, VAL2,", r",8, VAL1,", [sp], #-16")
+ TEST_RRP("strd r",6, VAL1,", r",7, VAL2,", [r",13, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!")
+ TEST_UNSUPPORTED("strd r6, r7, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!")
+ TEST_RRP("strd r",4, VAL1,", r",5, VAL2,", [r",14, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!")
TEST_UNSUPPORTED(__inst_thumb32(0xe9efec04) " @ strd r14, r12, [pc, #16]!")
TEST_UNSUPPORTED(__inst_thumb32(0xe8efec04) " @ strd r14, r12, [pc], #16")
@@ -821,14 +825,22 @@ CONDITION_INSTRUCTIONS(22,
TEST_RP( "str"size" r",14,VAL2,", [r",1, 256, ", #-128]!") \
TEST_RPR("str"size".w r",0, VAL1,", [r",1, 0,", r",2, 4,"]") \
TEST_RPR("str"size" r",14,VAL2,", [r",10,0,", r",11,4,", lsl #1]") \
+ TEST_UNSUPPORTED("str"size" r0, [r13, r1]") \
TEST_R( "str"size".w r",7, VAL1,", [sp, #24]") \
TEST_RP( "str"size".w r",0, VAL2,", [r",0,0, "]") \
+ TEST_RP( "str"size" r",6, VAL1,", [r",13, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"]!") \
+ TEST_UNSUPPORTED("str"size" r6, [r13, #-"__stringify(MAX_STACK_SIZE)"-8]!") \
+ TEST_RP( "str"size" r",4, VAL2,", [r",12, TEST_MEMORY_SIZE,", #-"__stringify(MAX_STACK_SIZE)"-8]!") \
TEST_UNSUPPORTED("str"size"t r0, [r1, #4]")
SINGLE_STORE("b")
SINGLE_STORE("h")
SINGLE_STORE("")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf801000d) " @ strb r0, [r1, r13]")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf821000d) " @ strh r0, [r1, r13]")
+ TEST_UNSUPPORTED(__inst_thumb32(0xf841000d) " @ str r0, [r1, r13]")
+
TEST("str sp, [sp]")
TEST_UNSUPPORTED(__inst_thumb32(0xf8cfe000) " @ str r14, [pc]")
TEST_UNSUPPORTED(__inst_thumb32(0xf8cef000) " @ str pc, [r14]")
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 6/7] kprobes: Pass the original kprobe for preparing optimized kprobe
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:28 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
From: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Pass the original kprobe for preparing an optimized kprobe arch-dep
part, since for some architecture (e.g. ARM32) requires the information
in original kprobe.
---
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
arch/x86/kernel/kprobes/opt.c | 3 ++-
include/linux/kprobes.h | 3 ++-
kernel/kprobes.c | 4 ++--
3 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/arch/x86/kernel/kprobes/opt.c b/arch/x86/kernel/kprobes/opt.c
index f1314d0..7028296 100644
--- a/arch/x86/kernel/kprobes/opt.c
+++ b/arch/x86/kernel/kprobes/opt.c
@@ -320,7 +320,8 @@ void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
* Target instructions MUST be relocatable (checked inside)
* This is called when new aggr(opt)probe is allocated or reused.
*/
-int arch_prepare_optimized_kprobe(struct optimized_kprobe *op)
+int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
+ struct kprobe *__unused)
{
u8 *buf;
int ret;
diff --git a/include/linux/kprobes.h b/include/linux/kprobes.h
index f7296e5..7ab2c93 100644
--- a/include/linux/kprobes.h
+++ b/include/linux/kprobes.h
@@ -308,7 +308,8 @@ struct optimized_kprobe {
/* Architecture dependent functions for direct jump optimization */
extern int arch_prepared_optinsn(struct arch_optimized_insn *optinsn);
extern int arch_check_optimized_kprobe(struct optimized_kprobe *op);
-extern int arch_prepare_optimized_kprobe(struct optimized_kprobe *op);
+extern int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
+ struct kprobe *orig);
extern void arch_remove_optimized_kprobe(struct optimized_kprobe *op);
extern void arch_optimize_kprobes(struct list_head *oplist);
extern void arch_unoptimize_kprobes(struct list_head *oplist,
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 3995f54..9f28aa7 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -717,7 +717,7 @@ static void prepare_optimized_kprobe(struct kprobe *p)
struct optimized_kprobe *op;
op = container_of(p, struct optimized_kprobe, kp);
- arch_prepare_optimized_kprobe(op);
+ arch_prepare_optimized_kprobe(op, p);
}
/* Allocate new optimized_kprobe and try to prepare optimized instructions */
@@ -731,7 +731,7 @@ static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
INIT_LIST_HEAD(&op->list);
op->kp.addr = p->addr;
- arch_prepare_optimized_kprobe(op);
+ arch_prepare_optimized_kprobe(op, p);
return &op->kp;
}
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 6/7] kprobes: Pass the original kprobe for preparing optimized kprobe
@ 2014-12-08 6:28 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: linux-arm-kernel
From: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Pass the original kprobe for preparing an optimized kprobe arch-dep
part, since for some architecture (e.g. ARM32) requires the information
in original kprobe.
---
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
---
arch/x86/kernel/kprobes/opt.c | 3 ++-
include/linux/kprobes.h | 3 ++-
kernel/kprobes.c | 4 ++--
3 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/arch/x86/kernel/kprobes/opt.c b/arch/x86/kernel/kprobes/opt.c
index f1314d0..7028296 100644
--- a/arch/x86/kernel/kprobes/opt.c
+++ b/arch/x86/kernel/kprobes/opt.c
@@ -320,7 +320,8 @@ void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
* Target instructions MUST be relocatable (checked inside)
* This is called when new aggr(opt)probe is allocated or reused.
*/
-int arch_prepare_optimized_kprobe(struct optimized_kprobe *op)
+int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
+ struct kprobe *__unused)
{
u8 *buf;
int ret;
diff --git a/include/linux/kprobes.h b/include/linux/kprobes.h
index f7296e5..7ab2c93 100644
--- a/include/linux/kprobes.h
+++ b/include/linux/kprobes.h
@@ -308,7 +308,8 @@ struct optimized_kprobe {
/* Architecture dependent functions for direct jump optimization */
extern int arch_prepared_optinsn(struct arch_optimized_insn *optinsn);
extern int arch_check_optimized_kprobe(struct optimized_kprobe *op);
-extern int arch_prepare_optimized_kprobe(struct optimized_kprobe *op);
+extern int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
+ struct kprobe *orig);
extern void arch_remove_optimized_kprobe(struct optimized_kprobe *op);
extern void arch_optimize_kprobes(struct list_head *oplist);
extern void arch_unoptimize_kprobes(struct list_head *oplist,
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 3995f54..9f28aa7 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -717,7 +717,7 @@ static void prepare_optimized_kprobe(struct kprobe *p)
struct optimized_kprobe *op;
op = container_of(p, struct optimized_kprobe, kp);
- arch_prepare_optimized_kprobe(op);
+ arch_prepare_optimized_kprobe(op, p);
}
/* Allocate new optimized_kprobe and try to prepare optimized instructions */
@@ -731,7 +731,7 @@ static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
INIT_LIST_HEAD(&op->list);
op->kp.addr = p->addr;
- arch_prepare_optimized_kprobe(op);
+ arch_prepare_optimized_kprobe(op, p);
return &op->kp;
}
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 6:27 ` Wang Nan
@ 2014-12-08 6:28 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: masami.hiramatsu.pt, tixy, linux; +Cc: lizefan, linux-kernel, linux-arm-kernel
This patch introduce kprobeopt for ARM 32.
Limitations:
- Currently only kernel compiled with ARM ISA is supported.
- Offset between probe point and optinsn slot must not larger than
32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
things complex. Futher patch can make such optimization.
Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
replace probed instruction by a 'b', branch to trampoline code and then
calls optimized_callback(). optimized_callback() calls opt_pre_handler()
to execute kprobe handler. It also emulate/simulate replaced instruction.
When unregistering kprobe, the deferred manner of unoptimizer may leave
branch instruction before optimizer is called. Different from x86_64,
which only copy the probed insn after optprobe_template_end and
reexecute them, this patch call singlestep to emulate/simulate the insn
directly. Futher patch can optimize this behavior.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
---
v1 -> v2:
- Improvement: if replaced instruction is conditional, generate a
conditional branch instruction for it;
- Introduces RELATIVEJUMP_OPCODES due to ARM kprobe_opcode_t is 4
bytes;
- Removes size field in struct arch_optimized_insn;
- Use arm_gen_branch() to generate branch instruction;
- Remove all recover logic: ARM doesn't use tail buffer, no need
recover replaced instructions like x86;
- Remove incorrect CONFIG_THUMB checking;
- can_optimize() always returns true if address is well aligned;
- Improve optimized_callback: using opt_pre_handler();
- Bugfix: correct range checking code and improve comments;
- Fix commit message.
v2 -> v3:
- Rename RELATIVEJUMP_OPCODES to MAX_COPIED_INSNS;
- Remove unneeded checking:
arch_check_optimized_kprobe(), can_optimize();
- Add missing flush_icache_range() in arch_prepare_optimized_kprobe();
- Remove unneeded 'return;'.
v3 -> v4:
- Use __mem_to_opcode_arm() to translate copied_insn to ensure it
works in big endian kernel;
- Replace 'nop' placeholder in trampoline code template with
'.long 0' to avoid confusion: reader may regard 'nop' as an
instruction, but it is value in fact.
v4 -> v5:
- Don't optimize stack store operations.
- Introduce prepared field to arch_optimized_insn to indicate whether
it is prepared. Similar to size field with x86. See v1 -> v2.
v5 -> v6:
- Dynamically reserve stack according to instruction.
- Rename: kprobes-opt.c -> kprobes-opt-arm.c.
- Set op->optinsn.insn after all works are done.
v6 -> v7:
- Using checker to check stack consumption.
v7 -> v8:
- Small code adjustments.
v8 -> v9:
- Utilize original kprobe passed to arch_prepare_optimized_kprobe()
to avoid copy ainsn twice.
- A bug in arch_prepare_optimized_kprobe() is found and fixed.
v9 -> v10:
- Commit message improvements.
v10 -> v11:
- Move to arch/arm/probes/, insn.h is moved to arch/arm/include/asm.
- Code cleanup.
- Bugfix based on Tixy's test result:
- Trampoline deal with ARM -> Thumb transision instructions and
AEABI stack alignment requirement correctly.
- Trampoline code buffer should start at 4 byte aligned address.
We enforces it in this series by using macro to wrap 'code' var.
v11 -> v12:
- Remove trampoline code stack trick and use r4 to save original
stack.
- Remove trampoline code buffer alignment trick.
- Names of files are changed.
v12 -> v13:
- Assume stack always aligned by 4-bytes in any case.
- Comments update.
v13 -> v14:
- Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
---
arch/arm/Kconfig | 1 +
arch/arm/{kernel => include/asm}/insn.h | 0
arch/arm/include/asm/kprobes.h | 29 +++
arch/arm/kernel/Makefile | 2 +-
arch/arm/kernel/ftrace.c | 3 +-
arch/arm/kernel/jump_label.c | 3 +-
arch/arm/probes/kprobes/Makefile | 1 +
arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
samples/kprobes/kprobe_example.c | 2 +-
9 files changed, 357 insertions(+), 6 deletions(-)
rename arch/arm/{kernel => include/asm}/insn.h (100%)
create mode 100644 arch/arm/probes/kprobes/opt-arm.c
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 89c4b5c..2471240 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -59,6 +59,7 @@ config ARM
select HAVE_MEMBLOCK
select HAVE_MOD_ARCH_SPECIFIC if ARM_UNWIND
select HAVE_OPROFILE if (HAVE_PERF_EVENTS)
+ select HAVE_OPTPROBES if !THUMB2_KERNEL
select HAVE_PERF_EVENTS
select HAVE_PERF_REGS
select HAVE_PERF_USER_STACK_DUMP
diff --git a/arch/arm/kernel/insn.h b/arch/arm/include/asm/insn.h
similarity index 100%
rename from arch/arm/kernel/insn.h
rename to arch/arm/include/asm/insn.h
diff --git a/arch/arm/include/asm/kprobes.h b/arch/arm/include/asm/kprobes.h
index 56f9ac6..50ff3bc 100644
--- a/arch/arm/include/asm/kprobes.h
+++ b/arch/arm/include/asm/kprobes.h
@@ -50,5 +50,34 @@ int kprobe_fault_handler(struct pt_regs *regs, unsigned int fsr);
int kprobe_exceptions_notify(struct notifier_block *self,
unsigned long val, void *data);
+/* optinsn template addresses */
+extern __visible kprobe_opcode_t optprobe_template_entry;
+extern __visible kprobe_opcode_t optprobe_template_val;
+extern __visible kprobe_opcode_t optprobe_template_call;
+extern __visible kprobe_opcode_t optprobe_template_end;
+extern __visible kprobe_opcode_t optprobe_template_sub_sp;
+extern __visible kprobe_opcode_t optprobe_template_add_sp;
+
+#define MAX_OPTIMIZED_LENGTH 4
+#define MAX_OPTINSN_SIZE \
+ ((unsigned long)&optprobe_template_end - \
+ (unsigned long)&optprobe_template_entry)
+#define RELATIVEJUMP_SIZE 4
+
+struct arch_optimized_insn {
+ /*
+ * copy of the original instructions.
+ * Different from x86, ARM kprobe_opcode_t is u32.
+ */
+#define MAX_COPIED_INSN DIV_ROUND_UP(RELATIVEJUMP_SIZE, sizeof(kprobe_opcode_t))
+ kprobe_opcode_t copied_insn[MAX_COPIED_INSN];
+ /* detour code buffer */
+ kprobe_opcode_t *insn;
+ /*
+ * We always copy one instruction on ARM,
+ * so size will always be 4, and unlike x86, there is no
+ * need for a size field.
