Skip to contentSkip to conversation

arm64: improve debuggability

PATCH Superseded Nadav Amit 3 participants 17 messages Jun 17, 2023 – Jun 26, 2023

Subsystems (3)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Nadav Amit

author
#0 lore raw

cover arm64: improve debuggability

From: Nadav Amit <[email protected]>

My recent experience in debugging ARM64 tests on EFI was not as fun as I expected it to be.

There were several reasons for that besides the questionable definition of "fun":

1. ARM64 is not compiled with frame pointers and there is no stack unwinder when the stack is dumped.

2. Building an EFI drops the debug information.

3. The addresses that are printed on dump_stack() and the use of GDB are hard because taking code relocation into account is non trivial.

The patches help both ARM64 and EFI for this matter. The image address is printed when EFI is used to allow the use of GDB. Symbols are emitted into a separate debug file. The frame pointer is included and special entry is added upon an exception to allow backtracing across exceptions.

Nadav Amit (6):
arm: keep efi debug information in a separate file lib/stack: print base addresses on efi
arm64: enable frame pointer and support stack unwinding
arm64: stack: update trace stack on exception
efi: Print address of image
arm64: dump stack on bad exception

 arm/Makefile.arm        |  3 ---
 arm/Makefile.arm64      |  1 +
 arm/Makefile.common     |  8 ++++++-
 arm/cstart64.S          | 13 ++++++++++
 lib/arm64/asm-offsets.c |  3 ++-
 lib/arm64/asm/stack.h   |  3 +++
 lib/arm64/processor.c   |  1 +
 lib/arm64/stack.c       | 53 +++++++++++++++++++++++++++++++++++++++++
 lib/efi.c               |  3 +++
 lib/stack.c             | 31 ++++++++++++++++++++++--
 10 files changed, 112 insertions(+), 7 deletions(-)
 create mode 100644 lib/arm64/stack.c
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 1/6 arm: keep efi debug information in a separate file

From: Nadav Amit <[email protected]>

Debugging tests that run on EFI is hard because the debug information is not included in the EFI file. Dump it into a separeate .debug file to allow the use of gdb or pretty_print_stacks script.

---
 arm/Makefile.common | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
arm/Makefile.common +4 -1
index d60cf8c..f904702 100644
@@ -69,7 +69,7 @@ FLATLIBS = $(libcflat) $(LIBFDT_archive) $(libeabi)
 ifeq ($(CONFIG_EFI),y)
 %.so: EFI_LDFLAGS += -defsym=EFI_SUBSYSTEM=0xa --no-undefined
 %.so: %.o $(FLATLIBS) $(SRCDIR)/arm/efi/elf_aarch64_efi.lds $(cstart.o)
-	$(CC) $(CFLAGS) -c -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
+	$(CC) $(CFLAGS) -c -g -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
 		-DPROGNAME=\"$(@:.so=.efi)\" -DAUXFLAGS=$(AUXFLAGS)
 	$(LD) $(EFI_LDFLAGS) -o $@ -T $(SRCDIR)/arm/efi/elf_aarch64_efi.lds \
 		$(filter %.o, $^) $(FLATLIBS) $(@:.so=.aux.o) \
@@ -78,6 +78,9 @@ ifeq ($(CONFIG_EFI),y)
 
 %.efi: %.so
 	$(call arch_elf_check, $^)
+	$(OBJCOPY) --only-keep-debug $^ [email protected]
+	$(OBJCOPY) --strip-debug $^
+	$(OBJCOPY) [email protected] $^
 	$(OBJCOPY) \
 		-j .text -j .sdata -j .data -j .dynamic -j .dynsym \
 		-j .rel -j .rela -j .rel.* -j .rela.* -j .rel* -j .rela* \
Signature · 1 line
2.34.1
#2 lore raw

Re: [kvm-unit-tests PATCH 1/6] arm: keep efi debug information in a separate file

Quoted 37 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Debugging tests that run on EFI is hard because the debug information is not included in the EFI file. Dump it into a separeate .debug file to allow the use of gdb or pretty_print_stacks script.

---
 arm/Makefile.common | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
arm/Makefile.common +4 -1
index d60cf8c..f904702 100644
@@ -69,7 +69,7 @@ FLATLIBS = $(libcflat) $(LIBFDT_archive) $(libeabi)
 ifeq ($(CONFIG_EFI),y)
 %.so: EFI_LDFLAGS += -defsym=EFI_SUBSYSTEM=0xa --no-undefined
 %.so: %.o $(FLATLIBS) $(SRCDIR)/arm/efi/elf_aarch64_efi.lds $(cstart.o)
-	$(CC) $(CFLAGS) -c -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
+	$(CC) $(CFLAGS) -c -g -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
 		-DPROGNAME=\"$(@:.so=.efi)\" -DAUXFLAGS=$(AUXFLAGS)
 	$(LD) $(EFI_LDFLAGS) -o $@ -T $(SRCDIR)/arm/efi/elf_aarch64_efi.lds \
 		$(filter %.o, $^) $(FLATLIBS) $(@:.so=.aux.o) \
@@ -78,6 +78,9 @@ ifeq ($(CONFIG_EFI),y)
 
 %.efi: %.so
 	$(call arch_elf_check, $^)
+	$(OBJCOPY) --only-keep-debug $^ [email protected]
+	$(OBJCOPY) --strip-debug $^
+	$(OBJCOPY) [email protected] $^
 	$(OBJCOPY) \
 		-j .text -j .sdata -j .data -j .dynamic -j .dynsym \
 		-j .rel -j .rela -j .rel.* -j .rela.* -j .rel* -j .rela* \
Signature · 1 line
2.34.1

Any reason to not also do this for x86?

One reason I ask, is that in order for pretty_print_stacks.py to make use of this from run_tests.sh we need a patch like

run_tests.sh +7 -1
index f61e0057b537..67b239f1adc7 100755
@@ -125,8 +125,14 @@ fi
 RUNTIME_log_stderr () { process_test_output "$1"; }
 RUNTIME_log_stdout () {
     local testname="$1"
+    local kernel
+
     if [ "$PRETTY_PRINT_STACKS" = "yes" ]; then
-        local kernel="$2"
+        if [ "$CONFIG_EFI" = "y" ]; then
+            kernel="${TEST_DIR}/${2}.efi.debug"
+        else
+            kernel="$2"
+        fi
         ./scripts/pretty_print_stacks.py "$kernel" | process_test_output "$testname"
     else
         process_test_output "$testname"

We'd have to special-case that CONFIG_EFI condition for arm64 if we don't also create .efi.debug files for x86.

Thanks,
drew

#3 lore raw

Re: [kvm-unit-tests PATCH 1/6] arm: keep efi debug information in a separate file

In reply to Nadav Amit
Quoted 37 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Debugging tests that run on EFI is hard because the debug information is not included in the EFI file. Dump it into a separeate .debug file to allow the use of gdb or pretty_print_stacks script.

---
 arm/Makefile.common | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
arm/Makefile.common +4 -1
index d60cf8c..f904702 100644
@@ -69,7 +69,7 @@ FLATLIBS = $(libcflat) $(LIBFDT_archive) $(libeabi)
 ifeq ($(CONFIG_EFI),y)
 %.so: EFI_LDFLAGS += -defsym=EFI_SUBSYSTEM=0xa --no-undefined
 %.so: %.o $(FLATLIBS) $(SRCDIR)/arm/efi/elf_aarch64_efi.lds $(cstart.o)
-	$(CC) $(CFLAGS) -c -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
+	$(CC) $(CFLAGS) -c -g -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
 		-DPROGNAME=\"$(@:.so=.efi)\" -DAUXFLAGS=$(AUXFLAGS)
 	$(LD) $(EFI_LDFLAGS) -o $@ -T $(SRCDIR)/arm/efi/elf_aarch64_efi.lds \
 		$(filter %.o, $^) $(FLATLIBS) $(@:.so=.aux.o) \
@@ -78,6 +78,9 @@ ifeq ($(CONFIG_EFI),y)
 
 %.efi: %.so
 	$(call arch_elf_check, $^)
+	$(OBJCOPY) --only-keep-debug $^ [email protected]
+	$(OBJCOPY) --strip-debug $^
+	$(OBJCOPY) [email protected] $^
 	$(OBJCOPY) \
 		-j .text -j .sdata -j .data -j .dynamic -j .dynsym \
 		-j .rel -j .rela -j .rel.* -j .rela.* -j .rel* -j .rela* \
Signature · 1 line
2.34.1

We also need a change to arm_clean to clean these new files up. Or, since we probably want them for x86 as well and we already have other efi cleanup to do, then maybe we should have a common efi_clean in the top- level Makefile which x86's and arm's clean use to clean up all efi related files.

Thanks,
drew

#4 lore raw

Re: [kvm-unit-tests PATCH 1/6] arm: keep efi debug information in a separate file

Quoted 42 lines · Andrew Jones

!! External Email

Quoted 35 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Debugging tests that run on EFI is hard because the debug information is not included in the EFI file. Dump it into a separeate .debug file to allow the use of gdb or pretty_print_stacks script.