+ */
+};
#endif /* _ARM_KPROBES_H */
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 40d3e00..1d0f4e7 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -52,7 +52,7 @@ obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o insn.o
obj-$(CONFIG_JUMP_LABEL) += jump_label.o insn.o patch.o
obj-$(CONFIG_KEXEC) += machine_kexec.o relocate_kernel.o
# Main staffs in KPROBES are in arch/arm/probes/ .
-obj-$(CONFIG_KPROBES) += patch.o
+obj-$(CONFIG_KPROBES) += patch.o insn.o
obj-$(CONFIG_OABI_COMPAT) += sys_oabi-compat.o
obj-$(CONFIG_ARM_THUMBEE) += thumbee.o
obj-$(CONFIG_KGDB) += kgdb.o
diff --git a/arch/arm/kernel/ftrace.c b/arch/arm/kernel/ftrace.c
index af9a8a9..ec7e332 100644
--- a/arch/arm/kernel/ftrace.c
+++ b/arch/arm/kernel/ftrace.c
@@ -19,8 +19,7 @@
#include <asm/cacheflush.h>
#include <asm/opcodes.h>
#include <asm/ftrace.h>
-
-#include "insn.h"
+#include <asm/insn.h>
#ifdef CONFIG_THUMB2_KERNEL
#define NOP 0xf85deb04 /* pop.w {lr} */
diff --git a/arch/arm/kernel/jump_label.c b/arch/arm/kernel/jump_label.c
index c6c73ed..35a8fbb 100644
--- a/arch/arm/kernel/jump_label.c
+++ b/arch/arm/kernel/jump_label.c
@@ -1,8 +1,7 @@
#include <linux/kernel.h>
#include <linux/jump_label.h>
#include <asm/patch.h>
-
-#include "insn.h"
+#include <asm/insn.h>
#ifdef HAVE_JUMP_LABEL
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
index bc8d504..76a36bf 100644
--- a/arch/arm/probes/kprobes/Makefile
+++ b/arch/arm/probes/kprobes/Makefile
@@ -7,5 +7,6 @@ obj-$(CONFIG_KPROBES) += actions-thumb.o checkers-thumb.o
test-kprobes-objs += test-thumb.o
else
obj-$(CONFIG_KPROBES) += actions-arm.o checkers-arm.o
+obj-$(CONFIG_OPTPROBES) += opt-arm.o
test-kprobes-objs += test-arm.o
endif
diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
new file mode 100644
index 0000000..afbfeef
--- /dev/null
+++ b/arch/arm/probes/kprobes/opt-arm.c
@@ -0,0 +1,322 @@
+/*
+ * Kernel Probes Jump Optimization (Optprobes)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2002, 2004
+ * Copyright (C) Hitachi Ltd., 2012
+ * Copyright (C) Huawei Inc., 2014
+ */
+
+#include <linux/kprobes.h>
+#include <linux/jump_label.h>
+#include <asm/kprobes.h>
+#include <asm/cacheflush.h>
+/* for arm_gen_branch */
+#include <asm/insn.h>
+/* for patch_text */
+#include <asm/patch.h>
+/* for stop_machine */
+#include <linux/stop_machine.h>
+/*
+ * NOTE: the first sub and add instruction will be modified according
+ * to the stack cost of the instruction.
+ */
+asm (
+ ".global optprobe_template_entry\n"
+ "optprobe_template_entry:\n"
+ ".global optprobe_template_sub_sp\n"
+ "optprobe_template_sub_sp:"
+ " sub sp, sp, #0xff\n"
+ " stmia sp, {r0 - r14} \n"
+ ".global optprobe_template_add_sp\n"
+ "optprobe_template_add_sp:"
+ " add r3, sp, #0xff\n"
+ " str r3, [sp, #52]\n"
+ " mrs r4, cpsr\n"
+ " str r4, [sp, #64]\n"
+ " mov r1, sp\n"
+ " ldr r0, 1f\n"
+ " ldr r2, 2f\n"
+ /*
+ * AEABI requires an 8-bytes alignment stack. If
+ * SP % 8 != 0 (SP % 4 == 0 should be ensured),
+ * alloc more bytes here.
+ */
+ " and r4, sp, #4\n"
+ " sub sp, sp, r4\n"
+ " blx r2\n"
+ " add sp, sp, r4\n"
+ " ldr r1, [sp, #64]\n"
+ " tst r1, #"__stringify(PSR_T_BIT)"\n"
+ " ldrne r2, [sp, #60]\n"
+ " orrne r2, #1\n"
+ " strne r2, [sp, #60] @ set bit0 of PC for thumb\n"
+ " msr cpsr_cxsf, r1\n"
+ " ldmia sp, {r0 - r15}\n"
+ ".global optprobe_template_val\n"
+ "optprobe_template_val:\n"
+ "1: .long 0\n"
+ ".global optprobe_template_call\n"
+ "optprobe_template_call:\n"
+ "2: .long 0\n"
+ ".global optprobe_template_end\n"
+ "optprobe_template_end:\n");
+
+#define TMPL_VAL_IDX \
+ ((unsigned long *)&optprobe_template_val - (unsigned long *)&optprobe_template_entry)
+#define TMPL_CALL_IDX \
+ ((unsigned long *)&optprobe_template_call - (unsigned long *)&optprobe_template_entry)
+#define TMPL_END_IDX \
+ ((unsigned long *)&optprobe_template_end - (unsigned long *)&optprobe_template_entry)
+#define TMPL_ADD_SP \
+ ((unsigned long *)&optprobe_template_add_sp - (unsigned long *)&optprobe_template_entry)
+#define TMPL_SUB_SP \
+ ((unsigned long *)&optprobe_template_sub_sp - (unsigned long *)&optprobe_template_entry)
+
+/*
+ * ARM can always optimize an instruction when using ARM ISA, except
+ * instructions like 'str r0, [sp, r1]' which store to stack and unable
+ * to determine stack space consumption statically.
+ */
+int arch_prepared_optinsn(struct arch_optimized_insn *optinsn)
+{
+ return optinsn->insn != NULL;
+}
+
+/*
+ * In ARM ISA, kprobe opt always replace one instruction (4 bytes
+ * aligned and 4 bytes long). It is impossible to encounter another
+ * kprobe in the address range. So always return 0.
+ */
+int arch_check_optimized_kprobe(struct optimized_kprobe *op)
+{
+ return 0;
+}
+
+/* Caller must ensure addr & 3 == 0 */
+static int can_optimize(struct kprobe *kp)
+{
+ if (kp->ainsn.stack_space < 0)
+ return 0;
+ /*
+ * 255 is the biggest imm can be used in 'sub r0, r0, #<imm>'.
+ * Number larger than 255 needs special encoding.
+ */
+ if (kp->ainsn.stack_space > 255 - sizeof(struct pt_regs))
+ return 0;
+ return 1;
+}
+
+/* Free optimized instruction slot */
+static void
+__arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
+{
+ if (op->optinsn.insn) {
+ free_optinsn_slot(op->optinsn.insn, dirty);
+ op->optinsn.insn = NULL;
+ }
+}
+
+extern void kprobe_handler(struct pt_regs *regs);
+
+static void
+optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
+{
+ unsigned long flags;
+ struct kprobe *p = &op->kp;
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+
+ /* Save skipped registers */
+ regs->ARM_pc = (unsigned long)op->kp.addr;
+ regs->ARM_ORIG_r0 = ~0UL;
+
+ local_irq_save(flags);
+
+ if (kprobe_running()) {
+ kprobes_inc_nmissed_count(&op->kp);
+ } else {
+ __this_cpu_write(current_kprobe, &op->kp);
+ kcb->kprobe_status = KPROBE_HIT_ACTIVE;
+ opt_pre_handler(&op->kp, regs);
+ __this_cpu_write(current_kprobe, NULL);
+ }
+
+ /* In each case, we must singlestep the replaced instruction. */
+ op->kp.ainsn.insn_singlestep(p->opcode, &p->ainsn, regs);
+
+ local_irq_restore(flags);
+}
+
+int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *orig)
+{
+ kprobe_opcode_t *code;
+ unsigned long rel_chk;
+ unsigned long val;
+ unsigned long stack_protect = sizeof(struct pt_regs);
+
+ if (!can_optimize(orig))
+ return -EILSEQ;
+
+ code = get_optinsn_slot();
+ if (!code)
+ return -ENOMEM;
+
+ /*
+ * Verify if the address gap is in 32MiB range, because this uses
+ * a relative jump.
+ *
+ * kprobe opt use a 'b' instruction to branch to optinsn.insn.
+ * According to ARM manual, branch instruction is:
+ *
+ * 31 28 27 24 23 0
+ * +------+---+---+---+---+----------------+
+ * | cond | 1 | 0 | 1 | 0 | imm24 |
+ * +------+---+---+---+---+----------------+
+ *
+ * imm24 is a signed 24 bits integer. The real branch offset is computed
+ * by: imm32 = SignExtend(imm24:'00', 32);
+ *
+ * So the maximum forward branch should be:
+ * (0x007fffff << 2) = 0x01fffffc = 0x1fffffc
+ * The maximum backword branch should be:
+ * (0xff800000 << 2) = 0xfe000000 = -0x2000000
+ *
+ * We can simply check (rel & 0xfe000003):
+ * if rel is positive, (rel & 0xfe000000) shoule be 0
+ * if rel is negitive, (rel & 0xfe000000) should be 0xfe000000
+ * the last '3' is used for alignment checking.
+ */
+ rel_chk = (unsigned long)((long)code -
+ (long)orig->addr + 8) & 0xfe000003;
+
+ if ((rel_chk != 0) && (rel_chk != 0xfe000000)) {
+ /*
+ * Different from x86, we free code buf directly instead of
+ * calling __arch_remove_optimized_kprobe() because
+ * we have not fill any field in op.
+ */
+ free_optinsn_slot(code, 0);
+ return -ERANGE;
+ }
+
+ /* Copy arch-dep-instance from template. */
+ memcpy(code, &optprobe_template_entry,
+ TMPL_END_IDX * sizeof(kprobe_opcode_t));
+
+ /* Adjust buffer according to instruction. */
+ BUG_ON(orig->ainsn.stack_space < 0);
+
+ stack_protect += orig->ainsn.stack_space;
+
+ /* Should have been filtered by can_optimize(). */
+ BUG_ON(stack_protect > 255);
+
+ /* Create a 'sub sp, sp, #<stack_protect>' */
+ code[TMPL_SUB_SP] = __opcode_to_mem_arm(0xe24dd000 | stack_protect);
+ /* Create a 'add r3, sp, #<stack_protect>' */
+ code[TMPL_ADD_SP] = __opcode_to_mem_arm(0xe28d3000 | stack_protect);
+
+ /* Set probe information */
+ val = (unsigned long)op;
+ code[TMPL_VAL_IDX] = val;
+
+ /* Set probe function call */
+ val = (unsigned long)optimized_callback;
+ code[TMPL_CALL_IDX] = val;
+
+ flush_icache_range((unsigned long)code,
+ (unsigned long)(&code[TMPL_END_IDX]));
+
+ /* Set op->optinsn.insn means prepared. */
+ op->optinsn.insn = code;
+ return 0;
+}
+
+/*
+ * Similar to __arch_disarm_kprobe, operations which removing
+ * breakpoints must be wrapped by stop_machine to avoid racing.
+ */
+static __kprobes int __arch_optimize_kprobes(void *p)
+{
+ struct list_head *oplist = p;
+ struct optimized_kprobe *op, *tmp;
+
+ list_for_each_entry_safe(op, tmp, oplist, list) {
+ unsigned long insn;
+ WARN_ON(kprobe_disabled(&op->kp));
+
+ /*
+ * Backup instructions which will be replaced
+ * by jump address
+ */
+ memcpy(op->optinsn.copied_insn, op->kp.addr,
+ RELATIVEJUMP_SIZE);
+
+ insn = arm_gen_branch((unsigned long)op->kp.addr,
+ (unsigned long)op->optinsn.insn);
+ BUG_ON(insn == 0);
+
+ /*
+ * Make it a conditional branch if replaced insn
+ * is consitional
+ */
+ insn = (__mem_to_opcode_arm(
+ op->optinsn.copied_insn[0]) & 0xf0000000) |
+ (insn & 0x0fffffff);
+
+ patch_text(op->kp.addr, insn);
+
+ list_del_init(&op->list);
+ }
+ return 0;
+}
+
+void arch_optimize_kprobes(struct list_head *oplist)
+{
+ stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
+}
+
+void arch_unoptimize_kprobe(struct optimized_kprobe *op)
+{
+ arch_arm_kprobe(&op->kp);
+}
+
+/*
+ * Recover original instructions and breakpoints from relative jumps.
+ * Caller must call with locking kprobe_mutex.
+ */
+void arch_unoptimize_kprobes(struct list_head *oplist,
+ struct list_head *done_list)
+{
+ struct optimized_kprobe *op, *tmp;
+
+ list_for_each_entry_safe(op, tmp, oplist, list) {
+ arch_unoptimize_kprobe(op);
+ list_move(&op->list, done_list);
+ }
+}
+
+int arch_within_optimized_kprobe(struct optimized_kprobe *op,
+ unsigned long addr)
+{
+ return ((unsigned long)op->kp.addr <= addr &&
+ (unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr);
+}
+
+void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
+{
+ __arch_remove_optimized_kprobe(op, 1);
+}
diff --git a/samples/kprobes/kprobe_example.c b/samples/kprobes/kprobe_example.c
index 623f864..e4e45f1 100644
--- a/samples/kprobes/kprobe_example.c
+++ b/samples/kprobes/kprobe_example.c
@@ -94,7 +94,7 @@ static int handler_fault(struct kprobe *p, struct pt_regs *regs, int trapnr)
static int __init kprobe_init(void)
{
int ret;
- kp.addr = kallsyms_lookup_name(kp.symbol_name);
+ kp.addr = kallsyms_lookup_name(kp.symbol_name) + 8;
kp.symbol_name = NULL;
kp.pre_handler = handler_pre;
kp.post_handler = NULL;
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 6:28 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 6:28 UTC (permalink / raw)
To: linux-arm-kernel
This patch introduce kprobeopt for ARM 32.
Limitations:
- Currently only kernel compiled with ARM ISA is supported.
- Offset between probe point and optinsn slot must not larger than
32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
things complex. Futher patch can make such optimization.
Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
replace probed instruction by a 'b', branch to trampoline code and then
calls optimized_callback(). optimized_callback() calls opt_pre_handler()
to execute kprobe handler. It also emulate/simulate replaced instruction.
When unregistering kprobe, the deferred manner of unoptimizer may leave
branch instruction before optimizer is called. Different from x86_64,
which only copy the probed insn after optprobe_template_end and
reexecute them, this patch call singlestep to emulate/simulate the insn
directly. Futher patch can optimize this behavior.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
---
v1 -> v2:
- Improvement: if replaced instruction is conditional, generate a
conditional branch instruction for it;
- Introduces RELATIVEJUMP_OPCODES due to ARM kprobe_opcode_t is 4
bytes;
- Removes size field in struct arch_optimized_insn;
- Use arm_gen_branch() to generate branch instruction;
- Remove all recover logic: ARM doesn't use tail buffer, no need
recover replaced instructions like x86;
- Remove incorrect CONFIG_THUMB checking;
- can_optimize() always returns true if address is well aligned;
- Improve optimized_callback: using opt_pre_handler();
- Bugfix: correct range checking code and improve comments;
- Fix commit message.
v2 -> v3:
- Rename RELATIVEJUMP_OPCODES to MAX_COPIED_INSNS;
- Remove unneeded checking:
arch_check_optimized_kprobe(), can_optimize();
- Add missing flush_icache_range() in arch_prepare_optimized_kprobe();
- Remove unneeded 'return;'.
v3 -> v4:
- Use __mem_to_opcode_arm() to translate copied_insn to ensure it
works in big endian kernel;
- Replace 'nop' placeholder in trampoline code template with
'.long 0' to avoid confusion: reader may regard 'nop' as an
instruction, but it is value in fact.
v4 -> v5:
- Don't optimize stack store operations.
- Introduce prepared field to arch_optimized_insn to indicate whether
it is prepared. Similar to size field with x86. See v1 -> v2.
v5 -> v6:
- Dynamically reserve stack according to instruction.
- Rename: kprobes-opt.c -> kprobes-opt-arm.c.
- Set op->optinsn.insn after all works are done.
v6 -> v7:
- Using checker to check stack consumption.
v7 -> v8:
- Small code adjustments.
v8 -> v9:
- Utilize original kprobe passed to arch_prepare_optimized_kprobe()
to avoid copy ainsn twice.