---
arm/Makefile.common | 5 ++++-

1 file changed, 4 insertions(+), 1 deletion(-)
arm/Makefile.common +0 -0
index d60cf8c..f904702 100644
@@ -69,7 +69,7 @@ FLATLIBS = $(libcflat) $(LIBFDT_archive) $(libeabi)
 ifeq ($(CONFIG_EFI),y)

%.so: EFI_LDFLAGS += -defsym=EFI_SUBSYSTEM=0xa --no-undefined %.so: %.o $(FLATLIBS) $(SRCDIR)/arm/efi/elf_aarch64_efi.lds $(cstart.o)

arm/Makefile.common +1 -1
-     $(CC) $(CFLAGS) -c -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
+     $(CC) $(CFLAGS) -c -g -o $(@:.so=.aux.o) $(SRCDIR)/lib/auxinfo.c \
             -DPROGNAME=\"$(@:.so=.efi)\" -DAUXFLAGS=$(AUXFLAGS)
     $(LD) $(EFI_LDFLAGS) -o $@ -T $(SRCDIR)/arm/efi/elf_aarch64_efi.lds \
             $(filter %.o, $^) $(FLATLIBS) $(@:.so=.aux.o) \
@@ -78,6 +78,9 @@ ifeq ($(CONFIG_EFI),y)
 

%.efi: %.so

     $(call arch_elf_check, $^)
arm/Makefile.common +3 -1
+     $(OBJCOPY) --only-keep-debug $^ [email protected]
+     $(OBJCOPY) --strip-debug $^
+     $(OBJCOPY) [email protected] $^
     $(OBJCOPY) \
             -j .text -j .sdata -j .data -j .dynamic -j .dynsym \
             -j .rel -j .rela -j .rel.* -j .rela.* -j .rel* -j .rela* \
--
Quoted 8 lines
Quoted 2 lines

2.34.1

We also need a change to arm_clean to clean these new files up. Or, since we probably want them for x86 as well and we already have other efi cleanup to do, then maybe we should have a common efi_clean in the top- level Makefile which x86's and arm's clean use to clean up all efi related files.

[ I aggregate my response to both of your emails ]

Thanks!

I will create .debug files for x86, but I am going to remove the ‘-g’ option. If anyone wants it, he would need to provide CFLAGS during configure.

I will cleanup the .debug files for each arch separately since other files are cleanup’d that way.

author
#5 lore raw

PATCH 2/6 lib/stack: print base addresses on efi

In reply to Nadav Amit

From: Nadav Amit <[email protected]>

Making sense from dumped stacks when running EFI tests is very hard due to the relocation. Fix it by adjusting the address back to the original address.

---
 lib/stack.c | 31 +++++++++++++++++++++++++++++--
 1 file changed, 29 insertions(+), 2 deletions(-)
lib/stack.c +29 -2
index bdb23fd..c3c7c24 100644
@@ -6,13 +6,38 @@
  */
 
 #include <libcflat.h>
+#include <stdbool.h>
 #include <stack.h>
 
 #define MAX_DEPTH 20
 
+#ifdef CONFIG_EFI
+extern char _text, _etext;
+
+static bool base_address(const void *rebased_addr, unsigned long *addr)
+{
+	unsigned long ra = (unsigned long)rebased_addr;
+	unsigned long start = (unsigned long)&_text;
+	unsigned long end = (unsigned long)&_etext;
+
+	if (ra < start || ra >= end)
+		return false;
+
+	*addr = ra - start;
+	return true;
+}
+#else
+static bool base_address(const void *rebased_addr, unsigned long *addr)
+{
+	*addr = (unsigned long)rebased_addr;
+	return true;
+}
+#endif
+
 static void print_stack(const void **return_addrs, int depth,
 			bool top_is_return_address)
 {
+	unsigned long addr;
 	int i = 0;
 
 	printf("\tSTACK:");
@@ -20,12 +45,14 @@ static void print_stack(const void **return_addrs, int depth,
 	/* @addr indicates a non-return address, as expected by the stack
 	 * pretty printer script. */
 	if (depth > 0 && !top_is_return_address) {
-		printf(" @%lx", (unsigned long) return_addrs[0]);
+		if (base_address(return_addrs[0], &addr))
+			printf(" @%lx", addr);
 		i++;
 	}
 
 	for (; i < depth; i++) {
-		printf(" %lx", (unsigned long) return_addrs[i]);
+		if (base_address(return_addrs[i], &addr))
+			printf(" %lx", addr);
 	}
 	printf("\n");
 }
Signature · 1 line
2.34.1
#6 lore raw

Re: [kvm-unit-tests PATCH 2/6] lib/stack: print base addresses on efi

Quoted 16 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Making sense from dumped stacks when running EFI tests is very hard due to the relocation. Fix it by adjusting the address back to the original address.