- A bug in arch_prepare_optimized_kprobe() is found and fixed.
v9 -> v10:
- Commit message improvements.
v10 -> v11:
- Move to arch/arm/probes/, insn.h is moved to arch/arm/include/asm.
- Code cleanup.
- Bugfix based on Tixy's test result:
- Trampoline deal with ARM -> Thumb transision instructions and
AEABI stack alignment requirement correctly.
- Trampoline code buffer should start at 4 byte aligned address.
We enforces it in this series by using macro to wrap 'code' var.
v11 -> v12:
- Remove trampoline code stack trick and use r4 to save original
stack.
- Remove trampoline code buffer alignment trick.
- Names of files are changed.
v12 -> v13:
- Assume stack always aligned by 4-bytes in any case.
- Comments update.
v13 -> v14:
- Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
---
arch/arm/Kconfig | 1 +
arch/arm/{kernel => include/asm}/insn.h | 0
arch/arm/include/asm/kprobes.h | 29 +++
arch/arm/kernel/Makefile | 2 +-
arch/arm/kernel/ftrace.c | 3 +-
arch/arm/kernel/jump_label.c | 3 +-
arch/arm/probes/kprobes/Makefile | 1 +
arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
samples/kprobes/kprobe_example.c | 2 +-
9 files changed, 357 insertions(+), 6 deletions(-)
rename arch/arm/{kernel => include/asm}/insn.h (100%)
create mode 100644 arch/arm/probes/kprobes/opt-arm.c
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 89c4b5c..2471240 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -59,6 +59,7 @@ config ARM
select HAVE_MEMBLOCK
select HAVE_MOD_ARCH_SPECIFIC if ARM_UNWIND
select HAVE_OPROFILE if (HAVE_PERF_EVENTS)
+ select HAVE_OPTPROBES if !THUMB2_KERNEL
select HAVE_PERF_EVENTS
select HAVE_PERF_REGS
select HAVE_PERF_USER_STACK_DUMP
diff --git a/arch/arm/kernel/insn.h b/arch/arm/include/asm/insn.h
similarity index 100%
rename from arch/arm/kernel/insn.h
rename to arch/arm/include/asm/insn.h
diff --git a/arch/arm/include/asm/kprobes.h b/arch/arm/include/asm/kprobes.h
index 56f9ac6..50ff3bc 100644
--- a/arch/arm/include/asm/kprobes.h
+++ b/arch/arm/include/asm/kprobes.h
@@ -50,5 +50,34 @@ int kprobe_fault_handler(struct pt_regs *regs, unsigned int fsr);
int kprobe_exceptions_notify(struct notifier_block *self,
unsigned long val, void *data);
+/* optinsn template addresses */
+extern __visible kprobe_opcode_t optprobe_template_entry;
+extern __visible kprobe_opcode_t optprobe_template_val;
+extern __visible kprobe_opcode_t optprobe_template_call;
+extern __visible kprobe_opcode_t optprobe_template_end;
+extern __visible kprobe_opcode_t optprobe_template_sub_sp;
+extern __visible kprobe_opcode_t optprobe_template_add_sp;
+
+#define MAX_OPTIMIZED_LENGTH 4
+#define MAX_OPTINSN_SIZE \
+ ((unsigned long)&optprobe_template_end - \
+ (unsigned long)&optprobe_template_entry)
+#define RELATIVEJUMP_SIZE 4
+
+struct arch_optimized_insn {
+ /*
+ * copy of the original instructions.
+ * Different from x86, ARM kprobe_opcode_t is u32.
+ */
+#define MAX_COPIED_INSN DIV_ROUND_UP(RELATIVEJUMP_SIZE, sizeof(kprobe_opcode_t))
+ kprobe_opcode_t copied_insn[MAX_COPIED_INSN];
+ /* detour code buffer */
+ kprobe_opcode_t *insn;
+ /*
+ * We always copy one instruction on ARM,
+ * so size will always be 4, and unlike x86, there is no
+ * need for a size field.
+ */
+};
#endif /* _ARM_KPROBES_H */
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 40d3e00..1d0f4e7 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -52,7 +52,7 @@ obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o insn.o
obj-$(CONFIG_JUMP_LABEL) += jump_label.o insn.o patch.o
obj-$(CONFIG_KEXEC) += machine_kexec.o relocate_kernel.o
# Main staffs in KPROBES are in arch/arm/probes/ .
-obj-$(CONFIG_KPROBES) += patch.o
+obj-$(CONFIG_KPROBES) += patch.o insn.o
obj-$(CONFIG_OABI_COMPAT) += sys_oabi-compat.o
obj-$(CONFIG_ARM_THUMBEE) += thumbee.o
obj-$(CONFIG_KGDB) += kgdb.o
diff --git a/arch/arm/kernel/ftrace.c b/arch/arm/kernel/ftrace.c
index af9a8a9..ec7e332 100644
--- a/arch/arm/kernel/ftrace.c
+++ b/arch/arm/kernel/ftrace.c
@@ -19,8 +19,7 @@
#include <asm/cacheflush.h>
#include <asm/opcodes.h>
#include <asm/ftrace.h>
-
-#include "insn.h"
+#include <asm/insn.h>
#ifdef CONFIG_THUMB2_KERNEL
#define NOP 0xf85deb04 /* pop.w {lr} */
diff --git a/arch/arm/kernel/jump_label.c b/arch/arm/kernel/jump_label.c
index c6c73ed..35a8fbb 100644
--- a/arch/arm/kernel/jump_label.c
+++ b/arch/arm/kernel/jump_label.c
@@ -1,8 +1,7 @@
#include <linux/kernel.h>
#include <linux/jump_label.h>
#include <asm/patch.h>
-
-#include "insn.h"
+#include <asm/insn.h>
#ifdef HAVE_JUMP_LABEL
diff --git a/arch/arm/probes/kprobes/Makefile b/arch/arm/probes/kprobes/Makefile
index bc8d504..76a36bf 100644
--- a/arch/arm/probes/kprobes/Makefile
+++ b/arch/arm/probes/kprobes/Makefile
@@ -7,5 +7,6 @@ obj-$(CONFIG_KPROBES) += actions-thumb.o checkers-thumb.o
test-kprobes-objs += test-thumb.o
else
obj-$(CONFIG_KPROBES) += actions-arm.o checkers-arm.o
+obj-$(CONFIG_OPTPROBES) += opt-arm.o
test-kprobes-objs += test-arm.o
endif
diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
new file mode 100644
index 0000000..afbfeef
--- /dev/null
+++ b/arch/arm/probes/kprobes/opt-arm.c
@@ -0,0 +1,322 @@
+/*
+ * Kernel Probes Jump Optimization (Optprobes)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2002, 2004
+ * Copyright (C) Hitachi Ltd., 2012
+ * Copyright (C) Huawei Inc., 2014
+ */
+
+#include <linux/kprobes.h>
+#include <linux/jump_label.h>
+#include <asm/kprobes.h>
+#include <asm/cacheflush.h>
+/* for arm_gen_branch */
+#include <asm/insn.h>
+/* for patch_text */
+#include <asm/patch.h>
+/* for stop_machine */
+#include <linux/stop_machine.h>
+/*
+ * NOTE: the first sub and add instruction will be modified according
+ * to the stack cost of the instruction.
+ */
+asm (
+ ".global optprobe_template_entry\n"
+ "optprobe_template_entry:\n"
+ ".global optprobe_template_sub_sp\n"
+ "optprobe_template_sub_sp:"
+ " sub sp, sp, #0xff\n"
+ " stmia sp, {r0 - r14} \n"
+ ".global optprobe_template_add_sp\n"
+ "optprobe_template_add_sp:"
+ " add r3, sp, #0xff\n"
+ " str r3, [sp, #52]\n"
+ " mrs r4, cpsr\n"
+ " str r4, [sp, #64]\n"
+ " mov r1, sp\n"
+ " ldr r0, 1f\n"
+ " ldr r2, 2f\n"
+ /*
+ * AEABI requires an 8-bytes alignment stack. If
+ * SP % 8 != 0 (SP % 4 == 0 should be ensured),
+ * alloc more bytes here.
+ */
+ " and r4, sp, #4\n"
+ " sub sp, sp, r4\n"
+ " blx r2\n"
+ " add sp, sp, r4\n"
+ " ldr r1, [sp, #64]\n"
+ " tst r1, #"__stringify(PSR_T_BIT)"\n"
+ " ldrne r2, [sp, #60]\n"
+ " orrne r2, #1\n"
+ " strne r2, [sp, #60] @ set bit0 of PC for thumb\n"
+ " msr cpsr_cxsf, r1\n"
+ " ldmia sp, {r0 - r15}\n"
+ ".global optprobe_template_val\n"
+ "optprobe_template_val:\n"
+ "1: .long 0\n"
+ ".global optprobe_template_call\n"
+ "optprobe_template_call:\n"
+ "2: .long 0\n"
+ ".global optprobe_template_end\n"
+ "optprobe_template_end:\n");
+
+#define TMPL_VAL_IDX \
+ ((unsigned long *)&optprobe_template_val - (unsigned long *)&optprobe_template_entry)
+#define TMPL_CALL_IDX \
+ ((unsigned long *)&optprobe_template_call - (unsigned long *)&optprobe_template_entry)
+#define TMPL_END_IDX \
+ ((unsigned long *)&optprobe_template_end - (unsigned long *)&optprobe_template_entry)
+#define TMPL_ADD_SP \
+ ((unsigned long *)&optprobe_template_add_sp - (unsigned long *)&optprobe_template_entry)
+#define TMPL_SUB_SP \
+ ((unsigned long *)&optprobe_template_sub_sp - (unsigned long *)&optprobe_template_entry)
+
+/*
+ * ARM can always optimize an instruction when using ARM ISA, except
+ * instructions like 'str r0, [sp, r1]' which store to stack and unable
+ * to determine stack space consumption statically.
+ */
+int arch_prepared_optinsn(struct arch_optimized_insn *optinsn)
+{
+ return optinsn->insn != NULL;
+}
+
+/*
+ * In ARM ISA, kprobe opt always replace one instruction (4 bytes
+ * aligned and 4 bytes long). It is impossible to encounter another
+ * kprobe in the address range. So always return 0.
+ */
+int arch_check_optimized_kprobe(struct optimized_kprobe *op)
+{
+ return 0;
+}
+
+/* Caller must ensure addr & 3 == 0 */
+static int can_optimize(struct kprobe *kp)
+{
+ if (kp->ainsn.stack_space < 0)
+ return 0;
+ /*
+ * 255 is the biggest imm can be used in 'sub r0, r0, #<imm>'.
+ * Number larger than 255 needs special encoding.
+ */
+ if (kp->ainsn.stack_space > 255 - sizeof(struct pt_regs))
+ return 0;
+ return 1;
+}
+
+/* Free optimized instruction slot */
+static void
+__arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
+{
+ if (op->optinsn.insn) {
+ free_optinsn_slot(op->optinsn.insn, dirty);
+ op->optinsn.insn = NULL;
+ }
+}
+
+extern void kprobe_handler(struct pt_regs *regs);
+
+static void
+optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
+{
+ unsigned long flags;
+ struct kprobe *p = &op->kp;
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+
+ /* Save skipped registers */
+ regs->ARM_pc = (unsigned long)op->kp.addr;
+ regs->ARM_ORIG_r0 = ~0UL;
+
+ local_irq_save(flags);
+
+ if (kprobe_running()) {
+ kprobes_inc_nmissed_count(&op->kp);
+ } else {
+ __this_cpu_write(current_kprobe, &op->kp);
+ kcb->kprobe_status = KPROBE_HIT_ACTIVE;
+ opt_pre_handler(&op->kp, regs);
+ __this_cpu_write(current_kprobe, NULL);
+ }
+
+ /* In each case, we must singlestep the replaced instruction. */
+ op->kp.ainsn.insn_singlestep(p->opcode, &p->ainsn, regs);
+
+ local_irq_restore(flags);
+}
+
+int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *orig)
+{
+ kprobe_opcode_t *code;
+ unsigned long rel_chk;
+ unsigned long val;
+ unsigned long stack_protect = sizeof(struct pt_regs);
+
+ if (!can_optimize(orig))
+ return -EILSEQ;
+
+ code = get_optinsn_slot();
+ if (!code)
+ return -ENOMEM;
+
+ /*
+ * Verify if the address gap is in 32MiB range, because this uses
+ * a relative jump.
+ *
+ * kprobe opt use a 'b' instruction to branch to optinsn.insn.
+ * According to ARM manual, branch instruction is:
+ *
+ * 31 28 27 24 23 0
+ * +------+---+---+---+---+----------------+
+ * | cond | 1 | 0 | 1 | 0 | imm24 |
+ * +------+---+---+---+---+----------------+
+ *
+ * imm24 is a signed 24 bits integer. The real branch offset is computed
+ * by: imm32 = SignExtend(imm24:'00', 32);
+ *
+ * So the maximum forward branch should be:
+ * (0x007fffff << 2) = 0x01fffffc = 0x1fffffc
+ * The maximum backword branch should be:
+ * (0xff800000 << 2) = 0xfe000000 = -0x2000000
+ *
+ * We can simply check (rel & 0xfe000003):
+ * if rel is positive, (rel & 0xfe000000) shoule be 0
+ * if rel is negitive, (rel & 0xfe000000) should be 0xfe000000
+ * the last '3' is used for alignment checking.
+ */
+ rel_chk = (unsigned long)((long)code -
+ (long)orig->addr + 8) & 0xfe000003;
+
+ if ((rel_chk != 0) && (rel_chk != 0xfe000000)) {
+ /*
+ * Different from x86, we free code buf directly instead of
+ * calling __arch_remove_optimized_kprobe() because
+ * we have not fill any field in op.
+ */
+ free_optinsn_slot(code, 0);
+ return -ERANGE;
+ }
+
+ /* Copy arch-dep-instance from template. */
+ memcpy(code, &optprobe_template_entry,
+ TMPL_END_IDX * sizeof(kprobe_opcode_t));
+
+ /* Adjust buffer according to instruction. */
+ BUG_ON(orig->ainsn.stack_space < 0);
+
+ stack_protect += orig->ainsn.stack_space;
+
+ /* Should have been filtered by can_optimize(). */
+ BUG_ON(stack_protect > 255);
+
+ /* Create a 'sub sp, sp, #<stack_protect>' */
+ code[TMPL_SUB_SP] = __opcode_to_mem_arm(0xe24dd000 | stack_protect);
+ /* Create a 'add r3, sp, #<stack_protect>' */
+ code[TMPL_ADD_SP] = __opcode_to_mem_arm(0xe28d3000 | stack_protect);
+
+ /* Set probe information */
+ val = (unsigned long)op;
+ code[TMPL_VAL_IDX] = val;
+
+ /* Set probe function call */
+ val = (unsigned long)optimized_callback;
+ code[TMPL_CALL_IDX] = val;
+
+ flush_icache_range((unsigned long)code,
+ (unsigned long)(&code[TMPL_END_IDX]));
+
+ /* Set op->optinsn.insn means prepared. */
+ op->optinsn.insn = code;
+ return 0;
+}
+
+/*
+ * Similar to __arch_disarm_kprobe, operations which removing
+ * breakpoints must be wrapped by stop_machine to avoid racing.
+ */
+static __kprobes int __arch_optimize_kprobes(void *p)
+{
+ struct list_head *oplist = p;
+ struct optimized_kprobe *op, *tmp;
+
+ list_for_each_entry_safe(op, tmp, oplist, list) {
+ unsigned long insn;
+ WARN_ON(kprobe_disabled(&op->kp));
+
+ /*
+ * Backup instructions which will be replaced
+ * by jump address
+ */
+ memcpy(op->optinsn.copied_insn, op->kp.addr,
+ RELATIVEJUMP_SIZE);
+
+ insn = arm_gen_branch((unsigned long)op->kp.addr,
+ (unsigned long)op->optinsn.insn);
+ BUG_ON(insn == 0);
+
+ /*
+ * Make it a conditional branch if replaced insn
+ * is consitional
+ */
+ insn = (__mem_to_opcode_arm(
+ op->optinsn.copied_insn[0]) & 0xf0000000) |
+ (insn & 0x0fffffff);
+
+ patch_text(op->kp.addr, insn);
+
+ list_del_init(&op->list);
+ }
+ return 0;
+}
+
+void arch_optimize_kprobes(struct list_head *oplist)
+{
+ stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
+}
+
+void arch_unoptimize_kprobe(struct optimized_kprobe *op)
+{
+ arch_arm_kprobe(&op->kp);
+}
+
+/*
+ * Recover original instructions and breakpoints from relative jumps.
+ * Caller must call with locking kprobe_mutex.
+ */
+void arch_unoptimize_kprobes(struct list_head *oplist,
+ struct list_head *done_list)
+{
+ struct optimized_kprobe *op, *tmp;
+
+ list_for_each_entry_safe(op, tmp, oplist, list) {
+ arch_unoptimize_kprobe(op);
+ list_move(&op->list, done_list);
+ }
+}
+
+int arch_within_optimized_kprobe(struct optimized_kprobe *op,
+ unsigned long addr)
+{
+ return ((unsigned long)op->kp.addr <= addr &&
+ (unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr);
+}
+
+void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
+{
+ __arch_remove_optimized_kprobe(op, 1);
+}
diff --git a/samples/kprobes/kprobe_example.c b/samples/kprobes/kprobe_example.c
index 623f864..e4e45f1 100644
--- a/samples/kprobes/kprobe_example.c
+++ b/samples/kprobes/kprobe_example.c
@@ -94,7 +94,7 @@ static int handler_fault(struct kprobe *p, struct pt_regs *regs, int trapnr)
static int __init kprobe_init(void)
{
int ret;
- kp.addr = kallsyms_lookup_name(kp.symbol_name);
+ kp.addr = kallsyms_lookup_name(kp.symbol_name) + 8;
kp.symbol_name = NULL;
kp.pre_handler = handler_pre;
kp.post_handler = NULL;
--
1.8.4
^ permalink raw reply related [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 6:28 ` Wang Nan
@ 2014-12-08 11:04 ` Jon Medhurst (Tixy)
-1 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 11:04 UTC (permalink / raw)
To: Wang Nan
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> This patch introduce kprobeopt for ARM 32.
>
> Limitations:
> - Currently only kernel compiled with ARM ISA is supported.
>
> - Offset between probe point and optinsn slot must not larger than
> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
> things complex. Futher patch can make such optimization.
>
> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
> replace probed instruction by a 'b', branch to trampoline code and then
> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
> to execute kprobe handler. It also emulate/simulate replaced instruction.
>
> When unregistering kprobe, the deferred manner of unoptimizer may leave
> branch instruction before optimizer is called. Different from x86_64,
> which only copy the probed insn after optprobe_template_end and
> reexecute them, this patch call singlestep to emulate/simulate the insn
> directly. Futher patch can optimize this behavior.
>
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
> Cc: Will Deacon <will.deacon@arm.com>
> ---
[...]
> v13 -> v14:
> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
Think we need to use stop_machine differently, see comments on code
below.
> ---
> arch/arm/Kconfig | 1 +
> arch/arm/{kernel => include/asm}/insn.h | 0
> arch/arm/include/asm/kprobes.h | 29 +++
> arch/arm/kernel/Makefile | 2 +-
> arch/arm/kernel/ftrace.c | 3 +-
> arch/arm/kernel/jump_label.c | 3 +-
> arch/arm/probes/kprobes/Makefile | 1 +
> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
> samples/kprobes/kprobe_example.c | 2 +-
The change kprobe_example.c doesn't apply and I guess wasn't meant to be
included in the patch?
[...]
> +/*
> + * Similar to __arch_disarm_kprobe, operations which removing
> + * breakpoints must be wrapped by stop_machine to avoid racing.
> + */
> +static __kprobes int __arch_optimize_kprobes(void *p)
> +{
> + struct list_head *oplist = p;
> + struct optimized_kprobe *op, *tmp;
> +
> + list_for_each_entry_safe(op, tmp, oplist, list) {
> + unsigned long insn;
> + WARN_ON(kprobe_disabled(&op->kp));
> +
> + /*
> + * Backup instructions which will be replaced
> + * by jump address
> + */
> + memcpy(op->optinsn.copied_insn, op->kp.addr,
> + RELATIVEJUMP_SIZE);
> +
> + insn = arm_gen_branch((unsigned long)op->kp.addr,
> + (unsigned long)op->optinsn.insn);
> + BUG_ON(insn == 0);
> +
> + /*
> + * Make it a conditional branch if replaced insn
> + * is consitional
> + */
> + insn = (__mem_to_opcode_arm(
> + op->optinsn.copied_insn[0]) & 0xf0000000) |
> + (insn & 0x0fffffff);
> +
> + patch_text(op->kp.addr, insn);
patch_text() itself may use stop_machine under certain circumstances,
and if it were to do so, I believe that would cause the system to
lock/panic. So, this should be __patch_text() instead, but we would also
need to take care of the cache_ops_need_broadcast() case, where all
CPU's need to invalidate their own caches and we can't rely on just one
CPU executing the code patching whilst other CPUs spin and wait. Though
to make life easier, we could just not optimise kprobes in the legacy
cache_ops_need_broadcast() case.
> +
> + list_del_init(&op->list);
> + }
> + return 0;
> +}
> +
> +void arch_optimize_kprobes(struct list_head *oplist)
> +{
> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> +}
I believe passing cpu_online_mask above will cause
__arch_optimize_kprobes to be executed on every CPU, is this safe? If it
is, it's a serendipitous optimisation if each CPU can process different
probes in the list. If it's not safe, this needs to be NULL instead so
only one CPU executes the code.
However, I wonder if optimising all probes under a single stop_machine
call is the best thing to do because stop_machine does what it says and
prevents everything else in the system from running, including interrupt
handlers. Perhaps for system responsiveness this should be a single
stop_machine per kprobe? Though of course that compounds the overhead of
stop_machine use and puts another delay of one scheduler tick per probe.
(stop_machine waits for the next tick to schedule the threads to perform
the work which is why the test code takes so long to run).
What do people think?
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 11:04 ` Jon Medhurst (Tixy)
0 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 11:04 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> This patch introduce kprobeopt for ARM 32.
>
> Limitations:
> - Currently only kernel compiled with ARM ISA is supported.
>
> - Offset between probe point and optinsn slot must not larger than
> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
> things complex. Futher patch can make such optimization.
>
> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
> replace probed instruction by a 'b', branch to trampoline code and then
> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
> to execute kprobe handler. It also emulate/simulate replaced instruction.
>
> When unregistering kprobe, the deferred manner of unoptimizer may leave
> branch instruction before optimizer is called. Different from x86_64,
> which only copy the probed insn after optprobe_template_end and
> reexecute them, this patch call singlestep to emulate/simulate the insn
> directly. Futher patch can optimize this behavior.
>
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
> Cc: Will Deacon <will.deacon@arm.com>
> ---
[...]
> v13 -> v14:
> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
Think we need to use stop_machine differently, see comments on code
below.
> ---
> arch/arm/Kconfig | 1 +
> arch/arm/{kernel => include/asm}/insn.h | 0
> arch/arm/include/asm/kprobes.h | 29 +++
> arch/arm/kernel/Makefile | 2 +-
> arch/arm/kernel/ftrace.c | 3 +-
> arch/arm/kernel/jump_label.c | 3 +-
> arch/arm/probes/kprobes/Makefile | 1 +
> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
> samples/kprobes/kprobe_example.c | 2 +-
The change kprobe_example.c doesn't apply and I guess wasn't meant to be
included in the patch?
[...]
> +/*
> + * Similar to __arch_disarm_kprobe, operations which removing
> + * breakpoints must be wrapped by stop_machine to avoid racing.
> + */
> +static __kprobes int __arch_optimize_kprobes(void *p)
> +{
> + struct list_head *oplist = p;
> + struct optimized_kprobe *op, *tmp;
> +
> + list_for_each_entry_safe(op, tmp, oplist, list) {
> + unsigned long insn;
> + WARN_ON(kprobe_disabled(&op->kp));
> +
> + /*
> + * Backup instructions which will be replaced
> + * by jump address
> + */
> + memcpy(op->optinsn.copied_insn, op->kp.addr,
> + RELATIVEJUMP_SIZE);
> +
> + insn = arm_gen_branch((unsigned long)op->kp.addr,
> + (unsigned long)op->optinsn.insn);
> + BUG_ON(insn == 0);
> +
> + /*
> + * Make it a conditional branch if replaced insn
> + * is consitional
> + */
> + insn = (__mem_to_opcode_arm(
> + op->optinsn.copied_insn[0]) & 0xf0000000) |
> + (insn & 0x0fffffff);
> +
> + patch_text(op->kp.addr, insn);
patch_text() itself may use stop_machine under certain circumstances,
and if it were to do so, I believe that would cause the system to
lock/panic. So, this should be __patch_text() instead, but we would also
need to take care of the cache_ops_need_broadcast() case, where all
CPU's need to invalidate their own caches and we can't rely on just one
CPU executing the code patching whilst other CPUs spin and wait. Though
to make life easier, we could just not optimise kprobes in the legacy
cache_ops_need_broadcast() case.
> +
> + list_del_init(&op->list);
> + }
> + return 0;
> +}
> +
> +void arch_optimize_kprobes(struct list_head *oplist)
> +{
> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> +}
I believe passing cpu_online_mask above will cause
__arch_optimize_kprobes to be executed on every CPU, is this safe? If it
is, it's a serendipitous optimisation if each CPU can process different
probes in the list. If it's not safe, this needs to be NULL instead so
only one CPU executes the code.
However, I wonder if optimising all probes under a single stop_machine
call is the best thing to do because stop_machine does what it says and
prevents everything else in the system from running, including interrupt
handlers. Perhaps for system responsiveness this should be a single
stop_machine per kprobe? Though of course that compounds the overhead of
stop_machine use and puts another delay of one scheduler tick per probe.
(stop_machine waits for the next tick to schedule the threads to perform
the work which is why the test code takes so long to run).
What do people think?
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 11:04 ` Jon Medhurst (Tixy)
@ 2014-12-08 11:15 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 11:15 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>> This patch introduce kprobeopt for ARM 32.
>>
>> Limitations:
>> - Currently only kernel compiled with ARM ISA is supported.
>>
>> - Offset between probe point and optinsn slot must not larger than
>> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
>> things complex. Futher patch can make such optimization.
>>
>> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
>> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
>> replace probed instruction by a 'b', branch to trampoline code and then
>> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
>> to execute kprobe handler. It also emulate/simulate replaced instruction.
>>
>> When unregistering kprobe, the deferred manner of unoptimizer may leave
>> branch instruction before optimizer is called. Different from x86_64,
>> which only copy the probed insn after optprobe_template_end and
>> reexecute them, this patch call singlestep to emulate/simulate the insn
>> directly. Futher patch can optimize this behavior.
>>
>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
>> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
>> Cc: Will Deacon <will.deacon@arm.com>
>> ---
> [...]
>> v13 -> v14:
>> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
>
> Think we need to use stop_machine differently, see comments on code
> below.
Well, yes, I experienced one deadlock at serval minutes before.
I'm not very sure the reason and working on it now. I think it may caused
by recursivly stop_machine().
>
>> ---
>> arch/arm/Kconfig | 1 +
>> arch/arm/{kernel => include/asm}/insn.h | 0
>> arch/arm/include/asm/kprobes.h | 29 +++
>> arch/arm/kernel/Makefile | 2 +-
>> arch/arm/kernel/ftrace.c | 3 +-
>> arch/arm/kernel/jump_label.c | 3 +-
>> arch/arm/probes/kprobes/Makefile | 1 +
>> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
>> samples/kprobes/kprobe_example.c | 2 +-
>
> The change kprobe_example.c doesn't apply and I guess wasn't meant to be
> included in the patch?
>
Yes. These 2 lines are introduced by mistake.
> [...]
>> +/*
>> + * Similar to __arch_disarm_kprobe, operations which removing
>> + * breakpoints must be wrapped by stop_machine to avoid racing.
>> + */
>> +static __kprobes int __arch_optimize_kprobes(void *p)
>> +{
>> + struct list_head *oplist = p;
>> + struct optimized_kprobe *op, *tmp;
>> +
>> + list_for_each_entry_safe(op, tmp, oplist, list) {
>> + unsigned long insn;
>> + WARN_ON(kprobe_disabled(&op->kp));
>> +
>> + /*
>> + * Backup instructions which will be replaced
>> + * by jump address
>> + */
>> + memcpy(op->optinsn.copied_insn, op->kp.addr,
>> + RELATIVEJUMP_SIZE);
>> +
>> + insn = arm_gen_branch((unsigned long)op->kp.addr,
>> + (unsigned long)op->optinsn.insn);
>> + BUG_ON(insn == 0);
>> +
>> + /*
>> + * Make it a conditional branch if replaced insn
>> + * is consitional
>> + */
>> + insn = (__mem_to_opcode_arm(
>> + op->optinsn.copied_insn[0]) & 0xf0000000) |
>> + (insn & 0x0fffffff);
>> +
>> + patch_text(op->kp.addr, insn);
>
> patch_text() itself may use stop_machine under certain circumstances,
> and if it were to do so, I believe that would cause the system to
> lock/panic. So, this should be __patch_text() instead, but we would also
> need to take care of the cache_ops_need_broadcast() case, where all
> CPU's need to invalidate their own caches and we can't rely on just one
> CPU executing the code patching whilst other CPUs spin and wait. Though
> to make life easier, we could just not optimise kprobes in the legacy
> cache_ops_need_broadcast() case.
>
>> +
>> + list_del_init(&op->list);
>> + }
>> + return 0;
>> +}
>> +
>> +void arch_optimize_kprobes(struct list_head *oplist)
>> +{
>> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
>> +}
>
> I believe passing cpu_online_mask above will cause
> __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
> is, it's a serendipitous optimisation if each CPU can process different
> probes in the list. If it's not safe, this needs to be NULL instead so
> only one CPU executes the code.
>
This stop_machine() call is copied from arch_disarm_kprobe, I think their
senario should be similar.
> However, I wonder if optimising all probes under a single stop_machine
> call is the best thing to do because stop_machine does what it says and
> prevents everything else in the system from running, including interrupt
> handlers. Perhaps for system responsiveness this should be a single
> stop_machine per kprobe? Though of course that compounds the overhead of
> stop_machine use and puts another delay of one scheduler tick per probe.
> (stop_machine waits for the next tick to schedule the threads to perform
> the work which is why the test code takes so long to run).
>
> What do people think?