---
 lib/stack.c | 31 +++++++++++++++++++++++++++++--
 1 file changed, 29 insertions(+), 2 deletions(-)
lib/stack.c +0 -0
index bdb23fd..c3c7c24 100644

lib/stack.c line 14

Quoted 9 lines
lib/stack.c +2 -0
@@ -6,13 +6,38 @@
  */
 
 #include <libcflat.h>
+#include <stdbool.h>
 #include <stack.h>
 
 #define MAX_DEPTH 20
 
+#ifdef CONFIG_EFI

We also relocate when building non-efi arm64 unit tests, so this should be

#if defined(CONFIG_EFI) || defined(__aarch64__)

Or, we could create a new config, CONFIG_RELOC, which gets set by --enable-efi and --arch=arm64.

Quoted 49 lines
lib/stack.c +27 -2
+extern char _text, _etext;
+
+static bool base_address(const void *rebased_addr, unsigned long *addr)
+{
+	unsigned long ra = (unsigned long)rebased_addr;
+	unsigned long start = (unsigned long)&_text;
+	unsigned long end = (unsigned long)&_etext;
+
+	if (ra < start || ra >= end)
+		return false;
+
+	*addr = ra - start;
+	return true;
+}
+#else
+static bool base_address(const void *rebased_addr, unsigned long *addr)
+{
+	*addr = (unsigned long)rebased_addr;
+	return true;
+}
+#endif
+
 static void print_stack(const void **return_addrs, int depth,
 			bool top_is_return_address)
 {
+	unsigned long addr;
 	int i = 0;
 
 	printf("\tSTACK:");
@@ -20,12 +45,14 @@ static void print_stack(const void **return_addrs, int depth,
 	/* @addr indicates a non-return address, as expected by the stack
 	 * pretty printer script. */
 	if (depth > 0 && !top_is_return_address) {
-		printf(" @%lx", (unsigned long) return_addrs[0]);
+		if (base_address(return_addrs[0], &addr))
+			printf(" @%lx", addr);
 		i++;
 	}
 
 	for (; i < depth; i++) {
-		printf(" %lx", (unsigned long) return_addrs[i]);
+		if (base_address(return_addrs[i], &addr))
+			printf(" %lx", addr);
 	}
 	printf("\n");
 }
Signature · 1 line
2.34.1

Thanks,
drew

#7 lore raw

Re: [kvm-unit-tests PATCH 2/6] lib/stack: print base addresses on efi

Quoted 10 lines · Andrew Jones

We also relocate when building non-efi arm64 unit tests, so this should be

#if defined(CONFIG_EFI) || defined(__aarch64__)

Or, we could create a new config, CONFIG_RELOC, which gets set by --enable-efi and --arch=arm64.

Will do. Thanks.

author
#8 lore raw

PATCH 3/6 arm64: enable frame pointer and support stack unwinding

In reply to Nadav Amit

From: Nadav Amit <[email protected]>

Enable frame pointers for arm64 and perform stack unwinding based on arm64 convention.

---
 arm/Makefile.arm      |  3 ---
 arm/Makefile.arm64    |  1 +
 arm/Makefile.common   |  3 +++
 lib/arm64/asm/stack.h |  3 +++
 lib/arm64/stack.c     | 37 +++++++++++++++++++++++++++++++++++++
 5 files changed, 44 insertions(+), 3 deletions(-)
 create mode 100644 lib/arm64/stack.c
arm/Makefile.arm +0 -3
index 2ce00f5..7fd39f3 100644
@@ -11,9 +11,6 @@ ifeq ($(CONFIG_EFI),y)
 $(error Cannot build arm32 tests as EFI apps)
 endif
 
-# stack.o relies on frame pointers.
-KEEP_FRAME_POINTER := y
-
 CFLAGS += $(machine)
 CFLAGS += -mcpu=$(PROCESSOR)
 CFLAGS += -mno-unaligned-access
arm/Makefile.arm64 +1 -0
index eada7f9..60385e2 100644
@@ -21,6 +21,7 @@ define arch_elf_check =
 endef
 
 cstart.o = $(TEST_DIR)/cstart64.o
+cflatobjs += lib/arm64/stack.o
 cflatobjs += lib/arm64/processor.o
 cflatobjs += lib/arm64/spinlock.o
 cflatobjs += lib/arm64/gic-v3-its.o lib/arm64/gic-v3-its-cmd.o
arm/Makefile.common +3 -0
index f904702..7fecfb3 100644
@@ -22,6 +22,9 @@ $(TEST_DIR)/sieve.elf: AUXFLAGS = 0x1
 ##################################################################
 AUXFLAGS ?= 0x0
 