>
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 11:15 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 11:15 UTC (permalink / raw)
To: linux-arm-kernel
On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>> This patch introduce kprobeopt for ARM 32.
>>
>> Limitations:
>> - Currently only kernel compiled with ARM ISA is supported.
>>
>> - Offset between probe point and optinsn slot must not larger than
>> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
>> things complex. Futher patch can make such optimization.
>>
>> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
>> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
>> replace probed instruction by a 'b', branch to trampoline code and then
>> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
>> to execute kprobe handler. It also emulate/simulate replaced instruction.
>>
>> When unregistering kprobe, the deferred manner of unoptimizer may leave
>> branch instruction before optimizer is called. Different from x86_64,
>> which only copy the probed insn after optprobe_template_end and
>> reexecute them, this patch call singlestep to emulate/simulate the insn
>> directly. Futher patch can optimize this behavior.
>>
>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
>> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
>> Cc: Will Deacon <will.deacon@arm.com>
>> ---
> [...]
>> v13 -> v14:
>> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
>
> Think we need to use stop_machine differently, see comments on code
> below.
Well, yes, I experienced one deadlock at serval minutes before.
I'm not very sure the reason and working on it now. I think it may caused
by recursivly stop_machine().
>
>> ---
>> arch/arm/Kconfig | 1 +
>> arch/arm/{kernel => include/asm}/insn.h | 0
>> arch/arm/include/asm/kprobes.h | 29 +++
>> arch/arm/kernel/Makefile | 2 +-
>> arch/arm/kernel/ftrace.c | 3 +-
>> arch/arm/kernel/jump_label.c | 3 +-
>> arch/arm/probes/kprobes/Makefile | 1 +
>> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
>> samples/kprobes/kprobe_example.c | 2 +-
>
> The change kprobe_example.c doesn't apply and I guess wasn't meant to be
> included in the patch?
>
Yes. These 2 lines are introduced by mistake.
> [...]
>> +/*
>> + * Similar to __arch_disarm_kprobe, operations which removing
>> + * breakpoints must be wrapped by stop_machine to avoid racing.
>> + */
>> +static __kprobes int __arch_optimize_kprobes(void *p)
>> +{
>> + struct list_head *oplist = p;
>> + struct optimized_kprobe *op, *tmp;
>> +
>> + list_for_each_entry_safe(op, tmp, oplist, list) {
>> + unsigned long insn;
>> + WARN_ON(kprobe_disabled(&op->kp));
>> +
>> + /*
>> + * Backup instructions which will be replaced
>> + * by jump address
>> + */
>> + memcpy(op->optinsn.copied_insn, op->kp.addr,
>> + RELATIVEJUMP_SIZE);
>> +
>> + insn = arm_gen_branch((unsigned long)op->kp.addr,
>> + (unsigned long)op->optinsn.insn);
>> + BUG_ON(insn == 0);
>> +
>> + /*
>> + * Make it a conditional branch if replaced insn
>> + * is consitional
>> + */
>> + insn = (__mem_to_opcode_arm(
>> + op->optinsn.copied_insn[0]) & 0xf0000000) |
>> + (insn & 0x0fffffff);
>> +
>> + patch_text(op->kp.addr, insn);
>
> patch_text() itself may use stop_machine under certain circumstances,
> and if it were to do so, I believe that would cause the system to
> lock/panic. So, this should be __patch_text() instead, but we would also
> need to take care of the cache_ops_need_broadcast() case, where all
> CPU's need to invalidate their own caches and we can't rely on just one
> CPU executing the code patching whilst other CPUs spin and wait. Though
> to make life easier, we could just not optimise kprobes in the legacy
> cache_ops_need_broadcast() case.
>
>> +
>> + list_del_init(&op->list);
>> + }
>> + return 0;
>> +}
>> +
>> +void arch_optimize_kprobes(struct list_head *oplist)
>> +{
>> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
>> +}
>
> I believe passing cpu_online_mask above will cause
> __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
> is, it's a serendipitous optimisation if each CPU can process different
> probes in the list. If it's not safe, this needs to be NULL instead so
> only one CPU executes the code.
>
This stop_machine() call is copied from arch_disarm_kprobe, I think their
senario should be similar.
> However, I wonder if optimising all probes under a single stop_machine
> call is the best thing to do because stop_machine does what it says and
> prevents everything else in the system from running, including interrupt
> handlers. Perhaps for system responsiveness this should be a single
> stop_machine per kprobe? Though of course that compounds the overhead of
> stop_machine use and puts another delay of one scheduler tick per probe.
> (stop_machine waits for the next tick to schedule the threads to perform
> the work which is why the test code takes so long to run).
>
> What do people think?
>
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 11:15 ` Wang Nan
@ 2014-12-08 11:50 ` Jon Medhurst (Tixy)
-1 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 11:50 UTC (permalink / raw)
To: Wang Nan
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> > On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> >> This patch introduce kprobeopt for ARM 32.
> >>
> >> Limitations:
> >> - Currently only kernel compiled with ARM ISA is supported.
> >>
> >> - Offset between probe point and optinsn slot must not larger than
> >> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
> >> things complex. Futher patch can make such optimization.
> >>
> >> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
> >> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
> >> replace probed instruction by a 'b', branch to trampoline code and then
> >> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
> >> to execute kprobe handler. It also emulate/simulate replaced instruction.
> >>
> >> When unregistering kprobe, the deferred manner of unoptimizer may leave
> >> branch instruction before optimizer is called. Different from x86_64,
> >> which only copy the probed insn after optprobe_template_end and
> >> reexecute them, this patch call singlestep to emulate/simulate the insn
> >> directly. Futher patch can optimize this behavior.
> >>
> >> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> >> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> >> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
> >> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
> >> Cc: Will Deacon <will.deacon@arm.com>
> >> ---
> > [...]
> >> v13 -> v14:
> >> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
> >
> > Think we need to use stop_machine differently, see comments on code
> > below.
>
> Well, yes, I experienced one deadlock at serval minutes before.
> I'm not very sure the reason and working on it now. I think it may caused
> by recursivly stop_machine().
>
> >
> >> ---
> >> arch/arm/Kconfig | 1 +
> >> arch/arm/{kernel => include/asm}/insn.h | 0
> >> arch/arm/include/asm/kprobes.h | 29 +++
> >> arch/arm/kernel/Makefile | 2 +-
> >> arch/arm/kernel/ftrace.c | 3 +-
> >> arch/arm/kernel/jump_label.c | 3 +-
> >> arch/arm/probes/kprobes/Makefile | 1 +
> >> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
> >> samples/kprobes/kprobe_example.c | 2 +-
> >
> > The change kprobe_example.c doesn't apply and I guess wasn't meant to be
> > included in the patch?
> >
>
> Yes. These 2 lines are introduced by mistake.
>
> > [...]
> >> +/*
> >> + * Similar to __arch_disarm_kprobe, operations which removing
> >> + * breakpoints must be wrapped by stop_machine to avoid racing.
> >> + */
> >> +static __kprobes int __arch_optimize_kprobes(void *p)
> >> +{
> >> + struct list_head *oplist = p;
> >> + struct optimized_kprobe *op, *tmp;
> >> +
> >> + list_for_each_entry_safe(op, tmp, oplist, list) {
> >> + unsigned long insn;
> >> + WARN_ON(kprobe_disabled(&op->kp));
> >> +
> >> + /*
> >> + * Backup instructions which will be replaced
> >> + * by jump address
> >> + */
> >> + memcpy(op->optinsn.copied_insn, op->kp.addr,
> >> + RELATIVEJUMP_SIZE);
> >> +
> >> + insn = arm_gen_branch((unsigned long)op->kp.addr,
> >> + (unsigned long)op->optinsn.insn);
> >> + BUG_ON(insn == 0);
> >> +
> >> + /*
> >> + * Make it a conditional branch if replaced insn
> >> + * is consitional
> >> + */
> >> + insn = (__mem_to_opcode_arm(
> >> + op->optinsn.copied_insn[0]) & 0xf0000000) |
> >> + (insn & 0x0fffffff);
> >> +
> >> + patch_text(op->kp.addr, insn);
> >
> > patch_text() itself may use stop_machine under certain circumstances,
> > and if it were to do so, I believe that would cause the system to
> > lock/panic. So, this should be __patch_text() instead, but we would also
> > need to take care of the cache_ops_need_broadcast() case, where all
> > CPU's need to invalidate their own caches and we can't rely on just one
> > CPU executing the code patching whilst other CPUs spin and wait. Though
> > to make life easier, we could just not optimise kprobes in the legacy
> > cache_ops_need_broadcast() case.
> >
> >> +
> >> + list_del_init(&op->list);
> >> + }
> >> + return 0;
> >> +}
> >> +
> >> +void arch_optimize_kprobes(struct list_head *oplist)
> >> +{
> >> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> >> +}
> >
> > I believe passing cpu_online_mask above will cause
> > __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
> > is, it's a serendipitous optimisation if each CPU can process different
> > probes in the list. If it's not safe, this needs to be NULL instead so
> > only one CPU executes the code.
> >
>
> This stop_machine() call is copied from arch_disarm_kprobe, I think their
> senario should be similar.
arch_disarm_kprobe is just executing __patch_text on each cpu, which
pokes a word of memory with a new value and flushes caches for it.
arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
iterating over a list of probes and removing each one in turn, if this
is happening on multiple cpu's simultaneously, it's not clear to me that
such an operation is safe. list_del_init calls __list_del which does
next->prev = prev;
prev->next = next;
so what happens if another cpu is at the same time updating any of those
list entries? Without even fully analysing the code I can see that with
the fact that the list handling helpers have no memory barriers, that
the above two lines could be seen to execute in the reverse order, e.g.
prev->next = next;
next->prev = prev;
so another CPU could find and delete next before this one has finished
doing so. Would the list end up in a consistent state where no loops
develop and no probes are missed? I don't know the answer and a full
analysis would be complicated, but my gut feeling is that if a cpu can
observe the links in the list in an inconsistent state then only bad
things can result.
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 11:50 ` Jon Medhurst (Tixy)
0 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 11:50 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> > On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> >> This patch introduce kprobeopt for ARM 32.
> >>
> >> Limitations:
> >> - Currently only kernel compiled with ARM ISA is supported.
> >>
> >> - Offset between probe point and optinsn slot must not larger than
> >> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
> >> things complex. Futher patch can make such optimization.
> >>
> >> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
> >> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
> >> replace probed instruction by a 'b', branch to trampoline code and then
> >> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
> >> to execute kprobe handler. It also emulate/simulate replaced instruction.
> >>
> >> When unregistering kprobe, the deferred manner of unoptimizer may leave
> >> branch instruction before optimizer is called. Different from x86_64,
> >> which only copy the probed insn after optprobe_template_end and
> >> reexecute them, this patch call singlestep to emulate/simulate the insn
> >> directly. Futher patch can optimize this behavior.
> >>
> >> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> >> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> >> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
> >> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
> >> Cc: Will Deacon <will.deacon@arm.com>
> >> ---
> > [...]
> >> v13 -> v14:
> >> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
> >
> > Think we need to use stop_machine differently, see comments on code
> > below.
>
> Well, yes, I experienced one deadlock at serval minutes before.
> I'm not very sure the reason and working on it now. I think it may caused
> by recursivly stop_machine().
>
> >
> >> ---
> >> arch/arm/Kconfig | 1 +
> >> arch/arm/{kernel => include/asm}/insn.h | 0
> >> arch/arm/include/asm/kprobes.h | 29 +++
> >> arch/arm/kernel/Makefile | 2 +-
> >> arch/arm/kernel/ftrace.c | 3 +-
> >> arch/arm/kernel/jump_label.c | 3 +-
> >> arch/arm/probes/kprobes/Makefile | 1 +
> >> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
> >> samples/kprobes/kprobe_example.c | 2 +-
> >
> > The change kprobe_example.c doesn't apply and I guess wasn't meant to be
> > included in the patch?
> >
>
> Yes. These 2 lines are introduced by mistake.
>
> > [...]
> >> +/*
> >> + * Similar to __arch_disarm_kprobe, operations which removing
> >> + * breakpoints must be wrapped by stop_machine to avoid racing.
> >> + */
> >> +static __kprobes int __arch_optimize_kprobes(void *p)
> >> +{
> >> + struct list_head *oplist = p;
> >> + struct optimized_kprobe *op, *tmp;
> >> +
> >> + list_for_each_entry_safe(op, tmp, oplist, list) {
> >> + unsigned long insn;
> >> + WARN_ON(kprobe_disabled(&op->kp));
> >> +
> >> + /*
> >> + * Backup instructions which will be replaced
> >> + * by jump address
> >> + */
> >> + memcpy(op->optinsn.copied_insn, op->kp.addr,
> >> + RELATIVEJUMP_SIZE);
> >> +
> >> + insn = arm_gen_branch((unsigned long)op->kp.addr,
> >> + (unsigned long)op->optinsn.insn);
> >> + BUG_ON(insn == 0);
> >> +
> >> + /*
> >> + * Make it a conditional branch if replaced insn
> >> + * is consitional
> >> + */
> >> + insn = (__mem_to_opcode_arm(
> >> + op->optinsn.copied_insn[0]) & 0xf0000000) |
> >> + (insn & 0x0fffffff);
> >> +
> >> + patch_text(op->kp.addr, insn);
> >
> > patch_text() itself may use stop_machine under certain circumstances,
> > and if it were to do so, I believe that would cause the system to
> > lock/panic. So, this should be __patch_text() instead, but we would also
> > need to take care of the cache_ops_need_broadcast() case, where all
> > CPU's need to invalidate their own caches and we can't rely on just one
> > CPU executing the code patching whilst other CPUs spin and wait. Though
> > to make life easier, we could just not optimise kprobes in the legacy
> > cache_ops_need_broadcast() case.
> >
> >> +
> >> + list_del_init(&op->list);
> >> + }
> >> + return 0;
> >> +}
> >> +
> >> +void arch_optimize_kprobes(struct list_head *oplist)
> >> +{
> >> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> >> +}
> >
> > I believe passing cpu_online_mask above will cause
> > __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
> > is, it's a serendipitous optimisation if each CPU can process different
> > probes in the list. If it's not safe, this needs to be NULL instead so
> > only one CPU executes the code.
> >
>
> This stop_machine() call is copied from arch_disarm_kprobe, I think their
> senario should be similar.
arch_disarm_kprobe is just executing __patch_text on each cpu, which
pokes a word of memory with a new value and flushes caches for it.
arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
iterating over a list of probes and removing each one in turn, if this
is happening on multiple cpu's simultaneously, it's not clear to me that
such an operation is safe. list_del_init calls __list_del which does
next->prev = prev;
prev->next = next;
so what happens if another cpu is at the same time updating any of those
list entries? Without even fully analysing the code I can see that with
the fact that the list handling helpers have no memory barriers, that
the above two lines could be seen to execute in the reverse order, e.g.
prev->next = next;
next->prev = prev;
so another CPU could find and delete next before this one has finished
doing so. Would the list end up in a consistent state where no loops
develop and no probes are missed? I don't know the answer and a full
analysis would be complicated, but my gut feeling is that if a cpu can
observe the links in the list in an inconsistent state then only bad
things can result.
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 11:50 ` Jon Medhurst (Tixy)
@ 2014-12-08 12:06 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 12:06 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>>>> This patch introduce kprobeopt for ARM 32.