+# stack.o relies on frame pointers.
+KEEP_FRAME_POINTER := y
+
 CFLAGS += -std=gnu99
 CFLAGS += -ffreestanding
 CFLAGS += -O2
lib/arm64/asm/stack.h +3 -0
index d000624..be486cf 100644
@@ -5,4 +5,7 @@
 #error Do not directly include <asm/stack.h>. Just use <stack.h>.
 #endif
 
+#define HAVE_ARCH_BACKTRACE_FRAME
+#define HAVE_ARCH_BACKTRACE
+
 #endif
lib/arm64/stack.c new file +37 -0
new file mode 100644
index 0000000..1e2568a
@@ -0,0 +1,37 @@
+/*
+ * backtrace support (this is a modified lib/x86/stack.c)
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2.
+ */
+#include <libcflat.h>
+#include <stack.h>
+
+extern char vector_stub_start, vector_stub_end;
+
+int backtrace_frame(const void *frame, const void **return_addrs, int max_depth) {
+	const void *fp = frame;
+	void *lr;
+	int depth;
+
+	/*
+	 * ARM64 stack grows down. fp points to the previous fp on the stack,
+	 * and lr is just above it
+	 */
+	for (depth = 0; fp && depth < max_depth; ++depth) {
+
+		asm volatile ("ldp %0, %1, [%2]"
+				  : "=r" (fp), "=r" (lr)
+				  : "r" (fp)
+				  : );
+
+		return_addrs[depth] = lr;
+	}
+
+	return depth;
+}
+
+int backtrace(const void **return_addrs, int max_depth)
+{
+	return backtrace_frame(__builtin_frame_address(0),
+			       return_addrs, max_depth);
+}
Signature · 1 line
2.34.1
#9 lore raw

Re: [kvm-unit-tests PATCH 3/6] arm64: enable frame pointer and support stack unwinding

Quoted 73 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Enable frame pointers for arm64 and perform stack unwinding based on arm64 convention.

---
 arm/Makefile.arm      |  3 ---
 arm/Makefile.arm64    |  1 +
 arm/Makefile.common   |  3 +++
 lib/arm64/asm/stack.h |  3 +++
 lib/arm64/stack.c     | 37 +++++++++++++++++++++++++++++++++++++
 5 files changed, 44 insertions(+), 3 deletions(-)
 create mode 100644 lib/arm64/stack.c
arm/Makefile.arm +0 -3
index 2ce00f5..7fd39f3 100644
@@ -11,9 +11,6 @@ ifeq ($(CONFIG_EFI),y)
 $(error Cannot build arm32 tests as EFI apps)
 endif
 
-# stack.o relies on frame pointers.
-KEEP_FRAME_POINTER := y
-
 CFLAGS += $(machine)
 CFLAGS += -mcpu=$(PROCESSOR)
 CFLAGS += -mno-unaligned-access
arm/Makefile.arm64 +1 -0
index eada7f9..60385e2 100644
@@ -21,6 +21,7 @@ define arch_elf_check =
 endef
 
 cstart.o = $(TEST_DIR)/cstart64.o
+cflatobjs += lib/arm64/stack.o
 cflatobjs += lib/arm64/processor.o
 cflatobjs += lib/arm64/spinlock.o
 cflatobjs += lib/arm64/gic-v3-its.o lib/arm64/gic-v3-its-cmd.o
arm/Makefile.common +3 -0
index f904702..7fecfb3 100644
@@ -22,6 +22,9 @@ $(TEST_DIR)/sieve.elf: AUXFLAGS = 0x1
 ##################################################################
 AUXFLAGS ?= 0x0
 
+# stack.o relies on frame pointers.
+KEEP_FRAME_POINTER := y
+
 CFLAGS += -std=gnu99
 CFLAGS += -ffreestanding
 CFLAGS += -O2
lib/arm64/asm/stack.h +3 -0
index d000624..be486cf 100644
@@ -5,4 +5,7 @@
 #error Do not directly include <asm/stack.h>. Just use <stack.h>.
 #endif
 
+#define HAVE_ARCH_BACKTRACE_FRAME
+#define HAVE_ARCH_BACKTRACE
+
 #endif
lib/arm64/stack.c new file +0 -0
new file mode 100644
index 0000000..1e2568a

lib/arm64/stack.c line 9

Quoted 9 lines
lib/arm64/stack.c new file +9 -0
@@ -0,0 +1,37 @@
+/*
+ * backtrace support (this is a modified lib/x86/stack.c)
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2.
+ */
+#include <libcflat.h>
+#include <stack.h>
+
+extern char vector_stub_start, vector_stub_end;

These aren't used until the next patch.

lib/arm64/stack.c line 11

Quoted 2 lines
lib/arm64/stack.c new file +2 -0
+
+int backtrace_frame(const void *frame, const void **return_addrs, int max_depth) {

'{' should be on its own line. I usually try to run the kernel's checkpatch since we use the same style (except we're even more forgiving for long lines).

Quoted 29 lines
lib/arm64/stack.c new file +26 -0
+	const void *fp = frame;
+	void *lr;
+	int depth;
+
+	/*
+	 * ARM64 stack grows down. fp points to the previous fp on the stack,
+	 * and lr is just above it
+	 */
+	for (depth = 0; fp && depth < max_depth; ++depth) {
+
+		asm volatile ("ldp %0, %1, [%2]"
+				  : "=r" (fp), "=r" (lr)
+				  : "r" (fp)
+				  : );
+
+		return_addrs[depth] = lr;
+	}
+
+	return depth;
+}
+
+int backtrace(const void **return_addrs, int max_depth)
+{
+	return backtrace_frame(__builtin_frame_address(0),
+			       return_addrs, max_depth);
+}
Signature · 1 line
2.34.1

Thanks,
drew

author
#10 lore raw

Re: [kvm-unit-tests PATCH 3/6] arm64: enable frame pointer and support stack unwinding

Quoted 14 lines · Andrew Jones
Quoted 2 lines

+extern char vector_stub_start, vector_stub_end;

These aren't used until the next patch.

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+int backtrace_frame(const void *frame, const void **return_addrs, int max_depth) {

'{' should be on its own line. I usually try to run the kernel's checkpatch since we use the same style (except we're even more forgiving for long lines).

I usually do use checkpatch. I guess I got sloppy. I will fix these 2 issues.

BTW: I used the get_maintainer script to get who to send the patches to and it included the depreciated [email protected] <mailto:[email protected]> .. Ugh.

#11 lore raw

Re: [kvm-unit-tests PATCH 3/6] arm64: enable frame pointer and support stack unwinding

Quoted 15 lines · Nadav Amit
Quoted 14 lines · Andrew Jones
Quoted 2 lines

+extern char vector_stub_start, vector_stub_end;

These aren't used until the next patch.

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+int backtrace_frame(const void *frame, const void **return_addrs, int max_depth) {

'{' should be on its own line. I usually try to run the kernel's checkpatch since we use the same style (except we're even more forgiving for long lines).

Quoted 4 lines

I usually do use checkpatch. I guess I got sloppy. I will fix these 2 issues.

BTW: I used the get_maintainer script to get who to send the patches to and it included the depreciated [email protected] <mailto:[email protected]> .. Ugh.

Ah, thanks for pointing that out. It's safe to declare the depreciation peroid over. I just pushed a patch dropping it now.

drew

author
#12 lore raw

PATCH 4/6 arm64: stack: update trace stack on exception

In reply to Nadav Amit

From: Nadav Amit <[email protected]>

Using gdb for backtracing or dumping the stack following an exception is not very helpful as the exact location of the exception is not saved.

Add an additional frame to save the location of the exception.

One delicate point is dealing with the pretty_print_stacks script. When the stack is dumped, the script would not print the right address for the exception address: for every return address it deducts "1" before looking for the instruction location in the code (using addr2line). As a somewhat hacky solution add "1" for the exception address when dumping the stack.

---
 arm/cstart64.S          | 13 +++++++++++++
 lib/arm64/asm-offsets.c |  3 ++-
 lib/arm64/stack.c       | 16 ++++++++++++++++
 3 files changed, 31 insertions(+), 1 deletion(-)
arm/cstart64.S +13 -0
index cbd6b51..61e27d3 100644
@@ -314,6 +314,13 @@ exceptions_init:
 	mrs	x2, spsr_el1
 	stp	x1, x2, [sp, #S_PC]
 
+	/*
+	 * Save a frame pointer using the link to allow unwinding of
+	 * exceptions.
+	 */
+	stp	x29, x1, [sp, #S_FP]
+	add 	x29, sp, #S_FP
+
 	mov	x0, \vec
 	mov	x1, sp
 	mrs	x2, esr_el1
@@ -349,6 +356,9 @@ exceptions_init:
 	eret
 .endm
 