>>>>
>>>> Limitations:
>>>> - Currently only kernel compiled with ARM ISA is supported.
>>>>
>>>> - Offset between probe point and optinsn slot must not larger than
>>>> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
>>>> things complex. Futher patch can make such optimization.
>>>>
>>>> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
>>>> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
>>>> replace probed instruction by a 'b', branch to trampoline code and then
>>>> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
>>>> to execute kprobe handler. It also emulate/simulate replaced instruction.
>>>>
>>>> When unregistering kprobe, the deferred manner of unoptimizer may leave
>>>> branch instruction before optimizer is called. Different from x86_64,
>>>> which only copy the probed insn after optprobe_template_end and
>>>> reexecute them, this patch call singlestep to emulate/simulate the insn
>>>> directly. Futher patch can optimize this behavior.
>>>>
>>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>>> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>>>> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
>>>> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
>>>> Cc: Will Deacon <will.deacon@arm.com>
>>>> ---
>>> [...]
>>>> v13 -> v14:
>>>> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
>>>
>>> Think we need to use stop_machine differently, see comments on code
>>> below.
>>
>> Well, yes, I experienced one deadlock at serval minutes before.
>> I'm not very sure the reason and working on it now. I think it may caused
>> by recursivly stop_machine().
>>
>>>
>>>> ---
>>>> arch/arm/Kconfig | 1 +
>>>> arch/arm/{kernel => include/asm}/insn.h | 0
>>>> arch/arm/include/asm/kprobes.h | 29 +++
>>>> arch/arm/kernel/Makefile | 2 +-
>>>> arch/arm/kernel/ftrace.c | 3 +-
>>>> arch/arm/kernel/jump_label.c | 3 +-
>>>> arch/arm/probes/kprobes/Makefile | 1 +
>>>> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
>>>> samples/kprobes/kprobe_example.c | 2 +-
>>>
>>> The change kprobe_example.c doesn't apply and I guess wasn't meant to be
>>> included in the patch?
>>>
>>
>> Yes. These 2 lines are introduced by mistake.
>>
>>> [...]
>>>> +/*
>>>> + * Similar to __arch_disarm_kprobe, operations which removing
>>>> + * breakpoints must be wrapped by stop_machine to avoid racing.
>>>> + */
>>>> +static __kprobes int __arch_optimize_kprobes(void *p)
>>>> +{
>>>> + struct list_head *oplist = p;
>>>> + struct optimized_kprobe *op, *tmp;
>>>> +
>>>> + list_for_each_entry_safe(op, tmp, oplist, list) {
>>>> + unsigned long insn;
>>>> + WARN_ON(kprobe_disabled(&op->kp));
>>>> +
>>>> + /*
>>>> + * Backup instructions which will be replaced
>>>> + * by jump address
>>>> + */
>>>> + memcpy(op->optinsn.copied_insn, op->kp.addr,
>>>> + RELATIVEJUMP_SIZE);
>>>> +
>>>> + insn = arm_gen_branch((unsigned long)op->kp.addr,
>>>> + (unsigned long)op->optinsn.insn);
>>>> + BUG_ON(insn == 0);
>>>> +
>>>> + /*
>>>> + * Make it a conditional branch if replaced insn
>>>> + * is consitional
>>>> + */
>>>> + insn = (__mem_to_opcode_arm(
>>>> + op->optinsn.copied_insn[0]) & 0xf0000000) |
>>>> + (insn & 0x0fffffff);
>>>> +
>>>> + patch_text(op->kp.addr, insn);
>>>
>>> patch_text() itself may use stop_machine under certain circumstances,
>>> and if it were to do so, I believe that would cause the system to
>>> lock/panic. So, this should be __patch_text() instead, but we would also
>>> need to take care of the cache_ops_need_broadcast() case, where all
>>> CPU's need to invalidate their own caches and we can't rely on just one
>>> CPU executing the code patching whilst other CPUs spin and wait. Though
>>> to make life easier, we could just not optimise kprobes in the legacy
>>> cache_ops_need_broadcast() case.
>>>
>>>> +
>>>> + list_del_init(&op->list);
>>>> + }
>>>> + return 0;
>>>> +}
>>>> +
>>>> +void arch_optimize_kprobes(struct list_head *oplist)
>>>> +{
>>>> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
>>>> +}
>>>
>>> I believe passing cpu_online_mask above will cause
>>> __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
>>> is, it's a serendipitous optimisation if each CPU can process different
>>> probes in the list. If it's not safe, this needs to be NULL instead so
>>> only one CPU executes the code.
>>>
>>
>> This stop_machine() call is copied from arch_disarm_kprobe, I think their
>> senario should be similar.
>
> arch_disarm_kprobe is just executing __patch_text on each cpu, which
> pokes a word of memory with a new value and flushes caches for it.
>
> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> iterating over a list of probes and removing each one in turn, if this
> is happening on multiple cpu's simultaneously, it's not clear to me that
> such an operation is safe. list_del_init calls __list_del which does
>
> next->prev = prev;
> prev->next = next;
>
> so what happens if another cpu is at the same time updating any of those
> list entries? Without even fully analysing the code I can see that with
> the fact that the list handling helpers have no memory barriers, that
> the above two lines could be seen to execute in the reverse order, e.g.
>
> prev->next = next;
> next->prev = prev;
>
> so another CPU could find and delete next before this one has finished
> doing so. Would the list end up in a consistent state where no loops
> develop and no probes are missed? I don't know the answer and a full
> analysis would be complicated, but my gut feeling is that if a cpu can
> observe the links in the list in an inconsistent state then only bad
> things can result.
>
I see the problem.
I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
stop_machine() is required when removing breakpoint, so I'd like to define
a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
Do you think it is a good idea?
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 12:06 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 12:06 UTC (permalink / raw)
To: linux-arm-kernel
On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>>>> This patch introduce kprobeopt for ARM 32.
>>>>
>>>> Limitations:
>>>> - Currently only kernel compiled with ARM ISA is supported.
>>>>
>>>> - Offset between probe point and optinsn slot must not larger than
>>>> 32MiB. Masami Hiramatsu suggests replacing 2 words, it will make
>>>> things complex. Futher patch can make such optimization.
>>>>
>>>> Kprobe opt on ARM is relatively simpler than kprobe opt on x86 because
>>>> ARM instruction is always 4 bytes aligned and 4 bytes long. This patch
>>>> replace probed instruction by a 'b', branch to trampoline code and then
>>>> calls optimized_callback(). optimized_callback() calls opt_pre_handler()
>>>> to execute kprobe handler. It also emulate/simulate replaced instruction.
>>>>
>>>> When unregistering kprobe, the deferred manner of unoptimizer may leave
>>>> branch instruction before optimizer is called. Different from x86_64,
>>>> which only copy the probed insn after optprobe_template_end and
>>>> reexecute them, this patch call singlestep to emulate/simulate the insn
>>>> directly. Futher patch can optimize this behavior.
>>>>
>>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>>> Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>>>> Cc: Jon Medhurst (Tixy) <tixy@linaro.org>
>>>> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk>
>>>> Cc: Will Deacon <will.deacon@arm.com>
>>>> ---
>>> [...]
>>>> v13 -> v14:
>>>> - Use stop_machine to wrap arch_optimize_kprobes to avoid a racing.
>>>
>>> Think we need to use stop_machine differently, see comments on code
>>> below.
>>
>> Well, yes, I experienced one deadlock at serval minutes before.
>> I'm not very sure the reason and working on it now. I think it may caused
>> by recursivly stop_machine().
>>
>>>
>>>> ---
>>>> arch/arm/Kconfig | 1 +
>>>> arch/arm/{kernel => include/asm}/insn.h | 0
>>>> arch/arm/include/asm/kprobes.h | 29 +++
>>>> arch/arm/kernel/Makefile | 2 +-
>>>> arch/arm/kernel/ftrace.c | 3 +-
>>>> arch/arm/kernel/jump_label.c | 3 +-
>>>> arch/arm/probes/kprobes/Makefile | 1 +
>>>> arch/arm/probes/kprobes/opt-arm.c | 322 ++++++++++++++++++++++++++++++++
>>>> samples/kprobes/kprobe_example.c | 2 +-
>>>
>>> The change kprobe_example.c doesn't apply and I guess wasn't meant to be
>>> included in the patch?
>>>
>>
>> Yes. These 2 lines are introduced by mistake.
>>
>>> [...]
>>>> +/*
>>>> + * Similar to __arch_disarm_kprobe, operations which removing
>>>> + * breakpoints must be wrapped by stop_machine to avoid racing.
>>>> + */
>>>> +static __kprobes int __arch_optimize_kprobes(void *p)
>>>> +{
>>>> + struct list_head *oplist = p;
>>>> + struct optimized_kprobe *op, *tmp;
>>>> +
>>>> + list_for_each_entry_safe(op, tmp, oplist, list) {
>>>> + unsigned long insn;
>>>> + WARN_ON(kprobe_disabled(&op->kp));
>>>> +
>>>> + /*
>>>> + * Backup instructions which will be replaced
>>>> + * by jump address
>>>> + */
>>>> + memcpy(op->optinsn.copied_insn, op->kp.addr,
>>>> + RELATIVEJUMP_SIZE);
>>>> +
>>>> + insn = arm_gen_branch((unsigned long)op->kp.addr,
>>>> + (unsigned long)op->optinsn.insn);
>>>> + BUG_ON(insn == 0);
>>>> +
>>>> + /*
>>>> + * Make it a conditional branch if replaced insn
>>>> + * is consitional
>>>> + */
>>>> + insn = (__mem_to_opcode_arm(
>>>> + op->optinsn.copied_insn[0]) & 0xf0000000) |
>>>> + (insn & 0x0fffffff);
>>>> +
>>>> + patch_text(op->kp.addr, insn);
>>>
>>> patch_text() itself may use stop_machine under certain circumstances,
>>> and if it were to do so, I believe that would cause the system to
>>> lock/panic. So, this should be __patch_text() instead, but we would also
>>> need to take care of the cache_ops_need_broadcast() case, where all
>>> CPU's need to invalidate their own caches and we can't rely on just one
>>> CPU executing the code patching whilst other CPUs spin and wait. Though
>>> to make life easier, we could just not optimise kprobes in the legacy
>>> cache_ops_need_broadcast() case.
>>>
>>>> +
>>>> + list_del_init(&op->list);
>>>> + }
>>>> + return 0;
>>>> +}
>>>> +
>>>> +void arch_optimize_kprobes(struct list_head *oplist)
>>>> +{
>>>> + stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
>>>> +}
>>>
>>> I believe passing cpu_online_mask above will cause
>>> __arch_optimize_kprobes to be executed on every CPU, is this safe? If it
>>> is, it's a serendipitous optimisation if each CPU can process different
>>> probes in the list. If it's not safe, this needs to be NULL instead so
>>> only one CPU executes the code.
>>>
>>
>> This stop_machine() call is copied from arch_disarm_kprobe, I think their
>> senario should be similar.
>
> arch_disarm_kprobe is just executing __patch_text on each cpu, which
> pokes a word of memory with a new value and flushes caches for it.
>
> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> iterating over a list of probes and removing each one in turn, if this
> is happening on multiple cpu's simultaneously, it's not clear to me that
> such an operation is safe. list_del_init calls __list_del which does
>
> next->prev = prev;
> prev->next = next;
>
> so what happens if another cpu is at the same time updating any of those
> list entries? Without even fully analysing the code I can see that with
> the fact that the list handling helpers have no memory barriers, that
> the above two lines could be seen to execute in the reverse order, e.g.
>
> prev->next = next;
> next->prev = prev;
>
> so another CPU could find and delete next before this one has finished
> doing so. Would the list end up in a consistent state where no loops
> develop and no probes are missed? I don't know the answer and a full
> analysis would be complicated, but my gut feeling is that if a cpu can
> observe the links in the list in an inconsistent state then only bad
> things can result.
>
I see the problem.
I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
stop_machine() is required when removing breakpoint, so I'd like to define
a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
Do you think it is a good idea?
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 12:06 ` Wang Nan
@ 2014-12-08 12:31 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 12:31 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On 2014/12/8 20:06, Wang Nan wrote:
> On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
>> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
[...]
>>
>> so another CPU could find and delete next before this one has finished
>> doing so. Would the list end up in a consistent state where no loops
>> develop and no probes are missed? I don't know the answer and a full
>> analysis would be complicated, but my gut feeling is that if a cpu can
>> observe the links in the list in an inconsistent state then only bad
>> things can result.
>>
>
> I see the problem.
>
> I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
> stop_machine() is required when removing breakpoint, so I'd like to define
> a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
> Do you think it is a good idea?
>
>
What I mean is something like this:
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 3a58db4..efd8ab1 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -163,19 +163,31 @@ void __kprobes arch_arm_kprobe(struct kprobe *p)
* memory. It is also needed to atomically set the two half-words of a 32-bit
* Thumb breakpoint.
*/
-int __kprobes __arch_disarm_kprobe(void *p)
-{
- struct kprobe *kp = p;
- void *addr = (void *)((uintptr_t)kp->addr & ~1);
-
- __patch_text(addr, kp->opcode);
+struct patch {
+ void *addr;
+ unsigned int insn;
+};
+static int __remove_breakpoint(void *data)
+{
+ struct patch *p = data;
+ __patch_text(p->addr, p->insn);
return 0;
}
+void __kprobes remove_breakpoint(void *addr, unsigned int insn)
+{
+ struct patch p = {
+ .addr = addr,
+ .insn = insn,
+ };
+ stop_machine(__remove_breakpoint, &p, cpu_online_mask);
+}
+
void __kprobes arch_disarm_kprobe(struct kprobe *p)
{
- stop_machine(__arch_disarm_kprobe, p, cpu_online_mask);
+ remove_breakpoint((void *)((uintptr_t)p->addr & ~1),
+ p->opcode);
}
void __kprobes arch_remove_kprobe(struct kprobe *p)
diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
index f88c79f..7b7c334 100644
--- a/arch/arm/probes/kprobes/core.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -30,6 +30,8 @@
#define KPROBE_THUMB16_BREAKPOINT_INSTRUCTION 0xde18
#define KPROBE_THUMB32_BREAKPOINT_INSTRUCTION 0xf7f0a018
+extern void remove_breakpoint(void *addr, unsigned int insn);
+
enum probes_insn __kprobes
kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
const struct decode_header *h);
diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
index afbfeef..a1a1882 100644
--- a/arch/arm/probes/kprobes/opt-arm.c
+++ b/arch/arm/probes/kprobes/opt-arm.c
@@ -28,8 +28,9 @@
#include <asm/insn.h>
/* for patch_text */
#include <asm/patch.h>
-/* for stop_machine */
-#include <linux/stop_machine.h>
+
+#include "core.h"
+
/*
* NOTE: the first sub and add instruction will be modified according
* to the stack cost of the instruction.
@@ -245,13 +246,8 @@ int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *or
return 0;
}
-/*
- * Similar to __arch_disarm_kprobe, operations which removing
- * breakpoints must be wrapped by stop_machine to avoid racing.