+vector_stub_start:
+.globl vector_stub_start
+
 vector_stub	el1t_sync,     0
 vector_stub	el1t_irq,      1
 vector_stub	el1t_fiq,      2
@@ -369,6 +379,9 @@ vector_stub	el0_irq_32,   13
 vector_stub	el0_fiq_32,   14
 vector_stub	el0_error_32, 15
 
+vector_stub_end:
+.globl vector_stub_end
+
 .section .text.ex
 
 .macro ventry, label
lib/arm64/asm-offsets.c +2 -1
index 53a1277..7b8bffb 100644
@@ -25,6 +25,7 @@ int main(void)
 	OFFSET(S_PSTATE, pt_regs, pstate);
 	OFFSET(S_ORIG_X0, pt_regs, orig_x0);
 	OFFSET(S_SYSCALLNO, pt_regs, syscallno);
-	DEFINE(S_FRAME_SIZE, sizeof(struct pt_regs));
+	DEFINE(S_FRAME_SIZE, (sizeof(struct pt_regs) + 16));
+	DEFINE(S_FP, sizeof(struct pt_regs));
 	return 0;
 }
lib/arm64/stack.c +16 -0
index 1e2568a..a48ecbb 100644
@@ -12,6 +12,8 @@ int backtrace_frame(const void *frame, const void **return_addrs, int max_depth)
 	const void *fp = frame;
 	void *lr;
 	int depth;
+	bool is_exception = false;
+	unsigned long addr;
 
 	/*
 	 * ARM64 stack grows down. fp points to the previous fp on the stack,
@@ -25,6 +27,20 @@ int backtrace_frame(const void *frame, const void **return_addrs, int max_depth)
 				  : );
 
 		return_addrs[depth] = lr;
+
+		/*
+		 * If this is an exception, add 1 to the pointer so when the
+		 * pretty_print_stacks script is run it would get the right
+		 * address (it deducts 1 to find the call address, but we want
+		 * the actual address).
+		 */
+		if (is_exception)
+			return_addrs[depth] += 1;
+
+		/* Check if we are in the exception handlers for the next entry */
+		addr = (unsigned long)lr;
+		is_exception = (addr >= (unsigned long)&vector_stub_start &&
+				addr < (unsigned long)&vector_stub_end);
 	}
 
 	return depth;
Signature · 1 line
2.34.1
#13 lore raw

Re: [kvm-unit-tests PATCH 4/6] arm64: stack: update trace stack on exception

Quoted 40 lines · Nadav Amit

From: Nadav Amit <[email protected]>

Using gdb for backtracing or dumping the stack following an exception is not very helpful as the exact location of the exception is not saved.

Add an additional frame to save the location of the exception.

One delicate point is dealing with the pretty_print_stacks script. When the stack is dumped, the script would not print the right address for the exception address: for every return address it deducts "1" before looking for the instruction location in the code (using addr2line). As a somewhat hacky solution add "1" for the exception address when dumping the stack.

---
 arm/cstart64.S          | 13 +++++++++++++
 lib/arm64/asm-offsets.c |  3 ++-
 lib/arm64/stack.c       | 16 ++++++++++++++++
 3 files changed, 31 insertions(+), 1 deletion(-)
arm/cstart64.S +7 -0
index cbd6b51..61e27d3 100644
@@ -314,6 +314,13 @@ exceptions_init:
 	mrs	x2, spsr_el1
 	stp	x1, x2, [sp, #S_PC]
 
+	/*
+	 * Save a frame pointer using the link to allow unwinding of
+	 * exceptions.
+	 */
+	stp	x29, x1, [sp, #S_FP]
+	add 	x29, sp, #S_FP
+
 	mov	x0, \vec
 	mov	x1, sp
 	mrs	x2, esr_el1

arm/cstart64.S exceptions_init: · line 360

Quoted 5 lines
arm/cstart64.S +2 -0
@@ -349,6 +356,9 @@ exceptions_init:
 	eret
 .endm
 
+vector_stub_start:
+.globl vector_stub_start

nit: I'd prefer the .globl directives above the labels to match the rest of the file.