- */
-static __kprobes int __arch_optimize_kprobes(void *p)
+void __kprobes arch_optimize_kprobes(struct list_head *oplist)
{
- struct list_head *oplist = p;
struct optimized_kprobe *op, *tmp;
list_for_each_entry_safe(op, tmp, oplist, list) {
@@ -277,16 +273,15 @@ static __kprobes int __arch_optimize_kprobes(void *p)
op->optinsn.copied_insn[0]) & 0xf0000000) |
(insn & 0x0fffffff);
- patch_text(op->kp.addr, insn);
+ /*
+ * Similar to __arch_disarm_kprobe, operations which
+ * removing breakpoints must be wrapped by stop_machine
+ * to avoid racing.
+ */
+ remove_breakpoint(op->kp.addr, insn);
list_del_init(&op->list);
}
- return 0;
-}
-
-void arch_optimize_kprobes(struct list_head *oplist)
-{
- stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
}
void arch_unoptimize_kprobe(struct optimized_kprobe *op)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
^ permalink raw reply related [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 12:31 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 12:31 UTC (permalink / raw)
To: linux-arm-kernel
On 2014/12/8 20:06, Wang Nan wrote:
> On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
>> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
[...]
>>
>> so another CPU could find and delete next before this one has finished
>> doing so. Would the list end up in a consistent state where no loops
>> develop and no probes are missed? I don't know the answer and a full
>> analysis would be complicated, but my gut feeling is that if a cpu can
>> observe the links in the list in an inconsistent state then only bad
>> things can result.
>>
>
> I see the problem.
>
> I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
> stop_machine() is required when removing breakpoint, so I'd like to define
> a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
> Do you think it is a good idea?
>
>
What I mean is something like this:
diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
index 3a58db4..efd8ab1 100644
--- a/arch/arm/probes/kprobes/core.c
+++ b/arch/arm/probes/kprobes/core.c
@@ -163,19 +163,31 @@ void __kprobes arch_arm_kprobe(struct kprobe *p)
* memory. It is also needed to atomically set the two half-words of a 32-bit
* Thumb breakpoint.
*/
-int __kprobes __arch_disarm_kprobe(void *p)
-{
- struct kprobe *kp = p;
- void *addr = (void *)((uintptr_t)kp->addr & ~1);
-
- __patch_text(addr, kp->opcode);
+struct patch {
+ void *addr;
+ unsigned int insn;
+};
+static int __remove_breakpoint(void *data)
+{
+ struct patch *p = data;
+ __patch_text(p->addr, p->insn);
return 0;
}
+void __kprobes remove_breakpoint(void *addr, unsigned int insn)
+{
+ struct patch p = {
+ .addr = addr,
+ .insn = insn,
+ };
+ stop_machine(__remove_breakpoint, &p, cpu_online_mask);
+}
+
void __kprobes arch_disarm_kprobe(struct kprobe *p)
{
- stop_machine(__arch_disarm_kprobe, p, cpu_online_mask);
+ remove_breakpoint((void *)((uintptr_t)p->addr & ~1),
+ p->opcode);
}
void __kprobes arch_remove_kprobe(struct kprobe *p)
diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
index f88c79f..7b7c334 100644
--- a/arch/arm/probes/kprobes/core.h
+++ b/arch/arm/probes/kprobes/core.h
@@ -30,6 +30,8 @@
#define KPROBE_THUMB16_BREAKPOINT_INSTRUCTION 0xde18
#define KPROBE_THUMB32_BREAKPOINT_INSTRUCTION 0xf7f0a018
+extern void remove_breakpoint(void *addr, unsigned int insn);
+
enum probes_insn __kprobes
kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
const struct decode_header *h);
diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
index afbfeef..a1a1882 100644
--- a/arch/arm/probes/kprobes/opt-arm.c
+++ b/arch/arm/probes/kprobes/opt-arm.c
@@ -28,8 +28,9 @@
#include <asm/insn.h>
/* for patch_text */
#include <asm/patch.h>
-/* for stop_machine */
-#include <linux/stop_machine.h>
+
+#include "core.h"
+
/*
* NOTE: the first sub and add instruction will be modified according
* to the stack cost of the instruction.
@@ -245,13 +246,8 @@ int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *or
return 0;
}
-/*
- * Similar to __arch_disarm_kprobe, operations which removing
- * breakpoints must be wrapped by stop_machine to avoid racing.
- */
-static __kprobes int __arch_optimize_kprobes(void *p)
+void __kprobes arch_optimize_kprobes(struct list_head *oplist)
{
- struct list_head *oplist = p;
struct optimized_kprobe *op, *tmp;
list_for_each_entry_safe(op, tmp, oplist, list) {
@@ -277,16 +273,15 @@ static __kprobes int __arch_optimize_kprobes(void *p)
op->optinsn.copied_insn[0]) & 0xf0000000) |
(insn & 0x0fffffff);
- patch_text(op->kp.addr, insn);
+ /*
+ * Similar to __arch_disarm_kprobe, operations which
+ * removing breakpoints must be wrapped by stop_machine
+ * to avoid racing.
+ */
+ remove_breakpoint(op->kp.addr, insn);
list_del_init(&op->list);
}
- return 0;
-}
-
-void arch_optimize_kprobes(struct list_head *oplist)
-{
- stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
}
void arch_unoptimize_kprobe(struct optimized_kprobe *op)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
^ permalink raw reply related [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 12:31 ` Wang Nan
@ 2014-12-08 13:22 ` Jon Medhurst (Tixy)
-1 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 13:22 UTC (permalink / raw)
To: Wang Nan
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On Mon, 2014-12-08 at 20:31 +0800, Wang Nan wrote:
> On 2014/12/8 20:06, Wang Nan wrote:
> > On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
> >> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
> >>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> >>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> [...]
> >>
> >> so another CPU could find and delete next before this one has finished
> >> doing so. Would the list end up in a consistent state where no loops
> >> develop and no probes are missed? I don't know the answer and a full
> >> analysis would be complicated, but my gut feeling is that if a cpu can
> >> observe the links in the list in an inconsistent state then only bad
> >> things can result.
> >>
> >
> > I see the problem.
> >
> > I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
> > stop_machine() is required when removing breakpoint, so I'd like to define
> > a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
> > Do you think it is a good idea?
> >
> >
>
> What I mean is something like this:
Yes, that should work, though as remove_breakpoint is a globally visible
symbol, I suggest a less generic name for it, perhaps
remove_kprobe_breakpoint ?
--
Tixy
> diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
> index 3a58db4..efd8ab1 100644
> --- a/arch/arm/probes/kprobes/core.c
> +++ b/arch/arm/probes/kprobes/core.c
> @@ -163,19 +163,31 @@ void __kprobes arch_arm_kprobe(struct kprobe *p)
> * memory. It is also needed to atomically set the two half-words of a 32-bit
> * Thumb breakpoint.
> */
> -int __kprobes __arch_disarm_kprobe(void *p)
> -{
> - struct kprobe *kp = p;
> - void *addr = (void *)((uintptr_t)kp->addr & ~1);
> -
> - __patch_text(addr, kp->opcode);
> +struct patch {
> + void *addr;
> + unsigned int insn;
> +};
>
> +static int __remove_breakpoint(void *data)
> +{
> + struct patch *p = data;
> + __patch_text(p->addr, p->insn);
> return 0;
> }
>
> +void __kprobes remove_breakpoint(void *addr, unsigned int insn)
> +{
> + struct patch p = {
> + .addr = addr,
> + .insn = insn,
> + };
> + stop_machine(__remove_breakpoint, &p, cpu_online_mask);
> +}
> +
> void __kprobes arch_disarm_kprobe(struct kprobe *p)
> {
> - stop_machine(__arch_disarm_kprobe, p, cpu_online_mask);
> + remove_breakpoint((void *)((uintptr_t)p->addr & ~1),
> + p->opcode);
> }
>
> void __kprobes arch_remove_kprobe(struct kprobe *p)
> diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
> index f88c79f..7b7c334 100644
> --- a/arch/arm/probes/kprobes/core.h
> +++ b/arch/arm/probes/kprobes/core.h
> @@ -30,6 +30,8 @@
> #define KPROBE_THUMB16_BREAKPOINT_INSTRUCTION 0xde18
> #define KPROBE_THUMB32_BREAKPOINT_INSTRUCTION 0xf7f0a018
>
> +extern void remove_breakpoint(void *addr, unsigned int insn);
> +
> enum probes_insn __kprobes
> kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
> const struct decode_header *h);
> diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
> index afbfeef..a1a1882 100644
> --- a/arch/arm/probes/kprobes/opt-arm.c
> +++ b/arch/arm/probes/kprobes/opt-arm.c
> @@ -28,8 +28,9 @@
> #include <asm/insn.h>
> /* for patch_text */
> #include <asm/patch.h>
> -/* for stop_machine */
> -#include <linux/stop_machine.h>
> +
> +#include "core.h"
> +
> /*
> * NOTE: the first sub and add instruction will be modified according
> * to the stack cost of the instruction.
> @@ -245,13 +246,8 @@ int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *or
> return 0;
> }
>
> -/*
> - * Similar to __arch_disarm_kprobe, operations which removing
> - * breakpoints must be wrapped by stop_machine to avoid racing.
> - */
> -static __kprobes int __arch_optimize_kprobes(void *p)
> +void __kprobes arch_optimize_kprobes(struct list_head *oplist)
> {
> - struct list_head *oplist = p;
> struct optimized_kprobe *op, *tmp;
>
> list_for_each_entry_safe(op, tmp, oplist, list) {
> @@ -277,16 +273,15 @@ static __kprobes int __arch_optimize_kprobes(void *p)
> op->optinsn.copied_insn[0]) & 0xf0000000) |
> (insn & 0x0fffffff);
>
> - patch_text(op->kp.addr, insn);
> + /*
> + * Similar to __arch_disarm_kprobe, operations which
> + * removing breakpoints must be wrapped by stop_machine
> + * to avoid racing.
> + */
> + remove_breakpoint(op->kp.addr, insn);
>
> list_del_init(&op->list);
> }
> - return 0;
> -}
> -
> -void arch_optimize_kprobes(struct list_head *oplist)
> -{
> - stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> }
>
> void arch_unoptimize_kprobe(struct optimized_kprobe *op)
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 13:22 ` Jon Medhurst (Tixy)
0 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-08 13:22 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, 2014-12-08 at 20:31 +0800, Wang Nan wrote:
> On 2014/12/8 20:06, Wang Nan wrote:
> > On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
> >> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
> >>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
> >>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
> [...]
> >>
> >> so another CPU could find and delete next before this one has finished
> >> doing so. Would the list end up in a consistent state where no loops
> >> develop and no probes are missed? I don't know the answer and a full
> >> analysis would be complicated, but my gut feeling is that if a cpu can
> >> observe the links in the list in an inconsistent state then only bad
> >> things can result.
> >>
> >
> > I see the problem.
> >
> > I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
> > stop_machine() is required when removing breakpoint, so I'd like to define
> > a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
> > Do you think it is a good idea?
> >
> >
>
> What I mean is something like this:
Yes, that should work, though as remove_breakpoint is a globally visible
symbol, I suggest a less generic name for it, perhaps
remove_kprobe_breakpoint ?
--
Tixy
> diff --git a/arch/arm/probes/kprobes/core.c b/arch/arm/probes/kprobes/core.c
> index 3a58db4..efd8ab1 100644
> --- a/arch/arm/probes/kprobes/core.c
> +++ b/arch/arm/probes/kprobes/core.c
> @@ -163,19 +163,31 @@ void __kprobes arch_arm_kprobe(struct kprobe *p)
> * memory. It is also needed to atomically set the two half-words of a 32-bit
> * Thumb breakpoint.
> */
> -int __kprobes __arch_disarm_kprobe(void *p)
> -{
> - struct kprobe *kp = p;
> - void *addr = (void *)((uintptr_t)kp->addr & ~1);
> -
> - __patch_text(addr, kp->opcode);
> +struct patch {
> + void *addr;
> + unsigned int insn;
> +};
>
> +static int __remove_breakpoint(void *data)
> +{
> + struct patch *p = data;
> + __patch_text(p->addr, p->insn);
> return 0;
> }
>
> +void __kprobes remove_breakpoint(void *addr, unsigned int insn)
> +{
> + struct patch p = {
> + .addr = addr,
> + .insn = insn,
> + };
> + stop_machine(__remove_breakpoint, &p, cpu_online_mask);
> +}
> +
> void __kprobes arch_disarm_kprobe(struct kprobe *p)
> {
> - stop_machine(__arch_disarm_kprobe, p, cpu_online_mask);
> + remove_breakpoint((void *)((uintptr_t)p->addr & ~1),
> + p->opcode);
> }
>
> void __kprobes arch_remove_kprobe(struct kprobe *p)
> diff --git a/arch/arm/probes/kprobes/core.h b/arch/arm/probes/kprobes/core.h
> index f88c79f..7b7c334 100644
> --- a/arch/arm/probes/kprobes/core.h
> +++ b/arch/arm/probes/kprobes/core.h
> @@ -30,6 +30,8 @@
> #define KPROBE_THUMB16_BREAKPOINT_INSTRUCTION 0xde18
> #define KPROBE_THUMB32_BREAKPOINT_INSTRUCTION 0xf7f0a018
>
> +extern void remove_breakpoint(void *addr, unsigned int insn);
> +
> enum probes_insn __kprobes
> kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_probes_insn *asi,
> const struct decode_header *h);
> diff --git a/arch/arm/probes/kprobes/opt-arm.c b/arch/arm/probes/kprobes/opt-arm.c
> index afbfeef..a1a1882 100644
> --- a/arch/arm/probes/kprobes/opt-arm.c
> +++ b/arch/arm/probes/kprobes/opt-arm.c
> @@ -28,8 +28,9 @@
> #include <asm/insn.h>
> /* for patch_text */
> #include <asm/patch.h>
> -/* for stop_machine */
> -#include <linux/stop_machine.h>
> +
> +#include "core.h"
> +
> /*
> * NOTE: the first sub and add instruction will be modified according
> * to the stack cost of the instruction.
> @@ -245,13 +246,8 @@ int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *or
> return 0;
> }
>
> -/*
> - * Similar to __arch_disarm_kprobe, operations which removing
> - * breakpoints must be wrapped by stop_machine to avoid racing.
> - */
> -static __kprobes int __arch_optimize_kprobes(void *p)
> +void __kprobes arch_optimize_kprobes(struct list_head *oplist)
> {
> - struct list_head *oplist = p;
> struct optimized_kprobe *op, *tmp;
>
> list_for_each_entry_safe(op, tmp, oplist, list) {
> @@ -277,16 +273,15 @@ static __kprobes int __arch_optimize_kprobes(void *p)
> op->optinsn.copied_insn[0]) & 0xf0000000) |
> (insn & 0x0fffffff);
>
> - patch_text(op->kp.addr, insn);
> + /*
> + * Similar to __arch_disarm_kprobe, operations which
> + * removing breakpoints must be wrapped by stop_machine
> + * to avoid racing.
> + */
> + remove_breakpoint(op->kp.addr, insn);
>
> list_del_init(&op->list);
> }
> - return 0;
> -}
> -
> -void arch_optimize_kprobes(struct list_head *oplist)
> -{
> - stop_machine(__arch_optimize_kprobes, oplist, cpu_online_mask);
> }
>
> void arch_unoptimize_kprobe(struct optimized_kprobe *op)
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 13:22 ` Jon Medhurst (Tixy)
@ 2014-12-08 13:48 ` Wang Nan
-1 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 13:48 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: masami.hiramatsu.pt, linux, lizefan, linux-kernel, linux-arm-kernel
On 2014/12/8 21:22, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 20:31 +0800, Wang Nan wrote:
>> On 2014/12/8 20:06, Wang Nan wrote:
>>> On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
>>>> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>>>>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>>>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>> [...]