Quoted 19 lines
arm/cstart64.S +4 -0
+
 vector_stub	el1t_sync,     0
 vector_stub	el1t_irq,      1
 vector_stub	el1t_fiq,      2
@@ -369,6 +379,9 @@ vector_stub	el0_irq_32,   13
 vector_stub	el0_fiq_32,   14
 vector_stub	el0_error_32, 15
 
+vector_stub_end:
+.globl vector_stub_end
+
 .section .text.ex
 
 .macro ventry, label
lib/arm64/asm-offsets.c +0 -0
index 53a1277..7b8bffb 100644

lib/arm64/asm-offsets.c int main(void) · line 29

Quoted 6 lines
lib/arm64/asm-offsets.c +2 -1
@@ -25,6 +25,7 @@ int main(void)
 	OFFSET(S_PSTATE, pt_regs, pstate);
 	OFFSET(S_ORIG_X0, pt_regs, orig_x0);
 	OFFSET(S_SYSCALLNO, pt_regs, syscallno);
-	DEFINE(S_FRAME_SIZE, sizeof(struct pt_regs));
+	DEFINE(S_FRAME_SIZE, (sizeof(struct pt_regs) + 16));
+	DEFINE(S_FP, sizeof(struct pt_regs));

It'd be good to comment this, something like

  ...
  OFFSET(S_ORIG_X0, pt_regs, orig_x0);
  OFFSET(S_SYSCALLNO, pt_regs, syscallno);

  /* FP and LR (16 bytes) go on the frame above pt_regs */
  DEFINE(S_FP, sizeof(struct pt_regs));
  DEFINE(S_FRAME_SIZE, (sizeof(struct pt_regs) + 16));

return 0;

Quoted 39 lines
lib/arm64/asm-offsets.c +0 -0
 	return 0;
 }
lib/arm64/stack.c +16 -0
index 1e2568a..a48ecbb 100644
@@ -12,6 +12,8 @@ int backtrace_frame(const void *frame, const void **return_addrs, int max_depth)
 	const void *fp = frame;
 	void *lr;
 	int depth;
+	bool is_exception = false;
+	unsigned long addr;
 
 	/*
 	 * ARM64 stack grows down. fp points to the previous fp on the stack,
@@ -25,6 +27,20 @@ int backtrace_frame(const void *frame, const void **return_addrs, int max_depth)
 				  : );
 
 		return_addrs[depth] = lr;
+
+		/*
+		 * If this is an exception, add 1 to the pointer so when the
+		 * pretty_print_stacks script is run it would get the right
+		 * address (it deducts 1 to find the call address, but we want
+		 * the actual address).
+		 */
+		if (is_exception)
+			return_addrs[depth] += 1;
+
+		/* Check if we are in the exception handlers for the next entry */
+		addr = (unsigned long)lr;
+		is_exception = (addr >= (unsigned long)&vector_stub_start &&
+				addr < (unsigned long)&vector_stub_end);
 	}
 
 	return depth;
Signature · 1 line
2.34.1

Thanks,
drew

author
#14 lore raw

PATCH 5/6 efi: Print address of image

In reply to Nadav Amit

From: Nadav Amit <[email protected]>

Using gdb to debug tests that are run from efi is very hard without knowing the base address in which the image was loaded. Print the address so it can be used while debugging.

---
 lib/efi.c | 3 +++
 1 file changed, 3 insertions(+)
lib/efi.c +3 -0
index 2e127a4..f42edd4 100644
@@ -18,6 +18,7 @@
 extern int __argc, __envc;
 extern char *__argv[100];
 extern char *__environ[200];
+extern char _text;
 
 extern int main(int argc, char **argv, char **envp);
 
@@ -363,6 +364,8 @@ efi_status_t efi_main(efi_handle_t handle, efi_system_table_t *sys_tab)
 		goto efi_main_error;
 	}
 
+	printf("Address of image is: 0x%lx\n", (unsigned long)&_text);
+
 	/* Run the test case */
 	ret = main(__argc, __argv, __environ);
 
Signature · 1 line
2.34.1
author
#15 lore raw

PATCH 6/6 arm64: dump stack on bad exception

In reply to Nadav Amit

From: Nadav Amit <[email protected]>

---
 lib/arm64/processor.c | 1 +
 1 file changed, 1 insertion(+)
lib/arm64/processor.c +1 -0
index 831207c..5bcad67 100644
@@ -130,6 +130,7 @@ static void bad_exception(enum vector v, struct pt_regs *regs,
 	printf("Vector: %d (%s)\n", v, vector_names[v]);
 	printf("ESR_EL1: %8s%08x, ec=%#x (%s)\n", "", esr, ec, ec_names[ec]);
 	printf("FAR_EL1: %016lx (%svalid)\n", far, far_valid ? "" : "not ");
+	dump_stack();
 	printf("Exception frame registers:\n");
 	show_regs(regs);
 	abort();
Signature · 1 line
2.34.1