>>>>
>>>> so another CPU could find and delete next before this one has finished
>>>> doing so. Would the list end up in a consistent state where no loops
>>>> develop and no probes are missed? I don't know the answer and a full
>>>> analysis would be complicated, but my gut feeling is that if a cpu can
>>>> observe the links in the list in an inconsistent state then only bad
>>>> things can result.
>>>>
>>>
>>> I see the problem.
>>>
>>> I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
>>> stop_machine() is required when removing breakpoint, so I'd like to define
>>> a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
>>> Do you think it is a good idea?
>>>
>>>
>>
>> What I mean is something like this:
>
> Yes, that should work, though as remove_breakpoint is a globally visible
> symbol, I suggest a less generic name for it, perhaps
> remove_kprobe_breakpoint ?
>
I don't think it is globally visible. Only files in arm/probes/kprobes
can include "core.h". However, I do agree that remove_breakpoint() is not
a good name. In my v15 patch, I'd like to rename it to kprobe_remove_breakpoint(),
due to may of names defined in core.h are called kprobes_xxx.
Thank you!
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-08 13:48 ` Wang Nan
0 siblings, 0 replies; 36+ messages in thread
From: Wang Nan @ 2014-12-08 13:48 UTC (permalink / raw)
To: linux-arm-kernel
On 2014/12/8 21:22, Jon Medhurst (Tixy) wrote:
> On Mon, 2014-12-08 at 20:31 +0800, Wang Nan wrote:
>> On 2014/12/8 20:06, Wang Nan wrote:
>>> On 2014/12/8 19:50, Jon Medhurst (Tixy) wrote:
>>>> On Mon, 2014-12-08 at 19:15 +0800, Wang Nan wrote:
>>>>> On 2014/12/8 19:04, Jon Medhurst (Tixy) wrote:
>>>>>> On Mon, 2014-12-08 at 14:28 +0800, Wang Nan wrote:
>> [...]
>>>>
>>>> so another CPU could find and delete next before this one has finished
>>>> doing so. Would the list end up in a consistent state where no loops
>>>> develop and no probes are missed? I don't know the answer and a full
>>>> analysis would be complicated, but my gut feeling is that if a cpu can
>>>> observe the links in the list in an inconsistent state then only bad
>>>> things can result.
>>>>
>>>
>>> I see the problem.
>>>
>>> I'm thinking about making core.c and opt-arm.c to share stop_machine() code.
>>> stop_machine() is required when removing breakpoint, so I'd like to define
>>> a "remove_breakpoint" function in core.c and make opt-arm.c to call it.
>>> Do you think it is a good idea?
>>>
>>>
>>
>> What I mean is something like this:
>
> Yes, that should work, though as remove_breakpoint is a globally visible
> symbol, I suggest a less generic name for it, perhaps
> remove_kprobe_breakpoint ?
>
I don't think it is globally visible. Only files in arm/probes/kprobes
can include "core.h". However, I do agree that remove_breakpoint() is not
a good name. In my v15 patch, I'd like to rename it to kprobe_remove_breakpoint(),
due to may of names defined in core.h are called kprobes_xxx.
Thank you!
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-08 11:50 ` Jon Medhurst (Tixy)
@ 2014-12-09 10:14 ` Masami Hiramatsu
-1 siblings, 0 replies; 36+ messages in thread
From: Masami Hiramatsu @ 2014-12-09 10:14 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: Wang Nan, lizefan, linux, linux-kernel, linux-arm-kernel
(2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> iterating over a list of probes and removing each one in turn, if this
> is happening on multiple cpu's simultaneously, it's not clear to me that
> such an operation is safe. list_del_init calls __list_del which does
>
> next->prev = prev;
> prev->next = next;
>
> so what happens if another cpu is at the same time updating any of those
> list entries? Without even fully analysing the code I can see that with
> the fact that the list handling helpers have no memory barriers, that
> the above two lines could be seen to execute in the reverse order, e.g.
>
> prev->next = next;
> next->prev = prev;
>
> so another CPU could find and delete next before this one has finished
> doing so. Would the list end up in a consistent state where no loops
> develop and no probes are missed? I don't know the answer and a full
> analysis would be complicated, but my gut feeling is that if a cpu can
> observe the links in the list in an inconsistent state then only bad
> things can result.
Just a comment, arch_optimize_kprobes() are only called under
kprobe_mutex held. No concurrent update happens :)
Thank you,
--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-09 10:14 ` Masami Hiramatsu
0 siblings, 0 replies; 36+ messages in thread
From: Masami Hiramatsu @ 2014-12-09 10:14 UTC (permalink / raw)
To: linux-arm-kernel
(2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> iterating over a list of probes and removing each one in turn, if this
> is happening on multiple cpu's simultaneously, it's not clear to me that
> such an operation is safe. list_del_init calls __list_del which does
>
> next->prev = prev;
> prev->next = next;
>
> so what happens if another cpu is at the same time updating any of those
> list entries? Without even fully analysing the code I can see that with
> the fact that the list handling helpers have no memory barriers, that
> the above two lines could be seen to execute in the reverse order, e.g.
>
> prev->next = next;
> next->prev = prev;
>
> so another CPU could find and delete next before this one has finished
> doing so. Would the list end up in a consistent state where no loops
> develop and no probes are missed? I don't know the answer and a full
> analysis would be complicated, but my gut feeling is that if a cpu can
> observe the links in the list in an inconsistent state then only bad
> things can result.
Just a comment, arch_optimize_kprobes() are only called under
kprobe_mutex held. No concurrent update happens :)
Thank you,
--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt at hitachi.com
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-09 10:14 ` Masami Hiramatsu
@ 2014-12-09 10:30 ` Jon Medhurst (Tixy)
-1 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-09 10:30 UTC (permalink / raw)
To: Masami Hiramatsu; +Cc: Wang Nan, lizefan, linux, linux-kernel, linux-arm-kernel
On Tue, 2014-12-09 at 19:14 +0900, Masami Hiramatsu wrote:
> (2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> > iterating over a list of probes and removing each one in turn, if this
> > is happening on multiple cpu's simultaneously, it's not clear to me that
> > such an operation is safe. list_del_init calls __list_del which does
> >
> > next->prev = prev;
> > prev->next = next;
> >
> > so what happens if another cpu is at the same time updating any of those
> > list entries? Without even fully analysing the code I can see that with
> > the fact that the list handling helpers have no memory barriers, that
> > the above two lines could be seen to execute in the reverse order, e.g.
> >
> > prev->next = next;
> > next->prev = prev;
> >
> > so another CPU could find and delete next before this one has finished
> > doing so. Would the list end up in a consistent state where no loops
> > develop and no probes are missed? I don't know the answer and a full
> > analysis would be complicated, but my gut feeling is that if a cpu can
> > observe the links in the list in an inconsistent state then only bad
> > things can result.
>
> Just a comment, arch_optimize_kprobes() are only called under
> kprobe_mutex held. No concurrent update happens :)
Except in the case of the code I was commenting on which was using
stop_machine to make all cpu's simultaneously do the work of
arch_optimize_kprobes :-)
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-09 10:30 ` Jon Medhurst (Tixy)
0 siblings, 0 replies; 36+ messages in thread
From: Jon Medhurst (Tixy) @ 2014-12-09 10:30 UTC (permalink / raw)
To: linux-arm-kernel
On Tue, 2014-12-09 at 19:14 +0900, Masami Hiramatsu wrote:
> (2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
> > iterating over a list of probes and removing each one in turn, if this
> > is happening on multiple cpu's simultaneously, it's not clear to me that
> > such an operation is safe. list_del_init calls __list_del which does
> >
> > next->prev = prev;
> > prev->next = next;
> >
> > so what happens if another cpu is at the same time updating any of those
> > list entries? Without even fully analysing the code I can see that with
> > the fact that the list handling helpers have no memory barriers, that
> > the above two lines could be seen to execute in the reverse order, e.g.
> >
> > prev->next = next;
> > next->prev = prev;
> >
> > so another CPU could find and delete next before this one has finished
> > doing so. Would the list end up in a consistent state where no loops
> > develop and no probes are missed? I don't know the answer and a full
> > analysis would be complicated, but my gut feeling is that if a cpu can
> > observe the links in the list in an inconsistent state then only bad
> > things can result.
>
> Just a comment, arch_optimize_kprobes() are only called under
> kprobe_mutex held. No concurrent update happens :)
Except in the case of the code I was commenting on which was using
stop_machine to make all cpu's simultaneously do the work of
arch_optimize_kprobes :-)
--
Tixy
^ permalink raw reply [flat|nested] 36+ messages in thread
* Re: [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
2014-12-09 10:30 ` Jon Medhurst (Tixy)
@ 2014-12-09 15:13 ` Masami Hiramatsu
-1 siblings, 0 replies; 36+ messages in thread
From: Masami Hiramatsu @ 2014-12-09 15:13 UTC (permalink / raw)
To: Jon Medhurst (Tixy)
Cc: Wang Nan, lizefan, linux, linux-kernel, linux-arm-kernel
(2014/12/09 19:30), Jon Medhurst (Tixy) wrote:
> On Tue, 2014-12-09 at 19:14 +0900, Masami Hiramatsu wrote:
>> (2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
>>> iterating over a list of probes and removing each one in turn, if this
>>> is happening on multiple cpu's simultaneously, it's not clear to me that
>>> such an operation is safe. list_del_init calls __list_del which does
>>>
>>> next->prev = prev;
>>> prev->next = next;
>>>
>>> so what happens if another cpu is at the same time updating any of those
>>> list entries? Without even fully analysing the code I can see that with
>>> the fact that the list handling helpers have no memory barriers, that
>>> the above two lines could be seen to execute in the reverse order, e.g.
>>>
>>> prev->next = next;
>>> next->prev = prev;
>>>
>>> so another CPU could find and delete next before this one has finished
>>> doing so. Would the list end up in a consistent state where no loops
>>> develop and no probes are missed? I don't know the answer and a full
>>> analysis would be complicated, but my gut feeling is that if a cpu can
>>> observe the links in the list in an inconsistent state then only bad
>>> things can result.
>>
>> Just a comment, arch_optimize_kprobes() are only called under
>> kprobe_mutex held. No concurrent update happens :)
>
> Except in the case of the code I was commenting on which was using
> stop_machine to make all cpu's simultaneously do the work of
> arch_optimize_kprobes :-)
Ah, right! stop_machine with cpu_online_mask cause that problem.
Thanks,
--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com
^ permalink raw reply [flat|nested] 36+ messages in thread
* [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32
@ 2014-12-09 15:13 ` Masami Hiramatsu
0 siblings, 0 replies; 36+ messages in thread
From: Masami Hiramatsu @ 2014-12-09 15:13 UTC (permalink / raw)
To: linux-arm-kernel
(2014/12/09 19:30), Jon Medhurst (Tixy) wrote:
> On Tue, 2014-12-09 at 19:14 +0900, Masami Hiramatsu wrote:
>> (2014/12/08 20:50), Jon Medhurst (Tixy) wrote:> arch_optimize_kprobes is calling __arch_optimize_kprobes, which is
>>> iterating over a list of probes and removing each one in turn, if this
>>> is happening on multiple cpu's simultaneously, it's not clear to me that
>>> such an operation is safe. list_del_init calls __list_del which does
>>>
>>> next->prev = prev;
>>> prev->next = next;
>>>
>>> so what happens if another cpu is at the same time updating any of those
>>> list entries? Without even fully analysing the code I can see that with
>>> the fact that the list handling helpers have no memory barriers, that
>>> the above two lines could be seen to execute in the reverse order, e.g.
>>>
>>> prev->next = next;
>>> next->prev = prev;
>>>
>>> so another CPU could find and delete next before this one has finished
>>> doing so. Would the list end up in a consistent state where no loops
>>> develop and no probes are missed? I don't know the answer and a full
>>> analysis would be complicated, but my gut feeling is that if a cpu can
>>> observe the links in the list in an inconsistent state then only bad
>>> things can result.
>>
>> Just a comment, arch_optimize_kprobes() are only called under
>> kprobe_mutex held. No concurrent update happens :)
>
> Except in the case of the code I was commenting on which was using
> stop_machine to make all cpu's simultaneously do the work of
> arch_optimize_kprobes :-)
Ah, right! stop_machine with cpu_online_mask cause that problem.
Thanks,
--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt at hitachi.com
^ permalink raw reply [flat|nested] 36+ messages in thread
end of thread, other threads:[~2014-12-09 15:13 UTC | newest]
Thread overview: 36+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-12-08 6:27 [PATCH v14 0/7] ARM: kprobes: OPTPROBES and other improvements Wang Nan
2014-12-08 6:27 ` Wang Nan
2014-12-08 6:27 ` [PATCH v14 1/7] ARM: probes: move all probe code to dedicate directory Wang Nan
2014-12-08 6:27 ` Wang Nan
2014-12-08 6:27 ` [PATCH v14 2/7] ARM: kprobes: introduces checker Wang Nan
2014-12-08 6:27 ` Wang Nan
2014-12-08 6:28 ` [PATCH v14 3/7] ARM: kprobes: collects stack consumption for store instructions Wang Nan
2014-12-08 6:28 ` Wang Nan
2014-12-08 6:28 ` [PATCH v14 4/7] ARM: kprobes: disallow probing stack consuming instructions Wang Nan
2014-12-08 6:28 ` Wang Nan
2014-12-08 6:28 ` [PATCH v14 5/7] ARM: kprobes: Add test cases for " Wang Nan
2014-12-08 6:28 ` Wang Nan
2014-12-08 6:28 ` [PATCH v14 6/7] kprobes: Pass the original kprobe for preparing optimized kprobe Wang Nan
2014-12-08 6:28 ` Wang Nan
2014-12-08 6:28 ` [PATCH v14 7/7] ARM: kprobes: enable OPTPROBES for ARM 32 Wang Nan
2014-12-08 6:28 ` Wang Nan
2014-12-08 11:04 ` Jon Medhurst (Tixy)
2014-12-08 11:04 ` Jon Medhurst (Tixy)
2014-12-08 11:15 ` Wang Nan
2014-12-08 11:15 ` Wang Nan
2014-12-08 11:50 ` Jon Medhurst (Tixy)
2014-12-08 11:50 ` Jon Medhurst (Tixy)
2014-12-08 12:06 ` Wang Nan
2014-12-08 12:06 ` Wang Nan
2014-12-08 12:31 ` Wang Nan
2014-12-08 12:31 ` Wang Nan
2014-12-08 13:22 ` Jon Medhurst (Tixy)
2014-12-08 13:22 ` Jon Medhurst (Tixy)
2014-12-08 13:48 ` Wang Nan
2014-12-08 13:48 ` Wang Nan
2014-12-09 10:14 ` Masami Hiramatsu
2014-12-09 10:14 ` Masami Hiramatsu
2014-12-09 10:30 ` Jon Medhurst (Tixy)
2014-12-09 10:30 ` Jon Medhurst (Tixy)
2014-12-09 15:13 ` Masami Hiramatsu
2014-12-09 15:13 ` Masami Hiramatsu
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.